Iterative modeling for invasive species

Product Type: Abstract

Year: 2009

Author(s): Jarnevich, C.S. and J.T. Morisette

Pages: 1

Suggested Citation: Jarnevich, C.S. and J.T. Morisette. 2009. Iterative modeling for invasive species abs.. In: 94th ESA Annual Meeting: Sunday, August 2 - Friday, August 7, 2009. Albuquerque, New Mexico. : Ecological Society of America. p. 1.

Abstract

Background/Question/Methods Species distribution models are an important component of invasive species risk analysis and management. Many scientists have suggested an iterative process for developing distribution models of invasive species, where: (1) existing data is used to develop models; (2) those models guide future sampling; (3) models are re-run, and (4) the process is repeated. We wanted to test the importance of iterative modeling as species spread and more data become available. Using Tamarix location data in the western United States collected at three different time periods, we created three corresponding distribution models using Maximum Entropy Modeling (Maxent). We ran each model 25 times, withholding a different 30% of data for testing each time. We divided our second phase of data collection into two subsets with one subset based on results from the first model’s standard deviation and clamping (‘smart’ subset) and the second subset randomly selected (random subset). We compared all the three different models using the time series of data collection and the two subsets using model performance metrics including AUC, kappa, sensitivity, specificity, correct classification rate, standard deviation, and area where clamping occurred. Results/Conclusions Subsequent phases consistently performed better than previous ones for all metrics, and the ‘smart’ subset outperformed the random subset in everything except similar AUC values (94.8 and 95, respectively). For example, the maximum standard deviation in the model was reduced from 0.37 to 0.23 (a 38% improvement) by using smart sampling based on previous model runs. Likewise, maximum clamping, a measure of the inequity of sample occurrence in space, was reduced from 0.88 to 0.57 (a 35% improvement) with smart sampling data. These data support using previously developed models to guide future sampling and the re-running of models in an iterative fashion. This approach may help guide early detection, rapid assessment, rapid response, and containment of harmful invasive plants, animals, and diseases at local to global scales.