<?xml version='1.0' encoding='utf-8'?>
<oai_dc:dc xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:oai_dc="http://www.openarchives.org/OAI/2.0/oai_dc/" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://www.openarchives.org/OAI/2.0/oai_dc/ http://www.openarchives.org/OAI/2.0/oai_dc.xsd">
  <dc:contributor>Kathryn Lawson</dc:contributor>
  <dc:contributor>Wenyu Ouyang</dc:contributor>
  <dc:contributor>Alison P. Appling</dc:contributor>
  <dc:contributor>Samantha K. Oliver</dc:contributor>
  <dc:contributor>Chaopeng Shen</dc:contributor>
  <dc:creator>Farshid Rahmani</dc:creator>
  <dc:date>2021</dc:date>
  <dc:description>&lt;div class="article-text wd-jnl-art-abstract cf"&gt;&lt;p&gt;Stream water temperature (&lt;i&gt;T&lt;/i&gt;&lt;sub&gt;s&lt;/sub&gt;) is a variable of critical importance for aquatic ecosystem health.&lt;span&gt;&amp;nbsp;&lt;/span&gt;&lt;i&gt;T&lt;/i&gt;&lt;sub&gt;s&lt;/sub&gt;&lt;span&gt;&amp;nbsp;&lt;/span&gt;is strongly affected by groundwater-surface water interactions which can be learned from streamflow records, but previously such information was challenging to effectively absorb with process-based models due to parameter equifinality. Based on the long short-term memory (LSTM) deep learning architecture, we developed a basin-centric lumped daily mean&lt;span&gt;&amp;nbsp;&lt;/span&gt;&lt;i&gt;T&lt;/i&gt;&lt;sub&gt;s&lt;/sub&gt;&lt;span&gt;&amp;nbsp;&lt;/span&gt;model, which was trained over 118 data-rich basins with no major dams in the conterminous United States, and showed strong results. At a national scale, we obtained a median root-mean-square error of 0.69°C, Nash–Sutcliffe model efficiency coefficient of 0.985, and correlation of 0.994, which are marked improvements over previous values reported in literature. The addition of streamflow observations as a model input strongly elevated the performance of this model. In the absence of measured streamflow, we showed that a two-stage model could be used, where simulated streamflow from a pre-trained LSTM model (&lt;i&gt;Q&lt;/i&gt;&lt;sub&gt;sim&lt;/sub&gt;) still benefited the&lt;span&gt;&amp;nbsp;&lt;/span&gt;&lt;i&gt;T&lt;/i&gt;&lt;sub&gt;s&lt;/sub&gt;&lt;span&gt;&amp;nbsp;&lt;/span&gt;model even though no new information was brought directly into the inputs of the&lt;span&gt;&amp;nbsp;&lt;/span&gt;&lt;i&gt;T&lt;/i&gt;&lt;sub&gt;s&lt;/sub&gt;&lt;span&gt;&amp;nbsp;&lt;/span&gt;model. The model indirectly used information learned from streamflow observations provided during the training of&lt;span&gt;&amp;nbsp;&lt;/span&gt;&lt;i&gt;Q&lt;/i&gt;&lt;sub&gt;sim&lt;/sub&gt;, potentially to improve internal representation of physically meaningful variables. Our results indicate that strong relationships exist between basin-averaged forcing variables, catchment attributes, and&lt;span&gt;&amp;nbsp;&lt;/span&gt;&lt;i&gt;T&lt;/i&gt;&lt;sub&gt;s&lt;/sub&gt;&lt;span&gt;&amp;nbsp;&lt;/span&gt;that can be simulated by a single model trained by data on the continental scale.&lt;/p&gt;&lt;/div&gt;</dc:description>
  <dc:format>application/pdf</dc:format>
  <dc:identifier>10.1088/1748-9326/abd501</dc:identifier>
  <dc:language>en</dc:language>
  <dc:publisher>IOP Science</dc:publisher>
  <dc:title>Exploring the exceptional performance of a deep learning stream temperature model and the value of streamflow data</dc:title>
  <dc:type>article</dc:type>
</oai_dc:dc>