Skill Assessment of Water Supply Forecasts for Western Sierra Nevada Watersheds

The western slope of the Sierra Nevada contains 13 major river basins with sufficient long-term seasonal forecast and runoff data to assess forecast skill. These seasonal forecasts are issued by the California Department of Water Resources on February 1, March 1, April 1, and May 1 of each year for the major watersheds in California. Annual average precipitation in these river basins goes from 1,500mm in the Yuba in the north, to 600mm in the Kern to the south. Average runoff fraction (April to July) for the various watersheds ranged from 0.1 to 0.5 with the lower-elevation watersheds, and the Kern in the south having lowest values. The difference between precipitation and runoff, an index of evapotranspiration, was highest in the lower-elevation Cosumnes and Mokelumne Basins. Approximately half of the April 1 annual forecasts had a percent bias of +/- 15%. Skill scores for the 13 watersheds showed low scores (0.3) for forecasts in February, increasing through the forecast season to 0.8 for forecasts issued May 1, with 1.0 being a perfect forecast. Correlation skill measures, such as the Nash Sutcliffe scores, also exhibited increases in skill through the season from 0.45 in February to 0.95 in May. A linear regression between Nash Sutcliff scores and watershed elevation yielded a strong relationship with a coefficient of determination of 0.77. This relationship between higher elevation basins and greater forecast skill reflects the stronger statistical relationship between snow accumulations at index sites and seasonal runoff, versus more rainfall dominance in lower elevation watersheds. April through July runoff for each year was classified as the lower 30%, the mid 40%, and upper 30%; categorical skill measures were computed on the three runoff categories. Increases in forecast skill during the forecast season were visible in the low-flow and high-flow years versus the midflow years. Over forecasting of flow in the middle category was especially apparent early in the season, illustrated by high early season false-alarm rate and over forecasting bias. Difficulty in making accurate forecasts for midflow runoff along with the under forecast of high-runoff years and the over forecast of low-runoff years are shown to be common difficulties in runoff forecasting, especially early in the forecast season. Forecast skill is shown to be elevation dependent and can be expected to decrease with increasing temperatures.