Ensemble Design for Forecasting Extreme Orographic Precipitation: Finer Horizontal Grid Spacing with Fewer Members?

August 4, 2026

The manuscript entitled “Ensemble Design for Forecasting Extreme Orographic Precipitation: Finer Horizontal Grid Spacing with Fewer Members?” was recently published in the journal Weather and Forecasting. The paper is led by CW3E’s Nora Mascioli and co-authored by Mohammadvaghef Ghazvinian (now at Lynker), Rachel Weihs, Caroline Papadopolous, Daniel Steinhoff, Matthew Simpson, and Luca Delle Monache. This work was supported by the Atmospheric River Program Phases III, IV, and V awarded by the California Department of Water Resources, and the U.S. Army Corps of Engineers Forecast-Informed Reservoir Operations. This work evaluates the impacts of ensemble size and horizontal resolution on precipitation forecast skill in the western U.S.

This work supports CW3E’s 2025-2029 Strategic Plan priority “Atmospheric Rivers and Extreme Precipitation Research, Prediction, and Applications”. Large ensembles, like CW3E’s 200-member ensemble, are powerful tools for forecasting extreme precipitation. However, running a large ensemble is computationally expensive, requiring trade-offs between ensemble size and ensemble grid spacing. This study evaluates the relative benefits of running a large ensemble with coarse horizontal grid spacing compared with a smaller ensemble at finer horizontal grid spacing.

This study compares two experimental versions of the operational West-WRF ensemble run at 9- and 3-km horizontal grid spacing, each with 72 members, run for January and February of 2017 and 2019 (Figure 1). The 3-km ensemble improves on the precipitation biases in the 9-km ensemble, especially over regions of complex terrain such as the Sierras. However, the 3-km ensemble is ~27 times more computationally expensive than the 9-km ensemble. In order to reduce the computational expense of the 3-km ensemble, the number of members is decreased to 45, 30, and 15. The forecast skill changes with decreasing ensemble size for all events ( Figure 2a) and weighted toward extreme events (Figure 2b). The 45-member 3-km ensemble captures most of the skill of the 72-member 3-km ensemble, but as we move to smaller ensemble sizes (30 and 15), the skill degrades significantly, especially for extreme events. The 15-member 3-km ensemble is comparable to the 72-member 9-km ensemble in terms of computational cost, but by lead day 3 has less skill when predicting extreme precipitation events. We conclude that reducing the horizontal grid-spacing significantly improves the ensemble skill, but a large ensemble with coarse horizontal grid-spacing may still outperform compared with a small ensemble with fine horizontal grid-spacing.

Figure 1. (a) PRISM storm total precipitation over the Sierras from the PRISM dataset, (b) ensemble-mean forecast errors for the 9-km ensemble (CTRL; CTRL – PRISM), (c) difference in terrain height between the 9- and 3-km model configuration, and (d) difference in storm total precipitation forecasts between 3km-72mem and CTRL. Storm period is 7–10 Jan 2017, and the model domain is subset to the Sierra Nevada and surrounding area. Figure 6 from Mascioli et al. (2026).

Figure 2. (a) Continuous ranked probability skill score (CRPSS) and (b) the 95th quantile threshold weighted continuous ranked probability skill score (twCRPSS), for the 24-h accumulated precipitation 1–4-day forecasts of 3km-72mem (red; diamonds), 3km-45mem (cyan; circles), 3km-30mem (green; triangles), 3km-15mem (purple; squares), and CTRL (black; x symbols). Note the difference in the y axes. All differences between the 3-km ensembles and CTRL are significant at 95% using a two-sided t-test. Figure 7 from Mascioli et al. (2026).

Citation:

Mascioli, N. R., Ghazvinian, M., Weihs, R., Papadopoulos, C., Steinhoff, D. F., Simpson, M., & Delle Monache, L. (2026). Ensemble Design for Forecasting Extreme Orographic Precipitation: Finer Horizontal Grid Spacing with Fewer Members?. Weather and Forecasting, 41(6), 1275-1288. https://doi.org/10.1175/WAF-D-25-0177.1