An evaluation of methods for modeling distribution of Patagonian insects

TOGNELLI, M.F., S.A. ROIG-JUÑENT, A.E. MARVALDI, G.E. FLORES, J. M. LOBO

REVISTA CHILENA DE HISTORIA NATURAL

SOC BIOLGIA CHILE

Año: 2009 vol. 82 p. 347 - 360

0716-078X

Various studies have shown that model performance may vary depending on the species being modelled, the study  area,  or  the  number  of  sampled  localities,  and  suggest  that  it  is  necessary  to  assess  which  model  is better  for  a  particular  situation.  Thus,  in  this  study  we  evaluate  the  performance  of  different  techniques  for modelling the distribution of Patagonian insects. We applied eight of the most widely used modelling methods (artificial  neural  networks,  BIOCLIM,  classification  and  regression  trees,  DOMAIN,  generalized  additive models,  GARP,  generalized  linear  models,  and  Maxent)  to  the  distribution  of  ten  Patagonian  insect  species. We  compared  model  performance  with  five  accuracy  measures.  To  overcome  the  problem  of  not  having reliable absence data with which to evaluate model performance, we used randomly selected pseudo-absences located outside of the polygon area defined by taxonomic experts. Our analyses show significant differences among modelling methods depending on the chosen accuracy measure. Maxent performed the best according to four out of the five accuracy measures, although its accuracy did not differ significantly from that obtained with  artificial  neural  networks.  When  assessed  on  per  species  basis,  Maxent  was  also  one  of  the  strongest performing methods, particularly for species sampled from a relatively low number of localities. Overall, our study identified four groups of modelling techniques based on model performance. The top-performing group is composed of Maxent and artificial neural networks, followed closely by the DOMAIN technique. The third group  includes  GARP,  GAM,  GLM,  and  CART,  and  the  fourth  best  performer  is  the  BIOCLIM  technique. Although  these  results  may  allow  obtaining  better  distributional  predictions  for  reserve  selection,  it  is necessary to be cautious in their use due to the provisional nature of these simulations.