Transferability of fine-scale habitat suitability models for temperate marine demersal fishes

Numerous models for predicting species distribution have been developed. Such approaches are useful for informing marine conservation policy (e.g. reserve network selection). The efficiency of these models is usually tested on the area for which they were developed. However, practitioners may have distribution models based on data from one area that they want to apply in other areas. Hence, an important question is: how accurate are models of species distribution when applied beyond the areas where they were developed? We tested the transferability of maximum entropy (MAXENT) for five temperate marine demersal fishes, between two study localities in south-eastern Australia. The MAXENT models were based on spatially-explicit multibeam-sonar derived seafloor variables and geo-located towed-video occurrence datasets. Comparisons of evaluations (via area under the curve of the receiver operating characteristic plot; AUC, and transferability index) and spatial predictions within and between regions were conducted in order to test if species meet the criteria of full transferability. By full transferability, we mean that: (1) the internal evaluation (i.e. internal AUC) of models fitted in region A and B must be similar; (2) a model fitted in region A must at least retain a comparable external evaluation (external AUC) when projected into region B, and vice-versa; and (3) internal and external spatial predictions have to match within both regions. Only three of the 10 models achieved the transferability requirements (i.e. when internal evaluation AUC > 0.7 and external evaluation AUC < 0.7). In addition, the transferability index and similarity between spatial predications suggested considerable differences for most models. The pronounced asymmetry in transferability between the two study regions may be due to differences in the ranges of environmental predictors, impact of fishing, or varied dependence on biotic interactions that are not properly incorporated into these models. Overall, the limited geographical transferability calls for caution when projecting these models for predicting the distribution of species beyond the areas where they were developed, and further highlights the importance of ground-truthing.