Author(s): Jiayan Xu; Shuguang Liu; Zhengzheng Zhou; Guihui Zhong
Linked Author(s): Guihui Zhong, Shuguang Liu, Zhengzheng Zhou
Keywords: Precipitation Forecasting Machine Learning ConvLSTM2D GPM Satellite Data Time Series Analysis
Abstract: Accurate precipitation forecasting is crucial for various applications including urban planning and disaster management. In this paper, we propose a machine learning approach to rainfall prediction by constructing a precipitation prediction model for Shanghai based on Global Precipitation Measurement (GPM) satellite data from 2001 to 2022 using ConvLSTM2D (2D Convolutional Long Short-Term Memory Network). The dataset is divided into two groups: the first 80% of the data is used as the training set, and the second 20% of the data is used as the validation set. The ConvLSTM2D deep learning model established in the study combines the advantages of convolutional neural network (CNN) and long short-term memory network (LSTM), which can simultaneously capture spatial and temporal dependencies, and is able to better process spatial and temporal sequence data and make rainfall prediction. In addition to the conventional metrics, spatial accuracy assessment metrics including structural similarity index (SSIM), peak signal-to-noise ratio (PSNR), Moran I, spatial root-mean-square error (Spatial RMSE), spatial distribution of relative errors, and correlation coefficients among the grid points are also added in the evaluation of rainfall prediction models. The research results can realize more accurate rainfall prediction, provide scientific theoretical basis for urban flood control planning, and provide technical support for cities to grasp the future climate change.
Year: 2025