Mountain precipitation forecasts from operational ensembles

An accurate quantitative precipitation forecast (QPF) in high terrain is a valuable service for many applications including transportation safety, avalanche mitigation and outdoor recreation. In Western Region forecast offices of the National Weather Service (NWS), the QPF process is typically centered on a first-guess grid calculated through a combination of high-resolution (~4–12-km) regional, and statistically downscaled global (~35–80-km) numerical guidance. However, in the mountainous Western US, there is limited verification information available for raw or downscaled QPF from current operational models, and little or no use of probabilistic QPF guidance in operations.

To address these issues, 24-h accumulated Snow Telemetry (SNOTEL) precipitation observations are used to evaluate 6 years of cool-season (Oct-Mar) QPF from operational, single- and multi-model global ensemble forecast systems, at lead times from 1 to 7 days. The nuances of cool-season precipitation measurement are carefully considered in order to perform quality control on this large observational dataset. Ensemble QPF is downscaled to 1/120-degree resolution using a simple technique based on the monthly Parameter-elevation Relationships on Independent Slopes Model (PRISM) climatology. This analysis will focus on 2 areas: (1) skill of raw and downscaled ensemble mean QPF; (2) reliability of raw and downscaled probabilistic QPF.

QPF skill varies widely across the Western US, with clear maxima over the Cascade and Sierra Ranges, and minima over the Great Basin and parts of Colorado. At many SNOTEL sites, downscaled ensemble mean QPF has a small bias and is more skillful than official NWS forecasts. Using downscaled QPF from a three-model, 93-member ensemble, probabilistic forecasts of precipitation events exceeding 25 mm (24 h)^-1 are highly reliable. However, relatively low skill over parts of the interior West suggests a marked need for improvements to numerical guidance. Future work is necessary to determine whether forecast errors in these areas are related to resolution, the upstream moisture budget, model parameterization issues, and/or other factors.