Coupled retrieval of turbulent heat fluxes and gross primary productivity via the assimilation of land surface temperature data from geostationary satellites

Abstract

Compared with those provided by polar-orbiting satellites/sensors (e.g., Landsat/MODIS), new-generation geostationary satellites, such as Himawari-8 and Geostationary Operational Environmental Satellite-R series (GOES-R), offer temporally continuous and much more frequent observations of the land surface over the course of the diurnal cycle. In this study, Himawari-8 land surface temperature (LST) data and GLASS leaf area index (LAI) were assimilated into a coupled two-source surface energy balance–vegetation dynamics model (TSEB-VDM) via the variational data assimilation (VDA) method to retrieve the regional sensible heat flux (H), latent heat flux (LE), and gross primary productivity (GPP) (hereafter referred to as the VDA_Himawari-8 scheme). Regional H, LE, and GPP values were estimated across the Heihe River Basin (HRB) in northwestern China, with a spatial resolution of 0.02° × 0.02°. Moreover, LST data from the MODIS, which is an instrument onboard polar-orbiting satellites (i.e., Terra and Aqua), were assimilated into the TSEB-VDM model (hereafter referred to as the VDA_MODIS scheme) for comparison with the VDA_Himawari-8 scheme. Four unknown parameters of the TSEB-VDM model, i.e., the neutral bulk heat transfer coefficient (C_HN), the evaporative fractions of the soil and canopy (EF_S and EF_C, respectively), and the specific leaf area, were optimized via the VDA approach. The estimated H and LE values were validated against ground measurements from the large-aperture scintillator, and the GPP estimates were validated against eddy covariance data at three sites (Arou, Daman, and Sidaoqiao) in the HRB. The results indicated that the assimilation of geostationary satellite-based LST data significantly improved the performance of the VDA model. The three-site-average root mean square errors (RMSEs) of the H, LE, and GPP estimates from the VDA_MODIS scheme were 45.27 W m⁻², 83.77 W m⁻², and 3.30 g C m⁻² d⁻¹, respectively. Compared with the VDA_MODIS scheme, the VDA_Himawari-8 scheme notably enhanced the performance of the VDA framework and decreased the RMSE by 15.0 % for H, 22.5 % for LE, and 38.5 % for the GPP, with an especially notable enhancement in humid (or vegetated) areas. The primary factors contributing to the enhancement in the performance of the VDA framework were the availability of much more frequent LST observations and the diurnal sampling capability of the Himawari-8 satellite.