Project Description

Accurate weather forecasting plays a crucial role in sectors such as agriculture, renewable energy, disaster management, and urban planning. However, the increasing variability in climate patterns, driven by global climate change, has made weather systems more complex, nonlinear, and uncertain. Traditional forecasting approaches, including statistical models and numerical weather prediction (NWP) techniques, often struggle to capture these dynamic patterns efficiently, especially for multi-step ahead predictions and region-specific variability.

Recent advances in artificial intelligence and data-driven modelling provide new opportunities to improve weather forecasting under complex atmospheric conditions. Modern deep learning techniques can effectively capture nonlinear temporal and spatial relationships present in meteorological observations, satellite products, and sky imaging datasets. In addition to conventional meteorological variables such as temperature, humidity, wind speed, and solar irradiance, emerging approaches involving cloud image processing and aerosol characterization can significantly enhance forecasting performance. Ground-based sky imagers and remote sensing observations provide valuable information on cloud dynamics, cloud cover variability, and atmospheric conditions that directly influence surface solar radiation. Furthermore, aerosols play a critical role in modifying the scattering and absorption of solar radiation, thereby affecting weather patterns and renewable energy generation. Integrating these heterogeneous datasets within physics-informed and AI-based frameworks offers a promising pathway for improving forecasting accuracy and reliability.

This research aims to develop hybrid physics-based and data-driven frameworks for weather forecasting across different climatic conditions. The study will investigate the application of machine learning and deep learning techniques for surface-based prediction of key atmospheric and renewable energy parameters, including solar irradiance, temperature, cloud cover, and precipitation. The proposed work will explore cloud image analysis, aerosol-radiation interactions, and spatiotemporal modelling approaches to better understand atmospheric variability and improve forecasting performance. The developed framework will incorporate data from multiple climatic regions to improve model generalization, robustness, and adaptability for renewable energy and climate-related applications.

The expected outcome of this research is the development of a reliable and scalable forecasting framework capable of supporting decision-making in climate-sensitive applications, particularly in renewable energy systems such as solar power generation. The proposed work will contribute to improving forecasting accuracy, reducing uncertainty, and enabling more efficient energy planning and climate adaptation strategies.