Science Inventory

Assessing the Accuracy of MODIS-NDVI Derived Land-Cover Across the Great Lakes Basin


IIAMES, J. S. AND R. S. LUNETTA. Assessing the Accuracy of MODIS-NDVI Derived Land-Cover Across the Great Lakes Basin. Presented at Fifth International Workshop on the Analysis of Multi-temporal Data (MultiTemp 2009), Groton, CT, July 29 - 30, 2009.




This research describes the accuracy assessment process for a land-cover dataset developed for the Great Lakes Basin (GLB). This land-cover dataset was developed from the 2007 MODIS Normalized Difference Vegetation Index (NDVI) 16-day composite (MOD13Q) 250 m time-series data. Traditional hyperspectral image classification techniques were applied to perform the GLB time-series based classification. To optimize results the GLB was stratified across 12 ecoregions, ten in the United States and two in Canada (Omernik, 1987) (Figure 1). The multi-temporal NDVI images were used as inputs to hyperspectral classification algorithms taking advantage of the phenological response of various cover types to segment the GLB into six broad land-cover classes (i.e. impervious, water, agriculture, deciduous forest, coniferous forest, and annual grasses) (Knight and Lunetta, 2009). The water and impervious classes, however, had proven problematic to classify in a previous study (Knight et al., 2006). Therefore, both were classified using alternative methods and then masked to the MODIS-NDVI composite. The water mask was created using multiple Landsat ETM+ scenes mosaiced over the entire GLB region. A simple binary water versus non-water mask was derived from individual band reflectance thresholds. The impervious and agricultural (row crops only) classes were generated from the 2001 National Land Cover Dataset (NLCD) by scaling the 30 m resolution to the 250 m resolution native to the MODIS NDVI time-series data. All 250 m pixels that were greater than 75% impervious or agriculture were labeled as such. Therefore, the land-cover classes remaining for classification were the annual grasses, deciduous and coniferous forests, and agriculture (hay).

Record Details:

Product Published Date: 07/30/2009
Record Last Revised: 12/08/2009
OMB Category: Other
Record ID: 203783