Thursday, August 5, 2010

PS 81-126: CANCELLED- The efficiency of niche envelope models in predicting species distributions: A comparison of species distribution modeling approaches using historical presence only data

Israel Del Toro, University of Massachusetts at Amherst

Background/Question/Methods

Niche envelope models are commonly used to predict species distributions at multiple spatial scales. Here I compare the efficiency of two commonly used modeling approaches, MaxEnt and GARP. The main objectives of this work were to identify the minimum number of sample sizes required to develop a reliable model for prediction, identify any relationships between sample size and species range area and compare the performance of both modeling approaches by varying sample sizes of three North American ant species. I used presence-only data of ants of the genus Liometopum, recently obtained from multiple museum and university entomological collections. I used these species because they have been heavily sampled and I have sufficient data to understand their total distribution. Using the niche modeling approach, I can identify the ecological variables that explain the species distribution patterns observed in nature. The methods developed from this exercise will serve as a baseline for understanding the dynamics of terrestrial arthropod species distributions. These applications will have implications for the fields of biogeography, conservation and environmental monitoring.  

Results/Conclusions

Previous works have suggested that MaxEnt generates a robust model of species potential distributions and typically predicts species distributions better than GARP. My findings demonstrate the both models behave similarly with large sample sizes (>100) and at large spatial scales (as is the case with L. luctuosum and L. apiculatum). By systematically reducing sample sizes I determined that MaxEnt consistently generated a reliable species distribution model for large-range species (L. luctuosum and L. apiculatum) with as little as 15 samples and as few as 12 samples for small-range species (L. occidentale).  GARP generated reliable species distribution models with as little as 25 samples for large range species and 18 samples for small range species. I also conclude that MaxEnt requires a minimum of 1 sample per 472 km2 of the total species range while GARP requires a minimum of 1 sample per 752 km2 of the total species range to develop an accurate niche envelope model. My future modeling attempts of ant community distributions and assemblages will use primarily MaxEnt to develop species distribution models.