National Aeronautics and Space Administration

Figure 1. The conceptual logic we are following for validation of
MOD17 GPP with fluxtower data where tower meteorology, ground
measured FPAR and more rigorous ecosystem models are
sequentially substituted to evaluate sources of error and variability
in MOD17 (based on Running et al. 1999).

Stage 3 validation has been achieved for the MODIS GPP product.

The original strategy for validation of the MOD17 GPP and NPP data, devised an overall plan to integrate MODIS Land data into global terrestrial monitoring Running et al. (1999). This paper laid the foundation for the interaction between the MODIS land team and the eddy covariance fluxtower community, illustrating the mutual advantage of fluxtowers providing direct measurement of carbon and water fluxes, while MODIS land products provided a spatial extrapolation of the tower data, Figure 1.

Figure 2. Annual tower GPP against MOD17A2 GPP for sites in 6
biome types. See Wang et al. (2017) for details.

Flux towers directly measure the CO2 flux, so the most direct comparison for MOD17 is with GPP, rather than NPP, a biomass measure.

A number of recent papers have used this type of input swapping logic to test MOD17 in multiple biome types. Wang et al. (2017) tested MOD17A2H, Collection 6 across 18 tower sites in 6 biomes (Figure 2), and found that much of the error at sites with GPP above 1000gCm^-2yr^-1 could be traced to cloud contaminated input data from MODIS LAI/FPAR (MOD15). With better LAI/FPAR data, correlation results were much stronger.

Figure 3. GPP from fluxtowers of forests worldwide compared
to MOD17A2. See Tang et al. (2015) for details.

Another study of grassland biomes has been published by Zhu et al. (2018), using MOD17A2 compared against flux tower data, see Figure 4.

A similar study by Tang et al. (2015) with 16 forest fluxtower sites in North America and Europe, tested MOD17A2 annual GPP against tower data. The sites included evergreen needleleaf, evergreen broadleaf and deciduous broadleaf forests and used Collection 5.5 data. Overall results for annual GPP can be found in Figure 3. Another study of grassland biomes has been published by Zhu et al. (2018), using MOD17A2 compared against flux tower data, see Figure 4.

Each of these studies, Tang et al. (2015), Wang et al. (2017), and Zhu et al. (2018), identified poor MOD15 data, predominantly from cloud contamination, as the source of most error in the MOD17 results. These errors are amplified in warm tropical regions that have both high NPP and high cloud cover.

Figure 4. GPP from MOD17 compared against fluxtower GPP for grasslands in tropical, temperate and alpine
environments. See Zhu et al. (2018) for details.