Ceratozygum horridum (Germar, 1839)

Perger, Robert, do Amaral, Karina Bohrer & Bianchi, Filipe Michels, 2021, Distribution modelling of the rare stink bug Ceratozygum horridum (Germar, 1839): isolated in small spots across the Neotropics or a continuous population?, Journal of Natural History 55 (11 - 12), pp. 649-663 : 658-659

Ceratozygum horridum distribution modelling: strengths and caveats

Building SDMs for species with poorly known distribution – either rare or endangered species – is challenging. In this study, the model was built following the best practices described by the literature (e.g. Araújo et al. 2019; Feng et al. 2019; Zurell and Engler 2019) aiming to improve the reliability of the outcomes. Our analyses reached the maximum of sensitivity, which is desirable to identify potential areas for rare species. However, data limitations may hamper some recommended practices, such as the definition of speciesspecific model settings ( Muscarella et al. 2014).

The background selection is pivotal to accurate modelling (see Elith et al. 2010; Merow et al. 2013). The AUC varies with the spatial extent used to select background points, reaching artificially higher AUC values in larger study areas ( Lobo et al. 2008; Jiménez- Valverde 2011; Hijmans and Elith 2017). The models could be overestimated, reflecting the loss of their accuracy extrapolating to non-analogous areas ( Jiménez-Valverde and Lobo 2007). We were aware of training models in a smaller area (than the whole distribution), where the environmental conditions tend to be more similar (but see Olson et al. 2001). Then, we projected them to other biomes. It is methodologically preferred rather than modelling for the whole distribution including all biomes at once. The biomes are widely recognised as a mosaic of unique features ( Olson et al. 2001; Dinerstein et al. 2017).

The performance of our models reached acceptable AUC values (higher than 0.7). Taking into account that the best possible practices were implemented, one of the most likely explanations for relatively low AUC values is due to the low sample size. The literature suggests that robust models may be obtained using a sample size>50 ( Stockwell and Peterson 2002),>30 ( Wisz et al. 2008), but useful models have been produced using 5–10 records ( Hernandez et al. 2006; Pearson et al. 2007). Besides methodological parameters and inputs, we also need to consider biological aspects of the species per se. The low AUC may be stressed by ecological plasticity, e.g. broad tolerance to abiotic conditions, which is already indicated by the wide distributional range. An alternative explanation for the low AUC could be the inclusion of data from two or more cryptic populations that may have been adapted to different ecological conditions.

MESS analysis allowed identifying areas with dissimilar conditions concerning those used for model training (see Elith et al. 2010), thus allowing a careful evaluation of the prediction values for areas in the dry diagonal and Atlantic Forest, which exhibited high occurrence probability. For instance, MESS indicated that the real presences in the ecotone between the Southwest Amazon forest, Chiquitano Dry Forest, Dry Chaco forest, and Atlantic Forest are located in a non-analogous environment to those used for model training. Such findings may explain the low values of occurrence probability obtained for these real presences. On the other hand, even with similar conditions within the Amazon areas, only some spots with reliable high occurrence probability were recovered in ‘new’ habitats, such as Tocantins /Pindare moist forest and Cordillera La Costa montane forest.

The studies focusing on little-known species (such as those known only from original descriptions) may raise more questions than light answers. Thus, even the small pieces of evidence are pertinent to the knowledge of these species. We must highlight our results as the first assessment of the potential distribution of the uncommon species C. horridum . Therefore, any extrapolation should be cautiously taken. Just the addition of more data will improve model performance and a better understanding of the ecological and biogeographical aspects of this species.