Robust identification of potential habitats of a rare demersal species (blackspot seabream) in the Northeast Atlantic

Species distribution models (SDM) are commonly used to identify potential habitats. When fitting them to heterogeneous, opportunistically collated presence/absence data, imbalance in the number of presence and absence observations often occurs, which could influence results. To robustly identify potential habitats for blackspot seabream (Pagellus bogaraveo) throughout its distribution area in the Northeast Atlantic and the western Mediterranean Sea, we used an ensemble species distribution modelling (eSDM) approach, modelling gridded presence–absence data with environmental predictors for two types of occurrence data sets. The first data set displayed the observed unbalanced spatially heterogeneous presence/absence ratio and the second a balanced presence/absence ratio. The data covered the full distribution area, including the European Atlantic shelf, the Azorean region and the Western Mediterranean Sea. Across these regions, populations display variable status. The main environmental predictors for potential habitats were bathymetry and annual maximum SST. The fitted ensemble compromise (eSDM) was projected over the whole grid to create a habitat suitability map. This map exhibited higher probabilities of presence for the balanced-ratio data set. A binary presence–absence map was then generated using optimized presence probability thresholds for four validation indices. Using the true skill statistic to optimize the threshold, the surface areas of the binary presence–absence map was 53% smaller for the balanced data set than for the observed unbalanced data set. However, the choice of validation index had an even greater impact (up to 15 000%). This indicates that studies using opportunistic data for SDM fitting need to pay attention to the effects of presence/absence data imbalance and the choice of validation index to fully evaluate uncertainty. •Ensemble species distribution modelling to identify potential habitat of rare fish.•Comparison of two occurrence data sets: imbalanced and balanced presence–absence.•Main predictors: bathymetry and sea surface temperature.•Higher presence probabilities for the balanced data set.•Effect of validation index on habitat map larger than data imbalance effect.