Seasonal course of a normalized differential vegetation index ‘NDVI’ derived from tower data

Vegetation Indices (VIs) like the Normalized Differential Vegetation Index (NDVI) are widely used to monitor seasonal, interannual, and long-term variations of structural, phenological, and biophysical parameters of land surface vegetation cover. They are spectral transformations of at least two spectral bands, chosen specifically to enhance the contribution of vegetation properties to surface reflectances. Remote sensing products generally produce information on GPP (or net primary productivity, NPP), in terms of a light use efficiency (ε) and the amount of absorbed visible sunlight

GPP=ε(T,θ,D)*fpar*Q_p (1)

In practice, light use efficiency (ε) is adjusted for seasonal changes in soil moisture, temperature and vapor pressure deficit and the fraction of absorbed photosynthetically active radiation, fpar, is inferred from the NDVI.

With the FLUXNET network database spanning a wide range of plant functional types, disturbance features, and climates we have the ability to revise and improve upon Equation 1 as a tool for converting remote sensing information to terrestrial biosphere carbon flux information. In particular we can explore the modulating effects of direct and diffuse radiation, temperature acclimation, soil water deficits, frost/freezing, phenology and growth on light use efficiency (e) using continuous and direct field measurements. And at field sites with up and downwelling quantum sensors, we can evaluate how well fpar is being assessed by the satellites.

At many FLUXNET sites we can assess a broad-band version of the normalized difference vegetation index (NDVI) using reflectance measurements of visible (Q_par) and shortwave (Rg) solar radiation to represent contributions from reflected near infrared and visible radiation, which in turn scales with fpar and leaf area index:

NDVI_tower = [(Rg - Q_par)|_nir - Q_par|_vis] / [(Rg - Q_par)|_nir + Q_par|_vis] (2.a)

Which simplifies to

NDVI_tower = 1 – 2 * Q_par / Rg (2.b)

While we measure Rg with a pyranometer and Qp with a quantum sensor, work by Ross and Sulev (2000) indicates that we can convert the reflected quantum flux density to an energy flux density with a conservative conversion factor (4.6 mmol m^-2 s^-1 (W m^-2)^-1). With this approach we are investigating the following hypotheses / questions:

· Use tower based NDVI to validate seasonality of MODIS and other remote sensing vegetation indices and products driven by NDVI or EVI?

· Does the mismatch between MODIS products from clear periods on one hand and tower NDVI and flux measurements for both overcast and clear condition on the other introduce a scaling bias, i.e. are there spectral differences in surface reflectances for direct and indirect radiation?

· Can we use tower NDVI and flux estimates to verify whether the tower fluxes are representative to the smallest MODIS grid scale, i.e. is the seasonality of reflectance detected by MODIS representative for the flux tower site? the eddy flux measurements have the same time and space stamp as the tower-based NDVI;

· Can we compare response functions from satellites with Tower data as a means of validation or even improve parameterizations that are currently in use? Several issues are investigated here: a) to test the GPP algorithm under both clear and cloudy skies, b) direct measurements of canopy carbon assimilation are used and c) the acquisition of long term data records at numerous field sites should allow us to quantify the effects of phenology, canopy structure, plant functional type, drought, and high and low temperature stresses on terms in Eq. 1.

Since FLUXNET Data Information System (DIS) is cataloguing the MODIS fpar, NDVI and the enhanced vegetation index (EVI) measurements around each tower site, we will develop transfer functions between the tower-based measurement of NDVI and satellite-based estimates of fpar, EVI and NDVI.

As a first example of this approach we compared data from the AMERIFLUX site at Vaira, CA, USA, an annual grassland that is surrounded by oak savanna with the MODIS NDVI product. During the winter and autumn, when the trees are leafless, the tower-based and MODIS-scale estimates of NDVI agree well. During the late spring and summer period, when the trees are green and the grass is dead, there is some disagreement between the two indices. Obviously, a better understanding of the sub-pixel heterogeneity of landscapes will be critical for utilizing Eq. 1 to compute local and regional scale carbon fluxes with MODIS.