The importance of biotic interactions in species distribution models: a test of the Eltonian noise hypothesis using parrots

Abstract

Aim To test the Eltonian noise hypothesis (ENH), that biotic interactions do not affect species distributions at large geographical scales.Location The Brazilian cerrado, a central South American savanna and biodiversity hotspot. MethodsWe modelled the distributions of 11 species of cerrado parrots using the software Maxent at four different spatial resolutions. We built models using abiotic variables, biotic variables (distribution of diet resources) and models combining abiotic and biotic variables. We compared model performance using the area under the curve of the receiver operating characteristic (AUC), retrieved from test data. We partitioned the variance between sets of predictors using a generalized linear model (GLM). Finally, we evaluated whether improvement in model performance (higher AUC values) in models with both abiotic and biotic variables, was related to the species' dietary niche breadth and/or spatial resolution of the models.Results We found that model performance was improved in most cases by the addition of biotic variables. Our variance-partitioning approach revealed that abiotic and biotic variables contribute independently to the final model. We found no relationship between model improvement and spatial resolution. We also found no relationship between dietary niche breadth and model improvement, indicating that dietary generalist and specialist species were not differently affected by the inclusion of biotic variables in the models. Main conclusionsOur results did not support the ENH. In this study, we explicitly incorporated a biotic variable (diet resource distribution) into species distribution models (SDMs), and we showed that these variables generally improve models and have independent contributions. These results agree with previous studies that incorporated biotic variables into SDMs. Ultimately, our results indicate that SDMs performed with abiotic variables only may depict only a partial representation of the geographical distribution of a species.

Division within the North American boreal forest: Ecological niche divergence between the Bicknell's Thrush (Catharus bicknelli) and Gray‐cheeked Thrush (C. minimus)

“…Four tree species (balsam fir, paper birch, black spruce, and white spruce) were tested as biotic correlatives to increase the predictability of my ecological niche models (Heikkinen et al, 2007;Wooten & Gibbs, 2012;de Araújo et al, 2014). Using occurrence data for tree species overlaid with the full set of 19 bioclimatic variables, I used MAxEnT 3.3.3k (Phillips, Anderson, & Schapire, 2006) to produce probability distributions for each tree species (Appendix S2).…”

“…Biotic factors are often important in defining a species' realized niche (de Araújo et al, 2014;Heikkinen et al, 2007;Pearson & Dawson, 2003), and here I included three tree species that frequently co-occur with Bicknell's Thrush or Gray-cheeked Thrush. The tree species layers were not significantly correlated with any climatic variable, indicating that these biotic variables are introducing new material into the models, a result also found by de Araújo et al (2014).…”

“…I compiled 320 occurrences for paper birch, 417 for balsam fir, 314 for black spruce, and 358 for white spruce (Appendices S1 and S3). Four tree species (balsam fir, paper birch, black spruce, and white spruce) were tested as biotic correlatives to increase the predictability of my ecological niche models (Heikkinen et al, 2007;Wooten & Gibbs, 2012;de Araújo et al, 2014). Using occurrence data for tree species overlaid with the full set of 19 bioclimatic variables, I used …”

“…Because ENMs may have environmental variables that are spatially autocorrelated (McCormack et al, 2009), I also employ a multivariate analysis, which compares the difference in means of niche axes (principal components) from actual and background occurrences. A secondary goal of this project is to include biotic factors in ecological niche models to more accurately depict and categorize a species' realized niche (de Araújo, Marcondes-Machado, & Costa, 2014;Heikkinen, Luoto, Virkkala, Pearson, & Körber, 2007;Pearson & Dawson, 2003). Models created from abiotic-only factors, abiotic and biotic factors, and biotic-only factors were compared to determine the most effective variable set to use in niche divergence tests.…”

Predicting the potential distribution area of Solidago ×niederederi (Asteraceae)

“…It is important that the algorithm based on maximum entropy is considered less sensitive to sample size than other methods used in SDM (Wisz et al, 2008). This method was successfully used in many ecological and biogeographical studies (de Araújo et al, 2014;Holzmann et al, 2015;Qin et al, 2015). For modeling the potential distribution of Solidago ×niederederi in Central and East Europe we used MAXENT software version 3.3.3 k. It was run with a maximum of 500 iterations, convergence threshold 0.00001, and all five auto feature classes (which are default values).…”