Tracking landscape scale vegetation change in the arid zone by integrating ground, drone and satellite data

Abstract A combined multiscale approach using ground, drone and satellite surveys can provide accurate landscape scale spatial mapping and monitoring. We used field observations with drone collected imagery covering 70 ha annually for a 5‐year period to estimate changes in living and dead vegetation of four widespread and abundant arid zone woody shrub species. Random forest classifiers delivered high accuracy (> 95%) using object‐based detection methods, with fast repeatable and transferrable processing using Google Earth Engine. Our classifiers performed well in both dominant arid zone landscape types: dune and swale, and at extremes of dry and wet years with minimal alterations. This highlighted the flexibility of the approach, potentially delivering insights into changes in highly variable environments. We also linked this classified drone vegetation to available temporally and spatially explicit Landsat satellite imagery, training a new, more accurate fractional vegetation cover model, allowing for accurate tracking of vegetation responses at large scales in the arid zone. Our method promises considerable opportunity to track vegetation dynamics including responses to management interventions, at large geographic scales, extending inference well beyond ground surveys.