Comparison of satellite-based models for estimating gross primary productivity in agroecosystems

Abstract

Satellite-based gross primary productivity (GPP) models have been widely used for simulating carbon exchanges of terrestrial ecosystems. However, the performances of various GPP models in agroecosystems have been rarely explored. In this study, we calibrated the model parameters and compared the performances of seven light use efficiency (LUE-GPP) models and five vegetation-index (VI-GPP) models for simulating daily GPP of agroecosystems over 106 crop growing seasons, and examined the effects of model structure on model performance. The simulations were carried out based on 19 eddy covariance (EC) sites from the global flux network and vegetation indices obtained from MODIS. The calibrated potential LUE (εmax) for C4 crop (summer maize, 2.59±0.94 g C MJ−1) was higher than that for C3 crops (1.42±0.58 g C MJ−1) in any LUE-GPP models. The performances of models differed across the crops. Generally, all models performed better for C3 crops than C4 crops, and for winter crops (winter wheat-Triticum aestivum L, rape-Brassica napus L, and winter barley-Hordeum vulgare L) than summer crops (summer maize-Zea mays L, potato-Solanum tuberosum L, rice-Oryza sativa L. and soybean-Glycine max (L.) Merr.). Cloudiness index-LUE (CI-LUE) model outperformed the other LUE-GPP models, and vegetation index (VEI) model outperformed the other VI-GPP models. LUE-GPP models demonstrated better performance than VI-GPP models due to the inclusion of water stress (Ws) and temperature stress (Ts). A comparison of the model structures showed that models only considering the effects of Ws produced smaller errors than those only considering the effects of Ts in simulating GPP. Ws algorithms generated the larger variations in LUE-GPP models compared to those of Ts, especially during the drought period. All models obtained higher R2 and smaller errors using the minimum method (Min (Ts, Ws)) than using the multiplication method (Ts × Ws) to integrate the effects of Ts and Ws on GPP, which suggested that the minimum method was better than the multiplication method to integrate Ts and Ws on LUE. These results showed that satellite-based models with calibrated crop-specific parameters have the potential to serve as the basis for estimation of agroecosystem GPP, and can provide direction for future model structure optimization.