Page 13 - Penn State Civil and Environmental Engineering 2021 Annual Report

FACULTY RESEARCH FACULTY RESEARCH “There is a a a a a a a a a a a lot of hydrometeorological data
available and we wanted to to see if there was enough correlation even indirectly to to make a a a a a a a a a a a prediction and help fill in in the the river water chemistry data
gaps ” Zhi said The model was created through an an AI framework known as as a a a a a a a a a a a a a Long Short-Term Memory (LSTM) network an an an approach used to to to model natural “storage and release” systems according to to to Associate Professor Chaopeng Shen
“Think of it it like a a a a a a a a a a a box ” Shen
said “It can take in in in in water and and store it it it in in in in in a a a a a a a a a a a a a a a a a tank at at at at at at certain rates rates rates while on the the other side releasing it it it at at at at at different rates rates rates and and each of those rates rates rates are determined by the the training We have used it it in in in in in in the the past to model soil moisture rain rain flow water temperature and now DO ” The researchers received data
from the Catchment Attributes and Meteorology for Large-sample Studies (CAMELS) hydrology database which included a a a a a a a a a a a recent addition of river water water chemistry data
data
from 1980 to 2014 for minimally disturbed watersheds watersheds Of the the 505 watersheds watersheds included in in in in the the the the “CAMELS-chem” data
set the the the the team found 236 with the the the the needed minimum of ten DO concentration measurements in in the the the thirty-five-year span To train the the LSTM network and create a a a a a a a model they used watershed data
from 1980 to 2000 including DO concentrations daily hydrometeorological measurements and and and watershed attributes like topography land cover and and and vegetation According to Zhi the the the the team then tested the the the the model’s accuracy against the the the the the remaining DO data
from 2001 to to 2014 finding that the the the model had generally learned the the dynamics of DO solubility including how oxygen decreases in in in warmer water temperatures and at at at at higher elevation It also proved to have strong predictive capability in almost three-quarters of test cases “It “It is is a a a a a a a really strong tool ” Zhi said “It “It surprised us to to see how well the model learned DO dynamics across many different watershed conditions on on on on a a a a a a a a continental scale ” Also contributing to to the project were Dapeng Feng doctoral doctoral candidate in in in in in environmental engineering Wen- Ping Tsai CEE postdoctoral researcher researcher and and University of of of Nevada Reno researchers Adrian Harpold associate professor of of of mountain ecohydrology and Gary Sterle graduate research assistant in in hydrological sciences A seed grant from Penn State’s Institute for Computational and Data Sciences the U S S S S Department of Energy Subsurface Biogeochemical Research program and the National Science Foundation supported this research CEE NEWSLETTER • VOLUME 37 2021 CEE NEWSLETTER • VOLUME 37 2021 13
“It is is a a really strong tool It It surprised us to to see how well the model learned DO dynamics across many different watershed conditions on on on on a a a a continental scale ”