气候变化领域集成服务门户

Aim Biodiversity prediction becomes increasingly important in the face of global diversity loss, whereas substantial challenges still exist in both conceptual and technical aspects. There exist many predictive models, and an integrative evaluation can help understand their performance in handling the multifacets of biodiversity. This study aims to evaluate the performance of these modelling approaches to predict both alpha- and beta-diversity in diverse ecological contexts. Location North Yellow Sea, China. Methods The biodiversity models follow three strategies, "assemble first, predict later", "predict first, assemble later" and "assemble and predict together". Hill diversity profile, Fisher's log-series parameter and the distance decay of similarity are used to measure alpha- and beta-diversity. The evaluation study is conducted based on seasonal bottom trawl surveys from October 2016 to August 2017 in North Yellow Sea, China, allocated to coastal and offshore areas. We evaluate the predictive power of the models using cross-validation. Results Following the "assemble first, predict later" approach, macroecological model (MEM) provided the most accurate prediction overall, whereas stacked species distribution model (SSDM) and joint species distribution model (JSDM), following the second and third modelling approaches, tended to overestimate alpha-diversity and underestimate beta-diversity. The performances of SSDM and JSDM could be improved by moderately down-weighting rare species. The relative performances of the three modelling approaches were consistent among seasons and spatial regions. Main conclusions The superior performances of MEM in a range of temporal and spatial contexts favour the "assemble first, predict later" approach and imply a tight community assembly in the studied area. The overall predictive powers of varying models suggest that the spatial pattern of marine biodiversity could be fairly well predicted with commonly accessible hydrologic data in a mesoscales. The approach of multi-model evaluations is applicable to a variety of ecosystems for biodiversity prediction.