FORECASTING OF METEOROLOGICAL DATA USING AI TECHNIQUES

Abstract

Weather forecasting has always been a tricky business. Even with decades of progress in numerical weather prediction, atmospheric systems remain highly nonlinear and chaotic, which makes short-term and medium-range forecasts vulnerable to error. In recent years, artificial intelligence has stepped in as a strong alternative, offering data-driven ways to capture patterns that traditional physics-based models sometimes miss. This paper looks into how AI techniques, particularly machine learning and deep learning, can be used to forecast meteorological parameters such as temperature, humidity, rainfall, wind speed, and atmospheric pressure. We worked with a historical dataset covering ten years of daily observations from a regional meteorological station and applied several models including Random Forest, Support Vector Regression, Long Short-Term Memory (LSTM) networks, and a hybrid CNN-LSTM model. The models were trained and tested using a chronological split to avoid data leakage, and performance was measured using RMSE, MAE, and R² scores. Results show that the hybrid CNN-LSTM model outperformed the others in temperature and rainfall prediction, while LSTM alone gave very competitive results for humidity and wind speed. The study also highlights how feature engineering and proper handling of missing values matter as much as the choice of model. Overall, our findings suggest that AI techniques can meaningfully improve forecasting accuracy when combined with thoughtful preprocessing and domain knowledge. The work contributes to ongoing efforts in building reliable, low-cost forecasting tools that can support agriculture, disaster management, and energy planning, especially in regions where high-resolution numerical models are not always accessible or affordable.