From Drops to Grid: Noise-Aware Spatio-Temporal Neural Process for Rainfall Estimation

2026-05-07

AI Summary

This work addresses the limitations of traditional rainfall observation—namely, low spatial resolution and high bias—which hinder accurate characterization of localized heavy precipitation. Existing deep learning approaches struggle to effectively fuse sparse and noisy ground station measurements with radar data. To overcome these challenges, the authors propose DropsToGrid, the first method to introduce noise-aware spatiotemporal neural processes into rainfall estimation. By integrating multimodal fusion, multiscale feature extraction, and temporal attention mechanisms, DropsToGrid enables efficient joint modeling of heterogeneously distributed personal weather stations and radar observations. The approach achieves robust, high-resolution rainfall field reconstruction even under few-station and cross-region conditions, while providing well-calibrated uncertainty quantification. Experiments on real-world data demonstrate that DropsToGrid significantly outperforms both operational systems and state-of-the-art deep learning baselines.

Abstract

High-resolution rainfall observations are crucial for weather forecasting, water management, and hazard mitigation. Traditional operational measurements are often biased and low-resolution, limiting their ability to capture local rainfall. Accurate high-resolution rainfall maps require integrating sparse surface observations, yet existing deep learning densification methods are hindered by rainfall's skewed, localized nature, noise, and limited spatio-temporal fusion. We present DropsToGrid, a Neural Process-based method that generates dense rainfall fields by fusing temporal sequences from noisy, irregularly distributed private weather stations with spatial context from radar. Leveraging multi-scale feature extraction, temporal attention, and multi-modal fusion, the model produces stochastic, continuous rainfall estimates and explicitly quantifies uncertainty. Evaluations on real-world datasets demonstrate that DropsToGrid outperforms both operational and deep learning baselines, generating accurate high-resolution rainfall maps with well-calibrated uncertainty, even when only few stations are available and in cross-regional scenarios.

Problem

• rainfall estimation
• spatio-temporal fusion
• noise-aware
• high-resolution mapping
• uncertainty quantification

Innovation

• Neural Process
• Rainfall Estimation
• Uncertainty Quantification
• Spatio-Temporal Fusion
• Noise-Aware Modeling