This page organizes papers related to artificial intelligence published around the world. This page is summarized using Google Gemini and is operated on a non-profit basis. The copyright of the paper belongs to the author and the relevant institution. When sharing, simply cite the source.
Precipitation forecasting ( precipitation forecasting within 3 hours) is crucial for disaster mitigation and real-time response planning. However, existing meteorological benchmarks rely on variables with strong periodicity, such as temperature and humidity, and thus fail to adequately reflect model performance in complex real-world weather scenarios such as precipitation forecasting. To address this gap, we propose RainfallBench, a specialized benchmark for precipitation forecasting within 3 hours. This benchmark addresses precipitation forecasting problems characterized by zero inflation, temporal decay, and nonstationarity. Using data on six key variables (including precipitation variable water vapor (PWV)) recorded at 15-minute intervals from over 12,000 GNSS stations over five years, we design a specialized evaluation strategy to evaluate model performance for meteorological challenges such as multi-scale forecasting and extreme precipitation events. We evaluate over 20 state-of-the-art models on RainfallBench and propose the Bi-Focus Precipitation Forecaster (BFPF), a plug-and-play module that integrates domain-specific prior information to address zero inflation and temporal decay. The code and dataset are available under https://anonymous.4open.science/r/RainfallBench-A710에서 .
Takeaways, Limitations
•
Proposal of RainfallBench, a precipitation prediction benchmark.
•
Focus on precipitation forecasting tasks considering zero inflation, temporal decay, and non-stationarity.
•
Specialized assessment strategies for multi-scale forecasting and extreme precipitation event assessment.
•
Including PWV variables important for precipitation prediction.
•
Improving model performance through the BFPF module.
