<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE article PUBLIC "-//NLM//DTD Journal Publishing with OASIS Tables v3.0 20080202//EN" "https://jats.nlm.nih.gov/nlm-dtd/publishing/3.0/journalpub-oasis3.dtd">
<article xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:oasis="http://docs.oasis-open.org/ns/oasis-exchange/table" xml:lang="en" dtd-version="3.0" article-type="research-article">
  <front>
    <journal-meta><journal-id journal-id-type="publisher">JECATS</journal-id><journal-title-group>
    <journal-title>Journal of Environmentally Compatible Air Transport System</journal-title>
    <abbrev-journal-title abbrev-type="publisher">JECATS</abbrev-journal-title><abbrev-journal-title abbrev-type="nlm-ta">J. Env. Com. Air Transp. Sys.</abbrev-journal-title>
  </journal-title-group><issn pub-type="epub">3053-9277</issn><publisher>
    <publisher-name>Copernicus Publications</publisher-name>
    <publisher-loc>Göttingen, Germany</publisher-loc>
  </publisher></journal-meta>
    <article-meta>
      <article-id pub-id-type="doi">10.5194/jecats-1-6-2026</article-id><title-group><article-title>Efficacy of scalable airline-led contrail avoidance</article-title><alt-title>Efficacy of scalable airline-led contrail avoidance</alt-title>
      </title-group>
      <contrib-group>
        <contrib contrib-type="author" corresp="yes" rid="aff1">
          <name><surname>Sankar</surname><given-names>Tharun</given-names></name>
          <email>tsankar@google.com</email>
        <ext-link>https://orcid.org/0000-0002-0557-0972</ext-link></contrib>
        <contrib contrib-type="author" corresp="no" rid="aff2">
          <name><surname>Dean</surname><given-names>Thomas</given-names></name>
          
        </contrib>
        <contrib contrib-type="author" corresp="no" rid="aff2">
          <name><surname>Abbott</surname><given-names>Tristan</given-names></name>
          
        </contrib>
        <contrib contrib-type="author" corresp="no" rid="aff3">
          <name><surname>Blickstein</surname><given-names>Jill</given-names></name>
          
        </contrib>
        <contrib contrib-type="author" corresp="no" rid="aff4">
          <name><surname>Martín Frías</surname><given-names>Alejandra</given-names></name>
          
        </contrib>
        <contrib contrib-type="author" corresp="no" rid="aff3">
          <name><surname>Galyen</surname><given-names>Mark</given-names></name>
          
        </contrib>
        <contrib contrib-type="author" corresp="no" rid="aff3">
          <name><surname>Grenham</surname><given-names>Rebecca</given-names></name>
          
        </contrib>
        <contrib contrib-type="author" corresp="no" rid="aff2">
          <name><surname>Hodgson</surname><given-names>Paul</given-names></name>
          
        <ext-link>https://orcid.org/0000-0002-8650-3906</ext-link></contrib>
        <contrib contrib-type="author" corresp="no" rid="aff1">
          <name><surname>McCloskey</surname><given-names>Kevin</given-names></name>
          
        <ext-link>https://orcid.org/0000-0001-9967-4117</ext-link></contrib>
        <contrib contrib-type="author" corresp="no" rid="aff3">
          <name><surname>Pechman</surname><given-names>Alan</given-names></name>
          
        </contrib>
        <contrib contrib-type="author" corresp="no" rid="aff2">
          <name><surname>Robarge</surname><given-names>Tyler</given-names></name>
          
        <ext-link>https://orcid.org/0009-0005-0665-4667</ext-link></contrib>
        <contrib contrib-type="author" corresp="no" rid="aff1">
          <name><surname>Sanekommu</surname><given-names>Dinesh</given-names></name>
          
        </contrib>
        <contrib contrib-type="author" corresp="no" rid="aff1">
          <name><surname>Sarna</surname><given-names>Aaron</given-names></name>
          
        <ext-link>https://orcid.org/0000-0001-8786-5581</ext-link></contrib>
        <contrib contrib-type="author" corresp="no" rid="aff1">
          <name><surname>Sonabend-W</surname><given-names>Aaron</given-names></name>
          
        </contrib>
        <contrib contrib-type="author" corresp="no" rid="aff2 aff5">
          <name><surname>Stettler</surname><given-names>Marc</given-names></name>
          
        <ext-link>https://orcid.org/0000-0002-2066-9380</ext-link></contrib>
        <contrib contrib-type="author" corresp="no" rid="aff4">
          <name><surname>Zopp</surname><given-names>Raimund</given-names></name>
          
        </contrib>
        <contrib contrib-type="author" corresp="no" rid="aff1">
          <name><surname>Geraedts</surname><given-names>Scott</given-names></name>
          
        <ext-link>https://orcid.org/0009-0001-6496-0153</ext-link></contrib>
        <aff id="aff1"><label>1</label><institution>Google Research, 1600 Amphitheatre Pkwy, Mountain View, CA, USA</institution>
        </aff>
        <aff id="aff2"><label>2</label><institution>Contrails.org, 4110 Carillon Point, Kirkland, WA, USA</institution>
        </aff>
        <aff id="aff3"><label>3</label><institution>American Airlines, 1 Skyview Drive, Fort Worth, TX, USA</institution>
        </aff>
        <aff id="aff4"><label>4</label><institution>Flightkeys, Herrengasse 2–4, 1010 Vienna, Austria</institution>
        </aff>
        <aff id="aff5"><label>5</label><institution>Imperial College London, Exhibition Road, South Kensington, London, UK</institution>
        </aff>
      </contrib-group>
      <author-notes><corresp id="corr1">Tharun Sankar (tsankar@google.com)</corresp></author-notes><pub-date><day>19</day><month>November</month><year>2026</year></pub-date>
      
      <volume>1</volume>
      <elocation-id>6</elocation-id>
      <history>
        <date date-type="received"><day>6</day><month>March</month><year>2026</year></date>
           <date date-type="rev-request"><day>21</day><month>April</month><year>2026</year></date>
           <date date-type="rev-recd"><day>30</day><month>October</month><year>2026</year></date>
           <date date-type="accepted"><day>2</day><month>November</month><year>2026</year></date>
      </history>
      <permissions>
        <copyright-statement>Copyright: © 2026 Tharun Sankar et al.</copyright-statement>
        <copyright-year>2026</copyright-year>
      <license license-type="open-access"><license-p>This work is licensed under the Creative Commons Attribution 4.0 International License. To view a copy of this licence, visit <ext-link ext-link-type="uri" xlink:href="https://creativecommons.org/licenses/by/4.0/">https://creativecommons.org/licenses/by/4.0/</ext-link></license-p></license></permissions><self-uri xlink:href="https://jecats.copernicus.org/articles/1/6/2026/jecats-1-6-2026.html">This article is available from https://jecats.copernicus.org/articles/1/6/2026/jecats-1-6-2026.html</self-uri><self-uri xlink:href="https://jecats.copernicus.org/articles/1/6/2026/jecats-1-6-2026.pdf">The full text article is available as a PDF file from https://jecats.copernicus.org/articles/1/6/2026/jecats-1-6-2026.pdf</self-uri>
      <abstract><title>Abstract</title>

      <p id="d2e257">Contrails account for a large portion of aviation's contribution to anthropogenic climate change. Navigational contrail avoidance is a promising solution to mitigate the warming caused by contrails. Prior trials testing navigational contrail avoidance have relied on bespoke integrations of contrail forecasts into airline operations. Here, we use a randomized controlled trial to test the feasibility of dispatcher-led contrail avoidance integrated into standard flight planning operations using a workflow that scales to an airline's entire network. We validated the efficacy of this intervention using satellite imagery and an automated flight-contrail attribution algorithm. Using this system, we observed an 11.6 % reduction in contrail formation rate for the 1232 flights marked as eligible for contrail avoidance (intent-to-treat) relative to the flights in the control group (<inline-formula><mml:math id="M1" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:mo>=</mml:mo><mml:mn mathvariant="normal">0.011</mml:mn></mml:mrow></mml:math></inline-formula>). In the 112 flights that flew contrail avoidance as planned (per-protocol flights), we observed a 62.0 % lower contrail formation rate relative to the flights in the control group (<inline-formula><mml:math id="M2" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:mo>&lt;</mml:mo><mml:mn mathvariant="normal">0.001</mml:mn></mml:mrow></mml:math></inline-formula>). No statistically significant difference in fuel usage was observed between the two groups.</p>
  </abstract>
    </article-meta>
  </front>
<body>
      

<sec id="Ch1.S1" sec-type="intro">
  <label>1</label><title>Introduction</title>
      <p id="d2e293">Aviation contributes to anthropogenic climate change through a combination of carbon dioxide (<inline-formula><mml:math id="M3" display="inline"><mml:mrow class="chem"><mml:msub><mml:mi mathvariant="normal">CO</mml:mi><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>) emissions and non-<inline-formula><mml:math id="M4" display="inline"><mml:mrow class="chem"><mml:msub><mml:mi mathvariant="normal">CO</mml:mi><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> effects <xref ref-type="bibr" rid="bib1.bibx22" id="paren.1"/>. Among these non-<inline-formula><mml:math id="M5" display="inline"><mml:mrow class="chem"><mml:msub><mml:mi mathvariant="normal">CO</mml:mi><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> effects, contrail cirrus – clouds formed by the condensation of water vapor on soot particles in the exhaust of aircraft engines – is estimated to be the largest single component of aviation's net effective radiative forcing (ERF) <xref ref-type="bibr" rid="bib1.bibx21" id="paren.2"/>. Unlike <inline-formula><mml:math id="M6" display="inline"><mml:mrow class="chem"><mml:msub><mml:mi mathvariant="normal">CO</mml:mi><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>, which persists in the atmosphere for centuries, contrails are short-lived climate forcers, lasting from minutes to hours. This transience presents a unique opportunity for rapid climate mitigation: by avoiding the specific atmospheric conditions necessary for the formation of persistent warming contrails, the climate impact of aviation could be significantly reduced on short timescales <xref ref-type="bibr" rid="bib1.bibx39" id="paren.3"/>.</p>
      <p id="d2e350">The formation of persistent contrails requires an aircraft to fly through regions of the atmosphere that are supersaturated with respect to ice (Ice Supersaturated Regions, or ISSRs) and sufficiently cold <xref ref-type="bibr" rid="bib1.bibx35" id="paren.4"/>. These conditions are not usually satisfied <xref ref-type="bibr" rid="bib1.bibx12 bib1.bibx29" id="paren.5"/>. Prior simulation studies quantified how unevenly contrail warming is distributed among the minority of flights that cause persistent contrails. For example, the Contrail Cirrus Prediction Model (CoCiP) <xref ref-type="bibr" rid="bib1.bibx36" id="paren.6"/> – which uses simplified zero-dimensional contrail–cirrus physics <xref ref-type="bibr" rid="bib1.bibx2" id="paren.7"/> and evolves individual contrails independently, typically without atmospheric moisture competition <xref ref-type="bibr" rid="bib1.bibx37" id="paren.8"/> – was used by <xref ref-type="bibr" rid="bib1.bibx30" id="text.9"/> to find that <inline-formula><mml:math id="M7" display="inline"><mml:mrow><mml:mn mathvariant="normal">20</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula> of individual contrails contribute <inline-formula><mml:math id="M8" display="inline"><mml:mrow><mml:mn mathvariant="normal">90</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula> of net contrail energy forcing; <xref ref-type="bibr" rid="bib1.bibx5" id="text.10"/> instead used a fully coupled climate model and reported that the largest <inline-formula><mml:math id="M9" display="inline"><mml:mrow><mml:mn mathvariant="normal">25</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula> of contrail-cirrus clusters contributed <inline-formula><mml:math id="M10" display="inline"><mml:mrow><mml:mn mathvariant="normal">68</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula> of radiative impact. Therefore, minor adjustments to flight altitudes or routes to avoid contrail forming regions –  a strategy known as navigational contrail avoidance –  could potentially eliminate a significant portion of this warming with minimal fuel penalties <xref ref-type="bibr" rid="bib1.bibx9 bib1.bibx6" id="paren.11"/>.</p>
      <p id="d2e423">Despite the theoretical promise, operationalizing contrail avoidance faces significant challenges, some of which are as yet unknown due to limited tests in real operational situations. Navigational avoidance requires accurately forecasting ISSRs in the upper troposphere and lower stratosphere. Recent studies have shown that the majority of contrail warming found in reanalysis products can be avoided through intelligent flight planning using forecast products at the lead times used by flight planners <xref ref-type="bibr" rid="bib1.bibx9 bib1.bibx6" id="paren.12"/>. However, these studies are based on simulations and do not capture whether the suggested contrail avoidance routes are operationally feasible. Additionally, several studies have called into question how well numerical weather prediction (NWP) models and their reanalysis products match real-world observations <xref ref-type="bibr" rid="bib1.bibx1 bib1.bibx12" id="paren.13"/>, which introduces additional uncertainties into simulation-based studies. To address this, recent efforts have leveraged machine learning (ML) approaches trained on satellite observations to improve prediction accuracy <xref ref-type="bibr" rid="bib1.bibx38" id="paren.14"/>. Accurate predictions are important: inaccurate predictions could lead to burning extra fuel to avoid a contrail region that does not exist, or even inadvertently creating additional contrails <xref ref-type="bibr" rid="bib1.bibx23" id="paren.15"/>. For this reason, evaluating the performance of contrail predictions using prospective trials with observational validation is essential.</p>
      <p id="d2e438">Modeling the efficacy of contrail avoidance in the real world would require a deep understanding of both the contrail prediction (and its uncertainties) as well as aircraft operations, which would be an extremely complicated task. An alternative method to measure the efficacy of an intervention on a complex system is a randomized controlled trial (RCT) <xref ref-type="bibr" rid="bib1.bibx31" id="paren.16"/>, and this approach has been applied to measure the efficacy of contrail avoidance. For instance, <xref ref-type="bibr" rid="bib1.bibx38" id="text.17"/> reported on a series of test flights where pilots altered climb and descent profiles based on contrail forecasts. Additionally, <xref ref-type="bibr" rid="bib1.bibx34" id="text.18"/> conducted a randomized trial in the Maastricht Upper Area Control airspace involving tactical altitude adjustments by air traffic controllers. While these trials found that their interventions reduced the number of observed contrails, they relied on bespoke communication channels and manual interventions that do not scale to the high-volume, automated workflows of modern commercial aviation. Furthermore, previous studies sometimes relied on manual inspection of satellite imagery for validation, a process that is labor-intensive.</p>
      <p id="d2e451">In this work, we present the results of a large-scale Randomized Controlled Trial designed to test the feasibility and efficacy of scalable dispatcher-led contrail avoidance integrated directly into standard airline operations. Through a collaboration between American Airlines Inc. (AAL), Flightkeys GmbH, Contrails.org, and Google LLC, we integrated an ML-based contrail forecast into the Flightkeys GmbH platform (hereafter referred to as “Flightkeys”), which AAL uses for creating flight plans in normal operations. This allowed for the generation of contrail-optimized flight plans across the airline's North Atlantic network without relying on labor-intensive planning methods, as the proposal of contrail-optimized flights plans was included in standard workflows.</p>
      <p id="d2e454">The goal of this trial was to reduce observable contrail formation using the satellite-based validation system described in Sect. <xref ref-type="sec" rid="Ch1.S2.SS5"/>, rather than relying on modeling results. In doing this, we aimed to test the end-to-end performance of the forecast system, operational implementation, and validation system. Per-contrail estimates of contrail warming from existing contrail process models are not yet well-validated. Due to this, we selected a contrail forecast that focuses on contrail formation, with less emphasis on the specific warming impact of those contrails. When the forecast predicts persistent contrails based on the underlying weather, it uses a averaged forcing estimate, described in Sect. <xref ref-type="sec" rid="Ch1.S2.SS1"/>, to predict the warming associated with those contrails. The averaged forcing estimate causes our avoidance to focus on contrails which are net warming and primarily during the evening or nighttime, since contrails during the daytime may have a net neutral or net cooling impact. We thereby minimized the risk that the trial increases contrail warming despite reducing persistent contrail formation.</p>
      <p id="d2e461">We validated the efficacy of these avoidance maneuvers using an automated satellite verification system based on the work of <xref ref-type="bibr" rid="bib1.bibx33" id="text.19"/>. This allows for an objective, blinded assessment of contrail formation at a scale previously unattainable. We compared the prevalence of observable contrails in flights that performed avoidance maneuvers against a control group of flights that did not consider contrails, providing statistically robust evidence of observable contrail reduction through scalable, integrated flight planning operations. We observed a 11.6 % reduction in contrail formation rate for the 1232 intent-to-treat flights relative to the control group, as well as a 62.0 % lower contrail formation rate for the 112 per-protocol flights relative to the control group.</p>
</sec>
<sec id="Ch1.S2">
  <label>2</label><title>Methods</title>
<sec id="Ch1.S2.SS1">
  <label>2.1</label><title>Contrail Forecast</title>
      <p id="d2e482">The contrail forecast used during the flight planning process was produced by a ML system trained on contrail detections in the GOES-East geostationary satellite from the Advanced Baseline Imager (ABI) <xref ref-type="bibr" rid="bib1.bibx13" id="paren.20"/>. The ML forecast model takes numerical weather forecast data from the European Center for Medium Range Weather Forecasts (ECMWF) as input and produces a probability of geostationary-observable contrail formation. This forecast system is similar to the one used in <xref ref-type="bibr" rid="bib1.bibx38" id="text.21"/>. More details about this forecast system are provided in Appendix <xref ref-type="sec" rid="App1.Ch1.S1"/>.</p>
      <p id="d2e493">For the training labels of the model, binary contrail labels were produced by a version of the system described in <xref ref-type="bibr" rid="bib1.bibx33" id="text.22"/>. This system was modified slightly in this work to include a filtering based on altitude, where (flight, contrail) pairs are not considered for matching if the radiometrically estimated altitude of the contrail is more than 1615 m different from the advected altitude of the flight. More details are in Appendix <xref ref-type="sec" rid="App1.Ch1.S2"/>.</p>
      <p id="d2e501">The probability of contrail formation derived from the weather forecast was multiplied by a averaged estimate of contrail energy forcing per flight distance obtained from the maps in <xref ref-type="bibr" rid="bib1.bibx30" id="text.23"/>. Energy forcing per flight distance is the radiative forcing (in units of <inline-formula><mml:math id="M11" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">W</mml:mi><mml:mspace width="0.125em" linebreak="nobreak"/><mml:msup><mml:mi mathvariant="normal">m</mml:mi><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">2</mml:mn></mml:mrow></mml:msup></mml:mrow></mml:math></inline-formula>) of a contrail, integrated over the area and lifetime of the contrail and divided by the flight distance which caused the contrail, leading to a result in units of <inline-formula><mml:math id="M12" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">J</mml:mi><mml:mspace linebreak="nobreak" width="0.125em"/><mml:msup><mml:mi mathvariant="normal">m</mml:mi><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:mrow></mml:math></inline-formula>. The map was generated by dividing the atmosphere into bins based on local time, latitude, and season, simulating all flights over a year in each bin using the Contrail Cirrus Prediction Model (CoCiP) <xref ref-type="bibr" rid="bib1.bibx36" id="paren.24"/>, and averaging the warming results. The result is a smoothly varying contrail impact map whose value is large for contrails expected to spend most of their lifetime at night when contrails are net warming, and small during the day when the warming impact is less certain. Incorporating this averaged forcing map into our forecast allowed us to focus on the contrails likely to have a net warming effect, while still enabling the primary goal of reducing observable contrail formation. The forecast used for this trial was set to 0 for all locations East of 30° W longitude since the GOES-East satellite was used for validation, and we did not want to perform contrail avoidance where it could not be validated. More details on the validation system are provided in Sect. <xref ref-type="sec" rid="Ch1.S2.SS5"/>.</p>
      <p id="d2e546">The warming estimates in <xref ref-type="bibr" rid="bib1.bibx30" id="text.25"/> is an instantaneous energy forcing per meter, but during flight planning we need to provide contrails in units of <inline-formula><mml:math id="M13" display="inline"><mml:mrow class="chem"><mml:msub><mml:mi mathvariant="normal">CO</mml:mi><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>-equivalent (<inline-formula><mml:math id="M14" display="inline"><mml:mrow class="chem"><mml:msub><mml:mi mathvariant="normal">CO</mml:mi><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>e) per meter. To do this, we first convert from instantaneous energy forcing to effective energy forcing by using the central estimate “ERF <inline-formula><mml:math id="M15" display="inline"><mml:mo>/</mml:mo></mml:math></inline-formula> RF” ratio of 0.42 from <xref ref-type="bibr" rid="bib1.bibx22" id="text.26"/> (this ratio was derived for radiative forcings rather than energy forcings, but division by a constant commutes with the conversion from radiative forcing to energy forcing so we can use the same factor for energy forcings). Note that this ERF <inline-formula><mml:math id="M16" display="inline"><mml:mo>/</mml:mo></mml:math></inline-formula> RF value is a global average; for a specific contrail, the value can range much more widely. We then apply the 100-year Absolute Global Warming Potentials (AGWP100) <xref ref-type="bibr" rid="bib1.bibx19" id="paren.27"/>, leading to a conversion factor of <inline-formula><mml:math id="M17" display="inline"><mml:mrow><mml:mn mathvariant="normal">3.36</mml:mn><mml:mo>×</mml:mo><mml:msup><mml:mn mathvariant="normal">10</mml:mn><mml:mn mathvariant="normal">9</mml:mn></mml:msup></mml:mrow></mml:math></inline-formula> <inline-formula><mml:math id="M18" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">J</mml:mi><mml:mspace linebreak="nobreak" width="0.125em"/><mml:msup><mml:mi mathvariant="normal">kg</mml:mi><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:mrow></mml:math></inline-formula>. Other methods of converting radiative forcing to <inline-formula><mml:math id="M19" display="inline"><mml:mrow class="chem"><mml:msub><mml:mi mathvariant="normal">CO</mml:mi><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>e exist in the literature, and which method to use for different applications is a subject of active research <xref ref-type="bibr" rid="bib1.bibx25 bib1.bibx4" id="paren.28"/>. In this work, one can think of <inline-formula><mml:math id="M20" display="inline"><mml:mrow class="chem"><mml:msub><mml:mi mathvariant="normal">CO</mml:mi><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>e as an equivalent unit of energy forcing, with the unit conversion being necessary for input to the optimizer: everywhere units of <inline-formula><mml:math id="M21" display="inline"><mml:mrow class="chem"><mml:msub><mml:mi mathvariant="normal">CO</mml:mi><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>e are reported in this work they can be converted back to Joules by multiplying by the provided factor if desired.</p>

      <fig id="F1" specific-use="star"><label>Figure 1</label><caption><p id="d2e667">Eastbound flight routes considered for this trial. The displayed routes are great circle paths between each city pair and do not necessarily reflect the actual paths that were flown. A full list of city pairs included in the trial and their applicable time periods is provided in Table <xref ref-type="table" rid="TC1"/>.</p></caption>
          <graphic xlink:href="https://jecats.copernicus.org/articles/1/6/2026/jecats-1-6-2026-f01.png"/>

        </fig>

</sec>
<sec id="Ch1.S2.SS2">
  <label>2.2</label><title>Scope</title>
      <p id="d2e686">For this trial, we considered all of AAL's eastbound flights from the United States to Europe, with certain flights removed from consideration due to operational constraints. All of the flights in the trial were part of AAL's standard schedule. The trial scope was explicitly defined as a set of origin-destination city pairs. Throughout the trial, 28 city pairs were considered, with certain city pairs added partway through the trial due to seasonal changes in AAL's flight schedule. Figure <xref ref-type="fig" rid="F1"/> shows a map of all city pairs considered for the trial. The eastbound flights to Europe primarily flew during the nighttime, meaning that any contrails produced by those flights were likely net warming. Therefore, this choice of scope also maximized the number of flights with high average contrail warming according to the forecast we used.</p>
      <p id="d2e691">Every 2–3 weeks, each city pair was randomly assigned to the treatment or control group with a 50 % probability. The treatment/control structure of the trial will be discussed in detail in Sect. <xref ref-type="sec" rid="Ch1.S2.SS4"/> and <xref ref-type="sec" rid="Ch1.S2.SS7"/>. The trial began with a “ramp-up” period, where city pairs were progressively added to the trial until week 5, when the trial included all candidate routes. A full list of city pairs and their applicable time periods is provided in Table <xref ref-type="table" rid="TC1"/>.</p>
      <p id="d2e700">The trial ran for 17 weeks beginning on 15 January 2025 and ending on  13 May 2025. The start and end dates were determined before the trial began and did not change depending on how many contrail avoidance flights were flown.</p>
</sec>
<sec id="Ch1.S2.SS3">
  <label>2.3</label><title>Flightkeys Optimizer</title>
      <p id="d2e711">The Flightkeys platform performs trajectory optimization using a gridded approach, leveraging established path-finding algorithms such as Dijkstra's and A* <xref ref-type="bibr" rid="bib1.bibx10 bib1.bibx9" id="paren.29"/>. In this work, we refer to plans generated by the Flightkeys optimizer that do not consider contrails as “non-avoidance” plans.</p>
      <p id="d2e717">For this study, forecasted contrail costs were included as an additional term in the cost function of the optimizer. The optimizer uses a marginal cost index to determine how heavily to weight contrails in the cost function. For this trial, we used a marginal cost value of USD 50 t <inline-formula><mml:math id="M22" display="inline"><mml:mrow class="chem"><mml:msub><mml:mi mathvariant="normal">CO</mml:mi><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>e<sup>−1</sup>. The Flightkeys platform coarse-grains the native <inline-formula><mml:math id="M24" display="inline"><mml:mn mathvariant="normal">0.25</mml:mn></mml:math></inline-formula>° contrail forecast resolution to <inline-formula><mml:math id="M25" display="inline"><mml:mn mathvariant="normal">1</mml:mn></mml:math></inline-formula>° by taking the average value. This resolution was chosen to balance between accuracy and compute resource usage. <xref ref-type="bibr" rid="bib1.bibx10" id="text.30"/> contains a further discussion of this design choice. The platform receives hourly updates to the contrail forecast with a forecast range of 24 h. In this work, we refer to plans generated by the Flightkeys optimizer that consider contrails as “contrail-optimized” plans.</p>
</sec>
<sec id="Ch1.S2.SS4">
  <label>2.4</label><title>Flight Selection and Operational Workflow</title>
      <p id="d2e768">The Flightkeys system generates one or more non-avoidance plans for every flight in AAL's environment beginning roughly 24 h before departure time. For every flight within scope of the trial, the forecasted EF (based on the forecast described in Sect. <xref ref-type="sec" rid="Ch1.S2.SS1"/>) of the non-avoidance plans was calculated. This process was repeated multiple times before departure. If the accumulated contrail impact across the flight path was less than 10 Mg <inline-formula><mml:math id="M26" display="inline"><mml:mrow class="chem"><mml:msub><mml:mi mathvariant="normal">CO</mml:mi><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>e (equivalent to 10 tonnes <inline-formula><mml:math id="M27" display="inline"><mml:mrow class="chem"><mml:msub><mml:mi mathvariant="normal">CO</mml:mi><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>e), the flight was not considered for the trial. The goal of this filter was to only suggest contrail avoidance when the original plan was forecasted to cause substantial contrail warming. The 10 Mg threshold was chosen by examining simulated plans for a set of 95 pre-trial flights. The authors of this work manually chose an appropriate value that filtered out flights where the optimizer did not find a plan with significant contrail benefit.</p>
      <p id="d2e795">For flights in the control group that exceeded the 10 Mg threshold, operations proceeded as normal without considering contrails, meaning that dispatchers were only presented with non-avoidance plans. Neither dispatchers nor pilots had any indication that a flight was included in the control group. These flights were later used for analysis to establish a baseline against which the treatment group was compared.</p>
      <p id="d2e798">In the treatment group, flights whose non-avoidance plan exceeded the 10 Mg threshold were re-optimized including the contrails cost term. The result of this procedure is a flight plan that minimizes a combination of contrail formation, fuel consumption, and other operational costs. Since the Flightkeys system is a cost-based system, the contrail-optimized plan may not avoid all forecasted contrails, if for example the extra fuel cost incurred by performing the contrail avoidance is too large.</p>
      <p id="d2e801">In AAL's operations, dispatchers exercise joint operational control with pilots and are responsible for planning flights and working with the pilots to ensure safe operations for the duration of the flight. Once the contrail-optimized plan was generated, dispatchers were presented with a single contrail-optimized plan and one or more non-avoidance plans for consideration. Having multiple candidate flight plans available is a standard workflow for dispatchers. Additionally, flights whose non-avoidance plan passed the 10 Mg threshold were indicated by a colored border in the Flightkeys user interface (UI), providing a visual cue for dispatchers to consider contrails for that flight. Dispatcher participation in the trial was voluntary, so the dispatcher for a given flight could decide to evaluate and release the contrail-optimized plan to the pilots or proceed as usual and evaluate and release a non-avoidance plan. If the dispatcher decided to release the contrail-optimized plan, the plan would be released with a tag indicating that contrails were considered for that flight and a remark informing the pilots that contrails were considered in the creation of the flight plan. Pilots were not provided any specific information about the location of contrail areas along or near the route.</p>
</sec>
<sec id="Ch1.S2.SS5">
  <label>2.5</label><title>Validation with Satellite Imagery</title>
      <p id="d2e813">After the trial completed, we used a modified version of the “contrail to flight” attribution algorithm “CoAtSaC” (short for “Contrail Attribution Sample Consensus”) <xref ref-type="bibr" rid="bib1.bibx33" id="paren.31"/>, with modifications described in Appendix <xref ref-type="sec" rid="App1.Ch1.S2"/>, to determine whether an observable contrail was created by each flight. This automated system also calculates the number of attributed observable contrail kilometers for each flight, which we used as the primary metric for comparing the treatment and control groups. In this work, we refer to the results of this system as “observable contrail formation”.</p>
      <p id="d2e821">The system takes input images from the GOES-East geostationary satellite's ABI <xref ref-type="bibr" rid="bib1.bibx13" id="paren.32"/>, and identifies contrails in those images using a machine learning computer vision algorithm <xref ref-type="bibr" rid="bib1.bibx27" id="paren.33"/>. The next step is to obtain Automatic Dependent Surveillance-Broadcast (ADS-B) flight path data. The data was a combination of terrestrial and satellite ADS-B data licensed from FlightAware LLC and Aireon LLC respectively. We process the ADS-B data using the process described in <xref ref-type="bibr" rid="bib1.bibx33" id="text.34"/>. The system then attributes each flight to contrails by simulating the advection of the flight path using wind data from ECMWF ERA5 reanalysis <xref ref-type="bibr" rid="bib1.bibx15" id="paren.35"/>. It then compares the advected flight tracks to the detected contrails and determines that the flight matches an observed contrail if they are close enough, according to thresholds determined in <xref ref-type="bibr" rid="bib1.bibx33" id="text.36"/>, with the difference that in this work we also considered whether the contrail and flights had similar altitudes (Appendix <xref ref-type="sec" rid="App1.Ch1.S2"/>). The algorithm considers all spatiotemporally nearby flights for attribution, not only the flights included in the trial.</p>
      <p id="d2e842">The performance of the attribution system is discussed in depth in prior work <xref ref-type="bibr" rid="bib1.bibx33 bib1.bibx11" id="paren.37"/>. In this work we are measuring the end-to-end performance of our forecast system, the optimizer, and the validation system.</p>
      <p id="d2e848">In order to keep the authors of this work blinded to the results, the satellite attribution was not performed until after the trial completed.</p>
</sec>
<sec id="Ch1.S2.SS6">
  <label>2.6</label><title>Contrail Warming Estimates</title>
      <p id="d2e859">Contrail warming cannot be observed directly, so we consider warming results based on CoCiP simulations to evaluate the efficacy of the trial in reducing contrail warming. A significant weakness in using such simulations to estimate the warming of individual flights is the difficulty in accurately predicting specific humidity <xref ref-type="bibr" rid="bib1.bibx12 bib1.bibx1" id="paren.38"/>. In this work this difficulty is reduced, because we have actually observed the flights making contrails; therefore, the observations give us information about the humidity. When estimating forcing, we assign zero forcing to flights which are not observed making contrails, and a nonzero forcing to those observed making contrails. We use two methods for estimating the energy forcing for the flights which make contrails based on observations.</p>
      <p id="d2e865">The “averaged forcing” method assumes that the avoided contrails are “average” contrails, with the properties of the average contrail based only on time of day, season and latitude, as given in <xref ref-type="bibr" rid="bib1.bibx30" id="text.39"/>. Roughly speaking, this leads to contrails at nighttime being assigned a warming of <inline-formula><mml:math id="M28" display="inline"><mml:mrow><mml:mo>∼</mml:mo><mml:mn mathvariant="normal">500</mml:mn></mml:mrow></mml:math></inline-formula> <inline-formula><mml:math id="M29" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">MJ</mml:mi><mml:mspace linebreak="nobreak" width="0.125em"/><mml:msup><mml:mi mathvariant="normal">m</mml:mi><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:mrow></mml:math></inline-formula> while daytime contrails are assigned around <inline-formula><mml:math id="M30" display="inline"><mml:mn mathvariant="normal">0</mml:mn></mml:math></inline-formula> <inline-formula><mml:math id="M31" display="inline"><mml:mrow class="unit"><mml:mi mathvariant="normal">MJ</mml:mi><mml:mspace linebreak="nobreak" width="0.125em"/><mml:msup><mml:mi mathvariant="normal">m</mml:mi><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1</mml:mn></mml:mrow></mml:msup></mml:mrow></mml:math></inline-formula>. In this work, we refer to this forcing estimate masked by observations as “observation-informed average forcing”.</p>
      <p id="d2e922">Another method is to use the forecast weather. We use the IFS ensemble forecast, and we leverage the fact that we observed a contrail by averaging the CoCiP EF of only the ensembles which predict a persistent contrail. In the (rare) case where we observe a contrail but no ensemble member predicts a persistent contrail, we still want to assign a nonzero warming value. Therefore, we also include in the average the averaged forcing estimate, to ensure that a nonzero forcing is assigned to all observed contrails. This follows a similar procedure to <xref ref-type="bibr" rid="bib1.bibx30" id="text.40"/> for estimating contrail impact assuming contrail formation. In this work, we refer to this warming estimate masked by observations as “observation-informed ensemble forcing”.</p>
</sec>
<sec id="Ch1.S2.SS7">
  <label>2.7</label><title>Randomized Controlled Trial</title>
      <p id="d2e936">From the treatment group determined by random city pair assignment and 10 Mg warming threshold above, we considered three subgroups for analysis. These subgroups were defined before any members of the team were unblinded to the satellite results. <list list-type="bullet"><list-item>
      <p id="d2e941"><italic>Level 1 (L1)</italic>. Entire treatment group</p></list-item><list-item>
      <p id="d2e947"><italic>Level 2 (L2)</italic>. Flights that were released as contrail-optimized plans (subset of L1)</p></list-item><list-item>
      <p id="d2e953"><italic>Level 3 (L3)</italic>. Flights that were released as contrail-optimized plans and flown as planned (subset of L2)</p></list-item></list> The Level 1 group represents the full intent-to-treat group. The Level 2 and 3 groups are per-protocol groups, meaning flights which adhered to the intervention we proposed, with differently strict definitions of protocol adherence <xref ref-type="bibr" rid="bib1.bibx24" id="paren.41"/>. The Level 2 group is the subset of the L1 group where the dispatcher released the contrail-optimized plan to the flight crew, meaning that the dispatcher voluntarily selected the contrail-optimized plan. Because operational circumstances sometimes result in the dispatcher releasing revised flight plans, in order to determine which plan was released to the flight crew we considered the last released plan before takeoff. The ratio of L2 to L1 flights (hereafter called the L2 <inline-formula><mml:math id="M32" display="inline"><mml:mo>/</mml:mo></mml:math></inline-formula> L1 take rate, where “take rate” indicates the voluntary dispatcher engagement with contrail avoidance) was <inline-formula><mml:math id="M33" display="inline"><mml:mrow><mml:mn mathvariant="normal">15.4</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula>. The Level 3 group is the subset of the L2 group where the flight adhered to the released flight plan. We considered a flight to have adhered to the contrail-optimized plan if the difference between the planned warming per flight km and the flown warming per flight km was less than 0.001 Mg <inline-formula><mml:math id="M34" display="inline"><mml:mrow class="chem"><mml:msub><mml:mi mathvariant="normal">CO</mml:mi><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>e km<sup>−1</sup>, where the warming was computed using the same forecast as described in Sect. <xref ref-type="sec" rid="Ch1.S2.SS1"/>. This threshold roughly corresponds to a 5 Mg <inline-formula><mml:math id="M36" display="inline"><mml:mrow class="chem"><mml:msub><mml:mi mathvariant="normal">CO</mml:mi><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula>e warming threshold for a 5000 km flight, which is the approximate length of many of the flights in the trial. This heuristic was determined before unblinding to the satellite results and was found to compare well with subjective human assessments of adherence performed by the authors of this work. These subjective assessments were performed by visualizing flight plans interpolated to the contrail forecast in the vertical, lateral, and time dimensions, and judging whether the flown ADS-B path matched the planned path. Examples of these visualizations are provided in Appendix <xref ref-type="sec" rid="App1.Ch1.S4"/>. This heuristic has the advantage of disregarding any non-adherence that was not near a contrail region. For example, if the flight climbed to its cruising altitude later than planned, but there were no contrail regions nearby, it could still be included in the L3 group assuming it performed the planned contrail avoidance maneuvers. The ratio of L3 to L1 flights (L3 <inline-formula><mml:math id="M37" display="inline"><mml:mo>/</mml:mo></mml:math></inline-formula> L1 take rate) was <inline-formula><mml:math id="M38" display="inline"><mml:mrow><mml:mn mathvariant="normal">7.8</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula>.</p>
      <p id="d2e1037">The reduction in contrail formation in the L1 group represents the average contrail reduction achieved over all the flights in the trial. However, it is strongly influenced by the relatively low take rate, which is in turn influenced by the design of this specific trial and may not be broadly applicable to other contrail avoidance efforts. The L2 group represents the contrail reduction achieved when dispatchers select and release a contrail avoidance flight plan. The L3 group represents the contrail reduction achieved when contrail avoidance routes were flown.</p>
      <p id="d2e1040">Each of these three treatment groups were evaluated against the control group using the observed per-flight contrail kilometers (normalized by flight distance) as the primary trial endpoint.</p>
</sec>
<sec id="Ch1.S2.SS8">
  <label>2.8</label><title>Statistical Analysis</title>
      <p id="d2e1051">To compare contrail formation in the treatment and control groups, we computed the ratio of total observed contrail kilometers (computed as the length of the flight that is attributed to the observed contrails) normalized by total flight kilometers, referred to hereafter as the Observed Contrail Rate (<inline-formula><mml:math id="M39" display="inline"><mml:mrow><mml:msubsup><mml:mi mathvariant="italic">λ</mml:mi><mml:mi>C</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mi>k</mml:mi><mml:mo>)</mml:mo></mml:mrow></mml:msubsup></mml:mrow></mml:math></inline-formula>):

            <disp-formula id="Ch1.E1" content-type="numbered"><label>1</label><mml:math id="M40" display="block"><mml:mrow><mml:msubsup><mml:mi mathvariant="italic">λ</mml:mi><mml:mi>C</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mi>k</mml:mi><mml:mo>)</mml:mo></mml:mrow></mml:msubsup><mml:mo>=</mml:mo><mml:mstyle displaystyle="true"><mml:mfrac style="display"><mml:mrow><mml:msub><mml:mo>∑</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>∈</mml:mo><mml:mi>k</mml:mi></mml:mrow></mml:msub><mml:msub><mml:mi>C</mml:mi><mml:mi>i</mml:mi></mml:msub></mml:mrow><mml:mrow><mml:msub><mml:mo>∑</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>∈</mml:mo><mml:mi>k</mml:mi></mml:mrow></mml:msub><mml:msub><mml:mi>L</mml:mi><mml:mi>i</mml:mi></mml:msub></mml:mrow></mml:mfrac></mml:mstyle><mml:mo>,</mml:mo></mml:mrow></mml:math></disp-formula>

          where <inline-formula><mml:math id="M41" display="inline"><mml:mi>i</mml:mi></mml:math></inline-formula> are the flights in either group <inline-formula><mml:math id="M42" display="inline"><mml:mi>k</mml:mi></mml:math></inline-formula> (and <inline-formula><mml:math id="M43" display="inline"><mml:mi>k</mml:mi></mml:math></inline-formula> is one of treatment or control), <inline-formula><mml:math id="M44" display="inline"><mml:mrow><mml:msub><mml:mi>C</mml:mi><mml:mi>i</mml:mi></mml:msub></mml:mrow></mml:math></inline-formula> represents the contrail distance in kilometers, and <inline-formula><mml:math id="M45" display="inline"><mml:mrow><mml:msub><mml:mi>L</mml:mi><mml:mi>i</mml:mi></mml:msub></mml:mrow></mml:math></inline-formula> represents the flight distance for flight <inline-formula><mml:math id="M46" display="inline"><mml:mi>i</mml:mi></mml:math></inline-formula> in kilometers. When computing <inline-formula><mml:math id="M47" display="inline"><mml:mrow><mml:msub><mml:mi>L</mml:mi><mml:mi>i</mml:mi></mml:msub></mml:mrow></mml:math></inline-formula>, we used only the flight distance west of 30° W longitude where the contrail forecast was available and where contrails could be observed by the GOES-East satellite.</p>
      <p id="d2e1190">To compare the fuel consumption between the treatment and control groups, we computed the Fuel Rate similarly to the Observed Contrail Rate (Eq. <xref ref-type="disp-formula" rid="Ch1.E1"/>):

            <disp-formula id="Ch1.E2" content-type="numbered"><label>2</label><mml:math id="M48" display="block"><mml:mrow><mml:msubsup><mml:mi mathvariant="italic">λ</mml:mi><mml:mi>F</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mi>k</mml:mi><mml:mo>)</mml:mo></mml:mrow></mml:msubsup><mml:mo>=</mml:mo><mml:mstyle displaystyle="true"><mml:mfrac style="display"><mml:mrow><mml:msub><mml:mo>∑</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>∈</mml:mo><mml:mi>k</mml:mi></mml:mrow></mml:msub><mml:msub><mml:mi>F</mml:mi><mml:mi>i</mml:mi></mml:msub></mml:mrow><mml:mrow><mml:msub><mml:mo>∑</mml:mo><mml:mrow><mml:mi>i</mml:mi><mml:mo>∈</mml:mo><mml:mi>k</mml:mi></mml:mrow></mml:msub><mml:msub><mml:mi>L</mml:mi><mml:mi>i</mml:mi></mml:msub></mml:mrow></mml:mfrac></mml:mstyle><mml:mo>,</mml:mo></mml:mrow></mml:math></disp-formula>

          where <inline-formula><mml:math id="M49" display="inline"><mml:mrow><mml:msub><mml:mi>F</mml:mi><mml:mi>i</mml:mi></mml:msub></mml:mrow></mml:math></inline-formula> represents the fuel usage per flight and <inline-formula><mml:math id="M50" display="inline"><mml:mrow><mml:msub><mml:mi>L</mml:mi><mml:mi>i</mml:mi></mml:msub></mml:mrow></mml:math></inline-formula> the total length of the flight (not restricted to west of 30° W). Fuel usage data is actual data provided by AAL. The quantity used for this analysis is the actual fuel used while the aircraft is in the air, derived by subtracting the fuel weight when the aircraft touches down from the fuel weight when the aircraft leaves the ground. This does not include the fuel used during the taxi phases before takeoff and after landing.</p>
      <p id="d2e1273">Due to operational constraints, city pairs were only randomized every 2–3 weeks. This caused an imbalance of aircraft type between treatment and control groups. For example, <inline-formula><mml:math id="M51" display="inline"><mml:mrow><mml:mn mathvariant="normal">74</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula> of the control group flights were Boeing 777s, whereas only <inline-formula><mml:math id="M52" display="inline"><mml:mrow><mml:mn mathvariant="normal">50</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula> of the treatment group flights were 777s. See Fig. <xref ref-type="fig" rid="FE1"/> for the aircraft type distribution across groups.</p>
      <p id="d2e1300">The physical properties of a contrail are dependent on the combination of aircraft and engine type involved in contrail formation <xref ref-type="bibr" rid="bib1.bibx20 bib1.bibx40" id="paren.42"/>. However, the effect of these differences on geostationary-observable contrail formation is an area of active research <xref ref-type="bibr" rid="bib1.bibx7 bib1.bibx14" id="paren.43"/>. Additionally, fuel consumption differs by aircraft type. Therefore, we treated aircraft type as a confounder for both contrail distance and fuel and adjusted the statistical tests accordingly. We report both marginal and adjusted results. To adjust the test statistic by aircraft type, we first computed the Observed Contrail and Fuel Rates per aircraft type. We then computed the weighted mean for treatment and control groups using the same weights, which are derived from the aircraft prevalence in the total study population. We computed the difference in both Observed Contrail and Fuel Rates: <inline-formula><mml:math id="M53" display="inline"><mml:mrow><mml:msup><mml:mi mathvariant="italic">λ</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mi mathvariant="normal">control</mml:mi><mml:mo>)</mml:mo></mml:mrow></mml:msup><mml:mo>-</mml:mo><mml:msup><mml:mi mathvariant="italic">λ</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mi mathvariant="normal">treatment</mml:mi><mml:mo>)</mml:mo></mml:mrow></mml:msup></mml:mrow></mml:math></inline-formula> as our test statistics. Statistical significance was assessed using a non-parametric stratified permutation test with 10,000 permutations <xref ref-type="bibr" rid="bib1.bibx28 bib1.bibx18" id="paren.44"/>. To generate the null distribution, treatment labels were randomly shuffled among flights strictly within their respective aircraft type, preserving the aircraft-type confounding structure. A <inline-formula><mml:math id="M54" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula> value was computed as the proportion of permuted statistics exceeding the observed reduction. For observed contrail formation, we used a one-sided <inline-formula><mml:math id="M55" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula> value, and for fuel consumption we used a two-sided <inline-formula><mml:math id="M56" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula> value.</p>
      <p id="d2e1363">To estimate the confidence interval, we performed a stratified bootstrap analysis with 10 000 bootstrap samples. We created new sample datasets by randomly drawing flights with replacement from the original treatment and control groups, doing so strictly within their respective aircraft strata to maintain the study's balance. For each iteration, we re-applied the aircraft prevalence weights and computed a new test statistic. The 95 % confidence interval was derived from the 2.5 and 97.5 percentiles these 10 000 simulated statistics.</p>
</sec>
<sec id="Ch1.S2.SS9">
  <label>2.9</label><title>Counterfactual Analysis</title>
      <p id="d2e1374">The L3 observable results in this work are the best estimate of how much reduction in contrail formation can be achieved with the methods used in this work when contrail-optimized routes are flown. The methods above compare the L2 and L3 groups to the control group. One drawback of the L2 and L3 analysis is that there might be some correlation between contrail formation and what causes a flight to be selected for the L2 and L3 groups. For example, in interviews conducted after the trial completed, dispatchers mentioned that the take rate depended on how busy they were. One factor that might make the dispatchers busy is more demanding weather conditions such as storms or turbulence, which could also be correlated with contrail formation. It may also be more difficult for the Flightkeys optimizer to offer a flyable contrail-optimized plan when a flight has to avoid large contrail-forming regions. In order to address these potential issues, we additionally performed a counterfactual analysis to confirm that the reduction in observed contrail km between the L2 and L3 groups is due to contrail avoidance. If the reduction factors from this counterfactual analysis were significantly smaller than the reduction factors in the observed results, we would have suspected that some amount of the observed reduction resulted from factors other than our intervention.</p>
      <p id="d2e1377">At many points in the flight planning process, the Flightkeys system generated both contrail-optimized and non-avoidance plans for a given flight. For flights where the contrail-optimized plan was released, we performed the additional counterfactual analysis to isolate the effect of our intervention without any additional confounders. To do this, we chose one of the non-avoidance plans as a counterfactual for the actual plan. In order to determine which non-avoidance plan to use as a counterfactual for the flight, we examined multiple choices of non-avoidance plan and compared them against the flown path for flights that did not perform contrail avoidance (specifically, flights from the treatment group that were not part of the L2 set, “L1 not L2”). The choices we evaluated were the minimum cost plan and the most recent plan. To find the “minimum cost” plan, we examined all non-avoidance plans generated before the flight's departure and chose the plan with the minimum total cost as calculated by Flightkeys. The total cost is a combination of fuel and other costs such as delays and overflight charges <xref ref-type="bibr" rid="bib1.bibx9" id="paren.45"/>. The results of that analysis are included in Sect. <xref ref-type="sec" rid="Ch1.S3.SS4"/>.</p>
      <p id="d2e1385">We used the choice of non-avoidance plan from that analysis to compute the differences in expected contrail distance between the counterfactual plan and the flown path, where expected contrail distance was computed using the ML formation probability forecast described in Sect. <xref ref-type="sec" rid="Ch1.S2.SS1"/>. Since the expected contrail distance is a forecast product, we also validated that the forecast agrees with observations to justify its use in this analysis.</p>
      <p id="d2e1390">We additionally performed a counterfactual analysis for the fuel usage. Lacking simulated fuel data for the flown path, we compared the simulated fuel value for the last-generated contrail-optimized plan to the non-avoidance plan computed nearest to it in time, whether preceding or following.</p>

<table-wrap id="T1" specific-use="star"><label>Table 1</label><caption><p id="d2e1397">Observable contrail formation results for the control and treatment groups. Reduction in contrail formation rate is calculated with respect to the contrail formation rate in the control group. The “Adjusted” rate column incorporates the aircraft type adjustment described in Sect. <xref ref-type="sec" rid="Ch1.S2.SS8"/>, whereas the “Unadjusted” rate column does not. The values in brackets represent the 95 % confidence interval of the column quantity. Bold font in this table refers to the most critical results.</p></caption><oasis:table frame="topbot"><oasis:tgroup cols="6">
     <oasis:colspec colnum="1" colname="col1" align="left"/>
     <oasis:colspec colnum="2" colname="col2" align="right"/>
     <oasis:colspec colnum="3" colname="col3" align="justify" colwidth="3cm"/>
     <oasis:colspec colnum="4" colname="col4" align="justify" colwidth="3cm"/>
     <oasis:colspec colnum="5" colname="col5" align="justify" colwidth="3cm"/>
     <oasis:colspec colnum="6" colname="col6" align="right"/>
     <oasis:thead>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1">Group</oasis:entry>
         <oasis:entry colname="col2"><inline-formula><mml:math id="M57" display="inline"><mml:mi>N</mml:mi></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3" align="left">Contrail Formation Rate (Unadjusted)</oasis:entry>
         <oasis:entry colname="col4" align="left">Reduction in Rate (Unadjusted)</oasis:entry>
         <oasis:entry colname="col5" align="left">Reduction in Rate (Adjusted)</oasis:entry>
         <oasis:entry colname="col6"><inline-formula><mml:math id="M58" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula> value (Adjusted)</oasis:entry>
       </oasis:row>
     </oasis:thead>
     <oasis:tbody>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1">Control</oasis:entry>
         <oasis:entry colname="col2"><inline-formula><mml:math id="M59" display="inline"><mml:mn mathvariant="normal">1172</mml:mn></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3" align="left"><inline-formula><mml:math id="M60" display="inline"><mml:mrow><mml:mn mathvariant="normal">5.9</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula> <inline-formula><mml:math id="M61" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">5.5</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">6.4</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col4" align="left">–</oasis:entry>
         <oasis:entry colname="col5" align="left">–</oasis:entry>
         <oasis:entry colname="col6">–</oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1">Treatment (L1)</oasis:entry>
         <oasis:entry colname="col2"><inline-formula><mml:math id="M62" display="inline"><mml:mn mathvariant="normal">1232</mml:mn></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3" align="left"><inline-formula><mml:math id="M63" display="inline"><mml:mrow><mml:mn mathvariant="normal">5.3</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula> <inline-formula><mml:math id="M64" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">4.9</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">5.6</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col4" align="left"><inline-formula><mml:math id="M65" display="inline"><mml:mrow><mml:mn mathvariant="bold">11.6</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula> <inline-formula><mml:math id="M66" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">2.1</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">20.1</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col5" align="left"><inline-formula><mml:math id="M67" display="inline"><mml:mrow><mml:mn mathvariant="bold">11.6</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula> <inline-formula><mml:math id="M68" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">2.1</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">20.3</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col6"><inline-formula><mml:math id="M69" display="inline"><mml:mn mathvariant="bold">0.011</mml:mn></mml:math></inline-formula></oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1">Treatment (L2)</oasis:entry>
         <oasis:entry colname="col2"><inline-formula><mml:math id="M70" display="inline"><mml:mn mathvariant="normal">190</mml:mn></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3" align="left"><inline-formula><mml:math id="M71" display="inline"><mml:mrow><mml:mn mathvariant="normal">3.8</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula> <inline-formula><mml:math id="M72" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">3.1</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">4.5</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col4" align="left"><inline-formula><mml:math id="M73" display="inline"><mml:mrow><mml:mn mathvariant="bold">36.4</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula> <inline-formula><mml:math id="M74" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">22.4</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">49.0</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col5" align="left"><inline-formula><mml:math id="M75" display="inline"><mml:mrow><mml:mn mathvariant="bold">29.5</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula> <inline-formula><mml:math id="M76" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">9.7</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">46.0</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col6"><inline-formula><mml:math id="M77" display="inline"><mml:mn mathvariant="bold">0.002</mml:mn></mml:math></inline-formula></oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">Treatment (L3)</oasis:entry>
         <oasis:entry colname="col2"><inline-formula><mml:math id="M78" display="inline"><mml:mn mathvariant="normal">112</mml:mn></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3" align="left"><inline-formula><mml:math id="M79" display="inline"><mml:mrow><mml:mn mathvariant="normal">2.1</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula> <inline-formula><mml:math id="M80" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">1.4</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">2.8</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col4" align="left"><inline-formula><mml:math id="M81" display="inline"><mml:mrow><mml:mn mathvariant="bold">64.5</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula>  <inline-formula><mml:math id="M82" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">52.0</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">75.7</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col5" align="left"><inline-formula><mml:math id="M83" display="inline"><mml:mrow><mml:mn mathvariant="bold">62.0</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula> <inline-formula><mml:math id="M84" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">47.1</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">75.3</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col6"><inline-formula><mml:math id="M85" display="inline"><mml:mrow><mml:mo mathvariant="bold">&lt;</mml:mo><mml:mn mathvariant="bold">0.001</mml:mn></mml:mrow></mml:math></inline-formula></oasis:entry>
       </oasis:row>
     </oasis:tbody>
   </oasis:tgroup></oasis:table></table-wrap>

</sec>
</sec>
<sec id="Ch1.S3">
  <label>3</label><title>Results</title>
<sec id="Ch1.S3.SS1">
  <label>3.1</label><title>Observable Contrails</title>
      <p id="d2e1920">Table <xref ref-type="table" rid="T1"/> shows the observed contrail formation rate for all the groups in our experiment. All subgroups of the treatment group showed a statistically significant reduction in observable contrail formation when compared against the control group. We rejected the null hypothesis of no change in observable contrail formation between the L1 treatment and control groups (one-sided <inline-formula><mml:math id="M86" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:mo>=</mml:mo><mml:mn mathvariant="normal">0.011</mml:mn></mml:mrow></mml:math></inline-formula>, stratified permutation test). We also rejected the null hypothesis of no change in observable contrail formation between the L2 treatment and control groups (one-sided <inline-formula><mml:math id="M87" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:mo>=</mml:mo><mml:mn mathvariant="normal">0.002</mml:mn></mml:mrow></mml:math></inline-formula>, stratified permutation test) and the L3 treatment and control groups (one-sided <inline-formula><mml:math id="M88" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:mo>&lt;</mml:mo><mml:mn mathvariant="normal">0.001</mml:mn></mml:mrow></mml:math></inline-formula>, stratified permutation test). We see that the results are not substantially changed by the adjustment for aircraft type. Additional observed contrail values are included in Appendix <xref ref-type="sec" rid="App1.Ch1.S6"/>.</p>
</sec>
<sec id="Ch1.S3.SS2">
  <label>3.2</label><title>Contrail Warming Estimates</title>
      <p id="d2e1971">To estimate the warming reduction from our trial, we used the two methods described in Sect. <xref ref-type="sec" rid="Ch1.S2.SS6"/>. The results of the observation-informed average forcing analysis are shown in Table <xref ref-type="table" rid="T2"/>. We found significant reductions in all treatment subgroups. We rejected the null hypothesis of no change in observation-informed average forcing between the L1 treatment and control groups (one-sided <inline-formula><mml:math id="M89" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:mo>=</mml:mo><mml:mn mathvariant="normal">0.006</mml:mn></mml:mrow></mml:math></inline-formula>, stratified permutation test). We also rejected the null hypothesis of no change in observation-informed average forcing between the L2 treatment and control groups (one-sided <inline-formula><mml:math id="M90" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:mo>&lt;</mml:mo><mml:mn mathvariant="normal">0.001</mml:mn></mml:mrow></mml:math></inline-formula>, stratified permutation test) L3 treatment and control groups (one-sided <inline-formula><mml:math id="M91" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:mo>&lt;</mml:mo><mml:mn mathvariant="normal">0.001</mml:mn></mml:mrow></mml:math></inline-formula>, stratified permutation test).</p>
      <p id="d2e2014">The results of the observation-informed ensemble forcing analysis are shown in Table <xref ref-type="table" rid="T3"/>. We rejected the null hypothesis of no change in observation-informed ensemble forcing between the L1 treatment and control groups (one-sided <inline-formula><mml:math id="M92" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:mo>=</mml:mo><mml:mn mathvariant="normal">0.045</mml:mn></mml:mrow></mml:math></inline-formula>, stratified permutation test). However, the confidence interval is relatively wide and crosses 0, indicating uncertainty in the warming reduction in the L1 group. We also rejected the null hypothesis of no change in observation-informed ensemble forcing between the L2 treatment and control groups (one-sided <inline-formula><mml:math id="M93" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:mo>&lt;</mml:mo><mml:mn mathvariant="normal">0.001</mml:mn></mml:mrow></mml:math></inline-formula>, stratified permutation test) L3 treatment and control groups (one-sided <inline-formula><mml:math id="M94" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:mo>&lt;</mml:mo><mml:mn mathvariant="normal">0.001</mml:mn></mml:mrow></mml:math></inline-formula>, stratified permutation test).</p>

<table-wrap id="T2" specific-use="star"><label>Table 2</label><caption><p id="d2e2058">Observation-informed average forcing estimates for the control and 3 treatment group levels. Reduction in observation-informed average forcing is calculated with respect to the observation-informed average forcing in the control group. The values in brackets represent the 95 % confidence interval of the column quantity. Bold font in this table refers to the most critical results.</p></caption><oasis:table frame="topbot"><oasis:tgroup cols="4">
     <oasis:colspec colnum="1" colname="col1" align="left"/>
     <oasis:colspec colnum="2" colname="col2" align="justify" colwidth="3cm"/>
     <oasis:colspec colnum="3" colname="col3" align="justify" colwidth="3cm"/>
     <oasis:colspec colnum="4" colname="col4" align="justify" colwidth="3cm"/>
     <oasis:thead>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1">Group</oasis:entry>
         <oasis:entry colname="col2" align="right">Observation-informed Average Forcing (J m<sup>−1</sup>, Unadjusted)</oasis:entry>
         <oasis:entry colname="col3" align="left">Reduction in Observation-informed Average Forcing (Unadjusted)</oasis:entry>
         <oasis:entry colname="col4" align="left">Reduction in Observation-informed Average Forcing (Adjusted)</oasis:entry>
       </oasis:row>
     </oasis:thead>
     <oasis:tbody>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1">Control</oasis:entry>
         <oasis:entry colname="col2" align="right"><inline-formula><mml:math id="M96" display="inline"><mml:mrow><mml:mn mathvariant="normal">2.19</mml:mn><mml:mo>×</mml:mo><mml:msup><mml:mn mathvariant="normal">10</mml:mn><mml:mn mathvariant="normal">7</mml:mn></mml:msup></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3" align="left">–</oasis:entry>
         <oasis:entry colname="col4" align="left">–</oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1">Treatment (L1)</oasis:entry>
         <oasis:entry colname="col2" align="right"><inline-formula><mml:math id="M97" display="inline"><mml:mrow><mml:mn mathvariant="normal">1.89</mml:mn><mml:mo>×</mml:mo><mml:msup><mml:mn mathvariant="normal">10</mml:mn><mml:mn mathvariant="normal">7</mml:mn></mml:msup></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3" align="left"><inline-formula><mml:math id="M98" display="inline"><mml:mrow><mml:mn mathvariant="bold">13.8</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula>  <inline-formula><mml:math id="M99" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">2.6</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">23.5</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col4" align="left"><inline-formula><mml:math id="M100" display="inline"><mml:mrow><mml:mn mathvariant="bold">13.7</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula>  <inline-formula><mml:math id="M101" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">3.1</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">23.4</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1">Treatment (L2)</oasis:entry>
         <oasis:entry colname="col2" align="right"><inline-formula><mml:math id="M102" display="inline"><mml:mrow><mml:mn mathvariant="normal">1.38</mml:mn><mml:mo>×</mml:mo><mml:msup><mml:mn mathvariant="normal">10</mml:mn><mml:mn mathvariant="normal">7</mml:mn></mml:msup></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3" align="left"><inline-formula><mml:math id="M103" display="inline"><mml:mrow><mml:mn mathvariant="bold">37.1</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula>  <inline-formula><mml:math id="M104" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">21.7</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">50.4</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col4" align="left"><inline-formula><mml:math id="M105" display="inline"><mml:mrow><mml:mn mathvariant="bold">35.5</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula> <inline-formula><mml:math id="M106" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">19.1</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">50.0</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">Treatment (L3)</oasis:entry>
         <oasis:entry colname="col2" align="right"><inline-formula><mml:math id="M107" display="inline"><mml:mrow><mml:mn mathvariant="normal">6.67</mml:mn><mml:mo>×</mml:mo><mml:msup><mml:mn mathvariant="normal">10</mml:mn><mml:mn mathvariant="normal">6</mml:mn></mml:msup></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3" align="left"><inline-formula><mml:math id="M108" display="inline"><mml:mrow><mml:mn mathvariant="bold">69.6</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula>  <inline-formula><mml:math id="M109" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">56.5</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">79.9</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col4" align="left"><inline-formula><mml:math id="M110" display="inline"><mml:mrow><mml:mn mathvariant="bold">69.3</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula>  <inline-formula><mml:math id="M111" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">56.1</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">80.9</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
       </oasis:row>
     </oasis:tbody>
   </oasis:tgroup></oasis:table></table-wrap>

<table-wrap id="T3" specific-use="star"><label>Table 3</label><caption><p id="d2e2421">Observation-informed ensemble forcing estimates for the control and 3 treatment group levels. Reduction in observation-informed ensemble forcing is calculated with respect to the observation-informed ensemble forcing in the control group. The values in brackets represent the 95 % confidence interval of the column quantity. Bold font in this table refers to the most critical results.</p></caption><oasis:table frame="topbot"><oasis:tgroup cols="4">
     <oasis:colspec colnum="1" colname="col1" align="left"/>
     <oasis:colspec colnum="2" colname="col2" align="justify" colwidth="3cm"/>
     <oasis:colspec colnum="3" colname="col3" align="justify" colwidth="3cm"/>
     <oasis:colspec colnum="4" colname="col4" align="justify" colwidth="3cm"/>
     <oasis:thead>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1">Group</oasis:entry>
         <oasis:entry colname="col2" align="right">Observation-informed Ensemble Forcing (J m<sup>−1</sup>, Unadjusted)</oasis:entry>
         <oasis:entry colname="col3" align="left">Reduction in Observation-informed Ensemble Forcing (Unadjusted)</oasis:entry>
         <oasis:entry colname="col4" align="left">Reduction in Observation-informed  Ensemble Forcing (Adjusted)</oasis:entry>
       </oasis:row>
     </oasis:thead>
     <oasis:tbody>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1">Control</oasis:entry>
         <oasis:entry colname="col2" align="right"><inline-formula><mml:math id="M113" display="inline"><mml:mrow><mml:mn mathvariant="normal">4.17</mml:mn><mml:mo>×</mml:mo><mml:msup><mml:mn mathvariant="normal">10</mml:mn><mml:mn mathvariant="normal">7</mml:mn></mml:msup></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3" align="left">–</oasis:entry>
         <oasis:entry colname="col4" align="left">–</oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1">Treatment (L1)</oasis:entry>
         <oasis:entry colname="col2" align="right"><inline-formula><mml:math id="M114" display="inline"><mml:mrow><mml:mn mathvariant="normal">3.55</mml:mn><mml:mo>×</mml:mo><mml:msup><mml:mn mathvariant="normal">10</mml:mn><mml:mn mathvariant="normal">7</mml:mn></mml:msup></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3" align="left"><inline-formula><mml:math id="M115" display="inline"><mml:mrow><mml:mn mathvariant="bold">14.3</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula>  <inline-formula><mml:math id="M116" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mo>-</mml:mo><mml:mn mathvariant="normal">4.5</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">29.9</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col4" align="left"><inline-formula><mml:math id="M117" display="inline"><mml:mrow><mml:mn mathvariant="bold">16.4</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula>  <inline-formula><mml:math id="M118" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mo>-</mml:mo><mml:mn mathvariant="normal">3.4</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">32.4</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1">Treatment (L2)</oasis:entry>
         <oasis:entry colname="col2" align="right"><inline-formula><mml:math id="M119" display="inline"><mml:mrow><mml:mn mathvariant="normal">2.18</mml:mn><mml:mo>×</mml:mo><mml:msup><mml:mn mathvariant="normal">10</mml:mn><mml:mn mathvariant="normal">7</mml:mn></mml:msup></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3" align="left"><inline-formula><mml:math id="M120" display="inline"><mml:mrow><mml:mn mathvariant="bold">47.4</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula>  <inline-formula><mml:math id="M121" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">28.7</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">62.3</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col4" align="left"><inline-formula><mml:math id="M122" display="inline"><mml:mrow><mml:mn mathvariant="bold">46.3</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula>  <inline-formula><mml:math id="M123" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">25.3</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">63.0</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">Treatment (L3)</oasis:entry>
         <oasis:entry colname="col2" align="right"><inline-formula><mml:math id="M124" display="inline"><mml:mrow><mml:mn mathvariant="normal">7.31</mml:mn><mml:mo>×</mml:mo><mml:msup><mml:mn mathvariant="normal">10</mml:mn><mml:mn mathvariant="normal">6</mml:mn></mml:msup></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3" align="left"><inline-formula><mml:math id="M125" display="inline"><mml:mrow><mml:mn mathvariant="bold">82.3</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula>  <inline-formula><mml:math id="M126" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">75.1</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">89.1</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col4" align="left"><inline-formula><mml:math id="M127" display="inline"><mml:mrow><mml:mn mathvariant="bold">83.0</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula>  <inline-formula><mml:math id="M128" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">74.8</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">89.4</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
       </oasis:row>
     </oasis:tbody>
   </oasis:tgroup></oasis:table></table-wrap>

</sec>
<sec id="Ch1.S3.SS3">
  <label>3.3</label><title>Fuel Usage</title>
      <p id="d2e2793">Contrail avoidance may incur additional costs, both monetary and climatological, primarily in fuel usage <xref ref-type="bibr" rid="bib1.bibx39 bib1.bibx9 bib1.bibx6" id="paren.46"/>. This extra fuel is critical to quantify due to the climate impact from burning extra fuel. As previously noted, the Flightkeys optimizer used for generating contrail routes considers multiple costs, including fuel, contrails (with the specific price assignment for this trial), and others such as delays and overflight charges <xref ref-type="bibr" rid="bib1.bibx9" id="paren.47"/>.</p>

<table-wrap id="T4" specific-use="star"><label>Table 4</label><caption><p id="d2e2805">Fuel usage for the three treatment group levels. Differences in fuel usage are calculated with respect to the fuel usage in the control group. The “Adjusted” usage column incorporates the aircraft type adjustment described in Sect. <xref ref-type="sec" rid="Ch1.S2.SS8"/>, whereas the “Unadjusted” usage column does not. Negative differences indicate that the treatment group used less fuel than the control group, and positive differences indicate that the treatment group used more fuel. Bold font in this table refers to the most critical results.</p></caption><oasis:table frame="topbot"><oasis:tgroup cols="5">
     <oasis:colspec colnum="1" colname="col1" align="left"/>
     <oasis:colspec colnum="2" colname="col2" align="right"/>
     <oasis:colspec colnum="3" colname="col3" align="justify" colwidth="3cm"/>
     <oasis:colspec colnum="4" colname="col4" align="right"/>
     <oasis:colspec colnum="5" colname="col5" align="justify" colwidth="3cm"/>
     <oasis:thead>
       <oasis:row>
         <oasis:entry colname="col1">Group</oasis:entry>
         <oasis:entry colname="col2"><inline-formula><mml:math id="M129" display="inline"><mml:mi>N</mml:mi></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3" align="left">Difference in Fuel</oasis:entry>
         <oasis:entry colname="col4"><inline-formula><mml:math id="M130" display="inline"><mml:mi>p</mml:mi></mml:math></inline-formula> value</oasis:entry>
         <oasis:entry colname="col5" align="left">Difference in Fuel</oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1"/>
         <oasis:entry colname="col2"/>
         <oasis:entry colname="col3" align="left">Usage (Adjusted)</oasis:entry>
         <oasis:entry colname="col4">(Adjusted)</oasis:entry>
         <oasis:entry colname="col5" align="left">Usage (Unadjusted)</oasis:entry>
       </oasis:row>
     </oasis:thead>
     <oasis:tbody>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1">Control</oasis:entry>
         <oasis:entry colname="col2"><inline-formula><mml:math id="M131" display="inline"><mml:mn mathvariant="normal">1172</mml:mn></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3" align="left">–</oasis:entry>
         <oasis:entry colname="col4">–</oasis:entry>
         <oasis:entry colname="col5" align="left">–</oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1">Treatment (L1)</oasis:entry>
         <oasis:entry colname="col2"><inline-formula><mml:math id="M132" display="inline"><mml:mn mathvariant="normal">1232</mml:mn></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3" align="left"><inline-formula><mml:math id="M133" display="inline"><mml:mrow><mml:mo mathvariant="bold">-</mml:mo><mml:mn mathvariant="bold">0.55</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula>   <inline-formula><mml:math id="M134" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1.06</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mo>-</mml:mo><mml:mn mathvariant="normal">0.03</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col4"><inline-formula><mml:math id="M135" display="inline"><mml:mn mathvariant="bold">0.044</mml:mn></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col5" align="left"><inline-formula><mml:math id="M136" display="inline"><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">7.2</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula>  <inline-formula><mml:math id="M137" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mo>-</mml:mo><mml:mn mathvariant="normal">8.5</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mo>-</mml:mo><mml:mn mathvariant="normal">5.7</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1">Treatment (L2)</oasis:entry>
         <oasis:entry colname="col2"><inline-formula><mml:math id="M138" display="inline"><mml:mn mathvariant="normal">190</mml:mn></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3" align="left"><inline-formula><mml:math id="M139" display="inline"><mml:mrow><mml:mn mathvariant="bold">0.30</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula>   <inline-formula><mml:math id="M140" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mo>-</mml:mo><mml:mn mathvariant="normal">0.65</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">1.28</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col4"><inline-formula><mml:math id="M141" display="inline"><mml:mn mathvariant="normal">0.642</mml:mn></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col5" align="left"><inline-formula><mml:math id="M142" display="inline"><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">9.7</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula>   <inline-formula><mml:math id="M143" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mo>-</mml:mo><mml:mn mathvariant="normal">12.5</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mo>-</mml:mo><mml:mn mathvariant="normal">6.9</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">Treatment (L3)</oasis:entry>
         <oasis:entry colname="col2"><inline-formula><mml:math id="M144" display="inline"><mml:mn mathvariant="normal">112</mml:mn></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3" align="left"><inline-formula><mml:math id="M145" display="inline"><mml:mrow><mml:mn mathvariant="bold">0.02</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula>   <inline-formula><mml:math id="M146" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mo>-</mml:mo><mml:mn mathvariant="normal">1.17</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">1.16</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col4"><inline-formula><mml:math id="M147" display="inline"><mml:mn mathvariant="normal">0.974</mml:mn></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col5" align="left"><inline-formula><mml:math id="M148" display="inline"><mml:mrow><mml:mo>-</mml:mo><mml:mn mathvariant="normal">10.9</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula>   <inline-formula><mml:math id="M149" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mo>-</mml:mo><mml:mn mathvariant="normal">14.7</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mo>-</mml:mo><mml:mn mathvariant="normal">7.1</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
       </oasis:row>
     </oasis:tbody>
   </oasis:tgroup></oasis:table></table-wrap>

      <fig id="F2" specific-use="star"><label>Figure 2</label><caption><p id="d2e3212">Histogram of fuel/flight distance for each flight in the control and treatment groups. Two peaks correspond to the two different aircraft types in the trial: Boeing 777 and Boeing 787. In panel <bold>(a)</bold>, the groups have different flight counts per peak; this imbalance is mostly removed in panel <bold>(b)</bold> after adjusting for aircraft type.</p></caption>
          <graphic xlink:href="https://jecats.copernicus.org/articles/1/6/2026/jecats-1-6-2026-f02.png"/>

        </fig>

      <p id="d2e3228">Using actual fuel data provided by AAL, we compared the fuel usage between the control group and treatment subgroups. The results from this analysis are shown in Table <xref ref-type="table" rid="T4"/>. The unadjusted numbers show a large reduction in fuel in the treatment group. The flights in the trial included two aircraft types: Boeing 777 and Boeing 787. Figure <xref ref-type="fig" rid="F2"/> visualizes the distribution of fuel per flight distance for each flight, demonstrating the aircraft type imbalance and effects of the adjustment described in Sect. <xref ref-type="sec" rid="Ch1.S2.SS8"/>.</p>
      <p id="d2e3237">Even after the aircraft type adjustment, we were able to reject the null hypothesis of no change in fuel usage between the L1 treatment and control groups with relatively high confidence (two-sided <inline-formula><mml:math id="M150" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:mo>=</mml:mo><mml:mn mathvariant="normal">0.044</mml:mn></mml:mrow></mml:math></inline-formula>, stratified permutation test). However, the difference is small (<inline-formula><mml:math id="M151" display="inline"><mml:mrow><mml:mo>&lt;</mml:mo><mml:mn mathvariant="normal">1</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula>) and the test is close to not being statistically significant at the <inline-formula><mml:math id="M152" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:mo>=</mml:mo><mml:mn mathvariant="normal">0.05</mml:mn></mml:mrow></mml:math></inline-formula> level. The remaining difference may be due to other residual confounders such as different city pairs having slightly different average fuel usage, differences in weather, or differences in load factor. We are unable to reject the null hypothesis of no change in fuel usage between the L2 treatment and control groups (two-sided <inline-formula><mml:math id="M153" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:mo>=</mml:mo><mml:mn mathvariant="normal">0.642</mml:mn></mml:mrow></mml:math></inline-formula>, stratified permutation test) and the L3 treatment and control groups (two-sided <inline-formula><mml:math id="M154" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:mo>=</mml:mo><mml:mn mathvariant="normal">0.974</mml:mn></mml:mrow></mml:math></inline-formula>, stratified permutation test).</p>
</sec>
<sec id="Ch1.S3.SS4">
  <label>3.4</label><title>Counterfactual Flightkeys-Generated Plans</title>
      <p id="d2e3309">In the previous section we found a small reduction in fuel usage in the treatment flights which implemented contrail avoidance. We suspect that this difference is due to uncontrolled confounders such as different weather patterns between treatment group days and control group days. In this section we study whether these confounders could also be responsible for the observed differences in contrail formation between the treatment and control groups.</p>
      <p id="d2e3312">To do this, we compare the contrail avoidance plans in the treatment group to the non-avoidance plans <italic>on the same flights</italic>, which we call the “counterfactual” plans.</p>
      <p id="d2e3318">The advantage of doing this counterfactual analysis is that since the flights are the same, confounders such as city pair and weather cannot be responsible for an observed reduction.  Figure <xref ref-type="fig" rid="F3"/> shows the results of the analysis described in Sect. <xref ref-type="sec" rid="Ch1.S2.SS9"/> to select which non-avoidance plan to use as a counterfactual. Since the “minimum cost” plan compares best against the flown values, we use this plan for subsequent analysis.</p>

      <fig id="F3" specific-use="star"><label>Figure 3</label><caption><p id="d2e3328">Bootstrap mean distributions of expected contrail distance for two potential choices of counterfactual plan (minimum cost plan and most recent plan before takeoff) compared against the flown path for flights in the L1 group where the contrail avoidance plan was not released (“L1 not L2” flights). The minimum cost counterfactual plan matches better to the flown values.</p></caption>
          <graphic xlink:href="https://jecats.copernicus.org/articles/1/6/2026/jecats-1-6-2026-f03.png"/>

        </fig>

      <fig id="F4" specific-use="star"><label>Figure 4</label><caption><p id="d2e3339">Comparison of the ML formation probability forecast vs. observed match distance per flight distance for the control and 3 treatment groups in the trial. The value of the bar is the mean match distance per flight distance for the group, with error bars showing the standard deviation. For each group, the forecast agrees with the observations.</p></caption>
          <graphic xlink:href="https://jecats.copernicus.org/articles/1/6/2026/jecats-1-6-2026-f04.png"/>

        </fig>

<table-wrap id="T5" specific-use="star"><label>Table 5</label><caption><p id="d2e3351">Contrail distance reduction factors of the two per-protocol groups evaluated against the counterfactual plans. Expected contrail distance uses the counterfactual plan. Observed results are reproduced here from Table <xref ref-type="table" rid="T1"/> as a baseline. The values in brackets represent the 95 % confidence interval of the column quantity. Bold numbers indicate that those are the primary results of the table. Italic font here is meant to draw attention to the fact that the point of comparison in the two rows of the table are different.</p></caption><oasis:table frame="topbot"><oasis:tgroup cols="3">
     <oasis:colspec colnum="1" colname="col1" align="justify" colwidth="4cm"/>
     <oasis:colspec colnum="2" colname="col2" align="right"/>
     <oasis:colspec colnum="3" colname="col3" align="right"/>
     <oasis:thead>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1" align="left"/>
         <oasis:entry colname="col2">L2</oasis:entry>
         <oasis:entry colname="col3">L3</oasis:entry>
       </oasis:row>
     </oasis:thead>
     <oasis:tbody>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1" align="left">Observed  Treatment vs. <italic>Control</italic> (baseline)</oasis:entry>
         <oasis:entry colname="col2"><inline-formula><mml:math id="M155" display="inline"><mml:mrow><mml:mn mathvariant="normal">29.5</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula> <inline-formula><mml:math id="M156" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">9.7</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">46.0</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3"><inline-formula><mml:math id="M157" display="inline"><mml:mrow><mml:mn mathvariant="normal">62.0</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula> <inline-formula><mml:math id="M158" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">47.1</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">75.3</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1" align="left">Expected Contrail Distance  Treatment vs. <italic>Counterfactual</italic></oasis:entry>
         <oasis:entry colname="col2"><inline-formula><mml:math id="M159" display="inline"><mml:mrow><mml:mn mathvariant="bold">25.0</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula> <inline-formula><mml:math id="M160" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">11.5</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">36.8</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3"><inline-formula><mml:math id="M161" display="inline"><mml:mrow><mml:mn mathvariant="bold">49.2</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula> <inline-formula><mml:math id="M162" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">34.9</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">60.9</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula></oasis:entry>
       </oasis:row>
     </oasis:tbody>
   </oasis:tgroup></oasis:table></table-wrap>

      <p id="d2e3541">A drawback of the conterfactual method is that since the counterfactual plan was not flown, we cannot use observations for this analysis. Instead, we evaluate the expected contrail distance from the contrail forecast. As discussed in Sect. <xref ref-type="sec" rid="Ch1.S2.SS1"/>, the final forecast output is the product of an ML-predicted contrail formation probability and an averaged forcing estimate. However, in the analysis shown in Sect. <xref ref-type="table" rid="T5"/> we report on the contrail length reductions because they are more easily directly observable. The contrail probability may not in general agree with observations on an individual flight, since the probability ranges from 0–1 while the observation-derived label from the CoAtSaC algorithm is either 0 (no contrail observed) or 1 (contrail observed). However, over a large number of flights these differences average out. As shown in Fig. <xref ref-type="fig" rid="F4"/>, the ML contrail forecast probability agrees with observations when aggregated over our treatment and control groups, justifying its use as a proxy for the observed results.</p>
      <p id="d2e3550">The results of this analysis are shown in Table <xref ref-type="table" rid="T5"/>. We see that when evaluated against the counterfactual plan, the reduction factors are smaller than the reductions seen in the observed results. However, the error bars between the counterfactual reductions and the observed reductions overlap significantly, and both treatment vs. counterfactual comparisons achieved statistical significance when evaluated using the permutation test (<inline-formula><mml:math id="M163" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:mo>&lt;</mml:mo><mml:mn mathvariant="normal">0.001</mml:mn></mml:mrow></mml:math></inline-formula>). Note that we do not use the stratified permutation test here since the counterfactual and treatment groups will have the same aircraft type distributions by construction. This counterfactual analysis demonstrates that the observed difference between the treatment and control groups cannot be explained by confounding differences in the flights chosen for each group.</p>
      <p id="d2e3568">In order to perform a similar analysis for fuel, we selected the most recently created contrail-optimized plan as the “treatment flight plan” and the non-avoidance plan that was created closest in time to the treatment flight plan to be the “control flight plan”. Similarly to the observed results, the simulated fuel comparisons do not achieve statistical significance for the L2 and L3 groups (L2 <inline-formula><mml:math id="M164" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:mo>=</mml:mo><mml:mn mathvariant="normal">0.855</mml:mn></mml:mrow></mml:math></inline-formula>, L3 <inline-formula><mml:math id="M165" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:mo>=</mml:mo><mml:mn mathvariant="normal">0.868</mml:mn></mml:mrow></mml:math></inline-formula>, permutation test).</p>
</sec>
</sec>
<sec id="Ch1.S4" sec-type="conclusions">
  <label>4</label><title>Conclusions</title>
      <p id="d2e3604">This study describes a scalable large-scale Randomized Controlled Trial that demonstrates the feasibility and efficacy of navigational contrail avoidance within a commercial airline's standard operational workflow. By integrating contrail forecasts directly into the Flightkeys flight planning software, we enabled dispatchers to execute avoidance maneuvers without the need for bespoke communication channels. The results provide strong evidence that navigational contrail avoidance is effective in reducing observable contrail formation and operationally feasible in realistic circumstances.</p>
      <p id="d2e3607">In the intent-to-treat group (Level 1, <inline-formula><mml:math id="M166" display="inline"><mml:mrow><mml:mi>n</mml:mi><mml:mo>=</mml:mo><mml:mn mathvariant="normal">1232</mml:mn></mml:mrow></mml:math></inline-formula>), we observed an 11.6 % <inline-formula><mml:math id="M167" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">2.1</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">20.3</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula> reduction in observable contrail formation relative to the control group. This value is statistically significant (<inline-formula><mml:math id="M168" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:mo>=</mml:mo><mml:mn mathvariant="normal">0.011</mml:mn></mml:mrow></mml:math></inline-formula>), though the uncertainty of the overall size of the effect is large. Based on these observations, we estimated the warming reduction to be 13.7 % <inline-formula><mml:math id="M169" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">3.1</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">23.4</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula> with the observation-informed average forcing method (<inline-formula><mml:math id="M170" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:mo>=</mml:mo><mml:mn mathvariant="normal">0.006</mml:mn></mml:mrow></mml:math></inline-formula>) and 14.3 % <inline-formula><mml:math id="M171" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mo>-</mml:mo><mml:mn mathvariant="normal">4.5</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">29.9</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula> (<inline-formula><mml:math id="M172" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:mo>=</mml:mo><mml:mn mathvariant="normal">0.045</mml:mn></mml:mrow></mml:math></inline-formula>) with the observation-informed ensemble forcing method.</p>
      <p id="d2e3727">These figures represent the aggregate benefit across the entire network subset, including flights where no action was taken. The efficacy of the intervention becomes most apparent in the per-protocol analysis (Level 3, <inline-formula><mml:math id="M173" display="inline"><mml:mrow><mml:mi>n</mml:mi><mml:mo>=</mml:mo><mml:mn mathvariant="normal">112</mml:mn></mml:mrow></mml:math></inline-formula>), which isolates flights that successfully flew the contrail-optimized plans. In this group, observable contrail length was reduced (<inline-formula><mml:math id="M174" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:mo>&lt;</mml:mo><mml:mn mathvariant="normal">0.001</mml:mn></mml:mrow></mml:math></inline-formula>) by 62.0 % <inline-formula><mml:math id="M175" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">47.1</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">75.3</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula>, observation-informed average forcing was reduced (<inline-formula><mml:math id="M176" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:mo>&lt;</mml:mo><mml:mn mathvariant="normal">0.001</mml:mn></mml:mrow></mml:math></inline-formula>) by 69.3 % <inline-formula><mml:math id="M177" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">56.1</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">80.9</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula>, and observation-informed average forcing was reduced (<inline-formula><mml:math id="M178" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:mo>&lt;</mml:mo><mml:mn mathvariant="normal">0.001</mml:mn></mml:mrow></mml:math></inline-formula>) by 83.0 % <inline-formula><mml:math id="M179" display="inline"><mml:mrow><mml:mo>[</mml:mo><mml:mn mathvariant="normal">74.8</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>,</mml:mo><mml:mn mathvariant="normal">89.4</mml:mn><mml:mspace width="0.125em" linebreak="nobreak"/><mml:mi mathvariant="italic">%</mml:mi><mml:mo>]</mml:mo></mml:mrow></mml:math></inline-formula>. This substantial gap between the intent-to-treat and per-protocol results highlights that while the physical mechanism of avoidance is highly effective, the operational execution  –  specifically the “take rate” of the plans  –  remains a primary bottleneck for network-wide scaling.</p>
      <p id="d2e3845">By utilizing the CoAtSaC  automated contrail-to-flight attribution algorithm  <xref ref-type="bibr" rid="bib1.bibx33" id="paren.48"/>, we were able to objectively validate performance across thousands of flights using GOES-East imagery. This methodology establishes a new standard for validating the efficacy of contrail avoidance trials, proving that contrail reductions can be quantified using peer-reviewed methodologies at large scales.</p>
      <p id="d2e3852">In this work we are reporting a <inline-formula><mml:math id="M180" display="inline"><mml:mrow><mml:mn mathvariant="normal">62.0</mml:mn><mml:mspace linebreak="nobreak" width="0.125em"/><mml:mi mathvariant="italic">%</mml:mi></mml:mrow></mml:math></inline-formula> reduction in contrails observable via geostationary satellite, but not all contrails are observable. In particular, the GOES ABI images have a nadir resolution of 2 km, which makes it difficult to detect young, narrow contrails. The reductions in total contrail distance and contrail warming are likely smaller if one assumes that some unobservable contrails are less affected by the contrail avoidance performed in this test. What fraction of contrails (and contrail warming) are observable is a subject of active research. For example, <xref ref-type="bibr" rid="bib1.bibx8" id="text.49"/> compared a set of GOES ABI images to higher-resolution low-Earth orbit (LEO) satellite images and found that 42 % of the higher-resolution contrail distance was visible in the nearest GOES scan. However, we should note that the CoAtSaC algorithm <xref ref-type="bibr" rid="bib1.bibx33" id="paren.50"/> only requires detecting the contrail in at least two scans captured during its lifespan, so having some frames where it is detectable in LEO but not in GOES is not inherently problematic. Furthermore, since more warming contrails tend to be larger with higher optical depth, they are also generally easier to detect in a GOES-ABI image, so the fraction of warming that is observable may be larger than the fraction of contrail distance. For example, <xref ref-type="bibr" rid="bib1.bibx7" id="text.51"/> used radiative transfer simulations to estimate that, in the clear sky case, 45 % of contrail distance is observable in a given GOES ABI frame, but 57 % of contrail distance and 97 % of contrail long-wave energy forcing comes from contrails observable at some point in their lifetime. Further research is needed to better quantify the fraction of contrail warming that is observable.</p>
      <p id="d2e3875">Operational integration presented specific challenges that influenced the trial's outcomes. The “take rates” of 15.4 % (flights released as contrail plans vs. total candidate flights) and 7.8 % (flights released as contrail plans and flown as planned vs. total candidate flights) are relatively low. Given the complexity of even routine daily aircraft operations, there are a large number of reasons why a contrail plan might not be released, or a released plan might not be flown. In interviews with participating dispatchers, a few common cases were discussed and summarized here as follows. Dispatchers found that contrail avoidance plans sometimes involved descents and ascents in the middle of a flight, which are safe and done in coordination with air traffic control, but which the dispatchers and flight crews preferred not to do. In the North America to Europe flights included in this trial, AAL preferred to adhere to the North Atlantic Organised Track System (OTS), and Flightkeys had special user interface features to make this easy for the dispatchers. These features were for technical reasons not able to be activated for the contrail flights during this trial, making contrail avoidance more difficult near the OTS tracks, which in many cases led to dispatchers choosing not to attempt contrail avoidance on those flights. Due to such issues, the take rate of dispatchers releasing a contrail-aware flight plan ranged widely per city pair. Dispatchers also mentioned that their take rate was typically lower when they and the aircraft crew were busier. As this trial was executed on a voluntary basis, a dispatcher could also choose to not take part for any reason. Dispatcher priorities were established with safety as the primary focus, followed by efficiency and finally contrails. The L2 <inline-formula><mml:math id="M181" display="inline"><mml:mo>/</mml:mo></mml:math></inline-formula> L1 take rate on a per-dispatcher basis ranged widely, suggesting that inter-dispatcher variation is important. Since city pairs were not evenly distributed between dispatchers, this likely had an effect on the difference in take rate per city pair. Finally, dispatchers mentioned the presence of turbulence, which is a safety concern, as a reason for not considering the contrail-optimized flight plan. Further research is necessary to understand the correlation between turbulence and contrail formation.</p>
      <p id="d2e3885">A few software improvements could improve the take rate for future trials. In the user interface available to dispatchers for this trial, there was only a top-down view of the contrail avoidance regions, but additionally including a vertical profile view would be beneficial. This made it difficult for dispatchers to understand why certain adjustments were happening.</p>
      <p id="d2e3888">Improvements to the user interface inside flight planning software could lead to a larger observed contrail reduction, even in the L3 group. Dispatchers frequently make manual adjustments to software-optimized plans, even when contrails are not considered. In a few (2–3) instances in the trial, dispatchers, without the benefit of the vertical profile view of the contrail avoidance regions, adjusted a contrail-optimized plan to fly through the contrail avoidance region, but these flights nonetheless ended up in the L3 group.</p>
      <p id="d2e3891">Around half of the released contrail-optimized flight plans were not flown as planned, as determined by the L3 criteria. This could potentially be ameliorated by displaying contrail information in tools accessible to pilots such as the “Electronic Flight Bag” (EFB), so that pilots can also understand why maneuvers are being made. Additionally, the automated flight planning process starts roughly 24 h prior to departure, and the dispatchers engage with the process closer to the departure time. The contrail forecast can change during that time. If a flight was initially exceeding the 10 Mg threshold, a dispatcher could start working on a contrail plan even if later updates to the weather forecast brought the flight below the 10 Mg threshold. This could have also occurred in the control group and therefore should not substantially affect the results, but it might have caused flights whose non-avoidance plans had low contrail impact to end up in both the treatment and control groups. Future improvements to flight planning software could ameliorate both of these problems. Additionally, in a system-wide deployment of contrail avoidance, the efficacy of the contrail avoidance regime may improve since dispatchers would only be presented with flight plans that consider contrails, rather than being presented with multiple non-avoidance and contrail-optimized plans from which to choose.</p>
      <p id="d2e3894">Our analysis of fuel burn revealed a counter-intuitive result: the Level 1 treatment group appeared to consume 0.55 % less fuel than the control group (adjusted <inline-formula><mml:math id="M182" display="inline"><mml:mrow><mml:mi>p</mml:mi><mml:mo>=</mml:mo><mml:mn mathvariant="normal">0.044</mml:mn></mml:mrow></mml:math></inline-formula>). As detailed in Sect. <xref ref-type="sec" rid="Ch1.S3.SS3"/>, this finding is likely driven by residual confounding within the randomization structure rather than a genuine fuel benefit of contrail avoidance. This “negative cost” result, along with the <inline-formula><mml:math id="M183" display="inline"><mml:mrow><mml:mo>≈</mml:mo><mml:mn mathvariant="normal">2</mml:mn></mml:mrow></mml:math></inline-formula> percentage-point-wide confidence intervals for fuel reduction in the L2/L3 groups, underscores the difficulty of isolating the fuel penalty of avoidance (estimated at 0.1 %–0.2 % across all of an airline's flights; <xref ref-type="bibr" rid="bib1.bibx39 bib1.bibx9 bib1.bibx6" id="altparen.52"/>) against the high variance of real-world fuel consumption data. Understanding the fuel penalty is critical for assessing the climate impact of contrail avoidance, since additional fuel usage is accompanied with higher emissions and climate impacts of <inline-formula><mml:math id="M184" display="inline"><mml:mrow class="chem"><mml:msub><mml:mi mathvariant="normal">CO</mml:mi><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> and non-<inline-formula><mml:math id="M185" display="inline"><mml:mrow class="chem"><mml:msub><mml:mi mathvariant="normal">CO</mml:mi><mml:mn mathvariant="normal">2</mml:mn></mml:msub></mml:mrow></mml:math></inline-formula> gases and aerosols. Larger sample sizes and stricter randomization protocols in future trials will be necessary to precisely quantify the fuel penalty trade-off.</p>
      <p id="d2e3948">Finally, while our counterfactual analysis suggests that the observed reductions were not driven by dispatchers selecting flights that would have formed fewer contrails even without our intervention, the infrequent randomization of city pairs remains a limitation. To further establish causality and refine the cost-benefit analysis, future work could expand beyond the North Atlantic to a global scope to ensure that the methods proposed here work in more congested airspace, increase the frequency of randomization, and further automate the optimization workflow to reduce labor burden on dispatchers. Future work should also explore different operational concepts for contrail avoidance such as coordination between airlines and Air Navigation Service Providers (ANSPs), ANSP-led trials, or coordination between multiple airlines. Despite these limitations, this trial provides a statistically robust point of evidence that commercial aviation can successfully avoid warming contrails using existing operational infrastructure.</p>
</sec>

      
      </body>
    <back><app-group>

<app id="App1.Ch1.S1">
  <label>Appendix A</label><title>Machine Learning Contrail Forecast</title>
      <p id="d2e3963">NWP models and their reanalysis products are known to contain biases in atmospheric humidity fields when compared to observations <xref ref-type="bibr" rid="bib1.bibx1 bib1.bibx12" id="paren.53"/>. In order to account for this uncertainty, we used a machine learning model to predict geostationary-observable contrail formation from NWP output fields, effectively acting as a debiasing operator against contrail observations on the NWP outputs. Similar approaches for ML debiasing of humidity have been discussed in the literature <xref ref-type="bibr" rid="bib1.bibx41" id="paren.54"/>.</p>
<sec id="App1.Ch1.S1.SSx1" specific-use="unnumbered">
  <title>Model Construction</title>
      <p id="d2e3977">The ML model was trained using weather fields as inputs and contrail formation labels from the automated flight-to-contrail attribution algorithm described in <xref ref-type="bibr" rid="bib1.bibx33" id="text.55"/> with the modifications from Appendix <xref ref-type="sec" rid="App1.Ch1.S2"/>. The training data for the model consisted of flight waypoints labeled by whether that flight waypoint formed a contrail.</p>

<table-wrap id="TA1"><label>Table A1</label><caption><p id="d2e3988">Weather variables and summary statistics provided as input features to the ML forecast model. Variables in italics are only provided at the original flight waypoint and are not summarized using the summary statistics.</p></caption><oasis:table frame="topbot"><oasis:tgroup cols="2">
     <oasis:colspec colnum="1" colname="col1" align="left"/>
     <oasis:colspec colnum="2" colname="col2" align="left"/>
     <oasis:thead>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1">Features</oasis:entry>
         <oasis:entry colname="col2">Summary Statistics</oasis:entry>
       </oasis:row>
     </oasis:thead>
     <oasis:tbody>
       <oasis:row>
         <oasis:entry colname="col1">relative humidity</oasis:entry>
         <oasis:entry colname="col2">minimum</oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">vertical velocity</oasis:entry>
         <oasis:entry colname="col2">5th percentile</oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">temperature</oasis:entry>
         <oasis:entry colname="col2">25th percentile</oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">specific humidity</oasis:entry>
         <oasis:entry colname="col2">mean</oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">u component of wind</oasis:entry>
         <oasis:entry colname="col2">standard deviation</oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">v component of wind</oasis:entry>
         <oasis:entry colname="col2">75th percentile</oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">divergence</oasis:entry>
         <oasis:entry colname="col2">90th percentile</oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">vorticity (relative)</oasis:entry>
         <oasis:entry colname="col2">maximum</oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">specific cloud ice water content</oasis:entry>
         <oasis:entry colname="col2"/>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">specific snow water content</oasis:entry>
         <oasis:entry colname="col2"/>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">fraction of cloud cover</oasis:entry>
         <oasis:entry colname="col2"/>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1"><italic>flight location latitude</italic></oasis:entry>
         <oasis:entry colname="col2"/>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1"><italic>flight location longitude</italic></oasis:entry>
         <oasis:entry colname="col2"/>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1"><italic>flight location pressure altitude</italic></oasis:entry>
         <oasis:entry colname="col2"/>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1"><italic>sine of day of year</italic></oasis:entry>
         <oasis:entry colname="col2"/>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1"><italic>cosine of day of year</italic></oasis:entry>
         <oasis:entry colname="col2"/>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1"><italic>cosine of solar minutes</italic></oasis:entry>
         <oasis:entry colname="col2"/>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1"><italic>sine of solar minutes</italic></oasis:entry>
         <oasis:entry colname="col2"/>
       </oasis:row>
     </oasis:tbody>
   </oasis:tgroup></oasis:table></table-wrap>

      <p id="d2e4177">In order to create input features for the model, each flight waypoint was advected in 3 dimensions with the NWP output winds using the third-order Runge-Kutta method <xref ref-type="bibr" rid="bib1.bibx3" id="paren.56"/> with 10 min timesteps for a total of 2 h. At each advected location, weather features were read from the NWP outputs. For each weather variable, its values across the advected path were summarized using a variety of statistics. These summary statistics were given to the ML model as input features. The model also received additional features about the flight location and time of day. A full list of input features and statistics is provided in Table <xref ref-type="table" rid="TA1"/>.</p>
      <p id="d2e4185">The ML model is a 4-layer deep neural network, trained using the AdamW optimizer with cross-entropy loss. The hyperparameters of the model were optimized using a hyperparameter search. We additionally applied a  calibration on the model outputs against a held-out set of the contrail formation labels to translate the model outputs into calibrated probabilities. More information about the ML forecast is available in <xref ref-type="bibr" rid="bib1.bibx38" id="text.57"/>.</p>
      <p id="d2e4192">The performance of the ML contrail forecast model is evaluated in ContrailBench <xref ref-type="bibr" rid="bib1.bibx17" id="paren.58"/>. In the operational workflow described in this trial, the forecast is treated as a continuous value; therefore, multiple operating points on the ContrailBench curve will be relevant in practice.</p>
      <p id="d2e4198">During development, the model was constructed and validated using the freshest available NWP forecast weather. The model's performance at operational forecast lead times is not assessed directly, but prior work suggests that contrail forecasts are performant at operationally-relevant lead times <xref ref-type="bibr" rid="bib1.bibx6" id="paren.59"/>. The trial described in this work is also designed to assess the forecast performance at operationally-relevant lead times.</p>
</sec>
</app>

<app id="App1.Ch1.S2">
  <label>Appendix B</label><title>Modifications to the Contrail-to-Flight Attribution Algorithm</title>
      <p id="d2e4214">For the validation system for the trial described in this paper, we used a modified version of the contrail-to-flight attribution algorithm described in <xref ref-type="bibr" rid="bib1.bibx33" id="text.60"/>. The modified version of this algorithm is also used for generating training data for the ML contrail formation forecast.</p>
<sec id="App1.Ch1.S2.SS1">
  <label>B1</label><title>Contrail Altitude Model</title>
      <p id="d2e4228">We developed a system to infer the altitude of a contrail in GOES imagery, similar to the method introduced in <xref ref-type="bibr" rid="bib1.bibx26" id="text.61"/>. The system uses a convolutional neural network (CNN) that takes radiances from tiles of the GOES imagery as input and infers an altitude at each pixel in the tile. This CNN is trained using altitudes from flight-to-contrail attributions as labels. Specifically, during the attribution process, the advection of exhaust plumes from flights are simulated in 3D space from their original locations to the times of satellite images, as described in <xref ref-type="bibr" rid="bib1.bibx33" id="text.62"/>. If the attribution algorithm finds a matched flight for a given detected contrail, the advected altitude of that flight is used as the label for the pixels spanned by the detected contrail. Detected contrails that do not have a matched flight are not included in the training data. The loss is only applied on pixels where the dataset includes a detected contrail with a corresponding matched flight.</p>
      <p id="d2e4237">In order to account for the effects of viewing angle on the perceived altitude by the satellite, we transform the advected barometric altitude of the flight into a “radiative path length”, which allows the altitude estimation model to take into account the amount of atmosphere through which the satellite-sensed radiation traveled between the earth surface and the observed contrail. To perform this transformation, we first interpolate the advected pressure level in ERA5 reanalysis data to find its corresponding geopotential height, and then convert that to geodetic height by dividing it by the acceleration due to Earth's gravity. We then apply trigonometric identities (visualized for a simplified 2D case in Fig. <xref ref-type="fig" rid="FB1"/>) to translate the geodetic height into a radiative path length. After the model performs inference, the inverse of this trigonometric transformation is used to translate the radiative path length back to a geodetic altitude.</p>
      <p id="d2e4242">During the training process, the model is validated against a held-out set of flight-to-contrail attributions. This dataset has the advantage of wide spatiotemporal coverage, but has the disadvantage of incorporating the noise introduced by the inaccuracies of the attribution algorithm. In order to ensure that the model performs well and generalizes, we additionally evaluate its performance on the LiDAR colocation dataset from <xref ref-type="bibr" rid="bib1.bibx26" id="text.63"/> that provides high-accuracy altitude estimations from the CALIOP instrument onboard the CALIPSO satellite. On this dataset, our model achieves a root mean squared error (RMSE) of 789 m, compared to the RMSE of 570 m from <xref ref-type="bibr" rid="bib1.bibx26" id="text.64"/>. We chose not to use the CALIOP data for training data generation in the model since CALIPSO was decommissioned in 2023. Since the RMSE of this altitude estimation is greater than 2 flight levels (2000 ft <inline-formula><mml:math id="M186" display="inline"><mml:mo>=</mml:mo></mml:math></inline-formula> 609.6 m), we use it as a coarse filter in the modified attribution algorithm described below.</p>
      <p id="d2e4258">At inference time, the model produces an inferred altitude for every pixel in the input satellite image, regardless of whether there is a contrail present. In order to use these inference results in the downstream attribution algorithm, we join the inferred altitudes with detected contrails from <xref ref-type="bibr" rid="bib1.bibx27" id="text.65"/>. Each linearized contrail is assigned the mean altitude of the inference results from the pixels it contains.</p>
</sec>
<sec id="App1.Ch1.S2.SS2">
  <label>B2</label><title>Modified Contrail-to-Flight Attribution Algorithm</title>
      <p id="d2e4272">At a high level, the system described in <xref ref-type="bibr" rid="bib1.bibx33" id="text.66"/> begins by considering attributions in each frame of the satellite imagery and looks for patterns in how those single-frame attributions evolve over time through multiple frames of the satellite image. The key modification we make to this algorithm is to filter out single-frame attributions where the advected altitude of the flight is significantly different from the inferred altitude of the detected contrail as determined by the CNN. We set the threshold as

            <disp-formula id="App1.Ch1.S2.E3" content-type="numbered"><label>B1</label><mml:math id="M187" display="block"><mml:mrow><mml:mn mathvariant="normal">2</mml:mn><mml:mo>⋅</mml:mo><mml:msqrt><mml:mrow><mml:msup><mml:msub><mml:mi mathvariant="normal">RMSE</mml:mi><mml:mi mathvariant="normal">altitude</mml:mi></mml:msub><mml:mn mathvariant="normal">2</mml:mn></mml:msup><mml:mo>+</mml:mo><mml:msup><mml:msub><mml:mi mathvariant="normal">RMSE</mml:mi><mml:mi mathvariant="normal">advection</mml:mi></mml:msub><mml:mn mathvariant="normal">2</mml:mn></mml:msup></mml:mrow></mml:msqrt><mml:mo>,</mml:mo></mml:mrow></mml:math></disp-formula>

          where <inline-formula><mml:math id="M188" display="inline"><mml:mrow><mml:msub><mml:mi mathvariant="normal">RMSE</mml:mi><mml:mi mathvariant="normal">altitude</mml:mi></mml:msub></mml:mrow></mml:math></inline-formula> is the root-mean-squared error of the altitude model from the previous section (800 m) and <inline-formula><mml:math id="M189" display="inline"><mml:mrow><mml:msub><mml:mi mathvariant="normal">RMSE</mml:mi><mml:mi mathvariant="normal">advection</mml:mi></mml:msub></mml:mrow></mml:math></inline-formula> is the vertical root-mean-squared error introduced during advection due to imperfect wind data.  <inline-formula><mml:math id="M190" display="inline"><mml:mrow><mml:msub><mml:mi mathvariant="normal">RMSE</mml:mi><mml:mi mathvariant="normal">advection</mml:mi></mml:msub></mml:mrow></mml:math></inline-formula> is computed independently of the advection age by comparing the advected altitude to the known synthetic altitude of contrails in SynthOpenContrails <xref ref-type="bibr" rid="bib1.bibx33" id="paren.67"/>, and is estimated at 111 m. Equation (<xref ref-type="disp-formula" rid="App1.Ch1.S2.E3"/>) therefore evaluates to 1615 m. If in a single satellite frame the inferred contrail altitude differs from the advected flight altitude by more than 1615 m, the single-frame attribution is not considered as a candidate match for the subsequent multitemporal aspects of the algorithm. Since 1615 m is greater than 5 flight levels (5000 ft), this is a coarse filter which conservatively removes highly unlikely candidate flight-contrail matches; the algorithm does not require strict agreement between the advected flight altitude and inferred contrail altitude. In order to evaluate the effect of this modification on the performance on CoAtSaC, we designed a version of the evaluation of CoAtSaC against SynthOpenContrails that assumes an RMSE of the altitude estimator and assigns the synthetic predicted altitude of a contrail to a value sampled from a Gaussian where the mean is the known altitude of the contrail and the standard deviation is the RMSE of the altitude estimator. This synthetic predicted altitude is used in place of the inferred altitude of the detected contrail in the modified algorithm described above. With this change, the “flight precision” on the tracking subset of SynthOpenContrails improves from 71.4 % to 79.4 %.</p><fig id="FB1"><label>Figure B1</label><caption><p id="d2e4351">Example of a transformation from geodetic altitude to radiative path length. This simplified example considers a contrail that is over the Equator at 40° longitude away from the nadir of the geostationary satellite. Its geodetic altitude is 10 km, which is translated into a radiative path length of 14.47 km using the triangle formed by the satellite, contrail, and center of the Earth and the purple triangle formed by the contrail, the point directly underneath it on the surface, and the point on the surface that the satellite is viewing for the pixel in which the contrail appears. Side lengths and angles other than the 40 degrees longitude, Earth radius, and geostationary orbit distance are derived via trigonometric identities.</p></caption>
          
          <graphic xlink:href="https://jecats.copernicus.org/articles/1/6/2026/jecats-1-6-2026-f05.png"/>

        </fig>


</sec>
</app>

<app id="App1.Ch1.S3">
  <label>Appendix C</label><title>City Pair Assignments</title>
      <p id="d2e4373">Table <xref ref-type="table" rid="TC1"/> shows the city pair assignments by week in the treatment and control groups.</p>

<table-wrap id="TC1"><label>Table C1</label><caption><p id="d2e4382">Flight route assignments by week and group. Week 1 began on Wednesday, 15 January 2025. Each week of the trial started on Wednesday and ended on the following Tuesday.</p></caption><oasis:table frame="topbot"><oasis:tgroup cols="3">
     <oasis:colspec colnum="1" colname="col1" align="justify" colwidth="3cm"/>
     <oasis:colspec colnum="2" colname="col2" align="justify" colwidth="6cm"/>
     <oasis:colspec colnum="3" colname="col3" align="justify" colwidth="6cm"/>
     <oasis:thead>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1" align="left">Week</oasis:entry>
         <oasis:entry colname="col2" align="left">Treatment Group Routes</oasis:entry>
         <oasis:entry colname="col3" align="left">Control Group Routes</oasis:entry>
       </oasis:row>
     </oasis:thead>
     <oasis:tbody>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1" align="left">Week 1 (Pretrial, Not Considered in Analysis)</oasis:entry>
         <oasis:entry colname="col2" align="left">JFK-LHR</oasis:entry>
         <oasis:entry colname="col3" align="left">–</oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1" align="left">Week 2</oasis:entry>
         <oasis:entry colname="col2" align="left">JFK-LHR, MIA-MAD, MIA-BCN, PHL-MAD</oasis:entry>
         <oasis:entry colname="col3" align="left">JFK-MAD, CLT-FRA, JFK-MXP, DFW-CDG</oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1" align="left">Week 3</oasis:entry>
         <oasis:entry colname="col2" align="left">JFK-LHR, MIA-MAD, MIA-BCN, PHL-MAD, PHL-BCN, CLT-MUC, CLT-FRA, MIA-LHR</oasis:entry>
         <oasis:entry colname="col3" align="left">ORD-LHR, CLT-LHR, CLT-MAD, PHL-LIS, PHL-CDG, PHL-AMS, JFK-MAD, DFW-CDG</oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1" align="left">Week 4</oasis:entry>
         <oasis:entry colname="col2" align="left">JFK-LHR, MIA-MAD, MIA-BCN, PHL-MAD, PHL-BCN, CLT-MUC, CLT-FRA, MIA-LHR, DFW-LHR, ORD-LHR, JFK-CDG, MIA-CDG, DFW-CDG, PHL-CDG, PHL-ZRH, JFK-MXP</oasis:entry>
         <oasis:entry colname="col3" align="left">PHL-LHR, PHL-LIS, CLT-MAD, BOS-LHR, DFW-FRA, LAX-LHR, CLT-LHR, PHL-AMS, PHL-DUB, JFK-MAD</oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1" align="left">Week 5–7</oasis:entry>
         <oasis:entry colname="col2" align="left">CLT-MUC, DFW-FRA, JFK-CDG, JFK-MAD, JFK-MXP, MIA-LHR, MIA-MAD, ORD-LHR, PHL-AMS, PHL-BCN, PHL-LHR, PHL-LIS, PHL-MAD, PHL-ZRH</oasis:entry>
         <oasis:entry colname="col3" align="left">JFK-LHR, MIA-BCN, CLT-FRA, MIA-CDG, DFW-CDG, PHL-CDG, CLT-MAD, BOS-LHR, LAX-LHR, CLT-LHR, PHL-DUB</oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1" align="left">Week 8–10</oasis:entry>
         <oasis:entry colname="col2" align="left">CLT-FRA, CLT-MAD, CLT-MUC, DFW-CDG, JFK-MAD, JFK-MXP, MIA-CDG, MIA-LHR, MIA-MAD, ORD-LHR, PHL-AMS, PHL-BCN, PHL-DUB, PHL-ZRH</oasis:entry>
         <oasis:entry colname="col3" align="left">JFK-LHR, MIA-BCN, PHL-MAD, JFK-CDG, PHL-CDG, PHL-LIS, PHL-LHR, BOS-LHR, DFW-FRA, LAX-LHR, CLT-LHR</oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1" align="left">Week 11–12</oasis:entry>
         <oasis:entry colname="col2" align="left">MIA-LHR, PHL-DUB, BOS-LHR, MIA-MAD, PHL-MAD, PHL-CDG, DFW-FRA, JFK-CDG, JFK-MAD, PHL-LIS, PHL-ZRH, DFW-CDG</oasis:entry>
         <oasis:entry colname="col3" align="left">JFK-LHR, MIA-BCN, PHL-BCN, CLT-MUC, CLT-FRA, MIA-CDG, JFK-MXP, PHL-LHR, CLT-MAD, PHL-AMS, DFW-FCO</oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1" align="left">Week 13–14</oasis:entry>
         <oasis:entry colname="col2" align="left">JFK-MXP, PHL-AMS, PHL-MAD, PHL-DUB, BOS-LHR, PHL-ZRH, MIA-LHR, MIA-MAD, PHL-BCN, DFW-MAD, PHL-LHR, MIA-BCN, PHL-FCO, DFW-FCO</oasis:entry>
         <oasis:entry colname="col3" align="left">JFK-LHR, CLT-MUC, CLT-FRA, JFK-CDG, MIA-CDG, DFW-CDG, PHL-CDG, PHL-LIS, CLT-MAD, DFW-FRA, JFK-MAD, PHL-ATH, DFW-BCN, JFK-BCN, JFK-FCO</oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1" align="left">Week 15–17</oasis:entry>
         <oasis:entry colname="col2" align="left">PHL-ZRH, PHL-LIS, CLT-FRA, DFW-BCN, CLT-MAD, DFW-CDG, PHL-DUB, PHL-LHR, PHL-BCN, JFK-MXP, PHL-FCO, PHL-MAD, PHL-CDG, JFK-BCN</oasis:entry>
         <oasis:entry colname="col3" align="left">JFK-LHR, MIA-BCN, CLT-MUC, MIA-LHR, JFK-CDG, MIA-CDG, BOS-LHR, DFW-FRA, PHL-AMS, JFK-MAD, DFW-FCO, PHL-ATH, DFW-MAD, JFK-FCO</oasis:entry>
       </oasis:row>
     </oasis:tbody>
   </oasis:tgroup></oasis:table></table-wrap>


</app>

<app id="App1.Ch1.S4">
  <label>Appendix D</label><title>Flight Adherence Visualizations</title>
      <p id="d2e4533">Figures <xref ref-type="fig" rid="FD1"/> and <xref ref-type="fig" rid="FD2"/> show examples of the visualization used by the authors of this work to assess flight plan adherence.</p>

      <fig id="FD1"><label>Figure D1</label><caption><p id="d2e4542">Example of an L2 flight that is not included in the L3 set. Dashed blue lines represent the flown path, and solid purple lines represent the planned path. The top image shows the vertical altitude profile, where the green colors represent the contrail forecast interpolated to the flown path. The bottom image shows the lateral profile, with the contrail forecast at a fixed 35 000 ft.</p></caption>
        
        <graphic xlink:href="https://jecats.copernicus.org/articles/1/6/2026/jecats-1-6-2026-f06.png"/>

      </fig>

<fig id="FD2"><label>Figure D2</label><caption><p id="d2e4557">Example of an L3 flight. Despite imperfect adherence to the vertical profile, the difference in forecasted contrail impact is not large enough to fail the L3 inclusion check. Dashed blue lines represent the flown path, and solid purple lines represent the planned path. The top image shows the vertical altitude profile, where the green colors represent the contrail forecast interpolated to the flown path. The bottom image shows the lateral profile, with the contrail forecast at a fixed 35 000 ft.</p></caption>
        
        <graphic xlink:href="https://jecats.copernicus.org/articles/1/6/2026/jecats-1-6-2026-f07.png"/>

      </fig>

</app>

<app id="App1.Ch1.S5">
  <label>Appendix E</label><title>Aircraft Type Distributions</title>
      <p id="d2e4576">Figure <xref ref-type="fig" rid="FE1"/> shows the distributions of aircraft type in the unadjusted treatment and control groups.</p>

      <fig id="FE1"><label>Figure E1</label><caption><p id="d2e4583">Distributions of aircraft types across the treatment and control groups before the adjustment described in Sect. <xref ref-type="sec" rid="Ch1.S2.SS8"/>.</p></caption>
        
        <graphic xlink:href="https://jecats.copernicus.org/articles/1/6/2026/jecats-1-6-2026-f08.png"/>

      </fig>


</app>

<app id="App1.Ch1.S6">
  <label>Appendix F</label><title>Additional Observed Contrail Values</title>
      <p id="d2e4607">Table <xref ref-type="table" rid="TF1"/> shows additional observed contrail values.</p>

<table-wrap id="TF1"><label>Table F1</label><caption><p id="d2e4615">Observed contrail formation results and total observed distances.</p></caption><oasis:table frame="topbot"><oasis:tgroup cols="3">
     <oasis:colspec colnum="1" colname="col1" align="left"/>
     <oasis:colspec colnum="2" colname="col2" align="right"/>
     <oasis:colspec colnum="3" colname="col3" align="right"/>
     <oasis:thead>
       <oasis:row>
         <oasis:entry colname="col1">Group</oasis:entry>
         <oasis:entry colname="col2"><inline-formula><mml:math id="M191" display="inline"><mml:mi>N</mml:mi></mml:math></inline-formula></oasis:entry>
         <oasis:entry colname="col3">Total Observed Contrail</oasis:entry>
       </oasis:row>
       <oasis:row rowsep="1">
         <oasis:entry colname="col1"/>
         <oasis:entry colname="col2"/>
         <oasis:entry colname="col3">Distance (km)</oasis:entry>
       </oasis:row>
     </oasis:thead>
     <oasis:tbody>
       <oasis:row>
         <oasis:entry colname="col1">Control</oasis:entry>
         <oasis:entry colname="col2">1172</oasis:entry>
         <oasis:entry colname="col3">299 317</oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">Treatment (L1)</oasis:entry>
         <oasis:entry colname="col2">1232</oasis:entry>
         <oasis:entry colname="col3">281 214</oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">Treatment (L2)</oasis:entry>
         <oasis:entry colname="col2">190</oasis:entry>
         <oasis:entry colname="col3">32 223</oasis:entry>
       </oasis:row>
       <oasis:row>
         <oasis:entry colname="col1">Treatment (L3)</oasis:entry>
         <oasis:entry colname="col2">112</oasis:entry>
         <oasis:entry colname="col3">10 640</oasis:entry>
       </oasis:row>
     </oasis:tbody>
   </oasis:tgroup></oasis:table></table-wrap>

</app>
  </app-group><notes notes-type="dataavailability"><title>Data availability</title>

      <p id="d2e4714">The accompanying dataset and code for this paper are available at <ext-link xlink:href="https://doi.org/10.5281/zenodo.19473496" ext-link-type="DOI">10.5281/zenodo.19473496</ext-link> <xref ref-type="bibr" rid="bib1.bibx32" id="paren.68"/>. Note that some data, such as per-flight fuel usage, is unavailable since this data is proprietary. ERA5 data are available from the Copernicus Climate Change Service Climate Data Store (CDS): <ext-link xlink:href="https://doi.org/10.24381/cds.bd0915c6" ext-link-type="DOI">10.24381/cds.bd0915c6</ext-link> <xref ref-type="bibr" rid="bib1.bibx16" id="paren.69"/>. Visualization of Contrail Detections on GOES-16 ABI data can be found at <uri>https://contrails.webapps.google.com/</uri> (last access: 13 November 2026). Raw GOES-16 data can be found at: <uri>https://console.cloud.google.com/marketplace/product/noaa-public/goes</uri> (last access: 13 November 2026).</p>
  </notes><notes notes-type="authorcontribution"><title>Author contributions</title>

      <p id="d2e4739">The study scope and randomized trial design were developed by TS, TD, SG, and ASW in collaboration with JB, MG, RG, and AP. JB, MG, RG, and AP led the operational implementation of the study at the airline, including communications with participants in the trial. AMF and RZ developed the integration of the contrail forecast into the flight planning software. AS, TS, and SG compiled and pre-processed the necessary flight and satellite data for the analysis. TS and ASW conducted the statistical analysis. TA, PH, KM, TR, DS, and MS provided scientific advising throughout the development of the study and the compiling of the results. TS interpreted the results and wrote the manuscript, including significant contributions from TD, TR, and SG. All authors reviewed the manuscript.</p>
  </notes><notes notes-type="competinginterests"><title>Competing interests</title>

      <p id="d2e4745">As denoted by their affiliations, some authors are employed by Google LLC, Flightkeys GmbH, and American Airlines. Contrails.org is operated by Breakthrough Energy, a family of organizations and activities committed to transitioning the world to net zero by 2050. All other authors declare no competing interests.</p>
  </notes><notes notes-type="disclaimer"><title>Disclaimer</title>

      <p id="d2e4751">Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. The authors bear the ultimate responsibility for providing appropriate place names. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.</p>
  </notes><ack><title>Acknowledgements</title><p id="d2e4757">We dedicate this paper to the memory of Darcy Freeman, a dispatcher who released many contrail flights during the course of this trial. We thank The International Council on Clean Transportation (ICCT) for providing grant funding to AAL to catalyze the trial. We would also like to thank Greg Thorstensen, Daniel Royal III, and Derek Voege, dispatchers who released many contrail flights and provided us with valuable insights about the opportunities and challenges of contrail avoidance; John Dudley, Deborah Hecker, Rondeau Flynn, and Kevin Macelhaney, pilots who supported the development and execution of our trial; Kevin Johle, who provided critical technical support for the operation of the trial; and Ethan Klapper, Chris D'Aloia, Ben Halle, and Thomas Adler for reviewing the manuscript. The authors of this work used Google Gemini AI to draft parts of the manuscript, which were then reviewed and heavily edited by the authors.</p></ack><notes notes-type="reviewstatement"><title>Review statement</title>

      <p id="d2e4762">This paper was edited by Simon Unterstrasser and reviewed by two anonymous referees.</p>
  </notes><ref-list>
    <title>References</title>

      <ref id="bib1.bibx1"><label>Agarwal et al.(2022)</label><mixed-citation>Agarwal, A., Meijer, V. R., Eastham, S. D., Speth, R. L., and Barrett, S. R.:  Reanalysis-driven simulations may overestimate persistent contrail formation  by 100 %–250 %, Environ. Res. Lett., 17, 014045,  <ext-link xlink:href="https://doi.org/10.1088/1748-9326/ac38d9" ext-link-type="DOI">10.1088/1748-9326/ac38d9</ext-link>, 2022.</mixed-citation></ref>
      <ref id="bib1.bibx2"><label>Akhtar Martínez et al.(2025)</label><mixed-citation>Akhtar Martínez, C., Eastham, S. D., and Jarrett, J. P.: Zero-dimensional contrail models could underpredict lifetime optical depth, Atmos. Chem. Phys., 25, 12875–12891, <ext-link xlink:href="https://doi.org/10.5194/acp-25-12875-2025" ext-link-type="DOI">10.5194/acp-25-12875-2025</ext-link>, 2025.</mixed-citation></ref>
      <ref id="bib1.bibx3"><label>Bogacki and Shampine(1989)</label><mixed-citation>Bogacki, P. and Shampine, L. F.: A 3 (2) pair of Runge-Kutta formulas, Appl. Math. Lett., 2, 321–325, <ext-link xlink:href="https://doi.org/10.1016/0893-9659(89)90079-7" ext-link-type="DOI">10.1016/0893-9659(89)90079-7</ext-link>, 1989.</mixed-citation></ref>
      <ref id="bib1.bibx4"><label>Borella et al.(2024)</label><mixed-citation>Borella, A., Boucher, O., Shine, K. P., Stettler, M., Tanaka, K., Teoh, R., and Bellouin, N.: The importance of an informed choice of CO<sub>2</sub>-equivalence metrics for contrail avoidance, Atmos. Chem. Phys., 24, 9401–9417, <ext-link xlink:href="https://doi.org/10.5194/acp-24-9401-2024" ext-link-type="DOI">10.5194/acp-24-9401-2024</ext-link>, 2024.</mixed-citation></ref>
      <ref id="bib1.bibx5"><label>Burkhardt et al.(2018)</label><mixed-citation>Burkhardt, U., Bock, L., and Bier, A.: Mitigating the contrail cirrus climate impact by reducing aircraft soot number emissions, npj Climate and Atmospheric Science, 1, 37, <ext-link xlink:href="https://doi.org/10.1038/s41612-018-0046-4" ext-link-type="DOI">10.1038/s41612-018-0046-4</ext-link>, 2018.</mixed-citation></ref>
      <ref id="bib1.bibx6"><label>Dean et al.(2025)</label><mixed-citation>Dean, T. R., Abbott, T. H., Engberg, Z., Masson, N., Teoh, R., Itcovitz, J. P., Stettler, M. E., and Shapiro, M. L.: Impact of forecast stability on navigational contrail avoidance, Environmental Research: Infrastructure and Sustainability, 5, 045 008, <ext-link xlink:href="https://doi.org/10.1088/2634-4505/ae1da5" ext-link-type="DOI">10.1088/2634-4505/ae1da5</ext-link>, 2025.</mixed-citation></ref>
      <ref id="bib1.bibx7"><label>Driver et al.(2025)</label><mixed-citation>Driver, O. G. A., Stettler, M. E. J., and Gryspeerdt, E.: Factors limiting contrail detection in satellite imagery, Atmos. Meas. Tech., 18, 1115–1134, <ext-link xlink:href="https://doi.org/10.5194/amt-18-1115-2025" ext-link-type="DOI">10.5194/amt-18-1115-2025</ext-link>, 2025.</mixed-citation></ref>
      <ref id="bib1.bibx8"><label>Euchenhofer et al.(2025)</label><mixed-citation>Euchenhofer, M. V., Prashanth, P., Parke, S. A., Eastham, S. D., and Waitz, I. A.: Contrail observation limitations using geostationary satellites, Geophys. Res. Lett., 52, e2025GL118386, <ext-link xlink:href="https://doi.org/10.1029/2025GL118386" ext-link-type="DOI">10.1029/2025GL118386</ext-link>, 2025.</mixed-citation></ref>
      <ref id="bib1.bibx9"><label>Frias et al.(2024)</label><mixed-citation>Frias, A. M., Shapiro, M. L., Engberg, Z., Zopp, R., Soler, M., and Stettler, M. E. J.: Feasibility of contrail avoidance in a commercial flight planning system: an operational analysis, Environmental Research: Infrastructure and Sustainability, 4, 015013, <ext-link xlink:href="https://doi.org/10.1088/2634-4505/ad310c" ext-link-type="DOI">10.1088/2634-4505/ad310c</ext-link>, 2024.</mixed-citation></ref>
      <ref id="bib1.bibx10"><label>Frias et al.(2025)</label><mixed-citation>Frias, A. M., Soler, M., Dean, T., and Zopp, R.: Evaluating Soft and Hard Contrail Avoidance Strategies in Commercial Flight Planning: Implementation and Effectiveness Analysis,  SSRN 6341458, <ext-link xlink:href="https://doi.org/10.1016/j.jairtraman.2026.103089" ext-link-type="DOI">10.1016/j.jairtraman.2026.103089</ext-link>, 2025.</mixed-citation></ref>
      <ref id="bib1.bibx11"><label>Geraedts et al.(2026)</label><mixed-citation>Geraedts, S., Sarna, A., Rohs, S., Teoh, R., and McCloskey, K.: Improving reanalysis weather for contrail validation by incorporating satellite observations, J. Env. Com. Air Transp. Sys., 1, 2, <ext-link xlink:href="https://doi.org/10.5194/jecats-1-2-2026" ext-link-type="DOI">10.5194/jecats-1-2-2026</ext-link>, 2026.</mixed-citation></ref>
      <ref id="bib1.bibx12"><label>Gierens et al.(2020)</label><mixed-citation>Gierens, K., Matthes, S., and Rohs, S.: How well can persistent contrails be predicted?, Aerospace, 7, 169, <ext-link xlink:href="https://doi.org/10.3390/aerospace7120169" ext-link-type="DOI">10.3390/aerospace7120169</ext-link>, 2020.</mixed-citation></ref>
      <ref id="bib1.bibx13"><label>Goodman et al.(2019)</label><mixed-citation>Goodman, S. J., Schmit, T. J., Daniels, J., and Redmon, R. J.: The GOES-R series: a new generation of geostationary environmental satellites, Elsevier, <ext-link xlink:href="https://doi.org/10.1016/C2015-0-06249-9" ext-link-type="DOI">10.1016/C2015-0-06249-9</ext-link>, 2019.</mixed-citation></ref>
      <ref id="bib1.bibx14"><label>Gryspeerdt et al.(2024)</label><mixed-citation>Gryspeerdt, E., Stettler, M. E. J., Teoh, R., Burkhardt, U., Delovski, T., Driver, O. G. A., and Painemal, D.: Operational differences lead to longer lifetimes of satellite detectable contrails from more fuel efficient aircraft, Environ. Res. Lett., 19, 084059, <ext-link xlink:href="https://doi.org/10.1088/1748-9326/ad5b78" ext-link-type="DOI">10.1088/1748-9326/ad5b78</ext-link>, 2024.</mixed-citation></ref>
      <ref id="bib1.bibx15"><label>Hersbach et al.(2020)</label><mixed-citation>Hersbach, H., Bell, B., Berrisford, P., Hirahara, S., Horányi, A., Muñoz-Sabater, J., Nicolas, J., Peubey, C., Radu, R., Schepers, D., Simmons, A., Soci, C., Abdalla, S., Abellan, X., Balsamo, G., Bechtold, P., Biavati, G., Bidlot, J., Bonavita, M., De Chiara, G., Dahlgren, P., Dee, D., Diamantakis, M., Dragani, R., Flemming, J., Forbes, R., Fuentes, M., Geer, A., Haimberger, L., Healy, S., Hogan, R. J., Hólm, E., Janisková, M., Keeley, S., Laloyaux, P., Lopez, P., Lupu, C., Radnoti, G., de Rosnay, P., Rozum, I., Vamborg, F., Villaume, S., and Thépaut, J.-N.: The ERA5 global reanalysis, Q. J. Roy. Meteor. Soc., 146, 1999–2049, <ext-link xlink:href="https://doi.org/10.1002/qj.3803" ext-link-type="DOI">10.1002/qj.3803</ext-link>, 2020.</mixed-citation></ref>
      <ref id="bib1.bibx16"><label>Hersbach et al.(2023)</label><mixed-citation>Hersbach, H., Bell, B., Berrisford, P., Biavati, G., Horányi, A., Muñoz Sabater, J., Nicolas, J., Peubey, C., Radu, R., Rozum, I., Schepers, D., Simmons, A., Soci, C., Dee, D., and Thépaut, J.-N.: ERA5 hourly data on pressure levels from 1940 to present, Copernicus Climate Change Service (C3S) Climate Data Store (CDS) [data set], <ext-link xlink:href="https://doi.org/10.24381/cds.bd0915c6" ext-link-type="DOI">10.24381/cds.bd0915c6</ext-link>, 2023.</mixed-citation></ref>
      <ref id="bib1.bibx17"><label>Hodgson and Abbott(2026)</label><mixed-citation>Hodgson, P. and Abbott, T.: Introducing ContrailBench: Measuring contrail  forecasts against reality, Contrails.org Notebook,  <uri>https://notebook.contrails.org/introducing-contrailbench/</uri>  (last access: 15 July 2026), 2026.</mixed-citation></ref>
      <ref id="bib1.bibx18"><label>Holt and Sullivan(2023)</label><mixed-citation>Holt, C. A. and Sullivan, S. P.: Permutation tests for experimental data, Exp. Econ., 26, 775–812, <ext-link xlink:href="https://doi.org/10.1007/s10683-023-09799-6" ext-link-type="DOI">10.1007/s10683-023-09799-6</ext-link>, 2023.</mixed-citation></ref>
      <ref id="bib1.bibx19"><label>IPCC(2021)</label><mixed-citation>IPCC: Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change, Cambridge University Press, Cambridge, UK and New York, NY, USA, <ext-link xlink:href="https://doi.org/10.1017/9781009157896" ext-link-type="DOI">10.1017/9781009157896</ext-link>, 2021.</mixed-citation></ref>
      <ref id="bib1.bibx20"><label>Jeßberger et al.(2013)</label><mixed-citation>Jeßberger, P., Voigt, C., Schumann, U., Sölch, I., Schlager, H., Kaufmann, S., Petzold, A., Schäuble, D., and Gayet, J.-F.: Aircraft type influence on contrail properties, Atmos. Chem. Phys., 13, 11965–11984, <ext-link xlink:href="https://doi.org/10.5194/acp-13-11965-2013" ext-link-type="DOI">10.5194/acp-13-11965-2013</ext-link>, 2013.</mixed-citation></ref>
      <ref id="bib1.bibx21"><label>Kärcher(2018)</label><mixed-citation>Kärcher, B.: Formation and radiative forcing of contrail cirrus, Nat. Commun., 9, 1824, <ext-link xlink:href="https://doi.org/10.1038/s41467-018-04068-0" ext-link-type="DOI">10.1038/s41467-018-04068-0</ext-link>, 2018.</mixed-citation></ref>
      <ref id="bib1.bibx22"><label>Lee et al.(2021)</label><mixed-citation>Lee, D., Fahey, D., Skowron, A., Allen, M., Burkhardt, U., Chen, Q., Doherty, S., Freeman, S., Forster, P., Fuglestvedt, J., Gettelman, A., De León, R., Lim, L., Lund, M., Millar, R., Owen, B., Penner, J., Pitari, G., Prather, M., Sausen, R., and Wilcox, L.: The contribution of global aviation to anthropogenic climate forcing for 2000 to 2018, Atmos. Environ., 244, 117 834, <ext-link xlink:href="https://doi.org/10.1016/j.atmosenv.2020.117834" ext-link-type="DOI">10.1016/j.atmosenv.2020.117834</ext-link>, 2021.</mixed-citation></ref>
      <ref id="bib1.bibx23"><label>Lee et al.(2023)</label><mixed-citation>Lee, D. S., Allen, M. R., Cumpsty, N., Owen, B., Shine, K. P., and Skowron, A.: Uncertainties in mitigating aviation non-CO<sub>2</sub> emissions for climate and air quality using hydrocarbon fuels, Environmental Science: Atmospheres, 3, 1693–1740, <ext-link xlink:href="https://doi.org/10.1039/d3ea00091e" ext-link-type="DOI">10.1039/d3ea00091e</ext-link>, 2023.</mixed-citation></ref>
      <ref id="bib1.bibx24"><label>McCoy(2017)</label><mixed-citation>McCoy, C. E.: Understanding the intention-to-treat principle in randomized controlled trials, Western Journal of Emergency Medicine, 18, 1075, <ext-link xlink:href="https://doi.org/10.5811/westjem.2017.8.35985" ext-link-type="DOI">10.5811/westjem.2017.8.35985</ext-link>, 2017.</mixed-citation></ref>
      <ref id="bib1.bibx25"><label>Megill et al.(2024)</label><mixed-citation>Megill, L., Deck, K., and Grewe, V.: Alternative climate metrics to the Global Warming Potential are more suitable for assessing aviation non-CO<sub>2</sub> effects, Commun. Earth Environ., 5, 249, <ext-link xlink:href="https://doi.org/10.1038/s43247-024-01423-6" ext-link-type="DOI">10.1038/s43247-024-01423-6</ext-link>, 2024.</mixed-citation></ref>
      <ref id="bib1.bibx26"><label>Meijer et al.(2024)</label><mixed-citation>Meijer, V. R., Eastham, S. D., Waitz, I. A., and Barrett, S. R. H.: Contrail altitude estimation using GOES-16 ABI data and deep learning, Atmos. Meas. Tech., 17, 6145–6162, <ext-link xlink:href="https://doi.org/10.5194/amt-17-6145-2024" ext-link-type="DOI">10.5194/amt-17-6145-2024</ext-link>, 2024.</mixed-citation></ref>
      <ref id="bib1.bibx27"><label>Ng et al.(2023)</label><mixed-citation>Ng, J. Y.-H., McCloskey, K., Cui, J., Meijer, V. R., Brand, E., Sarna, A., Goyal, N., Van Arsdale, C., and Geraedts, S.: Contrail Detection on GOES-16 ABI With the OpenContrails Dataset, IEEE T. Geosci. Remote, 62, <ext-link xlink:href="https://doi.org/10.1109/TGRS.2023.3345226" ext-link-type="DOI">10.1109/TGRS.2023.3345226</ext-link>, 2023.</mixed-citation></ref>
      <ref id="bib1.bibx28"><label>Pesarin and Salmaso(1993)</label><mixed-citation>Pesarin, F. and Salmaso, L.: The permutation testing approach: a review, STATISTICA, 70, 481–509, <ext-link xlink:href="https://doi.org/10.6092/issn.1973-2201/3599" ext-link-type="DOI">10.6092/issn.1973-2201/3599</ext-link>, 1993.</mixed-citation></ref>
      <ref id="bib1.bibx29"><label>Petzold et al.(2025)</label><mixed-citation>Petzold, A., Khan, N. F., Li, Y., Spichtinger, P., Rohs, S., Crewell, S., Wahner, A., and Krämer, M.: Most long-lived contrails form within cirrus clouds with uncertain climate impact, Nat. Commun., 16, 9695, <ext-link xlink:href="https://doi.org/10.1038/s41467-025-65532-2" ext-link-type="DOI">10.1038/s41467-025-65532-2</ext-link>, 2025.</mixed-citation></ref>
      <ref id="bib1.bibx30"><label>Platt et al.(2024)</label><mixed-citation>Platt, J. C., Shapiro, M. L., Engberg, Z., McCloskey, K., Geraedts, S., Sankar, T., Stettler, M. E. J., Teoh, R., Schumann, U., Rohs, S., Brand, E., and Van Arsdale, C.: The effect of uncertainty in humidity and model parameters on the prediction of contrail energy forcing, Environ. Res. Commun., 6, 095015, <ext-link xlink:href="https://doi.org/10.1088/2515-7620/ad6ee5" ext-link-type="DOI">10.1088/2515-7620/ad6ee5</ext-link>, 2024.</mixed-citation></ref>
      <ref id="bib1.bibx31"><label>Rubin(2008)</label><mixed-citation>Rubin, D. B.: For objective causal inference, design trumps analysis, Ann. Appl. Stat., 2, 808–840, <ext-link xlink:href="https://doi.org/10.1214/08-AOAS187" ext-link-type="DOI">10.1214/08-AOAS187</ext-link>, 2008.</mixed-citation></ref>
      <ref id="bib1.bibx32"><label>Sankar and Geraedts(2026)</label><mixed-citation>Sankar, T. and Geraedts, S.: Efficacy of Scalable Airline-led Contrail Avoidance (Supplemental), Zenodo [code/data set], <ext-link xlink:href="https://doi.org/10.5281/zenodo.19473497" ext-link-type="DOI">10.5281/zenodo.19473497</ext-link>, 2026.</mixed-citation></ref>
      <ref id="bib1.bibx33"><label>Sarna et al.(2025)</label><mixed-citation>Sarna, A., Meijer, V., Chevallier, R., Duncan, A., McConnaughay, K., Geraedts, S., and McCloskey, K.: Benchmarking and improving algorithms for attributing satellite-observed contrails to flights, Atmos. Meas. Tech., 18, 3495–3532, <ext-link xlink:href="https://doi.org/10.5194/amt-18-3495-2025" ext-link-type="DOI">10.5194/amt-18-3495-2025</ext-link>, 2025.</mixed-citation></ref>
      <ref id="bib1.bibx34"><label>Sausen et al.(2023)</label><mixed-citation>Sausen, R., Hofer, S. M., Gierens, K. M., Bugliaro Goggia, L., Ehrmanntraut, R., Sitova, I., Walczak, K., Burridge-Diesing, A., Bowman, M., and Miller, N.: Can we successfully avoid persistent contrails by small altitude adjustments of flights in the real world?, Meteorol. Z., <ext-link xlink:href="https://doi.org/10.1127/metz/2023/1155" ext-link-type="DOI">10.1127/metz/2023/1155</ext-link>, 2023.</mixed-citation></ref>
      <ref id="bib1.bibx35"><label>Schumann(1996)</label><mixed-citation>Schumann, U.: On conditions for contrail formation from aircraft exhausts, Meteorol. Z., 5, 4–23, <ext-link xlink:href="https://doi.org/10.1127/metz/5/1996/4" ext-link-type="DOI">10.1127/metz/5/1996/4</ext-link>, 1996.</mixed-citation></ref>
      <ref id="bib1.bibx36"><label>Schumann(2012)</label><mixed-citation>Schumann, U.: A contrail cirrus prediction model, Geosci. Model Dev., 5, 543–580, <ext-link xlink:href="https://doi.org/10.5194/gmd-5-543-2012" ext-link-type="DOI">10.5194/gmd-5-543-2012</ext-link>, 2012.</mixed-citation></ref>
      <ref id="bib1.bibx37"><label>Schumann et al.(2015)</label><mixed-citation>Schumann, U., Penner, J. E., Chen, Y., Zhou, C., and Graf, K.: Dehydration effects from contrails in a coupled contrail–climate model, Atmos. Chem. Phys., 15, 11179–11199, <ext-link xlink:href="https://doi.org/10.5194/acp-15-11179-2015" ext-link-type="DOI">10.5194/acp-15-11179-2015</ext-link>, 2015. </mixed-citation></ref>
      <ref id="bib1.bibx38"><label>Sonabend-W et al.(2024)</label><mixed-citation>Sonabend-W, A., Elkin, C., Dean, T., Dudley, J., Ali, N., Blickstein, J., Brand, E., Broshears, B., Chen, S., Engberg, Z., Galyen, M., Geraedts, S., Goyal, N., Grenham, R., Hager, U., Hecker, D., Jany, M., McCloskey, K., Ng, J., Norris, B., Opel, F., Rothenberg, J., Sankar, T., Sanekommu, D., Sarna, A., Schütt, O., Shapiro, M., Soh, R., Van Arsdale, C., and Platt, J. C.: Feasibility test of per-flight contrail avoidance in commercial aviation, Commun. Eng., 3, 184, <ext-link xlink:href="https://doi.org/10.1038/s44172-024-00329-7" ext-link-type="DOI">10.1038/s44172-024-00329-7</ext-link>, 2024.</mixed-citation></ref>
      <ref id="bib1.bibx39"><label>Teoh et al.(2020)</label><mixed-citation>Teoh, R., Schumann, U., Majumdar, A., and Stettler, M. E.: Mitigating the climate forcing of aircraft contrails by small-scale diversions and technology adoption, Environ. Sci. Technol., 54, 2941–2950, <ext-link xlink:href="https://doi.org/10.1021/acs.est.9b05608" ext-link-type="DOI">10.1021/acs.est.9b05608</ext-link>, 2020.</mixed-citation></ref>
      <ref id="bib1.bibx40"><label>Unterstrasser and Görsch(2014)</label><mixed-citation>Unterstrasser, S. and Görsch, N.: Aircraft-type dependency of contrail evolution, J. Geophys. Res.-Atmos., 119, 14015–14027, <ext-link xlink:href="https://doi.org/10.1002/2014JD022642" ext-link-type="DOI">10.1002/2014JD022642</ext-link>, 2014.</mixed-citation></ref>
      <ref id="bib1.bibx41"><label>Wang et al.(2025)</label><mixed-citation>Wang, Z., Bugliaro, L., Gierens, K., Hegglin, M. I., Rohs, S., Petzold, A., Kaufmann, S., and Voigt, C.: Machine learning for improvement of upper-tropospheric relative humidity in ERA5 weather model data, Atmos. Chem. Phys., 25, 2845–2861, <ext-link xlink:href="https://doi.org/10.5194/acp-25-2845-2025" ext-link-type="DOI">10.5194/acp-25-2845-2025</ext-link>, 2025.</mixed-citation></ref>

  </ref-list></back>
    <!--<article-title-html>Efficacy of scalable airline-led contrail avoidance</article-title-html>
<abstract-html/>
<ref-html id="bib1.bib1"><label>Agarwal et al.(2022)</label><mixed-citation>
      
Agarwal, A., Meijer, V. R., Eastham, S. D., Speth, R. L., and Barrett, S. R.:  Reanalysis-driven simulations may overestimate persistent contrail formation  by 100&thinsp;%–250&thinsp;%, Environ. Res. Lett., 17, 014045,  <a href="https://doi.org/10.1088/1748-9326/ac38d9" target="_blank">https://doi.org/10.1088/1748-9326/ac38d9</a>, 2022.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib2"><label>Akhtar Martínez et al.(2025)</label><mixed-citation>
      
Akhtar Martínez, C., Eastham, S. D., and Jarrett, J. P.: Zero-dimensional contrail models could underpredict lifetime optical depth, Atmos. Chem. Phys., 25, 12875–12891, <a href="https://doi.org/10.5194/acp-25-12875-2025" target="_blank">https://doi.org/10.5194/acp-25-12875-2025</a>, 2025.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib3"><label>Bogacki and Shampine(1989)</label><mixed-citation>
      
Bogacki, P. and Shampine, L. F.: A 3 (2) pair of Runge-Kutta formulas, Appl.
Math. Lett., 2, 321–325, <a href="https://doi.org/10.1016/0893-9659(89)90079-7" target="_blank">https://doi.org/10.1016/0893-9659(89)90079-7</a>, 1989.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib4"><label>Borella et al.(2024)</label><mixed-citation>
      
Borella, A., Boucher, O., Shine, K. P., Stettler, M., Tanaka, K., Teoh, R., and Bellouin, N.: The importance of an informed choice of CO<sub>2</sub>-equivalence metrics for contrail avoidance, Atmos. Chem. Phys., 24, 9401–9417, <a href="https://doi.org/10.5194/acp-24-9401-2024" target="_blank">https://doi.org/10.5194/acp-24-9401-2024</a>, 2024.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib5"><label>Burkhardt et al.(2018)</label><mixed-citation>
      
Burkhardt, U., Bock, L., and Bier, A.: Mitigating the contrail cirrus climate
impact by reducing aircraft soot number emissions, npj Climate and
Atmospheric Science, 1, 37, <a href="https://doi.org/10.1038/s41612-018-0046-4" target="_blank">https://doi.org/10.1038/s41612-018-0046-4</a>, 2018.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib6"><label>Dean et al.(2025)</label><mixed-citation>
      
Dean, T. R., Abbott, T. H., Engberg, Z., Masson, N., Teoh, R., Itcovitz, J. P.,
Stettler, M. E., and Shapiro, M. L.: Impact of forecast stability on
navigational contrail avoidance, Environmental Research: Infrastructure and
Sustainability, 5, 045&thinsp;008, <a href="https://doi.org/10.1088/2634-4505/ae1da5" target="_blank">https://doi.org/10.1088/2634-4505/ae1da5</a>, 2025.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib7"><label>Driver et al.(2025)</label><mixed-citation>
      
Driver, O. G. A., Stettler, M. E. J., and Gryspeerdt, E.: Factors limiting contrail detection in satellite imagery, Atmos. Meas. Tech., 18, 1115–1134, <a href="https://doi.org/10.5194/amt-18-1115-2025" target="_blank">https://doi.org/10.5194/amt-18-1115-2025</a>, 2025.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib8"><label>Euchenhofer et al.(2025)</label><mixed-citation>
      
Euchenhofer, M. V., Prashanth, P., Parke, S. A., Eastham, S. D., and Waitz,
I. A.: Contrail observation limitations using geostationary satellites,
Geophys. Res. Lett., 52, e2025GL118386,
<a href="https://doi.org/10.1029/2025GL118386" target="_blank">https://doi.org/10.1029/2025GL118386</a>, 2025.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib9"><label>Frias et al.(2024)</label><mixed-citation>
      
Frias, A. M., Shapiro, M. L., Engberg, Z., Zopp, R., Soler, M., and Stettler,
M. E. J.: Feasibility of contrail avoidance in a commercial flight planning
system: an operational analysis, Environmental Research: Infrastructure and
Sustainability, 4, 015013, <a href="https://doi.org/10.1088/2634-4505/ad310c" target="_blank">https://doi.org/10.1088/2634-4505/ad310c</a>, 2024.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib10"><label>Frias et al.(2025)</label><mixed-citation>
      
Frias, A. M., Soler, M., Dean, T., and Zopp, R.: Evaluating Soft and Hard
Contrail Avoidance Strategies in Commercial Flight Planning: Implementation
and Effectiveness Analysis,  SSRN 6341458,
<a href="https://doi.org/10.1016/j.jairtraman.2026.103089" target="_blank">https://doi.org/10.1016/j.jairtraman.2026.103089</a>, 2025.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib11"><label>Geraedts et al.(2026)</label><mixed-citation>
      
Geraedts, S., Sarna, A., Rohs, S., Teoh, R., and McCloskey, K.: Improving reanalysis weather for contrail validation by incorporating satellite observations, J. Env. Com. Air Transp. Sys., 1, 2, <a href="https://doi.org/10.5194/jecats-1-2-2026" target="_blank">https://doi.org/10.5194/jecats-1-2-2026</a>, 2026.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib12"><label>Gierens et al.(2020)</label><mixed-citation>
      
Gierens, K., Matthes, S., and Rohs, S.: How well can persistent contrails be
predicted?, Aerospace, 7, 169, <a href="https://doi.org/10.3390/aerospace7120169" target="_blank">https://doi.org/10.3390/aerospace7120169</a>, 2020.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib13"><label>Goodman et al.(2019)</label><mixed-citation>
      
Goodman, S. J., Schmit, T. J., Daniels, J., and Redmon, R. J.: The GOES-R
series: a new generation of geostationary environmental satellites, Elsevier,
<a href="https://doi.org/10.1016/C2015-0-06249-9" target="_blank">https://doi.org/10.1016/C2015-0-06249-9</a>, 2019.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib14"><label>Gryspeerdt et al.(2024)</label><mixed-citation>
      
Gryspeerdt, E., Stettler, M. E. J., Teoh, R., Burkhardt, U., Delovski, T.,
Driver, O. G. A., and Painemal, D.: Operational differences lead to longer
lifetimes of satellite detectable contrails from more fuel efficient
aircraft, Environ. Res. Lett., 19, 084059,
<a href="https://doi.org/10.1088/1748-9326/ad5b78" target="_blank">https://doi.org/10.1088/1748-9326/ad5b78</a>, 2024.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib15"><label>Hersbach et al.(2020)</label><mixed-citation>
      
Hersbach, H., Bell, B., Berrisford, P., Hirahara, S., Horányi, A.,
Muñoz-Sabater, J., Nicolas, J., Peubey, C., Radu, R., Schepers, D., Simmons,
A., Soci, C., Abdalla, S., Abellan, X., Balsamo, G., Bechtold, P., Biavati,
G., Bidlot, J., Bonavita, M., De Chiara, G., Dahlgren, P., Dee, D.,
Diamantakis, M., Dragani, R., Flemming, J., Forbes, R., Fuentes, M., Geer,
A., Haimberger, L., Healy, S., Hogan, R. J., Hólm, E., Janisková, M.,
Keeley, S., Laloyaux, P., Lopez, P., Lupu, C., Radnoti, G., de Rosnay, P.,
Rozum, I., Vamborg, F., Villaume, S., and Thépaut, J.-N.: The ERA5 global
reanalysis, Q. J. Roy. Meteor. Soc., 146,
1999–2049, <a href="https://doi.org/10.1002/qj.3803" target="_blank">https://doi.org/10.1002/qj.3803</a>, 2020.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib16"><label>Hersbach et al.(2023)</label><mixed-citation>
      
Hersbach, H., Bell, B., Berrisford, P., Biavati, G., Horányi, A., Muñoz Sabater, J., Nicolas, J., Peubey, C., Radu, R., Rozum, I., Schepers, D., Simmons, A., Soci, C., Dee, D., and Thépaut, J.-N.: ERA5 hourly data on pressure levels from 1940 to present, Copernicus Climate Change Service (C3S) Climate Data Store (CDS) [data set], <a href="https://doi.org/10.24381/cds.bd0915c6" target="_blank">https://doi.org/10.24381/cds.bd0915c6</a>, 2023.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib17"><label>Hodgson and Abbott(2026)</label><mixed-citation>
      
Hodgson, P. and Abbott, T.: Introducing ContrailBench: Measuring contrail  forecasts against reality, Contrails.org Notebook,  <a href="https://notebook.contrails.org/introducing-contrailbench/" target="_blank"/>  (last access: 15 July 2026), 2026.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib18"><label>Holt and Sullivan(2023)</label><mixed-citation>
      
Holt, C. A. and Sullivan, S. P.: Permutation tests for experimental data,
Exp. Econ., 26, 775–812, <a href="https://doi.org/10.1007/s10683-023-09799-6" target="_blank">https://doi.org/10.1007/s10683-023-09799-6</a>, 2023.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib19"><label>IPCC(2021)</label><mixed-citation>
      
IPCC: Climate Change 2021: The Physical Science Basis. Contribution of Working
Group I to the Sixth Assessment Report of the Intergovernmental Panel on
Climate Change, Cambridge University Press, Cambridge, UK and New York, NY,
USA, <a href="https://doi.org/10.1017/9781009157896" target="_blank">https://doi.org/10.1017/9781009157896</a>, 2021.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib20"><label>Jeßberger et al.(2013)</label><mixed-citation>
      
Jeßberger, P., Voigt, C., Schumann, U., Sölch, I., Schlager, H., Kaufmann, S., Petzold, A., Schäuble, D., and Gayet, J.-F.: Aircraft type influence on contrail properties, Atmos. Chem. Phys., 13, 11965–11984, <a href="https://doi.org/10.5194/acp-13-11965-2013" target="_blank">https://doi.org/10.5194/acp-13-11965-2013</a>, 2013.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib21"><label>Kärcher(2018)</label><mixed-citation>
      
Kärcher, B.: Formation and radiative forcing of contrail cirrus, Nat.
Commun., 9, 1824, <a href="https://doi.org/10.1038/s41467-018-04068-0" target="_blank">https://doi.org/10.1038/s41467-018-04068-0</a>, 2018.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib22"><label>Lee et al.(2021)</label><mixed-citation>
      
Lee, D., Fahey, D., Skowron, A., Allen, M., Burkhardt, U., Chen, Q., Doherty,
S., Freeman, S., Forster, P., Fuglestvedt, J., Gettelman, A., De León, R.,
Lim, L., Lund, M., Millar, R., Owen, B., Penner, J., Pitari, G., Prather, M.,
Sausen, R., and Wilcox, L.: The contribution of global aviation to
anthropogenic climate forcing for 2000 to 2018, Atmos. Environ., 244,
117&thinsp;834, <a href="https://doi.org/10.1016/j.atmosenv.2020.117834" target="_blank">https://doi.org/10.1016/j.atmosenv.2020.117834</a>, 2021.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib23"><label>Lee et al.(2023)</label><mixed-citation>
      
Lee, D. S., Allen, M. R., Cumpsty, N., Owen, B., Shine, K. P., and Skowron, A.:
Uncertainties in mitigating aviation non-CO<sub>2</sub> emissions for climate and air
quality using hydrocarbon fuels, Environmental Science: Atmospheres, 3,
1693–1740, <a href="https://doi.org/10.1039/d3ea00091e" target="_blank">https://doi.org/10.1039/d3ea00091e</a>, 2023.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib24"><label>McCoy(2017)</label><mixed-citation>
      
McCoy, C. E.: Understanding the intention-to-treat principle in randomized
controlled trials, Western Journal of Emergency Medicine, 18, 1075,
<a href="https://doi.org/10.5811/westjem.2017.8.35985" target="_blank">https://doi.org/10.5811/westjem.2017.8.35985</a>, 2017.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib25"><label>Megill et al.(2024)</label><mixed-citation>
      
Megill, L., Deck, K., and Grewe, V.: Alternative climate metrics to the Global
Warming Potential are more suitable for assessing aviation non-CO<sub>2</sub> effects,
Commun. Earth Environ., 5, 249,
<a href="https://doi.org/10.1038/s43247-024-01423-6" target="_blank">https://doi.org/10.1038/s43247-024-01423-6</a>, 2024.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib26"><label>Meijer et al.(2024)</label><mixed-citation>
      
Meijer, V. R., Eastham, S. D., Waitz, I. A., and Barrett, S. R. H.: Contrail altitude estimation using GOES-16 ABI data and deep learning, Atmos. Meas. Tech., 17, 6145–6162, <a href="https://doi.org/10.5194/amt-17-6145-2024" target="_blank">https://doi.org/10.5194/amt-17-6145-2024</a>, 2024.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib27"><label>Ng et al.(2023)</label><mixed-citation>
      
Ng, J. Y.-H., McCloskey, K., Cui, J., Meijer, V. R., Brand, E., Sarna, A.,
Goyal, N., Van Arsdale, C., and Geraedts, S.: Contrail Detection on GOES-16
ABI With the OpenContrails Dataset, IEEE T. Geosci.
Remote, 62, <a href="https://doi.org/10.1109/TGRS.2023.3345226" target="_blank">https://doi.org/10.1109/TGRS.2023.3345226</a>, 2023.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib28"><label>Pesarin and Salmaso(1993)</label><mixed-citation>
      
Pesarin, F. and Salmaso, L.: The permutation testing approach: a review,
STATISTICA, 70, 481–509, <a href="https://doi.org/10.6092/issn.1973-2201/3599" target="_blank">https://doi.org/10.6092/issn.1973-2201/3599</a>, 1993.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib29"><label>Petzold et al.(2025)</label><mixed-citation>
      
Petzold, A., Khan, N. F., Li, Y., Spichtinger, P., Rohs, S., Crewell, S.,
Wahner, A., and Krämer, M.: Most long-lived contrails form within cirrus
clouds with uncertain climate impact, Nat. Commun., 16, 9695,
<a href="https://doi.org/10.1038/s41467-025-65532-2" target="_blank">https://doi.org/10.1038/s41467-025-65532-2</a>, 2025.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib30"><label>Platt et al.(2024)</label><mixed-citation>
      
Platt, J. C., Shapiro, M. L., Engberg, Z., McCloskey, K., Geraedts, S., Sankar,
T., Stettler, M. E. J., Teoh, R., Schumann, U., Rohs, S., Brand, E., and
Van Arsdale, C.: The effect of uncertainty in humidity and model parameters
on the prediction of contrail energy forcing, Environ. Res.
Commun., 6, 095015, <a href="https://doi.org/10.1088/2515-7620/ad6ee5" target="_blank">https://doi.org/10.1088/2515-7620/ad6ee5</a>, 2024.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib31"><label>Rubin(2008)</label><mixed-citation>
      
Rubin, D. B.: For objective causal inference, design trumps analysis, Ann. Appl. Stat., 2, 808–840, <a href="https://doi.org/10.1214/08-AOAS187" target="_blank">https://doi.org/10.1214/08-AOAS187</a>, 2008.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib32"><label>Sankar and Geraedts(2026)</label><mixed-citation>
      
Sankar, T. and Geraedts, S.: Efficacy of Scalable Airline-led Contrail Avoidance (Supplemental), Zenodo [code/data set], <a href="https://doi.org/10.5281/zenodo.19473497" target="_blank">https://doi.org/10.5281/zenodo.19473497</a>, 2026.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib33"><label>Sarna et al.(2025)</label><mixed-citation>
      
Sarna, A., Meijer, V., Chevallier, R., Duncan, A., McConnaughay, K., Geraedts, S., and McCloskey, K.: Benchmarking and improving algorithms for attributing satellite-observed contrails to flights, Atmos. Meas. Tech., 18, 3495–3532, <a href="https://doi.org/10.5194/amt-18-3495-2025" target="_blank">https://doi.org/10.5194/amt-18-3495-2025</a>, 2025.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib34"><label>Sausen et al.(2023)</label><mixed-citation>
      
Sausen, R., Hofer, S. M., Gierens, K. M., Bugliaro Goggia, L., Ehrmanntraut,
R., Sitova, I., Walczak, K., Burridge-Diesing, A., Bowman, M., and Miller,
N.: Can we successfully avoid persistent contrails by small altitude
adjustments of flights in the real world?, Meteorol. Z.,
<a href="https://doi.org/10.1127/metz/2023/1155" target="_blank">https://doi.org/10.1127/metz/2023/1155</a>, 2023.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib35"><label>Schumann(1996)</label><mixed-citation>
      
Schumann, U.: On conditions for contrail formation from aircraft exhausts,
Meteorol. Z., 5, 4–23, <a href="https://doi.org/10.1127/metz/5/1996/4" target="_blank">https://doi.org/10.1127/metz/5/1996/4</a>, 1996.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib36"><label>Schumann(2012)</label><mixed-citation>
      
Schumann, U.: A contrail cirrus prediction model, Geosci. Model Dev., 5, 543–580, <a href="https://doi.org/10.5194/gmd-5-543-2012" target="_blank">https://doi.org/10.5194/gmd-5-543-2012</a>, 2012.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib37"><label>Schumann et al.(2015)</label><mixed-citation>
      
Schumann, U., Penner, J. E., Chen, Y., Zhou, C., and Graf, K.: Dehydration effects from contrails in a coupled contrail–climate model, Atmos. Chem. Phys., 15, 11179–11199, <a href="https://doi.org/10.5194/acp-15-11179-2015" target="_blank">https://doi.org/10.5194/acp-15-11179-2015</a>, 2015.


    </mixed-citation></ref-html>
<ref-html id="bib1.bib38"><label>Sonabend-W et al.(2024)</label><mixed-citation>
      
Sonabend-W, A., Elkin, C., Dean, T., Dudley, J., Ali, N., Blickstein, J.,
Brand, E., Broshears, B., Chen, S., Engberg, Z., Galyen, M., Geraedts, S.,
Goyal, N., Grenham, R., Hager, U., Hecker, D., Jany, M., McCloskey, K., Ng,
J., Norris, B., Opel, F., Rothenberg, J., Sankar, T., Sanekommu, D., Sarna,
A., Schütt, O., Shapiro, M., Soh, R., Van Arsdale, C., and Platt, J. C.:
Feasibility test of per-flight contrail avoidance in commercial aviation,
Commun. Eng., 3, 184, <a href="https://doi.org/10.1038/s44172-024-00329-7" target="_blank">https://doi.org/10.1038/s44172-024-00329-7</a>, 2024.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib39"><label>Teoh et al.(2020)</label><mixed-citation>
      
Teoh, R., Schumann, U., Majumdar, A., and Stettler, M. E.: Mitigating the
climate forcing of aircraft contrails by small-scale diversions and
technology adoption, Environ. Sci. Technol., 54, 2941–2950,
<a href="https://doi.org/10.1021/acs.est.9b05608" target="_blank">https://doi.org/10.1021/acs.est.9b05608</a>, 2020.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib40"><label>Unterstrasser and Görsch(2014)</label><mixed-citation>
      
Unterstrasser, S. and Görsch, N.: Aircraft-type dependency of contrail
evolution, J. Geophys. Res.-Atmos., 119, 14015–14027,
<a href="https://doi.org/10.1002/2014JD022642" target="_blank">https://doi.org/10.1002/2014JD022642</a>, 2014.

    </mixed-citation></ref-html>
<ref-html id="bib1.bib41"><label>Wang et al.(2025)</label><mixed-citation>
      
Wang, Z., Bugliaro, L., Gierens, K., Hegglin, M. I., Rohs, S., Petzold, A., Kaufmann, S., and Voigt, C.: Machine learning for improvement of upper-tropospheric relative humidity in ERA5 weather model data, Atmos. Chem. Phys., 25, 2845–2861, <a href="https://doi.org/10.5194/acp-25-2845-2025" target="_blank">https://doi.org/10.5194/acp-25-2845-2025</a>, 2025.

    </mixed-citation></ref-html>--></article>
