A comparative assessment of the accuracies of split-window algorithms for retrieving of land surface temperature using Landsat 8 data

Abstract

Land surface temperature (LST) is one of the most important physical parameters examined in modeling the land surface processes. There are several algorithms for estimating the LST using satellite imagery. The present study aims to evaluate the accuracy of ten split-window algorithms in estimating the LST using Landsat 8 images in the arid lands. The split-window algorithms were validated by employing the two methods of temperature-based (T-based) and cross-validation. In the T-based validation method, soil temperatures at depth of 5 cm within three meteorological stations were used at 6:30, 12:30, and 18:30 (local time). The land surface temperature was predicted at the moment when satellite overpass by implementing Fourier series. Results of the T-based validation indicated that the Li and Coll algorithms with RMSE values of 5.83 and 8.94 and MADE of 4.60 and 8.04 °C, have the lowest and highest errors, respectively. To conduct the cross-validation and prepare the RMSE statistical index images, the LST images associated with MODIS sensor and those of various split-window algorithms obtained via Landsat images were compared. Results of land-use analysis showed that areas with the RMSE of more than 5 °C, are located in valleys and regions with high humidity. The results of cross-validation showed that the Li and Jimenez’s split-window algorithms are the most accurate methods with RMSEs of 3.65 and 3.57 °C, respectively. The Parata, Price, Sobrino, Uliverii, Mcclain, Vidal, Kerr, and Coll algorithms are in the next grades, respectively. In general, using atmospheric water vapor increases the accuracy of retrieval LST in a split-window algorithm.