The predominant control of land surface interactions on dissolved oxygen in rivers at the conterminous United State
Abstract
Dissolved oxygen (DO) is a critical measure for water quality, aquatic productivity, and riverine nitrous oxide (N2O) emissions. The mechanistic understanding of its spatial-temporal variations however remains elusive at the continental scale. Here we expanded a previously published CAMELS-chem DO database of 236 basins to 580 basins, and used a deep learning approach to understand the DO dynamics. With a much larger number of basins, the model performed better than the previous model. Among all input parameters, the land surface parameters regarding energy balance and temperature regime were the dominant control, followed by hydrology parameters that have varied importance across different regions. Snow parameters are the most important in the mountainous West and least important in the South, whereas streamflow is a much more important input in the water-limited Midwest. The inclusion of biogeochemical processes (represented by water temperature and water quality data) led to minimal improvement in model performance, possibly due to limited data availability. Basin characteristics were crucial but only 1/6 of the measured basin characteristics is sufficient. The model performed better than the benchmark water quality model, the Weighted Regressions on Time, Discharge, and Season (WRTDS). When we held out 100 basins from the training data in different regions, The model showed promise in forecasting DO levels in the 100 chemically-ungauged basins in most parts of the US except the West, where the median Nash-Sutcliffe efficiency is below 0.5.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFM.H33J..09Z