Analysis and reduction of topographic effect induced errors in land surface temperature retrieval over the Tibetan Plateau

Land Surface Temperature (LST) plays a pivotal role in representing ground energy balance and understanding climate change. LST over the Tibetan Plateau (TP) significantly influences regional climate and environmental dynamics. Accurate LST data in the TP is vital for ecological monitoring and climate studies. However, most retrieval algorithms assume a flat-surface thermal infrared radiation transfer equation (TIRTE), which introduces inevitable topographic induced in the TP’s complex terrain. Additionally, the limited and sparse ground stations hinder pixel-level error analysis. These limitations restrict accurate characterization of topographic effects on LST errors and impede the effective application of current LST datasets. This study proposed a method to quantify the LST retrieval errors at a pixel level and innovatively introduced the radiation-topographic bias correction term (RTBC). The effectiveness of RTBC in reducing LST retrieval errors with only one atmospheric parameter was demonstrated theoretically. Random forest (RF) models were employed to assess the contribution of topographic effects to these errors. The sky view factor (SVF) was employed as an indicator of surface ruggedness. The results demonstrated that LST retrieval errors were predominantly due to topographic effect when surface ruggedness was high (SVF ≤ 0.25), with an R² value reaching up to 0.86. RTBC emerged as the primary factor influencing LST retrieval errors at SVF ≤ 0.25. In-situ LST analysis showed that when SVF decreased to 0.738, RTBC effectively reduced the root mean square error (RMSE) and the mean absolute error (MAE) and by an average of 1.2 K and 1.1 K, respectively. In comparison experiments with conventional methods, RTBC achieved approximately a 50 % reduction in RMSE. These findings highlight the significant impact of topography on LST retrieval accuracy and demonstrate the effectiveness of RTBC in reducing terrain-induced errors.