Using Landsat observations (1988–2017) and Google Earth Engine to detect vegetation cover changes in rangelands - A first step towards identifying degraded lands for conservation

Abstract

Abstract Globally, the area of agricultural land is shrinking in part due to environmental degradation. Acquisition and restoration of degraded lands no longer used for agriculture may present a major conservation opportunity with minimal social and political opposition. The ability to efficiently and accurately identify these lands from regional to global scales will aid conservation management, ultimately enhancing the global prospects of achieving the Sustainable Development Goals (SDGs). Remote Sensing provides a potential tool to identify areas where surface property changes can be mapped and linked with land degradation. In this study, we begin to tackle a small section of this challenge by presenting novel approach to mapping changes in vegetation cover amounts at the pixel level (30\u202fm), using Google Earth Engine (GEE). We illustrate our approach across large-scale rangelands in Queensland Australia, using three decades of Landsat satellite imagery (1988–2017) along with field observations of land condition scores for validation. The approach used an existing method for dynamic reference cover to remove the rainfall variability and focused on the human management effects on the vegetation cover changes. Results showed the identified vegetation cover changes could be categorized into five classes of decrease, increase or stable cover compared with a set reference level, which was obtained from locations of the most persistent ground cover across all dry years. In total, vegetation cover decrease was observed in 20% of our study area, with similar portion of lands recovering and the rest (~60%) staying stable. The lands with decrease in vegetation cover, covering a considerable area of ~2\u202f×\u202f105\u202fkm2, exhibited a markedly reduced resilience to droughts. The accuracy assessment yielded an overall classification accuracy of 82.6% (±3.32 standard error) with 75.0% (±5.16%) and 70.0% (±4.13%) producer s and user s accuracy for areas experiencing a significant decrease in vegetation cover, respectively. Identifying areas of degraded land will require multiple stages of spatial data analysis and this work provided the first stage for identifying vegetation cover changes in large-scale rangeland environment, and provides a platform for future research and development to identify degraded lands and their utility for achieving conservation endeavours.