Research on GNSS-PWV Retrieval and its Application in Rainfall Forecasting Based on Deep Learning

This thesis mainly studies the state-of-the-art methods and models of Precipitable Water Vapor (PWV) derived from ground-based Global Navigation Satellite Systems (GNSS) and its applications in rainfall forecast. Firstly, Convolutional Neural Networks (CNN) is utilized for feature extraction from GNSS-PWV data, leveraging their powerful image processing capabilities to identify spatial features in PWV. Subsequently, the Autoregressive Integrated Moving Average (ARIMA) model is applied for time series analysis, capturing the temporal dependencies in PWV variations. To further enhance the time series prediction performance, Long and Short Term Memory Neural Network (LSTM) is introduced to handle the long-term and short-term dependencies in PWV data, addressing the limitations of traditional time series models. Additionally, attention mechanisms are integrated into the LSTM models to improve their focus on critical time steps, thereby increasing prediction accuracy. To optimize the model parameters, the Slime Mold Algorithm (SMA) is employed. SMA, a novel optimization algorithm inspired by the foraging behavior of slime mold, efficiently searches for optimal solutions in high-dimensional spaces, thereby enhancing the convergence speed and prediction accuracy of the models. Experimental results demonstrate that the models integrating multiple deep learning techniques perform excellently in GNSS-PWV retrieval and rainfall forecasting, significantly outperforming single models. Keywords: global navigation satellite system, precipitable water vapor, rainfall forecasting, deep learning