Highlights What are the main findings? Atmospheric water vapor inputs from MERRA-2 and NCEP showed different error characteristics, and their use led to measurable differences in Landsat 8 land surface temperature retrieval. Normalized difference vegetation index threshold selection affected emissivity estimation and retrieval consistency, with cropland areas showing stronger sensitivity than desert and built-up areas. What are the implications of the main findings? Intermediate parameter settings should be carefully evaluated when applying single-channel land surface temperature retrieval. While the optimized framework guides region-adapted temperature monitoring, the limited case-study design underscores a critical need for broader, independent ground validation to verify absolute accuracy.Highlights What are the main findings? Atmospheric water vapor inputs from MERRA-2 and NCEP showed different error characteristics, and their use led to measurable differences in Landsat 8 land surface temperature retrieval. Normalized difference vegetation index threshold selection affected emissivity estimation and retrieval consistency, with cropland areas showing stronger sensitivity than desert and built-up areas. What are the implications of the main findings? Intermediate parameter settings should be carefully evaluated when applying single-channel land surface temperature retrieval. While the optimized framework guides region-adapted temperature monitoring, the limited case-study design underscores a critical need for broader, independent ground validation to verify absolute accuracy.Abstract Existing Google Earth Engine-based retrieval workflows often use fixed normalized difference vegetation index thresholds and coarse atmospheric water vapor inputs, which may limit their adaptability to regional surface and atmospheric conditions. This study evaluates how these two intermediate parameters influence Landsat 8 land surface temperature retrieval over northeastern Egypt using the generalized single-channel algorithm. Atmospheric water vapor was derived from MERRA-2 and NCEP reanalysis products, while land surface emissivity was estimated using ASTER Global Emissivity Dataset data and an NDVI-threshold framework. Reanalysis-derived total precipitable water was first compared with MODIS MOD05_L2. MERRA-2 showed a stronger correlation with MOD05_L2, whereas NCEP produced lower bias and root mean square error. The retrieved land surface temperature was then compared with the Landsat 8 Collection 2 Level 2 LST product as an internal consistency check. Using MERRA-2 reduced the overall root mean square error from 1.3977 K to 1.2615 K relative to NCEP, although it also increased the magnitude of the negative bias. A grid search of 24 normalized difference vegetation index threshold combinations showed that retrieval consistency was sensitive to threshold selection in cropland areas, while desert and built-up areas were largely insensitive. The best overall consistency with the Landsat product was obtained using a soil threshold of 0.20 and a vegetation threshold of 0.75, with a root mean square error of 1.2507 K and a bias of -0.7236 K. External validation at the Baseline Surface Radiation Network Gobabeb station showed a slight improvement when using MERRA-2 instead of NCEP, with root mean square error decreasing from 4.726 K to 4.441 K. Overall, the results show that intermediate parameter choices can affect Landsat land surface temperature retrieval, but the optimized settings should be interpreted as region-specific and relative to the Landsat product because independent validation remains limited.