Evaluating a Commonly Used Downscaling Technique for Precipitation Forecasts

The National Weather Service relies on basic downscaling techniques to create a quantitative precipitation forecast (QPF) over complex western US terrain from low-resolution global model forecasts. The most common method uses the PRISM monthly climatology dataset to introduce sub-grid-scale variability in precipitation amounts, with the amount of PRISM influence subject to forecaster intervention. Lacking objective verification, forecasters typically exercise substantial freedom with the amount of PRISM influence, in the interest of producing reasonable forecasts at several key points, and an overall grid that appears realistic in the given synoptic/mesoscale environment. In addition to the obvious challenges associated with subjective forecaster intervention, the downscaling technique itself implies that the monthly-mean precipitation-altitude relationship is appropriate for every individual precipitation event. In fact, these relationships can exhibit high variability from one event to the next, and within a single event.

Downscaled QPF from several deterministic models is evaluated against point observations and the Stage-IV precipitation analysis to determine how the current downscaling technique performs in a wide variety of scenarios, ranging from summertime convective events to cold-season events with strong orographic enhancement. Modifications to the existing approach are explored to improve the accuracy of QPF in both mountain and valley locations.