STAA: Spatio-Temporal Alignment Attention for Short-Term Precipitation Forecasting

TL;DR

This work proposes a short-term precipitation forecasting model based on spatiotemporal alignment attention, with self-attention for temporal alignment (SATA) as the temporal alignment module and spatiotemporal attention unit (STAU) as the spatiotemporal feature extractor to filter high-pass features from precipitation signals and capture multiterm temporal dependencies.

Abstract

There is a great need to accurately predict short-term precipitation, which has socioeconomic effects such as agriculture and disaster prevention. Recently, the forecasting models have employed multi-source data as the multi-modality input, thus improving the prediction accuracy. However, the prevailing methods usually suffer from the desynchronization of multi-source variables, the insufficient capability of capturing spatio-temporal dependency, and unsatisfactory performance in predicting extreme precipitation events. To fix these problems, we propose a short-term precipitation forecasting model based on spatio-temporal alignment attention, with SATA as the temporal alignment module and STAU as the spatio-temporal feature extractor to filter high-pass features from precipitation signals and capture multi-term temporal dependencies. Based on satellite and ERA5 data from the southwestern region of China, our model achieves improvements of 12.61\% in terms of RMSE, in comparison with the state-of-the-art methods.

Figures (10)

• Figure 1: STAA workflows: First, it encodes spatio-temporal data using Convolution Neural Networks (CNNs) while maintaining the independence of variable channels. Next, a SATA automatically learns the correlations between variables. Subsequently, STAU integrates the spatio-temporal features of various variables. Finally, CNNs are again used to decode the embedding back to the spatio-temporal domain. '$\times L$' means the module are stacked $L$ times.
• Figure 2: Results of performance comparison and ablation study. '${\uparrow}$' means the higher the better, and '${\downarrow}$' means the inverse. The values in bold are the top-1 results. The underlined values are sub-optimal results. 'IMP(%)' is the percent of improvements of STAA over the sub-optimal ones.
• Figure 3: Comparison of different models in hourly precipitation prediction as a case study. The red frame demonstrates the deficits in the prediction of different methods.
• Figure 4: An analysis of the case from 03:00 to 13:00 on August 25, 2021, within the region of 26°-31° N and 94°-99° E. A: Actual observations. B: STAA model prediction results. Numbers 1-10 represent specific time periods in this case study. For heavy rain prediction (accumulated rainfall reaching or exceeding 100 mm/hour), green indicates correct predictions, blue indicates false alarms, and red indicates missed predictions.
• Figure 5: ERA5-TP data lag correlation analysis for the flood seasons of 2017-2021.