Assessments of gross primary productivity estimations with satellite data-driven models using eddy covariance observation sites over the northern hemisphere

Abstract

The accurate quantification of gross primary productivity (GPP) has been a major challenge in global climate change research. Satellite data-driven models have been universally used as scientific tools for investigating the carbon cycle, including vegetation index (VI)-based models, light use efficiency (LUE) models, and process-based models. However, inconsistencies and uncertainties have been found in the GPP estimations from various models. The understanding of model behaviors under different climatic conditions remains unclear. In this study, three typical satellite data-driven models, namely, Moderate Resolution Imaging Spectroradiometer (MODIS) GPP (MOD17) model, Temperature and Greenness (TG) model and Boreal Ecosystem Productivity Simulator (BEPS), respectively, were compared to better understand discrepancies and uncertainties in GPP estimations at 119 northern eddy covariance (EC) sites. Due to the variations in climatic drivers of GPP, temperature, precipitation and incoming solar radiation were selected to describe climatic conditions. The results showed that BEPS and MOD17 exhibited similar performance in simulating GPP, with root-mean-square error (RMSE) values of 2.50 g C m−2 d−1 and 2.53 g C m−2 d−1, respectively, and performed slightly better than TG (RMSE = 2.98 g C m−2 d−1). Comparison between simulated GPP and EC GPP also revealed that model performance varied substantially among different vegetation types. The three models performed better for deciduous broadleaf forest, evergreen needleleaf forest, and mixed forest, in comparison to the results from evergreen broadleaf forest and crop. Specifically, all three models showed poor performance under the conditions of high temperature and low precipitation, revealing the models’ inability to characterize the impact of water stress on photosynthesis when drought occurs. Furthermore, our results indicated that GPP estimations from satellite data-driven models were also sensitive to remotely sensed data, suggesting that the high accuracy of remotely sensed data in describing vegetation canopy is important for carbon modeling. This study highlights the importance of understanding model behaviors in different vegetation types and climatic conditions, so that the model performances may be improved in future carbon cycle studies.