My summer project is entitled, "Modelling mammalian movements in fragmented landscapes: exploring and integrating least-cost models and agent-based simulations".
The general idea was re-implement a study already published (Driezen et al, 2007. Evaluating least-cost model predictions with empirical dispersal data: A case-study using radio-tracking data of hedgehogs. Ecological Modelling 209: 314-322) and then extend the study by creating an agent-based simulation with which to add additional realistic behaviours to 'hedgehogs' to see if this extended our ability to understand and explore the way hedgehogs move around the landscape.
I got involved with this paper as one of my supervisors (Patrick Doncaster) collected the data for the original study, rather than because I have any particular interest in hedgehogs (although they are pretty cute!).
Least-cost models essentially give a value to each habitat type in a landscape. The model is species-specific and generated from expert knowledge of how the species in question views the landscape. Values represent the cost of being in that habitat, with low values indicating preferred and more suitable habitats and high values indicating non-suitable habitat or that which may pose the greatest threat to the species (for example roads, water, arable land or urban areas may be examples of 'costly' habitats depending on the species in question). These models assume that individual move to areas that are of low cost.
The data was collected from several areas in and around Oxford, UK. GIS maps of these areas were supplied from one of the authors (Frank Adriaensen at the University of Antwerp) but the rest of the analysis was done by attempting to re-implement the statistical analysis of a number of least-cost models as detailed in the Driezen paper.
12 models were generated, each providing a different set of values for the habitat types found within the study areas.
Hedgehog locations were recorded as data points and overlaid on the GIS maps.
By amending the attribute table of each GIS raster dataset, the cost of each habitat could be added to the GIS layer. ArcGIS has good functionality and so an in-built function to generate a cost layer was used to create the leats-cost model from the original habitat map.
Each hedghog was released from a fixed location and this was used as the reference point for analysis. As you can see below, concentric circles were generated using each release point as the centre and each observed location from the release point as the radius. The spatial analyst extension in ArcGIS was used to generate summary statistics for each circle, from which a comparison between cell costs could be made and a z-score calculated from which to perform some quite complicated statistical analyses (look to later posts for reference and discussion on stats etc).
Fig 1. Original habitat map showing one hedgehog path and concentric circles used to score each least-cost model, on the left. On the right, an example least-cost model with concentric circle. As the cost of the landscape increases the colour changes from brown (low cost) to blue(high cost). The path taken by the hedgehog is shown in red:
The cost of the cell in which the hedgehogs were located was then compared to the costs of all the other cells located on the edge of that same circle. If the observed location cell cost is lower then average then it means the hedgehog is located in a relatively lower cost habitat. The lower the cost, in comparison to the average cost per circle and per individual hedgehog, the better the least-cost model is in explaining how hedgehogs move around the landscape.
By doing this same analysis for all of the regions and all 22 hedgehogs for which we had data, one least-cost model produced the best fit to the data. (In this case it was resistance set 10).
My results were different to those found by the original Driezen study and least-cost models seem sensitive to selection of empirical datasets. However, the selection and generation of least-cost models to test seemed random and arbitrary and perhaps more thought and a a full sensitivity analysis may provide more confidence in the selection of the most suitable least-cost models.
next post: extension to this work: using least-cost modelling in a simplified agent-based simulation in Netlogo.
No comments:
Post a Comment