Mapping and change detection of wetland and riparian ecosystems in the Gallatin Valley, Montana using landsat imagery

Abstract

The location and distribution of wetlands and riparian zones influences the ecological functions present on a landscape. Accurate and easily reproducible landcover maps enable monitoring of land management decisions and ultimately a greater understanding of landscape ecology. Multi-season Landsat ETM+ imagery from 2001 combined with ancillary topographic and soils data was used to map wetland and riparian systems in the Gallatin Valley of Southwest Montana. Classification Tree Analysis (CTA) and Stochastic Gradient Boosting (SGB) decision-tree based classification algorithms were used to distinguish wetlands and riparian areas from the rest of the landscape. CTA creates a single classification tree using a one-step-look-ahead procedure to reduce variance. SGB utilized classification errors to refine tree development and incorporated the results of multiple trees into a single best classification. The SGB classification (86.0% overall accuracy) was more effective than CTA (61.7% overall accuracy) at detecting a variety of wetlands and riparian zones present on this landscape. A change detection analysis was performed for the years 1988 and 2001. The change detection used Landsat-based Tasseled Cap (TC) components and change vector analysis (CVA) to identify locations of wetland/riparian gain or loss in the 13-year period. CVA of TC brightness, greenness, and wetness components reduces the compound errors of multi-date classifications by using a threshold value to separate land cover change from spectral variability between 1988 and 2001 imagery. Only the highly changed pixels were classified using 1988 Landsat imagery and ancillary data. These change pixels were then merged with the 2001 classified image to develop a wetland/riparian map for 1988. The high overall accuracy of the 1988 classification (81%) developed with this procedure showed the benefits of this technique for mapping historical landcover features. Comparison of the 1988 and 2001 classifications identified locations where wetlands/riparian areas increased, decreased, or remained stable between these years. TC based CVA had an overall change detection accuracy of 75.8% and was able to identify areas of isolated and contiguous wetland/riparian change.