Mapping & Monitoring
Land Cover Mapping in the Chesapeake Bay Watershed
Maps of land cover types were derived from multi-temporal Landsat imagery, including a total of 60 scenes capturing Spring, Summer, and Fall conditions for circa 2000. All scenes were radiometrically calibrated, converted to top-of-atmosphere reflectance, orthographically rectified using USGS 30m digital elevation data sets, corrected for topographic illumination effects, temporally normalized between scenes and cloud and shadow masked.
The land cover type mapping was done using a decision tree approach. The algorithm searches for a dependent variable that, if used to split a population of pixels into two groups, explains the largest portion of the deviation of the independent variable. At each new split in the tree, the same exercise is conducted and the tree is grown until it reaches the terminal nodes, each representing a unique set of image areas that are then assigned a specific land cover class based on the training information. Over 3,800 field sites were sampled for training data, allowing land cover to be mapped into 21 classes.
The map produced from this classification is an improvement over previous land cover maps of the region, particularly the discrimination between agricultural crop types and pasture. Overall map accuracy is 86%, although some of the specific classes that attempt to discriminate land use may be less accurate.
Access these data here:
ftp://ftp.whrc.org/GIS-RS/cbw/landcover
The land cover / use maps are being used, together with the impervious surface and tree cover maps, to inform Chesapeake Bay watershed models and restoration efforts.
Landcover in the Chesapeake Bay watershed in 2000. Click the image for a larger map.
