Computationally-efficient deep learning models for nowcasting of precipitation: A solution for the Weather4cast 2025 challenge

TL;DR

This work tackles precipitation nowcasting by learning spatiotemporal dynamics directly in the satellite observation space with a ConvGRU encoder–decoder, then translating predictions to radar-derived rainfall through an empirically calibrated transformation $R = \alpha \cdot \max(0, 300 - \mathcal{T})^{\beta}$. A two-stage training strategy forecasts up to 4 hours ahead, using four horizon-specific ConvGRU models and transferring knowledge from BOXI_0015 to ROXI regions to enhance regional generalization. For downstream tasks, 3D event detection is performed on upsampled rainfall fields to characterize precipitation events. The method achieves 2nd place in the Weather4Cast 2025 cumulative rainfall task and shows comparable event-prediction performance to the baseline, demonstrating a computationally efficient and transferable approach for satellite-to-radar nowcasting with strong cross-domain generalization.

Abstract

This study presents a transfer-learning framework based on Convolutional Gated Recurrent Units (ConvGRU) for short-term rainfall prediction in the Weather4Cast 2025 competition. A single SEVIRI infrared channel (10.8 μm wavelength) is used as input, which consists of four observations over a one-hour period. A two-stage training strategy is applied to generate rainfall estimates up to four hours ahead. In the first stage, ConvGRU is trained to forecast the brightness temperatures from SEVIRI, enabling the model to capture relevant spatiotemporal patterns. In the second stage, an empirically derived nonlinear transformation maps the predicted fields to OPERA-compatible rainfall rates. For the event-prediction task, the transformed rainfall forecasts are processed using 3D event detection followed by spatiotemporal feature extraction to identify and characterize precipitation events. Our submission achieved 2nd place in the cumulative rainfall task. Further, the same model was used out-of-the-box for the event prediction task, and resulted in similar scores as the baseline model to the competition.

Figures (3)

• Figure 1: ConvGRU Weather Nowcasting Architecture.
• Figure 2: Comparison of ConvGRU, ConvLSTM, and Persistence models.
• Figure 3: Model comparison across lead time.