Evaluation of modelled net primary production using MODIS and landsat satellite data fusion

To improve estimates of net primary production for terrestrial ecosystems of the continental United States, we evaluated a new image fusion technique to incorporate high resolution Landsat land cover data into a modified version of the CASA ecosystem model. The proportion of each Landsat land cover type within each 0.004 degree resolution CASA pixel was used to influence the ecosystem model result by a pure-pixel interpolation method.

Seventeen Ameriflux tower flux records spread across the country were combined to evaluate monthly NPP estimates from the modified CASA model. Monthly measured NPP data values plotted against the revised CASA model outputs resulted in an overall R² of 0.72, mainly due to cropland locations where irrigation and crop rotation were not accounted for by the CASA model. When managed and disturbed locations are removed from the validation, the R² increases to 0.82.

The revised CASA model with pure-pixel interpolated vegetation index performed well at tower sites where vegetation was not manipulated or managed and had not been recently disturbed. Tower locations that showed relatively low correlations with CASA-estimated NPP were regularly disturbed by either human or natural forces.