Perceptually Constrained Precipitation Nowcasting Model

Wenzhi Feng, Xutao Li, Zhe Wu, Kenghong Lin, Demin Yu, Yunming Ye, Yaowei Wang

Abstract: Most current precipitation nowcasting methods aim to capture the underlying spatiotemporal dynamics of precipitation systems by minimizing the mean square error (MSE). However, these methods often neglect effective constraints on the data distribution, leading to unsatisfactory prediction accuracy and image quality, especially for long forecast sequences. To address this limitation, we propose a precipitation nowcasting model incorporating perceptual constraints. This model reformulates precipitation nowcasting as a posterior MSE problem under such constraints. Specifically, we first obtain the posteriori mean sequences of precipitation forecasts using a precipitation estimator. Subsequently, we construct the transmission between distributions using rectified flow. To enhance the focus on distant frames, we design a frame sampling strategy that gradually increases the corresponding weights. We theoretically demonstrate the reliability of our solution, and experimental results on two publicly available radar datasets demonstrate that our model is effective and outperforms current state-of-the-art models.

Lay Summary:

Most Current precipitation nowcasting methods mainly focus on reducing prediction errors to improve performance. However, they often overlook how precipitation is actually distributed, which can lead to blurry forecast images and lower accuracy — especially for longer time periods.

To tackle this problem, we introduce a new approach that adds perceptual constraints to the forecasting process. This helps the model not only reduce errors, but also better understand and reproduce how precipitation really looks and behaves. We first predict the general trend of future rainfall, and then use a rectified flow model to adjust the results so they better match real-world patterns. We also design a strategy that encourages the model to pay more attention to frames further in the future.

Experiments on several public radar datasets show that our method produces clearer, more accurate forecasts compared to existing approaches.

Primary Area: Applications->Chemistry, Physics, and Earth Sciences

Keywords: Precipitation Nowcasting, Spatio-Temporal Prediction, Generative model

Submission Number: 3747