Spatiotemporal Precipitation Trends and Hybrid Machine Learning Prediction in the Humid Rainforest Climatic Zone of Nigeria

Abstract

This study investigated the spatiotemporal trends and hybrid machine learning prediction of precipitation in Nigeria’s humid rainforest climatic zone using 45 years (1981–2025) of monthly rainfall data from Calabar, Yenagoa, Benin City, and Ikeja. Data obtained from the NASA POWER database were analyzed using statistical, spatial, and machine learning techniques. The Mann–Kendall test and Sen’s slope estimator assessed the significance and magnitude of rainfall trends, while Kernel Density Estimation (KDE) and spatiotemporal analyses evaluated spatial variability. Hybrid machine learning models, Random Forest–Support Vector Regression (RF-SVR) and Convolutional Neural Network–Long Short-Term Memory (CNN-LSTM), were employed for precipitation prediction. Model performance was evaluated using Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). Results revealed decreasing but non-significant rainfall trends at Calabar, Yenagoa, and Ikeja, while Benin City exhibited a significant increasing trend. Spatial analysis demonstrated considerable variability in precipitation characteristics across the study area. RF-SVR outperformed CNN-LSTM, with RMSE values of 0.1765–0.4608 and MAE values of 0.1433–0.3658, compared with RMSE values of 1.5639–1.7747 and MAE values of 1.2868–1.5055 for CNN-LSTM. CNN-LSTM forecasts for 2026–2035 indicated decreasing precipitation at Calabar, Yenagoa, and Ikeja, but an increasing trend at Benin City. RF-SVR predicted decreasing precipitation at Calabar, relative stability at Yenagoa, moderate variability at Benin City, and substantial interannual variability at Ikeja, without a significant long-term decline by 2035. These findings highlight rainfall variability as an important concern for agriculture, water resources, food security, and sustainable development in Nigeria’s humid rainforest zone.