BUSH ENCROACHMENT MONITORING USING MULTI-TEMPORAL LANDSAT DATA AND RANDOM FORESTS

摘要: It is widely accepted that land degradation and desertification (LDD) are serious global threats to humans the environment. Around a third of savannahs in Africa affected by LDD processes may lead substantial declines ecosystem functioning services. Indirectly, can be monitored using relevant indicators. The encroachment woody plants into grasslands, subsequent conversion open woodlands shrublands, has attracted lot attention over last decades been identified as potential indicator LDD. Mapping bush large areas only effectively done Earth Observation (EO) data techniques. However, accurate assessment large-scale savannah through with satellite imagery remains formidable task due fact on vegetation variability response highly variable rainfall patterns might obscure underlying processes. Here, we present methodological framework for monitoring encroachment-related environment Northwest Province South Africa. We utilise multi-temporal Landsat TM ETM+ (SLC-on) from 1989 until 2009, mostly dry-season, ancillary GIS then use machine learning classification approach random forests identify extent 20-year period. results show area study, alarming permanent loss. year 2009 validated yielding low commission omission errors high k-statistic values grasses classes. Our step towards rigorous effective assessment.