Modeling continental US stream water quality using Long-Short Term Memory (LSTM) and Weighted Regressions on Time, Discharge, and Season (WRTDS)
Abstract
The temporal dynamics of solute export from catchments is extremely challenging to understand and to model due to confounding hydrological and biogeochemical processes. Conventionally, the concentration and discharge relationship (C-Q) and statistical approaches to describe it, such as the Weighted Regressions on Time, Discharge and Seasons (WRTDS), have been widely used. Recently, deep learning (DL) approaches, especially Long Short-Term Memory (LSTM) models, have shown predictive capability for discharge and stream-dominated variables. However, it is not clear if such advances could be expanded to water quality variables that are driven by complex subsurface biogeochemical processes. This work evaluates the performance of LSTM and WRTDS for 20 water quality variables across ~500 catchments in the continental US. We find that LSTM does not markedly outperform WRTDS. Both models present a common pattern across water quality variables, with the LSTM displaying better performance for nutrient variables and worse on weathering-related solutes. Also, LSTM does not benefit from the data that drives the stream generating process. For further evaluation, we introduced a "simplicity index" which considers both the seasonality in the concentration pattern and the linearity in the C-Q relationship, or the C-Q-t pattern. The simplicity index can explain both the similarity and differences between modeled solute behaviors, and arguably, the underlying controls on water quality dynamics. The DL experiments and model-intercomparison highlight the strengths and deficiencies of existing frameworks, pointing to the need to develop further hydrogeochemical theories that are amenable to complex basins and solutes.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFM.H13D..05F