SYMP 13-2
Assessing the uncertainty in ecological niche modeling: Locality, climate, and topography

Wednesday, August 12, 2015: 2:00 PM
307, Baltimore Convention Center
Otto Alvarez, University of California, Merced
Qinghua Guo, School of Engineering and Sierra Nevada Research Institute, University of California Merced, Merced, CA
Robert C. Klinger, Western Ecological Research Center, U.S. Geological Survey, Oakhurst, CA
Yanjun Su, Engineering, University of California Merced, MERCED, CA

The use of species/ecological niche models (ENM) has been key in understanding the impact of climate change. ENM builds a statistical relationship between the occurrences of the species to environment predictors, allowing it to predict future distribution under climate change and topography layers. Unfortunately, there are many issues related with uncertainties in the data inputted to generate the predictions. First, one needs to give the location of either the species, present only or both present and absence. Second, environmental layers, these layers are typically climate surfaces and topography. They are very limited studies that include all of these (point, topography, and Climate) uncertainty in trying to predict species niches, by understanding all the uncertainties that propagate from the algorithms and the surfaces, there is a higher probability that we can build models that better represent the niche or distribution of species. The point locality was obtain from the Global Biodiversity Information Facility, we selected 40 mammalians around North America. For topography (elevation) the data was obtain from STRM and the climate surfaces from ClimSurf. All three sources contain the adequate uncertainty to use in each model run. A Monte Carlo simulation was run in seven different variations of each set of models to determine the impact of uncertainty.  Multi statistics were ran to determine the actual differences between each model run.


The results show that the uncertainties from different sources (point locality, topography and climate surfaces) do influence the predicted distributions of the majority species tested in this study. We used the prediction result without introducingany kind of uncertainties as the base. Through the McNemar’s test, it showed that all other predictions results with introducing different combinations of uncertaintiesare significantly different from the base at the significance level of 0.05. We further run the factor importance analysis to test the contribution of different uncertainty sources to the niche modelling results. For the majority of the tested 40 mammalians, the uncertainty of climate surfaces has the most significant influence on the predicted distributions. Moreover, the influence of the uncertainty in elevation on the niche modelling results is slightly larger than the uncertainty in point locality. The results indicate that the uncertainties of different input variablesshould be carefully considered in the niche modelling process. Especially, accurate climate surface inputs should be the first priority to be consider, which may significantly improve the accuracy of niche modeling results.