The BIOFRAG project developed out of a project that was looking for a new biodiversity impact metric designed to capture impact of habitat fragmentation in fragmented forested landscapes. That was quite some time ago and involved people like Prof David Coomes from Cambridge and his postdoc and Dr Rob Ewers from Imperial College London. Funded through the ERC (2012-2017), we have moved away considerably from the original thinking behind the BIOFRAG metric, started to set up the BIOFRAG project compiling data from fragmentation scientists around the world, and generating new, funky, smarter way to analyse how biodiversity responds to habitat fragmentation.
The original year 2010 BIOFRAG metric (values ranging from 0 to 1) was a new index that tried to describe the impact of forest fragmentation in a landscape on biodiversity at community level. Thereby ‘BIOFRAG = 1’ equalled no biological effect of fragmentation and ‘BIOFRAG = 0’ equalled a strong effect of fragmentation.
The metric exploited the exptectation that communities in sites affected by fragmentation may be more different to communities in sites unaffected by fragmentation (‘controls’) in their species composition. For example, an ensemble of bird species found in the interior of a large continuous forest will differ from an ensemble in a nearby forest patch that is separated from the main continuous forest reflecting a response of birds to fragmentation.
The computation of BIOFRAG for a given forested landscape required as input: a species-abundance matrix containing the abundance of species measured for each plot in a given landscape, the coordinates of each of these plots in that landscape and a forest-cover map (presence – absence of forest).
The main steps in the calculation of the metric were:
- Compute a Bray Curtis – Index (0: no species in common; 1: identical abundance of allspecies) based community dissimilarity matrix following square-root transformation of the species abundance data
- Use multivariate partitioning (multivariate regression tree modelling) to split plots measured in the forest patches into control plots and fragmented plots: sites are clustered according to their community similarity (response) with respect to a known environmental gradient (predictor), which in our case is ‘Distance to forest edge from the specific forest plot measured’
- Delineation of forst fragments from the forest maps based on a watershed-algorithm and computation of new fragmentation metrics (see Vero’s poster presented at the Science@SAFE workshop, Silwood Park Campus, September 2012). Link.
- Statistically link the community similarity of each fragmentation-affected plot to the fragmentation metric including smoothness, compactness and distance to forest edge. At the moment, the final selection of the minimum-adequate model (beta-logistic regression) describing this relationship is based on the AIC criterion.
- Ue the resultant models and their coefficients to compute community similarity (BC value) for each forest pixel (and simulate to get the confidence intervals)
- Compute BIOFRAG as the sum of BC over all forest cells divided by the sum of the maximum BC (defined similarity among control groups)
2010 Biodiversity Indicators Partnership. 2010. Forest Fragmentation: Identifying a biodiversity-relevant indicator. 2010 BIP, Cambridge, UK. [BIP Forest Fragmentation report]
Lafortezza R, Coomes DA, Kapos V, Ewers RM. 2010. Assessing the impacts of fragmentation on plant communities in New Zealand: scaling from survey plots to landscapes. Global Ecology and Biogeography, 19, 741-754.
Ewers RM, Marsh CJ, Wearn OR. 2010. Making statistics biologically relevant in fragmented landscapes. Trends in Ecology and Evolution, 25, 699-704.