Next-Generation Weather Forecasting: A Survey on Integrating AI Models for Accurate and Scalable Climate Predictions

Abstract

The growing necessity for precise, fine-grained, and real-time weather predictions has exposed several limitations in conventional numerical weather prediction (NWP) methods. These traditional models, while physically grounded, are computationally expensive and often less reliable when dealing with incomplete or noisy atmospheric data. In light of this, machine learning (ML) and deep learning (DL) approaches have gained prominence for their ability to learn from historical patterns and handle the complex, nonlinear dynamics of weather systems. This survey examines a broad spectrum of ML and DL models, such as Support Vector Regression (SVR), Random Forest (RF), Gradient Boosting (GBM/XGBoost), Long Short-Term Memory (LSTM), and Convolutional Neural Networks (CNN), along with hybrid models like CNN-LSTM and XGBoost-LSTM. In addition, the paper explores advanced mechanisms such as attention layers, physics-aware ML frameworks, and preprocessing techniques including wavelet transforms and empirical mode decomposition (EMD) that contribute to improved prediction accuracy. Key applications reviewed include rainfall forecasting, temperature estimation, flood prediction, and solar radiation modelling. The survey also highlights ongoing challenges, including overfitting, lack of interpretability, uneven data distribution, and difficulties in transferring models across different climatic zones. By synthesizing recent advancements, this paper aims to provide a valuable reference point for researchers and practitioners seeking to enhance atmospheric forecasting using intelligent, data-driven approaches.