SRTM-DEM AND LANDSAT ETM+ DATA FOR MAPPING TROPICAL DRY FOREST COVER AND BIODIVERSITY ASSESSMENT IN NICARAGUA

Abstract

Tropical dry and deciduous forest comprises as much as 42% of the world’s tropical forests, but hasreceived far less attention than forest in wet tropical areas. Land use change threatens to greatly reducethe extent of dry forest that is known to contain high levels of plant and animal diversity. Forest fragmentationmay further endanger arboreal mammals that play principal role in the dispersal of large seeded fruits, plantcommunity assembly and diversity in these systems. Data on the spatial arrangement and extent of dryforest and other land cover types is greatly needed to enhance studies of forest fragmentation effects onanimal populations. To address this issue, we compared two Random Forest decision tree models forland cover classification in a Nicaraguan tropical dry forest landscape with and without the use of terrainvariables derived from Space Shuttle Radar and Topography Mission digital elevation data (SRTM-DEM).Landsat Enhanced Thematic Mapper (ETM+) bands and vegetation indices were the principle source ofspectral variables used. Overall classification accuracy for nine land cover types improved from 82.4% to87.4% once terrain and spectral predictor variables were combined. Error matrix comparisons showedthat class accuracy was significantly greater (z = 2.57, p-value < 0.05) with the inclusion of terrain variables(e.g., slope, elevation and topographic wetness index) in decision tree models. Variable importance metricsindicated that a corrected Normalized Difference Vegetation Index (NDVIc) and terrain variables improveddiscrimination of forest successional types and wetlands in the study area. Results from this study demonstratethe capability of terrain variables to enhance land cover classification and habitat mapping useful tobiodiversity assessment in tropical dry forest.

Keywords

• STRM-DEM
• Landsat ETM+
• Random Forest classifier
• tropical dry forest
• land cover