Assessing the performance of operational CFSv2 & CWRF downscaled seasonal ensemble predictions over CONUS

The CFSv2 operational seasonal predictions are downscaled using a state-of-the-art coupled regional climate model CWRF (Climate-Weather Research Forecasting). The performance of CWRF and CFSv2 over the contiguous United States (CONUS) is assessed using the gridMET observations for 2013-2022. The CFSv2 ensemble consists of two forecast realizations initialized on 00Z of the first day of January and February each year, while the CWRF ensemble includes five physics members with different cumulus and radiation schemes times two initial conditions (at 00Z on the first day of January and February) for spring and three initial conditions (at 00Z on the first day of March, April, and May) for summer. The ensemble of CFSv2 showed a systemic cold bias in 2-meter daily maximum temperature, whereas a warm bias in minimum temperature over most regions for spring and summer seasons. CWRF ensemble significantly reduced these cold and warm biases, although notable regional biases persist, especially during summer. For precipitation, both CFSv2 and CWRF have a systematic wet bias over most CONUS except near Louisiana, with a dry bias during spring. However, summer predictions do not improve the wet bias in the west and dry bias in the east. CWRF drastically reduced the root mean square error of CFSv2 temperature forecasts, especially in the eastern CONUS region. CFSv2 temperature forecasts show high inter-annual anomaly correlation (IACC) scores (greater than 95% significance) in the Midwest and Northeast. CWRF downscaling predictions further improve these scores for spring, but no improvement is observed in summer. CFSv2 and CWRF show mixed precipitation IACC scores with positive and negative correlations over the CONUS. Overall, the CWRF ensemble has better prediction skills in spring and comparable skills during summer relative to the CFSv2 ensemble. These CWRF climates are used to get the crop growth forecasts derived from standalone and coupled crop models.