Improving the global MODIS GPP model by optimizing parameters with FLUXNET data

Abstract

The global gross primary productivity (GPP) product derived from the Moderate Resolution Imaging Spectroradiometer (MODIS) is perhaps the most widely used GPP product. However, there is still a large uncertainty associated with the MODIS GPP product partly due to the uncertainty in the default Biome specified Parameters Look-Up Table (BPLUT) of the MODIS photosynthesis (PSN) model. Here, we used the Bayesian inference with the Markov chain Monte Carlo (MCMC) approach and FLUXNET data from 110 sites to estimate the parameters of the MODIS PSN model (maximum light use efficiency: ɛmax; temperature scalar-related parameters: Tminmin and Tminmax; water scalar-related parameters: VPDmin and VPDmax) through individual and joint optimization. The spread of the posterior probability density function (PDF) of the parameters allowed for the calculation of parameter means and uncertainty estimates and also provided information on the behavior of the parameters. Each model parameter varied not only across sites but also across plant functional types (PFTs). The means of the optimized parameter values within each PFT were used to update the BPLUT. We also generated parameter estimates for wetlands and C4/C3 croplands in the BPLUT. Parameters from the joint optimization were more representative and less variable. The optimization improved the performance of the MODIS PSN model by 15% for deciduous broadleaf forests, 8% for savannas, and 3% for grasslands with well-constrained parameters. The performance of the optimized model depended on the effectiveness of parameter optimization. Our study is an effort towards quantifying and reducing parameter uncertainty of the MODIS PSN model and improving the global MODIS GPP product for better understanding global ecosystem carbon dynamics, plant productivity, and carbon-climate feedbacks.