As a society, we need to know whether our conservation management is effective. Assessing the effectiveness of vegetation restoration is often hampered by insensitive baseline data. The data used to justify investing in restoration are rarely suitable for conducting sensitive assessments of change through time, even if sensitive, quantitative data are available from subsequent surveys. But can we salvage something from such situations, given that they are common, and that the ideal—high quality data before and after management—is rare?
For a while, Dave Duncan and I have been working away this problem (as has our student, Chris Jones, in his PhD problem). And in a paper to be published in Ecological Applications, we describe an approach for a first-cut analysis of initial, semi-quantitative, rapid assessments, and later, quantitative surveys. The case study is an analysis of investment in vegetation management on private land. And in no way do we intend that this would replace sensitive BACI-type or case-control longitudinal monitoring designs. This is an attempt to say something from a post hoc analysis where ordinarily we’d be forced to admit we can’t say anything about change, let alone the effect of management on that change.
The analysis is Bayesian. We model the change between two sets of values, unknown historical values, and the later observations. Those unknowns are informed by various assumptions, made explicit in the form of Bayesian prior distributions. We try various forms of prior information: the ranges implied by the historical semi-quantitative rapid assessments, and predictions from regional models, and a combination of the two. We show that some variables show changes that are robust to our (explicit) beliefs about the past.
Something this work taught me about Bayesian priors is that the more refined ones’ model, the more sensitive it is to prior assumptions. Models become tuned to the selection of covariates. Two simple things about survey designs for rapid vegetation condition assessment are the need for area-based assessment, rather than roving whole-of site assessment, and how difficult some people find counting the number of large trees.