Subseasonal Errors and Skill in FIM-iHYCOM Model Precipitation: Sensitivity to Convective Parameterization and Model Resolution

The FIM-iHYCOM model has produced both real-time and retrospective forecasts at subseasonal timescales for NOAA's Subseasonal Experiment (SubX). Previous work has found that the SubX version of FIM-iHYCOM yields subseasonal (weeks 3 and 4) precipitation forecasts with some systematic biases, and - like other models at this timescale - very little predictive skill. Because sustained subseasonal precipitation anomalies can have significant impacts to society, any improvements to these forecasts are of great value. This study examines the sensitivity of FIM-iHYCOM simulated precipitation to the choice of convective parameterization (Grell-Freitas vs. Simplified Arakawa-Schubert), and to horizontal resolution ( 60 km vs. 30 km). Understanding these sensitivities is critical towards making improved subseasonal precipitation forecasts.

Simulated precipitation is verified against reanalysis in terms of bias (absolute error), deterministic skill, and probabilistic skill. While there are noticeable differences in biases amongst the different model configurations, the results in terms of deterministic and probabilistic forecast skill show much smaller differences. This indicates that a global reduction in precipitation bias is not sufficient to yield marked improvements in the skill of precipitation forecasts. This conclusion - along with confirmation of earlier work showing the stubbornly low predictive skill of precipitation at subseasonal timescales - suggests that it may be more worthwhile to investigate the impact of convective parameterization (and model resolution) on the errors and skill of the large-scale general circulation.