Development of a machine learning model for local to national river temperature modeling

The impacts of disturbance events (climate change, wildfires, urbanization) on stream water temperature, and their subsequent effects on ecosystem health, fish species, and water quality, continue to be poorly constrained in hydrologic prediction models. This project aims to address this critical knowledge gap by way of analysis and application of innovative stream temperature models for both natural and urban systems.

The overarching goal of this CIROH project is to improve local to national predictions of stream temperature through enhancement and coupling within the NextGen framework. The Year 1 objective was to develop machine learning models capable of predicting stream temperature across different spatial scales that can be integrated into the NextGen framework. The Year 2 objective is to demonstrate how the machine learning models can be used to evaluate the impacts of disturbance (climate change, fire, urbanization) on stream temperature, and how that in turn may impact ecosystem health and habitat availability for fish species in the contiguous U.S.

To enable local- to national-scale river temperature modeling, we will develop or adapt a total of three data-driven river temperature models.  The first model, TempEst 2.0, extends an existing river temperature remote sensing model (TempEst) to estimate daily historic river temperatures using satellite remote sensing data.  We are currently finalizing TempEst 2.0.  Using TempEst 2.0 outputs, the second model, NEXT (“Near-term EXpected Temperatures”), will use National Weather Service forecasts to forecast near-term (e.g., 1-10 days) and seasonal to sub-seasonal river temperatures. Finally, a third model will use TempEst 2.0 and NEXT outputs to provide predictions at high resolution within a single stretch of river and predict the effects of river-scale disturbances.  We will then apply the three models to disturbance and climate change case studies within the CONUS.  All three models will be fully compatible with the NextGen national water model framework.  Key outcomes include the ability to efficiently, and without on-site field data, model current and future river temperatures at a range of scales, as well as novel insights into case studies derived using that new modeling capability.

This project, with straightforward integration into the NextGen framework, will directly operationalize river temperature analysis and forecasting at a range of spatial and temporal scales, including both current/historical conditions and disturbance scenarios.