THE KANSAS NEXT-GENERATION LAND USE/LAND COVER MAPPING INITIATIVE

The Kansas Next-Generation Land Cover Mapping Initiative is a two-phase 18-month mapping endeavor to be accomplished over a three-year period. The mapping methodology uses a hybrid, hierarchical classification of multi-temporal, multi-resolution imagery (Landsat TM and MODIS NDVI time-series) to develop modified Anderson Level I and Anderson Level II land cover maps of Kansas. During Phase I (July ’06 – Dec. ’07), a modified Level I land cover digital dataset was produced from multiseasonal Landsat TM imagery using an unsupervised classifier. A formal accuracy assessment reported the map to have an overall accuracy level of 90.7%. In Phase II (Jan. ’08 – June ’09), subclasses of cropland and grassland are being mapped using a decision tree classifier to produce a modified Level II digital dataset. The Kansas GAP database and the attributed USDA Common Land Unit (CLU) dataset were used for training and validation. MODIS NDVI imagery was used to map cropland subclasses and multi-seasonal Landsat TM imagery was used to map grassland subclasses. Cropland and grassland were separated using the 30-meter Level I map as a mask to isolate cropland pixels in the MODIS imagery and grassland pixels in the Landsat TM imagery. Using results from the decision tree classifier, cropland and grassland pixels in the Level I map will be reassigned to their respective subclasses to produce the 30-meter modified Anderson Level II map of Kansas. Preliminary classification results show the cropland map to have an average model accuracy of 93.7%. Irrigation status is also being mapped during Phase II using time-series MODIS NDVI imagery, USDA Census of Agriculture data, and unattributed CLU boundaries. The average maximum NDVI value per CLU was calculated and used to identify irrigated cropland. Irrigation status will be added to the cropland subclasses. A formal accuracy assessment of the Level II map will be performed.