I am currently doing species distribution models using presence-only data, and I would like to perform spatial filtering of presence points in order to lower the sampling bias and to make sure that the data for building/evaluating models (I'm doing the data partition-based evaluation) are independent.
I have R function which leaves in one occurence point and removes other points in specified nearest-neihborhood distance area. However, choosing the distance within which the data will be rarefied is often arbitrary.
My concern now is how to justify or to select the distance threshold. As of now, I am using 1km threshold (duplicate points within 1km buffer are removed). My study area is ~250x50km, it is quite spatially heterogeneous, and resolution of my most coarse environmental variable is 1km (downsampled to 30 meters).
I feel that 1km threshold is somewhat adequate, but I cannot reasonabily justify this choice. Does anyone has tips on this issue or some articles to direct me towards. One method that I've came across in the literature so far is using variogram range where the points become spatially independent, but I am still not clear on how to use my environmental variables to build the semivariogram, so any tips here would be very appreciated.
I would suggest you the variogram method you cited. For every environmental variable you have, use the presence points of your species as sampling points to evaluate in which distance the autocorrelation would disappear. The threshold would depend on the Range value for every variable. I would probably chose the highest Range value, but it depends on your results.
I suggest these papers,
Fourcade, Y.; Engler, J.O.; Rödder, D.; Secondi, J. 2014. Mapping Species Distributions with MAXENT Using a Geographically Biased Sample of Presence Data: A Performance Assessment of Methods for Correcting Sampling Bias. PLoS ONE 9(5): e97122. doi: 10.1371/journal.pone.0097122
Ranjitkar, S.; Kindt, R.; Sujakhu, N.M.; Hart, R.; Guo, W.; Yang, X.; Shrestha, K.K.; Xu, J.; Luedeling, E. 2014. Separation of the bioclimatic spaces of Himalayan tree rhododendron species predicted by ensemble suitability models. Global Ecology and Conservation 1: 2-12. doi: 10.1016/j.gecco.2014.07.001
Jiménez-Valverde, A. 2014. Threshold-dependence as a desirable attribute for discrimination assessment: implications for the evaluation of species distribution models. Biodiversity and Conservation 23(2): 369-385. doi: 10.1007/s10531-013-0606-1
Boria, R.A.; Olson, L.E.; Goodman, S.M.; Anderson, R.P. 2014. Spatial filtering to reduce sampling bias can improve the performance of ecological niche models. Ecological Modelling 275: 73-77. doi: 10.1016/j.ecolmodel.2013.12.012
You too should be consider the species that you want modelling. The requirements for invertebrates (eg. insects) can be very different to the requirements of plants or vertebrates (eg. birds, reptiles).The distance threshold to use in spatial filtering (in the same habitat) can vary between some of these groups according to their ecological characteristics.
Besides the data-driven (geographic) approach as referenced by Joaquin, ecological considerations may be considered in parallel, Some heuristic distances are used in the attached. For mobile animals (bear) the risk of double counting the same individual seems high in surrounding pixels/cells at your spatial resolutions. Less so for plants. Evidently, the spatial precision of the presence data may also play its role as has been tackled in the endemic plant SDM study.
Article Fine resolution distribution modelling of endemics in Majell...
Article Where the bears roam in Majella National Park, Italy
- Badges
- Science topic