An Alternative Scaling Method for Cross-Validation of Land Surface Temperature from Remote Sensing Data: A Case Study of Fars Province

Land surface temperature (LST) is one of the key parameters in a wide range of geosciences and environmental studies. Remote sensing technology, can monitor the space and time variations of this quantity in large areas. But evaluating the accuracy of this parameter is the most challenging issues. On the one hand, rapid changes in temperature occur in space and time and the mismatch between the spatial scale of satellite and ground-based sensors, validation of LST is difficult by ground based temperature data. Validation is a process of independently assessing the uncertainty of the data derived from the system outputs. Without validation, no methods, algorithms, or parameters derived from remotely sensed data can be used with confidence. As the retrieved LSTs from satellite TIR data involve corrections to the satellite-observed radiances to account for atmospheric effects and non-unity LSEs, it is necessary to assess the accuracy of the retrieval to provide potential LST users with reliable information regarding the quality of the LST product. However, validation of satellite-derived products is often a challenge in itself. This is true for the sensors having a coarse resolution and even more with the thermal bands over land. Although many algorithms have been proposed and developed over recent decades to retrieve the LST from satellite TIR data, validation of satellite-derived LSTs remains far behind the development of new LST retrieval algorithms due to the difficulties. The important difficulties including: (i) it is difficult to obtain representative LST data at the scale of a pixel, which usually covers an area of about several square kilometers, due to the spatial variability of landscape and the large spatial–temporal variations in the LST itself. (ii) The other difficulty concerns the temporal sampling. Surface temperature is characterized by a very high temporal frequency. In a matter of seconds, the surface skin temperature may change by several degrees due to wind, shadow, etc. Generally, ground radiometers often integrate the temperature over several minutes, but the sensor onboard the satellite acquires the data in a fraction of a second. (iii) In addition, it is difficult to conduct in situ LST measurements at local scale, because LST, LSE, and the atmospheric and environmental radiances are coupled. Therefore, to obtain LST, one has to measure simultaneously the atmospheric and environmental radiances and the LSE, which is time-consuming and hard to monitor. Although there exist many problems and difficulties in the validation of the LST derived from satellite TIR data, in recent years, many efforts and studies have been performed to validate the derived LSTs from different satellite sensors. The methods utilized to validate LST values retrieved from space may be roughly grouped into four categories: the temperature-based method (T-based), the radiance-based method (R-based), the cross-validation method, and the indirect validation method. Among these methods, due to development of thermal sensors, cross-validation procedure is the most common method to evaluate the accuracy of surface temperature. Due to the large spatial and temporal variations in the LST, geographic coordinate matching, time matching, and view zenith angle matching are basic requirements of the cross-validation. In this paper, an alternative scaling method was proposed for cross-validation between LST product of LDCM and MODIS sensors. Because of the wide coverage and taken LST product twice per day by each of the MODIS sensors (Terra and Aqua), this product was selected as the reference temperature. The LST products of MODIS sensors have been validated with in situ measurements and by various methods in more than 50 clear-sky cases taking into account the higher accuracies within 1 °K for both Terra and Aqua. The results of the proposed method showed that evaluation of LST accuracy in areas with high homogeneity and in the first examined LDCM scene, in terms of mean differences and mean absolute differences measures, is 0.6 °K and 1.63 °K, respectively. For the second examined LDCM scene, the accuracy measures were 0.94 °K and 1.27 °K in the LST, respectively. Generally, the experimental results on two LDCM data demonstrated that the proposed cross-validation method not only is a robust and accurate LST validation approach, but also it is applicable for each thermal sensor and every time and place.