Creating a Land-use Classification for Iowa using MODIS 250-meter Imagery
暂无分享,去创建一个
Using NOAA AVHRR or MODIS imagery to create land-use classifications has been attempted for many years. Unfortunately, most of these classifications do not differentiate crop types. If one is to extract pixel information as input to a crop model it is critical that the pixel be correctly identified with its crop type. In this study a Landsat TM classification was aggregated to the 250-meter pixel size of MODIS. Only pixels that were at least 80% of a particular crop were used as ground truth for the MODIS classification. The MODIS classification was performed using both standard techniques and decision tree techniques. The MODIS classifications were compared at the county, agricultural statistics district and state levels to the 80% mask and to the original TM land-use classification. When compared with the 80% mask, 60-80% accuracies were obtainable at the state level. However, the MODIS classification was only approximately 40% accurate when compared with the Landsat TM classification. It does appear that decision tree methodology resulted in better accuracies in most cases.
[1] Philip H. Swain,et al. Remote Sensing: The Quantitative Approach , 1981, IEEE Transactions on Pattern Analysis and Machine Intelligence.
[2] Rick Mueller,et al. Creating a Cropland Data Layer for an Entire State , 2002 .
[3] R. DeFries,et al. Classification trees: an alternative to traditional land cover classifiers , 1996 .
[5] STATE-LEVEL CROP MAPPING IN THE U.S. CENTRAL GREAT PLAINS AGROECOSYSTEM USING MODIS 250-METER NDVI DATA , 2005 .