VTechWorks staff will be away for the Independence Day holiday from July 4-7. We will respond to email inquiries on Monday, July 8. Thank you for your patience.
 

Process-Guided Deep Learning Predictions of Lake Water Temperature

Simple item page
dc.contributor.authorRead, Jordan S.en
dc.contributor.authorJia, Xiaoweien
dc.contributor.authorWillard, Jareden
dc.contributor.authorAppling, Alison P.en
dc.contributor.authorZwart, Jacob A.en
dc.contributor.authorOliver, Samantha K.en
dc.contributor.authorKarpatne, Anujen
dc.contributor.authorHansen, Gretchen J. A.en
dc.contributor.authorHanson, Paul C.en
dc.contributor.authorWatkins, Williamen
dc.contributor.authorSteinbach, Michaelen
dc.contributor.authorKumar, Vipinen
dc.contributor.departmentComputer Scienceen
dc.date.accessioned2020-06-12T17:46:06Zen
dc.date.available2020-06-12T17:46:06Zen
dc.date.issued2019-11-08en
dc.description.abstractThe rapid growth of data in water resources has created new opportunities to accelerate knowledge discovery with the use of advanced deep learning tools. Hybrid models that integrate theory with state-of-the art empirical techniques have the potential to improve predictions while remaining true to physical laws. This paper evaluates the Process-Guided Deep Learning (PGDL) hybrid modeling framework with a use-case of predicting depth-specific lake water temperatures. The PGDL model has three primary components: a deep learning model with temporal awareness (long short-term memory recurrence), theory-based feedback (model penalties for violating conversation of energy), and model pretraining to initialize the network with synthetic data (water temperature predictions from a process-based model). In situ water temperatures were used to train the PGDL model, a deep learning (DL) model, and a process-based (PB) model. Model performance was evaluated in various conditions, including when training data were sparse and when predictions were made outside of the range in the training data set. The PGDL model performance (as measured by root-mean-square error (RMSE)) was superior to DL and PB for two detailed study lakes, but only when pretraining data included greater variability than the training period. The PGDL model also performed well when extended to 68 lakes, with a median RMSE of 1.65 degrees C during the test period (DL: 1.78 degrees C, PB: 2.03 degrees C; in a small number of lakes PB or DL models were more accurate). This case-study demonstrates that integrating scientific knowledge into deep learning tools shows promise for improving predictions of many important environmental variables.en
dc.description.adminPublic domain – authored by a U.S. government employeeen
dc.description.notesSee supporting information for data access, extended methods details, and example code. See https://doi.org/10.5066/P9AQPIVD for this study's data release and https://doi.org/10.5281/zenodo.3497495 for the versioned code repository. This research was funded by the Department of the Interior Northeast and North Central Climate Adaptation Science Centers, a Midwest Glacial Lakes Fish Habitat Partnership grant through F&WS, NSF Expedition in Computing Grant 1029711 to the University of Minnesota, a postdoctoral fellowship awarded to J.A.Z. under NSF EAR-PF-1725386, and a seed grant from the Digital Technology Center at the University of Minnesota. Access to computing facilities was provided by the University of Minnesota Supercomputing Institute and USGS Advanced Research Computing, USGS Yeti Supercomputer (https://doi.org/10.5066/F7D798MJ). We thank North Temperate Lakes Long-Term Ecological Research (NSF DEB-1440297) and Global Lake Ecological Observatory Network (NSF 1702991) for modeling discussions and data sharing, and Arka Daw, Randy Hunt, Jeff Sadler, Emily Read, and Mike Fienen for the PGDL discussions and ideas. We thank Luke Winslow, Noah Lottig, Madeline Magee, and along with MN DNR and WI DNR for temperature and bathymetric data, with special thanks to Pete Jacobson, Katie Hein, and Madeline Humphrey for collating thousands of temperature records, and Dave Wolock, the editorial group at WRR, and three anonymous reviewers for input that was used to improve this paper.en
dc.description.sponsorshipDepartment of the Interior Northeast Climate Adaptation Science Center; Midwest Glacial Lakes Fish Habitat Partnership grant through FWS; NSF Expedition in Computing Grant [1029711]; NSFNational Science Foundation (NSF) [EAR-PF-1725386]; Digital Technology Center at the University of MinnesotaUniversity of Minnesota System; Department of the Interior North Central Climate Adaptation Science Center; North Temperate Lakes Long-Term Ecological Research [NSF DEB-1440297]; Global Lake Ecological Observatory Network [NSF 1702991]en
dc.format.mimetypeapplication/pdfen
dc.identifier.doihttps://doi.org/10.1029/2019WR024922en
dc.identifier.eissn1944-7973en
dc.identifier.issn0043-1397en
dc.identifier.urihttp://hdl.handle.net/10919/98833en
dc.language.isoenen
dc.rightsCreative Commons CC0 1.0 Universal Public Domain Dedicationen
dc.rights.urihttp://creativecommons.org/publicdomain/zero/1.0/en
dc.subjectdeep learningen
dc.subjectlake modellingen
dc.subjecttemperature predictionen
dc.subjectprocess-guided deep learningen
dc.subjecttheory-guided data scienceen
dc.subjectdata scienceen
dc.titleProcess-Guided Deep Learning Predictions of Lake Water Temperatureen
dc.title.serialWater Resources Researchen
dc.typeArticle - Refereeden
dc.type.dcmitypeTexten
dc.type.dcmitypeStillImageen

Files

Original bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
2019WR024922.pdf
Size:
1.75 MB
Format:
Adobe Portable Document Format
Description:
Download

Collections

Scholarly Works, Computer Science