<b>Machine Learning Techniques for Predicting Surface Water Availability in Gombe State, Nigeria</b>

Water resource availability in semi-arid regions is increasingly threatened by climate variability, population growth, and unsustainable management practices. This study develops a big data and machine learning techniques for predicting water resource availability in Gombe State, Nigeria. The research employs three machine learning algorithms Random Forest (RF), XGBoost, and Long Short-Term Memory (LSTM) to model the complex, non-linear relationships between climatic variables and water availability. Historical climate and hydrological data spanning 2005–2025, integrated with remote sensing indices including Normalized Difference Vegetation Index (NDVI), Standardized Precipitation Index (SPI), and land surface temperature, were utilized for model training and validation. The dataset was processed through systematic data cleaning, feature engineering, and an 80:20 train-test split following the CRISP-DM methodology. Findings reveal that XGBoost outperformed both Random Forest and LSTM, achieving the lowest Root Mean Squared Error (4.090), lowest Mean Absolute Error (3.258), and highest coefficient of determination (R² = 0.9397), explaining approximately 94% of variance in the Water Availability Index (WAI). Random Forest demonstrated comparable performance (RMSE = 4.193, R² = 0.9366), while LSTM exhibited comparatively lower accuracy (RMSE = 5.513, R² = 0.8901), likely due to limited temporal depth in the training dataset. Correlation analysis identified rainfall as the dominant predictor of water availability (r = 0.93), followed by NDVI (r = 0.82) and SPI-3 (r = 0.72). Projections for 2026–2036 indicate a gradual decline in water availability, with WAI decreasing from approximately 31.0 to 28.7, representing a 7.4% reduction over the decade. Seasonal analysis reveals a characteristic unimodal pattern aligned with the West African monsoon, with severe dry season water stress (WAI ~12.5–16.8) contrasting with peak wet season abundance (WAI ~43.5–47.5). Spatial distribution mapping for 2036 demonstrates a pronounced north-south gradient, with northern regions (latitudes 11.0°–11.5°N) exhibiting significantly higher vulnerability to water stress compared to southern areas.