Woodland habitat metrics (edge shape index etc.) and bat activity

[JonMoore]

Hi guys,

I'm really hoping someone may be able to help me with my MSc research.  Research has already shown that bats like to forage along woodland edges in the UK, but I'm hoping to examine data I have available to me to see whether other woodland metrics may effect the level of bat activity, particularly edge to core ratios.

The bat activity data was collected using remote detectors at static point samples and I have created points in ArcMap with the associated data in the point's attribute table.  I have the woodland mapped as polygons and have calculated the habitat metrics using the V-Late extension e.g. woodland edge, edge to core ratios, shape index etc. the calculated metrics are now in the polygon attribute table. I have data on 12 sites with 5-8 locations at each site spread around the UK. 

I've trawled websites and journal articles but I've been really struggling to work out how to perform the analysis.  So am hoping any readers of this post may be able to give me a few ideas.  I was hoping to keep the data in polygons to keep a higher resolution, but should I scrap polygons, convert to raster and use fragstats?  Should I be doing a nearest neighbour analysis i.e. nearest woodland to the sample points? Should I take an entirely different approach?!!

I only have an ArcGIS desktop licence, so some of the possible methods I have come across have been ruled out. However, I do have access to a statistical package to export results to for analysis if necessary.  Any ideas or guidelines to perform this analysis would be thoroughly appreciated.

[JeffreyEvans]

I am not entirely clear on what your data looks like. Are the sample-locations outside of the region that you calculated the landscape metrics in? If the hypothesis that you are testing is that bat activity is influenced by the proximal structural characteristics represented by your metrics then you could just assign the metric value from the closest polygon to your sample-locations. However, I would also include a distance from edge covariate. You could produce synthetic variables that characterize the variance of the metrics as well.    

This certainly sounds like a question that you can tackle in a regression framework. One word of caution is that, from the explanation of your study design, it sounds like your data is pseudo-replicated. As such, the degrees of freedom will be based on the number of sample-sites and not the number of observations. Regardless of the approach that you eventually settle on, this adds a wrinkle that you will have to sort out. Given that you only have 12 observations "spread around the UK" it is unlikely that your data is autocorrelated and thus is not appropriate for a spatial statistical model. With only 12 observations you will have a lack of power issue as well. Aside from lack of power, you may be able to address the pseudo-replication problem in a mixed effects model where the variable sample-size at each sample location is treated as a random effect. If you use AIC to test competing hypothesis, I would recommend using the small sample size corrected version AICc.

[JonMoore]

Jeffrey, many thanks for the very helpful advice!

As to help clarify the data, it was collected at multiple sample locations at 12 individual sites for EIA purposes.  These sites are around 1-2km square and their locations and the sample points within each site have not been randomly generated. However, the sites are reasonably spread across the north and central areas of England. At each of the 12 sites, between 5 and 8 observations were made at sample points across the site, giving a total of 92 sample points across all sites e.g. site 1 contains 6 sample points. To give you an idea of the sampling effort involved, the data contains 10,000+ nocturnal hours of recording effort over 950+ nights, containing 50,000+ records of bat passes. What I think I would prefer to do is analyse the 92 sample points, and perhaps summarise by site. would be sufficient power for a regression framework?

I should note that all sample points were on the exterior of woodland, but the sample points and the woodland patches lie within the same landscape. Woodland metrics have been calculated using V-late for patches within a range of buffer zones up to 1 km from each sample point.  The woodland patches within this 1 km buffer zone, range from 25m sq to 19 hectares. 

Unfortunately, I had no hand in the design of the surveys, I have simply been given access to the results of surveys for EIAs, so you could define it as secondary data. Not all the surveys were carried out to the same methodology e.g. period of survey, time of year, repetition. Each static remote detector collected between 4 and 14 nights of data per survey session at the sample points and not all sample point survey sessions were repeated on multiple occasions over the bat activity season. The sample points were not randomly generated but the location chosen to collect information about the specific location.  Any thoughts on whether the unrepeated sample points should be excluded from the study?  This would reduce the number of sample points to 46.

Thanks again Jeffrey