Retrievals of all-weather daytime air temperature from MODIS products

It is well known that remote sensing techniques hold the potential to explore the spatial estimation of air temperature (Ta) with fine spatial and temporal resolution across the world. However, because of the complex interaction of land-atmosphere system and the contamination of cloud cover, the retrieval of daytime Ta exclusively from remote sensing data is still far from straight forward, especially under cloudy sky conditions. In this paper, we presented a simple parameterization scheme of daytime Ta under all-weather conditions entirely based on Moderate Resolution Imaging Spectroradiometer (MODIS) products. To evaluate its applicability, the scheme was demonstrated in two regions with totally different geomorphological and climatic conditions, the east part of the Qaidam Basin (EQB) in China and the Southern Great Plains (SGP) in the United States of America. The instantaneous Ta under clear sky conditions (Ta , clear) was determined as the average of near surface air temperature (TaS) retrieved from MOD07_L2 product and land surface temperature (Ts) retrieved from MOD06_L2 product. Then a regression model between Ta , clear and Ts was established, and the instantaneous Ta under cloudy sky conditions (Ta , cloudy) was estimated by applying the regression model to Ts retrieved under cloudy sky conditions. The results showed that the averaging parameterization scheme has significantly improved the accuracy of Ta , clear retrievals with MAE = 1.95 °C, RMSE = 2.50 °C, and B = 0.02 in the EQB, and MAE = 2.02 °C, RMSE = 2.56 °C, and B = 0.01 in the SGP. The Ta , cloudy estimates also showed good agreement with Ta observations in both regions with a correlation coefficient (r) higher than 0.91. The values of RMSE calculated for the EQB and SGP were 3.42 °C and 2.91 °C, respectively. The accuracy of both Ta , clear and Ta , cloudy estimates has reached a level comparable with other traditional statistical approaches that adopt ancillary Ta measurements as training dataset. Therefore, it is feasible to estimate daytime Ta under all-weather conditions entirely based on MODIS products.