Forecasting Climate and Weather Variability Using AI-Based Hybrid LSTM and XGBoost Model

Abstract Accurate forecasting of climate and weather changes is crucial for effective disaster risk management in the world as well as in Eastern Ethiopia (Dire Dawa, Harar and Jijiga), a region where highly vulnerable to climate variability and extreme weather events such as drought, high heat wave and high flooding. This paper introduces a frame work based forecasting approach using advanced artificial AI techniques specifically, integrated Long Short-Term Memory (LSTM) and Extreme Gradient Boosting (XGBoost) to compare, analysis and predict key meteorological parameters, Earth skin temperature, relative humidity, precipitation and wind speed. The model were evaluated using comparative analysis of Meteorological data across these variables, with performance assessed through standard and statistical metrics including the coefficient of determination (R²), root mean square error (RMSE), and mean square error measures. The result shows both models are advanced of capturing multi-layer patterns and long term climatic change effectively. Specially, the integrated LSTM and XGBoost model outperformed LSTM and XGBoost over by 95% in handling complex time-dependent relationships high correlation and accuracy. For precipitation prediction, the hybrid model records a mean absolute error of 0.632, a root mean square error of 0.890, a MAPE of 1.6%, an R² of 0.989, and a correlation coefficient of 0.91, indicating excellent predictive fidelity. Temperature forecasting achieved an RMSE as low as 0.003 and R² of 0.998 almost a minimum error and a great prediction, denoting almost perfect correspondence to observed data. The forecasting results provide critical insights for the potential of flood risks, drought and high heat wave, a supportive mechanism for the development of early warning systems and informed adaptive strategies, with a great climate resilience and disaster preparedness.