Predictive analytics for urban temperature dynamics in Uganda using Smart Models: An ARIMAX-LSTM and damped local trend approaches

Urban temperature dynamics are increasingly critical to sustainable planning amid rapid population growth and climate variability. This project aims to forecast the urban temperature in a few selected Ugandan cities using smart models such as the Damped Local Trend (DLT) model and ARIMAX-LSTM. The model used monthly temperature, precipitation, and relative humidity datasets (2017–2023) from the Uganda National Meteorological Authority and World Bank annual urban population data (1961–2023). The DLT model fitted with the LBFGS algorithm using Orbit performed better and detected temperature trends more slowly, as evidenced by smaller forecast errors (MAE = 3.239, RMSE = 3.270, BIC = 272.682). The ARIMAX-LSTM model well represented nonlinear interactions, with humidity significantly lowering temperature (Coef = -0.0911, p < 0.001) but producing higher error metrics (MAE = 4.07, RMSE = 10.33). Forecasts for 2025 revealed that DLT offered more consistent and plausible temperature trajectories, while ARIMAX-LSTM showed irregularities such as an anomalous drop in January (-4.89°C). These findings underscore the relevance of integrating hybrid deep learning with structural Bayesian methods in urban analytics. The study contributes to intelligent computing by offering data-driven insights for climate-resilient urban policy, highlighting the potential of sensor-augmented forecasting systems as a future research frontier.