Uncertainty Estimates of Atmospheric River-Induced Extreme Precipitation from Numerical Precision in a Regional Weather Model

Accurate forecasts of extreme precipitation are critical when making decisions to protect life, property, and ecology as it is frequently associated with river, tributary, and lowland flooding in the U.S. West. The Russian River watershed, in Northern California, is a susceptible area to flooding, in part, because of the mountainous terrain and rainfall climatology. It receives the majority of its annual precipitation over the course of a few days, on average. These bursts of heavy precipitation are frequently associated with long, narrow, and moist corridors of moisture called atmospheric rivers (ARs). Skillful atmospheric modeling of ARs is essential to predicting the extreme precipitation especially in areas like the Russian River where decisions about flooding risk and water supply are vital to the public safety, economy, and ecology.

In order to better predict the horizontal transport of moisture in association with ARs, a customized version of the Weather Research and Forecasting (WRF) model was developed at the Center for Western Weather and Water Extremes (CW3E). The WRF model allows the flexibility to control the spatial and temporal resolution and relevant physical dynamics for features of interest. The customized model (herein West-WRF) is designed to capture the synoptic scale features of the AR, including the parent extratropical cyclone commonly associated with the AR and broad moisture plumes, as well as the mesoscale orographic ascent over land. Deterministic forecasts of weather along the West are produced each year during the cool season between December and March in order to provide high resolution forecast information.

The WRF model, like other NWP models, consists of parameterized physics and dynamics on a distributed grid that is typically run on supercomputers. The calculations can run in parallel to distribute memory between multiple processors in order to reduce computational load and overall expense. However, because of the unique discretization of the physics equations (especially when thresholds are used), and other factors surrounding numerical precision, the forecasts of precipitation can change simply due to differences in computer architecture, software, and memory distribution. This study presents an estimate of the variance in extreme precipitation in atmospheric rivers due to variations in the computer components that affect numerical precision.

The West-WRF will be used to simulate 3 major AR cases and its resulting precipitation. Each simulation will be performed using 7 different sets of processing nodes on the United States Army Corps of Engineers (USACE) High Performance Computer (HPC) cluster. In addition, each case will also be perturbed by turning on the capability to flip the last digit precision bit of all input values. The forecast fields will be analyzed to understand the scales of precipitation affected by numerical precision and the impact on hourly and total precipitation accumulations in the Russian River watershed. These case studies will be compared to the spread of precipitation distributions generated by other standard ensemble methodologies. These experiments will help provide a minimum error estimate on the precipitation skill in West-WRF deterministic forecasts.