Improving prediction of rare species’ distribution from community data

摘要: Species distribution models (SDMs) have been increasingly used to predict the geographic of a wide range organisms; however, relatively fewer research efforts concentrated on rare species despite their critical roles in biological conservation. The present study tested whether community data may improve modelling by sharing information among common and ones. We chose six SDMs that treat different ways, including two traditional single-species (random forest artificial neural network) four joint incorporate associations implicitly (multivariate random multi-response or explicitly (hierarchical communities generalized attribute model). In addition, we evaluated approaches arrangement, filtering conditional prediction, enhance selected models. model predictions were using cross validation based empirical collected from marine fisheries surveys, effects comparing for species. results demonstrated improved species' distributions certain extent but might also be unhelpful some cases. could appropriately predicted terms occurrence, whereas abundance tended underestimated most substantially benefited predictive performances multiple- models, respectively. conclude both algorithms need carefully order deliver improvement highlights opportunity challenges prediction making data.