The recent advancements in weather forecasting and climate prediction have seen a significant shift towards leveraging deep learning models, particularly those incorporating attention mechanisms and multi-modal data integration. The field is increasingly focusing on developing scalable, high-resolution models that can handle complex spatio-temporal dynamics, often necessitated by the need for localized and timely predictions in response to environmental challenges such as climate change and natural disasters. Innovations in model architectures, such as the use of transformers and graph neural networks, are being employed to enhance the accuracy and reliability of forecasts, particularly in areas like precipitation nowcasting, nuclear radiation forecasting, and diabetes risk assessment. These models are not only improving prediction accuracy but also contributing to more efficient and scalable solutions, often through cloud-based implementations. Notably, the integration of physical models with data-driven approaches is becoming a common theme, aiming to bridge the gap between global large-scale models and locally accurate predictions. This trend underscores a move towards more holistic and context-aware forecasting systems that can better inform decision-making and emergency response strategies.