Simulation of Spatiotemporal Distribution and Variation of 30 m Resolution Permafrost in Northeast China from 2003 to 2021

Abstract

The high-resolution permafrost distribution maps have a closer relationship with engineering applications in cold regions because they are more relative to the real situation compared with the traditional permafrost zoning mapping. A particle swarm optimization algorithm was used to obtain the index η with 30 m resolution and to characterize the distribution probability of permafrost at the field scale. The index consists of five environmental variables: slope position, slope, deviation from mean elevation, topographic diversity, and soil bulk density. The downscaling process of the surface frost number from a resolution of 1000 m to 30 m is achieved by using the spatial weight decomposition method and index η. We established the regression statistical relationship between the surface frost number after downscaling and the temperature at the freezing layer that is below the permafrost active layer base. We simulated permafrost temperature distribution maps with 30 m resolution in the four periods of 2003–2007, 2008–2012, 2013–2017, and 2018–2021, and the permafrost area is, respectively, 28.35 × 10⁴ km², 35.14 × 10⁴ km², 28.96 × 10⁴ km², and 25.21 × 10⁴ km². The proportion of extremely stable permafrost (<−5.0 °C), stable permafrost (−3.0~−5.0 °C), sub-stable permafrost (−1.5~−3.0 °C), transitional permafrost (−0.5~−1.5 °C), and unstable permafrost (0~−0.5 °C) is 0.50–1.27%, 6.77–12.45%, 29.08–33.94%, 34.52–39.50%, and 19.87–26.79%, respectively, with sub-stable, transitional, and unstable permafrost mainly distributed. Direct and indirect verification shows that the permafrost temperature distribution maps after downscaling still have high reliability, with 83.2% of the residual controlled within the range of ±1 °C and the consistency ranges from 83.17% to 96.47%, with the identification of permafrost sections in the highway engineering geological investigation reports of six highway projects. The maps are of fundamental importance for engineering planning and design, ecosystem management, and evaluation of the permafrost change in the future in Northeast China.

1. Introduction

The permafrost region of Northeast China, located at the southern edge of the permafrost region of Eurasia, is the main distribution region of permafrost in the middle and high latitudes of China. It is also a significant and sensitive area of climate change. The permafrost is distributed in the form of islands and is easily affected by external environmental changes, and its thermal state is extremely unstable [1,2,3,4]. The duration of the existence of permafrost varies greatly due to regional differences in climate conditions. In a certain region, permafrost distribution is not completely continuous, the distribution characteristics will also change within a short distance, and it cannot be directly observed with remote sensing. Therefore, a spatial evaluation of permafrost on a large scale is challenging [5,6,7]. The permafrost distribution in northern Northeast China is highly variable and discrete in a small range and the temperature, thickness, thermal stability, and sensitivity to climate and environmental changes at different spatiotemporal scales are significantly different [8,9].

China’s research on the permafrost mapping began in the 1960s. Due to the limitations of observation data and survey methods before the 1980s, the mapping was mainly based on the limited permafrost investigation and survey data as well as on the understanding of the temperature conditions and topographic and geomorphological features affecting the formation and distribution of permafrost; it was artificially sketched out, and the accuracy of the distribution maps was low. The first generation of permafrost maps published at this stage was influenced by the permafrost classification system of the former Soviet Union and North America and used continuity as the criterion for dividing the permafrost zones, which is a combination of latitudinal zonation and vertical zonation and the continuity used to define the permafrost classification system. The permafrost regions are not 100% developed permafrost, and the permafrost area reflects the area of the permafrost regions. Therefore, the permafrost area in Northeast China in the early period differs greatly from recent research results. In recent years, with the optimization of survey methods, the accumulation and open sharing of monitoring data, the increasing abundance of remote sensing data, and the improvement of physical models, favorable conditions have been provided for understanding the distribution and change of permafrost.

The surface frost number model is an empirical model, which was proposed by Nelson [10] and does not have obvious physical significance. It is calculated from the freezing index and melting index and can be used to analyze, simulate, and predict the permafrost distribution [10]. The model has been used to evaluate permafrost in different spatial scales in Central Canada [11], Russia [12], China’s Qinghai Tibet Plateau [13,14,15], and Northeast China [16,17,18]. However, the frost number values of the plane boundary dividing seasonal frozen soil and permafrost as well as continuous and discontinuous permafrost in permafrost regions are often adjusted based on experience rather than being applied to all regions with a fixed value. In addition, due to the lack of local factors, such as vegetation, altitude, and terrain in the model, there is significant uncertainty in the simulation results in some applications [19,20].

The permafrost in Northeast China is mainly distributed in the Great and Lesser Khingan Mountains, with dense vegetation. As a sensitive component of the terrestrial ecosystem, vegetation plays a connecting role among the atmosphere, water, and soil and an important regulatory role in the terrestrial carbon balance, climate system, etc. [21]. The response of permafrost to climate change is also complicated by linkages and mutual feedback with various components of terrestrial ecosystems that influence the sensitivity of permafrost to external disturbances [22,23]. Combining the variables that affect the occurrence of permafrost in the region and adding factors that affect the formation and development process of permafrost to the model, such as the vegetation factors [24,25], can better simulate the real permafrost distribution in Northeast China.

It is difficult to evaluate the permafrost distribution over a large area and fully consider the spatial heterogeneity of the environment. The environmental changes under the influence of natural and human activities determine the water, heat, and material balance of the surface; will change the land–atmosphere water, energy, and carbon cycle; and will even cause climate change. Thus, the environment variables are often simplified or selected with large influence weights in physical or empirical models. These environmental variables, from land cover categories to surface materials, involve ecosystems and geomorphic processes. The parameters that describe environmental variables are rarely on the same temporal and spatial resolution conditions. Therefore, the selection of environmental variables (or parameters) as well as the accuracy and quantifiability should also be considered as key factors in evaluating the permafrost distribution models. The distribution of permafrost in Northeast China generally follows a certain latitude zone pattern, which is a reflection of multiple factors, such as lithology, water content, vegetation, snow cover, surface water, groundwater, slope, aspect, and altitude. There are broad island permafrost regions [26,27,28]. Due to differences in the above factors, the mean annual ground temperature is also inconsistent and even has significant changes within small regions [29]. The MODIS LST products, as an important source of surface temperature data, are widely used in the top temperature of the permafrost model [30,31,32] and surface frost number model [17,18,25,33]. However, they still meet the application requirements for spatial resolution of permafrost distribution in basic engineering in cold regions.

The particle swarm optimization algorithm starts from a random solution and searches for the global optimal solution through iteration. It has the characteristics of fast iteration speed and fast convergence. Through this algorithm, the optimal solution can be selected from multiple environmental variables that affect the distribution of permafrost in Northeast China, with some environmental variables having a relatively large impact weight to characterize the probability of large-scale permafrost distribution and further optimize the surface frost number model. The spatial weight decomposition method is simple and easy to implement, with fewer adjustable parameters, and can achieve downscaling of the model. Thus, the two problems of limited parameters in the frozen soil distribution mapping model and low spatial resolution of data were solved. Under the influence of climate change and human activities, the global average surface temperature has increased by about 1 °C since 1850–1900 [34], and the warming effect is more significant in high-latitude and high-altitude areas [35,36,37,38], leading to permafrost degradation in these areas. The temperature in Northeast China has generally increased by 0.9~2.2 °C in the last 50 years, the southern boundary of the permafrost region moves northward, and the area continues to decrease [28]. Frozen soil is a product of the interaction and balance between the earth and air; its physical properties largely depend on the environmental temperature. When the heat absorbed by the soil layer exceeds that released at the interannual scale, the excess heat will cause the soil layer temperature to increase, leading to an increase in the thickness of the active layer or the melting of frozen soil [39,40,41,42]. On the contrary, it may cause thinning or refreezing. Therefore, with the temperature change, the permafrost area fluctuates in the short-term and decreases in the long-term. The dynamic changes of permafrost are manifested in the spatial distribution and thermal conditions of the active layer as well as fluctuations in the temperature of the permafrost layer and even the continuous changes of the southern and lower boundaries. Judging the presence or absence of permafrost at a 1 km resolution is of little use in guiding and informing engineering construction in cold regions. The relatively accurate permafrost temperature is also an important basis for judging the degree of the phase change of permafrost and evaluating the stable state of permafrost.

The objective of this paper is to produce large-scale and high-resolution permafrost distribution maps by downscaling existing data. First, the normalized difference vegetation index and forest canopy parameters, which represent the growth state and coverage degree of vegetation, are introduced into the surface frost number model to improve the accuracy of the model in the study area. Based on the 1 km resolution MODIS land surface temperature data, the distribution of surface frost number from 2003 to 2021 is obtained. Then, taking the geological investigation reports of six highways as sample data of permafrost distribution, a particle swarm optimization algorithm was used to obtain the index η with 30 m resolution and used to characterize the distribution probability of permafrost at the field scale. Index η consists of five geomorphic environmental variables: Slope position (SP), slope, deviation from mean elevation (DEV), topographic diversity (TD), and soil bulk density (SBD). Using the method of spatial weight decomposition, the process of reducing the surface frost number from a 1 km to 30 m resolution is realized by using η. Finally, the regression statistical relationship between the surface frost number and permafrost temperature after downscaling is established, the large-scale and high-resolution spatial distribution map of the permafrost temperature in Northeast China is simulated, and its spatial distribution characteristics and changes with time are analyzed.

2. Data and Methods

2.1. Study Region

The administrative scope of Northeast China includes Heilongjiang Province, Jilin Province, Liaoning Province, Hulunbuir City, Hinggan League, Tongliao City, Chifeng City, and Xilingol League of the Inner Mongolia Autonomous Region, with a total area of 147 × 10⁴ km². The geographical location is roughly 38°40′ N~53°30′ N and 110°45′ E~135°02′ E, with a north–south span of nearly 15° and an east–west span of nearly 20°; it is also located at the northernmost end of the East Asian monsoon, which is affected by alternating high and low pressures and monsoons in the inland and ocean. The winter is cold, dry, and long, and the summer is hot and rainy. It belongs to the temperate continental monsoon climate [43,44]. The climate is gradually stronger from east to west, the annual precipitation drops from 1000 mm to less than 300 mm from southeast to northwest, and it transits from humid and semi-humid areas to semi-arid areas.

The geomorphic types in the study area vary from flat coastal plains to undulating mountains, with elevations ranging from 0 to 2500 m. The terrain is dominated by plains, hills, and mountains, with the Changbai Mountains in the southeast, Lesser Khingan in the northeast, and Greater Khingan in the northwest. Between the hills, the Northeast Plain and the Inner Mongolian Plateau are distributed. The terrain is generally distributed in a semi-ring shape, high in the northeast, southeast, and northwest and low in the south [45,46]. The spatial distribution of vegetation types is closely related to the terrain. The vegetation of the Greater Khingan Mountains is mostly coniferous forest; the Lesser Khingan Mountains and Changbai Mountains are coniferous, mainly broad-leaved mixed forest; the southern Lesser Khingan Mountains and part of the south of Greater Khingan Mountains are mostly broad-leaved forest; and the western and southwestern slopes of the Greater Khingan Mountains are typical concentration areas of grassland and grassland meadow. Due to the temperature inversion layer in winter and the surface covering vegetation (moss, grass cover, or forest) having significant effects on the hydrothermal process and spatial distribution of permafrost [47,48,49,50,51], the permafrost is generally developed in the mountain depressions, valley terraces, and swamps covered by mud carbon.

2.2. Monitoring Data

The temperature data of the permafrost layer are from the ground temperature monitoring data of the Ministry of Education Observation and Research Station of Permafrost Geo-Environment System in Northeast China (MEORS-PGSNEC). There are 18 monitoring sections (47 boreholes) with the monitoring time from 2016 to now, and the depth range is 5~21.5 m, located in Genhe to Labudalin section of National Highway G332, Genhe to Mangui section of provincial highway S204 of Inner Mongolia, Walagan to Zhangling section of national highway G111, Bei’an to Heihe section of national highway G1211, and Yiershi to Chaiqiao section of provincial highway S308 of Inner Mongolia, respectively. To avoid the interference of the monitoring equipment installation process on the potential heat of the ground temperature distribution, the data of the second year after starting monitoring is selected. The temperature of the permafrost layer is selected to be 1 m below the base of the permafrost active layer. Figure 1 shows the geographical location of boreholes.

2.3. Surface Frost Number Model

The surface vegetation is densely covered in Northeast China, and the seasonal and spatial distribution of vegetation determines the surface temperature distribution. Therefore, the surface vegetation factor E_t (Equation (1)), based on the normalized differential vegetation index (NDVI) and forest canopy, which represent the growth state and coverage degree of vegetation, was added to the surface frost number model (Equation (4)) [24,25]. DDT was the surface thawing index, which was calculated from the daily cumulative value of surface temperature greater than 0 °C from 1 January to 31 December of the current year, as shown in Equation (3); DDF was the surface freezing index, which was calculated from the cumulative value of the daily absolute value of surface temperature less than 0 °C from 1 July of each year to 30 June of the next year, as shown in Equation (2). The influence of clouds is a major problem in optical/thermal infrared satellite data. MODIS land surface temperature (LST) daily products may be missing part of their values due to cloud coverage, but it is very unlikely that there are clouds in the same area for 8 consecutive days, and the correlation between remote sensing LST and ground temperature of weather stations is significantly better on the time scale of long period [52,53]. Therefore, MODIS LST 8-day synthesis products MOD11A2, MYD11A2, and NDVI monthly synthesis products MOD13A3 were selected, with spatial resolutions of 1000 m.

The surface vegetation factor E_t is shown in Equation (2):

$$E_t={\epsilon }_{fcc}\times \left({\left({MANDVI}_G\right)}_t+1\right),$$

(1)

where t is the year (t = 2003, 2004, ……, 2020) and ε_fcc is the forest canopy closure. According to the stand spatial structure [54,55] of the study area and the regulations of FAO, the average ε_fcc in the study area is moderate canopy closure with a range from 0.2 to 0.69. The ε_fcc was taken as a constant value of 0.56 [25] for convenient calculation. The mean annual value of NDVI in the growing season (MANDVI_G) was selected to characterize the overall condition of interannual vegetation cover.

$$DDF=\sum _{182}^n\left|{\overline{LST}}_t\right|+\sum _1^{181}\left|{\overline{LST}}_{t+1}\right| \left(\overline{LST}<0 °\mathrm{C}\right)$$

(2)

$$DDT=\sum _1^n\overline{LST} \left(\overline{LST}>0 °\mathrm{C},n\le 365\right)$$

(3)

We defined the surface frost number (SFn) under the influence of E_t-SFnc, as shown in Equation (4) [25].

$${SFn}_C=E_t\times \left(\frac{\sqrt{DDF}}{\sqrt{DDF}+\sqrt{DDT}}\right),$$

(4)

where $\overline{LST}$ was the average value of MOD11A2 and MYD11A2, n was the number of days corresponding to the image, and t was the year.

2.4. Surface Frost Number Downscaling

The distribution characteristics of permafrost are influenced by environmental variables, such as climate, geology, topography, surface cover, and soil properties [56,57]. Therefore, evaluating the permafrost distribution should consider its formation mechanism and performance characteristics from multiple perspectives [37]. The application of remote sensing to obtain environmental variables, such as land surface temperature, vegetation, snow cover, organic layer, and soil characteristics, has been applied to the practice of permafrost mapping [58]. In addition, the resolution of variable data is also a key consideration in high-resolution permafrost mapping research, where its resolution should be similar to the results. In this study, particle swarm optimization was first used to obtain multiple environmental variables that represent the relative probability of permafrost distribution at the field scale η. Then, the spatial weight decomposition method to achieve the downscaling process of downscaling to 30 m resolution was based on the 1 km surface frost number and indicators η.

Environmental variables that will not undergo significant changes over time in a short period and can also reflect the specificity of permafrost regions are selected. Including soil organic content (SOC), coarse fragment content (CFC), and soil bulk density (SBD), which are related to soil properties and obtained from Soil Grids250 (https://soilgrids.org (accessed on 1 January 2023)) with a spatial resolution of 250 m. The variables were calculated from seven standard depths (0, 5, 15, 30, 30, 60, 100, and 200 cm) using a weighted average of depth intervals. The environmental variables were associated with terrain including the Digital Elevation Model (DEM), slope, aspect, roughness (RN), slope position (SP), solar radiation (SR) [59,60], deviation from mean elevation (DEV) [61,62,63], and Topographic Position Index (TPI) [61,62]. DEM data (SRTMGL1_003) are freely available from NASA’s Land Process Distributed Active Archive Center, and the other variables were obtained through topographic analysis. The variables were associated with ecological parameters, including Continuous Heat-Insolation Load Index (CHILI), multi-scale Topographic Position index (mTPI), and Topographic Diversity (TD) [64].

The environment variables were combined with basic function (logarithmic function, trigonometric function, exponential function, power function) relation. Then, an index η representing the relative probability of field scale permafrost distribution was obtained with particle swarm optimization algorithm. The index is expressed by Equation (5), which contains 5 environmental variables, namely SP (Equation (6)), Slope, DEV, TD, and SBD, and their spatial distribution is shown in Figure 2.

$$\mathsf{\eta }=1-\mathrm{S}\mathrm{P}-5\times \sin \left(\mathrm{S}\mathrm{l}\mathrm{o}\mathrm{p}\mathrm{e}\right)-3\times \mathrm{D}\mathrm{E}\mathrm{V}-1.5\times \mathrm{T}\mathrm{D}-1.5\times \sqrt{5}\times \sqrt{\mathrm{S}\mathrm{B}\mathrm{D}},$$

(5)

where SP is used to indicate the relative position of any point on the slope on which it is located; η is the dimensionless variable.

$$\mathrm{S}\mathrm{P}=\frac{\mathrm{D}\mathrm{E}\mathrm{M}-{\mathrm{D}\mathrm{E}\mathrm{M}}_{\mathrm{m}\mathrm{i}\mathrm{n}}}{{\mathrm{D}\mathrm{E}\mathrm{M}}_{\mathrm{m}\mathrm{a}\mathrm{x}}-{\mathrm{D}\mathrm{E}\mathrm{M}}_{\mathrm{m}\mathrm{i}\mathrm{n}}},$$

(6)

where DEM_max is the highest elevation of the slope where it is located; DEM_min is the lowest elevation of the slope on which it is located.

The index η was applied to downscale the surface frost number of 1 km resolution using the spatial weight decomposition method. Its model is shown in Equation (7), and the specific downscaling workflow is shown in Figure 3.

$${SFn}_{c\_30m}={SFn}_{c\_1km}\times \frac{{\mathsf{\eta }}_{\mathrm{a}}}{{\mathsf{\eta }}_{\mathrm{b}}},$$

(7)

where ${SFn}_{c\_30m}$ and ${SFn}_{c\_1km}$ are surface frost number with resolutions of 30 m and 1 km, respectively; η_a is the pixel value of pixel a in the image η with 30 m resolution, η_b is the average value of η within pixel b at a resolution of 1 km surface frost number where pixel a is located, and $\frac{{\eta }_a}{{\eta }_b}$ constitutes spatial weights for downscaling.

2.5. Surface Frost Number and Permafrost Temperature

The SFn_{c_30m} after downscaling had the possibility of distinguishing the permafrost temperatures and SFn_{c_30m} at different monitoring positions on the same cross-section. The latest ground temperature monitoring data began in 2018, and SFn_{c_30m} was selected during the same period (2018–2021). Finally, a total of 45 sets of monitoring data were used, and the linear fitting results are shown in Figure 4. The linear regression equation is shown in Equation (8):

$$T_p=-32.103\times {SFn}_{c\_30m}+16.254,$$

(8)

where T_p is the permafrost temperature and SFn_c is the surface frost number.

3. Results and Tests

3.1. Spatial Distribution and Variation Characteristics of Permafrost Temperature

The permafrost temperature from 2003 to 2021 is obtained according to Equation (7). In order to reflect the spatiotemporal dynamic changes of permafrost over the past 20 years, the period 2003–2021 is divided into four periods, namely 2003–2007, 2008–2012, 2013–2017, and 2018–2021, and their distribution is shown in Figure 5. The corresponding permafrost areas are 28.35 × 10⁴ km², 35.14 × 10⁴ km², 28.96×10⁴ km², and 25.21×10⁴ km² (excluding lakes), respectively. The area of different permafrost temperatures at intervals of 0.5 °C was plotted, as shown in Figure 6. The proportions of the total area of extremely stable permafrost (<−5.0 °C), stable permafrost (−3.0~−5.0 °C), sub-stable permafrost (−1.5~−3.0°C), transitional permafrost (−0.5~−1.5 °C), and unstable permafrost (>−0.5 °C) [65] are 0.50–1.27%, 6.77–12.45%, 29.08–33.94%, 34.52–39.50%, and 19.87–26.79%, respectively (in

Table 1. Area of different permafrost stability in Northeast China.

Permafrost Stability Types	Extremely Stable	Stable	Sub-Stable	Transitional	Unstable	Area (×104 km2)	Minimum Tp (°C)	Mean Annual Air Temperature (°C)
2003–2007	0.27	2.90	9.25	10.14	5.79	28.35	−13.3	3.2
2008–2012	0.45	4.37	11.93	12.13	6.98	35.14	−13.9	2.6
2013–2017	0.18	2.34	8.98	10.94	7.76	28.96	−13.1	3.1
2018–2021	0.13	1.71	7.33	9.96	6.09	25.21	−12.9	4.0

). The permafrost temperature is mainly of sub-stable, transitional, and unstable types, and the sub-stable permafrost is mainly distributed in the temperature range of −1.5 °C to −2.0 °C. In general, there is still extremely stable and stable permafrost in Northeast China on a 30 m resolution. Although it only accounts for 7.27% to 13.72% of the total area, it is mainly distributed in the Daxing’anling regions, from Mohe at the northernmost end to Aershan Mountain in the middle of the ridge and also some in Xiaoxing’anling regions, Changbai Mountain, and Shuangfeng Forest Farm. Other stable types of permafrost coexist and are distributed throughout permafrost regions.

The distribution area changes of permafrost within different temperatures are shown in Figure 6. The permafrost area changes dynamically with time, but the overall trend is still downward. The area of other stable types of permafrost also shows a decreasing trend and the same dynamic change pattern as the total area, except for a slight increase in the distribution area of unstable permafrost. Compared with 2003–2007, the area of transitional and unstable permafrost increased from 2013 to 2017, and only the area of unstable permafrost increased from 2018 to 2021. This indicates that, overall, the temperature of the permafrost is increasing and its stability is decreasing.

3.2. Indirect Test

3.2.1. The Relative Accuracy of Permafrost Temperature

The monitoring data and simulation data were compared and analyzed to verify the accuracy of the simulation results on a single-point scale, as shown in Figure 7. The correlation coefficient between them is 0.7882, the root-mean-square deviation is 0.1929, and 83.2% of the residual is controlled within ±1 °C. These values indicate that downscaled large-scale simulations of the permafrost temperature still have a high accuracy. There are some uncertainty factors that are partly responsible for the source of the error. On the one hand, it can be seen from Figure 1 that the spatial distribution of monitoring stations covers the Greater and Lesser Khingan and Yilehuli Mountains, including all types of permafrost regions. Although the ground temperature monitoring data of 47 boreholes have been used in the simulation process, which is much higher than the amount of data used in the previous research on permafrost distribution in the study area, the monitoring stations used in the study are mainly linear distribution and lack dispersion in terms of local distribution. In addition, the monitoring stations are mainly located in the areas with elevations of 200~1000 m and lack monitoring data in higher elevation areas. Furthermore, the sporadic permafrost is widely distributed from the Songnen Plain to the Xiaoxing’anling regions. The permafrost may melt during the drilling process, and therefore, there is also a lack of monitoring sites in the permafrost transition and degradation region. On the other hand, it can be seen from Figure 7a that the residuals are larger in the relatively high-temperature interval (−0.5 to 0 °C) and that the spatial heterogeneity of permafrost is stronger in the relatively high-temperature interval, leading to some representativeness errors for individual boreholes compared to the lower-temperature interval and at the scale of the 30 m grid cells used in the study.

The optimal location for monitoring ground temperature is the field with less interference from human activities. However, more monitoring stations are based on linear engineering, the permafrost temperature increases after the disturbance, and the monitoring data reflect the permafrost temperature in the temperature field after reaching the new equilibrium.

There is also some uncertainty in the environmental variables used in the downscaling process. For example, there is little ground survey data of the soil bulk density data in the study area; these data errors will also have a certain impact on the simulation results. The resolution of the data is reduced from 250 m to 30 m after resampling, and certain errors will be generated in the resampling process. In addition, the environmental variables, such as the aspect, are difficult to quantify and assign weights to and are not considered in the downscaling process. The scientific and technical challenges of further breakthroughs in more accurate ground temperature simulations in the future will depend on further data accumulation, especially in improving the spatial representativeness of boreholes in the permafrost region of Northeast China.

3.2.2. The Relative Accuracy of Permafrost Range

At the regional scale, the simulation results from 2018 to 2021 are compared with the distribution of permafrost sections in six linear engineering geological investigation reports, as shown in Figure 8. There are no temperatures in the reports, so the comparison focuses more on the consistency of distinguishing the range of permafrost sections at the regional scale. Figure 8b shows a total of 48 permafrost sections in investigation reports, accounting for nearly 50% of the road length, which is from the Genhe to Mangui section of the provincial highway S204. The road passes through relatively low-terrain valleys. The simulation results show that the permafrost within the road area is continuously distributed, with temperatures mainly between −4.5 °C and −2 °C. Some road sections have melting zones. Figure 8c shows a total of 14 permafrost sections in the investigation reports, which are from Shiwei to Labudalin on National Highway G331. The permafrost temperature within the road area has a large span. Near Shiwei and Labudalin, there are permafrost degradation areas, and the permafrost temperature is mainly between −1.5 and −0.5 °C, with only nine permafrost sections within the 90 km road and a length of less than 500 m of each section. There are 27 permafrost sections with about 1 km length of each section in the middle part of the 60 km length road, and the permafrost temperature is between −3.5 and −2 °C. Figure 8d shows that there are 21 permafrost sections in the investigation reports, which are the Yiershi to Chaiqiao section of provincial Highway S308. The temperature span of permafrost in the road area is large, mainly between −3.5 °C and −0.5 °C. The permafrost is distributed in the form of islands near the Yiershi road area, the temperature is mainly between −2 and −0.5 °C, and the length of the permafrost section is less than 400 m. However, there are many melting areas in the road area near Chaiqiao, and the permafrost temperature is relatively low, which is mainly between −3.5 °C and −1.5 °C. The longest permafrost section can reach 1.9 km. Figure 8d shows there are 34 permafrost sections in the investigation reports, which are the Kubuchun Forest Farm to the Genhe sections of National Highway G332. About 2/3 of the road is distributed in the valley area with relatively low terrain; the permafrost is widely distributed within the road area with a relatively low temperature ranging from −3.5 °C to −0.5 °C. The road passes through a number of melting zones. Figure 8e shows that there are 14 permafrost sections in the investigation reports, which are the Jiageda to Changqing Forest Farm sections of National Highway G331. The road passes through a relatively low-terrain valley area, mainly located in the permafrost degradation area, with permafrost temperatures ranging from −2.5 °C to −0.5 °C. A longer permafrost section corresponds to relatively low temperatures (around −2 °C). Figure 8f shows there are 129 permafrost sections in the investigation reports, which are the Walagan to Xilinji sections of National Highway G111. Taking the Zhangling section in the middle as the boundary, the Xilinji to Zhangling sections are mainly located in a continuous permafrost distribution area, with island-shaped melting areas along the road. The length of the permafrost section is relatively long, and the temperature within the road area is between −2.5 °C and −1 °C. The Zhangling to Walagan sections are mainly located in a permafrost degradation area, with a relatively short length of permafrost sections and a temperature of −1.5 °C to −0.5 °C within the road area.

A comparison of the simulation results and the investigation reports of the distribution permafrost section shows that the total length of the simulated permafrost sections is larger than that of the reports, which is limited by the number of boreholes, and the way single-point boreholes cannot fully reflect the distribution of permafrost along the road, especially for the identification of permafrost that has a sporadic or island distribution.

Table 2. Comparison of permafrost distribution between investigation data and simulation results.

No.	Location	Road Length (km)	Number of Boreholes	Drilling Depth (m)	Permafrost Section Length (km)	Drilling Time	Same Length as Investigation Data	Consistency
1	Walagan–Xilinji section of National Highway G111	157.01	1112	3.5~40	47.26	September. 2017	45.59	96.47%
2	Genhe–Mangui section of provincial highway S204	255.00	522	5~40	112.98	May. 2020	108.04	95.63%
3	Jiageda–Changqing Forest Farm section of National Highway G331	52.85	161	5~25	11.52	May. 2020	10.16	88.19%
4	Kubuchun Forest Farm–Genhe section of National Highway G332	59.69	382	5~35	34.58	December. 2020	30.23	87.74%
5	Shiwei–Labudalin Highway section of National Highway G331	153.74	425	5~40	29.01	July. 2012	24.14	83.21%
6	Yiershi–Chaiqiao section of provincial highway S308	87.08	243	5~40	5.35	July. 2017	4.45	83.17%

shows that the consistencies of six roads are up to 96.47% and down to 83.21%. From a local perspective, the differences are mainly reflected in the following aspects. In island permafrost regions, the determination of whether permafrost exists in a road section can be consistent, but there are differences in the determination of the start and end points of the permafrost section, as shown in Figure 9a. In continuous permafrost regions, there are often a number of melting zones distributed in the permafrost sections with long lengths and relatively low temperatures. Drilling surveys missed out on melting zones, resulting in differences as shown in Figure 9b. In permafrost degradation regions, some sporadic permafrost was missed through drilling surveys, as shown in Figure 9c. Simulation results show that there are several permafrost sections with relatively high temperatures and less than 1 km length on the road due to different drilling locations missing the determination of sporadic permafrost distribution. The permafrost section distribution of investigated and simulated results are matched best in the region of temperatures below −2 °C. There are some slight differences in the melting zones, the transition zones, and the edge of the permafrost zones, with temperatures between −1 and 0 °C. The permafrost sections in the investigation reports were formed by connecting boreholes, which were not evenly distributed. Thus, there is a limitation to fully reflecting the permafrost distribution through the survey. In addition, the simulation results show more significant topographic features in the gully area, which is influenced by the downscaling parameters and individual pixels of the remote sensing data.

Since the end of the Tertiary and the beginning of the Quaternary, the Greater and Lesser Xingan Mountains have been under slow differential uplift on the basis of old structures [66]. Under the action of long-term denudation and planation, the loose sediments on the mountaintops and hillsides are very thin, with generally only 1~2 m eluvium. Mountain stream basins and gullies are the main accumulation areas, and the thickness of the loose sediments often reaches about 10 m. The sediments are sub-sandy soil, sub-clay, sand gravel, etc., and the surface layer has a large thickness of peat and peat, which becomes a good occurrence condition of island permafrost. Therefore, island permafrost develops in river terraces, gully marshes, and wetlands. Due to the influence of natural factors, such as temperature inversion, differences in the type and thickness of the loose layer, vegetation, slope direction, surface water, and groundwater, permafrost has spatial differences in different topographic and geomorphologic parts of the same area. The simulation results show that the island permafrost in the gully region presents an irregular network distribution at a spatial resolution of 30 m after scaling down, which is spatially correlated with the elevation (Figure 10b) and surface temperature (Figure 10c). By a comparison with Figure 10d, it can be found that the influence of human activities and the thawing of permafrost lead to multiple ground subsidence, which is also spatially correlated with the permafrost temperature. Although the resolutions of the two are different, the spatial correlation is significant with the thermal infrared surface temperature’s low value and the surface thaw settling boundary, which shows the distribution characteristics of permafrost in a typical island frozen soil region [67].

4. Discussion

4.1. Permafrost Area in Northeast China

Permafrost maps widely used in different periods in Northeast China have been collected, which are summarized in

Table 3. Permafrost area from different sources in Northeast China.

Permafrost Map Name	Area (104 km2)	Age	Source
Distribution of permafrost in Northeast China	40~50	1950s	Xin KuiDe, Ren QiJia [69], 1956
Division of permafrost regions in Daxiao Hinggan Ling Northeast China (1:200,000)	38.67	1970s	Guo Dongxin, et al. [27], 1981
Map of snow, ice, and frozen ground in China (1:400,000)	38–39	1990s	Shi Yafeng, Mi Desheng [70], 1988
Geocryological regionalization and classification map of the frozen soil in China (1:1000,000)	39	1990s	Zhou Youwu, et al. [71], 2000
Map of the Glaciers, Frozen Ground and Deserts in China (1:400,000)	39	1980–1990s	Nan zhuotong, et al. [72], 2002; Wang Tao [73], 2006
Frozen soil map of China	24.0	2000s	Ran Youhua, Li Xin [2] (2018)
No Name	25.7	2000s	Wei Zhi, et al. [74]
No Name	27.10	2003–2019	Shan, et al. [24]

. These results were published at different times, and the area changes reflect information about permafrost changes. However, considering the accuracy and classification criteria used in the early permafrost maps [68], there are differences between them and the maps adopted by GIS technology and integrated with remote sensing observation data after 2000. In the context of climate warming, there are still no long-term and large amounts of direct evidence to show whether the degree of permafrost thawing has led to the large-scale disappearance of the permafrost. Therefore, it is not compared with the permafrost area of earlier studies, and concerning the research results after 2000, they are also based on remotely sensed surface temperature observation data. In this paper, the influence of vegetation is taken into account in the permafrost mapping model, which improves the applicability of the surface frost number model in Northeast China. At the same time, with the help of more extensive ground observation data and the downscaling of the frost number through multiple environmental variables, the permafrost temperature maps with a spatial resolution of 30 m are obtained. Compared with the relevant research results of the same period, the simulated permafrost in this paper is more widely distributed and the area is larger. In addition to the differences caused by different observation times, the downscaling process includes both topographic factors and superimposed soil information, which is more detailed than the previous description of permafrost distribution, and also highlights the details ignored by 1 km remote sensing data. Therefore, the 30 m permafrost area is slightly larger but closer to the actual spatial distribution of permafrost.

The warming effect is amplified at high latitudes, and the regional degradation of permafrost [75] leads to an overall decreasing trend in the permafrost area in Northeast China. The response of permafrost to air temperature change has a significant hysteresis on the decadal and monthly scales, and the change of ground temperature lags behind the change in air temperature. With the deepening of the depth, the response of the ground temperature to air temperature change gradually lags behind; this hysteresis is not obvious on the interannual scale. Due to local conditions, the permafrost temperature in individual regions hardly changed and even tended to cool down [76]. The transformation of some seasonal permafrost into permafrost may be manifested as an increase in permafrost area [77]. Therefore, on a short time scale, the distribution pattern of permafrost shows a small fluctuating change within a certain range with the influence of environmental factors.

4.2. Analysis of Permafrost Temperature Distribution

We outlined three ridgelines along the Greater and Lesser Hinggan Mountains and Changbai Mountains and extended a length, as shown in Figure 11. The temperature of the pixel points to the corresponding elevation, and the latitude of the extracted curve is drawn as shown in Figure 12. To maintain the consistency of elevation, latitude, and temperature data and facilitate comparison, ground temperature data greater than 0° in the simulation results are also retained for recent reference and not used for discussion.

From the correlation coefficients between the permafrost temperature and elevation and latitude, it can be seen that the permafrost temperature in the Daxinganling region is mainly affected by elevation, such as Arshan and Huanggangliang, which have an altitudinal advantage in the same latitudinal range and have even lower temperatures. The latitude has a more significant effect on the permafrost temperature in the Xiaoxinganling and the Changbai Mountain range. For example, there is island permafrost on the Sanjiang Plain; the “Snow Town” is at the same latitude as Huanggangliang and also has the advantage of altitude and lower permafrost temperatures compared to the surrounding area. The correlation of the permafrost temperature with elevation and latitude in different regions also shows that the temperature in Northeast China is dominated by many factors, and elevation and latitude are not the absolute main factors. This also suggests that a more accurate and spatially resolved mapping of permafrost distribution may require geographic zoning simulations.

Drawing on the permafrost mapping method of an equivalent latitude and equivalent elevation, it was based on the regression relationship between the permafrost temperature and elevation and latitude (the formula is shown in Figure 12). The projected permafrost temperature (Tpp) corresponding to the three curves (the light blue curves in Figure 12) has a large discrepancy with the simulation results, and it does not reflect the variability of the permafrost temperatures in the region. At the same time, it is not able to characterize the temperature-reachable limit value. This method has also been used to simulate the annual mean ground temperature distribution on the Tibetan Plateau [72], but it has been shown that the mapping method that mainly relies on the elevation and latitude has a large amount of uncertainty. Permafrost temperatures may be underestimated and overestimated at the Sanjiang Plain with relatively low elevation and the northern foothills of the Daxing’anling Mountains with the highest latitude. Simulation results show that the lowest temperature can be as low as −12.9 °C, which has not been obtained in the existing permafrost observation. Therefore, the extremely low-temperature region still needs further verification, and we will also add it in future research.

4.3. The Southern Boundary of Permafrost Region in the Northeast

The line connecting the southernmost edge of the permafrost island is called the southern physical geographic boundary of the permafrost region. According to a large number of paleo periglacial phenomena found in the middle and late stages of the last glacial period in North China, Ordos Plateau in Inner Mongolia, and Northeast China, the southern boundary of the permafrost at a high latitude in China at the peak of the last glacial period is preliminarily determined, as shown in Figure 13b. It is from Dalian (39° N) to Beijing and Datong (40° N), protrudes south in the Ordos area to the vicinity of Yulin and Jingbian (37° N), and enters Jiuquan and Dunhuang (39° N~40° N) of the Hexi Corridor through Zhongning (37.4° N). North of this line, the permafrost developed in the late Pleistocene [78]. By the 1850s, according to the existing survey and meteorological data and with reference to the permafrost distribution line between the Soviet Union and the Mongolian People’s Republic, the first southern boundary of permafrost in Northeast China was preliminarily determined. It was from Ateka and Huduke and Nuotuoaili in the Mongolian People’s Republic to NanXing ‘an in China, passing the south of BuhaQi and Boligen and, after passing through the placer gold lands of Buteha, south of Boligen, Dedu, and Duru rivers. It crossed the Amur into the Soviet Union [69]. In the 1970s, field investigations were carried out in Jiayin, Dedu, Arshan, and New Balhu Right Banner; the northern Greater Khingan Mountains; the Hulunbuir Grassland; the Changbai Mountains; and Huanggangliang. Based on a large number of investigation data, the southern boundary of the permafrost in Northeast China was revised. Although the mean annual temperature of the Mongolia Plateau is lower than 0 °C, there is no relevant data to prove the existence of permafrost. In the area north of the left and right Banner of New Balhu in Hulunbuir Plateau, there is sporadic permafrost distribution in the low-lying and humid zone, and there is no permafrost in the vast area between the Kelulun River and the Wuerxun River. The island permafrost is distributed in the northern Songnen Plain. The southern boundary of permafrost is determined to be Haraha River–Arshan–Wuchagou–Chaihe–north of Arong Banner–Laolai–Dedu–Qing’an–Tiansheng–Jiayin [77], as shown in Figure 13a. At this time, the southern boundary is not horizontally distributed along the latitudinal direction, mainly due to the influence of topography and mountain trends. Permafrost develops in Arshan Mountain in the middle of the Greater Khingan Mountains (elevation of 700–1200 m) and the southern of the Lesser Khingan Mountains (elevation of 400–700 m) with a higher elevation than the neighboring areas and superimposed with latitude factors. The southern boundary protrudes southward along the two ridgelines with a “W” shape. According to the simulation results, Huanren County, Liaoning Province, which is about 300 km southwest of Changbai Mountain (Figure 13c), also has sporadic permafrost, and the temperature can reach a minimum of −3.4 °C. In areas south of the traditional southern boundary, the roads in Jilin and Liaoning provinces are affected by icing diseases [79,80], which confirms that the southern boundary of the permafrost in the northeast is more southerly than previously recognized, indicating the reliability of the results in this paper. High-precision permafrost data can better describe the permafrost distribution in remote areas, which previous studies have failed to focus on.

4.4. Application Prospect of 30 m Resolution Permafrost Map

The simulation results of this study synthesize the current high-quality remote sensing land surface temperature data with both temporal and spatial resolution. Under the condition of the occurrence of regional permafrost, the normalized differential vegetation index and forest canopy are introduced into the surface frost number model to characterize the vegetation cover. The field permafrost probability index represented by multiple parameters is obtained to achieve a downscaling process of surface frost number. At the same time, the permafrost monitoring data in several regions were used. The distribution map with relatively high accuracy and higher spatial resolution is used for the planning, design, ecological planning, and management of engineering in cold regions to improve the social adaptability of perennial permafrost degradation in the context of global warming. First, the refined permafrost distribution map is the basic data for engineering planning and design in terms of coping with engineering geohazards. Table 1 shows that the study area is distributed and dominated by sub-stable, transitional, and unstable types, with temperatures mainly between −2 to 0 °C. The permafrost with lower temperatures thawed and transitioned to the less stable permafrost. Permafrost degradation induces disasters, such as frost heave and thaw collapse, which adversely affect the operation of regional highways, railroads, oil pipelines, and airports and increase maintenance costs. Therefore, it is important to make a prediction of the changing trend based on the existing permafrost distribution when planning for engineering construction. The land should be rationally utilized, and active engineering measures to enhance the stability of the roadbed and reduce engineering disasters should be taken. When confronted with permafrost with poor stability and high temperature, the traditional design concepts and methods of protecting permafrost foundations are not adapted to the highway construction and maintenance practice in the northeast permafrost area. A new type of roadbed structure based on the concept of destroying permafrost design and the thermodynamic characteristics of the foundation structure of the block layer, which uses the large-grained block layer to refill the permafrost layer of the foundation, can reduce the total heat that enters the deep foundations and unfrozen water content migrating from the deep foundation to the active layer. It can also increase the hydrothermal convection flux inside and outside the block layer, thus increasing the upper limit of the permafrost and reducing the deformation of the roadbed. These measures have been successful in the engineering practice of the Walagan to Xilinji section of National Highway G111 and Yiershi to Chaiqiao section of provincial highway S308 in the Inner Mongolia Autonomous Region. Second, permafrost degradation is closely related to groundwater recharge, runoff, and discharge and has a controlling or important influence on the hydrogeological environment at various scales, which is the basis for maintaining water balance in the permafrost region. For example, thawing permafrost may reduce the lake area by creating drainage channels leading to surface water decline and increasing soil erosion in lakes [82]. Furthermore, the permafrost is an important reservoir of organic carbon and plays a very active role in the global carbon cycle. The permafrost thawing leads to an accelerated release of organic carbon from the soil layer [83,84], and the increase in greenhouse gas concentrations is closely related to global warming [85]. In addition, permafrost thawing processes increase the frequency of forest fires [36,86], and the loss of forests may accelerate the permafrost thaw, leading to an increased release of soil carbon deposited over many years, triggering a warming feedback loop [87]; the degradation of the permafrost due to forest fires may be irreversible [88]. Therefore, the permafrost temperature distribution map with a 30 m resolution can be used to support regional engineering construction, rational planning of structure and function, and disaster management as well as being the basic data for evaluating permafrost change and ecohydrology research. With the accumulation of remote sensing data and ground monitoring data, permafrost mapping will develop towards demand on a large-scale and with high precision. This study is an attempt to move the mapping towards the large-scale with 30 m resolution, and it also breaks the barriers of traditional mapping thinking. Meanwhile, some problems are reflected in the process of the study to improve the accuracy through more quantifiable factors.

5. Conclusions

The probability index η of permafrost distribution at the field scale, which is characterized by five environmental variables affecting permafrost occurrence, is obtained using the particle swarm optimization algorithm. The surface frost number with a spatial resolution of 1 km is reduced to 30 m through the descending method of spatial weight decomposition. The linear relationship between it and the ground monitoring data is established, and the 30 m resolution permafrost temperature distribution map in Northeast China from 2003 to 2021 is simulated. The following conclusions can be drawn from the simulation results:

• During 2003–2007, 2008–2012, 2013–2017, and 2018–2021, the permafrost area is 28.35 × 10⁴ km², 35.14 × 10⁴ km², 28.96 × 10⁴ km², and 25.21 × 10⁴ km², respectively (excluding lakes).
• The proportion of permafrost of extremely stable, stable, sub-stable, transitional, and unstable types is 0.50~1.27%, 6.77~12.45%, 29.08~33.94%, 34.52~39.50%, and 19.87~26.79%, respectively. The permafrost area shows a decreasing trend, the temperature shows an increasing trend, and stability declines.
• The permafrost temperature in the Greater Khingan Mountain is mainly affected by elevation while the latitude has a more significant effect on the permafrost temperature in the Lesser Khingan Mountains and Changbai Mountain.

The results of this paper can be used to support the planning, design, and ecological planning and management of cold region projects and can be used as a data benchmark for evaluating the future permafrost changes in Northeast China.