Consistency between In Situ, Model-Derived and High-Resolution-Image-Based Soil Temperature Endmembers: Towards a Robust Data-Based Model for Multi-Resolution Monitoring of Crop Evapotranspiration

摘要

Due to their image-based nature, “contextual” approaches are very attractive to estimate evapotranspiration (ET) from remotely-sensed land surface temperature (LST) data. Their application is however limited to highly heterogeneous areas where the soil and vegetation temperature endmembers (Tends) can be observed at the thermal sensor resolution. This paper aims to develop a simple theoretical approach to estimate Tends independently from LST images. Soil Tends are simulated by a soil energy balance model forced by meteorological data. Vegetation Tends are obtained from soil Tends and air temperature. Model-derived soil Tends are first evaluated with in situ measurements made over an irrigated area in Morocco. The root mean square difference (RMSD) between modeled and ground-based soil Tends is estimated as 2.4 ∘ C. Model-derived soil Tends are next compared with the soil Tends retrieved from 90-m resolution ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) data collected over two irrigated areas in Mexico and Spain. Such a comparison reveals a strong consistency between model-derived and high-resolution image-based soil Tends. A recent contextual ET model (SEB-1S) is then applied to 90-m resolution and to 1-km resolution (aggregated) ASTER data using the model-derived or image-based Tends as the input. The RMSD between 90-m resolution SEB-1S and in situ ET is estimated as 65 and 82 W·m - 2 , and the RMSD between 1-km resolution SEB-1S and aggregated SEB-1S ET is estimated as 78 and 56 W·m - 2 for the image-based and model-derived Tends, respectively. In light of the above results, Tends should be estimated a priori when contextual models are applied to low resolution images. Moreover, the consistency over highly heterogeneous areas between model-derived and high-resolution image-based Tends provides a meaningful basis for developing mixed modeling observational approaches.