Tools in the SDM tackle box: How to maximize model performance when predicting fish distributions

Date

2021-05

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

Fishes play an important role in global ecosystems, in establishing stable states and cascading trophic interactions. L(Muška et al. 2018)arge rates of commercially fished species that we heavily rely upon are in decline and concerns over habitat alterations and climate change. Understanding the distributions of fishes provides insights into the regions essential for survival and reproduction and the habitats and resources relied upon within these areas. Furthermore, invasive fish species contribute to biodiversity loss, ecosystem degradation, and impairment of ecosystem services worldwide. To conserve the many imperiled and manage the invasive species, managers first must understand where the fish occur. Species distribution models (SDMs) provide a method to predict ranges of coastal species by relating observations of occurrences to underlying environmental factors. My research involved investigating the knowledge and distribution of coastal and freshwater fishes in North America and in the state of Texas. First, I chose a suite of coastal fishes (Anchoa mitchilli, Ariopsis felis, Caranx hippos, Cynoscion nebulosus, Hippocampus zosterae, Lutjanus campechanus, Micropogonias undulatus, Mugil cephalus, Orthopristis chrysoptera, Paralichthys lethostigma, Pogonias cromis, and Trichiurus lepturus) with varying body sizes, habitat use, diets, vagility, and life-history traits that occur along the Texas Coast. I created and compared SDMs in Maxent for the twelve fishes with occurrence data from various sources, including the Coastal Fisheries Division of Texas Parks & Wildlife (TPWD; random and systematic) and Fishes of Texas Database (FoTx; opportunistic), to better understand the influence of data collection method on SDM performance. FoTx data resulted in stronger model predictive performances (higher AUC) than TPWD, regardless of species. However, TPWD data had a much higher sample size than FoTx. Subsampling TPWD resulted in AUC values that were closer, but not equal to, models based on FoTx data, indicating that differences in sample size explained only some of the variation between the two data sources. Next, I used the previously created SDMs to determine if species traits influenced model predictive strength as indicated by AUC. Out of the traits I considered, only commonness significantly impacted model accuracy. Rarer fish tended to produce SDMs with a better model fit. Likely, this is caused by species having fewer occurrence data, a result of overfitting the model. This also indicates that multiple factors, outside of species traits, are influential to species distribution predictions. For my freshwater fish analyses, I reviewed the literature for the tilapia species O. aureus, O. macrochir, O. mossambicus, O. niloticus, O. urolepis, S. melanotheron, T. mariae, T. rendalli, T. sparrmanii, and T. zillii that are already present or have been previously documented in the wild in North America to summarize current global knowledge on the diet, reproduction, and physiological tolerances of the ten species of tilapia documented in North America. In Texas, four Oreochromis spp. are allowed for stocking without a permit, so I predicted the potential distribution of invasive tilapia to promote informed conservation and management policies, especially regarding potential impacts introduced tilapia may have on native fish species of greatest conservation need (SGCN). I developed SDMs and compiled a list from the literature to identify the fish SGCN in Texas that are most at risk of negative impacts of tilapia introduction. I identified 21 SGCN at high risk from O. aureus, 15 SGCN at high risk from O. mossambicus, and one SGCN at high risk from O. niloticus. My findings support the conclusion that the success of invasive tilapia is due to wide environmental tolerances, robust reproductive strategies, and large dietary breath, and a well-documented history of negative impacts on local fauna. Improving our understanding of uncertainties in modelling techniques, including inputs into the model (e.g., data type, species traits, physiological tolerances), will improve reliability and application, particularly in the face of threats.

Description

Keywords

SDMs, ENMs, Species Distributions, Niche, Fishes, Fish, Estuaries, Texas, Tilapia, Coastal, North America, Invasive, Imperiled, Management, Conservation, SGCN, Physiological Tolerances, Oreochromis, Traits, Modeling

Citation