A Two-step Time-dependent Enhanced Informer method for numerical weather prediction post-processing

Reliable precipitation forecasts are crucial for water resource management and flood disaster early warning. However, numerical weather prediction (NWP) products often suffer from systematic biases, limiting their applicability across different regions. To address this issue, this study proposes a Two-step Time-dependent Enhanced Informer (TTEInformer) method for precipitation post-processing. This method employs a two-step classification and regression correction framework. It classifies precipitation into wet and dry days, then performs regression correction on samples identified as wet days, while dry day samples are maintained as zero values. To better capture temporal dependencies, TTEInformer augments the Informer model with a feature extraction module and a bidirectional gated recurrent unit (BiGRU) module. The study evaluates the proposed method over the Yalongjiang River basin upstream of the Yajiang hydrological station and compare it with multiple deep learning baselines. The results indicate that all corrected products substantially reduce forecast errors relative to the raw NWP precipitation. The proposed model demonstrates outstanding performance, achieving an R value of over 0.9 and significantly reducing forecast errors compared to other models. Moreover, the two-step correction framework effectively enhances model correction accuracy compared to traditional direct correction strategies, with notable improvements in correcting light precipitation events. The work provides a reliable post-processing method for hydrometeorological applications in precipitation forecasting.