Ecological niche modeling (ENM) has been extensively applied as a reliable tool in conservation biology. Still, challenges abound in generating optimal models, especially when using limited occurrence data. The bluehead shiner, Pteronotropis hubbsi, a threatened species of concern, was modeled throughout its range within the U.S. South Central Plains Ecoregion. The portions of states that overlap this region include Texas, Oklahoma, Arkansas, and Louisiana. I used the Maxent software package (Phillips et al., 2006), as the ENM algorithm for this project. A maximum of 14 geospatial environmental layers (climatic, hydrologic, and geologic) were chosen to determine the species' association with its environment. Numerous sources suggest Maxent's default settings do not generate optimal model performance. Because of this, I compared models by first examining the effects of spatial filtering. Then, I tuned Maxent's features (linear and hinge) and regularization multipliers across seven extents. All unfiltered datasets exhibited heavy overfitting and did not produce a model with an acceptable omission rate. For the tuning experiments using filtered datasets, all default settings experienced model overfitting, which constrains the algorithm's predictive performance. Generally, models with regularization multipliers greater than three lose their discriminative ability where maps predict unrealistic habitat suitability within a majority of the study's extent. The majority of optimal models with limited sample sizes required the following applications: spatial filtering of occurrence data, use of linear features, and a Regularization multiplier greater than the default. A jackknife test of variable importance determined that each extent relied on a unique combination of variables to predict habitat suitability, but Geology, Strahler Stream Order, River Basin, and Soils were the most consistent top four predictors throughout the various extents. This project demonstrates that tuning the Maxent algorithm, spatial filtering, and data reduction are required to generate optimal models. This information can be used in the effort to evaluate the conservation status of a rare, aquatic species by efficiently planning surveys to discover unknown populations.

Date of publication

Fall 12-1-2015

Document Type




Persistent identifier


Included in

Biology Commons