Breeding habitat characteristics of Saunders’s Gull (Saundersilarus saundersi) in the Liao River Estuary National Nature Reserve using MaxEnt Authors: Lu Yang, Zhi Zhang, Weikang Zhang, Tong Zhang, Huan Meng, Yucheng Wang, Beixuelai Zhou, Jinshuang Wang, Zhou Zhou Published: 2026 ABSTRACT Conservation of breeding habitats is essential for sustaining populations of endangered waterbirds, yet quantitative links between breeding behavior and environmental drivers remain poorly understood. Saunders’s Gull (Saundersilarus saundersi), a globally endangered species, has its largest breeding population in the Liao River Estuary National Nature Reserve (LRENNR), China. We integrated field survey data (53 sampling points, 24 environmental variables) with Maximum Entropy (MaxEnt) modeling and GIS analysis to identify suitable breeding habitats and key environmental drivers. Wet grassland was the dominant land-use type (40.17% of the study area). Elevation, wind speed, soil pH, and soil electrical conductivity collectively explained 70.90% of breeding habitat suitability, with higher predicted breeding occurrence probability associated with ranges of 13.80–16.97 m, 19.50–23.78 km/h, pH 6.65–7.50, and > 7,821.00 µS/cm, respectively. Only 2.05% (0.0802 km2) of the study area was identified as highly suitable habitat, highlighting the limited availability of potential breeding sites. These findings provide quantitative evidence for spatially targeted conservation planning and demonstrate the value of combining species distribution modeling with field-based behavioral analysis for endangered waterbird conservation.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1038/s41598-026-62146-6. Introduction Birds serve as key ecological indicators1, and their conservation should align with their ecological habits while minimizing human interference2,3. Wetlands support approximately 20% of the world’s bird species for feeding, breeding, or wintering4, and many endangered birds depend on wetland habitats5,6. The diversity of waterbird populations is essential for evaluating wetland health7. Habitat quality directly affects bird population size and resilience to external pressures8, and natural environmental factors such as topography and climate fundamentally shape breeding ground selection9. The study of bird habitat10 is an important theoretical basis for protecting bird populations and diversity11–13, while human interference, landscape connectivity14 and food abundance also have a great impact on bird habitat suitability. The Saunders’s Gull (Saundersilarus saundersi, formerly Larus saundersi) is a globally endangered seabird listed in the Red data book15 and Red list of threatened species16. Its breeding range is restricted to China and South Korea, with wintering grounds extending to Japan17. The species was first described in 187118, and its breeding grounds were not discovered until the late 20th century19. Its breeding season extends from late March to mid-July, with chicks hatching around mid-June. Saunders’s Gull is one of the few waterbirds that like to live on land, which is conducive to the development of this study. The northernmost breeding grounds in China are located in the Liao River Estuary National Nature Reserve (LRENNR)20, which is the largest breeding habitat for Saunders’s Gull worldwide. As of the 2022 breeding season census conducted by the LRENNR Saunders’s Gull monitoring station, approximately 11,088 individuals were recorded in the Nanxiaohe breeding ground, representing approximately 77% of the global population of 14,400 mature individuals21. Therefore, this study takes it as the research area for investigation. While previous studies have documented population dynamics22, general habitat characteristics23, and breeding ecology24,25, quantitative analyses linking breeding behavior to specific environmental variables remain lacking. Research on food sources26, genetics and populations also laid the foundation for the protection of Saunders’s Gull at the genetic level27–29. These studies have identified key threats such as habitat degradation, predation, and food limitation, and have provided valuable insights into reproductive behavior and population structure. However, existing studies have three critical limitations: (1) they rarely quantify the relative importance of environmental variables for breeding habitat suitability; (2) they do not identify optimal environmental thresholds for breeding occurrence; and (3) they fail to link specific breeding behaviors to measurable habitat characteristics. Consequently, conservation planning lacks the quantitative evidence needed for spatially targeted habitat management. In addition, although species distribution models (SDMs), particularly the Maximum Entropy (MaxEnt) model30,31, have been widely applied to assess habitat suitability, their application to Saunders’s Gull has been limited and rarely incorporates field-based behavioral data. As a result, the mechanisms by which environmental factors influence breeding habitat selection and behavior remain insufficiently understood. To address these gaps, this study integrates field survey data with MaxEnt modeling and GIS analysis to quantitatively evaluate breeding habitat suitability and its environmental drivers for Saunders’s Gull in LRENNR. Furthermore, this study links habitat characteristics with different breeding behaviors, providing a more comprehensive understanding of species-environment relationships, and the suggestions for its breeding habitat protection and planning are proposed (Supplementary Fig. S1). By combining spatial modeling with field-based ecological observations, this research advances current approaches to habitat assessment for endangered waterbirds. The following are the primary concerns addressed: (1) Which environmental factors are associated with the breeding habitat suitability of Saunders’s Gull? (2) Where are the highly suitable breeding areas, and what are their key environmental thresholds? (3) How do environmental characteristics vary across different breeding behaviors? Materials and methods Study area LRENNR is located at the southernmost tip of Panjin City, Liaoning Province, on the northern shore of Liaodong Bay. The reserve was mainly formed by the alluvial deposits of Liao River and the Xiaoling River. The area is flat and the elevation gradually increases as it gets farther from the coastline. It belongs to the warm temperate continental semi-humid monsoon climate zone, with rain and heat in the same season and four distinct seasons. As shown in Fig. 1, the study site, Nanxiaohe breeding ground, occupies an area of 3.9074 km2 and is situated on the southernmost portion of the LRENNR coastline. Since it is an artificially constructed breeding ground, it consists of four ponds (Ponds NO.1, 2, 3 and 4, which serve as Saunders’ Gull habitat maintenance areas) and a water surface area at the south end. Every year, a significant number of Saunders’ Gulls breed here, making up half of the global total. However, Saunders’s Gull safety is frequently threatened because of the growth of nearby human construction land and oil production. Fig. 1Location of the Nanxiaohe breeding ground within the Liao River Estuary National Nature Reserve (LRENNR), Liaoning Province, China. (A) Regional context showing the position of LRENNR on the northern shore of Liaodong Bay, Bohai Sea. (B) The study site, consisting of four ponds (Nos. 1–4) and a southern water surface area (total area: 3.91 km2). Map generated by the authors using ArcGIS 10.7 (Esri, Redlands, CA, USA; https://www.esri.com) based on publicly available Landsat 8 OLI/TIRS imagery obtained from the Geospatial Data Cloud (https://www.gscloud.cn/). Location of the Nanxiaohe breeding ground within the Liao River Estuary National Nature Reserve (LRENNR), Liaoning Province, China. (A) Regional context showing the position of LRENNR on the northern shore of Liaodong Bay, Bohai Sea. (B) The study site, consisting of four ponds (Nos. 1–4) and a southern water surface area (total area: 3.91 km2). Map generated by the authors using ArcGIS 10.7 (Esri, Redlands, CA, USA; https://www.esri.com) based on publicly available Landsat 8 OLI/TIRS imagery obtained from the Geospatial Data Cloud (https://www.gscloud.cn/). Species distribution points sampling Sampling is based on the breeding ground’s accessibility, with the preservation of Saunders’ Gull and the minimization of disturbance as the criteria. Sample plots selected also refers to the “1996–2010 Sino-Japanese Cooperation Saunders’s Gull Survey and Banding Report” jointly published by the National Bird Banding Center, the Kitakyushu Saunders’s Gull Research Association of Japan, and other units (https://zh.z-lib.gs/book/30165275/967d56). Points were set up in four ponds of the Nanxiaohe breeding ground so that they were dispersed within the sample plots. Observe each sampling point for signs of Saunders’s Gull. Try to ensure that there is a water source near the sampling point. Use GPS equipment to record the travel trajectory and the longitude, latitude and altitude of each sampling point. To obtain more comprehensive data, 61 sampling points were set up to record nests, feces, food residue or carcasses, as shown in Supplementary Fig. S2. Environmental factors selection The primary reproductive behaviors of Saunders’s Gull involved mate selection, mating, nesting, raising chicks, foraging, and flying based on the founding of breeding habits22. Following initial screening and analysis, the primary factors influencing their behavior were 29 environmental factors in five categories: vegetation conditions, water source physical and chemical properties, soil physical and chemical properties, meteorological factors, and terrain factors. These factors were determined combined with the research area scale and the MaxEnt for species habitat assessment process. The 29 environmental variables evaluated in this study are as follows. Vegetation condition: vegetation height (VH); vegetation color—red (VR), green (VG), red-green (VRG); vegetation species—Suaeda salsa (L.) Pall. (S. salsa), Phragmites australis (Cav.) Trin. ex Steud. (P. australis), Sonchus oleraceus L. (S. oleraceus). Water physicochemical properties: total dissolved solids (TDS), water electrical conductivity (WEC), water salinity (WS), dissolved oxygen (DO), water pH (WpH), oxidation-reduction potential (ORP), seawater specific gravity (SG), water temperature (WT), nutrient solution fertility (NSF), water resistivity (WR), chemical oxygen demand (COD), chloride ion content (Cl). Soil physicochemical properties: soil temperature (ST), soil pH (SpH), soil electrical conductivity (SEC), soil moisture (SM), soil salinity (SS). Meteorological factors: wind speed (WS), temperature (T), relative humidity (RH), illuminance (I). Topographic factor: elevation (El). Full definitions, measurement methods, and units are provided in Supplementary Table S1. Vegetation types are mainly based on on-site records. Since the Nanxiaohe breeding ground is a mudflats, the vegetation types are relatively single, most of which are S. salsa, with only a small number of other wetland herbaceous vegetation. Here, P. australis represent taller wetland herbs, and S. oleraceus represents other wetland herbs. Vegetation color was used to differentiate S. salsa from other wetland plant species because the red coloration is primarily characteristic of S. salsa at mature stages. Vegetation color categories (VR, VG, VRG) were derived from manual visual classification conducted at each 1 m × 1 m plot. VR indicates plots where > 70% of S. salsa individuals displayed red coloration; VG indicates plots where > 70% remained green; and VRG indicates mixed or transitional plots. This classification is based on visual observation rather than spectral indices and was consistently applied by the same observer team at the Saunders’s Gull monitoring station in LRENNR. Data source Geographic information data The Landsat 8 OLI/TIRS satellite digital products (data identifier: LC81200322016184LGN01, spatial resolution 30 m) of Xinglongtai District, Panjin City, Liaoning Province were obtained from the open source database Geospatial Data Cloud (https://www.gscloud.cn/search) with a cloud cover of ≤ 5% for the analysis in the Nanxiaohe breeding ground. The 2016 image was selected because it represents the most recent cloud-free (< 5% cloud cover) scene during the breeding season for the study area, and land-use types in this strictly protected nature reserve have remained stable, with no major habitat conversion recorded between 2016 and 2022 (LRENNR Management Bureau, unpublished monitoring data). The land-use classification was ground-validated during the 2022 field survey. The downloaded Landsat 8 OLI/TIRS images were processed in ArcGIS 10.7 (Esri, Redlands, CA, USA; https://www.esri.com) for image clipping, supervised land-use classification, and delineation of the study area boundary. These spatial data were subsequently used for land-use mapping and as the geographic framework for visualization of the MaxEnt model outputs and field survey data. Field investigation Field surveys of habitat and Saunders’s Gull (survival traces and reproductive behaviors) were conducted daily at the Nanxiaohe breeding ground from July 1 to July 20, 2022 (20 consecutive days), corresponding to the late breeding (chick-rearing) stage. Sampling was conducted between 07:00 and 17:00 each day to capture diurnal variation. Although this period spans approximately three weeks and does not cover the full breeding season (late March to mid-July), the late breeding stage encompasses the complete range of breeding behaviors (mate selection, nesting, chick-rearing, and foraging), with vegetation and environmental conditions remaining relatively stable, thereby providing representative habitat use patterns. See Supplementary Table S2 for the specific survey content. For all soil and water variables, three repeated measurements were taken at each sampling point, and the mean value was used for analysis to ensure accuracy and representativeness. MaxEnt modeling Model tuning was conducted using the ENMeval R package32. We tested combinations of five feature classes—linear (L), linear + quadratic (LQ), hinge (H), linear + quadratic + hinge (LQH), and linear + quadratic + hinge + product (LQHP)—and regularization multipliers ranging from 0.5 to 4.0 (in increments of 0.5). The optimal model was selected based on the lowest AICc (corrected Akaike Information Criterion). The final model used linear and quadratic features (LQ) with a regularization multiplier of 1.0, consistent with recommendations for small sample sizes to avoid overfitting33. The processed environmental data and distribution data were imported into MaxEnt 3.4.3 for the final modeling analysis. Species distribution data removal To reduce spatial autocorrelation and avoid redundant sampling, a minimum distance threshold was applied to the sampling points. Based on the relationship between body size and flight initiation distance in waterbirds, Saunders’s Gull (body length ≈ 32 cm) has an estimated disturbance or alert distance of approximately 50 m34,35. Therefore, a 50 m minimum spacing was adopted as a criterion for sampling point selection. A 50 m buffer was generated around each sampling point in ArcGIS to identify overlapping areas (Supplementary Fig. S3). Overlapping buffers indicate spatial redundancy among sampling points. Eight points exhibited buffer intersections and were therefore removed to ensure independence among samples. After filtering, a total of 53 sampling points were retained for subsequent modeling. To further assess residual spatial autocorrelation, Moran’s I was calculated on the MaxEnt model residuals using inverse-distance weights36(R package ‘spdep’). The test did not detect significant spatial autocorrelation (Moran’s I = − 0.04, p = 0.72), indicating that the 50-m minimum distance criterion adequately reduced spatial dependence among sampling points. While this result supports the assumption of sample independence, we acknowledge that spatial block cross-validation32(e.g., via ENMeval) would provide a more rigorous approach and should be considered in future studies with larger sample sizes. Correlation analysis of environmental variables To minimize multicollinearity and overfitting in the MaxEnt model37, environmental variables of the same type were initially subjected to Pearson correlation analysis with a clear threshold set at |r| > 0.7 (Supplementary Fig. S4). Variables showing high correlation were further evaluated using variance inflation factor (VIF) analysis, and only those with VIF < 5 were retained to ensure statistical independence while maintaining biological relevance for Saunders’s Gull reproduction. And SG, WS, DO, SS, and T were excluded. Consequently, 24 independent environmental variables, including 7 vegetation conditions, 4 soil physicochemical properties, 9 water source physicochemical properties, 3 meteorological factors, and 1 topographic factor, were selected for model construction. Among the excluded variables, four merit ecological note. WS and SG were highly correlated with WEC (|r| = 0.89 and 0.91, respectively), and their exclusion does not imply that salinity is unimportant but rather that its variation is adequately captured by WEC and SEC. DO and WT were substantially correlated (|r| = 0.76), and temperature was retained given its stronger mechanistic link to breeding phenology. WpH and SpH were moderately correlated (|r| = 0.62) and both were retained as they reflect distinct chemical environments (water column vs. substrate). The excluded variables’ ecological roles are subsumed by the retained ones and their exclusion is statistically necessary to avoid multicollinearity. Although Pearson correlation (|r| > 0.7) and VIF (< 5) screening reduced the variable set from 29 to 24, the ratio of sampling points (n = 53) to predictors (p = 24) remains approximately 2.2:1, which is below the commonly recommended minimum of 10:1 for species distribution models38. A low sample-to-variable ratio may increase the risk of model overfitting despite cross-validation. To mitigate this concern, we (1) applied k-fold cross-validation (k = 5) to evaluate model performance on independent subsets, (2) relied primarily on the Jackknife test to identify the top contributing variables rather than interpreting all 24 response curves, and (3) acknowledge that model performance metrics from small samples should be interpreted as conservative estimates. We further note that the dominant four variables (elevation, wind speed, soil pH, soil electrical conductivity), which together contribute 70.9% of the model, are robustly estimated relative to the sample size (53/4 ≈ 13:1). Further variable reduction was considered but would risk excluding ecologically meaningful predictors given that the remaining 20 variables, though individually less influential, collectively represent the full range of habitat dimensions (vegetation, water, soil, meteorology, topography) relevant to breeding site selection.” Results Model accuracy verification The five most common model evaluation indicators of the species distribution model are: sensitivity, AUC area, overall accuracy, TSS and specificity39. The AUC value was used to evaluate the MaxEnt model, and the test method is widely used in species distribution models and can be used to evaluate the performance of the model under different threshold conditions40. In addition to AUC, this study included additional evaluation metrics such as TSS and Kappa, which together can provide a more comprehensive assessment of model accuracy. During the model simulation process, the operating environment was configured as follows: 75% of the dataset was used for training and 25% for testing. An AUC of 0.8–0.9 indicates adequate predictive performance, while 0.9–1.0 indicates excellent performance, and a TSS above 0.6 and a Kappa above 0.6 are generally considered to reflect good model accuracy and reliability similarly41,42. In this study, model performance was evaluated using k-fold cross-validation (k = 5). The mean AUC on the test set was 0.805 (95% CI: 0.800–0.810), TSS was 0.614 (95% CI: 0.609–0.619), and Kappa was 0.614 (95% CI: 0.611–0.617), indicating that the model was both accurate and stable in classifying breeding habitat suitability. The three evaluation metrics provide complementary information: AUC (0.805) measures overall discriminatory ability independent of any threshold; TSS (0.614) evaluates classification accuracy at the optimal threshold accounting for both sensitivity and specificity; and Kappa (0.614) measures agreement beyond chance. The moderate TSS and Kappa values (both ~ 0.6) relative to the adequate AUC (0.805) suggest that while the model discriminates well between suitable and unsuitable habitat overall, its binary classification at any single threshold is more conservative. This pattern is expected for a species with a spatially restricted and ecologically specialized breeding habitat: AUC captures the model’s ability to rank habitat quality across the entire landscape, while the lower TSS and Kappa reflect the inherent difficulty of binary classification when optimal habitat is rare (2.05%). Environment analysis of Saunders’s Gull breeding ground Land use type The land use types of the Nanxiaohe breeding ground are divided into water surface, mudflats, bare land, and wet grassland, as shown in Supplementary Fig. S5. Mudflats with an area of 1.09 km2, accounted for 27.91% of the total area, providing for the survival of Helice tridens tientsinensis, sandworms, shells and snails, which are the main food sources of Saunders’s Gull. Mudflats also support the gull’s activity and predation. Bare land is 0.25 km2, accounting for 6.35%, which primarily consists of roads and monitoring stations made of relatively dry land. The remaining 25.57% is made up of water surface, which is 1.00 km2, with most in shallow pits at pond connections and other tiny water surfaces on the mudflats. The Bohai Sea in the south serves as the primary water source and supplies a humid climate for birds to construct nests. Wet grassland is 1.57 km2, accounting for 40.17% of this area. Wetland vegetation can provide birds with a relatively hidden and safe space, and plant branches and leaves can also be used as nesting materials. Environmental factor range The vegetation, physical and chemical properties of water and soil, meteorology Index and topography in Nanxiaohe Saunders’ Gull breeding ground are shown in Supplementary Fig. S6 to S10. In all, it is nearly completely covered by S. salsa (coverage rate of about 50%), except for the water area. The plants in ponds No. 3 and No. 4 are significantly higher than others, with an overall height between 2.62 and 22.43 cm, with a mean of 10.16 ± 4.61 cm (95% CI: 8.98–11.34 cm). The three colors are distributed fairly evenly (they represent the maturity of S. salsa). The water was generally alkaline, with a mean WpH of 9.12 ± 0.48 (95% CI: 8.85–9.39) and a mean WT of 27.95 ± 1.38 °C (95% CI: 27.13–28.73 °C) and showed little variation, indicating that these conditions are favorable for the reproduction and growth of aquatic organisms. DO levels were high, with a mean of 7.30 ± 0.72 mg/L (95% CI: 7.22–8.02 mg/L), thereby supporting the survival of small fish, shrimps and shellfish, which are valuable food sources. TDS averaged 11.07 ± 5.78 mg/L (95% CI: 7.43–13.83 mg/L), and COD averaged 5.37 ± 0.79 mg/L (95% CI: 5.23–6.10 mg/L), indicating that the concentration of organic pollutants in the waters is low, with a mean ORP of −322.49 ± 141.93 MV (95% CI: −400.9 to −244.1 MV). WEC averaged 21,826.37 ± 11,009.30 µS/cm (95% CI: 14,606.32-26,799.82 µS/cm), indicating relatively high salinity (the maximum values reached 2.47% of WS and 41,977.80 µS/cm of EC at the junction of ponds 2 and 3, where seawater flows in), with a mean WS of 1.27 ± 0.74% (95% CI: 0.79–1.99%) and Cl of 18.50 ± 13.66‰ (95% CI: 10.94–26.06‰). WR averaged 0.473 ± 0.194 MΩ (95% CI: 0.330–0.544 MΩ), indicating relatively low water purity. SG was 1.008 ± 0.006 (95% CI: 1.005–1.010), and NSF remained at 3, suggesting relatively high nutrient availability, with no pronounced variation observed across the study area. ST averaged 27.95 ± 1.54 °C (95% CI: 27.43–28.22 °C). In most areas of Nanxiaohe breeding ground, the SpH was around 7.0, with a mean of 6.62 ± 0.75 (95% CI: 6.08–7.16), while a small portion of the area was acidic. Since the coastal wetland area is close to the ocean, the seawater contains a large amount of salt and other electrolytes that will dissolve into the soil, resulting in a high SEC with a mean of 4714.15 ± 1288.64 µS/cm (95% CI: 4398.84–5315.03 µS/cm), and SS is relatively high, with a mean of 2570.65 ± 855.21 mg/L (95% CI: 2483.79–3973.07 mg/L). the survey period is in the rainy season, and most of SM is 100%, with a mean of 97.58 ± 12.39% (95% CI: 92.95–100.00%), but some areas with higher terrain have low water content due to rapid evaporation, such as the embankment in the west of the No. 2 pond, with a water content of only 26.86–36.79%, and it was approximately 89.52% in the transitional zone. The mean Temp in this season is 26.9 ± 1.2 °C (95% CI: 26.16–27.64 °C) within a relatively stable range. High wind speed is a significant meteorological feature of the area, with a mean WS of 15.07 ± 1.96 km/h (95% CI: 12.19–17.95 km/h). RH is relatively high, with a mean of 74.90 ± 4.9% (95% CI: 71.87–77.93%), which is mainly due to the unique characteristics of the geographical location and the influence of the marine climate. Due to the constant humidity and frequent haze, the sunshine intensity in this area is weak, and the number of sunny hours in this season is relatively low, with a mean of 10594.99 ± 2765 lx (95% CI: 7830.0–13360.0 lx). Because of its location in a swamp area and the intertidal zone formed by seawater invasion, the terrain is relatively flat and the altitude is relatively low, with an average altitude of 10.99 ± 3.08 m (95%CI:10.20–11.78 m). However, the terrain of Nanxiaohe breeding ground is also affected by seawater erosion and siltation, resulting in different altitudes in different areas. Contribution rate of environmental factors and division of breeding suitability area in Saunders’s Gull breeding ground Breeding suitability area division and area ratio calculation Habitat suitability was classified into three levels using the maximum training sensitivity plus specificity (MTSS) threshold approach43. This method was chosen over alternatives (equal training sensitivity and specificity, minimum distance to ROC corner) because it balances omission and commission errors and is widely recommended for presence-only data44. The MTSS threshold was calculated independently for each of the 10 replicate runs, and the mean threshold value was applied. Pixels with logistic output values below the MTSS threshold were classified as low suitability; values between the MTSS threshold and the maximum sensitivity-specificity equality threshold were classified as medium suitability; and values above the latter were classified as high suitability. The resulting classification (low: 90.56%, medium: 7.39%, high: 2.05%) reflects the conservative nature of the threshold approach and the limited extent of optimal habitat within the study area. The study area is divided into low, medium and high breeding suitability areas, of which the low suitability area accounts for 90.56% and the area is 3.5385 km2, the medium suitability area accounts for 7.39% and the area is 0.2887 km2, and the high suitability area accounts for 2.05% and the area is 0.0802 km2. Although the high suitability area is very limited, it represents the most critical locations for Saunders’s Gull breeding activities. The limited extent of high-suitability habitat has direct implications for population carrying capacity. Considering the observed nest density in high-suitability areas (approximately 1 nest per 200–300 m2 based on nearest-neighbor distances of ~ 5 m), and assuming similar habitat occupancy patterns, this area could theoretically support approximately 270–400 breeding pairs. This estimate represents a potential habitat capacity approximation rather than a validated population carrying capacity. With the recorded breeding population of ~ 11,088 individuals (approximately 5,544 pairs; LRENNR Saunders’s Gull monitoring station in 2022), this suggests that high-suitability habitat within the Nanxiaohe breeding ground may be approaching its carrying capacity, potentially contributing to density-dependent competition and sub-adult mortality. This finding underscores the urgency of habitat expansion through artificial breeding islands and adjacent restoration areas. Overall, due to the protected status of the area, the entire Nanxiaohe breeding ground is suitable for Saunders’s Gull to breed or survive, and combined with ArcGIS layer processing, the most suitable areas for survival and breeding are mainly located in Pond No. 1, Pond No. 3 and Pond No. 4, as shown in Fig. 2. Fig. 2Breeding habitat suitability classification for Saunders’s Gull at the Nanxiaohe breeding ground, Liao River Estuary National Nature Reserve (LRENNR), China. Suitability was classified into three levels using the maximum training sensitivity plus specificity (MTSS) threshold averaged across 10 MaxEnt replicate runs: low suitability (green, 90.56%, 3.5385 km2), medium suitability (yellow, 7.39%, 0.2887 km2), and high suitability (red, 2.05%, 0.0802 km2). The high-suitability areas are concentrated in Ponds 1, 3, and 4. The prediction results were visualized and mapped in ArcGIS 10.7 (Esri, Redlands, CA, USA; https://www.esri.com). The study area boundary was delineated from publicly available Landsat 8 OLI/TIRS imagery obtained from the Geospatial Data Cloud (https://www.gscloud.cn/search). Breeding habitat suitability classification for Saunders’s Gull at the Nanxiaohe breeding ground, Liao River Estuary National Nature Reserve (LRENNR), China. Suitability was classified into three levels using the maximum training sensitivity plus specificity (MTSS) threshold averaged across 10 MaxEnt replicate runs: low suitability (green, 90.56%, 3.5385 km2), medium suitability (yellow, 7.39%, 0.2887 km2), and high suitability (red, 2.05%, 0.0802 km2). The high-suitability areas are concentrated in Ponds 1, 3, and 4. The prediction results were visualized and mapped in ArcGIS 10.7 (Esri, Redlands, CA, USA; https://www.esri.com). The study area boundary was delineated from publicly available Landsat 8 OLI/TIRS imagery obtained from the Geospatial Data Cloud (https://www.gscloud.cn/search). Environmental factor contribution rate and response curve The Jackknife test is used to determine the weight of different environmental variables in contributing to the prediction of the suitable breeding area for Saunders’s Gull. Through MaxEnt simulation, the following table shows the 11 environmental variables that contribute more than 1% to the simulation results, namely, the influence on the reproduction of Saunders’s Gull from large to small are: EL, WS, SpH, SEC, VH, ST, ORP, WpH, WR, WT and Cl. The total contribution rate of these 11 environmental factors is 94.10%, among which the total contribution rate of EL, WS, SpH and SEC is 70.90%, which is the main factor affecting the survival and reproduction of Saunders’s Gull, as shown in Supplementary Fig. S11. Variable contribution values reported are from the final tuned model and may exhibit variability across replicate runs. This study identifies the four environmental factors with the highest contribution rates in the study area, and simulates the probability-of-occurrence function curve for Saunders’s Gull. It is generally believed that when the predicted occurrence probability exceeds 0.5, the corresponding environmental range represents conditions associated with relatively higher predicted breeding suitability in the model45. Response curves represent statistical associations between environmental variables and predicted breeding occurrence probability. They should be interpreted as correlative patterns rather than direct evidence of ecological preference or causation. Elevation Elevation Elevation As the altitude increases, the probability of Saunders’s Gull breeding first decreases and then increases. The probability is lowest at about 12.70 m, and at 13.80 m, the probability begins to be greater than 0.5, then rises and reaches the highest at 16.97 m. This indicates that the predicted breeding occurrence probability was highest within the elevation range of 13.80 to 16.97 m in the MaxEnt model, as shown in Fig. 3 (A). (2)Wind speed (2)Wind speed Wind speed When the wind speed is between 10.90 and 16.00 km/h, the probability of Saunders’s Gull breeding is the lowest. When the wind speed is between 16.00 and 19.00 km/h, the probability increases, but is still less than 0.5. When the wind speed reaches 19.50 km/h, the probability is greater than 0.5, and the probability reaches a peak at 23.78 km/h. Higher predicted breeding occurrence probability was associated with wind speeds ranging from 19.50 to 23.78 km/h, as shown in Fig. 3 (B). (3)Soil pH (3)Soil pH Soil pH When the soil pH was 5.60, the probability of Saunders’s Gull reproduction was the lowest, the probability was 0.5 at 6.65, and the probability reached its peak at 7.50, indicating that the predicted probability of breeding occurrence increased under relatively neutral to alkaline soil conditions (pH 6.65 − 7.50), as shown in Fig. 3 (C). (4)Soil electrical conductivity (4)Soil electrical conductivity Soil electrical conductivity The soil electrical conductivity of the breeding probability of Saunders’s Gull began to gradually decrease at 320.4µS/cm, and the probability was the lowest at around 3,714µS/cm. It was greater than 0.5 at 7,821µS/cm and finally reached a peak at 9,107.9µS/cm. Electrical conductivity represents salt content, indicating that higher predicted breeding occurrence probability was associated with relatively elevated soil electrical conductivity (> 7,821 µS/cm), as shown in Fig. 3 (D). Areas characterized by these environmental conditions showed higher predicted habitat suitability in the MaxEnt model. At this time, Saunders’s Gull showed a higher predicted probability of breeding occurrence under these modeled conditions, which is conducive to the stability and development of Saunders’s Gull population. The response curves are presented as indicative statistical associations, and while bootstrapped confidence intervals were not applied, they align with known ecological patterns and provide useful insights for conservation planning. Fig. 3Response curves of the four dominant environmental variables from MaxEnt modeling (EL, WS, SpH, and SEC; cumulative contribution rate: 70.90%). (A) Elevation (m), (B) Wind speed (km/h), (C) Soil pH, (D) Soil electrical conductivity (µS/cm). Response curves are derived from the final tuned MaxEnt model; variability across the five cross-validation folds is not shown but was qualitatively consistent across folds. The dashed horizontal line at y = 0.5 indicates the threshold used for interpreting higher predicted suitability in the model. X-axis ranges are clipped to the 5th–95th percentile of observed values. Response curves of the four dominant environmental variables from MaxEnt modeling (EL, WS, SpH, and SEC; cumulative contribution rate: 70.90%). (A) Elevation (m), (B) Wind speed (km/h), (C) Soil pH, (D) Soil electrical conductivity (µS/cm). Response curves are derived from the final tuned MaxEnt model; variability across the five cross-validation folds is not shown but was qualitatively consistent across folds. The dashed horizontal line at y = 0.5 indicates the threshold used for interpreting higher predicted suitability in the model. X-axis ranges are clipped to the 5th–95th percentile of observed values. Environment characteristics under different reproductive behaviors of Saunders’s Gull The multiscale environmental characteristics corresponding to different reproductive behaviors are clarified, which will guide the protection of Saunders’s Gull and the restoration of their habitats46. Mate selection and mating Mate selection and mating Mate selection and mating Saunders’s Gull usually selects and mates near low vegetation on intertidal beaches, or in mudflat gullies with undulating terrain, as shown in Supplementary Fig. S12 (A). During this breeding period, Saunders’s Gulls are more vigilant and choose areas with more hidden structures (e.g. within a S. salsa community). At this time, the environmental characteristics are reflected in the wetland vegetation structure. (2)Nesting and raising chicks (2)Nesting and raising chicks Nesting and raising chicks Saunders’s Gull nests are shallow, disc-shaped ground nests with an average size of 216 mm × 208 mm, a height of 35 mm, and a depth of 20 mm. Nest material composition was systematically recorded for all 53 nests identified across the four ponds. P. australis reed leaves constituted the primary nesting material across all ponds (mean 72 ± 14% of nest dry mass), followed by S. salsa branches (mean 21 ± 11%). The remaining material (< 7%) consisted of other herbaceous wetland plants. Pond-level variation was observed: nests in Pond 1 contained a higher proportion of P. australis (84 ± 8%) compared to Pond 4 (58 ± 15%), where S. salsa branches were more prevalent (35 ± 12%), suggesting flexibility in nest material use depending on local vegetation availability. Nests are grouped closely, sometimes only 5 m apart. In poor habitats, simple pits with messy nesting materials are used. Nests are built within 20 cm high porous vegetation for safety and near water sources for drinking. In this stage, suitable habitats feature complete, interspersed environment variables and appropriate structures to support survival and reproduction, as shown in Supplementary Fig. S12 (B) and (C). (3)Foraging and flight (3)Foraging and flight Foraging and flight The flight process of chicks is closely related to foraging behavior. The main food of Saunders’s Gull is sandworms, crabs, fish and shrimps, all of which grow in mudflats or places with abundant water sources, as shown in Supplementary Fig. S12 (D). The first stage of the flight process of chicks is to move around near the nest; the second stage is to leave the nest at a certain distance from the nest; the third stage is to move around in a farther place, learning to forage and practice flying, as shown in Supplementary Fig. S12 (E). These behaviors are more related to environmental variables. Foraging is mainly near mudflats and water sources, while the flight is mainly in the sky. Restoration and management of Saunders’s Gull breeding grounds Construct an artificial breeding island Construct an artificial breeding island Construct an artificial breeding island The breeding population of Saunders’s Gulls around Nanxiaohe has reached 11,088 due to the strengthening of ecological and environmental protection in recent years (from the ecological monitoring station in 2022). However, a high mortality rate of sub-adult birds still exists, which is most likely caused by competition and predation in the relatively concentrated breeding area. The construction of artificial islands (Fig. 4) could serve as a potential habitat enhancement strategy by increasing available breeding space, reducing local competition pressure, and providing additional suitable sites for Saunders’s Gulls breeding. Fig. 4Schematic Diagram of Restoration and Management of Saunders’s Gull Breeding Ground. Schematic Diagram of Restoration and Management of Saunders’s Gull Breeding Ground. (2)Planting native species (2)Planting native species Planting native species Delta Reservoir and Shilixi have been the abandoned areas of Saunders’s Gull breeding, where S. salsa, S. glauca and reeds are the main vegetation. However, in the 1990 s, in addition to the Suaeda community, there were also Nitraria tangutorum Bobrov and Tamarix chinensis Lour in the area47. Areas dominated by these native plant species may provide suitable nesting substrates for Saunders’s Gull. The wetland native species can be artificially planted to restore and increase the breeding area, as shown in Fig. 4. In the restoration process of the breeding area, the N. tangutica and S. salsa community and the T. chinensis and S. salsa community would be configured in the Delta Reservoir and in the tidal ditch area near the existing reed beach in Shilixi, respectively. (3)Habitat management (3)Habitat management Habitat management The monitoring and management of Saunders’s Gull breeding grounds should be strengthened promptly to ensure the healthy growth of wetland vegetation and maintain the sustainable development of habitat functions. The variables and structures in the new islands should strictly refer to the environmental characteristics of strong breeding suitability inside Nanxiaohe breeding ground to provide a relatively stable breeding environment for Saunders’s Gull (Fig. 3). The overall height of the vegetation should be kept at around 20 cm, which is suitable for the habitat selection of Saunders’s Gull. The water depth should be promptly adjusted according to the growth needs of plants to maintain the stability of the vegetation community. (4)Core area protection (4)Core area protection Core area protection Building on our key finding that only 2.05% of the study area is highly suitable, future conservation strategies will focus on targeted habitat enhancement, such as reinforcing protection measures in Ponds 1, 3, and 4, restoring native vegetation, and minimizing human disturbance to ensure the persistence of Saunders’s Gull breeding sites. Discussions Saunders’s Gull habitat assessment Habitat assessment is a method of evaluating the resources and conditions required for species survival and reproduction in the natural environment and is of great significance to the assessment of ecosystem value and the designation of priority protection areas48. This study uses species habitat assessment to provide more accurate assessment ideas in the calculation of biodiversity value services. The soil temperature in the Nanxiaohe breeding ground was the same as the water source temperature during the assessment. This is due to the unique geographical conditions and long-term precipitation, which have kept soil water content close to saturation, leading to relatively stable soil temperature. The soil in some areas of Nanxiaohe breeding ground is acidic as opposed to the coastal soil, which is alkaline49. This pattern may be associated with localized inputs of organic matter. One possible contributor is the decomposition of bird carcasses during the late breeding period of Saunders’s Gull, when chick mortality and predator activity increase50. Previous studies have indicated that the decomposition of animal remains can alter soil chemical properties, including pH, through microbial processes, although such effects are typically spatially limited and context-dependent51. However, the magnitude and spatial scale of this effect at Nanxiaohe breeding ground remain uncertain, as wetland soil pH is primarily governed by hydrology, vegetation type, and sediment chemistry52. Alternative explanations, including anaerobic microbial activity in waterlogged soils producing organic acids from decomposing plant roots53, and localized differences in sediment composition, cannot be ruled out. We therefore present this as a hypothesis requiring further investigation rather than a confirmed mechanism. The status of this area as the largest known breeding ground of Saunders’s Gull is supported by long-term monitoring and previous survey reports22,34,35, rather than being solely attributable to the environmental variables examined in this study. Therefore, the identified environmental conditions should be interpreted as potentially favorable factors associated with breeding habitat selection, rather than direct causal determinants. Through 10 simulations of the MaxEnt, this study identified the four environmental factors and their range values that have a greater impact on the reproduction of Saunders’s Gull and divided the breeding-suitable areas of different levels in the Saunders’s Gull breeding grounds. MaxEnt is often used for species simulation in large-scale research areas47,54. The research area of this paper is small and cannot be simulated using open-source data. Data can only be collected through field surveys. Although the number of presence records is limited, we focused on ecologically meaningful predictors, and spatial autocorrelation was minimal, suggesting that the model results remain robust. This can provide more accuracy and obtain the required environmental data by combining the breeding habits of Saunders’s Gull flexibly, solving the limitations of data acquisition for species distribution simulation in small-scale plots. Comparison of environmental characteristics of habitats in the breeding areas of Saunders’s Gull The LRENNR is the northernmost limit of the Saunders’s Gull breeding ground in China and is where the Saunders’s Gull breeds the most55,56. But it is not the only breeding ground. There are also Saunders’s Gull breeding grounds in the Yellow River Delta National Nature Reserve in Shandong Province (YRD), the Yancheng National Rare Bird Nature Reserve in Jiangsu Province (YNRBNR), and the Yalu River Estuary Coastal Wetland National Nature Reserve in Liaoning (YRE)20,34,57. Compared with Saunders’s Gull breeding ground at the same latitude, the YRE has lost the reed community due to improper irrigation, and the original nesting area has been flooded58,59. Vegetation is significant for the reproduction of Saunders’s Gull. Although the Spartina in the YRD and YNRBNR is also suitable for Saunders’s Gull, the area where S. salsa is the dominant population, is still favored22,60–62. Although YRD and YNRBNR have warmer climates, more abundant rainfall, and higher biodiversity, which are more suitable for the reproduction of Saunders’s Gull, also the number of natural enemies such as raptors and small carnivores has increased. This may be related to the higher number of Saunders’s Gulls in LRENNR. The breeding season of Saunders’s Gull is mainly from March to July each year. The climate conditions in LRENNR are more suitable for the reproduction of Saunders’s Gull63. Therefore, LRENNR has become the largest breeding ground compared to other protected areas and its environmental characteristics are mainly reflected in the variables and structure, such as land and climate, especially the wetland vegetation. Regarding the environmental ranges associated with higher predicted habitat suitability identified in this study (elevation 13.8–17.0 m, wind speed 19.5–23.8 km/h, soil pH 6.65–7.5, and soil conductivity > 7,821 µS/cm), direct quantitative comparison with other breeding colonies is limited by the absence of comparable high-resolution environmental data. For example, the YNRBNR breeding grounds are at lower elevations (< 5 m) with lower soil salinity due to higher freshwater input, with reported nest densities of approximately 1.6 nests/ha in core breeding areas22, compared with ~ 5 nests/ha estimated in Nanxiaohe high-suitability areas, yet they also support breeding populations, suggesting that the ranges identified here should not be interpreted as universal ecological thresholds. We therefore interpret these environmental ranges as site-specific associations derived from the breeding ground dataset, which may reflect local habitat associations rather than species-wide physiological tolerances. Multi-site comparative studies with standardized environmental measurements are needed to establish generalizable thresholds for the species. Due to the influence of climate, sea level, topography and other factors, the Nanxiaohe breeding ground may also experience changes and evolution in the topography in the future. Inland erosion or seawater invasion may occur, which will have an important impact on the ecological environment and human society of coastal wetlands. The spatial changes in the Nanxiaohe breeding ground can be simulated through environment variables47. They should be comprehensively considered when restoring or planning the Saunders’s Gull protection belt. Strategies for the breeding grounds conservation of Saunders’s Gull It is the most crucial and essential method of protecting Saunders’ Gull by restoring the estuary wetlands and expanding the natural habitat area while maintaining the current breeding grounds62. Establishing artificial ecological floating islands would increase microhabitat availability and habitat heterogeneity64. The construction of islands could provide some co-benefits for Saunders’s Gull. The establishment of islands increases community composition and diversity of the macroinvertebrates, and thus influences the biodiversity of their predators, such as fish and waterbirds. There are large areas of oil fields and aquaculture sites that are under exploitation around the Nanxiaohe breeding ground65. Therefore, periodic water quality monitoring should be performed to avoid water quality problems directly affecting the production of aquatic organisms by periodical water monitoring, which ensures the stability of the food and water source of Saunders’s Gull. Establishing floating islands, expanding the breeding and rationally utilising the seawater resources in offshore areas would ensure a sufficient, safe and sustainable environment for Saunders’s Gull. The nesting sites of Saunders’s Gull were associated with specific ranges of soil pH and electrical conductivity, among which electrical conductivity was highly correlated with salinity (Supplementary Fig. S4); however, whether these soil properties directly influence nesting success or merely co-vary with other habitat features remains to be experimentally verified. Rather than direct soil chemical interventions, we recommend maintaining natural hydrological processes that regulate soil salinity and pH, including the preservation of tidal exchange and freshwater inputs. In areas where soil conditions have been degraded by anthropogenic activities (e.g., oil extraction, construction), passive restoration through the removal of disturbance sources and the re-establishment of native vegetation is preferred over active chemical amendments. Any active intervention should be preceded by pilot studies to evaluate ecological impacts and should comply with the management regulations of LRENNR as a national nature reserve. Given the uncertainty around the causal role of soil properties in nest-site selection, these management recommendations are presented as precautionary measures based on observed habitat associations, pending confirmation through controlled field studies. Wetland vegetation has a great impact on the breeding and survival of Saunders’s Gull, especially the plant height. Regularly monitor the vegetation community and appropriately intervene in the number and height of dominant species. S. salsa, S. glauca and P. australis are the main vegetation species in the area, which are relatively single, but the coverage is large. Preventing the invasion of alien species is considered essential for maintaining the stability of the native vegetation community. Protecting and planting native species can restore or promote vegetation renewal in the degraded or damaged areas. This study found that only 2.05% of the site was a high-suitability area. This small proportion of highly suitable habitat may lead to a spatial concentration of breeding activities, potentially increasing local density and competition. It also points out the importance of these core areas, which conservation and management efforts should prioritize. Limitations and future work Environmental variables are essential for modelling the habitat suitability of waterfowls. However, these environmental ranges associated with higher predicted suitability should be interpreted as site-specific patterns derived from the current dataset, rather than universal thresholds, and may vary under different ecological contexts or across other breeding colonies. Furthermore, the 20-day field survey period, while capturing the late breeding stage comprehensively, does not represent early-season breeding activities such as courtship and nest initiation. Future studies would benefit from multi-season surveys covering the entire breeding cycle. Additionally, the temporal offset between the 2016 satellite imagery and the 2022 field data may introduce minor uncertainties in land-use classification, although the protected status of LRENNR limits land-cover change during this period. Low or not discolored wetland herbs were classified into mudflats by the limited accuracy of satellite images, and the wetland grassland refers to a large area of S. salsa that has completely turned red. Regional hydrological conditions and bird breeding ground protection restrictions, the research data are not continuous and is greatly affected by the weather, resulting in abnormal assessment results of some environmental data. Monitoring equipment should be improved and monitoring of Saunders’s Gull should be strengthened to obtain more accurate data. The habitat requirements of migratory shorebirds vary depending on whether they are migrating or breeding. The records reflect the presence of Saunders’s Gull at that location with the behavioral states of foraging and resting. The suitable area modelled in this study is a probability description of the breeding habitat quality of Saunders’s Gull. In the future, we will focus the research on migration with satellite tracking66 and field survey data to analyze the environmental characteristics of Saunders’s Gull habitat in different life stages or behaviors. Moreover, the limited sample size (n = 53) relative to the number of predictors (p = 24) may have contributed to model overfitting despite the use of cross-validation and regularization; this ratio may also compromise model stability, and therefore, ecological interpretations based on individual variables should be considered exploratory and require validation using larger spatial and temporal datasets; uncertainty related to small sample size and model transferability will be further addressed in future studies as more continuous and multi-site data become available. Although Moran’s I did not indicate significant spatial autocorrelation in model residuals, this test has limited power with small sample sizes. Future studies should employ spatial block cross-validation to explicitly account for potential residual spatial structure. Although model performance uncertainty was quantified using confidence intervals, uncertainties in variable contributions and response curves were not directly estimated; Jackknife results and response curves are presented as single-run outputs without variability metrics, which represents a limitation of the current study. Future studies should address this using bootstrap resampling, replicate MaxEnt runs, or larger multi-year datasets. Conclusions This study demonstrates that only 2.05% (0.0802 km2) of the Nanxiaohe breeding ground provides high-suitability habitat for the globally endangered Saunders’s Gull, with elevation, wind speed, soil pH, and soil electrical conductivity collectively explaining 70.9% of breeding habitat suitability. The estimated carrying capacity of this limited high-suitability area (270–400 breeding pairs) suggests that the current population (~ 11,088 individuals) may be approaching habitat saturation, highlighting the potential value of expanding protected breeding habitat through measures such as artificial islands and native vegetation restoration. We recommend that future conservation planning in LRENNR prioritize the identified core areas (Ponds 1, 3, and 4) for immediate protection and that adjacent degraded wetlands (Delta Reservoir, Shilixi) be restored as satellite breeding sites. Standardized multi-site environmental monitoring is needed to determine whether the environmental associations identified here are site-specific or generalizable to other Saunders’s Gull colonies. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 7 Supplementary Material 7 Supplementary Material 8 Supplementary Material 8 Supplementary Material 9 Supplementary Material 9 Supplementary Material 10 Supplementary Material 10 Supplementary Material 11 Supplementary Material 11 Supplementary Material 12 Supplementary Material 12 Supplementary Material 13 Supplementary Material 13 Supplementary Material 14 Supplementary Material 14 Supplementary Material 15 Supplementary Material 15 Supplementary Material 1 Supplementary Material 1 Supplementary Material 2 Supplementary Material 2 Supplementary Material 3 Supplementary Material 3 Supplementary Material 4 Supplementary Material 4 Supplementary Material 5 Supplementary Material 5 Supplementary Material 6 Supplementary Material 6