Previous Page Table of Contents Next Page

MODIS Validation at Flux Tower Sites, Markus Reichstein and Steffen Gruenler

The use of eddy covariance tower sites data
to compare MODIS sensor GPP estimates


There are number of FLUXNET sites, where simultaneous observations of canopy structure, leaf-area index (LAI) development, net ecosystem exchange and soil respiration are made. These sites provide an excellent basis for validating the Moderate Resolution Imaging Spectroradiometer (MODIS) remote sensing products that are related to ecosystem structure and carbon and water balance. The MODIS-GPP/NPP algorithm integrates 1°×1°-assimilated meteorological data, the remotely sensed fraction of photosynthetically active radiation (fPAR) absorbed by the vegetation, LAI and landcover information to achieve a global estimate of daily to annual gross and net primary productivity (GPP; NPP) on a 1 km2 grid. Over homogeneous terrain (1 ha to 1 km2), the FLUXNET eddy covariance stations sample net ecosystem carbon exchange, which can be split into ecosystem respiration and GPP flux components. Hence, the FLUXNET eddy covariance stations provide an excellent means to evaluate the MODIS-GPP product and other remote-sensing-driven carbon balance estimates.


In GTOS, there is an ongoing effort to compare GPP estimated using the MODIS sensor with ground-observed data at various European FLUXNET/CARBOEUROPE eddy covariance tower sites. The sites selected for comparison range from 38° to 67° N and comprise boreal and temperate conifer forests (spruce, pine), temperate and Mediterranean deciduous forests (beech, oak), Mediterranean evergreen broadleaf forests, and a savannah-type Mediterranean ecosystem.

Figure 1. Net Primary Productivity

The analysis is based on the assumption that a sensible evaluation of the MODIS-GPP estimate must account for all error sources that occur during the computation. Thus, in a factorial approach, we analyse (1) the effect of driving the MODIS-GPP model with 1° by 1° assimilated meteorological data versus local meteorological data; (2) the error introduced by remotely sensed estimates of seasonal fPAR/LAI development; and (3) the bias introduced by the MODIS-GPP radiation-use-efficiency model itself.


Given the independent nature (not fitted against flux data) and the simplicity of the MODIS-GPP model, its overall performance in predicting GPP is remarkable under normal conditions (r2 between 0.7 and 0.95). The assimilated meteorology does not capture all day-to-day variation, but matches the local tower data well on an eight-day scale. However, at certain sites the meteorological bias influences estimates of GPP significantly. Particularly at high latitudes, the correction of cloud-contaminated fPAR/LAI values enhances GPP estimates considerably. At sites with understorey or a herbaceous spring layer, springtime GPP is often overestimated by the MODIS-GPP model since it cannot account for differences in radiation-use efficiency by canopy and understorey. Furthermore, there is potential for considerable improvements of the GPP algorithm by better accounting for soil drought effects, by reducing the radiation-use efficiency under high-radiation conditions, and by introducing more geo-biological variability. It has been shown that these parts of the MODIS-GPP algorithm can be re-parameterized using CARBOEUROPE eddy covariance data, so the synergistic use of MODIS and CARBOEUROPE data will improve the ability of a global terrestrial observation system.

Previous Page Top of Page Next Page