UPDATING THE NATIONAL WETLAND INVENTORY IN MINNESOTA BY INTEGRATING AIR PHOTO-INTERPRETATION, OBJECT-ORIENTED IMAGE ANALYSIS AND MULTI-SOURCE DATA FUSION

摘要

In 2006, several agencies from the state of Minnesota along with federal agencies collaboratively developed a comprehensive wetland assessment, monitoring, and mapping strategy for Minnesota. One of the key components to this strategy is to update the National Wetland Inventory Data for Minnesota using the best available data and techniques. Planning and pilot testing for the NWI update was initiated in 2008 and production level mapping efforts began in 2010. Methods employ a combination of object oriented image analysis, RandomForest? algorithm based classification, and manual photo-interpretation. Image segmentation of high resolution color IR aerial photography from both spring and summer was used to extract wetland and upland feature boundaries. The geometry and spectral characteristics from the extracted features were combined with additional data including PALSAR satellite RADAR, airborne LiDAR derived DEMs, and soils data to train a RandomForest algorithm to classify wetland types. The segmentation based geometry and RandomForest classification results are then integrated into a photo-interpretation workflow to create the final updated wetland inventory.