Habitat Condition Assessment System: a new way to assess the condition of natural habitats for terrestrial biodiversity across whole regions using remote sensing data

Consistent and repeatable estimation of habitat condition for biodiversity assessment across large areas (i.e. regional to global) with limited field observations presents a major challenge for remote sensing (RS). RS can describe what a site looks like and how it behaves (using time series), but is unable to distinguish anthropogenic impacts from natural dynamics. Consequently, it is possible to mistake a heavily degraded habitat for a natural habitat, for example a logged forest may appear identical to an intact open woodland. This problem is compounded by the existence of multiple natural states in any given environment, and spatial variation in the natural composition and structure of vegetation as a function of variation in environment. Uncertainty in assessing habitat condition from RS is often further exacerbated by sparseness in the spatial coverage of training data. We describe a novel generic, RS-based algorithm called Habitat Condition Assessment System, designed to address the above sources of uncertainty and to be highly flexible in its application. It allows for variability in the definition of condition and in the type and quantity of input data employed. Here, we demonstrate the mechanics of the new algorithm in a simple worked example and its practical application in a case study using inferred natural-only' reference data, reflective remotely sensed data, and associated environmental data, to map condition for Australia at a 001 degrees resolution. We assess the behaviour and shortcomings of the method, and compare the national case study with estimates from two existing national data sets, and field measured condition data observed at 16967 sites across the State of Victoria. The modelled predictions outperform both of the existing national data sets, explaining 52% of the variability in field observations for well-sampled cells (relative to 8% and 12% for the existing products). The methodology can potentially address some of the key pitfalls of condition modelling and could be applied in other regions with sufficient coverage of reference data. The approach also has good potential to be extended to work with reference data for which condition is measured on a continuous scale, for example from field-based condition assessment initiatives.