Quantifying the uncertainty of precipitation forecasting

Lei Xu,Nengcheng Chen,Chao Yang

2021-01-01

Abstract:Precipitation forecasting is an important mission in weather science. In recent years, data9 driven precipitation forecasting techniques could complement numerical prediction, such as precipitation 10 nowcasting, monthly precipitation projection and extreme precipitation event identification. In data11 driven precipitation forecasting, the predictive uncertainty arises mainly from data and model 12 uncertainties. Current deep learning forecasting methods could model the parametric uncertainty by 13 random sampling from the parameters. However, the data uncertainty is usually ignored in the forecasting 14 process and the derivation of predictive uncertainty is incomplete. In this study, the input data uncertainty, 15 target data uncertainty and model uncertainty are jointly modeled in a deep learning precipitation 16 forecasting framework to estimate the predictive uncertainty. Specifically, the data uncertainty is 17 estimated a priori and the input uncertainty is propagated forward through model weights according to 18 the law of error propagation. The model uncertainty is considered by sampling from the parameters and 19 is coupled with input and target data uncertainties in the objective function during the training process. 20 Finally, the predictive uncertainty is produced by propagating the input uncertainty and sampling the 21 weights in the testing process. The experimental results indicate that the proposed joint uncertainty 22 modeling and precipitation forecasting framework exhibits comparable forecasting accuracy with 23 existing methods, while could reduce the predictive uncertainty to a large extent relative to two existing 24 joint uncertainty modeling approaches. The developed joint uncertainty modeling method is a general 25 uncertainty estimation approach for data-driven forecasting applications. 26