A method for predicting native vegetation condition at regional scales

Abstract

Regional-scale mapping of vegetation characteristics such as extent, configuration, composition and condition are critical for managing native vegetation. The extent and configuration of native vegetation is typically mapped using remote sensing, and plant species and communities are typically mapped using statistical models built with explanatory variables derived from GIS layers. Such research has paid limited attention to the 'condition' of native vegetation and rarely are explanatory variables derived from satellite remote sensing and GIS layers used together to spatially predict vegetation characteristics. We calculated two independent metrics of vegetation condition using field data measured at each of 239 0.1 ha plots. These metrics of vegetation condition were used to develop two continuous maps of vegetation condition across an area of 260,000 ha using statistical models (generalised additive models, GAMs) built with explanatory variables derived from a range of sources including digital elevation models (DEMs), metrics of landscape connective, land use mapping and satellite remote sensing. Both models included significant explanatory variables that were derived from satellite remote sensing and GIS layers. Using a cross-validation technique based on bootstrapping, correlations between observed plot data and predicted data for the two measures of vegetation condition were only reasonable (0.47-0.56). Improved stratified sampling which captures disturbance gradients is a priority for improving models of this type. Crown