Spatiotemporal Temperature Trends and Hybrid Machine Learning Prediction Across the Coastal Rainforest Zone of Nigeria

Abstract

This study assessed spatiotemporal temperature trends and machine-learning-based temperature prediction across the coastal rainforest zone of Nigeria. Monthly and annual temperature data for Calabar, Yenagoa, Benin City and Ikeja were obtained from the NASA POWER database for the period 1981–2025. The dataset was processed using descriptive statistics, visual inspection, box plots, Mann–Kendall trend testing and Sen’s slope estimation. Hybrid machine-learning prediction was performed using CNN–LSTM, while RF–SVR model performance was also evaluated using root mean square error and mean absolute error. The results showed statistically significant increasing temperature trends across all four locations. Sen’s slope values were 0.031°C/year for Calabar, 0.026°C/year for Yenagoa, 0.017°C/year for Benin City and 0.024°C/year for Ikeja. The strongest monotonic trend was observed in Calabar, while Benin City recorded the highest overall temperature distribution among the stations. Box-plot analysis indicated relatively narrow temperature variability, reflecting the stable thermal character of the humid rainforest environment. Model evaluation showed that RF–SVR produced lower prediction errors than CNN–LSTM across all locations, with the lowest RF–SVR error recorded for Benin City. Overall, the findings indicate a consistent warming tendency across the coastal rainforest zone and demonstrate the usefulness of combining statistical trend analysis with machine-learning evaluation for regional temperature assessment.