Between 2006 and 2017, ground temperature monitoring was systematically initiated at eight industrial project sites located between 27 and 34° S and within approximately 25 km of the Chilean–Argentine border (Fig. 1). The compiled dataset presented in this study includes measurements collected from 53 boreholes distributed across the region, with 27 located in Chile and 26 in Argentina. The boreholes were installed to depths varying from 10 to 100 m at surface elevations ranging from 3625 to 5251 m, in areas with and without permafrost. Of the total, 30 boreholes intercepted permafrost, while the remaining 23 were installed within unfrozen or non-cryotic ground (Fig. 2) subject only to seasonal freeze–thaw. Due to data interruptions and short monitoring duration, some boreholes in cryotic ground did not meet the 2-year criterion of continuous ground temperature monitoring to confirm the presence of permafrost (van Everdingen, 1998). However, given that measurements were available from below the DZAA (i.e., from depths ≥ 10 m), it is reasonable to infer the presence of permafrost in boreholes where the temperature is at or below 0 °C, even without a complete 2-year record. Surface morphologies mapped at the borehole locations during thermistor installation include bedrock, colluvium, rock glaciers (containing ground ice) and landslide deposits (Table S1). Except for those in rock glaciers, the boreholes typically intercept bedrock within ∼ 10–20 m of the ground surface.

Ground temperatures were monitored along the profile of each borehole using negative temperature coefficient (NTC) thermistor strings (models YSI 4400, RST TH00 or Geoprecision TNode series). Thermistors were accurate within a range of ± 0.1 to ± 0.5 °C from base temperature, with precision ranging between 0.1 and 0.25 °C. Sensors were positioned at varying depths, starting from the ground surface (0 m) and spaced from 0.5 to 15 m (depending on the objectives and maximum depth of the borehole). Loggers were used for data collection at most locations, although manual readings were made on occasion. The frequency of data collection varied with location, ranging from hourly to daily measurements.

The duration of monitoring varied by borehole, ranging from less than 1 month to 9 years (Fig. 2). Several locations experienced interruptions to data collection due to factors such as electrical storms, instrument malfunctions, inaccessibility for download or maintenance due to remoteness, adverse weather, slope instability, and/or changing regulatory requirements. In some cases, interruptions occurred simply because a borehole was temporarily not being monitored as part of the project's objectives. A comprehensive discussion of data collection challenges, monitoring gaps, known data artifacts and filtering rationale is included in Sect. 3.2. Gaps in individual monitoring records were filled as outlined in Sect. 3.3 prior to interpretation of the data. Raw data for each borehole are presented in the Supplement (Figs. S1–53).

Long-term operation of borehole infrastructure in high-mountain regions can be hindered by natural factors including erosion, mass movements or meteorological effects, all of which may lead to instrument malfunction that can reduce data quality or create gaps in monitoring records. Challenging terrain, remoteness and lack of financial support can limit site access, leading to poor instrument maintenance and data loss. Other factors affecting ground temperature monitoring unrelated to mountainous regions may include instrument damage by wildlife, vandalism, construction or damage during instrument installation.

Many of these challenges were encountered across the study area, with most interferences or malfunctions identified by field staff before analysis. The most common causes of data interruptions were battery power loss, faulty connections or sensor failures between maintenance visits. Irregular funding for ground temperature monitoring based on specific project needs led to inconsistent and unpredictable field work and maintenance schedules, making it challenging to perform routine upkeep (e.g., battery replacement of sensor repairs) proactively or in a timely manner.

Drilling activities were frequently identified as a source of interference to monitoring. While several thermistors were installed in pre-existing exploration boreholes, more than half of the locations were drilled explicitly for ground temperature monitoring. This included all boreholes at Sites 1, 4, 6 and 8, as well as boreholes 3-8, 3-11, 7-4, 7-5, 7-6 and 8-3. Dry sonic methods were used to drill all boreholes except at Site 1, which were advanced by diamond coring using water or polymer fluid. Early temperature measurements in these boreholes were likely overestimated due to thermal disturbances from the drilling process, especially where fluid circulation was involved. Elevated temperatures at Site 1 are evident at the onset of monitoring, particularly from depths of 10 m and below (Figs. S1–9). Although temperatures gradually decrease as drilling disturbances dissipated, it remains uncertain whether ground temperatures at Site 1 had fully stabilized within the roughly 1-year monitoring period. Early measurements from Site 1 and other locations with approximately 1 year of data (i.e., boreholes 6-8 and 6-11, 7-4, 7-5 and 7-6) were treated with caution in the analyses, particularly those collected during the initial 2 to 3 months, as they are unlikely to represent average ground temperature conditions. Drilling interferences were less of a concern for boreholes 3-8, 3-11 and 8-3 and most boreholes at Site 6, which have longer monitored records.

At several monitoring locations, changes in ground conditions may have compromised measurement quality. For example, construction activities that altered local runoff patterns caused a large gully to form near borehole 5-1 (Fig. 3). Field staff first noted the gully in 2017, but anomalies in the data and multiple sensor failures suggested that erosional activity had affected borehole temperatures since late 2015. Consequently, measurements beyond September 2015 were excluded from analysis. Another instance of ground disturbance influencing monitoring was noted during a 2017 field visit, where a previously unseen debris flow was observed near borehole 2-2 (Fig. 4). As with borehole 5-1, erroneous data were identified and excluded from analyses.

Project-driven optimization of monitoring sometimes led to data interruptions (or shortened record lengths) and informed additional filtering. For example, data collection ceased at select boreholes at Site 2 once permafrost was determined to be absent, and thermistor strings were relocated to higher elevations where encountering permafrost was more likely. Two thermistors were moved (from boreholes 2-4 and 2-5 to boreholes 2-6 and 2-7), resulting in the termination of monitoring at the original locations. During the relocation of the thermistor from borehole 2-4 to 2-6, it became evident that the sensor at ∼ 24 m depth was damaged. Initially, anomalously high measurements recorded by this sensor at its original location were considered plausible, possibly due to warm groundwater or exothermic reactions at depth. However, similar anomalies persisted at the new location (borehole 2-6), suggesting measurements were artifacts and should be excluded from analysis.

The Supplement accompanying this paper displays all raw data except for cases where artifacts were confidently identified and removed (e.g., related to erosion events or known instrument failures). Thermal disturbances from drilling and occasional unexplained artifacts are visible in the figures but were omitted from analyses. These anomalies could be the result of water or air flow through blocky materials or electrical storms, which are common in central Chile (e.g., Montana et al., 2021). As there was no clear indication that they were erroneous, these anomalies were retained on the plots for transparency and dataset completeness.

Gaps in ground temperature time series were interpolated using non-linear least squares regression to enhance data visualization and remove seasonal bias from analysis. At thermistors that showed seasonal variation, a sinusoidal function with a superimposed linear trend was fit to filtered data to approximate seasonal and longer-term temperature variations. For sensors located below the DZAA, linear interpolation was used. Initial conditions for each sensor were specified by setting function parameters (i.e., period, amplitude, phase shift, offset, slope) that produced a reasonable visual match to observed data. Parameters were then optimized using the generalized reduced gradient (GRG) non-linear solver in Microsoft Excel to minimize the sum of squared residuals, targeting a normalized root mean square (NRMS) below 10 %. The quality of fit was first assessed visually, with manual adjustments made occasionally to improve the overall match of the solution to the data. Missing values were then filled on a daily time step using the fitted equation and are plotted alongside raw data in the Supplement (Fig. S1 through Fig. S53).

It is noted that this approach has limitations in estimating temperatures within the active layer or in boreholes containing ground ice, as it cannot represent latent heat effects during phase changes. Additionally, it is not suitable for long records with complex warming or cooling trends, which may vary over the monitoring period. Despite these limitations, the method provided a useful way to estimate missing values and visualize seasonal patterns and short-term variations in the data, making it appropriate for this study, where data records at each instrument are less than 20 years.

A total of 22 permafrost and 12 non-permafrost boreholes have continuous records spanning at least 1 year, making them suitable for examining seasonal temperature variations (Fig. 5). Most sensors near 10 m depth (Fig. 5a) show clear fluctuations, with greater seasonal amplitudes in unfrozen ground (∼ 0.5 to 1 °C) compared to frozen ground (< 0.5 °C). At depths near 20 m (Fig. 5b), sensors are mostly below the DZAA, with some exceptions in non-cryotic ground. Seasonal variations at both depth horizons are less pronounced where temperatures hover around 0 °C, reflecting latent heat effects from annual freezing and thawing of ground ice. Similarly, attenuated temperature fluctuations at shallow depths (< 2 m) were noted during winter at 14 locations (Table S1, Fig. S1 through Fig. S53) and are likely attributable to snow cover.

Unlike permafrost regions in the Northern Hemisphere, identification of any consistent trend of rising ground temperatures within the Andean dataset to date is not possible due to its short duration. It instead reflects baseline local topo-climatic conditions and short-term climate fluctuations associated with the mountainous terrain. Both short-term warming and cooling are noted in the dataset, with no correlation to location, altitude or surface substrate (Fig. 5). Figure 6 illustrates the wide variation in short-term rates of temperature change from locations with at least 2 years of data (19 permafrost and 9 non-permafrost boreholes). Warming (0 to 0.05 °C yr⁻¹) was noted in approximately half of the boreholes examined (15), with the rest showing short-term cooling. Non-permafrost boreholes exhibited greater variability in temperature change over the period of record.

Figure 6 includes representative long-term warming trends from other permafrost regions, as documented in Smith et al. (2022). This includes continuous or “cold” Arctic permafrost (temperatures below -2 °C, warming rates from ∼ 0.04 to 0.11 °C yr⁻¹, monitored since the 1980s), discontinuous or “warm” Arctic permafrost (temperatures between -2 and 0 °C, warming rates from∼ 0.01 to 0.05 °C yr⁻¹, monitored since the late 1970s to early 1980s), and mountain permafrost within the Swiss Alps (average: ∼ 0.02 °C yr⁻¹, monitored since in the late 1980s to early 1990s). These trends are presented for reference only, but it is notable that the short-term warming rates in the Andes align with trends estimated for the Swiss Alps and the warm Arctic regions. However, when making this reference we strongly caution the reader, as Northern Hemisphere studies rely on significantly longer (decadal or multi-decadal) datasets than the limited records in this study. The longest record in the Andean analysis was approximately 9 years (borehole 3-11), with most spanning between 2 and 7 years, making reliable comparisons with Northern Hemisphere studies premature. To establish causal relationships between ground temperatures and long-term climate variability in the Andes, ongoing monitoring of ground temperatures alongside local climate variables is needed. Given the constraints of the current monitoring dataset, a comprehensive analysis of this scale is beyond the scope of this study and presents an opportunity for climate researchers.

ALT in permafrost boreholes was estimated by linearly interpolating measurements between thermistors above and below the zero-degree isotherm at the time of maximum annual thaw. While this approach may slightly overestimate ALT (e.g., Riseborough, 2008), a lack of shallow temperature measurements at the Andes sites prevents reliable extrapolation of the zero-degree isotherm from above. Additionally, snow cover varies considerably across boreholes; in areas with little snow, atmospheric gradients strongly influence temperatures in the active layer, whereas snow-covered areas insulate the ground (e.g., basal temperature of snow (BTS) method by Haeberli, 1978). Given these complexities and the fact that the goal was to track potential changes in permafrost depth and compare boreholes, linear interpolation was considered a reasonable approach, allowing for relative comparisons rather than estimating absolute ALT values.

Of the 30 permafrost boreholes, 20 were considered in this analysis. Two rock glaciers (boreholes 6-1 and 6-4) were assessed as depth to permafrost table, as the depth to permafrost in these boreholes exceeded the maximum annual freeze–thaw depth. For boreholes 1-2 and 1-8, which lacked 2 consecutive years of data, the maximum annual thaw depth was estimated in a similar manner to ALT (or permafrost table), as the records encompassed a complete freeze–thaw cycle with subsequent refreezing. Permafrost locations with shorter monitoring records were not considered. Two permafrost locations (boreholes 3-7 and 3-12) were also excluded because the depth of thaw penetration remained above the shallowest sensors (< 1 m deep).

Figure 7 shows that thaw depth and ALT typically range from ∼ 0.5 to less than 4 m, although ALT at boreholes 3-5 and 6-2 reached depths ≥ 6 m. Consistent with field observations of advanced degradation in these landforms (though not necessarily representative of the region), depth to permafrost within rock glaciers is the highest in the dataset, ranging from 7.3 to 17.4 m in 2021. There is no consistent increase in ALT over time across the dataset, and in some locations, the active layer may be shallowing (e.g., boreholes 3-6 and 3-9, and possibly 210). In contrast, the top of permafrost is deepening in rock glaciers, at rates of approximately 0.4 m yr⁻¹ at borehole 6-1, 0.8 m yr⁻¹ at borehole 6-2 and 1.5 m yr⁻¹ at borehole 6-4. These results reflect short-term fluctuations in climate.

Contour diagrams of borehole temperature evolution with time (Fig. 8) reveal supra-permafrost taliks at boreholes 6-1 and 6-4. At borehole 6-4, the top of permafrost was fully decoupled from the active layer throughout monitoring. In contrast, the formation of the talik at borehole 6-1 began forming in mid-2019.

The temperature profile within a borehole is shaped by factors such as regional geothermal heat flux, lithology variations, local topo-climatic variations and historical fluctuations in ground surface temperatures. These factors result in a wide range in ground temperatures (from approximately -7 to 7 °C) and varied profile curvatures (Fig. 9), reflecting the complex thermal landscape of the Andes. Variations in profile curvature within individual boreholes suggest ground temperatures are not in equilibrium with modern climate conditions. Instead, they represent the present balance between surface and geothermal heat fluxes controlled by thermal properties of the ground, which vary with location and depth. Borehole temperature gradients within the upper 30 m are both positive (warming with depth) and negative (cooling with depth), with some locations showing nearly isothermal conditions. Several boreholes in warm permafrost (3-5, 6-2, 6-5, 6-8, 6-11, 8-3) exhibit a composite temperature profile within the upper 30 m, characterized by a nearly isothermal region close to 0 °C, below which temperatures increase with depth due to the geothermal heat flux. The temperature profiles shown in Fig. S1 through Fig. S53 reveal the presence of thin permafrost layers in several rock glaciers, ranging from approximately 2 to 40 m thick. Permafrost thickness in boreholes that did not intercept the base of permafrost (but exhibited warming with depth) was estimated to range from 40 to > 500 m, based on the projection of thermal gradients to the zero-degree depth intercept.

There is no clear relationship between profile shape or permafrost thickness and latitude (i.e., site number), likely due to the wide variation in ground elevations and surface slope orientations within a few tens of metres at each project site (Table S1 and Sect. 3.3). Such topographic heterogeneity results in significant variations in surface solar radiation and microclimatic conditions, which have a greater influence on ground temperatures than latitude. There also does not appear to be a relationship between profile curvature or permafrost thickness with surface morphology, likely because most boreholes intercept bedrock at depths of 10–20 m (BGC Engineering Inc., personal communication, 2023). As such, temperature profiles generally reflect thermal properties of shallow bedrock, with minimal influence from surface substrate.

Potential exceptions to this are noted in rock glaciers at Site 6 (i.e., boreholes 6-2, 6-5, 6-8 and 6-11), which, as noted above, are characterized by shallow isothermal conditions near ∼ 0 °C then increasing temperatures with depth. The coexistence of air and ice in the pore space near the ground surface and within the active layer of these coarse blocky landforms results in a significantly lower thermal conductivity compared to bedrock. In addition, the relatively large pore space promotes air convection, which can significantly cool the active layer, leading to temperatures ∼ 1 °C colder than without convection (e.g., Wicky and Hauck, 2020). Below the active layer, the ice–rock mixture is likely to be less sensitive to variations in the near-surface thermal regime (although still distinct from bedrock), leading to a unique profile shape despite possibly similar thermal histories at the ground surface. As these particular landforms are known to contain ice-rich permafrost near the phase change temperature, vertical segments of the profiles may reflect the melting of ground ice, slowed due to latent heat effects.

Thermal gradients of individual boreholes provide insight into the recent thermal evolution of the ground in response to changing surface temperatures. The attenuation of historical surface temperature anomalies may be indicated by warm (cool) deviations of shallow borehole temperatures from the linear trajectory of the deep gradient. This temperature deviation (or offset) was estimated as the difference between projected surface temperatures from shallow (< 30 m) and deep (30–100 m) gradients. Offsets estimated in this depth range reflect recent decade-scale shifts in surface temperatures, assuming equilibrium conditions and uniform thermal properties with depth (e.g., Lachenbruch and Marshall, 1986). Since thermal responses to surface warming at depths greater than ∼ 100 m likely lag several decades, the offset provides a first approximation of secular temperature changes with time.

A total of 19 boreholes (10 permafrost and 9 non-permafrost boreholes) in the Andean dataset extend to depths beyond 30 m and were considered in this analysis. Rock glaciers were excluded due to their unique thermal profiles (Sect. 4.3), as were boreholes 2-4 and 2-6 due to erroneous measurements at ∼ 24 m (Sect. 3.2). Thermal gradients were estimated from the linear segments of profiles at intermediate depths (below the DZAA to 30 m) and from 30 m to the final depth of each borehole, except at borehole 8-1. The shallower gradient at borehole 8-1 was estimated from the DZAA to 60 m and the deep gradient from 60 m onward due to a slight decline observed in the thermal gradient at approximately this depth (Fig. 9).

Results of the gradient analysis (Fig. 10) are presented in a similar manner to the temperature changes derived from the interpolation of time-series data (Fig. 6) for conceptual comparison of the two analyses. As with the time-series analysis, the gradient analysis indicates both warming and cooling in recent history, with greater variability in the non-permafrost boreholes. Warm-side deviations from deep thermal gradients are evident for 14 of the boreholes (7 permafrost and 7 non-permafrost) and range from 0.05 to 0.9 °C. The remaining boreholes (2 permafrost and 3 non-permafrost) indicate cooling in recent decades, with offsets ranging from approximately -0.05 to -1.08 °C. It is emphasized that these estimates are derived from many simplifying assumptions and that three-dimensional analysis accounting for ground thermal properties is necessary for a more accurate understanding of near-surface ground temperature changes in recent history.

Representative ground temperatures across the study region show no correlation with latitude, but significant variation occurs over short lateral distances at the site level (Fig. 11). Transitions from cold to warm permafrost (< -2 to 0 °C) and/or to non-cryotic ground occurs with 1 km, highlighting the greater influence of catchment-scale topo-climatic variability over regional climate. Ground temperatures generally decrease with increasing altitude, showing no clear relationship with surface morphology, as most measurements represent thermal conditions of shallow bedrock (Fig. 12a). Rock glaciers, however, consistently show lower temperatures than boreholes at similar elevations (< 0 °C in rock glaciers vs. 2–6 °C in other boreholes). They also mark the lowest altitudinal occurrence of permafrost within the dataset (slightly below 3600 m) due to a sustained presence of ground ice compared to the broader dataset. Excluding rock glaciers, ground temperature correlates more strongly with altitude at permafrost boreholes (r2 = 0.65) than for the complete dataset (r2=0.53), with lapse rates of approximately -4.3 and -5.7 °C km⁻¹, respectively.

Consistent with cooler temperatures at higher altitudes, both depth to permafrost and permafrost thickness decrease as elevation increases (Fig. 12b and c). Again, no correlation with surface morphology is evident except in rock glaciers, which show the most variable depth to permafrost and lowest estimated thickness within the dataset, reflecting the advanced state of degradation of these landforms (e.g., supra-permafrost talik at boreholes 6-1 and 6-4, Fig. 8). Excluding rock glaciers, the depth to top of permafrost decreases by approximately 1.9 m km⁻¹ elevation gain (r2 = 0.30), while permafrost thickness increases at approximately 300 m km⁻¹ (r2=0.45).

Incorporating slope aspect into the analysis reveals that the zero-degree isotherm for ground temperature occurs at higher elevations on northeast-facing slopes, ranging from below 3700 m (within rock glaciers) to approximately 5000 m (Fig. 13a). This asymmetry is partly due to variations in average incident solar radiation with aspect (Fig. 13b), and more broadly to the geographical position of the study area within the Southern Hemisphere. Coincidentally, all monitoring locations within rock glaciers are on south-facing slopes, where lower solar radiation may help sustain permafrost in these landforms.

This study provides new insights into permafrost dynamics in the Central Andes, some which were previously hypothesized based on studies from other permafrost regions but lacked sufficient data to confirm their broader relevance to South America. It represents the largest and most regionally extensive compilation of ground temperature data from high-altitude (> 3500 m) sites with permafrost in South America to date, filling a critical knowledge gap in permafrost research. Using data from 53 boreholes (10 to 100 m deep) within permafrost and non-permafrost zones, the compilation provides a unique snapshot of ground thermal state within the Andean Cordillera of Chile and Argentina. Adequate depth of monitoring throughout the dataset enables characterization of average temperatures below the DZAA, visualization of borehole thermal gradients, and first-order estimates of the permafrost thickness and depth. This level of insight surpasses that of the few existing monitoring studies in the Andes, which generally are limited to boreholes only a few metres deep (Table 1).

An important implication of this work is that the data can be used to validate existing permafrost distribution models in the region, which were previously developed without any borehole temperature data (e.g., Arenson and Jakob, 2010; Ruiz and Trombotto, 2012). Additionally, having ground temperature data within permafrost and non-permafrost zones helps reduce the risk of site-selection bias towards permafrost presence, which can complicate evaluations of spatially distributed models predicting permafrost. Beyond validating existing permafrost distribution models in the region, data from this study may be utilized to support upscaling endeavors like those of Mathys et al. (2022), to quantify ice content in Andean permafrost regions and inform future water resource planning amidst the challenges posed by climate change.

Several important insights can be derived from the presented dataset that broaden the understanding of permafrost thermal state in both the Andes and the global permafrost context. Some of these findings are consistent with observations in other permafrost regions, while others may uniquely represent the Central Andes or, more precisely, the region within the altitudinal and latitudinal constraints of the dataset (i.e., 3625 to 5251 m and 27–34° S). These are discussed further in the sections that follow.

The Andean data compilation shows no consistent trend of warming in recent history. Instead, warming and cooling of the ground are inferred from time series (Sect. 4.1) and thermal gradient analysis (Sect. 4.4). Limited by the short monitoring period (< 10 years), the time-series analysis captures short-term fluctuations driven by seasonal and interannual climate variability and the study area's mountainous topography. While thermal gradient analysis may offer more reliable decadal-scale estimates of ground temperature change, its failure to show a consistent trend could stem from oversimplified assumptions about borehole geology and varying climatic conditions across locations.

Due the short record length, geologic complexity and topo-climatic variability between sites, it is not possible to determine whether the ground thermal regime in the study area follows long-term warming trends observed in other permafrost regions. However, it is worth noting that similar deviations from long-term trends have been observed in global datasets and are linked to short-term local meteorological influences (e.g., Biskaborn et al., 2019; Etzelmüller et al., 2020; Haberkorn et al., 2021). Localized changes in snow cover, vegetation and soil moisture content, especially in mountain environments, are known to significantly impact the thermal state of the ground, as evidenced by periods of permafrost cooling in the Alps in 2016 and 2023, attributed to anomalously low snow conditions during those years (PERMOS, 2023). Also in the Alps, summer heat waves during 2016 and 2019 were shown to correlate with short-term increases in active layer thickness (PERMOS, 2023), and water percolation has been linked to accelerated permafrost degradation (Luethi et al., 2017). Interannual variability in snow cover is inferred from near-surface isothermal conditions within the active layer at several of our Andean boreholes (Fig. S1 through Fig. S53). The resulting irregular insulation of the ground during winters, along with inconsistent infiltration of meltwater in the spring, likely contributes to both temporal and spatial variability of ground temperatures in the dataset, although latent heat released from the annual freezing and thawing of porewater may also play a role. The significant dryness of the Central Andes compared to other mountain environments, which has been exacerbated in recent years by the megadrought (Garreaud et al., 2020), also leads to comparatively less water available from snowmelt to infiltrate the ground and promote permafrost degradation. Long-term warming trends may also be obscured by temporary, localized surface temperature inversions, which can cause air temperatures at lower elevations to be cooler than expected, and vice versa. Such inversions have been shown to create unique permafrost conditions in near-proximity dissimilar valleys in Yukon, Canada, where ground temperatures at some high-altitude sites are significantly warmer than those at adjacent valley bottoms (Lewkowicz et al., 2012; Noad and Bonnaventure, 2023).

At a more regional scale, the combined influence of SAM, ENSO and PDO on South American climate patterns (e.g., Mantua and Hare, 2002; Montecinos and Aceituno, 2003; Vuille et al., 2015; Saavedra et al., 2018; Garreaud et al., 2020; Yoshikawa et al., 2020; González-Reyes et al., 2020; King et al., 2023) can temporarily influence the ground thermal regime across the study area. This would occur primarily through changes to spatial and temporal snowpack distribution, snow–albedo feedback, and variations in water infiltration that impact latent heat absorption. Although temperature and precipitation patterns were directly related to SAM, ENSO and PDO, and their potential impacts to the ground thermal regime were not examined in this paper, some patterns are apparent for the Central Andes and may be applicable to the study area. This includes temporary cooling trends observed in shallow (2 m deep) ground temperatures simultaneously with deceleration of several rock glaciers at a site with permafrost in central Chile (within the range of this study) aligning with lower MAATs during the same period (2010–2015). These cooling trends have been linked to the predominance of La Niña and neutral ENSO conditions since 2009 (Vivero et al., 2021).

With a monitoring record that currently falls short of the ideal length to assess impacts of atmospheric warming on ground temperatures (i.e., 20 years or more), air temperature data collected in the study region may offer complementary insights to the observations presented in this work. A summary of MAATs in mountainous regions in South America by Hock et al. (2019), which is based on very limited monitoring data, indicates lower warming rates or even slight cooling trends when compared to global studies. One meteorological station located at Site 3 (∼ 5000 m, ∼ 29° S) demonstrates relatively stable air temperatures over 20 years of monitoring (1999–2021, Fig. 14). This apparent stability and/or slight lowering of MAATs in South America suggests that the trajectory of ground temperatures in the Andes is likely to be unique from other regions that show clear signs of warming. The unique topo-climatic attributes of the Andean cryosphere – characterized by arid conditions, high solar radiation, minimal vegetative cover and organic matter, and less massive ice (except for rock glaciers) together with mountain topography – may expedite energy transfer and reduce latency of temperature change compared to other permafrost regions, in both mountains and the Arctic. Regardless of the specific mechanisms driving ground temperature evolution in the Andes, the preceding discussions emphasize the importance of continued monitoring to establish robust connections with climate change. This effort must also consider local climate and geographic conditions, as well as relationships with oceanic phenomena.

Although the topo-climatic conditions of the Andes are unique among permafrost zones, the data compiled in this study reveal several characteristics shared with other mountain environments (e.g., Etzelmüller et al., 2020; PERMOS, 2023). High spatial heterogeneity in ground temperatures, correlations with altitude and slope aspect, and the distinct thermal characteristics of rock glaciers compared to other landforms suggest that similar processes are shaping the current and evolving ground thermal regime in the Central Andes.

Significant variations in ground temperature over short horizontal distances – shifting from cold to warm permafrost (< -2 to 0 °C) and/or non-permafrost ground within less than 5 km – emphasize the strong microclimatic influence on the distribution of permafrost at catchment level over regional climate stressors. The pattern of decreasing ground temperatures, greater depth to permafrost and increasing permafrost thicknesses with increasing altitude reflects the orographic influence of air temperatures, a common feature of mountainous regions.

The spatial pattern of ground temperatures with respect to slope aspect reflects the asymmetry of incident solar radiation of the geographic region, resulting in a higher elevation of the zero-degree ground temperature isotherm on northeast-facing slopes. These characteristics were previously inferred for the Central Andes based on regional climate and mapping studies (e.g., Saito et al., 2016) and general knowledge of mountain permafrost environments (e.g., Haeberli, 1973). However, until now, these assumptions were not extensively validated with ground temperature data specific to the Andes.

Rock glaciers emerge as thermally unique permafrost landforms in the Andean dataset, a characteristic that has been widely observed in other mountain permafrost regions (e.g., Barsch, 1977). This uniqueness arises from their coarse blocky morphology and high porosity, which enable substantial variations in air, water and ground ice content – factors known to significantly influence ground temperatures, both spatially and temporally. This effect is particularly pronounced when rock glaciers contain substantial ground ice near the phase change temperature (e.g., Haeberli et al., 2006). Ground temperatures within rock glaciers in the Andean dataset do not follow general altitudinal relationships of other boreholes – instead, they consistently show lower temperatures than other boreholes at similar elevations, likely due to reduced thermal conductivity, convective cooling in the active layer and the sustained presence of ground ice from latent heat. Depth to permafrost and permafrost thickness in rock glaciers also deviate from the broader altitudinal relationships observed in the dataset, exhibiting the greatest and most variable depths to permafrost and lowest permafrost thickness within the dataset. Within rock glaciers, permafrost depth was estimated to be > 7 m and permafrost thicknesses ranged between 2 and 40 m (compared to depths < 4 m and thicknesses typically > 40 m in other boreholes). The formation of a supra-permafrost talik at two locations (boreholes 6-1 and 6-4) reflects an advanced state of degradation in these landforms.

Finally, rock glaciers appear to mark the lowest altitudinal limit of permafrost in the Andean dataset, with the lowest measurements slightly below 3700 m at Site 6. The remaining measurements within rock glaciers cluster near 0 °C between elevations of 3700 and 3800 m, and other permafrost boreholes are located at altitudes > 4250 m. Although this estimate roughly aligns with previously reported limits in the region (i.e., ranging between 2900 and 3200 m and occasionally reaching elevations of 3700 m, Saito et al., 2016), caution is warranted due to sample coverage bias associated with industry-driven data collection, rather than permafrost delineation objectives. The same caution applies to the asymmetry of the interpreted zero-degree ground temperature isotherm with respect to aspect for the full dataset. While rock glaciers occur on south-facing slopes and in areas that experience lower solar radiation, their absence on north-facing slopes may partially be a consequence of sampling bias rather than solar radiation effects.