This study area including the waters 50 km offshore of Fujian, Guangdong, and Guangxi provinces, as well as Hainan and Taiwan Islands, to encompass the known distribution range of S. chinensis in China (Yong et al., 2023). The precise coordinates of this region are 17.71 - 27.23° N, 107.97 - 122.49° E. This area has a coastline of approximately 17,216 km, and comprises 33 bays, over 100 estuaries with width more than 200 m. Currently, there are eight regular populations of S. chinensis (population size >10) and a few scattered individuals in the study area ( Figure 1 ; Appendix 1 , Supplementary Table S1 ).

Historical records were digitized from maps that extracted from documents including academic dissertations, journal papers, monographs, scientific reports and conference papers published since 1950 (Wang, 1965; Huang et al., 1978; Jefferson, 2000; Liu and Huang, 2000; Yang and Deng, 2006; Chen et al., 2009, 2010; Wang, 2011; Hung, 2012; Wu, 2013; Wu et al., 2014; Chen et al., 2016a; Wang et al., 2017; Huang et al., 2018; Bao et al., 2019; Li et al., 2019; Liu et al., 2020, 2021; Peng et al., 2020; Zeng, 2021; Guo et al., 2022). For records with only toponyms, we determined the latitude and longitude coordinates on the map using ArcGIS Pro 3.0 (ESRI), based on our knowledge of the area and S. chinensis.

The interview survey was conducted in coastal communities outside the current distribution range of S. chinensis, covering 49 coastal villages or harbors in Fujian, Guangdong, and Hainan provinces from March 2010 to November 2011. Each interviewee was informed about the interview purpose and that all data were anonymous before interview started. After obtaining consent, interviewees reporting S. chinensis sightings were further asked to describe key morphological traits (body color, dorsal fin, long beak) and identify S. chinensis from a set of photographs of related dolphin species, which was to verify identification accuracy. Maps were provided to enable interviewers to indicate their fishing areas and the locations where S. chinensis had been sighted. All descriptions by the interviewee were elicited without any prompts to ensure the validity of their responses and reduce recall bias. We evaluated the reliability of reported sightings by cross-referencing interviewee’s description with established ecological knowledge of S. chinensis. The questionnaire included interviewees’ fishing experience, the time and location of last sighting of S. chinensis, and the number of individuals observed per sighting. Each questionnaire took 10–15 minutes to complete. A total of 540 valid questionnaires were obtained, of which 297 provided sighting information about S. chinensis, yielding 252 sighting records after cross-referencing. The historical occurrences from interviews were categorized into two datasets: (1) the sightings caught by interviewees after 2009, and (2) sightings before 2009 but never observed again.

The occurrences from documents and interviews were classified based on location precision: Accuracy-1, explicit latitude and longitude available (e.g., digitized from maps); Accuracy-2, noted with place names from documents; Accuracy-3, digitized on the map based on S. chinensis sighting locations provided by interviewees. This dataset was used in three primary applications: (1) all occurrences were used to reconstruct the historical distribution of S. chinensis during before 1990, 1990–2009 and 2010-2022. To clearly illustrate the distribution pattern, we developed a 5 km × 5 km grid layer over the study area, and then overlapped it with all historical occurrences. Grids containing at least one occurrence were used to represent historical distribution of S. chinensis during the three periods. In detail, all the occurrences from documents during 1990–2022 were assumed stable unless disappearances have been documented, and together with occurrences during 2010–2022 from questionnaires, were used to infer current distribution (during 2010-2022) of S. chinensis by overlapping with grids. The current distribution and the grids overlapped with historical occurrences from documents or sightings caught before 2010 (or 1990) but S. chinensis was never observed again were used as the distribution during before 2010 (or 1990). If a grid contained occurrences from multiple periods, it was classified into the last period. (2) Data with Accuracy-1 were used to simulate the baseline suitable habitats in 1990 and develop 95% minimum convex polygon (MCP). (3) Occurrences with Accuracy-2 and 3 were used to validate the habitats outside the current distribution range ( Table 1 ; Figure 2 ).

Here, we defined baseline suitable habitat as the maximum extent of suitable areas under natural conditions, thus corresponding to the fundamental niche (Kearney and Porter, 2004). Seven natural variables were used to represent habitat configurations of S. chinensis ( Table 2 ).

Chlorophyll a (Chla) serves as a proxy for primary productivity in aquatic ecosystems, reflecting potential food availability for S. chinensis (Bao et al., 2019; Lin et al., 2020). We obtained annual mean sea surface Chla concentrations and sea surface temperature (SST) data (4 km resolution) for 2002–2022 from the National Aeronautics and Space Administration (NASA) OceanColor Database (https://oceancolor.gsfc.nasa.gov/). Near-continental SST grids with anomalous values were replaced by the average of adjacent grids. Sea surface salinity (SSS) data (5 km resolution) for the 2000s and 2010s were downloaded from Bio-ORACLE’s global dataset (Assis et al., 2024).

To evaluate potential impacts on Chla, SST and SSS from climate change, we conducted trend analysis using linear fits to check if any significant changes happen during the past decades. Our results revealed no significant temporal changes in Chla across most of the study area, while SST showed a nonsignificant warming trend ( Appendix 4 , Supplementary Figure S1 ). Consequently, we derived long-term mean values for Chla and SST (2002-2022) as static environmental variables respectively. SSS data, representing decadal averages (2000s and 2010s), were aggregated to align temporally with the Chla/SST baseline, and also was calculated the average values. Our approach of using the mean values for those dynamic oceanographic variables can also eliminate the interannual fluctuations.

Bathymetry (hereafter, Bathy) data were downloaded from the General Bathymetric Chart of the Oceans (https://www.gebco.net/), with a spatial resolution of approximately 500 m. Considering that habitat in 1990 likely approximated baseline suitable habitats, a 50 km offshore buffer was generated using coastline in 1990, developed based on 30 m resolution Landsat images (Unpublished manuscript), and served as a spatial mask for extracting natural variables and cover occurrences since 1990. Euclidean distance to coastline (EDC) has been proven to be an important indicator of S. chinensis habitat (Wu et al., 2017; Chen et al., 2020). We calculated the EDC in ArcGIS Pro 3.0 using the Euclidean Distance Tool and the coastline in 1990. The distance to estuaries is an important factor influencing S. chinensis habitat selection (Chen et al., 2016b). Numerous narrow estuaries (< 200 m width) along the study area exhibit spatial congruence with coastal proximity metrics (EDC) at regional scales. Given the heightened susceptibility of these fluvial features to anthropogenic geomorphic alterations, EDC was prioritized in our analytical framework as a stable spatial proxy. We than extracted the locations of estuaries with widths ≥ 200 m using Google Remote Sensed imagery with a 4 m spatial resolution (https://www.google.com/earth). To further assess the influences of estuary sizes, we classified estuaries into two categories based on width: 200–400 m and > 400 m. We then calculated the Euclidean distance to estuaries of 200–400 m (EDE24) and estuaries > 400 m (EDE4) respectively.

All variables were prepared using ArcGIS Pro 3.0 and resampled to a 500 m resolution. Pearson correlation coefficient (r) among the seven natural variables were calculated, and for each pair with high correlation (|r| > 0.75) (Kumar and Stohlgren, 2009; Kramer-Schadt et al., 2013), retained the variable most ecologically important for S. chinensis ( Appendix 5 , Supplementary Figure S2 ). All seven variables were ultimately retained for habitat simulations.

Maxent (version 3.4.4) was employed to simulate suitable habitats for S. chinensis (Phillips et al., 2006) due to its ability to generate robust projections with varying sample sizes, and its capacity to incorporate correlated variables (Wisz et al., 2008; Elith et al., 2011). To eliminate sampling bias caused by inhomogeneous survey effort (Phillips and Dudík, 2008; Kumar and Stohlgren, 2009; Kramer-Schadt et al., 2013), a spatial filtering method was applied to remove duplicate presence records within the same 2 km × 2 km grid ( Appendix 3 , Supplementary Table S3 ). To simulate the baseline suitable habitat and mitigate potential niche truncation when modeling distributions affected by anthropogenic range contractions (Pang et al., 2022), we tried two handling methods: (1) we firstly constructed models with samplings incorporating documented occurrences and questionnaire-derived occurrences (Accuracy-1, 2 and 3). However, this method introduced bias in local areas, e.g., failing to capture key areas like Sanya waters. In fact, prediction bias can arise from integrating datasets of heterogeneous accuracy levels, a process that may exacerbate geographic sampling bias due to uneven record counts (Elith et al., 2011; Bao et al., 2019). As an alternative, we exclusively employed post-1990 Accuracy-1 occurrences as model inputs. This approach resulted in only a marginal prediction divergence (1.2%) compared to models incorporating questionnaire-derived occurrences, while ensuring data quality control throughout the analytical framework. (2) we strategically adjusted the regularization multiplier to 1.5 to prevent overfitting (Elith et al., 2010, Merow et al., 2013). Only the hinge feature was used in Maxent to generate smooth models, improving model performance without significantly increasing model complexity. Cross validate was selected to run the model 10 times (Phillips and Dudík, 2008; Elith et al., 2010). Overall model performance was evaluated by Area Under the Curve (AUC). A random prediction model should have an AUC value of 0.5, while a perfect prediction model should have an AUC value of 1, and thus an AUC > 0.75 can be considered an effective model (Phillips et al., 2006). In addition, maximizing the sum of sensitivity and specificity (Max SSS) threshold selection method was used to classify predicted habitats into suitable or unsuitable habitats (Liu et al., 2013).

The baseline suitable habitat predicted by Maxent was validated by overlapping it with historical occurrences from documents and questionnaires (Accuracy-2 and 3), and counted the percentage of occurrences as predicted habitats ( Figure 2 ).

To assess the impacts of five anthropogenic disturbances on distribution shrinkage of S. chinensis, we developed five indices to quantify anthropogenic disturbances in the study area ( Table 3 ).

The historical distribution grids showed that S. chinensis were widely distributed (474 grids) in coastal waters along southeast China before 1990. During 1990-2009, S. chinensis disappeared from Putian waters, resulting a total of 469 grids in this period. During 2010-2022, distribution grids of S. chinensis shrank to 351 ( Figure 3 ).

All models had an AUC > 0.9, indicating good performance of those models. Suitable habitats of S. chinensis under natural conditions showed a relatively continuous distribution pattern across Fujian, Guangdong and Guangxi provinces, but were only found in the coastal waters of Haikou, western Hainan Island, and western Taiwan Island. However, only 167 of 286 (58.4%) occurrences (Accuracy-2 and 3) were located in predicted suitable habitat, indicating potential underestimation of suitable habitat in the waters north of Ningde, Putian, from Xiamen to Shantou, Shanwei, from Yangjiang to Zhanjiang, as well as southern Leizhou Peninsula and southeastern Hainan Island. In addition, all the 56 grids contained S. chinensis sightings reported by interviewees before 2009 but not afterward fell outside the current distribution boundary delineated by representative ranges, with 23 grids in Fujian, 21 in Guangdong and 12 in Hainan. Notably, only 60.7% (n = 34) of these were situated within baseline suitable habitat (15 grids in coastal waters of Fujian and 4 grids in Shanwei waters), suggesting that the population may have disappeared despite suitable natural conditions; while 39.3% (n = 22) occupied environmentally unsuitable zones, suggesting the spatial extent of habitat contraction has surpassed model projections ( Figure 4 ).

Among the natural variables, Chla exhibited the highest permutation importance (56.9%), followed by EDE4 (11.8%), Bathy (11%), SSS (8.9%), and EDC (8.1%). EDE24 and SST had the lowest permutation importance (< 5%). The probability of S. chinensis presence increased with Chla but decreased when Chla exceeds 7 mg/m³. S. chinensis was most prevalent in waters up to 10 km offshore and 40 km from estuaries, with depth < 30 m. The probability of S. chinensis presence increased with SSS within the range of 12 to 20‰. The S. chinensis adapted well when SSS varied between 18‰ and 34‰, and SST remained between 20°C and 27°C ( Figure 5 ).

A single candidate model was retained based on the AICc criterion (ΔAICc < 4), with an Akaike weight of 0.953, indicating robust model support. All five anthropogenic disturbances demonstrated statistically significant negative effects on the current distribution of S. chinensis in suitable habitats. The standardized regression coefficients revealed a clear impact hierarchy: PMA exerted the greatest effect (β = - 1.358 ± 0.150), followed by MFCP (β = - 1.231 ± 0.135), LCI (β = - 0.754 ± 0.091), CRI (β = - 0.522 ± 0.127), and SDI (β = -0.257 ± 0.092) ( Figure 6 ).

Understanding distribution shrinkage and its driving factors is critical for conservation management of threatened species but is constrained by scarce historical survey data. In this study, we reconstructed the historical distribution of S. chinensis in southeast China by integrating LEK from questionnaires combined with SDM. The reconstructed distributions revealed that, outside current presence areas, S. chinensis may have disappeared from northern Ningde, Putian waters, Shanwei, Yangjiang to Zhanjiang, southwestern Leizhou Peninsula, and Haikou waters since 1990, despite suitable natural conditions, suggesting that increased anthropogenic disturbances contributed to habitat loss in these regions.

The distribution of baseline suitable habitat predicted by Maxent indicated a high connectivity among regular populations in the absence of anthropogenic disturbances. Previous IUCN assessments of S. chinensis suggested a continuous potential distribution range along the Chinese coast while acknowledging a possible distribution gap (Jefferson and Smith, 2016). Our SDM predictions address this distribution gap and offer new insights into distribution dynamics of S. chinensis. Notably, spatial heterogeneity was observed in the primary environmental constraints of unsuitable regions. Deeper waters and lower Chla restricted the suitable habitat between Putian and Ningde, while lower Chla limited the continuous distribution of suitable habitat between Xiamen and Shantou. In addition, habitat discontinuities between Shantou and eastern Hong Kong, as well as between Haikou and western Hainan Island were primarily constrained by larger EDE4, while the southeastern waters of Hainan Island were restricted by low Chla ( Appendix 7 , Supplementary Figure S4 ).

Our findings potentially underestimated the range of baseline suitable habitat. For instance, 41.6% (119 of 286) historical occurrence records (Accuracy-2 and 3) were located outside predicted suitable habitats, which may be partially explained by temporally truncated ecological niches resulting from historical distributional contraction (Pang et al., 2022). Additionally, occurrences along eastern Hainan Island overlapped with the IUCN-suggested range but not with predicted habitats. This inconsistency likely stems from two mechanisms: (1) the existence of a potential narrow migratory corridor for S. chinensis along eastern Hainan Island requires further investigation; and (2) the recognition of two distinct ecotypes of S. chinensis may exclude eastern Hainan waters from predicted suitable habitats due to ecological differentiation (Lin et al., 2022a), thereby causing a mismatch with historical occurrence records.

Among natural variables, Chla was the most important variable affecting S. chinensis distribution in southeast China, similar to previous localized findings (Huang et al., 2018; Bao et al., 2019; Chen et al., 2020; Wu et al., 2021), due to its association with prey availability for S. chinensis (Wu et al., 2017; Lin et al., 2023). Higher probability of S. chinensis presence associated with Chla of 4–12 mg/m³, along with a peak at 7 mg/m³. This pattern reflects the ecological constraints of estuarine habitats where elevated Chla levels (>7 mg/m³) predominantly occur within 3 km of river mouths and heightened exposure to anthropogenic stressors. Additionally, sustained Chla elevation signaled escalating harmful algal bloom proliferation, and 10 mg/m³ can be a threshold for formal eutrophication (Chen et al., 2014). These blooms drive hypoxic conditions through microbial degradation of senescent biomass, consequently inducing distribution shifts of food sources (Guo et al., 2022). We also found that the probability of S. chinensis presence decreased as distances to the coast and estuary increased, confirming the coastal resident of this species (Jefferson and Smith, 2016). In fact, smaller EDC values were associated with higher Chla (r = -0.61), lower Bathy (r = -0.49), which are also characteristics of waters with higher fish densities (Zhou et al., 2016; Lin et al., 2023). In addition, S. chinensis is not strictly confined to estuarine habitats, for instance, a population inhabits deep waters with low Chla and high SSS along southwest Hainan (Li et al., 2016; Lin et al., 2022a), which may partly explain why eastern waters of Hainan were not predicted as suitable habitat by our Maxent model. Additionally, larger estuaries (> 400 m) had higher importance than smaller estuaries (200–400 m) in this study, probably due to their higher fish densities (Zhou et al., 2016; Wu et al., 2023). Meanwhile, the limited trophic pooling effect of small estuaries and the adaptability to larger variation of salinity may have contributed to non-estuarine ecotype of S. chinensis population in southwest Hainan (Lin et al., 2022a). The contribution of SST to S. chinensis distribution was limited in this study.

The fishery industry including mariculture and fishing negatively associated with S. chinensis presence in suitable habitat, due to competition for space and food resources, as well as increased health risks and mortality for S. chinensis (Lin et al., 2021). Our findings indicate that mariculture exerts the most significant adverse impact on the distribution shrinkage of S. chinensis. The mariculture area along the Chinese coast increased dramatically by 6,191.78 km² during 1990 to 2022 (Yin et al., 2023). Notably, over 80% of mariculture was located in shallow waters < 20 m, and more than 90% of mariculture was distributed within 20 km offshore over the past three decades (Liu et al., 2023). This spatial pattern indicates a highly overlap with S. chinensis habitats that directly caused habitat loss, and other risks including entanglement, entrapment, behavioral modification and pesticide poisoning (Bath et al., 2023). Despite this, the negative effects of mariculture could still be underestimated because our dataset, which was developed using remote sensing data, did not account for stick and rack culturing methods commonly used for oysters—one of the most significant mariculture species along China’s coast (Yin et al., 2023; Yu et al., 2023). The maximum fishing intensity in China occurred within 30 km offshore (Geng et al., 2023), and overfishing had caused a drastic decline in coastal fish stocks since the 1990s (Liang and Pauly, 2017; Ding et al., 2021). The MFCP in this study reflects the intensity of regional fishery development and the associated entanglement risk to S. chinensis. In fact, small-scale fisheries have a greater global impact on small cetaceans than large-scale fisheries. It is estimated that approximately 22% of small cetacean populations globally are at risk of extinction due to the threats posed by small-scale fisheries (Temple et al., 2024). Field surveys revealed high spatial density of fixed and drifting gillnets in estuaries, which effectively exclude S. chinensis from their habitats in the estuaries. Furthermore, the negative impact of MFCP may still be underestimated due to insufficient data on small-scale fisheries and unreported incidents of illegal fishing methods such as electrofishing and piscicide use (Thomas et al., 2019; Zhang et al., 2023). Therefore, mariculture and inshore fisheries were likely the primary disturbances that have removed S. chinensis from the study area, given the continues distributions of this species in southeast China 30 years ago (Wang et al., 2012; Wu et al., 2014). For instance, the distribution shrinkage validation showed that all 15 grids disappearing in Fujian coast post-2010 were located in high PMA waters, whereas the 4 grids disappearing in Shanwei were located in high MFCP waters.

Previous localized-scale studies demonstrated S. chinensis behavioral adaptations to reclamation pressures, such as artificial shoreline avoidance, displacement from construction zones, and shifts in core habitat use (Wang et al., 2017; Karczmarski et al., 2017; Huang et al., 2022). Our regional-scale analysis revealed significant negative effects of both LCI and CRI on S. chinensis distribution. Collectively, this multi-scale evidence demonstrates that cumulative anthropogenic modifications across terrestrial and marine coastal zones, as well as other terrestrial stressors like pollution, have been a primary driver of S. chinensis range contraction over the past three decades. Notably, LCI/CRI effect sizes were lower than those of fisheries, reflecting different intrinsic mechanisms. Compared to large-scale habitat exclusion by fisheries, the coastal urbanization and reclamation predominantly caused habitat encroachment at the periphery of the S. chinensis representative habitats. On the other hand, LCI/CRI impacts exhibit temporal decoupling, with ecological consequences lagging behind disturbance initiation, a critical consideration routinely neglected in conservation planning. The delayed response, exemplified by cumulative pollution and irreversible alterations to biogeochemical cycling (Lin and Yu, 2018), coastal hydrology, sediment dynamics, poses distinctive management challenges (Huang et al., 2022). Such anthropogenic modifications not only permanently alter benthic topography and disrupt hydrological connectivity but also initiate cascading degradation of essential habitat features in long term. Absent science-informed regulatory frameworks, synergistic alterations across physical environments—including sediment geomorphology shifts, toxicant bioaccumulation, and resultant trophic cascade disruptions—may collectively exceed critical ecological resilience thresholds and result irreversible habitat collapse (Huang et al., 2022).

The present study documents significant distributional avoidance of S. chinensis in areas with intense shipping activities, aligning with localized findings from Sanniang Bay (Lin et al., 2022b). Paradoxically, persistent populations coexist with dense shipping in major estuarine systems (e.g., Pearl River Estuary, Xiamen Bay) (Guo et al., 2022; Zhang et al., 2025). Unlike spatial exclusion caused by mariculture, fishing or reclamation, shipping impacts manifest through two primary pathways: (1) acute mortality from vessel collisions and chronic physiological stress from underwater noise pollution (Li et al., 2018; Yuen et al., 2025); (2) bioenergetic constraints through reduced foraging efficiency that diminishes population fitness (Lin et al., 2021). However, these chronic stressors may drive population declines that ultimately manifest as detectable range contractions only after multidecadal latency (e.g., >30 years) —a temporal scale exceeding that of most observational studies (Jefferson et al., 2023). Furthermore, estuaries with dense shipping channels often incidentally create ephemeral refugia or corridors by excluding competing anthropogenic activities (e.g., mariculture, fishing), where temporarily enhanced prey availability partially offsets shipping-derived ecological costs (Zhang et al., 2025). This dual framework delineates the nonlinear, temporally mediated interactions between acute shipping impacts (direct mortality, behavioral disruption) and persistent anthropogenic pressures (habitat degradation from coastal development).

The shrinkage of S. chinensis distribution was detected by LEK. However, questionnaires may be limited by the availability of experienced interviewees (McMillan et al., 2019). Second, predicted baseline habitat for S. chinensis, and thus the distribution shrinkage, may still be underestimated due to niche truncation caused by early anthropogenic disturbances. Third, quantifying anthropogenic disturbances remains challenging, especially the inshore fishing—a major threat to coastal cetaceans, due to insufficient data. Low resolution MFCP data limits the capacity for fine-scale interpretation of localized areas and cross-regional comparability, while potentially diminishing its relative importance rankings in GLM. Continued decline of remnant populations under persistent anthropogenic disturbances ( Appendix 1 , Supplementary Table S1 ) may drive future distribution shrinkage, but this demographic process beyond the scope of our current framework. Crucially, prey availability dynamics fundamentally influence S. chinensis distribution (Lin et al., 2021; Guo et al., 2022; Lin et al., 2023), the absence of spatially explicit, time-series data on prey biomass across our study domain precluded quantification of this driver.

Future research should focus on three interconnected priorities to address coastal cetacean conservation under climate change: (1) unraveling synergistic interactions among climate change, prey dynamics, and anthropogenic stressors in shaping population resilience and distribution; (2) advancing mechanistic models to quantify climate-mediated habitat shifts, such as those driven by ocean acidification, sea-level rise, and prey biogeography changes (Lawlor et al., 2024), and their cascading impacts on cetacean behavioral plasticity and demographic responses (Pirotta et al., 2018; Celine et al., 2021); and (3) designing dynamic marine protected areas (MPAs) that account for spatiotemporal mismatches between prey availability and cetacean habitat use. Simultaneously, high-resolution local-scale fisheries data should be integrated to address compounding threats to coastal cetacean population dynamics arising from overfishing and cumulative stressors in nearshore mariculture. Given the projected compounding effects of future climate change and intensified anthropogenic disturbances, coupled with the current severe deficiency in MPAs coverage across China’s coastal waters (Bohorquez et al., 2021; Huang et al., 2022), future efforts can prioritize developing resilient MPA networks using Artificial Intelligence (AI)-enabled decision-support tools to evaluate cost-benefit trade-offs of conservation actions; and fostering interdisciplinary frameworks that leverage AI for data integration, predictive modeling, and adaptive management (Luo et al., 2024). These strategies will bridge theoretical-practical gaps, enhance understanding of cetacean distribution dynamics, and support adaptive conservation strategies for coastal marine megafauna, including S. chinensis.