Decision tree classification of dambo wetlands using remotely sensed multispectral and topographic data
Wetlands are the greatest single source of atmospheric methane (CH4), an important greenhouse gas. While the CH4 emissions from the expansive dambo wetlands of tropical Africa are likely substantial, they have not been reliably measured or estimated. Vegetation, topography, and soils vary among the following dambo land system zones: uplands, margins, floors, and bottoms. CH4 production also presumably varies among these zones, with the greatest emissions most likely originating from the bottoms. Multispectral and topographic remote sensing data were used to create a 20 m resolution classified map of dambo land system zones within a 2,214 sq km study area located in central Uganda. Training and accuracy assessment ground truth data were collected during a 4-week field campaign in the Ugandan study area. Multispectral inputs included reflectance values, vegetation indices, and spectral mixture modeling fractions from SPOT 4 satellite images acquired December, 2006, and February, 2007. Topographic inputs consisted of a digital elevation model (DEM), slope, and 20 relative elevation layers calculated using moving windows of various sizes. A binary decision tree (BDT) was used to create decision rules for the study area classification. Decision rules were based upon the following input variables: the Normalized Difference Vegetation Index (NDVI); the Normalized Difference Infrared Index (NDII); the shortwave infrared (SWIR) image band; northing value, in terms of Universal Transverse Mercator (UTM) projected coordinates; slope; and two relative elevation layers. The overall classification accuracy of 75.5% and Kappa (K) coefficient of 0.67.