Skill evaluation of water supply forecasts in western Sierra Nevada and Colorado River basins
- Author(s): Harrison, Brent
- Advisor(s): Bales, Roger
- et al.
Runoff records from thirteen major river basins on the western slope of the Sierra Nevada in California were compared to runoff forecasts for those watersheds to determine the skill of those runoff forecasts. The forecasts, some dating back to the 1930's, were made at the beginning of the months of February, March, April and May. An array of summary, correlation and categorical skill measures were computed for each forecast and associated observation. The same array of skill measures were computed for 28 basins tributary to the Colorado River, again with some forecasts dating back to the 1930's. The skill measures were reviewed for each region and compared to watershed characteristics to develop explanations for the trends and results. Finally the Sierra Nevada results were directly compared with the watersheds in the Colorado River basin, thereby developing explanations for the trends, similarities and differences obtained.
A strong relationship between increasing watershed elevation and increased forecast skill was shown by the western Sierra forecasts but was not found in the Colorado locations. This result can be explained by snow dominance of the Sierra Nevada runoff. The snow dominance was shown to increase to the south as the elevation of the watersheds increased. The summary measures included a skill score based on Mean Absolute Error which showed clearly the increase in skill during the forecast season, with both the skill scores starting at 0.3 for both regions. This increase in skill over the forecast season is attributed to the increasing knowledge of climatology during the forecast season. The skill of the western Sierra forecasts ends the season somewhat higher at 0.8 compared to 0.6 for the Colorado Basin. The monthly increase in NS score was higher at 0.19 for the western Sierra watersheds when compared to watersheds elsewhere in the western United States, which ranged from 0.08 to 0.12 per month. These results can be explained by the snow magnitude and dominance of runoff in the Sierra Nevada. The under forecasting of high runoff years was illustrated in both the low False Alarm Rate and the under forecast Bias for high flow runoff years.