Bat Habitat Suitability Model Basics

Model Philosophy

MaxEnt models need a set of presence samples and a grid model description of the habitat where those samples were recorded. Sounds simple doesn't it? The real issue you face before setting up a model of this type is how do you describe the habitat because this habitat description will be dependant on the "scale" (i.e. the grid cell size) of your model. In the case of my previous work, the grid cell size was approximately 10mx10m, and it was possible to resolve individual tree species and very detailed habitat changes, however a model resolved at this scale is going to be too big to run in a reasonable time and would actually be very difficult to define the grids to this level of accuracy. In this case I'm therefore forced into a larger scale and a grid cell size of approximately 50mx50m, which in turn means using a different set of habitat indicators.

After reviewing what other researchers had used in similar models (I recommend you look up the work of Chloe Bellamy and John Altringham) I decided on the following approach for the initial model:

1. Woodland cover would be represented by:
   1. Deciduous
   2. Coniferous/pine
   3. Mixed coniferous/deciduous
2. Land cover would be represented by:
   1. Built-up
   2. Open farmland (could be arable or pasture at this stage)
   3. Inland freshwater
3. Altitude would be included as a variable
4. The following "proximity" grids would also be used:
   1. Distance from woodland edges
   2. Distance from building lines
   3. Distance from inland water edges

For the land cover grids, the grid was built using the assumption the grid cell would assume a value based on whatever occupied 51% or greater of that cell area (except the mixed woodland definition). Proximity and altitude grids where built using continuous variable and a simple bi-linear interpolation was made to derive the final grid values. All the grids were derived from Open source OS data combined with additional detail where this was available (notably the plantation plans provided by the Forestry Commission estates, but also on site survey where open source aerial photography did not enable the habitat cover to be estimated). All the grids were initially built up in a distance preserving "Transverse Mercator" projection, however due to problems with converting the WGS-84 coordinates of my recordings to Mercator with sufficient accuracy, I converted all the final grid projections into WGS-84 projections to use with MaxEnt. All the grids used in this project have been built up and managed using QGIS 2.8.2 and GRASS 6.4.2 (both of these are open source and free for non-commercial use).

The 4th model evolution includes two additional grids, a terrain roughness index (TRI) grid and a compass aspect grid. Both of these grids have been derived from the elevation grid using the DEM (Digital Elevation Model) tools in QGIS. The terrain roughness variable is an indicator of how much height variation there is between adjacent cells, so a TRI of 0 represents flat ground, and the higher it is the bigger change in height from cell to cell. This is potentially a more useful measure of the landscape height changes than the just measuring the gradient and has been incorporated as continuous variable in the model. The compass aspect is the compass direction of the normal to the elevation grid cell and as with the Parkhurst Forest model, I've employed this as a continuous variable.

The presence samples can be obtained from a variety of means, in this case the project use a combination of acoustic survey (transect + fixed) roost survey and bat rescue data. I will say a bit more regarding how to manage bias issues and mixing the presence data types in a separate section on Survey protocols in due course.