Perspectives on dynamic water quality modeling across continental and watershed scales

Stream water-quality and its drivers vary across time and space, but we only monitor a small fraction of streams consistently over long periods of time. Such limited monitoring necessitates the development and application of spatially explicit and dynamic models to predict water quality at unmonitored locations. Historically, data and computational limitations have hindered temporally variable prediction efforts across large spatial scales. However, hybrid statistical and process models, such as Spatially Referenced Regression on Watershed attributes (SPARROW), can provide spatially explicit, accurate predictions of water quality constituents with substantially lower computational cost than process-only models while retaining process-level information that can be obscured within machine learning models. An emerging next generation of such hybrid models moves beyond temporally static predictions into dynamic predictions. Here, we present regional- and continental-scale dynamic SPARROW models developed across the United States to simulate annual salinity and seasonal nutrient loads and concentrations over decades. Dynamic SPARROW models account for temporal variability of constituent sources and processes that deliver constituents from the landscape to streams. In addition, dynamic SPARROW models quantify lagged delivery of contaminants to streams that may have accumulated in soils, groundwater, and vegetation. Results quantify that legacy sources can vary by constituent, location, and time, and provide inference into river responses and lags to management activities. For example, groundwater storage contributes between 66 and 82% of the dissolved solids load to streams in the Upper Colorado River Basin, while lagged storage contributes on average between 20% to nearly 50% of the total nutrient load to Illinois River Basin streams. Ongoing work to expand dynamic representation of loading up to the continental United States will provide further insight into the continually evolving impacts of legacy and other sources on riverine water quality. Dynamic representation of key processes across spatial scales provides new opportunities for more informed management that can improve water quality for human and ecosystem uses.