Carreon et al. Modeling the spatial distribution of dung beetle species with different nesting strategies and body sizes under climate change scenarios
Description
Previous studies have used climate niche models (CNM) to predict shifts in the distribution of suitable habitats of dung beetles without considering whether the effects of climate change vary depending on their nesting strategy (telecoprids, paracoprids and endocoprids) and body size. In this study, we used CNM to estimate the future distribution of climatically suitable habitats for 15 of dung beetle species that differ in these features. We focused on the near-term climate change predictions (period 2041–2060) in the Americas, hypothesizing that climatically suitable will decrease for all species, but the magnitude of these negative effects will be greater for large-sized species than for small-sized ones. The resulting models are provided in a KMZ file "Climate Niche Models". For each species, we included a map whose pixels indicate a probability of occurrence under the current climate and four other maps indicating the probability of occurrence under four Shared Socio-economic Pathways (SSP). In all maps, Highly Suitable Habitats (HSH) are indicated in green, Moderate Suitable Habitats (MSH) in yellow and Unsuitable Habitats (UH) as empty area.
Files
Steps to reproduce
First, 15 species were selected that differ in body size and nesting strategy. The occurrence data for each species was obtained using the GBIF (Global Biodiversity Information Facility) database, with the data limited to the Americas. Separately and for each species, the points of occurrence were visualized and refined in Quantum GIS 3.22. To avoid model overfitting, a 5-km radius buffer was applied to all the occurrence points; when two or more points overlapped with each other, only one of them was retained. Second, CNM were calibrated using MaxEnt 3.4. To obtain the climate data associated with the occurrence data, we used the WorldClim database, which includes 19 bioclimatic variables available at different spatial resolutions. Considering that spatial resolution may affect the model results, we chose to use a resolution of 2.5 arcmin (~21 km2/pixel). To avoid model overfitting, it is necessary to sort out the bioclimatic variables that are correlated with each other. Spearman’s correlation tests were used, and we retained the bioclimatic variables that were highly correlated (Spearman's coefficient > 0.7) with the largest number of variables. We calibrated the CNM using 75% of the occurrence points as training data and the rest for testing. This selection was conducted 100 times using bootstrap resampling. In each run, the ROC curve was computed with its corresponding area under the curve (hereafter, AUC) value. Third, to estimate the current distribution of each species separately, each CNM was projected on the land area of the American continents (excluding Canada) along with the corresponding occurrence data used to calibrate the model, while considering the current values for the climate variables of the region. This was used to produce a map for each species whose pixels indicate a probability of occurrence under the current climate. The probabilities associated with each point of occurrence were then extracted. This set of probabilities was categorized into deciles, which were grouped into ranges of low, moderate, and high probability of species presence according to the percentage of occurrence probabilities in each range. Finally, for statistical validation of the future distributions, we used the multivariate environmental similarity surface analysis (hereafter, MESS) developed by Elith et al. (2010). We performed the MESS analysis for each bootstrap run (100 in total) using the MaxEnt algorithm. These results were averaged to obtain a single MESS per SSP. To assess whether the predictions for the future distribution are valid, we overlapped the CNM projections for each SSP with its respective MESS. If multiple regions classified as highly suitable habitats overlap with areas where the MESS analysis indicates values less than zero (areas that are climatically dissimilar and where the species is assumed not to be present) then it is concluded that the CNM has a low efficiency to predict future changes.