QuantWeather: Quantile-Aware Probabilistic Forecasting for Subseasonal Precipitation

2026-05-11

AI Summary

Subseasonal precipitation forecasting remains challenging due to systematic biases and reliance on costly reforecast data, which hinder existing ensemble methods from reliably quantifying uncertainty. This work proposes QuantWeather—the first end-to-end probabilistic forecasting framework that operates without reforecasts—employing a dual-head neural network architecture to jointly optimize deterministic and quantile-based objectives. By integrating a quantile-aware loss function with stochastic forward sampling, the model generates well-calibrated probabilistic forecasts in a single forward pass. QuantWeather maintains competitive forecast skill while substantially reducing computational overhead and storage requirements during inference, demonstrating superior performance across multiple probabilistic verification metrics.

Abstract

Subseasonal precipitation forecasting is inherently uncertain due to chaotic atmospheric dynamics, making reliable uncertainty estimation essential for real-world applications. Existing approaches typically represent uncertainty through ensemble forecasts rather than directly modeling predictive distributions. However, due to systematic model biases, raw ensemble outputs are often not well calibrated and cannot be directly interpreted as reliable uncertainty estimates. As a result, operational systems rely on post-hoc calibration based on reforecast datasets, which are computationally expensive to generate and maintain. To address these limitations, we propose QuantWeather, an end-to-end probabilistic forecasting framework with a dual-head design. The probabilistic and deterministic heads are supervised with separate objectives and optimized jointly. The framework further supports stochastic sampling, enabling probabilistic outputs even with a single stochastic forward pass and allowing optional multi-sample aggregation. Extensive experiments show that QuantWeather demonstrates superior probabilistic forecasting skill while substantially reducing inference-time computational and storage costs.

Problem

• subseasonal precipitation forecasting
• uncertainty estimation
• ensemble calibration
• reforecast datasets
• probabilistic forecasting

Innovation

• probabilistic forecasting
• quantile-aware
• dual-head architecture
• subseasonal precipitation
• stochastic sampling