The study combined four modeling domains: a global model, or G-SDM, and independently calibrated models for the Asia–Pacific native range, Europe and North America. Each geographic model had separate terrestrial and nearshore components. The terrestrial component represented environmental conditions associated with the crab’s freshwater phase, whereas the nearshore component represented brackish-water conditions potentially relevant to reproduction and early development. This division reflected the species’ catadromous life cycle, in which juvenile and growth phases occur in freshwater before reproductive migration toward marine environments. The nearshore outputs were interpreted as potentially suitable environments, not confirmed spawning grounds or nursery habitats.

The researchers compiled 4,408 occurrence records from the Global Biodiversity Information Facility, published literature and field investigations. These were records gathered before spatial filtering, not the final number used by every model. To reduce spatial sampling bias and autocorrelation, the records were thinned to retain no more than one record per 10 × 10 km grid cell, then intersected with the appropriate terrestrial or nearshore predictor layers. After thinning and intersection, the native-range model retained 56 terrestrial and 31 nearshore records; the global model retained 685 and 271; the European model retained 472 and 216; and the North American model retained 67 and 28. Eight algorithms—including artificial neural networks, classification trees, generalized boosting, generalized linear models, multivariate adaptive regression splines, MaxEnt, random forests and surface range envelopes—were fitted with repeated pseudo-absence sampling and random holdout validation.

Random holdout validation indicated generally good discrimination, but spatial block cross-validation produced lower performance, as expected when nearby records cannot be divided randomly between training and testing data. Across the four geographic models, mean spatial-validation ROC-AUC values ranged from 0.77 to 0.94 and mean true skill statistic values from 0.55 to 0.77, indicating moderate to good discriminatory ability under spatially separated validation. The study also used a multivariate environmental similarity surface analysis to identify where future conditions fell outside the current calibration range. Negative similarity values occurred mainly in Western Europe and along the North Sea coast, parts of coastal North America, and coastal East Asia. Some of these extrapolation areas overlapped cells projected as suitable, making those predictions less secure.

Regional models identified different environmental associations for terrestrial suitability. In the native range, the leading predictors were annual mean temperature, distance to the shoreline, isothermality, mean temperature of the wettest quarter and annual precipitation. Distance to the shoreline ranked first in Europe, followed by annual mean temperature, wettest-quarter temperature, isothermality and annual precipitation. In North America, annual precipitation was the leading predictor, followed by annual mean temperature and precipitation of the warmest quarter. Using the central 95% ranges associated with high suitability, the native range extended from 9.3°C to 20.4°C in annual mean temperature, compared with 4.3°C to 15.7°C in Europe and 3.8°C to 15.4°C in North America. The invaded regions therefore contained cooler conditions within their modeled high-suitability ranges. The authors cautioned that such differences can arise from regional environmental availability, range occupancy, environmental filtering and calibration coverage; they do not by themselves demonstrate climatic adaptation, niche evolution or greater physiological tolerance.

The nearshore models showed more consistent predictor rankings. Water depth, mean chlorophyll concentration, mean sea-surface temperature and mean salinity were the principal predictors in all three regional models. The central 95% high-suitability range for mean sea-surface temperature was 13.8°C to 25.5°C in the native range, 7.3°C to 18.7°C in Europe and 7.0°C to 24.5°C in North America. Chlorophyll ranges were narrower in both invaded regions than in the native range. Despite this relative consistency, the terrestrial and nearshore outputs frequently failed to overlap. For example, some parts of Newfoundland and the northern Mediterranean were classified as suitable by only one component. The study did not show that these mismatches restrict reproduction, because the models did not include hydrological connectivity, actual dispersal pathways or confirmed reproductive sites.

Under the low-emissions SSP1-2.6 pathway, projected changes were relatively limited. With SSP2-4.5 and especially SSP5-8.5, terrestrial suitability shifted northward in all three regional models. Europe projected gains across Scandinavia and northern European Russia, including extensive areas north of 60°N in the later high-emission periods. North America projected increasing suitability in eastern Canada, Newfoundland, Labrador, the Gulf of St Lawrence and southern Alaska. In the native-range domain, suitability increased in northeastern China, the Korean Peninsula and other regions north of 40°N while declining across eastern and southern China. By the 2090s under SSP5-8.5, high-suitability areas in eastern China were nearly absent in the native-range projection. The global model likewise projected new high-latitude areas in northern Europe, northern North America and the Russian Far East, together with losses along the lower-latitude margins of the current modeled range.

Nearshore projections changed less dramatically than terrestrial ones, but this did not produce complete life-cycle overlap. In Europe, terrestrial suitability expanded into northern Scandinavia and northern European Russia without a corresponding expansion of suitable nearshore brackish-water conditions under later SSP2-4.5 and SSP5-8.5 scenarios. North American projections showed a similar discordance: new terrestrial areas appeared around Newfoundland, Labrador and farther north even where suitable nearshore conditions were absent. A shared principal-component analysis of the terrestrial projections quantified changes in environmental space. Current–future overlap generally declined toward later periods and higher-emission scenarios, reaching its lowest value in all three regions under SSP5-8.5 in the 2090s. The native range’s 95% kernel-density area declined from about 6.6 to 4.1 squared principal-component units, while its convex-hull area fell from about 16.5 to 10.7 squared units. North America’s corresponding areas declined from approximately 9.2 to 6.3 and from 23.0 to 16.0 squared units under the same scenario.

Europe followed a different trajectory in the environmental-space analysis. Its 95% kernel-density breadth remained comparatively stable, while occupied grid-cell counts and convex-hull extent increased under several future scenarios, peaking at approximately 575 cells and 24.3 squared principal-component units under SSP1-2.6 in the 2090s. Expansion into environmental space outside the current suitable range remained small everywhere, ranging from 0.2% to 2.7%. The unfilling index—the portion of current suitable environmental density not covered under future conditions—rose to 0.032–0.170 in the native range and 0.063–0.237 in North America by the 2090s, compared with 0.018–0.042 in Europe. These are terrestrial projections within predefined regional domains, so the native-range contraction does not imply loss across a wider area beyond that model’s northern boundary.

The results point to a geographically uneven reorganization of potential habitat rather than a uniform increase in the crab’s modeled range. Europe retained more of its current terrestrial environmental space, whereas the native range and North America showed stronger contraction and redistribution under late high-emission conditions. River-network post-processing identified candidate areas in Europe and North America where predicted terrestrial suitability, selected river systems and predicted nearshore suitability spatially coincided, but these were screening zones rather than estimates of actual dispersal distance or hydrological connectivity. The study’s main limitation is that the models estimate environmental suitability rather than actual establishment: they omit dispersal constraints, biotic interactions, verified reproductive sites, river-flow direction and demographic responses. Independent field observations, reproductive records and process-based investigations will therefore be needed to determine whether the projected shifts correspond to changes in Chinese mitten crab populations.

Subject of Research: Climate-driven redistribution of potential Chinese mitten crab habitat across native and invaded regions

Article Title: Assessing the Potential Distribution of Eriocheir sinensis Under Climate Change: Regional Contrasts and Spatial Mismatches in Terrestrial and Nearshore Suitability

Article References: Gong, H., Yang, X., Bidegain, G., Wang, Y., Jiang, Y., Han, T., Dong, J., & Zhang, X. (2026). Assessing the Potential Distribution of Eriocheir sinensis Under Climate Change: Regional Contrasts and Spatial Mismatches in Terrestrial and Nearshore Suitability. Ecology and Evolution, 16(10), Article e74351. https://doi.org/10.1002/ece3.74351

Image Credits: AI Generated

DOI: 10.1002/ece3.74351

Keywords: Chinese mitten crab, Eriocheir sinensis, species distribution models, climate change, habitat suitability, nearshore ecology, freshwater ecosystems, CMIP6 projections, invasive species, catadromous life cycle, Assessing, Potential