Satellite remote sensing to monitor land condition and dynamics in arid Australia: letting the landscape speak for itself.

Abstract

The natural arid regions of Australia hold special value because their ecosystems are relatively intact. They play an important role in carbon cycling, provide ecosystem services, deliver benchmark information about ecosystem structure and function in unmodified landscapes, and are often the last stronghold of threatened species. Many of these regions are also homelands for Aboriginal traditional owners. These regions are under increasing external pressures from mining, tourism, localised grazing and invasive species. Careful management is needed to maintain their ecological values. Monitoring land condition is vital for management, but in extensive remote regions collecting field data to adequately represent land systems and processes is time consuming and costly. The high spatio-temporal variability of the arid landscape further confounds data interpretation. Long-term patterns of variability in vegetation response need to be understood to interpret management effects as distinct from natural variability. These long-term patterns cannot be understood from field data alone. In contrast, satellite-based monitoring offers potential monitoring solutions, with spatially comprehensive and consistent coverage over wide regions at relatively low cost. This overall aim of the research was to improve arid land condition monitoring through use of satellite remote sensing. Vegetation cover and soil exposure were used as indicators of land condition throughout the study. The study focused on the Alinytjara Wilurara (AW) Natural Resources Management (NRM) region in the far west of South Australia. This region is 261,180 km² in extent, much of it in near-pristine or wilderness condition, and is co-managed by the South Australian Government and the Aboriginal traditional owners. The landscape is extremely varied, incorporating calcarenite cliffs and dunes along the southern coast, the flat limestone Nullarbor Plain, red dune fields of the Great Victoria Desert, and the granitic Central Ranges and associated alluvial fans and plains in the north. The research comprised three components to address the overall aim. Specific objectives were to characterize and better understand the patterns of long term spatiotemporal variability in vegetation growth over the Australian arid zone; to interpret these patterns, their significance and implications for monitoring and management across the AW study region; and to evaluate the potential of high-temporal frequency low-spatial resolution fractional cover imagery for rapid land condition monitoring in the region. To address the difficulties of field validation, a further objective was to develop the use of high-spatial resolution satellite data as a tool for evaluation of low-spatial resolution fractional cover products. To detect long-term patterns of variability in vegetation growth, 25 years of twice-monthly AVHRR NDVI data were analysed with principal components analysis. The main components that underlie Australia-wide arid zone variability were revealed as total vegetation growth, seasonal response, and erratic east-west response driven by cyclonic activity. These factors were used as the basis to classify the Australian arid region into 24 classes. The new spatio-temporal classes, which represent long-term vegetation function, were compared to the existing Interim Biogeographic Regionalisation for Australia (IBRA), which describes vegetation in terms of structure and composition. Some classes showed close correspondence with IBRA regions, but in other areas the classes revealed variation in functional response within and between IBRA regions. Subsequently, focusing on the AW region, the four dominant classes in this region showed distinct characteristics in relation to average amount and temporal variability of vegetation growth, timing of growth cycles, and vegetation type. These distinctions can improve interpretation of on-ground data and have important implications for site selection and monitoring protocols such as timing and frequency of monitoring. Lastly, a MODIS fractional cover product, designated for Australia-wide usage, was tested for suitability over the AW region for ongoing monitoring of land condition. In the absence of field data, remotely-sensed surrogates were created, which classified high-spatial resolution (2.5m) ALOS PRISM data into fractions of bare soil and vegetation cover. These were up-scaled to MODIS resolution. Weak correlation was found between the surrogate and the MODIS fractional cover product, implying that the MODIS product is not suitable in its current form for use in the AW region. The finding of a slightly stronger correlation with increased vegetation cover suggests that the lack of relationship may be due to the generally low NDVI response of the arid perennial vegetation in this region. The novel method employed to create the soil exposure surrogate for this evaluation warrants further development and application for validating low spatial resolution image products in arid regions worldwide. This research shows how remote sensing can be used to define the high spatio-temporal variability of the Australian arid zone and provide new spatio-temporal information to improve regional environmental monitoring and management. We recommend that satellite remote sensing, because of its temporal capacity and comprehensive nature, be included as an essential component for monitoring remote arid environments.