Geospatial based Comparison of malaria and climate variability in Dar es salaam Using GAM and Random Forest

Background: Malaria is public health burden in tropical and subtropical regions like Dar es Salaam, such that climate variability exerts a major influence on malaria transmission dynamics. Traditional field-based mapping and modeling techniques are often limited by expense, spatial incompleteness, and reliance on linear or non-linear systems. These gaps resolved by applying a geospatially integrated models to analyze compounded effects of climate variables on malaria incidence. Methods: The research utilized confirmed malaria case data from 2015 to 2023 from 1,794 health facilities, alongside climate variables (temperature, rainfall, humidity) from TerraClimate cross-verified with TMA records. Bilinear interpolation and inverse distance weighting applied to both climate variables and Malaria for spatial compatibility a mapping. The GAM applied to identify non-linear relationships, while RF models to examine linear as well as non-linear relationships. Partial dependence plots and Pearson correlations were used to evaluate model fit and climate impact. Results: Malaria cases decreased over time, but spatial heterogeneity persisted. Within 2015, 2019 and 2023 GAM explained 23.4%, 39.1%, and 40.2% deviance respectively, with partial correlation of 0.484, 0.626, and 0.634. Given that temperature increases the influence over time, with effective degrees of freedom rising from 8.287 to 8.828. RF models were superior to GAM, forecasting 71.07%, 71.36%, and 74.37% of variance for corresponding years, and with Pearson correlations of 0.977, 0.975, and 0.980. Partial dependence plots showed strong positive associations of malaria cases with temperature (26.4 to 27.2°C) and rainfall (75 to 130 mm), and a more complex, typically inverse relationship of humidity (20.5 to 22.0 mmHg). Conclusions: The Random Forest model was more highly predictive than GAM, confirming the reality of malaria transmission in Dar es Salaam being moderated by linear and non-linear relationships with climate factors. These results identify geospatially integrated modeling paradigms to be more effective to enhance malaria surveillance and inform climate-resilient public health policy.