Distribution patterns of Caspian seal island haul-outs under Caspian Sea regression

Abstract

The Caspian seal (Pusa caspica Gmelin, 1788) is the sole marine mammal and an indicator of ecosystem health in the Caspian Sea. This allopatric endemic species is listed as Endangered. Aerial surveys conducted in 2020, 2022, and 2023 showed that a shallow, inaccessible area in the northeastern Caspian Sea served as the primary seal concentration site. During spring, up to 98% of all molting Caspian seals hauled out on these island haul-out sites. To determine the distribution patterns of seal haul-outs, we conducted a comprehensive analysis of geospatial data from Sentinel-2 and Landsat-8 satellite imagery across a designated large study area (130×250 km) divided latitudinally into 16 transects. A 2.28 m drop in the North Caspian Sea level from 1995 to 2023 shifted the coastline westward by an average of 44.1 to 163.7 km. The maximum shift reached 175.4 km due to the drying of the Kaydak and Komsomolets bays. Island area decreased more than 15-fold during this period. We found that island evolution under sea-level regression follows a triad: emergence, growth, and merger with the mainland. An ecosystem shift is occurring: the shallow-water ecosystem is transforming into a mainland ecosystem. The rate of coastline shift has recently exceeded 23 km per year. Combined with increasing shelf depths, this accelerates the triad cycle and drives the disappearance of islands in the northeastern part of the sea. Because islands serve as primary seal haul-out sites, their disappearance forces seals to relocate to newly emerging shallow areas characterized by active shipping, fishing, and planned oil and gas development. This inevitably exacerbates the conflict of interests between industrial shelf development and the conservation objectives for this endemic species. Conserving Caspian seal habitats requires establishing ecological corridors along migration routes between existing marine protected areas, alongside rapid and adaptable protection of identified marine mammal haul-out sites. Because the westward coastline shift has caused mass haul-out sites to extend beyond the territorial waters of the Republic of Kazakhstan, coordination among Caspian littoral states is necessary to grant protective status to these waters, including their designation as an Important Marine Mammal Area (IMMA), and to establish a transboundary nature reserve.

1 Introduction

The Caspian seal (Pusa caspica Gmelin, 1788) is the sole marine mammal in the Caspian Sea. The species is monotypic, comprising a single population, and is an allopatric endemic to the Caspian Sea. In 2008, the IUCN Red List classified the species as Endangered (Goodman and Dmitrieva, 2016). In 2020, the Republic of Kazakhstan added the Caspian seal to its List of Rare and Endangered Animal Species (Baimukanov et al., 2023). The species is similarly listed in all Caspian littoral states (Bizikov et al., 2021; Rustamov et al., 2021; Eybatov and Gadzhiev, 2022.).

The Caspian Sea, situated at the crossroads of Europe and Asia, is the largest lake on Earth and has no connection to the world’s oceans. It is referred to as a sea because of its immense size and the oceanic type of the Earth’s crust beneath it (Poletaeva and Poletaev, 2015). A remarkable feature of the Caspian is the instability of its water level. These sea-level fluctuations are determined by a combination of geological, climatic, and anthropogenic factors. Although the rise and fall of the sea level is described as a cyclical process, the overall trend during the Holocene over the past 6,000 years has been downward (Rychagov, 2011).

Over the past century, the decline in sea level has been linked to climate warming and decreased river inflow, mainly caused by the consumptive use of water resources. Predictions indicate that this downward trend will continue until the end of the 21st century (Prange et al., 2020), potentially causing serious negative impacts on the economy and biodiversity of the Caspian basin (Rychagov, 2011; Prange et al., 2020; Court et al., 2025).

The Caspian Sea is conventionally divided into three parts: the North, Middle, and South Caspian (Figure 1). The North Caspian features shallow waters, with depths of less than 5 meters covering about 70% of its area (Water Balance and Fluctuations of the Caspian Sea Level. Modeling and Forecast, 2016). The ongoing sea regression most severely impacts this region, especially its eastern side, which is part of the Kazakhstan sector of the North Caspian. In this study, the Kazakhstan sector is defined according to the Agreement between the Republic of Kazakhstan and the Russian Federation, which establishes the seabed boundary between the two countries along the median line.

Figure 1

The North Caspian serves as a vital habitat for the Caspian seal. During winter, seals gather on the ice of the North Caspian to breed, while in the spring molting season, they haul out on coastal islands and shalygas (a local Caspian term for small seabed elevations that become exposed during offshore wind surges and submerge again during onshore wind setups) in the shallow northeastern part of the sea. In autumn, after their foraging and migrations, they form pre-winter aggregations in this same area (Bizikov et al., 2021; Baimukanov et al., 2022, 2023).

The increasing frequency of warm winters and the reduction in ice cover mainly affect the seal breeding aggregations, which form from late January to late February (Naurozbaeva, 2025). Therefore, the current study does not investigate the impact of sea regression on winter haul-out sites.

Historical records document a shift in island haul-outs during a period of sea-level decline in the 1930s and 1940s. At that time, commercial hunting targeted the species, and researchers noted this spatial shift with the pragmatic goal of harvesting animals on the newly forming islands (Badamshin, 1950).

Currently, considering the catastrophic state of the Caspian seal population and the ongoing sea-level regression, we must more thoroughly investigate the loss of established haul-outs and assess the species’ ability to adapt to unstable conditions. Remote sensing methods offer access to satellite imagery, while GIS technologies enable us to calculate and analyze the rate of ongoing changes in the Caspian’s configuration (Vos et al., 2019). This enables us to determine the distribution patterns of seals during their time on island haul-outs under conditions of continuous seabed exposure and coastline retreat. These insights are essential for developing dynamic strategies and tactics to conserve Caspian seal habitats.

Studies on seal distribution conducted since 2008 indicate that the shallow waters of the northeastern Caspian have maintained suitable conditions for seal haul-outs during spring and autumn (Dmitrieva et al., 2016; Baimukanov et al., 2020). Aerial surveys over island haul-outs in the northeastern part of the sea during the first half of March in 2020 and 2022 confirmed that approximately 98% of all seals hauling out across the entire Caspian during this period were concentrated specifically in this region (Baimukanov et al., 2023).

The present study aims to examine the distribution patterns of island seal haul-outs in relation to sea regression within the Kazakhstan sector of the North Caspian. This information will be significant for predicting future aggregation sites and applying a dynamic approach to the conservation of key habitats.

To achieve this goal, we addressed the following objectives:

• Assess the relationship between long-term sea-level dynamics and coastline changes;

• Evaluate changes in the mainland and island areas in the northeastern Caspian Sea;

• Analyze the distribution of island Caspian seal haul-outs in response to sea regression.

2 Materials and methods

2.1 Assessment of long-term sea-level dynamics

We obtained average monthly sea-level data for the period from 1995 to 2023 from the National Hydrometeorological Service of the Republic of Kazakhstan, “Kazhydromet.” These data were collected from the Peshnoy and Kulaly stations, located in the north and south of the Kazakhstan sector of the North Caspian, respectively (Figure 1). We calculated the average annual sea-level indicators based on the data from these stations. All calculations were performed in Microsoft Excel.

2.2 Assessment of ice conditions

To analyze and assess the ice conditions before the aerial seal surveys, we analyzed Landsat 8–9 and Sentinel-2 satellite imagery (Sentinel-Hub, 2024).

2.3 Determination of the study area

Using QGIS software, we outlined a study area based on the Gauss-Kruger rectangular coordinate system, which makes it easier to perform measurements in meters or kilometers (Berlyant, 2002; Deakin et al., 2010). We then used this area to calculate coastline shifts, changes in mainland and island surface areas, and seal distribution in the northeastern Caspian. We set the northern, southern, and western boundaries of the area based on spring and autumn seal haul-out data collected during aerial surveys from 2009 to 2023. We drew these boundaries at a distance of 10 km from the furthest points of seal distribution. The eastern boundary follows the meridian, located 5 km from the extreme southeastern point of the 1995 coastline (Supplementary Figure 1).

We marked parallels with a width of 7,500 m in QGIS using the “Grid” tool in the View > Decorations menu. Since the distance between the northern and southern boundaries is 130 km, this created 16 transects. The distance from the northern boundary of the first transect (segment 1) to the corresponding northern boundary of the study area was 4,500 m; from the southern boundary of the 16th transect (segment 17) to the corresponding southern boundary of the study area was 5,500 m. To determine the coordinates of the endpoints of the transects, we used the “Extract Vertices” tool in QGIS (Supplementary Figure 1). The coordinates of the endpoints of the study area and the western and eastern boundary points of the segments (transect boundaries) are shown in Supplementary Tables 1 and 2.

2.4 Satellite imagery processing methodology

To minimize the impact of cloud cover, shadows, and atmospheric interference on spatial analysis quality, we applied the median composite method (Roberts et al., 2017). Archival imagery from Landsat-5 (TM), Landsat-8 (OLI), and Sentinel-2 (MSI) satellites (Sentinel-Hub, 2024) was used as source data, acquired between March 1 and May 31 (spring hydrological season) for 1995–2023, as well as the autumn of 2023 (September 1 to November 30). The years 2001–2007 and 2012 were excluded from the analysis due to the absence or critically defective archival data. The selection of satellite imagery depended on the year being analyzed. Landsat-5 images were used to produce median composites from 1995 to 2011. Landsat-8 and Sentinel-2 images were used from 2013 onward, selected based on the least cloud cover and absence of any distortions for median creation. Median composite synthesis was performed using the Google Earth Engine (GEE) cloud platform (Roberts et al., 2017), enabling automated cloud filtering by metadata (QA channels) and efficient processing of large raster datasets (Mao et al., 2021; Geographic median image compositing in Google Earth Engine, 2024; Hermosilla et al., 2024). This study employed remote sensing methods (Vos et al., 2019). Different code scripts were applied for processing images from each satellite [Supplementary Material 1 (Code 1, Code 2)]. Due to incomplete territorial coverage in 1998, 1999, 2013, 2014, and 2018, median composites covering both spring and summer were used for these years.

The GEE preprocessing algorithm included the following steps:

• Spatiotemporal filtering: limiting the sample to the region of interest (ROI) polygon and target calendar dates.

• Cloud and shadow masking: for Landsat imagery, we applied the built-in per-pixel CFMask algorithm (based on the QA_PIXEL channel); for Sentinel-2, we used the S2cloudless algorithm (based on the QA60 mask). This enabled automatic exclusion of pixels with a high probability of cloud or shadow presence prior to the median calculation step.

• Per-band median computation: for each pixel in the resulting image, we calculated the median spectral reflectance value from the entire cloud-cleared image stack for the season. This method ensures the robustness of the composite against random outliers and atmospheric haze.

To create a binary image of the water surface and extract coastlines and land/water polygons, we used the Normalized Difference Vegetation Index (NDVI) (Sun et al., 2012; Bishop-Taylor et al., 2019). Pixels with NDVI values below −0.1 were classified as water, producing a binary raster with values of 0 (land) and 1 (water). The use of a stricter threshold compared to the simple criterion NDVI < 0 was aimed at reducing the risk of erroneously classifying wet sands, exposed mudflats, and temporarily moistened surfaces as water. The threshold of −0.1 was selected as a conservative empirical threshold based on visual interpretation of multispectral composites and comparative analysis of classification results obtained using NDVI, NDWI, and MNDWI (Sun et al., 2012; Bishop-Taylor et al., 2019). The raster image was then converted to vector format to extract coastlines and land/water polygons.

2.5 Method for calculating coastline shift

We chose 1995 as the baseline year for analyzing coastline changes because the Caspian Sea level reached its maximum level in the past 30 years before starting to decline. This enabled us to monitor how the coastline changed as the sea level fluctuated in the following years.

We measured all coastline shifts along the 17 segments (transect boundaries) using the “Measure Line” tool. For each of the 17 segments, we marked the intersection points with the coastline for every year in the analyzed period - from 1996 to 2023. We then measured the distance from the intersection point of the 1995 coastline to these subsequent intersection points. We assigned a minus sign (-) to values corresponding to coastlines located west of the 1995 coastline, while we assigned a plus sign (+) to those to the east.

We determined the coastline shift rate as the difference between the average shift values across all 17 segments (transect boundaries) of a given year and the previous year. We recorded the absolute shift values and the calculated rates in meters in a Microsoft Excel spreadsheet.

2.6 Method for assessing mainland and island areas

We measured the land area within the defined study area for the spring periods from 1996 to 2023 and additionally for the autumn of 2023, and differentiated it into mainland and island areas.

We defined the mainland as the territory bounded by the eastern border of the study area and the coastline of the corresponding year to the west. We designated the water area as the space extending from the coastline to the western boundary of the study area.

We defined islands as landmasses surrounded by water. We calculated the island area based on QGIS polygon data located within the water area. Because the vector data were divided into land and water, we deleted all water polygon data within the sea area and water bodies on the mainland to obtain a clear representation of the land area. If coastline retreat caused large water spaces to become separated and located within the mainland, we still classified them as water areas, as they could potentially remain connected to the sea. To create demonstration maps illustrating coastline shifts and the formation of land and islands, we used the 2GIS map basemap, accessed through the “QuickMapServices” plugin.

2.7 Aerial survey method

2.7.1 Development of the aerial survey route

To determine the aerial survey route, we assessed potential seal haul-out sites using Sentinel-2 and Landsat-8 satellite images with a spatial resolution of up to 10 meters (Sentinel-Hub, 2024). We reviewed images taken approximately 10 days before the scheduled survey date. This allowed us to analyze images where cloud cover did not obstruct the view of potential haul-outs. We also examined images captured during offshore (wind set-down) and onshore (wind set-up) wind surges in the “Prorva” area (Figure 1) to evaluate the presence of islands and shalygas under different hydrological conditions in shallow waters. We selected the image with the clearest visibility and placed waypoints on islands and shalygas within the study area where seals were most likely to haul out. In addition to the main study area, the flight route included the Southwest shalygas (potential haul-out sites outside the study area) in 2022 and 2023, as well as the Tyuleniy Islands archipelago. According to flight regulations, the maximum number of waypoints was limited to 32. Connecting these waypoints sequentially formed the flight route (Figure 1).

We conducted the aerial surveys using a twin-engine turboprop short-haul De Havilland DHC8–315 aircraft, following the established methodology for studying the Caspian seal in the Republic of Kazakhstan (Order of the Minister of Environment and Water Resources of the Republic of Kazakhstan No. 104-Ө, 2014). On Approval of the Rules for the Preparation of Biological Justification for the Use of Wildlife; Baimukanov and Ryskulov, 2022). Flights departed from and returned to the Atyrau airport, located approximately 200 km from the nearest point on the North Caspian coast. The general characteristics of the aerial surveys are presented in Supplementary Table 3.

Five specialists participated in the aerial survey. Before the flights, we synchronized the time on all photo and video equipment and GPS navigators to within one second. This ensured precise georeferencing of the photo and video data to the flight route waypoints during later data processing.

During aerial reconnaissance, we used two Sony FDR-AX43 4K video cameras, a Canon EOS R5, and a Nikon D7000. Throughout the flight, we maintained a dictaphone recording and tracked the route with two Garmin 66S GPS units. Each observer had a route map with marked waypoints. The forward observer monitored the aircraft’s geographical position via the GPS and used the cabin microphone to inform the camera operators in advance when approaching specific waypoints marked on the map. When necessary, this observer coordinated the route with the pilots to cover haul-outs that were not initially indicated on the flight map. We recorded continuous video from both sides of the aircraft during the entire marine segment of the route. Two operators, one on each side, took photographs of seals upon detection and photographed their potential haul-out sites.

2.7.2 Analysis of aerial survey data

We examined all photographic materials to identify the presence of seals. We counted and numbered the seal images in Adobe Photoshop using the “Count Tool.” An operator paused the video footage every second to inspect the frame for the possible presence of seals. When seals were found, we saved screenshots and performed the counting and numbering in Adobe Photoshop.

We grouped the seal images into two categories: 1) Haul-outs – aggregations of seals on an island or groups of 10 or more seals swimming in the water; 2) Solitary seals – fewer than 10 seals either alone or scattered across the water area. In some cases, there was no nearby island, while in others, an island was present but lacked seal aggregations.

We reviewed the dictaphone recordings and recorded visual counts of seals that were not captured on photos or videos. During this expert assessment, we classified the animals into haul-outs and solitary seals based on the criteria described above. Due to wind and salt fog during the 2020 survey, some photos were of low quality. For these images, we estimated the number of seals by extrapolating the average seal size to the aggregation area, or by extrapolating a known haul-out density to the haul-out area.

Supplementary Table 4 presents the number of seals recorded during the aerial surveys.

To analyze the distribution of the recorded seals, we determined their survey coordinates by matching the time of the photo or video capture using metadata (via the Exiftool program) with the flight track logs from GPS devices in “.gpx” format. Using the gpsbabel program, we increased the number of track points to ensure each second of the flight had a corresponding time and location coordinate. We accomplished this using the “interpolate between trackpoints” filter. We selected the coordinates that matched the seal detection times, converted the output into a “.csv” file in QGIS, and subsequently saved it as a Shapefile for further analysis.

Within the study area, we numbered haul-outs and solitary seals from each aerial survey separately, from north to south: haul-outs received one set of sequential numbers, and solitary seals received another. Seals at each haul-out were counted by numbering them sequentially from No. 1 onward in Adobe Photoshop using the Count Tool. We used this approach to count and determine the locations of haul-outs and solitary seals relative to the transects. Based on the results of the 2023 aerial survey, haul-outs and solitary seals located outside the study area received numbers that continued the respective sequences.

We calculated the location of all seal encounter points, including haul-outs and solitary seals, within the study area relative to the 2020 coastline. We generated lines of latitude using the “View” > “Decorations” > “Grid” panel at 200-meter intervals. We conducted measurements with the “Measure Line” tool in the QGIS application. We chose the 2020 coastline as the reference point because it marked the year when the first aerial survey was conducted and haul-outs and solitary seals were first discovered in the study area.

To evaluate the shift in seal distribution from the northeastern Caspian Sea to the Tyuleniy Islands, we calculated the mean coordinates of seal haul-outs in both areas. For the northeastern region, we used data from aerial reconnaissance flights on March 8, 2020, March 13, 2022, and November 4, 2023. For the Tyuleniy Islands, we used data from November 4, 2023. We generated these coordinates using the “Mean coordinate(s)” tool in QGIS, and measured the distance between them with the “Measure Line” tool.

2.8 Statistical methods

To describe the dynamics of sea-level changes and changes in mainland and island areas, we performed regression analysis in Microsoft Excel. We assessed the type of relationship between the data (X and Y) by comparing the coefficients of determination (R²) for different trend lines (linear, logarithmic, power, exponential, polynomial). When the approximation reliability values for various trend lines were similar, we selected the linear relationship. To avoid ambiguous interpretation, only visual trend lines are shown on the graphs as an illustrative element. Regression equations and coefficients of determination are provided in Supplementary Table 5.

We employed a graphical method to evaluate the normality of data distribution. We conducted statistical analysis in the R software environment (version 4.5.2). To assess the diversity of values within the samples, we reported their range and calculated the mean with standard deviation error (Mean ± SE) and the coefficient of variation (CV, %). In cases of non-normal distribution, we employed the Mann-Whitney U test (Zar, 2010) to determine the significance of differences between independent samples, reporting the results as medians (Me) with the interquartile range [Q1; Q3]. Since the parameters studied did not follow a normal distribution, we used Spearman’s rank correlation coefficient (Zar, 2010) to evaluate the relationships between sea level values, coastline shifts, and the areas of the mainland and islands.

3 Results

3.1 Sea-level dynamics

Sea-level dynamics are analyzed for the period from 1995 to 2023. We selected 1995 as the starting point because that year recorded the highest sea level since 1935, reaching -26.59 m BS (Baltic System). Starting in 1996, the sea level began to decline (Figure 2). A brief phase of sea-level rise occurred from 2001, followed by stabilization around -27 m BS between 2004 and 2007, and then a steady decline until 2016. A slight increase was noted in 2017. Between 2020 and 2023, the sea level dropped sharply by 0.64 m (Table 1).

Figure 2

Overall, during the analyzed period, the sea level dropped by 2.28 meters. Regression analysis of the sea-level dynamics indicates that a sixth-degree polynomial trendline provides the most reliable approximation (R² = 0.9911) (Supplementary Table 5).

3.2 Ice conditions

An assessment of ice conditions using satellite imagery prior to all conducted aerial surveys revealed an absence of ice in the North Caspian (Supplementary Figure 2).

3.3 Coastline shift

We assessed the coastline shift rate from 1996 to 2023 relative to the 1995 coastline, which was taken as a relative zero, based on spring data using two methods: 1 - Shift along individual boundary segments of the transects (Figure 3); 2 - Shift over time (Figure 4).

Figure 3

Figure 4

The shift along individual boundary segments of the transects indicates that the peak indicators for several transects show values above 0; despite the sea-level decline, the coast advanced eastward for some time. This is especially evident in segments 5 and 6, 10-12, 14, and 17. The maximum eastward advance values exceeded 8.5 km and 11 km. However, overall, a notable westward coastline shift was observed by 2023. This is particularly apparent starting from the 13th boundary segment (Figure 3). The trend line is best modeled by a fifth-degree polynomial curve with a high approximation reliability value (R² = 0.9512) (Supplementary Table 5).

The average coastline shift over all years for boundary segments 1–12 was 44.08 ± 6.82 km (CV = 15.5%), while for segments 13-17, it was 163.73 ± 12.26 km (CV = 7.5%). In other words, the westward movement of the coastline from the 1st to the 12th boundary segment was about four times less than from the 13th to the 17th transects. A comparison of the medians shows a similar magnitude of difference and is highly significant (-41,763.4 [-44,776.7; -39,919.4]; -165,584.4 [-169,335.5; -165,294.0]; W = 60, p < 0.01). The greatest retreat occurred in the area of the 15th transect, where the coastline had receded by 175.4 km by 2023.

According to Figure 4, an eastward movement of the coastline in some areas was observed in 1996 and 1997, followed by a significant westward retreat in 1998 - more than 30 km on average. In 1999 and 2000, the coastline moved east again, reaching roughly the 1996 level. Given the lack of median data for 2001-2007, we can state that since 2008, the coastline has been steadily shifting westward with small fluctuations (Figures 4, 5). The sharpest decline was observed in 2023.

Figure 5

An analysis of the coastline retreat rate over different years shows that the highest westward shift rate occurred in 1998, followed by a rapid eastward movement in 1999. The lack of annual data for 2001–2007 prevents us from assessing the coastline shift rate during those years, but the shift occurred in a westward direction by more than 30 km.

Until 2020, the shift was recorded in both westward and eastward directions, but overall, the coastline moved westward, with the average calculated rate across all segments being 2.13 ± 2.09 km per year (CV = 98.5%); Me = -1.18 [0.87; 1.75].

Starting in 2020, a consistent westward shift has been observed. The rate for the 2020–2021 period was over 8.6 km; it dropped significantly the following year, but increased to more than 23 km in 2022-2023. The average westward rate across all transects was 11.08 ± 7.49 km per year (CV = 67.6%); Me = -9.24 [6.37; 11.72]. Therefore, the shift rate for the 2020–2023 period increased by a factor of 7.8 compared to all previous years (W = 275, p < 0.001) (Figure 6).

Figure 6

The westward coastline retreat rate across all transects for the entire study period averaged 2.83 ± 2.03 km per year (CV = 71.7%); Me = -1.57 [1.47; 5.11].

3.4 Dynamics of mainland and island areas

The mainland area (hereinafter referred to as land) increased by a factor of 1.72 during the study period and shows a linear trend for further growth (Figure 7). The approximation reliability value is high (R² = 0.9058) (Supplementary Table 5), with an average increase of more than 432.48 km² per year.

Figure 7

Schematic maps of the study area showing the variability of mainland and island areas for specific years are provided in Supplementary Figure 3. A significant increase in the mainland area is noteworthy in 1998, followed by a sharp decline in 2000. Subsequently, however, the land area increases again with minor fluctuations.

At the same time, we observed a decrease in the island area by a factor of 15.76. In 1998, the island area demonstrated a severe decrease. In 1999, the island area increased, followed by a consistent decline with varying degrees of interannual fluctuations; the trend line is close to a linear relationship. The approximation reliability value is above average (R² = 0.6901) (Supplementary Table 5), with the island area decreasing at an average rate of 27.838 km² per year (Figure 8).

Figure 8

3.5 Correlation among various parameters

Correlation analysis reveals a strong pairwise relationship between all variables with high confidence limits - p < 0.001 (Supplementary Table 6).

3.6 Seal population distribution

During the aerial surveys conducted in 2020, 2022, and 2023, no seals were observed on the Southwest shalygas. Sightings were recorded within the study area and near the Tyuleniy Islands archipelago.

General information regarding the number of recorded seals is provided in Supplementary Table 4. The number of haul-outs in 2022 was significantly higher than in 2020. In 2023, the number of haul-outs fell between the 2020 and 2022 counts. We determined that most seals were hauled out on islands (i.e., in aggregations), while an insignificant proportion was encountered as solitary seals: 0.22% in 2020, 0.49% in 2022, and 0.33% in 2023 of the total number of seals recorded each year. Despite the small number of recorded solitary seals, the geographic range of seal distribution expanded significantly specifically due to these solitary individuals, according to the 2022 and 2023 survey data, covering transects 3, 15, and 16, where no haul-outs were observed (Figures 9, 10).

Figure 9

Figure 10

Distribution graphs for both haul-outs and solitary seals show a common pattern: no seals were found on transect 11, and in 2023, seals were detected outside the designated study area near the Tyuleniy Islands. This allows us to divide the seal distribution into three zones: the first (northern) zone from transects 1 to 10; the second (southern) zone from transects 12 to 16; and the third - the Tyuleniy Islands (Figure 9). Determining the mean coordinates of seal distribution in the northeastern part of the sea, combining the first and second zones, yielded 45° 53.719’ N 54° 28.802’ E, while the third zone yielded 45° 15.033’ N 52° 23.130’ E.

A comparative analysis of the quantitative indicators of seal haul-outs shows immense variation in the number of hauled-out seals on the islands (CV varies from 103% to 229%). Haul-outs can contain a minimal number of seals - 10 or 30 individuals. The maximum number was recorded in 2020 at a haul-out located in transect 13, comprising 9,806 hauled-out seals. In subsequent years, the maximum number of individuals in a single haul-out was significantly smaller - approximately half that amount. At the same time, the wide range of seal abundance across haul-outs demonstrates their inequality (Figure 11a).

Figure 11

An analysis of population distribution within the study area across haul-outs allows for a conventional division into three categories: small (from 10 to 100 individuals), medium (from 101 to 900 individuals), and large (901 or more individuals). In terms of the number of haul-outs, small and large haul-outs dominated in 2020. In 2022, the number of small haul-outs increased markedly, while large haul-outs decreased but still outnumbered medium haul-outs (Supplementary Table 7; Figure 11b). The total number of seals in small haul-outs was 1.2% in 2020, doubled in 2022, but dropped to 0.3% in 2023.

In 2023, large haul-outs dominated. In terms of the total number of seals, large haul-outs occupied the dominant position, although their relative share decreased in favor of medium haul-outs in 2023.

The number of haul-outs lacked a normal distribution, which is evident from both a graphical review of the entire spectrum of haul-outs with varying population sizes (Figure 11a) and a more detailed breakdown of the large haul-outs (Figure 11b).

Consequently, the standard errors of the mean values for the number of seals at haul-outs are very high, making comparisons between mean values unreliable (Supplementary Table 8).

The seal population distribution graph based on 2020 data (March 8) shows that the primary concentrations were located in the southern zone from transects 12 to 14, accounting for 86.2% of all individuals recorded on haul-outs that year. However, the number of haul-outs in the northern zone on transects 5 to 10 accounted for about 48% of the total recorded (Figure 10; Supplementary Table 8), which results from the uneven distribution of seals across the haul-outs. For example, the mean number of hauled-out seals within transects 5–10 was 5.8 times smaller than within transects 12-14; a comparison of the medians shows highly significant differences in abundance between the northern and southern zones at the p < 0.001 level (Supplementary Table 8).

The seal distribution in 2022 (March 13) appears different: 69.3% of the haul-outs were located in the northern zone within transects 4-10; the abundance in this area also became predominant, accounting for 61.72% of all hauled-out seals. Regarding the number of hauled-out seals, transects 6, 7, 8, and 14 stand out, collectively hosting 95.3% of all individuals recorded that year. It should be noted that in the southern zone, transect 14 clearly dominated in both the number of hauled-out seals and haul-outs, where 37.5% of all recorded seals were concentrated across 18 haul-outs, with a mean of 1,285.9 individuals per haul-out.

Overall, the mean number of seals per haul-out in the southern zone was higher than in the northern zone (1,028.2 versus 733.7 individuals per haul-out), but the differences between the medians were not statistically significant (Supplementary Table 8).

A comparative assessment of the median abundance on haul-outs between 2020 and 2022 in the northern zone (4-10) showed no significant differences, while in the southern zone (12-14), the mean number of hauled-out seals per haul-out was higher in 2020 than in 2022.

In autumn 2023 (November 4), the vast majority of seals were located in the northern zone - on transects 5, 6, and 7 - accounting for 82.8%. A significant portion (15%) was on transect 9, which aligns with the distribution of haul-outs: 95% of the haul-outs were located within transects 5–7 and 9 (Figures 10, 12). No haul-outs were encountered in the southern zone; only 1 individual was recorded in the water (transect 13). In the third zone near the Tyuleniy Islands, 1.9% of the seal population was recorded across 3 haul-outs. Solitary seals were encountered from transect 3 to 10, with the majority (62%) on transects 5 and 7; solitary seals were also observed near the Tyuleniy Islands.

Figure 12

A comparative median assessment of abundance showed significant differences in the number of hauled-out seals per haul-out within transects 4–10 between 2020 and 2023, and between 2022 and 2023. The mean number of seals per haul-out increased significantly in 2023.

3.7 Distribution of haul-outs and solitary seals relative to the coastline

An assessment of the distribution of haul-outs and solitary seals relative to the 2020 coastline indicates that, in spring 2020, haul-outs were located at distances ranging from 16.4 km to 38.5 km. A large number (14 haul-outs) were situated on transect 6 at a mean distance of 32.3 km, while another concentration center was on transect 12 (15 haul-outs) at a mean distance of 21.5 km. A substantial number of solitary seals recorded on transects 5 and 6 were located at distances ranging from 17.2 to 39 km, whereas sightings of solitary seals on transects 12 and 13 were recorded at distances between 19.1 and 16.5 km (Figure 13; Supplementary Tables 9, 10).

Figure 13

In spring 2022, seal haul-outs in the northern region were located relative to the 2020 coastline at distances from 15.3 to 41.5 km, and in the southern region from 20.1 to 45.4 km. A comparison of the haul-out locations between 2022 and 2020 reveals that the minimum distance is shorter than the 2020 minimum, and the maximum is greater than the 2020 maximum. Overall, some haul-outs were closer than in 2020, but the majority shifted further from the 2020 coastline, with the mean distance metric demonstrating a 0.67 km shift westward. Solitary seals were, on average, encountered 14.1 km further seaward than the 2020 encounters (Figure 13; Supplementary Tables 9, 10).

A significantly larger westward shift for both haul-outs and solitary seals was observed in autumn 2023. The minimum distance for haul-out locations was 25.9 km, and the maximum was 66.6 km; solitary seals were encountered approximately within these same bounds. Overall, based on mean metrics, seal haul-outs shifted by approximately 10–11 km from their locations in 2020 and 2022.

4 Discussion

In the 20th century, two oppositely directed processes were observed in the long-term sea-level dynamics of the Caspian Sea: a prolonged decline from 1930 to 1977, followed by a prolonged rise from 1978 to 1995 (Water Balance and Fluctuations of the Caspian Sea Level. Modeling and Forecast, 2016). During the decline phase, the sea level dropped by 3 m, reaching its minimum level in the past 500 years at -29 m BS. The subsequent rise amounted to 2.5 m; thus, as a result of the regression and transgression that occurred, the sea level decreased by a net 0.5 m. Since 1996, the sea has again been in a regression phase, which is largely driven by climate warming, decreased river inflow, and increased evaporation from the water surface (Rychagov, 2011; Prange et al., 2020). The negative socio-economic consequences of these cyclical processes can be considerable, and efforts are underway to find ways to forecast and mitigate them - for example, by constructing dams to protect coastal economic facilities from the encroaching sea (Water Balance and Fluctuations of the Caspian Sea Level. Modeling and Forecast, 2016).The assessment of ecological risks is also of great importance, given the threat of bioproductivity loss in marine areas due to the drying of water spaces, habitat pollution of flora and fauna (primarily due to the flooding of oil and gas wells) (Panasenko et al., 2019), as well as the loss of habitats for endangered endemic sturgeon species and the Caspian seal (Duan et al., 2025).

Forecasts indicate that the regressive trend of the Caspian Sea will persist until the end of the 21st century (Prange et al., 2020). Global warming could lead to a further sea-level drop of 21 m by 2100. However, even a 5–10 m drop will critically disrupt key ecosystems, necessitating a shift from traditional static Caspian Sea biodiversity conservation planning to a proactive, dynamic approach that allows protected areas to assist endemic and rare species in adapting to ongoing changes (Court et al., 2025). The materials analyzed in this article contribute to the scientific basis for such a dynamic, flexible approach to conserving the endangered endemic Caspian seal.

Long-term seal abundance monitoring data collected since 2008 (Baimukanov et al., 2020, 2023) in the North and Middle Caspian, as well as seal migration studies (Dmitrieva et al., 2016), have demonstrated the immense importance of the northeastern part of the sea for seal haul-outs during the spring and autumn periods. When referring to the northeastern Caspian, we mean the marine area from the mouth of the Zhaiyk (Ural) River, extending along the eastern coast of the North Caspian to the Buzachi Peninsula, encompassing Kaydak Bay and Komsomolets Bay (Figure 1).

A retrospective analysis, coupled with modern data, indicates that conditions suitable for Caspian seal haul-outs have persisted in the northeastern Caspian. The fragility of these conditions will be discussed further. It should be noted that on the Southwest shalygas islands located near the Zhaiyk (Ural) River mouth, haul-outs of many thousands of seals existed in the first third of the 19th century (Karelin, 1883), which were actively harvested. A century later, however, no more than a hundred individuals were encountered on these islands (Badamshin, 1950). During helicopter flights on October 16 and November 1, 2010, only 10 and 5 seals were observed, respectively (Baimukanov et al., 2022). Aerial surveys conducted in 2020, 2022, and 2023 showed a complete absence of seals on them. The main reasons for this include bycatch in fishing nets, poaching, and disturbance factors such as shipping near seal haul-outs and human visits, which cause the animals to flee (Badamshin, 1950; Baimukanov et al., 2022).

Research from the 1930s and 1940s also highlights the high significance of the marine area along the eastern coast of the North Caspian as a seal habitat. It was noted that due to the sea-level drop, the Dead Kultuk and Kaydak bays located here became separated from the sea, the coastline shifted westward, and most islands merged with the mainland, became overgrown with vegetation, and could no longer serve as seal haul-outs (Badamshin, 1950). Instead, new small islands formed in the shallow waters, where large numbers of seals began to haul out. Five major haul-out groups were identified: Rakushechnye, Suendykovskie, Bolashovskie, Kolkhoznye, and Urtaespinskie shalygas, where seals hauled out constantly from early spring until late autumn. The highest abundance was observed in spring (from late March to early May) and in autumn (from early October until ice freeze-up, November 15-29). Data on the total absolute abundance of marine mammals were based on seal harvest calculations from a specific area, as well as visual observations (“by eye”) and corresponding extrapolations to the total haul-out area (Badamshin, 1950). The total abundance across all shalygas could reach up to 25,000 individuals in April and up to 34,000 individuals in November. Although these calculations are highly approximate, we assume that the abundance of hauled-out seals during the described historical period was certainly no less. It was possibly significantly greater, considering that visual counting typically underestimates seals located in the back rows.

What causes such a high concentration of seals in this part of the sea? The shelf in the eastern North Caspian has a low-gradient slope, and depths do not exceed 4 m up to the “Ural Trench” depression (Sadykov et al., 1995; Mitina et al., 2016), while along the coast in a strip several kilometers wide, depths are less than 1 m. The shallow waters near the coast make the islands and shalygas that form in this marine space highly inaccessible to humans. This inaccessibility reduces or eliminates the impact of the aforementioned negative factors on seal haul-outs, serves as a barrier that protects their haul-outs from anthropogenic disturbance, and promotes the formation of an ecosystem favorable to the Caspian seal. The question arises: to what extent can the coastline regression affect this condition, and how critical is the sea-level drop for seal distribution? To address this question, we conducted a comprehensive analysis of the hydrological and hydro-morphological conditions, as well as seal distribution, within a specific study area where seal haul-outs previously formed and currently form. These studies aim to explore the relationship between sea-level dynamics, changes in mainland and island areas, and the abundance and distribution of seals.

The northeastern Caspian is characterized by an extremely shallow coastal plain with a low seabed slope. In such conditions, even relatively small wind-driven surge and set-down fluctuations in water level can lead to significant horizontal displacement of the coastline. During set-down events, vast shallow areas become exposed, forming wet sands, mudflats, salt marshes, residual pools, and temporarily moistened surfaces. These transitional surface types create significant uncertainty in spectral water extraction, as their reflectance properties can be similar to those of shallow or turbid water.

Taking these features into account, this study applied a conservative NDVI-based threshold approach to delineate the main open water surface. Although NDVI is traditionally used as a vegetation index, water surfaces are generally characterized by low or negative NDVI values due to strong near-infrared absorption (Sun et al., 2012; Bishop-Taylor et al., 2019). In the context of this study, this property of NDVI was particularly important, as some Caspian seal haul-out areas in the shallow coastal zone may become overgrown with reeds and other aquatic-riparian vegetation. Therefore, the chosen index needed not only to separate open water from exposed or moistened surfaces, but also to allow the exclusion of vegetated areas from the open water class. Reed and wetland vegetation areas are typically characterized by higher, often positive NDVI values, making this approach useful for distinguishing open water from vegetated coastal complexes.

It should be noted that independent quantitative validation of the chosen threshold against ground truth data or higher spatial resolution imagery was not conducted within this study. The threshold used should therefore be regarded not as a universal water extraction value, but as a locally adapted, expert-selected criterion for the conditions of the Northeastern Caspian. This approach may lead to underestimation of some very shallow, turbid, or bottom-reflectance-affected water; however, it was considered preferable for the purposes of this study, as it reduces the likelihood of overestimating the water area due to wet sands, mudflats, salt marshes, and temporarily moistened surfaces.

Observation results from two hydrometeorological stations from 1995 to 2023 show that the sea-level drop in the Kazakhstan sector of the North Caspian amounted to 2.28 m, with the level once again approaching the -29 m BS mark. During this period, a brief phase of sea-level rise and stabilization was observed, but there is an overall downward trend, with the rate increasing in recent years (2020-2023), reaching an average of 0.21 m per year.

In this regard, a westward coastline retreat is observed, which proceeds unevenly and is associated with the geographical features of the eastern North Caspian. This area hosted Dead Kultuk, the largest bay for the described marine sector, which was already marked as dried up and turned into a salt marsh on maps based on the 1962 situation (Map of the Caspian Sea, 1985). The 1995 coastline, taken as the relative zero, ran along this vast salt marsh and encompassed Kaydak Bay to the south, which indented deeply into the mainland. Even at the beginning of the last century, it was noted that the depths in this part of the sea, despite being shallow, are not uniform - there are elevations and depressions ranging from 0.5 to 1.6 m (Kiselevich, 1914). When interpreting coastline variability, this unevenness in sea depth within the study area, which could affect the assessment of the coastline shift, must be kept in mind. One must also consider wind surge phenomena, which can shift the coastline up to 30 km or more in either direction over a short period of just a few days (Sadykov et al., 1995). However, calculating the median allowed for an annual robust assessment of the coastline’s central position in the available satellite imagery dataset for the March-May period (Berlyant, 2002; Deakin et al., 2010).

The assessment of coastline shift along east-west extending transects within the study area boundaries shows a relatively uniform rate on transects 1-12, where the retreat from 1995 to 2023 ranged from 38.05 to 62.62 km, averaging 44.08 km. For transects 13 to 16, which cover the northern part of Kaydak Bay, the retreat during this same time was significantly larger - from 143.04 to 175.39 km, averaging 163.73 km, indicating the process of this bay drying out.

It is worth noting that the coastline shift on individual transects was not westward every year. The presence of the large Kaydak Bay and the gentle slope of the seabed determined the specific nature of the coastline shift near the bay during the initial period of the regression phase. The accumulation of a significant volume of water in the bay during the first years of the sea-level drop could “hold back” the coastline at approximately the 1995 position. Thus, a sea-level drop of up to 0.5 m in 1996–1997 did not lead to a noticeable coastline shift; exceeding this value in 1998 resulted in its significant westward shift, but the following year, with a 0.2 m sea-level rise, the coastline returned to approximately its previous position (Figures 2, 4). Eastward coastline shift was also observed in some other years, albeit with a predominant overall westward trend (Figures 3, 4, 5). Overall, until 2020, the westward shift fluctuated from a few tens of meters to 7.4 km, averaging 2.7 km per year. Starting in 2020, a continuous westward shift has been observed, with its maximum rate achieved between 2022 and 2023, exceeding 23 km per year (Figure 6).

Important insights were gained from analyzing the changes in the mainland area and the forming islands. While the coastline shift led to a significant 1.72-fold increase in the mainland area from 1995 to 2023, the island area, conversely, decreased by a factor of 15.76 (Figures 7, 8). The growth of the mainland area is quite logical - the retreat of the sea naturally leads to an increase in the mainland area. The westward coastline shift led to the separation and subsequent drying of Kaydak Bay in 2017 (Supplementary Figure 4). In 2023, Komsomolets Bay completely dried up as well, except for a small inlet on transect 13 (Figure 8).

But why does the retreat of the sea lead to a significant reduction in island area? One of the reasons for this may be the merging of large islands with the mainland. The most striking example is the evolution of Durnev Island in Komsomolets Bay. Initially, an island emerged on the site of an underwater shoal, the so-called “bank,” which was first described in 1910 (Kiselevich, 1914). Later, it was mentioned that in the 1920s, more islands and shalygas formed nearby, which became known as the Durnev islands (Badamshin, 1950). During the previous regression period in the 1930s-1970s, these islands grew and were not flooded during the transgression that lasted until 1995. The new regression phase resulted in the islands located in the northern part of Komsomolets Bay ceasing to exist by 2009, having merged with the mainland (Figure 8). This caused the island area within the study area to halve between 1995 and 2009.

The fluctuations in the increase and decrease of island area repeat cyclically, as shown in the island area dynamics graph (Figure 8). The cycle ranges from 3 to 5 years. This cyclicity is explained by the fact that as some islands merge with the mainland, new islands emerge in the sea, which then increase in size, merge with one another, and in turn also merge with the mainland.

The newly formed islands in the shallow waters are small in area; they are elongated - up to several hundred meters in length - and narrow, ranging from a few tens of meters to 100 m in width. Their origin is associated with winter ice drift. Large, massive hummocky blocks of sea ice characteristic of the North Caspian – stamukhas - gouge into the shallow seabed. The underwater part of stamukhas can reach 20 m. Driven by wind and currents, they move, pushing and shifting the seabed soil, forming furrows and elevations (Bukharitsin, 2019, 2021) (Supplementary Figure 5).

During wind set-downs and sea-level drops, these shallow-water elevations turn into exposed islets - shalygas. Year after year, stamukhas and drifting ice can pile more and more marine soil onto the previously formed seabed elevations, facilitating their growth and transforming them into permanent islands. Many shalygas and islands are attractive for seal haul-outs and serve as their haul-outs. However, due to the declining sea level, the newly formed islands near the coast initially increase in size, incorporating other nearby islands, and then merge with the mainland. It is worth noting that the construction of bird nests (cormorants, various gull species) and reed growth also contribute to the increase in island size (Supplementary Figure 6).

Thus, the evolution of islands during the regression of the Caspian Sea follows an evolution triad: Emergence - Growth - Merger with the mainland. Evaluating the rate of triad cycle turnover indicates how rapidly the shallow-water ecosystem of the northeastern Caspian is transforming under the influence of sea regression.

As the distance from the 1995 coastline increases, the shelf bottom slope and sea depth increase, which apparently reduces the impact of stamukhas on the seabed soil. Furthermore, this is facilitated by climate warming, which reduces the area and thickness of the ice cover (Lobanov and Naurozbaeva, 2021) and, consequently, decreases stamukha formation. Sea waves also smooth out the elevations until they disappear. And as the coastline approaches the depth zone - the Ural Trench area - an acceleration of the triad cycle and a reduction in the total island area are observed.

The correlation between all the analyzed parameters: sea-level drop, westward coastline shift, mainland area increase, and island area decrease is high and indicates a tight interconnection between the described phenomena (Supplementary Table 6).

As a result of the triad, the area of potential Caspian seal haul-outs is shrinking. This is well demonstrated by the situation with the haul-outs on the western Durnev islands. On these islands, located in Komsomolets Bay, seal haul-outs numbering approximately 25,000 and 20,000 individuals were discovered on April 13, 2009, and April 9, 2011, respectively (International study of the Caspian seal, 2010; 2011; 2012).

In late April and early May 2016, seals were absent from these islands; the depth near these islands in a strip several kilometers wide was no more than 0.1-0.2 m, which prevented the seals from approaching them to haul out. Seals were found at the edge of the northern part of Komsomolets Bay, where depths near the islands increased to 0.4–1.0 m (Baimukanov et al., 2020). The maximum abundance of hauled-out seals on one of the reed-covered islands during the survey days of April 29-May 1 was 391 individuals. Seals continued to haul out at this site until 2019 (67 individuals on April 20, 2019) (Study of habitat conditions, 2020). These numbers do not reflect the true abundance of hauled-out seals since the survey was conducted at the end of the molting period, but they demonstrate that the possibility for animals to haul out existed.

It was during this period, from 2009 to 2019, that intensive sea-level decline occurred; the level stabilized slightly in 2019–2020 before plunging downward again from 2021 (Figure 2). As visible in Figure 9, the spring 2020 coastline shifted slightly eastward due to water ingress through deeper areas. Large new islands formed near Komsomolets Bay, surrounded by thin water passages separating them from the mainland. However, the overall sea-level decline prevented seals from reaching these islands, and 2019 became the last year when seals hauled out in Komsomolets Bay. By 2021, these large islands had merged with the mainland (Supplementary Figure 3), significantly increasing the mainland area in the northeastern Caspian (Figure 7).

The results of the conducted aerial surveys clearly demonstrate the patterns of seal haul-out dependence on coastline shift and the island evolution triad.

Considering other alternative factors that could have influenced the shift in seal haul-out locations, we note those most likely to be significant. As previously mentioned, the northeastern Caspian is shallow, and in the areas where seals haul out, factors such as fishing and shipping are either absent or reduced to a minimum. Furthermore, the northern part of the Caspian Sea is subject to a nature reserve regime, which also restricts oil extraction activities on the shelf near seal haul-out sites and aircraft overflight (Ecological Code of the Republic of Kazakhstan, 2021). Anthropogenic factors are therefore minimized and cannot significantly influence the shift in haul-out locations.

Among natural factors, we note the possibility of changes in food resources. The Caspian seal is an obligate piscivore, and dietary studies (Baimukanov et al., 2025) have established that during spring and autumn haul-out periods at island sites, the primary prey items are goby fish, which predominantly lead a sedentary, bottom-dwelling lifestyle. Their migrations are local in character and limited to short distances. Therefore, this factor also cannot lead to a significant shift in seal haul-out locations.

The timing of the aerial surveys was also not chosen randomly.

Abundance counting took place during seal concentrations, when they gather on the islands. It is important to note that in the life cycle of the Caspian seal, these phenomena repeat periodically and are vital for the species’ existence. In spring, they gather on islands to complete molting; in autumn, they congregate on them again to wait for ice formation and climb onto the ice for reproduction. In both cases, the researcher has an excellent opportunity to count the seals in haul-outs. In addition, seals observed in the water, both near the haul-outs and far from them, were also counted. Overall, this provided an opportunity to assess the abundance distribution of both hauled-out animals and those at sea.

The Caspian seal is sensitive to the periods of ice formation and melting in the North Caspian and is not encountered on island haul-outs even in the presence of minimal ice cover areas, both in spring and in autumn (authors’ unpublished data).

The 2020 and 2022 aerial surveys were conducted in early spring - on March 8 and March 13, respectively. The main condition for selecting the flight dates by the researchers was the absence of ice on the sea. It was important to survey the seals gathered at haul-outs immediately after the ice melted for molting. Doing so captured the locations and the maximum abundance of seals at the haul-outs, thereby increasing the objectivity of the data regarding the distribution of hauled-out seals. Had the surveys been conducted later, the number of hauled-out animals would have decreased by the number of seals that had already finished molting.

In autumn, it was equally important to conduct the aerial survey during the peak assembly of seals at haul-outs in the eastern part of the North Caspian. It is known that seals gather at island haul-outs in the North Caspian during the pre-winter period, starting in the second half of September, and actively throughout October and November until ice freeze-up (Badamshin, 1950). As shown by aerial reconnaissance on November 20, 2021, seals abandon the islands upon the formation of initial ice - thin and brittle fast ice (authors’ unpublished data). Initial ice more commonly forms in the second half of November; therefore, November 4 was selected for the 2023 aerial survey, at which point coastal ice had not yet formed.

The spring 2020 aerial survey was the first in the 21st century to discover numerous seal haul-outs stretching along the coast of the northeastern Caspian. Therefore, the location of haul-outs and seal distribution in this and subsequent years was assessed relative to the 2020 coastline.

In 2020, based on mean data, seal haul-outs were located at distances ranging from 21.5 to 37.4 km, with a mean distance of 30 km from the 2020 coastline; solitary seals were encountered slightly further away - from 22.5 to 38.2 km, with a mean distance of 32 km. Even though the sea level in 2022 dropped by 39 cm compared to 2020, no noticeable shift in haul-outs occurred in the spring of 2022, amounting to a mean westward shift of only 0.67 km. Encounters with solitary seals, however, were recorded 14.1 km further west compared to similar encounters in 2020.

A significant change in haul-out locations was observed in the autumn of 2023: compared to 2020 and 2022, haul-outs shifted westward by a mean of approximately 10.5 km (Supplementary Table 11). Additionally, some haul-outs were discovered outside the study area - near the Tyuleniy Islands, having shifted in a southwestern direction by approximately 179.5 km. This shift is driven by the further drop in sea level - by 25 cm compared to 2022, which led to a sharp decline in island area. While the island area decreased by a factor of 1.3 over two years (2020-2022), it plummeted by a factor of 4.1 in just one year (2022-2023) (Figure 8). At the same time, vast spaces with new islands are forming near the Tyuleniy Islands archipelago, which have become attractive for seal haul-outs.

The distribution of seal abundance within the study area follows a distinct pattern. Firstly, there is a geographic division of seal distribution within the study area: no seal haul-outs were discovered in transect 11, in the band between latitudes 45°50.804′ and 45°46.756′ N, across all study years. This allows the study area to be divided into two zones: the northern zone (Zone 1, between latitudes 46°31.293′ and 45°50.804′ N) and the southern zone (Zone 2, between latitudes 45°46.756′ and 45°26.500′ N). Secondly, seal haul-outs are shifting outside the study area into the Tyuleniy Islands archipelago region, forming a third zone of seal haul-out distribution.

In 2020, the number of seal haul-outs was distributed roughly evenly between the northern and southern zones, with a slight predominance in the south (48% and 52%, respectively). However, the abundance of seals was predominantly in the second zone — over 86,2% of all seals recorded that year - located between latitudes 45°46.756′ and 45°34.600′ N (transects 12-14). The mean number of hauled-out seals also dominated in the southern zone, with the maximum abundance at a single haul-out reaching nearly 10,000 individuals.

In 2022, the situation changed; both the number of haul-outs and the number of seals were predominantly in the northern zone. In the southern zone, transect 14 (between latitudes 45°38.653′ and 45°34.600′ N) maintained its position, but the number of hauled-out seals per haul-out was almost half that of 2020, although still higher than in the first zone.

In the autumn of 2023, the southern zone completely lost its significance for seal haul-outs. The vast majority of seals hauled out only in the northern zone (98.1%), while the remaining hauled-out seals were located near the Tyuleniy Islands.

As we can see, over the three years of aerial surveys, multidirectional phenomena in the location of seal haul-outs are observed: most of the seals concentrate in a northwestern direction within the study area, while the others are located far to the southwest.

Overall, the number of haul-outs in 2020 was lower than in 2022, but the mean number of hauled-out seals per island was almost twice as large, given the roughly equal total abundance of hauled-out seals in the compared years. This indirectly suggests that the haul-out area occupied by seals was larger in 2020 than in 2022, which aligns with the framework that during a sea regression, islands are unstable - they emerge and soon disappear, and newly formed islands are small in size.

In the future, it is necessary to measure the area of haul-outs and assess the density of seals on them. The present study recorded a lack of normal distribution regarding seal abundance in haul-outs. The reasons for this lie in the varying haul-out conditions. Under unstable conditions associated with small-area islands, small haul-outs form. Their number can be large, as observed in 2020 and 2022. The sea-level drop and depth increase lead to a decrease in the number of newly forming islands; at the same time, previously formed islands grow in area. The latter offer better conditions for seal haul-outs, leading seals to aggregate there in large numbers. This results in the formation of numerous haul-outs containing over 1,000 individuals, and sometimes reaching up to 10,000. The intensifying sea regression accelerates the merging of large islands with the mainland, which also leads to a reduction in seal haul-outs.

As the coastline approaches the relatively deep-water Ural Trench basin, where depths sharply plunge to 4 m and more, the possibility of island formation will near zero. Consequently, in the near future, the entire coast of the northeastern Caspian may lose its significance for spring and autumn seal haul-outs. New haul-out sites will be located near the Tyuleniy Islands and further south.

However, while the northeastern coast was characterized by its inaccessibility to humans and its waters were relatively undeveloped by economic activities (within the study area), the waters around the Tyuleniy Islands host fishing grounds and developed shipping routes, and oil and gas field development is planned north of the islands (Timurziev, 2020). Therefore, the conflict of interest between anthropogenic activities and Caspian seal conservation will intensify. Proactive actions are needed to conserve seal habitats, including the reservation of shallow water areas suitable for island formation and prompt, flexible protection of identified seal haul-outs, coupled with continuous monitoring of their abundance and distribution.

Overall, the water area around the Tyuleniy Islands is considered a key Caspian seal habitat, assigned the international IUCN status of Important Marine Mammal Area (IMMA) (IUCN, 2026) for the conservation of the Caspian seal population. This serves as an additional argument for the special protection of this area.

The present studies clearly demonstrate the influence of sea regression on seal haul-out locations. Given the conservation status of the Caspian seal as an endangered species at the apex of the food chain — and therefore an indicator species of the Caspian Sea ecosystem — continuous aerial surveying of seals at haul-out sites is essential for making timely decisions on identifying haul-outs and granting them special protection. Experience in organizing aerial surveys shows that survey conditions are not always favorable: for example, on March 8, 2020, a salt storm interfered with obtaining high-quality photographs of the haul-outs, and in spring 2021 and 2022, haul-outs had not yet formed and surveys could not be conducted in full — which is why these data were not included in the present study. Future survey planning should therefore schedule two or three surveys per season. Unfortunately, the primary limiting factor is a lack of adequate funding. Securing co-financing from interested organizations will be necessary to improve the coverage and objectivity of future surveys.

5 Conclusion

The Caspian Sea level is characterized by pronounced nonlinear dynamics: sharp declines alternate with periods of temporary stabilization and partial recovery. It is precisely this unpredictability that constitutes a source of ecological stress for the Caspian seal, as it generates instability in island habitats. The present study documents this stochastic reality rather than modelling the future.

The northeastern Caspian is of utmost importance for Caspian seal island haul-outs during the spring and autumn periods. The comprehensive analysis of geospatial data using Sentinel-2 and Landsat-8 satellite imagery and multi-year aerial survey materials allowed us to assess the scale of the Caspian Sea regression’s impact on the distribution of island seal haul-outs in the study area. We established that a sea-level drop of 2.28 m over the 1995–2023 period led to a westward coastline shift and a 15.76-fold reduction in island area. We determined that the evolution of islands in the shallow-water zone follows an “ Emergence – Growth – Merger with the mainland “ cycle. The ongoing acceleration of this cycle leads to the degradation of habitats suitable for animal haul-outs. The described phenomenon clearly demonstrates the ongoing ecosystem shift, of which the Caspian seal is an indicator. The shallow-water ecosystem transforms into a terrestrial mainland ecosystem, passing through the triad phase.

The loss of established haul-outs forces seals to alter their spatial distribution. The southern zone of seal haul-outs, which held the greatest value, is completely degrading due to the drying of Komsomolets Bay and the ongoing sea regression. It is predicted that the northern zone may also soon become shallow and lose its significance for seal haul-outs. Unlike the highly inaccessible shallow waters of the northeastern coast, new seal concentration sites are discovered approximately 179.5 km to the southwest, but they fall within a zone of high anthropogenic pressure. In the Tyuleniy Islands archipelago region, shipping and fishing are actively developed, and the exploitation of new oil and gas fields is planned. The relocation of seals to this region inevitably leads to an exacerbation of the conflict of interest between the industrial development of the shelf and the objectives of conserving the endemic species.

The Republic of Kazakhstan has established the state nature reserve “Kaspiy Itbalyqy” (Decree of the Government of the Republic of Kazakhstan No. 884, 2024) to conserve seal habitats, but there is a need to create ecological corridors along seal migration routes and implement prompt, flexible protection of identified haul-outs. Given the observed westward coastline shift, mass seal haul-out sites are already extending beyond the territorial waters of the Republic of Kazakhstan; therefore, coordination among other Caspian littoral states is necessary to designate these areas as specially protected key habitats of the Caspian seal, taking into account their IUCN status as an Important Marine Mammal Area (IMMA), and to establish a transboundary nature reserve.

References

• 1

  (2021). Ecological Code of the Republic of Kazakhstan, Law No. 400-VI ZRK, Dated January 2, 2021. Available online at: https://adilet.zan.kz/rus/docs/K2100000400 (Accessed May 20, 2026).

  • Google Scholar
  • View reference in article

• 2

  (2024). Geographic Median Image Compositing in Google Earth Engine. Available online at: https://gis.stackexchange.com/questions/444516/geometric-median-image-compositing-in-google-earth-engine (Accessed October 05, 2024).

  • Google Scholar
  • View reference in article

• 3

  (2010). International Study of the Caspian Seal - Report on Helicopter Flights in 2008-2009 (Atyrau: Agip Kazakhstan North Caspian Operating Company).

  • Google Scholar
  • View reference in article

• 4

  (2011). International Study of the Caspian Seal - Report on Helicopter Flight and Tagging of the Caspian Seal Using Satellite Telemetry in 2009-10 ( Agip Kazakhstan North Caspian Operating Company).

  • Google Scholar
  • View reference in article

• 5

  (2012). International Study of the Caspian Seal - Report on Helicopter Flight and Tagging of the Caspian Seal Using Satellite Telemetry, Tagging 2011-12 ( NCPOC).

  • Google Scholar
  • View reference in article

• 6

  (1985). Map of the caspian sea, scale 1:1,000,000 (Almaty: AZR RGKP "Yuzhgeodeziya").

  • Google Scholar
  • View reference in article

• 7

  (2020). Study of Habitat Conditions, Clarification of Patterns of Haul-Out Formation and Population Abundance of the Caspian Seal in the Kazakhstan Part of the Caspian Sea. (Report on Research Work 2018-2020) (Almaty: Institution "Institute of Hydrobiology and Ecology).

  • Google Scholar

• 8

  (2016). Water Balance and Fluctuations of the Caspian Sea Level. Modeling and Forecast (Moscow: Triada ltd).

  • Google Scholar

• 9

  BadamshinB. I. (1950). Some data on the island haul-outs of seals in the northern Caspian. Tr. Kasp. Basseynovogo Filiala VNIRO11, 201–221.

  • Google Scholar
  • View reference in article

• 10

  BaimukanovM. T.BaimukanovaA. M.BaimukanovT. T.IsbekovK. B.DauenevE. S.RyskulovS. E. (2020). “ Results of the census of Caspian seals (Pusa caspica) on island haul-outs in the Kazakhstan zone of the Caspian Sea in 2015–2018”, in: Marine Mammals of the Holarctic (Moscow: Marine Mammal Council), 2, 48–59.

  • Google Scholar
  • View reference in article

• 11

  BaimukanovM. T.RyskulovS. E. (2022). Methodological Guide for Determining the Abundance of the Caspian Seal (Pusa Caspica Gmelin, 1788) (Almaty).

  • Google Scholar
  • View reference in article

• 12

  BaimukanovM.ShagilbayevA.IskakovA.BaimukanovaA. (2025). Assessment of the fish diet of the Caspian seal (Pusa caspica Gmelin 1788) during haul-out periods on island sites: results, challenges, and perspectives. Front. Mar. Sci12, 2025. doi: 10.3389/fmars.2025.1565902

  • CrossRef
  • Google Scholar
  • View reference in article

• 13

  BaimukanovM. T.ShagilbayevA. U.RyskulovS. E.IskakovA. A.SydykovaZ. A.KuznetsovaT. V.et al. (2023). “ Main results and prospects of research on the Caspian seal (Pusa caspica Gmelin 1788) population during the haul-out periods in the Kazakhstan part of the Caspian Sea”, in: Zoological Research in Kazakhstan in the 21st Century: Results, Problems, and Prospects (Almaty: Institute of Zoology of the Republic of Kazakhstan), 715–722.

  • Google Scholar
  • View reference in article

• 14

  BaimukanovM. T.SirazhitdinovaM. K.BaimukanovaA. M. (2022). Atlas of Haul-Outs of the Caspian Seal (Pusa Caspica Gmelin, 1788) in the Kazakhstan Part of the Caspian Sea (Almaty: Institute of Hydrobiology and Ecology). Available online at: https://ihe.kz/ru/publikatsii (Accessed July 29, 2025).

  • Google Scholar
  • View reference in article

• 15

  BerlyantA. M. (2002). Cartography. Textbook for Universities (Moscow: Aspekt Press).

  • Google Scholar
  • View reference in article

• 16

  Bishop-TaylorR.SagarS.LymburnerL.AlamI.SixsmithJ. (2019). Sub-pixel waterline extraction: characterising accuracy and sensitivity to indices and spectra. Remote Sens.11, 2984. doi: 10.3390/rs11242984

  • CrossRef
  • Google Scholar

• 17

  BizikovV. A.ChernookV. I.SidorovV. K.ShipulinS. V.KlimovF. V.BelyaevV. A.et al. (2021). Assessment of the Caspian seal population abundance based on the results of instrumental aerial surveys on the ice in the northern part of the Caspian Sea in 2012, 2020, and 2021. Ispolzovanie i okhrana prirodnykh resursov168, 81–92.

  • Google Scholar
  • View reference in article

• 18

  BukharitsinP. I. (2019). Studies of Caspian Ice (Chisinau, Moldova: Palmarium Academic Publishing).

  • Google Scholar
  • View reference in article

• 19

  BukharitsinP. I. (2021). Drifting Ice, Hummocks, and Stamukhas of the Caspian Sea (Chisinau, Moldova: Palmarium Academic Publishing).

  • Google Scholar
  • View reference in article

• 20

  CourtR.LattuadaM.ShumeykoN.BaimukanovM.EybatovT.KaidarovaA.et al. (2025). Rapid decline of Caspian Sea level threatens ecosystem integrity, biodiversity protection, and human infrastructure. Commun. Earth Environ.6, 261. doi: 10.1038/s43247-025-02212-5

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 21

  DeakinR. E.HunterM. N.KarneyC. F. F. (2010). “ The gauss-krüger projection”, in: Proceedings of the 23rd Victorian Regional Survey Conference. ( Warrnambool, VIC, Australia: The Institution of Surveyors), 1–20. Available online at: http://www.mygeodesy.id.au/documents/Gauss-Krueger%20Warrnambool%20Conference%20V2.pdf (Accessed March 15, 2026).

  • Google Scholar
  • View reference in article

• 22

  Decree of the government of the republic of Kazakhstan no. 884 ( On the establishment of the republican state institution "State Nature Reserve 'Kaspiy Itbalyqy' of the Fisheries Committee of the Ministry of Agriculture of the Republic of Kazakhstan"). Available online at: https://adilet.zan.kz/rus/docs/P2400000884 (Accessed December 2, 2024).

  • Google Scholar
  • View reference in article

• 23

  DmitrievaL.JüssiM.JüssiI.KasymbekovY.VerevkinM.BaimukanovM.et al. (2016). Individual variation in seasonal movements and foraging strategies of a land-locked, ice-breeding pinniped. Mar. Ecol. Prog. Ser.554, 241–256. doi: 10.3354/meps11804

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 24

  DuanZ.WangG.HuJ.YuT.ChenS.ZhangY.et al. (2025). Spatiotemporal dynamics of northern Caspian shorelines, (1985–2023) and implications for coastal management: Lessons from the Aral Sea. PloS One20, e0325546. doi: 10.1371/journal.pone.0325546

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 25

  EybatovT. M.GadzhievD. V. (2022). Fossil and modern pinnipeds of Azerbaijan. ANAS Trans. Earth Sci.1, 106–118. doi: 10.33677/ggianas20220100077

  • CrossRef
  • Google Scholar
  • View reference in article

• 26

  GoodmanS.DmitrievaL. (2016). Pusa caspica. IUCN Red List Threatened Species 2016: e.T41669A45230700. doi: 10.2305/IUCN.UK.2016-1.RLTS.T41669A45230700.en

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 27

  HermosillaT.FranciniS.NicolauA. P.WulderM. A.WhiteJ. C.CoopsN. C.et al. (2024). “ Clouds and image compositing”, in: Cloud-Based Remote Sensing with Google Earth Engine (Cham: Springer). doi: 10.1007/978-3-031-26588-4_15

  • CrossRef
  • Google Scholar
  • View reference in article

• 28

  IUCN (2026). Caspian Seal Breeding Area Imma. Available online at: https://www.marinemammalhabitat.org/wp-content/uploads/imma-factsheets/BlackSeaTurkishStraitsCaspianSea/Caspian-Seal-Breeding-Area-BlackSeaTurkishStraitsCaspianSea.pdf (Accessed February 22, 2026).

  • Google Scholar
  • View reference in article

• 29

  KarelinG. S. (1883). Journeys of G. s. Karelin on the Caspian Sea (St. Petersburg: Typography of the Imperial Academy of Sciences), vol. 10.

  • Google Scholar
  • View reference in article

• 30

  KiselevichK. (1914). Excursion to the tsesarevich bay (Dead kultuk). Tr. Ikhtiol. Laboratorii3, 35–88.

  • Google Scholar
  • View reference in article

• 31

  LobanovV. A.NaurozbaevaZ. K. (2021). The Impact of Climate Change on the Ice Regime of the North Caspian (St. Petersburg: RGGMU).

  • Google Scholar
  • View reference in article

• 32

  MaoY.HarrisD.XieZ.PhinnS. (2021). Efficient measurement of large-scale decadal shoreline change with increased accuracy in tide-dominated coastal environments with Google Earth Engine. ISPRS J. Photogramm. Remote Sens.181, 385–399. doi: 10.1016/j.isprsjprs.2021.09.021

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 33

  MitinaN. N.MalashenkovB. M.TelitchenkoL. A. (2016). Underwater Landscapes of the North Caspian: Structure, Hydroecology, Conservation (Moscow: Tipografiya Rosselkhozakademii).

  • Google Scholar
  • View reference in article

• 34

  NaurozbaevaZ. K. (2025). “ Caspian seals on ice: monitoring and conservation under climate change”, in: Ecosystem of the Caspian Sea: Key Problems and Solutions (Moscow), 124–134.

  • Google Scholar

• 35

  Order of the Minister of Environment and Water Resources of the Republic of Kazakhstan No. 104-Ө (Accessed 02 December 2024). On Approval of the Rules for the Preparation of Biological Justification for the Use of Wildlife. Available online at: https://adilet.zan.kz/rus/docs/V1400009307 (Accessed December 02, 2024).

  • Google Scholar

• 36

  PanasenkoN. N.SinelshchikovA. V.YakovlevP. V. (2019). Technogenic risks of construction and operation of oil and gas complexes in the Caspian Sea. Vestn. AGTU. Ser. Morskaya tekhnika i tekhnologiya4, 46–59. doi: 10.24143/2073-1574-2019-4-46-59

  • CrossRef
  • Google Scholar

• 37

  PoletaevaE. V.PoletaevA. V. (2015). Study of the South Caspian crust from geophysical data. Uchenye zapiski Tambovskogo otdeleniya RoSMU3, 177–187.

  • Google Scholar

• 38

  PrangeM.WilkeT.WesselinghF. P. (2020). The other side of sea level change. Commun. Earth Environ.1, 1–4. doi: 10.1038/s43247-020-00044-z

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 39

  RobertsD.MuellerN.McintyreA. (2017). High-dimensional pixel composites from earth observation time series. IEEE Trans. Geosci. Remote Sens.55, 6254–6264. doi: 10.1109/TGRS.2017.2723644

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 40

  RustamovE. A.ShcherbinaA. A.BelousovaA. V.MammedovS. B. (2021). The state of the Caspian seal in the Turkmen sector of the Caspian Sea 2012-2021. Aktual. vopr. Zool. ekol. i okhrany prirody3, 133–138.

  • Google Scholar

• 41

  RychagovG. I. (2011). Fluctuations of the Caspian Sea level: causes, consequences, forecast. Vestn. Mosk. un-ta2, 4–12.

  • Google Scholar

• 42

  SadykovZ. S.GolubtsovV. V.KuandykovB. M. (1995). The Caspian Sea and its Coastal Zone (Natural Conditions and Ecological State) (Almaty).

  • Google Scholar

• 43

  Sentinel-Hub (2024). Eo Browser. Available online at: https://apps.sentinel-hub.com/eo-browser (Accessed May 12, 2024).

  • Google Scholar

• 44

  SunF.SunW.ChenJ.GongP. (2012). Comparison and improvement of methods for identifying waterbodies in remotely sensed imagery. Int. J. Remote Sens.33, 6854–6875. doi: 10.1080/01431161.2012.692829

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 45

  TimurzievA. I. (2020). Prospects of the North Caspian shelf based on the results of analyzing the distribution of oil and gas potential in the continental part of Western Kazakhstan. Geologiya nefti i gaza3, 29–41.

  • Google Scholar

• 46

  VosK.SplinterK.HarleyM.SimmonsJ.TurnerI. L. (2019). CoastSat: A Google Earth Engine-enabled Python toolkit to extract shorelines from publicly available satellite imagery. Environ. Model. Software122, 104528. doi: 10.1016/j.envsoft.2019.104528

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 47

  ZarJ. H. (2010). Biostatistical Analysis. 5th Edn (Upper Saddle River, NJ: Prentice-Hall).

  • Google Scholar