气候变化领域集成服务门户

This study investigated elements important to regional landscape assessments: (a) appropriate mapping spatial resolutions (regional versus subregional), and (b) accuracy assessment procedures (point-based versus area-based). The study used MODIS NDVI time-series data to derive landcover products (2007) in a study area within the Laurentian Great Lakes Basin (as). Area-based reference data (i.e., "maplets") were varied in size and number to assess landcover proportionality agreement and to provide accuracy assessment metrics generated by point-based methodology. High spatial resolution Landsat ETM data was used to assess pixel purity (PP) for the MODIS 250 m pixels imbedded within the maplets (i.e., percent homogeneous) for the dominant cover type. Comparisons between the maplet reference data found a 21.7 percent variation in accuracy values between PP50 percent (67.9 percent accuracy) and PP100 percent (89.6 percent accuracy). Point-based accuracy assessments typically use 100 percent homogeneous reference pixels to assess landcover products, positively biasing the accuracy values. Our area-based methodology allows for the assessment at varying reference pixel homogeneity.