Figure 2 illustrates the spatial distribution changes in the Bumbat aufeis area. Bumbat aufeis typically occurs around the pond, with affected areas ranging from 0.008 to 0.029 km², showing variability in size. However, during the 2023–2024 period, aufeis was observed in the lower section of the pond, within the settlement area, encompassing an area of 0.044 km², 1.5 to 5.5 times larger than in previous years. During fieldwork, residents reported that the settlement area had never previously experienced aufeis, and several households required relocation because of the impact of this phenomenon.

Figure 3 illustrates the monthly changes in the area of the aufeis formation during the 2023–2024 period. The dynamics of Bumbat aufeis exhibit significant variation, with aufeis commencing in November and increasing progressively in size until it reaches its maximum extent in March. After reaching this peak, the ice begins to melt and melts completely by early May.

The locations of the ERT measurements conducted in the study area are shown in Figure 4. Resistivity values were recorded across all ERT profiles, ranging from 45 to 1094 Ω·m. Sections with resistivity values between 45 and 112 Ω·m are classified as moist alluvium, while those between 112 and 440 Ω·m represent low-moisture alluvium. Surface layers with resistivity values of 694 Ω·m or higher indicate permafrost. Deeper layers exhibiting similarly high resistivity values represent bedrock (Figure 5).

Profile 1 (A-A′) was conducted west to east, crossing the spring source along the spring stream (Figure 4). The results revealed three primary layers: moist alluvium, permafrost, and bedrock. A fault was identified between the permafrost and bedrock, coinciding with the location of the spring source (Figure 5A).

Profiles 2 (B-B′) and 3 (C-C′) were acquired in areas prone to aufeis formation, following the cross-section outlined in Figure 4. The results revealed two main layers: low-moisture alluvium and moist alluvium. In addition, Profile 3 identified two small permafrost layers (Figures 5B,C).

Profile 4 (D-D′) revealed three main layers: moist alluvium, low-moisture alluvium, and bedrock. In 2022–2023, aufeis also occurred in areas with low soil moisture content; however, aufeis was observed in moist zones during the 2023–2024 period (Figure 5D).

Profiles 5 (E-E′) and 6 (F-F′) identified three primary layers: moist alluvium, low-moisture alluvium, and bedrock. In Profile 6, a fault was detected between the bedrock layers, and during the 2023–2024 period, aufeis predominantly developed in areas with higher soil moisture contents (Figures 5E,F).

Borehole BH-1 was drilled in an aufeis-prone area with the following stratigraphy: (see borehole log in Figure 5G).

0–0.5 m: Black silty loam.

0.5–2.5 m: Yellow clay loam with fine-grained gravel.

2.5–6.0 m: Brown loam with medium-grained sand.

Borehole BH-2 was drilled in the non-aufeis area with the following stratigraphy: (see borehole log in Figure 5G).

0–2.5 m: Black silty loam.

2.5–3.5 m: Dark brown sandy clay loam with gravel.

3.5–6.0 m: Yellow clay loam with fine-grained gravel.

In BH-1, a 2 m-thick yellow clay loam with fine-grained gravel was identified between depths of 0.5 and 2.5 m. In BH-2, a similar yellow clay loam layer with fine-grained gravel was encountered at a depth of approximately 3.5 m. The upper 0–3.5 m in BH-2 consisted primarily of silty loam and sandy clay loam.

Figure 7 illustrates the mean daily GSTs, air temperatures, and freezing n-factor values. Data for GST-1 correspond to the GST recorded at the RX3000 weather station (Figure 1C). In contrast, the other data points represent measurements obtained using temperature data loggers installed in areas associated with aufeis formation (Figure 1C).

For GST-1, GST-2, and GST-3, the freezing n-factor values ranged from 0.05 to 0.11 (Figures 7A–C). This low n-factor rate can be attributed to the thermal insulation effect provided by snow cover. Snow cover depth, duration, and timing considerably influence the ground surface thermal regime in the winter months.

GST-4 exhibits a higher freezing n-factor value of 0.21, indicating a relatively stronger thermal interaction than the other points. Notably, between 14 December 2023, and 15 March 2024, a marked decrease in GSTs was observed, coinciding with the period during which the spring stream froze (Figure 7D).

Figure 8 illustrates the relationship between the aufeis area and several climatic factors, including MAAT, total annual precipitation, FDD_A, and TDD_A for the corresponding years. Upon examining the correlation between changes in aufeis area and annual mean air temperature, FDD_A, and TDD_A, a weak correlation was observed, with R ² values ranging from 0.0006 to 0.17, indicating that the air temperature regime does not directly influence the aufeis area (Figures 8A,C,D). In contrast, a moderate relationship was found between total annual precipitation and changes in the aufeis area, with an R ² value of 0.53 (Figure 8B), indicating that the extent of the aufeis area is influenced by the total amount of annual precipitation.

Figure 9A presents the total monthly and seasonal snowfall amounts for the winter seasons from 2009 to 2024. These data indicate considerable fluctuations in monthly snowfall, with certain years experiencing earlier snowfall (e.g., 2016 and 2023) and other years later snowfall (e.g., 2015). The total snowfall during this period ranged from 37 mm (2018–2019) up to 498 mm (2023–2024). Notably, the winter of 2023–2024 experienced snowfall amounts 1.8 to 13.5 times greater than those in previous years, marking an exceptional anomaly. Snowfall distribution during this winter showed considerable accumulation in November, followed by sustained high levels of snowfall in the subsequent months. This notable shift in the timing and quantity of snowfall is expected to have implications for the region’s seasonal hydrological and GST regimes. However, when comparing total snowfall during winter to the dynamics of the aufeis area, a correlation coefficient of R² = 0.31 was observed (Figure 9B), indicating that snowfall does not significantly impact fluctuations in the aufeis area.

Resistivity is strongly influenced by moisture content, with dry materials exhibiting significantly higher resistivity than moist or saturated ones (Telford et al., 1990). Based on ERT measurements (Figure 5), four distinct subsurface layers were identified. Sections with resistivity values between 45.0 Ω·m and 112 Ω·m are classified as moist alluvium. This is due to the presence of dissolved ions in pore water, which enhance electrical conductivity and thus reduce resistivity (Reynolds, 2011; Fitts, 2012). In contrast, sections with resistivity values between 112 Ω·m and 440 Ω·m correspond to low-moisture alluvium.

Surface layers with high resistivity values (>694 Ω·m) are interpreted as permafrost, since frozen ground lacks liquid water, thereby significantly reducing ionic conductivity (Hauck, 2002; Reynolds, 2011). According to the permafrost distribution map of Mongolia, the study area is located within a sporadic permafrost zone (Yamkhin et al., 2022). Previous studies have also documented permafrost at depths of 2–6 m in clay and clay loam soils within moist, swampy valleys of small rivers in Ulaanbaatar, including the Uliastai, Selbe, and Tolgoit Rivers (Dashtseren, 2008).

Subsurface layers at greater depths showing similarly high resistivity values are interpreted as bedrock. Dense igneous and metamorphic rocks—such as granite, basalt, and marble—typically exhibit high resistivity due to their low porosity and minimal water content (Palacky, 1987; Telford et al., 1990; Reynolds, 2011). Although permafrost and bedrock have similar resistivity values, their primary distinction lies in their depth of existence. According to our ERT results, bedrock is inferred to be present at depths ranging from 20 to 45 m below the surface (Figure 5), whereas permafrost typically occurs at much shallower depths of 2–6 m in this region (Dashtseren, 2008).

The results from ERT Profile 1 identified a fault between the permafrost and bedrock, coinciding with the location of the spring source (Figure 5A). In general, tectonic faults in hard rock terrain are common sources of groundwater. The chemical analysis showed relatively high-water hardness, confirming that the spring water originates from a groundwater source (Table 3; Todd and Mays, 2005; Fitts, 2012; Sengupta, 2013). The water temperature measurement findings indicate that the minimum temperature of spring water is 1.51°C, confirming the perennial spring (Figure 6A). During winter, the perennial spring water loses heat and freezes along the stream, contributing significantly to annual aufeis formation (Figures 6C,E; Harden et al., 1977; Turcotte et al., 2023).

Chemical analysis reveals that the hardness values of spring water, river water, and pond outlet water are 3.56 mg-eq/dm³, 1.72 mg-eq/dm³, and 1.80 mg-eq/dm³, respectively, indicating that the pond is fed by spring and river water (Table 3). The relatively high hardness of spring water shows that it has likely percolated through mineral-rich soils and rocks, dissolving calcium and magnesium ions and increasing water hardness (Todd and Mays, 2005; Fitts, 2012). The spatial distribution of aufeis (Figure 2) shows that, except for the winter of 2023–2024, aufeis predominantly occurs around the pond, indicating the contribution of river water to the aufeis process. However, during the winter of 2023–2024, aufeis was observed downstream of the pond, encompassing an area 1.5 to 5.5 times larger than in previous years (Figure 2).

The ERT surveys revealed that, in previous years, aufeis mainly occurred in areas with low soil moisture contents (Figures 5C,D), while in 2023–2024, it shifted to regions with relatively higher soil moisture contents (Figures 5D–F). We conducted electrical resistivity tomography (ERT) measurements along Profile 4 at the same location during the autumns of 2023 and 2024 (see Supplementary Material for ERT results). The total precipitation prior to the measurements differed between the two years, 338.5 mm in 2023 and 217.4 mm in 2024. However, the ERT results were similar for both years, suggesting that the soil moisture regime in the study area is more influenced by underlying geological conditions than by annual variations in precipitation.

According to the borehole data from BH-2, the subsurface consists of silty loam from 0 to 2.5 m depth, sandy clay loam from 2.5 to 3.5 m, and clay loam with fine-grained gravel from 3.5 to 6 m (Figure 5G). Since silty loam and sandy clay loam have higher permeability than clay loam, these upper layers (0–3.5 m) likely allow for greater water infiltration, resulting in the presence of moist alluvium. In contrast, the low-permeability clay loam layer found at depths of 3.5–6 m prevents downward water infiltration and retains water above it.

In BH-1, a low-permeability yellow clay loam layer was identified at depths of 0.5–2.5 m (Figure 5G). This layer likely impedes the downward movement of surface water, thereby reducing moisture content in the underlying strata and resulting in higher electrical resistivity values.

Aufeis in these areas may influence air and ground temperatures more significantly than snow-covered surfaces because of the higher thermal conductivity of ice than snow (Arenson et al., 2021). This enhanced heat transfer can lead to more rapid freezing of the soil. As the soil freezes, it undergoes expansion and creates upward pressure on unfrozen soil water. This water migrates through overlying, less dense soil layers and accumulates near the surface, contributing to the development of ground aufeis (Figure 10; Carey, 1973; Kane, 1981; Yu et al., 2005; Woo, 2012).

Drone imagery revealed that areas where soil water emerged were closely associated with pit latrines and the ground beneath yurts (Figure 10). The hollow spaces beneath pit latrines facilitated the upward movement of soil water toward the surface. Additionally, the ground temperature beneath yurts remained above freezing during the winter because of indoor heating, maintaining the ground temperatures above freezing, and allowing soil water to migrate toward the surface. Although this study did not directly observe overflow events, the differences in soil moisture have contributed to aufeis development by facilitating water migration toward the surface through pit latrines and thawed zones beneath yurts, thus expanding the area affected by aufeis formation (Figures 10, 11).

The analysis of aufeis dynamics and its correlation with climatic factors over the past years indicates that air temperature regimes have minimal impact on the extent of aufeis. In contrast, the amount of precipitation clearly influenced the changes in aufeis dynamics (Figure 8B). While several studies suggest that precipitation and air temperature regimes affect aufeis dynamics (Morse and Wolfe, 2015; Lainis et al., 2024), this discrepancy may stem from variations in aufeis sources. Specifically, aufeis formed by springs is less influenced by seasonal weather changes (Turcotte et al., 2023) because the thermal regime of groundwater is primarily controlled by underground conditions, rendering it less susceptible to direct changes in air temperature.

The analysis reveals a strong correlation between spring stream temperatures and air temperatures, with the spring streams freezing during specific winter periods (Figures 6C–F). Generally, the water emerging from perennial springs freezes along the stream under cold conditions unless it discharges directly into major rivers or lakes or rapidly infiltrates into the ground (Turcotte et al., 2023). When the spring water stream freezes, it blocks the flow of spring water into ponds, causing the water to surface and freeze, thus leading to aufeis (Carey, 1973; Yoshikawa et al., 2007; Ensom et al., 2020).

During the winter of 2023–2024, snowfall amounts were exceptionally high compared with previous years (Figure 9). The freezing n-factor values recorded at GST-1, GST-2, and GST-3 during this period ranged from 0.05 to 0.11, highlighting the insulating effect of snow cover (Figures 7A–C). Snow cover is critical in influencing the thermal regime between the atmosphere and the ground surface during the cold season (Zhang, 2005). In Mongolia, a snow layer 10–20 cm thick can increase the surface temperature by 4°C–6°C during winter (Sharkhuu and Anarmaa, 2015).

Consequently, spring water that surfaced onto unfrozen ground flowed along the snow-covered surfaces and accumulated in low-lying areas. Evidence of this phenomenon was observed on 4 January 2024, when drilling in the study area revealed water flowing beneath the snow cover (Figure 12). Additionally, data from the GST-4 surface temperature (Figure 7D) demonstrated a reduction in surface temperature when the spring stream was frozen, indicating that a decrease in surface temperature resulted from spring water flowing over the surface, potentially forming a thin aufeis layer. Drone surveys conducted on 15 April 2024, after the snow cover had melted (Figure 3), indicated that this thin layer of aufeis likely melted along with the snow cover.

These findings illustrate that thick snow cover can significantly alter the surface thermal regime, leading to changes in the spatial distribution of aufeis fed by perennial springs.

Finally, the thick snow cover during the winter of 2023–2024 altered the ground surface thermal regime. When the spring stream froze, spring water flowed beneath the snow and accumulated in low-lying areas of the settlement area, resulting in a shift in the spatial distribution of aufeis. Consequently, the ground thermal regime in the area changed, causing soil water to rise to the surface through the ground beneath yurts and pit latrines, thus expanding the aufeis area (Figures 10, 11).

Interviews with residents and satellite imagery analysis indicated that in previous years, aufeis near pond areas did not threaten nearby settlements (Figure 2). However, aufeis occurred within settlement areas during the winter of 2023–2024, significantly damaging many households and displacing numerous families, as their yurts, furniture, and belongings were rendered unusable by aufeis (Figures 10, 13).

Another serious concern linked to the Bumbat aufeis is NH₄ pollution. Chemical analyses of the aufeis samples revealed elevated NH₄ concentrations, particularly in the lower sections of the aufeis, where levels exceeded background concentrations (Table 3). Soil water migrating toward the surface through pit latrines contributed to an expansion in the aufeis area (Figure 10).

Aufeis does not typically exhibit elevated NH₄ levels. In some regions of High Mountain Asia, aufeis is actively utilized in local irrigation systems and is not considered problematic (Schmidt and Nüsser, 2017; Brombierstäudl et al., 2024). Nevertheless, in this particular case, the aufeis is fed by sewage-contaminated water due to inadequate sanitation infrastructure. Wastewater from human activities accumulates in pit latrines, increasing NH₄ concentrations in the soil water (Oyunbat et al., 2022), which contributes to aufeis formation and expansion.

As a result, aufeis polluted with NH₄ poses a health risk to residents in the affected areas and the broader population. High NH₄ concentrations are known to cause respiratory irritation and inflammation (Roney and Fernando, 2016) and can increase the risk of asthma in preschool children (Holst et al., 2018).

This study utilized satellite imagery data to examine the natural factors influencing changes in the spatial distribution of Bumbat aufeis over the past 15 years. Due to the inherent limitations associated with satellite products, particularly cloud and snow cover, estimating the aufeis distribution for 2012, 2015, 2016, 2017, 2021, and 2022 could not be performed.

Additionally, drone surveys provided high-accuracy measurements of the aufeis area and thickness. However, when estimating the thickness of the aufeis using a DEM generated from drone imagery, distinguishing the snow cover from the underlying ice was impossible. This limitation reduced the accuracy of the aufeis thickness measurements. Nevertheless, since this study focused on changes in the aufeis area rather than its thickness, the impact of this issue on the overall research findings was minimal.

In this study, the Wenner-Schlumberger array was selected for ERT measurements due to its balanced sensitivity to both vertical and horizontal resistivity variations (Pazdirek and Bláha, 1996), making it suitable for characterizing subsurface structures related to aufeis. However, this array has moderate sensitivity to shallow features compared to other configurations, such as dipole-dipole (Dahlin and Zhou, 2004), which may limit the resolution of near-surface moisture variations important for aufeis formation.

Additionally, the resolution of deep subsurface structures may be limited by the increasing electrode spacing, particularly in areas with heterogeneous geology. This may affect the accuracy of delineating the boundary between permafrost and bedrock, especially when both materials exhibit similarly high resistivity values. Furthermore, inversion artifacts can occur depending on the subsurface conditions and data quality, potentially leading to misinterpretation of resistivity anomalies. In our case, the interpretation of high-resistivity layers at depths of 20–45 m as bedrock and those at 2–6 m as permafrost is based on a combination of resistivity values and prior studies.

Future studies may benefit from the integration of multiple arrays (e.g., combining Wenner-Schlumberger with dipole-dipole) or three-dimensional ERT surveys to improve spatial resolution and reduce interpretational ambiguity.

Despite these limitations, the authors believe this research study provides valuable insights into the dynamics of aufeis processes in this region.