Mapping regional evapotranspiration in cloudy skies via variational assimilation of all-weather land surface temperature observations

Recently, a number of studies estimated evapotranspiration (ET) via variational assimilation of land surface temperature (LST) data from Moderate Resolution Imaging Spectroradiometer (MODIS). Their unknown parameters are neutral bulk heat transfer coefficient (C_HN) (that scales the sum of sensible and latent heat fluxes), and evaporative fraction (EF) (that represents the partitioning of available energy between sensible and latent heat fluxes). The variational data assimilation (VDA) approaches estimate C_HN and EF by minimizing the difference between the MODIS LST observations and model estimates. The applicability of these studies is limited only to clear-sky conditions where MODIS LST data are available. Even when the sky is clear, they cannot robustly update the initial guess of EF (i.e., the a priori EF value) because of the low temporal resolution of MODIS LST data. This study overcomes these shortcomings by 1) using the random forest (RF) method to obtain a reasonable the a priori EF value, and 2) assimilating the all-weather LST data (which are obtained by merging thermal infrared and passive microwave observations) into the VDA approach. The VDA approach is applied to the Source Regions of Rivers (SRR) in southwest China with heavy cloud covers. Results show that the RF method obtains a reasonably accurate the a priori EF value. Compared to assimilating the MODIS LST product, assimilation of all-weather LST data lead to an improvement in the ET estimates, especially in regions with dense clouds. Comparison of ET estimates with the measurements at four sites (i.e., Dangxiong, Linzhi, Naqu, and Qomolangma) in the SRR shows that the VDA approach can accurately estimate ET in cloudy conditions. Finally, the three-cornered hat (TCH) method is employed to assess the relative uncertainty of ET estimates over the SRR.