An assessment of NRCS seasonal streamflow forecast performance in Idaho

Abstract

This report summarizes recent NRCS seasonal streamflow forecast performance in Idaho. Considering only April 1st forecasts for the April-July period, I first attempted to quantify forecast performance from a usersí perspective, using three simple statistical parameters: (1) the standard error derived from each forecast equation, (2) the percent error of the median forecast in relation to the observed flow, and (3) a count of how frequently the observed streamflow volume fell outside the forecasted range of values. Regional differences in these forecast error metrics were apparent, with points in northern basins showing lower standard errors and percent errors but higher out-of-range counts, and points in southern basins showing higher standard errors and percent errors and generally lower out-of-range counts. The high out-of-range counts across the state in certain years can be attributed to extreme spring weather events that could not be predicted using the current statistical forecast techniques. Future weather remains the largest source of uncertainty in seasonal streamflow forecasts.

I then used the Nash-Sutcliffe skill score (NS), which accounts for the annual variability of observed streamflow and allows for direct comparison of forecast skill between basins with highly-variable streamflow and basins with more annual consistency in streamflow volumes. Regional differences in forecast error were filtered out using the NS skill score, and points with lower scores were identified, possibly revealing more systematic errors in the forecasts. I then attempted to isolate influences on forecast quality by comparing basin- or region-scale metrics to forecast performance. Of the factors considered in this analysis, climatic region and the occurrence of extreme spring weather events remain the strongest influences on streamflow forecast quality and performance.