Wetland mapping using classification trees to combine TM imagery and climato-topographical index in Zoige plateau

摘要

The “wetness” is the fundamental characteristics of a wetland. Using precipitation records of 28 surface stations, ANUSPLIN was employed to interpolate rainfall data to surfaces. By replacing the upslope area by the volume of precipitation, a climato-topographical index was derived from 30m×30m DEM. Based on wetlands map obtained from the Yellow River Conservancy Commission, a set of training pixels were randomly selected in the proportion of the area of each class to establish the classification trees. Landsat TM imagery combined with climato-topographical index were employed as main discriminators. Classification trees were built to classify wetlands into five types: riverine (WR), lacustrine (WL), Alpine swamp meadow (WSM), scrub-shrub swamp (WSS), and seasonal flood plain (WFP). In addition, as a typical degenerated cover type, bare black beach (DBB) was also extracted. The wetlands map of 2007 was produced, and its accuracy was evaluated. The overall accuracy was 83.40%, whereas omission error rates for classification of WSS and WSM are respectively 15.52% and 12.60%. Generally, the developed method is relatively easy to implement, and should be applicable in other settings.