Search papers, labs, and topics across Lattice.
This paper introduces FlashBack Memory (FB), a novel module designed to enhance precipitation nowcasting by dynamically retrieving and integrating key historical states through an adaptive fusion gate. By incorporating FB into various recurrent models, the authors demonstrate significant improvements in prediction accuracy, particularly for high-intensity rainfall and long sequences, while also reducing false alarms and missed events. The results show that FB enhances the spatiotemporal representation capabilities of these models, leading to better temporal consistency and spatial localization in precipitation forecasts.
FlashBack Memory reduces false alarms and missed events in precipitation nowcasting by effectively leveraging historical data, achieving superior prediction accuracy for high-intensity rainfall.
Accurate precipitation nowcasting is crucial for disaster mitigation and socio-economic planning, yet existing methods often struggle with false alarms, missed events, and long range dependency modeling at high spatiotemporal resolution. To address these challenges, we propose FlashBack Memory (FB), a module that dynamically retrieves key historical states and integrates them via an adaptive fusion gate, enhancing the spatiotemporal representation capability of recurrent-based models. We incorporate FB into PredRNN, PredRNNpp, MIM, MotionRNN, and PredRNN-V2, and evaluate on CIKM2017, Shanghai2020, and SEVIR datasets. Experimental results demonstrate that FB significantly improves MSE, MAE, SSIM, and CSI metrics, particularly for high-intensity rainfall and long-sequence predictions, while reducing false alarms and missed events and enhancing temporal consistency and spatial localization. The proposed method provides a general and efficient memory enhancement mechanism, improving the overall performance of recurrent-based precipitation nowcasting models.
