Sampling Design and Implementation
The BigFoot sample design calls for 100 ground validation measurements of land cover, LAI, fAPAR, and NPP at each site (see Figure 2, below). Plot size is 25 x 25 m, chosen to roughly correspond to the pixel size of ETM+ data and neatly nesting at various increments up to 1 km. Between 60 and 80 plots will be concentrated within a 1 km cell centered on the site's eddy flux tower, with the balance of the 100 plots located outside the tower cell, but within the 5 x 5 km BigFoot footprint. This density of plots within the tower footprint ensures adequate characterization of the vegetation properties within that footprint, a critical accomplishment if flux data are to be properly interpreted and used to assess scaled carbon and water flux estimates from biogeochemical models. The 20-40 plots outside of the tower footprint (i.e., within the 24 external cells) are apportioned within basic land cover strata to enable independent validation of BigFoot surface products over the full BigFoot footprint.
To facilitate a geospatial understanding of the ecology of the tower footprint, the plot design is a nested spatial series (see sampling design diagram above). This permits explicit examination of spatial covariation among field-measured ecosystem properties using variograms and cross-variograms (Cressie 1993). Further, the nested cyclical design provides a distribution of plots that is efficient at maximizing the number of plots at each lag (i.e., separation distance) in increments of 25 m up to nearly 1 km, an important consideration if the data are to be used for geostatistical analyses (Figure 4a-b). This also facilitates an examination of the effects of observation grain size on MODIS NPP estimates, an essential element of the BigFoot validation protocol. In high spatial frequency (e.g., heterogeneous) landscapes, functionally-important, but small vegetation patches cannot be detected above a certain grain size of observation. The use of geostatistical techniques will play an important role in increasing our understanding of the effect of image pixel resolution on coupled remote sensing–modeling characterizations within a biome (Milne and Cohen 1999). Moreover, we can assess if there is a fundamental grain size above which functionally-important vegetation patches can not be resolved and modeling errors accelerate, and, how the fundamental grain size varies among biomes.
