Choosing the most appropriate split-window algorithms for estimating land surface temperature by use of MODIS sensor case study: Kerman Plain

The use of remote sensing data for land surface temperature estimation is regarded as a new method which significantly reduces costs of temperature estimation in classical method. For this purpose, for checking the land surface temperature in Kerman plain, 12 cloudless images from MODIS sensor in the summertime of 1392 and 10 split-window algorithm were used to estimate the land surface temperature. Also, to compare these split-window algorithms, the temperature in 5cm soil depth at satellite passing moment in 12 selected days was estimated by a digital thermometer and a handheld GPS device in the observed points. The results showed that MODIS sensor has the ability to detect heterogeneous surfaces and peaks and is good at Fractionation compliance reflective and thermal images. The model of Cole and Casillas (1978) with MARE of 4/762 and MBE of -0/075 has the highest accuracy among the other models. The Price model with MAE of 4/987 and MBE of 0/0817 is located on the second place with accuracy close to the Cole model. Bikroly model (1990) has the lowest accuracy among 10 algorithm, with MAE of 6/214 and MBE of 3/739. R2 of 0/8091 also shows the accuracy of Cole and Casillas (1978) algorithm between the results derived from ground conclusion and the temperature derived from satellite data. Then , thermal classification map of Kerman county plain were obtained by use of Support Vector Machine method which is a powerful tool in the Zoning.