Abstract The modeling of Near-Surface Temperature Lapse Rate (NSTLR) is of great importance in various environmental applications. This study proposed a new approach for modeling the NSTLR based on the Normalized Land Surface Temperature (NLST). A set of remote sensing imagery including Landsat images, MODIS products, and ASTER Digital Elevation Model (DEM), land cover maps, and climatic data recorded in meteorological stations and self-deployed devices located in the three study area were used for modeling and evaluation of NSTLR. First, the Split Window (SW) and Single Channel (SC) algorithms were used to estimate LST, and the spectral indices were used to model surface biophysical characteristics. The solar local incident angle was obtained based on topographic and time conditions for different dates. In the second step, the NSTLR value was calculated based on the LST-DEM feature space at the regional scale. The LST was normalized relative to the surface characteristics based on Random Forest (RF) regression and the NSTLR was calculated based on the NLST-DEM feature space. Finally, the coefficient of determination (R2) and Root Mean Square Error (RMSE) between the modeled NSTLR and the observed NSTLR were calculated to evaluate the accuracy of the modeled NSTLR. The mean values of R2 between DEM and NLST were improved 0.3, 0.42 and 0.35, rather than between DEM and LST for the study area A, B and C, respectively. The R2 and RMSE between the observed NSTLR and the Landsat derived NSTLR based on NLST for the study area A (B, C) were improved 0.30 (0.26, 0.35) and 0.81 (0.80, 0.94) °C Km−1, respectively, rather than the Landsat derived NSTLR based on LST. Also, for the study area A, the R2 and RMSE between the observed NSTLR and the MODIS derived NSTLR based on NLST in spring, summer, autumn and winter were improved 0.17, 0.12, 0.10, and 0.22; and 0.51, 0.44, 0.27, and 0.51 °C Km−1, respectively, rather than the MODIS derived NSTLR based on LST. Model assessment results (R2 and RMSE) and comparing modeled NSTLR (all strategies) with observed NSTLR, for both Landsat and MODIS, showed that the use of NLST instead of LST, significantly improved the accuracy of the obtained NSTLR in the mountainous regions.

中文翻译：

---

基于归一化地表温度数据的近地表温度递减率建模新方法

摘要 近地表温度衰减率 (NSTLR) 的建模在各种环境应用中具有重要意义。本研究提出了一种基于归一化地表温度 (NLST) 对 NSTLR 进行建模的新方法。使用包括Landsat影像、MODIS产品和ASTER数字高程模型（DEM）在内的一套遥感影像、土地覆盖图以及位于三个研究区的气象站和自部署设备记录的气候数据进行建模和评估NSTLR 的。首先，分裂窗口 (SW) 和单通道 (SC) 算法用于估计 LST，光谱指数用于模拟表面生物物理特性。太阳局地入射角是根据不同日期的地形和时间条件获得的。第二步，NSTLR 值是基于区域尺度的 LST-DEM 特征空间计算得出的。LST 基于随机森林 (RF) 回归相对于表面特征进行归一化，NSTLR 基于 NLST-DEM 特征空间计算。最后，计算建模的 NSTLR 和观察到的 NSTLR 之间的决定系数 (R2) 和均方根误差 (RMSE)，以评估建模的 NSTLR 的准确性。DEM 和 NLST 之间的 R2 平均值分别提高了 0.3、0.42 和 0.35，而不是研究区域 A、B 和 C 之间的 DEM 和 LST。研究区 A (B, C) 观测到的 NSTLR 与基于 NLST 的 Landsat 导出的 NSTLR 之间的 R2 和 RMSE 分别提高了 0.30 (0.26, 0.35) 和 0.81 (0.80, 0.94) °C Km−1，而不是比基于 LST 的 Landsat 派生 NSTLR。还，对于研究区A，春、夏、秋、冬季观测到的NSTLR与基于NLST的MODIS推导的NSTLR之间的R2和RMSE分别提高了0.17、0.12、0.10和0.22；和 0.51、0.44、0.27 和 0.51 °C Km-1，而不是基于 LST 的 MODIS 导出的 NSTLR。Landsat 和 MODIS 的模型评估结果（R2 和 RMSE）以及模拟的 NSTLR（所有策略）与观测的 NSTLR 的比较表明，使用 NLST 代替 LST，显着提高了山区获得的 NSTLR 的准确性。