|Abstract:||Freshwaters support over 40% of fish species diversity, as well as one-third of all vertebrate species, yet remain one of the most threatened habitats globally. Anthropogenic disturbances have caused many negative impacts throughout history, and continue to do so today. After the dust bowl we began to inch our way toward smarter management of our watersheds. This eventually spurred the development of best management practices (BMPs) to combat non-point source pollution. Voluntary private lands programs such as the Conservation Reserve Program (CRP) look to offer monetary incentives to landowners willing to implement conservation practices on their lands. Biological goals, such as increased native bird or fish populations, are sometimes included in programs like CRP but little has been done to evaluate whether those goals are being achieved or not. Sampling can often be expensive for these endeavors, so alternative measures for obtaining this information are valuable. Species distribution modeling (SDM) has provided us with a chance to gain more information about communities without additional sampling effort. I look to balance sampling efforts with species distribution modeling to investigate the effects of CRP of stream fish species richness.
In this study, I use data from two Illinois fisheries datasets in combination with GIS environmental data to predict the presence or absence of 64 fish species across the Kaskaskia River basin using random forest classification. Of the 64 modeled species, 52 SDMs met my model performance requirements (TSS>0.2). These 52 SDMs were then stacked to obtain an index of species richness across the basin, and then the species richness values were compared with observed richness of modeled species, via regression, for accuracy. The regression deviated from the ideal 1:1 line, but Theil’s Inequality Coefficient indicated a very strong matchup between observed and predicted richness (U=0.012). Based on this, I concluded that my SDMs were able to provide a reasonable representation of species richness when the predictions of individual species models were stacked.
I then developed a novel standardization method using a house-neighborhood framework. “Neighborhoods”, all stream reaches within a given waterway distance from a site, were built around a group of fish sampling sites in the Kaskaskia River basin, Illinois. The species richness of the neighborhood was then used to standardize species richness at fish sampling sites. It is expected that a site in a neighborhood with high species richness would have more species than a site in a neighborhood with low species richness. Standardization based on the neighborhood species richness removes this species pool effect. Logit regression was then used to assess the effect of local habitat variables including CRP on species richness. Proportion of CRP lands within the local watershed for sampled sites ranged from 0% to 45.13%. Using the dredge function within the MuMIn package in R, all possible models were explored. R2 values were low across all models, ranging from R2 = 0.0915 to R2 = 0.2367. The best models (ΔAIC<2) took various combinations of in-stream habitat characteristics with large substrate consistently being ranked as one of the most important variables for species richness. The proportion of CRP lands in the local watershed was not taken as a predictor for any of the top models, while local habitat variables were found to be the most common factors influencing species richness. In conclusion, my study was unable to detect any major influence from CRP on stream fish species richness, and shows that local habitat factors are drivers of species richness when removing species pool effects from models. More rigorous targeting in the CRP implementation plans may help to increase the effect that CRP lands can have on fish species richness.