Approaches to Bayesian occupancy modelling for habitat quality assessment

Bayesian ‘occupancy’ models (BOM) are a powerful tool that have recently been adapted to deal with ‘opportunistic’ species data (i.e. biological records). Individual species trends from any models can be aggregated to produce habitat indicators; here this is demonstrated for BOMs and Frescalo (Hill, 2012) using two examples. Example one demonstrates the production of habitat-specific trends using NPMS indicator species and subsets of 1 x 1 km grid cells predicted to contain the habitat of interest from Land Cover Mapping. Decisions around whether to include subsets of habitat-containing cells, or all cells within a political boundary, will be important for trend interpretation: habitat subsets of cells may lead to biases depending on true habitat change over time. Example two compares BOMs to the Frescalo method, as well as investigating the impacts of decisions for indicator production (e.g. weighting or not weighting by a species national frequency) on the trends produced. In this example weighted trends for 18 Sphagnum species typical of blanket bog were much more similar than unweighted trends. In the case of contradictory habitat (or species) trends it will not normally be possible to know which model is correct (at least in the absence of an unbiased dataset to which to refer). Given that BOMs, as currently used, may contain significant bias, a prudent approach would be to compare the outputs of several methods before making decisions.