Uncertainty in simulating gross primary production of cropland ecosystem from satellite-based models

Authors

Wenping Yuan, Beijing Normal UniversityFollow
Wenwen Cai, Beijing Normal University
Anthony L. Nguy-Robertson, University of Nebraska-LincolnFollow
Huajun Fang, Chinese Academy of Sciences, BeijingFollow
Andrew E. Suyker, University of Nebraska-LincolnFollow
Yang Chen, Beijing Normal University
Wenjie Dong, Beijing Normal University
Shuguang Liu, Central South University of Forestry and Technology, Changsha
Haicheng Zhang, Beijing Normal University

Document Type

Article

Date of this Version

4-2015

Citation

Published in Agricultural and Forest Meteorology 207 (2015), pp. 48–57; doi: 10.1016/j.agrformet.2015.03.016

Comments

Copyright © 2015 Elsevier B.V. Used by permission.

Abstract

Accurate estimates of gross primary production (GPP) for croplands are needed to assess carbon cycle and crop yield. Satellite-based models have been developed to monitor spatial and temporal GPP patterns. However, there are still large uncertainties in estimating cropland GPP. This study compares three light use efficiency (LUE) models (MODIS-GPP, EC-LUE, and VPM) with eddy-covariance measurements at three adjacent AmeriFlux crop sites located near Mead, Nebraska, USA. These sites have different croprotation systems (continuous maize vs. maize and soybean rotated annually) and water management practices (irrigation vs. rainfed). The results reveal several major uncertainties in estimating GPP which need to be sufficiently considered in future model improvements. Firstly, the C4 crop species (maize) shows a larger photosynthetic capacity compared to the C3 species (soybean). LUE models need to use different model parameters (i.e., maximal light use efficiency) for C3 and C4 crop species, and thus, it is necessary to have accurate species-distribution products in order to determine regional and global estimates of GPP. Secondly, the 1 km sized MODIS fPAR and EVI products, which are used to remotely identify the fraction of photosynthetically active radiation absorbed by the vegetation canopy, may not accurately reflect differences in phenology between maize and soybean. Such errors will propagate in the GPP model, reducing estimation accuracy. Thirdly, the water-stress variables in the remote sensing models do not fully characterize the impacts of water availability on vegetation production. This analysis highlights the need to improve LUE models with regard to model parameters, vegetation indices, and water-stress inputs.