Delineation and Classification of Wetlands in the Northern Jarrah Forest, Western Australia Using Remote Sensing and Machine Learning

Abstract

Wetlands are under increasing pressure from threatening processes. Efforts to protect and monitor wetlands are hampered without datasets capturing the extent, type, and condition. The purpose of this study is to map the distribution of wetland type, vegetation type and vegetation condition for wetlands in the Northern Jarrah Forest region, Western Australia. A random forest algorithm implemented via Google Earth Engine (GEE) was used to classify wetlands and vegetation condition using satellite imagery, topographic indices, and soil mapping. Wetland type was classified using a hierarchical approach incorporating increasing level of detail. Wetland type was mapped as system type from the Interim Australian National Aquatic Ecosystem (ANAE) Classification framework and at hydroperiod level, with overall accuracy of 83% and 82% respectively. Vegetation type was mapped with an accuracy of 78.3%. Mapping of vegetation condition using the Vegetation Assets, States and Transitions (VAST) framework achieved an overall accuracy of 79.6%. Results show that wetlands occur in greater concentration as narrow seasonally waterlogged sites in the west, more sparsely and seasonally inundated sites in the northeast, and as broad seasonally waterlogged sites in the southeast of the study area. Wetland degradation determined through vegetation condition is concentrated in the east, and highest in seasonally waterlogged wetlands. Overall, the wetlands mapping framework implemented in this study can be used by land managers and other interested parties seeking to identify threatened and high conservation value wetlands in other areas.