Radar meteorological data analysis and aviation meteorological prediction model based on graph neural network

Abstract Radar meteorological data offers real-time, high-resolution information required for reliable aviation weather forecasts. However, the dynamic and nonlinear nature of atmospheric processes presents considerable hurdles for standard forecasting models, particularly in capturing complex spatial-temporal dependencies and interactions among multiple meteorological variables. This research aims to improve the accuracy and resilience of aviation meteorological forecasts by creating a unique model that successfully captures both spatial and temporal dynamics from radar data. Radar reflectivity, Doppler velocity, wind speed, temperature, and humidity were gathered from a network of meteorological stations. Noise reduction and z-score normalization were employed for the preprocessing process. Principal Component Analysis (PCA) reduces dimensionality and extracts prominent patterns from multivariate radar information. This research proposes a Gorilla Troops Optimized Graph Attention Network (GTO-GANet) method that incorporates Graph Neural Networks (GNN) to learn spatial structure, an attention (Attn) mechanism to dynamically weight node influence, and the Gorilla Troops Optimization (GTO) algorithm to optimize model parameters and learning weights for improving learning efficiency and prediction accuracy. Experimental results show that the suggested method significantly improves forecasting performance, with Mean Squared Error (MSE) of 18.92 compared to baselines. The proposed model accurately reflects complicated radar meteorological patterns, making it a dependable tool for precise and fast aviation weather forecasting.