Separating land surface component temperatures from Low Earth Orbit (LEO) satellite data by coupling of dual-time and multi-pixel data

The component temperature encapsulates more physical meaning than Land Surface Temperature (LST) and better meets the requirements of estimating evapotranspiration, monitoring drought and other studies. The polar-orbit satellites can observe the entire globe with a high spatial resolution and a modest temporal resolution from 1980 to present, and therefore have more wide applications than geostationary satellites. For these reasons, the study focuses on the methodology for estimating vegetation and soil component temperatures from polar-orbit satellite data.To meet operational and accurate requirements, the study proposed to use multi-temporal and multi-pixel data to separate the vegetation and soil component temperature. Specifically, a well-studied Diurnal Temperature Cycles (DTC) model was applied to link the two observations on one day, and then the moving-window technology was used to add available observations for solving the retrieval model. In addition, a spatial weighting matrix was adopted to improve the limitation of using multi-pixel data.The proposed algorithm was implemented by using Moderate Resolution Imaging Spectroradiometer (MODIS) data, and was evaluated by using in-situ measurements on Skukuza site and high-resolution Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) data, respectively. In the case of the validation of field data, the separation accuracy of component temperatures is about 2 K, and RMSEs of daytime vegetation, nighttime vegetation, daytime soil, and nighttime soil are 2.3 K, 2.5 K, 1.5 K and 1.9 K, respectively. The better performance at daytime is resulted from the fact that DTC model cannot describe the temperature decrease at night well. Regarding with the validation of ASTER data, the separation accuracies of the vegetation and soil component are 1.4 K and 1.7 K, respectively. The vegetation component is slightly overestimated (bias = 0.3 K) while the soil component is slightly underestimated (bias = -0.7 K), which is because of the systematic error between MODIS LST and ASTER LST. Moreover, this study also analyzed the influence of different time groups. Firstly, the combinations of one daytime moment and one nighttime moment can provide same estimation with high accuracy while the performance of the combination of two daytime moments is worse. The result is expected because two daytime moments are close to the maximum temperature moment, and therefore more sensitivity to temperature variation. Secondly, the performance of the time group from two sensors or one sensor is basically same, indicating that the time group is not limited by the sensor.This study proposed an algorithm for separating vegetation and soil component temperatures from polar-orbit satellite land surface temperatures. The practical method need only two observations from single or different sensors, i.e., one in daytime and the other one in nighttime, which makes it available for almost all sensors. The validation of field data and high-resolution data indicated that the separation accuracy is about 2 K and the best up to 1.4 K. Considering its accuracy, operationality and robustness, the proposed method would be an effective tool for separating component temperatures. © 2021, Science Press. All right reserved.