Evaluating the Reliability of Satellite and Reanalysis Climate Datasets for Local-Scale Applications in Bangladesh Shuvo

Reliable information is essential for environmental assessment, disaster risk reduction, public health and local planning in Bangladesh, where ground observations remain spatially sparse. This study evaluated ERA5, MODIS, CHIRPS and GFS against contemporaneously matched observations from eight Bangladesh Meteorological Department stations using station pixel gridded data collected and analyzed for January 2019 – December 2022. Rainfall was retained as monthly accumulated precipitation and was not replaced with climatological means. Primary temperature and relative humidity analyses used complete cases, with isolated single month interpolation examined only in a sensitivity analysis. Performance was assessed using mean bias error (MBE), mean absolute error (MAE), root mean squared error (RMSE), Pearson correlation, Lin's concordance correlation coefficient (CCC), Bland-Altman limits, 95% station year cluster bootstrap confidence intervals, paired tests, station level variability and extreme rainfall analysis. ERA5 performed best for temperature (RMSE = 1.30 °C, r = 0.972, CCC = 0.934) and relative humidity (RMSE = 5.19 percentage points, r = 0.808, CCC = 0.782), whereas GFS showed poor humidity agreement (CCC = -0.113). Rainfall RMSE was similar for CHIRPS and ERA5 (177.8 and 179.3 mm), although ERA5 had lower MAE (100.0 versus 121.6 mm; Holm-adjusted p < 0.001). Both products underestimated high rainfall months. Leave one year out quantile mapping improved temperature and ERA5 humidity but increased rainfall RMSE; the modest GFS humidity RMSE reduction did not restore meaningful temporal agreement (corrected CCC ≈ 0.02). ERA5 is therefore the most defensible source for temperature and humidity, while rainfall products require cautious, application specific evaluation. Urban Heat Island Trends Across Southwestern Nigeria Using Modis (2013– 2020)