Predictor complexity and feature selection affect Maxent model transferability: Evidence from global freshwater invasive species

Abstract

Aim Ecological niche models (ENMs) are widely used to address urgent real‐world problems such as climate change effects or invasive species; however, the generality of models when projected through space and/or time, that is transferability, remains a key challenge. Here, we explored the effects of complex predictors and feature selection on ENM transferability in a widely employed algorithm, Maxent, using five globally invasive freshwater species as case studies. Location Global. Methods We modelled the global distributions of five notorious freshwater invasive species (African sharptooth catfish Clarias gariepinus, Mozambique tilapia Oreochromis mossambicus, American bullfrog Lithobates catesbeianus, red swamp crayfish Procambarus clarkii, and Australian redclaw crayfish Cherax quadricarinatus), using three predictor datasets of varying complexities derived from two commonly used climatic data sources (WorldClim and IPCC) and three methods of model tuning that differentially incorporated feature selection. Spatially explicit transferability assessments were then conducted using a suite of evaluation metrics previously used to quantify Maxent model performance. Results We show that in the absence of detailed biological knowledge of focal species, simpler predictor datasets produce models that are more accurate than those calibrated using comprehensive “bioclimatic” datasets. Additionally, we find that tuning models for both optimal regularization parameters as well as feature‐class combinations led to the greatest increases in transferability and geographic niche conservatism. Results indicate a tenuous link between model transferability and Akaike's information criterion corrected for small sample sizes (AICc), suggesting that the indiscriminate use of AICc as an estimate of model parsimony may lead to erratic model performance. Main conclusions Our findings demonstrate that methodological considerations can drastically affect the reliability of spatial and possibly temporal projections, which has severe implications when ENMs are used to infer species’ niches, and quantify ecological or evolutionary change across impacted landscapes.

Influence of the Environment on the Distribution and Quality of Gentiana dahurica Fisch.

“…To map the environmental suitability of G. dahurica , the BRT modeling procedure also required background points as input data. Based on the literatures, areas where the annual precipitation was < 250 mm or > 400 mm, or where the elevation was < 800 m, were considered less suitable for planting G. dahurica and were used as the basis for screening the background points ( Guo et al, 2019 ; Li et al, 2019 ; Low et al, 2020 ). A total of 246 background points reflecting unsuitable environmental conditions for growing G. dahurica were randomly selected.…”

Predicting the Habitat Suitability of Melaleuca cajuputi Based on the MaxEnt Species Distribution Model

“…Nonetheless, like any modelling approach, the quality of input data (i.e., the reliability of environmental and species presence data) and the parameterization employed for modelling significantly impact the model's results using MaxEnt [56,57]. Issues of multicollinearity among the predictor variables that could affect the model performance were highlighted in the previous literatures [58] and reducing the number of variables could reduce the model complexity and benefits the operation time and model interpretability. In this study, we carefully selected the variables by comparing the list of variables favored by step-wise VIF and variable permutation importance.…”

“…When comparing the model complexity, the most used metrics assessments are information criterion such as Akaike information criterion (AIC) or AICc for smaller samples. However, the use of AIC as an estimate of model parsimony may lead to uncertain or confusing performance of the MaxEnt model [58]. Using AIC is acceptable for application with another model such as GLM or GAM but in the case of the MaxEnt model, the use of AIC or AICc is debatable because of a philosophical disconnection between machine learning and classical modeling [44].…”

Wolf–Hunting Dog Interactions in a Biodiversity Hot Spot Area in Northern Greece: Preliminary Assessment and Implications for Conservation in the Dadia-Lefkimi-Soufli Forest National Park and Adjacent Areas

“…Data partitioning during bootstrapping used 75% of all available cases for training and 25% for testing. Model transferability and overall performance of replicated models was evaluated with four criteria, which have been used and shown to be very consistent with and related to model transferability [ 121 , 126 ]: AUC test , AUC diff , OR 10 , and OR min . AUC diff is the difference between AUC train and AUC test .…”