Xiao B et al. (2025), Improving Distribution Prediction by Integratin...

ECB-ART-54062

Ecol Evol 2025 Jul 06;157:e71747. doi: 10.1002/ece3.71747.

Show Gene links Show Anatomy links

Improving Distribution Prediction by Integrating Expert Range Maps and Opportunistic Occurrences: Evidence From Japanese Sea Cucumber.

Xiao B , Yuan S , Bede-Fazekas Á , Zhou J , Song X , Lin Q , Cui L , Zhang Z .

???displayArticle.abstract???
In an era of biodiversity crisis, it is critical to perform biodiversity assessments to better inform conservation strategies. In this regard, species distribution models (SDMs) represent a widely used tool for biodiversity assessment. Despite their popularity, the accuracy of SDM predictions has long been criticized because we have incomplete or biased information on species distribution. To overcome this limitation, researchers have proposed improving predictions of SDMs by integrating different types of distribution data, but this idea has rarely been explored in the marine realm. In this study, we explored the idea of data integration using the Japanese sea cucumber, whose distribution is known to be restricted by freshwater discharge of the Yangtze River. We first fitted SDMs for this species based on opportunistic occurrence records via four modeling algorithms, then built two types of ensemble models using stacked generalization: an ensemble model that solely used four model predictions and an expert-informed ensemble model that further accounted for distance to the IUCN expert range map. Our results showed that integrating an expert range map into the opportunistic occurrence model improved distribution prediction by avoiding overprediction in the south of the dispersal barrier for this species. Our study highlights the benefits of integrating expert range maps into opportunistic occurrence SDMs, which improve the reliability of species' spatial distributions.

???displayArticle.pubmedLink??? 40625339
???displayArticle.pmcLink??? PMC12229810
???displayArticle.link??? Ecol Evol

???attribute.lit??? ???displayArticles.show???

	FIGURE 1. Study design for this study.
	FIGURE 2. Predictive performance of six species distribution models fitted by different strategies. GAM, generalized additive model; GLM, generalized linear model; MaxEnt, maximum entropy.
	FIGURE 3. Comparison of habitat suitability predicted by the two types of ensemble models. Habitat suitability predicted by the two ensemble models (a) within and (b) outside expert range map. The dashed lines in panels (a) and (b) show 1:1 reference lines. The color legend in panel (b) indicates the distance to expert range polygon.
	FIGURE 4. Spatial distribution patterns of the Japanese sea cucumber predicted by species distribution models. (a) Opportunistic occurrence records, (b) IUCN expert range map, (c) spatial prediction by generalized linear model, (d) spatial prediction by generalized additive model, (e) spatial prediction by maximum entropy, (f) spatial prediction by random forest, (g) spatial prediction by occurrence‐only ensemble model, and (h) spatial prediction by expert‐informed ensemble model. To improve visualization of maps, we aggregated habitat suitability maps into 30 arcmin spatial resolution. High‐resolution figures can be found in Figure S6.