Spatial prediction of vegetation integrity indicators to support ecosystem protection and restoration planning

The integrity of native vegetation is widely used as an indicator of the status of ecosystems and biodiversity. Spatial information describing vegetation integrity is useful for land management decision-making and guiding progress toward conservation targets, such as the global ‘30 by 30’ area-based protection goal. Here we developed, evaluated, and applied a method for predicting continuous variation in vegetation integrity indicators in southeastern Australia that integrates vegetation site data, remote sensing, and environmental covariates. Standardised site-level measures ( N = 54,334) of five vegetation integrity indicators were derived from vegetation survey data. These were calculated from in situ vegetation structure, composition, and function estimates, and expressed relative to best-on-offer benchmark conditions. We trained statistical models of vegetation integrity and compared prediction uncertainty estimates across quantile regression and model instability approaches. Using five-fold cross-validation, machine learning models explained 37–58% of variance across the five vegetation integrity indicators, with land use and optical remote sensing indices the most important predictors. Model performance was highest for indicators describing vegetation composition and function, while vegetation structure was more challenging to predict. With reference to the global ‘30 by 30’ conservation target, we differentiated ecosystems in southeastern Australia that are already protected and high-integrity (e.g., heathlands), those with cost-effective opportunities for protection of high-integrity land (e.g., woodlands), and those requiring restoration investment (e.g., tussock grasslands). Analysis of indicator reliability revealed that measurement uncertainty and temporal variability in vegetation site data constrain the accuracy of integrity indicators and their spatial predictions. • Vegetation site data were used to train models of vegetation integrity indicators. • Diverse environmental and remote sensing predictors captured key integrity drivers. • Performance was reliable for composition, function, and overall integrity indicators. • Site measurement uncertainty, particularly for structure, constrained performance. • Broadscale spatial predictions help prioritise ecosystem protection and restoration.