Accurate and unbiased precipitation forecasts are vitally important for the prediction of streamflow. Streamflow forecasts benefit greatly from probabilistic precipitation forecasts that can yield the likelihood of exceeding flood stage. Precipitation in the Southwest U.S. during the North American Monsoon varies greatly in both time and space, and accurately predicting the nature of this variation requires methods that can realistically account for the uncertainty in the spatial and temporal variation of precipitation. We present recent work on ensemble 24-hour precipitation forecasting for the Verde Basin of north-central Arizona at 1.8 km spatial resolution and hourly temporal resolution, produced by constructing a non-parametric PDF that relates model forecasted daily basin-average precipitation to observed precipitation. Each day's forecasts are disaggregated into ensemble precipitation forecasts by selecting historical analog days from the National Centers for Environmental Prediction Stage-IV precipitation analysis weighted by the conditional PDF for each model forecast.

A Monte-Carlo style simulation was performed to determine how significant precipitation forecast improvements are, compared to both the original model forecast and climatology. For 90% of the samples there was a decrease in the root mean square error when comparing the adjusted mean forecast with the model forecast. The probability forecasts for basin-average precipitation were more skillful than the climatological probability distribution, based on the continuous rank probability skill-score. Additional results demonstrate how our method compares with typical precipitation forecasting methods fused by National Weather Service River Forecast Centers to generate spatially and temporally disaggregated ensemble forecasts.

The disaggregated forecasts were also used to generate probabilistic streamflow forecasts for a sub-basin of the Verde River basin in North-Central Arizona using the VIC hydrologic model. The results indicate that these forecasts tend to over-predict on most days, while significantly lagging behind observed streamflow. These streamflow forecasting errors represent both VIC model calibration errors and precipitation forecasting errors. Further robust verification of the streamflow forecasts will also be presented.