High spatial resolution vegetation gross primary production product: Algorithm and validation

Vegetation gross primary production (GPP) in terrestrial ecosystems is a key element of the carbon cycle, and its estimation highly determines the accuracy of carbon budget assessments. Currently, several global datasets of vegetation production are available at coarse or moderate spatial resolutions, but globally they still have large uncertainties, which hindered the application of GPP, especially in strongly heterogeneous agriculture ecosystems and mountainous areas. Here, we used the Markov chain Monte Carlo (MCMC) approach with the footprints of FLUXNET data to optimize the parameters of the high resolution of Global LAnd Surface Satellite (Hi-GLASS) GPP algorithm (i.e., the revised EC-LUE model) using 30 m spatial resolution Landsat data as driving data. Then, we generated a new set of algorithm parameters for high resolution GPP estimates. We used the optimized parameters with integrating footprint to calculate Hi-GLASS GPP based on Landsat data and our results revealed that on average, Hi-GLASS GPP explained 76% of variance in tower GPP at the total of 78 sites across ten vegetation types. Moreover, compared with previous 500 m GPP product such as GLASS GPP and MODerate Resolution Imaging Spectroradiometer (MODIS) GPP, our optimized Hi-GLASS algorithm using Landsat data had large superiority in simulating GPP for wetlands, savannas, shrubland and C3, C4 cropland ecosystems, and had a slightly improvement for deciduous broadleaf forests and evergreen broadleaf forest ecosystem. Our study is an effort to optimize and quantify parameter uncertainty of Hi-GLASS algorithm using high spatial resolution (30 m) Landsat data and improve the high resolution GPP estimation for better understanding global ecosystem carbon dynamics and carbon-climate feedbacks.