UCLA

In recent years, quantitative precipitation forecast became viable for the emerging needs around the world. The quality of the precipitation forecast would be crucial for further application. The purpose of this study is to understand the accuracy of the seasonal precipitation forecast and explore the potential of improving the forecast, as well as understanding the precipitation forecast performance from lead time 1 day to 10 years.To understand the accuracy of the seasonal forecast around the world, the forecast of North American Multi-Model Ensemble (NMME) in the Blue Nile basin, Western United States, Tigris-Euphrates River basin and Brazil is evaluated, and compared with gridded precipitation observation products. Results show that NMME overall can explain from 0% to up to 55% of the wet season precipitation variance, varying different models and regions. Also, in all selected basins, atmospheric climate indices can be used as a reliable predictor to forecast the wet season precipitation in linear regression. By combining the climate indices and NMME, the newly formed statistical model can have better performance, compared to NMME model results. This opens the possibility of utilizing machine learning techniques to reliably improve the seasonal precipitation forecast. The analysis of precipitation forecast across different lead time shows that the performance from lead time 1 day to 10 years is not dropping as lead time increases. When two products’ lead time overlaps, using the longer-range forecast product can have better forecast results, except decadal forecast.