Introduction

Permafrost is defined as a soil or rock body with a temperature below 0 °C that persists for at least two years1. Permafrost changes have important hydrological, ecological and environmental impacts2,3,4. There has been a notable progress in understanding the effects of long-term climate warming on permafrost changes based on observations and simulations. It is worth noting that short-term extreme events also have a significant impact on the permafrost environment, which has received widespread attention, especially in the current trend towards a gradual increase in the frequency and intensity of extreme high temperatures and heat waves5,6.

The existing studies on the impact of extreme events on the permafrost environment focuses on the following three aspects, including the impact on permafrost changes, disaster-causing mechanism and ecosystem processes. Firstly, extreme high temperature events could not only cause ground temperature to increase and ground ice to melt7,8,9,10,11, but also induce snow drought or wildfire events, which significantly altered the impact process of the buffer layer on the thermal state of permafrost12,13. Furthermore, in the temperature-sensitive alpine ice-rich permafrost regions, extreme high temperatures or heat waves have been observed to trigger rockfall, slope failures and landslides by melting shallow ground ice and weakening slope stability14,15,16. Additionally, extreme winter warming events have been shown to alter ecosystem processes by reducing the following summer vegetation root growth and gross primary productivity, and inhibiting the normal growth of soil fauna, bryophyte and lichen species in the sub-Arctic permafrost regions17,18.

Nevertheless, above these studies mainly focused on the effects at event scale, the principal processes by which extreme events affect the thermal state of the active layer and permafrost at seasonal and interannual scales remain uncertain. In the summer of 2022, unprecedented extreme high temperatures or heat wave exceeding 40 °C were observed in many regions of the Northern Hemisphere19, and the temperatures in some regions were 2 to 4 °C higher than the average for the past same period20,21,22,23. This can serve as a typical case for understanding the above impact processes and mechanisms.

In this study, the permafrost region in the central Qinghai-Tibet Plateau (QTP) was selected as the study area. Long-term monitoring data from six active layer and three permafrost boreholes were used (Fig. 1), in combination with field survey and reanalysis data to focus on the following objectives: (1) Did the study area also experience the summer heat wave in 2022? (2) If so, did the active layer and permafrost absorb more heat as a result of the extreme conditions, resulting in a thicker active layer or higher permafrost temperature? Were there significant differences in the response observed across the study area? (3) How much do these heat wave events contribute to seasonal thaw depth of the active layer? The results obtained can provide a more comprehensive understanding of permafrost changes and their impacts from the perspective of the effects of extreme events, and facilitate improvement in permafrost modelling.

Fig. 1: Location of active layer, borehole sites, and meteorological station in the Qinghai-Tibet Plateau (QTP).
figure 1

The permafrost distribution map was derived from Zou et al.38.

Full size image

Results

Variations of summer air temperature (SAT) and degree days thawing (DDT) from 1961 to 2022

From 1961 to 2022, the mean SAT at all sites showed significant warming trends, ranging from 0.17 to 0.34 °C·decade−1. It is noteworthy that the mean SAT in 2022 at all sites was the highest records during the last 62 years (Fig. 2). The 1961–2021 mean SAT ranged from 1.7 ± 0.8 to 6.5 ± 0.6 °C (mean: 3.9 °C), while the 2022 mean SAT reached 4.7 to 8.2 °C (mean: 6.5 °C), which was 3.2 °C above that of the 1961–1990 reference period for climate change (Supplementary Table 1). It is evident that during the summer of 2022, the daily air temperature surpassed the average temperature recorded during 1961–2021 by 5 °C on multiple occasions (Fig. 2b, d, f, h, j, l, and n). The World Meteorological Organization (WMO) defines a heat wave as when the daily maximum temperature surpasses the average maximum temperature by more than 5 °C for at least five consecutive days. The above data records demonstrate that the study area experienced the summer heat wave in 2022. In addition, the average warming rate of the study area for the 1961–2021 mean SAT was 0.27 °C·decade−1, while the 1961–2022 warming rate was 0.3 °C·decade−1 (Supplementary Table 2).

Fig. 2: Temporal characteristics of summer air temperature (SAT).
figure 2

a, c, e, g, i, k, m Characteristics of interannual variability of mean summer air temperature (SAT) from 1961 to 2022. b, d, f, h, j, l, n Comparison of summer air temperature (SAT) in 2022 with the 1961–2021 mean. Red circles represent the maximum SAT from 1961 to 2022. “k” is the magnitude of the linear trend, “R” is the correlation coefficient, and “p” is the Mann–Kendall test value.

Full size image

DDT in the study area also showed significant increasing trends during 1961–2022 with 24.28 to 38.56 °C·day·decade−1. The DDT in 2022 was significantly higher than that of 1961–2021 for all sites and represented the largest value (Fig. 3). During 1961–2021, the mean value of DDT was 450.6 °C·day, while the DDT in 2022 reached 715 °C·day, which was 264.4 °C·day higher, i.e., the DDT in 2022 was about 1.6 times higher than the mean value during 1961–2021 (Supplementary Table 1). In addition, the increase rates of DDT during 1961–2021 ranged from 21.47 to 35.74 °C·day·decade−1 (mean: 27.96 °C·day·decade−1), while the 1961–2022 increase rates ranged from 24.28 to 38.56 °C·day·decade−1 (mean: 30.69 °C·day·decade−1), i.e., the increase rate of DDT during 1961–2022 was 10% higher than the increase rate during 1961–2021, which was likely to be associated with the very high DDT caused by the summer heat wave in 2022 (Supplementary Table 2).

Fig. 3: Temporal characteristics of degree days thawing (DDT).
figure 3

a, c, e, g, i, k, m Characteristics of interannual variability of degree days thawing (DDT) from 1961 to 2022. b, d, f, h, j, l, n Comparison of degree days thawing (DDT) in 2022 with the 1961–2021 mean. Red circles represent the maximum DDT from 1961 to 2022. “k” is the magnitude of the linear trend, “R” is the correlation coefficient, and “p” is the Mann–Kendall test value.

Full size image

Observed changes in thermal conditions of the active layer

There were significant differences in the onset, offset and maximum depth of the seasonally freezing and thawing processes at six active layer sites (Supplementary Figure 1). From 2003 to 2022, the thaw depths of the active layer showed gradually increasing trends, with China06, QT01, QT06 and QT09 showing more significant increases. Especially, above four sites showed the higher soil temperature and greater thaw depth in 2022 than that of other previous years (Supplementary Fig. 1c, d, e, and f).

The summer mean ground temperature gradually decreased with increasing depths at the six active layer sites. At the four sites, including China06, QT01, QT06, and QT09, the mean summer ground temperature in 2022 was significantly higher than in previous years. In the depth interval of 0–50 cm, the mean summer ground temperature in 2022 was 1.5 °C higher than in other years, and this difference gradually decreased as depth increased (Fig. 4). Specifically, at China06, the mean summer ground temperature in 2022 were 0.3 to 2 °C higher than the average temperature recorded during 2005–2021 (Fig. 4c). Similarly, at QT01, the mean summer ground temperature in 2022 were 0.3 to1.3 °C above the average during 2004–2021 in the depth interval of 30–210 cm (except 50 cm) (Fig. 4d). At QT06 and QT09, the mean summer ground temperature in 2022 were respectively higher than the average of previous years by 0.7 to1.8 °C in the depth interval of 2–120 cm and 0.1 to 1.9 °C in the depth interval of 5–140 cm (Fig. 4e, f). At China01, the mean summer ground temperature in 2022 was higher than the values recorded during 2003–2021, but only in the depth interval of 50–120 cm (Fig. 4a). In contrast, China04 did not exhibit the same characteristics despite being described above. The mean summer ground temperature in 2022 at China04 was consistent with the range of summer mean ground temperature during 2002–2021, and did not show higher temperature features in comparison to other years despite the summer heat wave in 2022 (Fig. 4b).

Fig. 4: Changes in the mean summer ground temperature of active layer.
figure 4

a Observed mean summer ground temperature from 2003 to 2022 at China01. b Observed mean summer ground temperature from 2002 to 2022 at China04. c Observed mean summer ground temperature from 2005 to 2022 at China06. d Observed mean summer ground temperature from 2004 to 2022 at QT01. e Observed mean summer ground temperature from 2004 to 2022 at QT06. f Observed mean summer ground temperature from 2011 to 2022 at QT09. The red line represents the ground temperature profile in 2022, and the grey lines represent other years.

Full size image

From 2000 to 2022, the active layer showed gradual thickening trends at different rates, among which, the four sites, China06, QT01, QT06, and QT09, showed relatively larger thickening rates with 16.26, 14.85, 17.49, and 11.23 cm·decade−1, respectively (Fig. 5c, d, e, and f). These thickening rates were much larger than that in the Arctic and even in the Northern Hemisphere, where the thickening rate of the active layer in the Arctic was 1.5 cm·decade−1 during 1990–2019 and in the Northern Hemisphere was 6.5 cm·decade−1 during 2000–2018, respectively4,24.

Fig. 5: Temporal characteristics of active layer thickness (ALT).
figure 5

a Characteristics of interannual variability in active layer thickness (ALT) from 2000 to 2022 at China01. b Characteristics of interannual variability in active layer thickness (ALT) from 2000 to 2022 at China04. c Characteristics of interannual variability in active layer thickness (ALT) from 2000 to 2022 at China06. d Characteristics of interannual variability in active layer thickness (ALT) from 2000 to 2022 at QT01. e Characteristics of interannual variability in active layer thickness (ALT) from 2000 to 2022 at QT06. f Characteristics of interannual variability in active layer thickness (ALT) from 2000 to 2022 at QT09. Red circles represent the maximum ALT during monitoring period. “k” is the magnitude of the linear trend, “R” is the correlation coefficient, and “p” is the Mann–Kendall test value.

Full size image

It is noteworthy that the maximum ALT at all four sites occurred in 2022, and the ALT in 2022 were 29.4, 33.1, 37.9, and 26.7 cm higher than the mean ALT values during 2000–2021, respectively (Fig. 5), and the ALT in 2022 was 1.2 times higher than the 2000–2021 average (Supplementary Table 1). The mean thickening rate were 14.96 cm·decade−1 and 12.97 cm·decade−1 during 2000–2022 and 2000–2021, respectively (Fig. 5c–f; Supplementary Table 2), this difference was not only related to the different ending years of the linear trend25,26, but also probably to the extreme high temperatures in 2022. In addition, the ALT at China01 and China04 showed thickening trends during 2000–2022 with small rate of 2.46 and 5.98 cm·decade−1, respectively (Fig. 5a, b). The maximum ALT for China01 occurred in 2016 (167.6 cm) (Fig. 5a); China04’s ALT in 2022 (119.9 cm) was very close to the maximum ALT that occurs in 2013 (120 cm) (Fig. 5b).

Observed changes in thermal conditions of permafrost

At three borehole sites, seasonally thawing process of shallow soil mainly occurred in the depth interval of 0–3 m, but the seasonally freezing process varied greatly (Supplementary Fig. 2), among which the seasonally freezing process of QTB01 and QTB06 mainly occurred in the depth interval of 0–2 m (Supplementary Fig. 2a, b), while the seasonally freezing process of QTB09 could reach a depth of 6 m (Supplementary Fig. 2c). Both shallow ground temperature and thaw depth in 2022 at three sites reached the maximum. In the permafrost layers, ground temperature of QTB01 and QTB06 presented relatively high average temperatures of −0.33 and −0.55 °C, respectively (Supplementary Fig. 2a, b), while QTB09 was even lower, reaching −2.5 °C (Supplementary Fig. 2c).

The summer mean ground temperature at the three borehole sites exhibited significantly year-to-year variation. Additionally, the negative temperature ranges showed considerable variability, with QTB01 and QTB06 having negative temperature ranges of no greater than −1 °C, and QTB09 having an even lower value (not less than −3 °C) (Fig. 6). The mean summer ground temperature in 2022 was overall higher than that of the other previous years, and this difference gradually decreased with the depth, where in the depth interval of 0–2 m, the summer ground temperature of the three boreholes were higher than that of the other previous years by about 1.95 °C and 0.25 °C in the depth interval of 2–18 m, respectively. Specifically, at QTB01, summer ground temperature at different depths in 2022 were 0.03–2.14 °C higher than those of 2011–2020 (Fig. 6a); at QTB06, summer ground temperature at different depths in 2022 were 0.01–2.44 °C higher than that of 2006–2021 (Fig. 6b); at QTB09, summer ground temperature at different depths in 2022 were 0.13–2.56 °C higher than those of 2011–2021 (Fig. 6c).

Fig. 6: Changes in the mean summer ground temperature of permafrost.
figure 6

a Observed mean summer ground temperature from 2011 to 2022 at QTB01 borehole. b Observed mean summer ground temperature from 2006 to 2022 at QTB06 borehole. c Observed mean summer ground temperature from 2011 to 2022 at QTB09 borehole. The red line represents the ground temperature profile in 2022, and the grey lines represent other years.

Full size image

The MAGT in 2022 at QTB01, QTB06, and QTB09, were the largest during the whole monitoring period, being −0.26, −0.24, and −2.34 °C (mean: −0.95 °C), respectively, which were higher than the means of the other previous years (−0.32 ± 0.04, −0.36 ± 0.07, and −2.48 ± 0.09 °C, mean: −1.05 °C), i.e., in 2022, the MAGT in the study area was 1.1 times higher than the mean of the other previous years (Supplementary Table 1). Combining the multiple linear regression statistics and existing studies, the high values of MAGT in 2022 were not only caused by the extreme high temperatures or the summer heat wave in 2022, but probably by the cumulative effect of the climatic conditions in the previous years27,28,29 (Supplementary information). The MAGT at the three sites all showed increasing trends. It may be dominated by winter climate change rather than summer, due to the higher winter warming rate and the limited buffering effect of snow cover30,31,32 (Supplementary Table 3). If the year 2022 was excluded, the MAGT at the three borehole sites demonstrated a significant increasing trend at rates of 0.085, 0.138, and 0.298 °C·decade−1, respectively. However, if 2022 was included, the MAGT increasing rates are 0.088, 0.139, and 0.289 °C·decade−1, and all these trends passed the 0.01 significance test despite missing data in individual years (Fig. 7). This difference was not only influenced by the different ending years of the linear trend25,26, but was likely to be related to the extreme high temperatures of 2022 and the cumulative effect of climatic conditions in the previous years.

Fig. 7: Temporal characteristics of mean annual ground temperature (MAGT).
figure 7

a Characteristics of interannual variability in mean annual ground temperature (MAGT) from 2011 to 2022 at QTB01 borehole. b Characteristics of interannual variability in mean annual ground temperature (MAGT) from 2006 to 2022 at QTB06 borehole. c Characteristics of interannual variability in mean annual ground temperature (MAGT) from 2012 to 2022 at QTB09 borehole. Red circles represent the maximum MAGT during monitoring period. “k” is the magnitude of the linear trend, “R” is the correlation coefficient, and “p” is the Mann–Kendall test value.

Full size image

Contribution of heat wave to the seasonal thaw depth of the active layer

The contribution of heat wave to the seasonal thaw depth of the active layer was quantified using the Stefan physics-based model, which was expressed as the contribution fraction. The mean contribution fractions of six sites differed significantly, with their mean values ranging from 6.6% to 13.6%. China01, China06 and QT09 showed relatively larger contribution fractions, with 11.1% ± 4.4%, 13.6% ± 6.6% and 11.8% ± 6.1%, respectively, all of which were over 10%. The mean contribution fractions of China04, QT01 and QT06 were all below 10%, with China04 having the smallest value at 6.6% ± 2.9% (Fig. 8).

Fig. 8: Box plots of contribution faction at six active layer sites.
figure 8

The dotted line in the boxes represents the average of contribution fraction. The values in the boxes before and after the symbol “±” represent the mean value and standard deviation of contribution faction from 2000 to 2022, respectively.

Full size image

The contribution fractions of six sites exhibited significant increasing trends from 2000 to 2022, with increase rates ranging from 1.77%·decade−1 to 4.52%·decade−1 (Fig. 9). The increase rates at China01, China04, China06, and QT06 were 3.07, 2.78, 4.52, and 2.42%·decade−1, respectively, and all passed the 0.05 significance test. However, the trends at QT01 (1.77%·decade−1) and QT09 (3.66%·decade−1) did not pass the significance test. The maximum contribution fraction at all six sites occurred in 2022. Among them, China01, China06, QT01 and QT09 had a contribution fraction of more than 20% in 2022, with values of 22.8%, 31.3%, 20.9% and 28.9%, respectively. China04 and QT06 reached their maximum in 2022 with values of 12.4% and 14.5%, respectively (Fig. 9). In 2022, the average contribution fraction at all six sites was 21.8%, which is 2.3 times higher than the average contribution during 2000–2021 (9.5%). The average increase rate of contribution fraction during 2000–2022 at all six sites was 3%·decade−1, but only 1.9%·decade−1 during 2000–2021.

Fig. 9: Temporal characteristics of contribution factions.
figure 9

a Characteristics of interannual variability in contribution factions from 2000 to 2022 at China01. b Characteristics of interannual variability in contribution factions from 2000 to 2022 at China04. c Characteristics of interannual variability in contribution factions from 2000 to 2022 at China06. d Characteristics of interannual variability in contribution factions from 2000 to 2022 at QT01. e Characteristics of interannual variability in contribution factions from 2000 to 2022 at QT06. f Characteristics of interannual variability in contribution factions from 2000 to 2022 at QT09. Red circles represent the maximum contribution faction during monitoring period. “k” is the magnitude of the linear trend, “R” is the correlation coefficient, and “p” is the Mann–Kendall test value.

Full size image

Discussion

From the above results, it could be seen that despite the significant positive correlations between SAT, \(\sqrt{DDT}\) and the ALT at different sites (Supplementary Fig. 3 and Supplementary Fig. 4), there were large differences in the response the summer heat wave in 2022 among the 6 active layer sites. Therefore, we elucidated the main mechanisms by assessing the impacts of four factors on ALT, including climate, vegetation, soil, and topography factors among the six sites.

During the thawing period, the heat flux penetrated into the active layer due to the thermogradient between air and soil temperature, while the accumulated heat flux determines the total thaw depth33. More heat accumulation can promote the thickening of the active layer. Wu and Zhang34 found that the increase in SAT was the main influencing factor for the thickening of the active layer in the QTP. We observed significant differences in SAT among the six sites, with China04 having the highest SAT (6.8 ± 0.6 °C), and China06 having the lowest SAT (2.4 ± 0.9 °C) (Fig. 10a). The numerical characteristics of the DDT were relatively similar to those of SAT, i.e., DDT was the highest in China04 and the smallest in China06 (Fig. 10b). The maximum difference in mean SAT and DDT among the six sites amounted to 4.4 °C and 563.1 °C·day, respectively. Such large differences inevitably led to differences in heat accumulation and thawing processes among different sites.

Fig. 10: Correlation between various influencing factors and active layer thickness (ALT).
figure 10

a Correlation between SAT and ALT. b Correlation between DDT and ALT. c Correlation between snow depth and ALT. d Correlation between vegetation coverage and ALT. e Correlation between above-ground biomass and ALT. f Correlation between below-ground biomass and ALT. g Correlation between soil organic matter and ALT. h Correlation between soil moisture and ALT. i Correlation between proportion of soil particle size and ALT. j Correlation between altitude and ALT. k Correlation between slope aspect and ALT. l Correlation between slope gradient and ALT. “k” is the magnitude of the linear trend, “R” is the correlation coefficient, and “p” is the Mann–Kendall test value. The red straight line depicts the fitted values line for six sites, while the blue line represents the fitting line for the remaining five sites after China04’s exclusion in (a, b) and QT06’s exclusion in (c), respectively.

Full size image

The effect of snow cover on the thermal condition of the active layer mainly depends on its thickness, density, timing, duration, and structure, etc35,36,37. Generally, thicker snow cover resulted in higher albedo, higher thermal emissivity, higher absorptivity, and lower thermal conductivity. The mean annual snow depth showed considerable variability, ranging from 0.33 to 0.46 cm across the six sites (Fig. 10c), and the maximum daily snow depth did not exceed 2 cm. In addition, the duration of snow cover is very short due to strong solar radiation and winds, especially in the hinterland and northern regions of the QTP30, resulting in a limited buffering effect between the air and the ground surface, and a relatively small effect of snow cover on local ground temperature38. It can be reasonably assumed that the impact of thin snow cover on ALT in the study area is likely to be insignificant.

Vegetation with high coverage can create a shading effect that weakens the solar radiation reaching the ground to some extent, and the vegetation transpiration process also consumes some of the heat, which effectively lowering the near-surface temperature and resulting in a shallower ALT39. The lower the vegetation coverage, the more sensitive the active layer soil temperature in response to changes in air temperature40,41. In the permafrost region of the QTP, as the vegetation type gradually changes from alpine swamp meadow to alpine meadow, alpine grassland, and alpine desert grassland, the surface vegetation coverage gradually decreases, while the ALT tends to increase42. In the study area, the vegetation type was alpine swamp meadow at China01 and China04, alpine meadow at QT01 and QT09, and alpine steppe at China06 and QT06 (Table 1, Supplementary Figure 5). Statistical results showed that for every 10% decrease in vegetation coverage in the permafrost regions of the QTP, ALT increased by approximately 17.8 cm43. Vegetation coverage in this study area ranged from 24.2% to 89.6%, with China04 having significantly higher vegetation coverage (89.6% ± 7.8%) than the other sites (Supplementary Fig. 5; Supplementary Fig. 6a), followed by QT09 (79.9% ± 7.6%), and China06 having the lowest vegetation coverage (24.2% ± 5.9%), which was less than 1/3 of that of China04 (Fig. 10d). Biomass is also one of the important factors influencing the thermal condition of the active layer44. The values of above-ground biomass values ranged from 0.06 to 0.32 kg·m−2 and below-ground biomass values ranging from 0.94 to 21.73 kg·m−2, with China04 having the largest above-ground and below-ground biomass of 0.32 ± 0.07 kg·m−2 and 21.73 ± 2.3 kg·m−2, respectively (Fig. 10e, f). Additionally, the below-ground biomass in the study area was overall higher than the below-ground biomass, and higher below-ground biomass indicates a richer root system, which to some extent enhances heat transfer through the soil.

Table 1 Information on active layer and borehole sites
Full size table

The active layer with a thicker organic layer tends to be insensitive to changes in air temperature due to their buffering effect37. The thick the soil organic matter content can increase the soil water-holding capacity and significantly increase the soil moisture in the active layer. This, in turn, will decrease the soil thermal conductivity, which effectively attenuates the amplitude of the warming of the active layer. Thus, under their combined effects, ALT tends to be thinner with increasing soil organic matter content39,45. The high and thick organic matter content has been considered as one of the main causes for the shallow ALT in the circumpolar Arctic46. Soil organic matter content of China04 was significantly higher than that of the other five sites (Supplementary Fig. 6b, c), reaching to 110.2 ± 73.7 g·kg−1, which is 7 to 28 times higher than that of the other sites, followed by China01 (15.4 ± 1.0 g·kg−1), while China06 had the lowest soil organic matter content of 4.0 ± 1.1 g·kg−1 (Fig. 10g). Thus, we infer that the high organic matter content is the main reason that China04 exhibit less sensitivity to the summer heat wave when compared to other sites.

Soil moisture can alter the energy exchange processes in the soil by affecting soil heat capacity and thermal conductivity, which ultimately affects the ALT47. Previous observations and modelling results suggested that higher soil moisture tends to result in a shallower ALT48. Recent results suggested that higher near-surface soil moisture can lead to the increase of thermal conductivity, which will enhance the energy exchange between air and soil, allowing more heat penetrate into the active layer and provide positive feedbacks to ALT. However, increasing soil moisture in the middle or bottom active layer may result in negative feedbacks in ALT through increasing the latent heat of fusion used for thawing processes49. In particular, the variations in latent heat associated with soil moisture may have a greater effect on ALT than the variations in thermal conductivity50. In the study area, the mean annual observed soil moisture at different depths of the six active layer sites also exhibited considerable variability, with values ranging from 0.02 m3·m−3 to 0.39 m3·m−3 (Supplementary Fig. 7). QT06 had the highest soil moisture value of 0.28 ± 0.1 m3·m−3 (Fig. 10h), and it was significantly higher than the other sites in the depth interval below 90 cm, reaching a maximum of 0.39 m3·m−3, and gradually increasing with depth (Supplementary Fig. 7e). The soil moisture at China01 and China04 were followed by QT06, with 0.20 ± 0.05 m3·m−3 and 0.19 ± 0.05 m3·m−3, respectively (Fig. 10h; Supplementary Fig. 6d), and both gradually decreased at depths below 95 cm (Supplementary Fig. 7a, b). It is worth noting that the field survey revealed the presence of significant amounts of ground ice near permafrost table at China04 (Supplementary Fig. 6e, f). The values of QT09, QT01 and China06 were closer and relatively small, 0.15 ± 0.05, 0.14 ± 0.06 and 0.13 ± 0.07 m3·m−3, respectively (Fig. 10h). Soil moisture at QT09 were relatively low in the 5 cm depth interval at the surface and the 120–160 cm depth interval below ground, averaging around 0.1 m3·m−3 (Supplementary Fig. 7f). QT01’s soil moisture in the 30–120 cm depth interval ranged from 0.1 m3·m−3 to 0.16 m3·m−3 (Supplementary Fig. 7d). At China06, soil moisture in the 10–80 cm depth interval was the lowest of the six sites, with a mean value of only 0.06 m3·m−3 (Supplementary Fig. 7c).

The finer the soil particles, the lower the ALT. This is because the finer the soil particles, the larger the surface energy, the stronger the water-holding capacity, the more water is retained in the soil body under the same conditions of external water replenishment, and the evaporation of water consumes a large amount of heat, resulting in a relatively small amount of heat used to warm the soil body, which lead to a decrease in soil temperature and a shallower ALT51. Under the same climate warming background, the active layer dominated by coarse soils had a larger change magnitude than the active layer dominated by fine soils50. In this study area, the proportion of soil particle size that less than 0.054 mm ranged from 0.9% ± 0.5% to 13.3% ± 4.6%, with the smallest and largest percentages at China06 and China01, respectively, and closer values at China04, QT01, and QT09, which were 7.2% ± 2.6%, 9.6% ± 3.6%, and 8.1% ± 5.1%, respectively (Fig. 10i).

Other things being equal, the higher the altitude, the lower the ALT. For every 100 m increase in altitude in the permafrost regions along the Qinghai-Tibet Highway, the ALT decreases by about 40 cm43. The altitudes in the study area ranged from 4538 to 4896 m, with a maximum altitude difference of 358 m (Fig. 10j). In general, ALT is greater on sunny slopes than on shady slopes; for the same slope aspect and different slope gradient, the greater the slope gradient, the smaller the ALT42. Among the six active layer sites, China01 and QT01 are located on shady slopes, while the other four sites are located on sunny slopes (Fig. 10k). The slope gradients of all sites ranged from 1.1° to 2.5°, with the largest slope gradient at QT09 (2.5°) and the smallest at QT06 (1.2°) (Fig. 10l).

SAT and DDT exhibited weak positive correlations with ALT as indicated by red fitted lines in Fig. 10a, b. Upon excluding China04, it was observed that significant positive correlations existed at the remaining five sites. This was evidenced by the blue fitted lines in Fig. 10a, b. As a corollary, China04’s response to changes in SAT and DDT is not sensitive. There was a clear negative correlation between snow depth and ALT (Fig. 10c). This is in contrast to previous studies and may be due to the fact that, on the one hand, snow cover is not a dominant factor influencing ALT in the study area due to the thin snow depth and short duration. On the other hand, the magnitude of the snow depth at a given site has a significant effect on ALT, which increases with increasing snow depth. Nevertheless, the analysis of this fitting relationship for six different sites within the study area may not yield reasonable results. Additionally, higher vegetation coverage in the study area was associated with smaller ALT (Fig. 10d). The greater the above-ground and below-ground biomass, the lower the ALT, particularly at China04. This site boasted the highest vegetation coverage, as well as the largest above-ground and below-ground biomass, resulting in the smallest ALT (Fig. 10e, f). Soil organic matter exhibited an inverse relationship with ALT across all sites, with China04 having the highest concentration of this factor and concurrently the smallest ALT (Fig. 10g). After the exclusion of QT06, the fitted relationships for the remaining five sites demonstrated that an increase in soil moisture corresponded to a decrease in ALT (Fig. 10h). There was a negative, but insignificant correlation between the proportion of soil particle size that below 0.054 mm and ALT (Fig. 10i). Among the topography factors, there was a slight negative correlation between altitude, slope gradient, and ALT (Fig. 10j, l), while the effect of slope aspect on ALT was not noticeably significant (Fig. 10k).

In summary, the ALT was a result of various factors associated with climate, vegetation, soil and topography. Changes in temperature or other individual factors did not always lead to corresponding significant changes in the ALT. Therefore, the diverse response of the ALT to the summer heat wave in 2022 in the study area occurred. For instance, the ALT of China04 showed little response to SAT and DDT, while being significantly affected by vegetation, soil organic matter and ground ice near permafrost table (Supplementary Fig. 6). The ALT at QT06 was primarily determined by climate factors and was relatively less impacted by vegetation and soil factors. Furthermore, in certain areas, the ALT did not display sensitivity to climate warming, which may be attributed to a greater amount of subsurface ice content. This is due to the fact that a significant amount of heat transported by climate warming is consumed in thawing the subsurface ice and causing ground subsidence37. Nonetheless, current ALT monitoring inadequately reflects genuine changes in the ALT because it fails to incorporate observations of ground subsidence52.

The monitoring of the active layer in the future must consider the integrated changes in the ALT and ground subsidence. Investigation should focus on response differences among various types of active layer and permafrost to long-term climate warming and short-term extreme events. This will enable the further exploration of the influence process and mechanism of extreme events on permafrost environments and lead to better permafrost modelling concerning extreme events.

Methods

Calculation of meteorological and soil metrics

We chose four meteorological and soil metrics, including the mean summer air temperature (SAT), degree days thawing (DDT), active layer thickness (ALT), mean annual ground temperature (MAGT). We calculated the SAT by using the daily air temperature acquired from June to August during 1961–2022. DDT was calculated by summing positive daily air temperature during each thawing period, it can effectively reflect the cumulative amount of energy transfer between the ground and air in the permafrost regions, and it is widely used to drive statistical or physical models related to permafrost53. Based on observed soil temperature at different depths, 0 °C isotherm can be obtained and its maximum depth in a given year is the ALT for that year. MAGT is the ground temperature at the depth of zero annual amplitude (DZAA) of permafrost profile, which is acquired based on observed daily ground temperature at different depths.

In this study, it is impossible to judge the ALT of some active layer sites due to data missing in the in-situ soil temperature in some years. Therefore, we used the Stefan model and ERA-5 Land reanalysis data to estimate the ALT in the year of missing data for the six active layer sites during 2000–2022. The Stefan model had been widely used to estimate ALT in the QTP24,54. This calculation assumes that the entire active layer is a homogeneous subsurface medium with constant hydro-thermal conditions. Based on soil hydro-thermal and related physical parameters, ALT can be expressed as:

$${ALT}=\sqrt{(2{\lambda }_{t}{DDT}) / \rho \omega L)}$$
(1)

where ALT is the thaw depth of the active layer. λt is the thermal conductivity of the unfrozen soil. ρ is the soil dry density. ω and L are the soil moisture and the latent heat of fusion, respectively. Degree days thawing (DDT) is the cumulative number of positive degree days. By introducing an edaphic factor (E), including unfrozen soil thermal conductivity, soil moisture and latent heat of fusion, and soil dry density55, a simplified Stefan model is as follows:

$${ALT}=E\sqrt{{DDT}}$$
(2)

We firstly acquired the observed ALT by interpolating the layered soil temperature data. Based on Eq. (2), using the observed ALT and the corresponding DDT, we acquired the averaged E for the years having abundant data. Then the acquired E was taken as a known and assumed that it was constant in years with missing data, and the value of DDT in the year of missing temperature data was inputted into Eq. 2 again, which will give us the data missing year of ALT.

Based on the simulated ALT obtained by the simplified Stefan model, we quantitatively evaluated the simulation accuracy using the observed ALT, and found that the R2 of the observed ALT versus the simulated ALT was 0.9, the RMSE was 12.96, and the MAE was 10.31 cm (Supplementary Fig. 8), which were close to the existing studies24,55,56. Therefore, using the simplified Stefan model for the missing data years of ALT filling has sufficient reliability in this study.

Estimating the contribution fraction of heat wave to the seasonal thaw depth of the active layer

In general, the downward thawing process of the active layer in the Qinghai-Tibet Plateau begins in late April or May, this process finishes until reaching the maximum thaw depth in late October or early November. We assumed that the active layer thawing process for a year can be divided into two parts, one was the normal thawing process dominated by regularly seasonal air temperature variations; the other was the cumulative thawing process caused by only heat wave events with short duration and high intensity.

Under such assumption, we could obtain the seasonal thaw depth without the influence of heat wave events (TDN_HW). The main practices were as follows: firstly, we replaced the initial temperature series including the influences of heat wave events with the temperature series that without the influence of heat wave events by using the averages of all temperatures on the same day of the year corresponding to the day of the heat wave event when the same day is a non-heat wave event; Next, we used the Stefan model, a physics-based models, to simulate TDN_HW by inputting the temperature series without the influence of heat wave events.

According to our assumption, the ALT in a given year was mainly composed of the seasonal thaw depth without the influence of heat wave events (TDN_HW) and the seasonal thaw depth influenced by heat wave events (TDHW). We therefore could obtain the TDHW based on the known ALT and the TDN_HW obtained above, as follows.

$${{TD}}_{{HW}}={ALT}-{{TD}}_{{N\_HW}}$$
(3)

We introduced the contribution fraction indicator to quantify the contribution of heat wave events to the seasonal thaw depth in a given year, and we could obtain the contribution fraction (Unit: %) by calculating the ratio between TDHW and ALT, which is as follows:

$${Contribution\; fraction}=\frac{{{TD}}_{{HW}}}{{ALT}}\times 100 \%$$
(4)

Multiple linear regression analyses for the MAGT

As DDT and DDF are the most important influences on MAGT in the QTP and are important input factors for the TTOP (temperature at the top of permafrost model) and FROSTNUM models, we conducted multiple linear regression analyses with DDT and DDF at the three boreholes as independent variables and MAGT as the dependent variable. It is important to note that vegetation, snow cover, soil texture and soil moisture are also key elements affecting MAGT, but we do not have sufficient continuous observational data for these elements. We designed a total of eight potential fitting schemes as follows.

$${\rm{Scheme}}\,1:{MAGT}={t}_{1}{{DDT}}_{0}+{f}_{1}{{DDF}}_{0}+c$$
$${\rm{Scheme}}\,2:{MAGT}={t}_{1}{{DDT}}_{0}+{t}_{2}{{DDT}}_{-1}+{f}_{1}{{DDF}}_{0}+{f}_{2}{{DDF}}_{-1}+c$$
$${\rm{Scheme}}\,3:{MAGT}={t}_{1}{{DDT}}_{0}+{t}_{2}{{DDT}}_{-1,-2}+{f}_{1}{{DDF}}_{0}+{f}_{2}{{DDF}}_{-1,-2}+c$$
$$\begin{array}{ll}{\rm{Scheme}}\,4:{MAGT}={t}_{1}DDT_{0}+{t}_{2}DDT_{-1,-2,-3}\\\qquad\qquad\qquad\qquad\qquad+\,{f}_{1}DDF_{0}+{f}_{2}DDF_{-1,-2,-3}+c\end{array}$$
$$\begin{array}{l}{\rm{Scheme}}\,5:MAGT={t}_{1}DDT_{0}+{t}_{2}DDT_{-1,-2,-3,-4}+{f}_{1}DDF_{0}\\\qquad\qquad\qquad\qquad\qquad+\,{f}_{2}DDF_{-1,-2,-3,-4}+c\end{array}$$
$$\begin{array}{l}{\rm{Scheme}}\,6:MAGT={t}_{1}DDT_{0}+{t}_{2}DDT_{-1}+{t}_{3}DDT_{-2}+{f}_{1}DDF_{0}\\\qquad\qquad\qquad\qquad\quad+\,{f}_{2}DDF_{-1}+{f}_{3}DDF_{-2}+c\end{array}$$
$$\begin{array}{l}{\rm{Scheme}}\,7:{MAGT}={t}_{1}DDT_{0}+{t}_{2}DDT_{-1}+{t}_{3}DDT_{-2}+{t}_{4}DDT_{-3}\\\qquad\qquad\qquad\qquad\qquad+\,{f}_{1}DDF_{0}+{f}_{2}DDF_{-1}+{f}_{3}DDF_{-2}+{f}_{4}DDF_{-3}+c\end{array}$$
$$\begin{array}{l}{\rm{Scheme}}\,8:{MAGT}={t}_{1}DDT_{0}+{t}_{2}DDT_{-1}+{t}_{3}DDT_{-2}+{t}_{4}DDT_{-3}\\\qquad\qquad\qquad\qquad\quad+\,{t}_{5}DDT_{-4}+{f}_{1}DDF_{0}+{f}_{2}DDF_{-1}+{f}_{3}DDF_{-2}\\\qquad\qquad\qquad\qquad\quad+\,{f}_{4}DDF_{-3}+{f}_{5}DDF_{-4}+c\end{array}$$

where DDT0 is the value of DDT in the same year as MAGT, DDT-1 is the value of DDT in the past year 1, DDT-2 is the value of DDT in the past year 2, and DDT-3 is the value of DDT in the past year 3; DDT-1, -2 represents the sum of DDT in the past years 1 and 2, and DDT-1, -2, -3 is the sum of DDT in the past year 1, year 2 and year 3, and so on. DDF is the same as this. t1, t2, …, t5 and f1, f2, …, f5 are the coefficients of the above independent variables, respectively. c is a constant.

Other methods

The Pearson correlation analysis method was used to estimate the correlation coefficient (R). The least squares method was used to quantitatively estimate the linear trend. The Mann-Kendall test was used to determine the significance level of the changing trends. Root mean square error (RMSE) and mean absolute error (MAE) were used to quantitatively assess the difference between site-observed and ERA5-Land temperature data, observed ALT and simulated ALT, respectively.

$${RMSE}=\sqrt{{\sum }_{i=1}^{n}{{d}_{i}}^{2}/n}$$
(5)
$${MAE}=\mathop{\sum}\limits_{i=1}^{n}\left|{d}_{i}\right|/n$$
(6)

where di is the difference between the site-observed and ERA5-Land, or the observed ALT and simulated ALT using the Stefan model, n is the number of values.

Meteorological and soil measurements

In this study, six active layer sites and three borehole sites were selected in the permafrost regions of the central Qinghai-Tibet Plateau (QTP) (Fig. 1 and Supplementary Fig. 1), with altitudes above 4500 m a.s.l. Vegetation types dominated by alpine steppe and alpine meadows. Soil temperature and moisture in the active layer were monitored using 105 T thermocouple temperature sensors and Stevens Hydro probes with accuracies of ± 0.1 °C and ± 3%, respectively. The ground temperature in boreholes were observed using thermistor probes with accuracy of ± 0.1 °C, which were produced and calibrated by the State Key Laboratory of Frozen Soil Engineering, Chinese Academy of Sciences. CR10X and CR3000 data loggers (Campbell Scientific, Inc.) were used to record above data every 30 minutes. The in-situ monitoring data was obtained from the Cryosphere Research Station on the Qinghai-Tibet Plateau, Chinese Academy of Sciences (http://www.crs.ac.cn). The mean annual air temperature (MAAT) ranged from –8.9 to –3.6 °C, while the mean annual ground temperature (MAGT) ranged from –2.47 to –0.31 °C. The active layer thickness (ALT) ranged from 114.4 to 244.8 cm (Table 1). Soil samples were collected from six active layer sites by digging pits, and soil organic content was measured using the dichromate oxidation method after pretreatment of air-dried soil samples by grinding and passing through a 2 mm screen.

To overcome the data gapping problems in the in-situ observations, air temperature during 1961–2022 from the ERA5-Land reanalysis dataset was used due to its high spatial and temporal resolution and high reliability, and this dataset has been widely used in accessing permafrost state and dynamics in the QTP50,57,58. To reduce the data uncertainties in local scales due to the complex topography of the study area, in-situ air temperature observations were used to assess the reliability of the ERA5-Land. The results of the assessment showed that ERA5-Land can better reflect the interannual variation characteristics of air temperature (Supplementary Fig. 9, Supplementary Fig. 10, and Supplementary Fig. 11), and has a smaller error and higher reliability in summer (Supplementary Fig. 12 and Supplementary Fig. 13; Supplementary Table 4). In particular, ERA5-Land can well reflect the high temperature characteristics of the study area in the summer of 2022. The detailed assessment process and results are presented in the Supplementary information. Additionally, ERA5-Land soil temperature data are significantly underestimated in the QTP, leading to large errors in modelling active layer thickness and permafrost area59. The reliability of ERA5-Land soil temperatures was significantly improved after the modification of the snow densification parameterization60. Therefore, ERA5-Land soil temperature data should be treated with caution when used in permafrost related studies. Moreover, in consideration of the exemplary reliability of the long-term series of daily snow depth dataset in China (1979–2023)61, snow depth during 2000–2022 from this dataset were used to access the response of ALT to snow cover.

Vegetation, biomass and environmental datasets

To access the effects of vegetation and biomass on ALT, vegetation coverage, above and below-ground biomass data were used, which were acquired during field surveys from 2013 to 2023. The detailed survey methods are described in Yue et al.44. SRTM DEM data (https://srtm.csi.cgiar.org/srtmdata/) was used to calculate the topography factors, including altitude, slope aspect and slope gradients. Then, the three factors were interpolated to the corresponding sites using bilinear interpolation.