Chin J Plant Ecol ›› 2017, Vol. 41 ›› Issue (3): 337-347.DOI: 10.17521/cjpe.2016.0182

Special Issue: 生态遥感及应用

• Research Articles • Previous Articles     Next Articles

Application and comparison of remote sensing GPP models with multi-site data in China

Ke-Qing WANG, He-Song WANG*(), Osbert Jianxin SUN   

  1. College of Forestry, Beijing Forestry University, Beijing 100083, China
  • Online:2017-03-10 Published:2017-04-12
  • Contact: He-Song WANG
  • About author:KANG Jing-yao(1991-), E-mail:


Aims Estimation of gross primary productivity (GPP) of vegetation at the global and regional scales is important for understanding the carbon cycle of terrestrial ecosystems. Due to the heterogeneous nature of land surface, measurements at the site level cannot be directly up-scaled to the regional scale. Remote sensing has been widely used as a tool for up-saling GPP by integrating the land surface observations with spatial vegetation patterns. Although there have been many models based on light use efficiency and remote sensing data for simulating terrestrial ecosystem GPP, those models depend much on meteorological data; use of different sources of meteorological datasets often results in divergent outputs, leading to uncertainties in the simulation results. In this study, we examines the feasibility of using two GPP models driven by remote sensing data for estimating regional GPP across different vegetation types. Methods Two GPP models were tested in this study, including the Temperature and Greenness Model (TG) and the Vegetation Index Model (VI), based on remote sensing data and flux data from the China flux network (ChinaFLUX) for different vegatation types for the period 2003-2005. The study sites consist of eight ecological stations located in Xilingol (grassland), Changbaishan (mixed broadleaf-conifer forest), Haibei (shrubland), Yucheng (cropland), Damxung (alpine meadow), Qianyanzhou (evergreen needle-leaved forest), Dinghushan (evergreen broad-leaved forest), and Xishuangbanna (evergreen broad-leaved forest), respectively. Important findings All the remote sensing parameters employed by the TG and VI models had good relationships with the observed GPP, with the values of coefficient of determination, R2, exceeding 0.67 for majority of the study sites. However, the root mean square errors (RMSEs) varied greatly among the study sites: the RMSE of TG ranged from 0.29 to 6.40 g·m-2·d-1, and that of VI ranged from 0.31 to 7.09 g·m-2·d-1, respectively. The photosynthetic conversion coefficients m and a can be up-scaled to a regional scale based on their relationships with the annual average nighttime land surface temperature (LST), with 79% variations in m and 58% of variations in a being explainable in the up-scaling. The correlations between the simulated outputs of both TG and VI and the measured values were mostly high, with the values of correlation coefficient, r, ranging from 0.06 in the TG model and 0.13 in the VI model at the Xishuangbanna site, to 0.94 in the TG model and 0.89 in the VI model at the Haibei site. In general, the TG model performed better than the VI model, especially at sites with high elevation and that are mainly limited by temperature. Both models had potential to be applied at a regional scale in China.

Key words: gross primary productivity, remote sensing, enhanced vegetation index, land surface temperature, eddy covariance