In-situ soil moisture data improve seasonal streamflow forecast accuracy in rainfall-dominated watersheds

Abstract

Seasonal streamflow forecasting methods are less skillful in rainfall-dominated catchments than snow-dominated catchments, where measurements of water storage in the snowpack enhance predictability. Recent research in snow-dominated catchments showed that forecasts can be further enhanced by also including soil moisture measurements, but the impact of soil moisture data on forecast performance in rainfall-dominated watersheds remains unknown. Our objective was to evaluate the potential improvements gained by including in-situ soil moisture data in seasonal streamflow forecasting models in rainfall-dominated watersheds. Precipitation and soil moisture data from four watersheds in the U.S. were incorporated into a modified principal components analysis and regression method to predict seasonal (4-month) streamflow totals at 0-, 1-, 2-, and 3-month lead times. Forecasts derived from antecedent precipitation alone were often statistically insignificant and explained less than 30% of the variance in seasonal streamflow, as indicated by the Nash-Sutcliffe efficiency coefficient. Conversely, forecast models that included soil moisture information explained up to 87% of seasonal streamflow variance at the 0-month lead time, up to 81% at the 1-month lead time, up to 71% at the 2-month lead time, and up to 52% at the 3-month lead time. The root mean square errors for forecasts which included soil moisture data were on average 55% lower than for those based on antecedent precipitation alone. The soil moisture-based forecasts for rainfall-dominated watersheds exhibited accuracies comparable to those previously reported in snow-dominated watersheds. This new forecast method shows strong potential for use in surface water management in rainfall-dominated regions.