A permafrost state gradient identification method based on thermal state prior probability

Abstract

The application discloses a permafrost state gradient identification method based on thermal state prior probability, comprising the following steps: step 1, multi-source data acquisition, unified gridding and physical preprocessing; step 2, snow correction thermal state set simulation and thermal state prior probability generation; step 3, annual prior probability fusion, long-term deviation learning and target time window probability reconstruction; step 4, active layer structure response simulation and thermal state-structure response basic discrimination; and step 5, time perception dynamic correction, one-way probability conservative correction and spatial post-processing. The application integrates annual average ground temperature, active layer thickness and permafrost occurrence probability into a unified discrimination framework, and overcomes the limitations in the prior art that thermal state simulation, probability mapping and active layer thickness calculation are independent of each other and are difficult to be combined to serve engineering discrimination.

Description

Technical Field

[0001] This invention relates to the field of permafrost state identification technology, and in particular to a method for permafrost state gradient identification based on thermal state prior probability. Background Technology

[0002] Identifying the state of permafrost is fundamental for site selection, foundation design, and stability evaluation in cold-region engineering projects. Accurately understanding the spatial distribution, thermal state, and changing trends of permafrost directly impacts the rationality of building foundation schemes, the long-term service safety of infrastructure, and the prevention and control of engineering disaster risks. For a long time, the classification and mapping of permafrost has primarily relied on its continuous distribution on a plane, such as continuous permafrost, discontinuous permafrost, scattered permafrost, and island permafrost. Mapping methods have evolved from empirical models to physical models, and then to statistical learning and multi-source data fusion. Currently, commonly used methods include thermal state simulation based on the TTOP model, probabilistic mapping based on remote sensing and machine learning, and simulation of active layer thickness using various process models.

[0003] However, the aforementioned classification and mapping methods primarily focus on describing the geographical distribution of permafrost, rather than meeting the needs of engineering applications. Classification systems based on distribution continuity are too macroscopic and fail to reflect the differences in thermal state and structural response characteristics within permafrost. For example, even within discontinuous permafrost regions, there may be stable permafrost with temperatures below -2°C and high-temperature unstable permafrost with temperatures near 0°C, exhibiting fundamental differences in foundation bearing capacity and thaw settlement risk. Existing permafrost classification methods mainly rely on multi-year average ground temperature for stability grading, failing to reflect the interannual dynamic changes of the permafrost system. Furthermore, existing mapping methods typically treat the existence probability, thermal state, and active layer thickness of permafrost as independent outputs, lacking a framework that organically integrates these three factors to perform engineering-scale state grading of permafrost.

[0004] This crude approach to state identification poses real risks to engineering construction in cold regions. In engineering design practice, misclassifying high-temperature unstable permafrost as stable permafrost may underestimate the risk of foundation thaw settlement, leading to uneven foundation settlement or even structural damage. Conversely, misclassifying stable permafrost as non-permafrost or high-temperature permafrost may result in over-design and unnecessary increased engineering costs. With global warming, the ground temperature in mid- and high-latitude permafrost regions continues to rise, and the active layer thickens, making the state gradient within permafrost more complex. Traditional binary classification (permafrost and non-permafrost) or zoning methods based on distribution continuity are no longer sufficient to meet the practical needs of refined design and disaster prevention in cold region engineering.

[0005] To address the above problems, this invention provides a method for identifying the state gradient of permafrost based on thermal state dominance and probabilistic conservative correction. Summary of the Invention

[0006] The purpose of this invention is to address the technical deficiencies in the existing technology by providing a method for identifying the state gradient of permafrost based on the prior probability of thermal state.

[0007] The technical solution adopted to achieve the purpose of this invention is: A method for identifying the state gradient of permafrost based on prior probability of thermal state includes the following steps: Step 1: Obtain multi-source spatiotemporal data for multiple years in the target region. Perform unified gridding, resampling, and quality control on the multi-source spatiotemporal data sequentially to obtain a unified multi-source spatiotemporal dataset. And calculate the frozen living days Living by melting ; Step 2: Assign values ​​to the freeze-thaw parameters, correct them based on the snow accumulation parameters, and obtain the effective freeze-thaw factor. and Calculate the annual thermal state The parameters are perturbed to calculate the multi-year average ground temperature. and thermal state prior probability ; Step 3, based on and Calculate the environmental suitability score Calculate the annual prior fusion probability Long-term bias prediction field is obtained based on spatial bias learning. , combined and Calculate the average probability of permafrost occurrence within the target time window. ; Step 4: Using soil thermal parameters as input, simulate the annual active layer thickness. Calculate the multi-year average active layer thickness ,according to and Determine the basic condition category of permafrost; Step 5, Construct a general dynamically unstable indicator and strong dynamic instability indicators The basic state category of permafrost is dynamically corrected by perception to obtain the time-corrected category. Introducing the result from step 3 right After applying a probabilistic conservative correction, the final state category of permafrost is obtained. ,for By performing neighborhood mode replacement on isolated connected patches, the state gradient map of permafrost is obtained. .

[0008] In the above technical solution, in step 1, the multi-source spatiotemporal data includes surface temperature, near-surface air temperature, snow cover, soil moisture, vegetation index, soil properties, land cover category, and topographic data; The Represented as: ; In the formula, For passing through the target area grid Inner The first pixel Year Multi-source spatiotemporal data after unifying the grid for each data category For the target area grid, For the spatial location index of the effective cell set, To be the set of valid cells that pass quality control and enter the computation, For the year, This represents the total number of years. For multi-source spatiotemporal data category indexing, This represents the total number of categories in multi-source spatiotemporal data. in, The calculation formula is: ; In the formula, The original multi-source spatiotemporal data, For resampling operators corresponding to multi-source spatiotemporal data types; The The calculation formula is: ; In the formula, For the first Frozen in a pixel-by-pixel life For the first One pixel is used to calculate the daily average temperature driving force for the freeze-thaw index. For freezing season days or thawing season days, To gather for the freezing season, This is a function to find the maximum value. The The calculation formula is as follows: ; In the formula, For the first The melting of individual pixels to survive the day Gathering for the melting season.

[0009] In the above technical solution, in step 2, the... The calculation formula is: ; In the formula, The cumulative distribution function of the standard normal distribution. This represents the average annual thermal state after the disturbance. The standard deviation of the annual thermal state after the disturbance. It is a numerically stable term; in, The calculation formula is: ; In the formula, The annual thermal state after the disturbance. For set members, The total number of members in the set; The calculation formula is: ; in, The calculation formula is: ; In the formula, For the annual thermal state mapping function, Parameters to be disturbed The A set of members; The calculation formula is: ; In the formula, As the baseline parameter, Zero-mean disturbance; The The calculation formula is: ; In the formula, For the first The annual heat status of each pixel, This is the ratio of thermal conductivity in the frozen state to that in the thawed state. The length of the annual cycle, For the freeze period factor, For the first Annual Image Melting period factor, For the first The heat determination item for each pixel; in, The calculation formula is: ; The calculation formula is: ; In the formula, For pixels In the The annual surface melting degree days calculated from surface temperature For pixels In the Annual air melting degree days calculated from near-surface air temperature; The calculation formula is: ; In the formula, This is the baseline value for the freeze period factor. For land cover categories, It is a natural exponential function. This is the snow depth attenuation coefficient. This represents the average snow depth during the freezing season. For the continuous proportion of snow cover, The snow cover duration proportional decay coefficient; The calculation formula is: .

[0010] In the above technical solution, in step 3, The calculation formula is: ; In the formula, For the first The probability of permafrost occurrence per year For the year, This represents the total number of years. in, The calculation formula is: ; In the formula, This is an interval clipping function. For long-term deviation prediction field, The annual prior fusion probability; The calculation formula is: ; In the formula, For the ensemble weights of the ridge regression model, Ridge regression model for pixels The predicted value of the systematic bias, For gradient boosting regression tree models, the pixels The predicted value of the system bias; The calculation formula is: ; In the formula, For ridge regression model, This is the bias prediction feature vector; The calculation formula is: ; In the formula, For gradient boosting regression tree models; in, The calculation formula is: ; In the formula, This represents the annual average probability. The standard deviation of the annual average probability. For elevation, Slope direction, Northwards, Normalized Difference Vegetation Index (NDVI) This represents the average snow depth over many years. This represents the volumetric water content of the root zone. Soil organic matter content, The interaction features are probability-temperature, probability-elevation, probability-snow cover, probability-soil moisture, and probability mean-interannual variability.

[0011] In the above technical solution, The calculation formula is: ; In the formula, For the annual prior fusion probability, For the year, This represents the total number of years. The calculation formula is: ; in, The calculation formula is: ; In the formula, For the first Annual Image The prior probability, For pixels The local neighborhood, The number of neighboring pixels, Weights for the center pixel For any pixel in the neighborhood, For the first Annual neighborhood pixels The annual pixel prior probability; in, The calculation formula is: ; In the formula, It is a natural exponential function. For annual integration score; in, The calculation formula is: ; In the formula, For the weight of the thermal state term, For the thermal state term, For the environmental modulation term weight, The intercept; in, The calculation formula is: ; In the formula, It is the natural logarithm function. The prior probability of the thermal state after trimming; in, The calculation formula is: ; In the formula, For probability-based pruning stability parameters, To find the minimum value function, This is a function to find the maximum value. The calculation formula is: ; In the formula, The total number of environmental variables involved in the fusion. For variable weights, For the first Multi-source spatiotemporal data of environmental variables after robust standardization Direction coefficient; in, The calculation formula is: ; In the formula, The median, Interquartile range, For numerically stable terms, For passing through the target area grid Inner The first pixel Year Multi-source spatiotemporal data after unifying the grid for each data category; The calculation formula is: ; In the formula, For soil layer indexing, The total number of soil layers included in the root zone moisture content weighted calculation. For the first Layer thickness, For the first The layer in the first Annual Image Volumetric water content; The calculation formula is: ; In the formula, For pixels Slope aspect expressed in radians It is a cosine function.

[0012] In the above technical solution, step 4 involves using an activity layer thickness model to simulate the annual activity layer thickness. The active layer thickness model is the Kudryavtsev model, the Stefan model, or the GIPL model; The The calculation formula is: ; In the formula, For the year, This represents the total number of years. The thickness of the annual activity layer; in, The calculation formula is: ; In the formula, Thermal conductivity in the molten state The latent heat of phase change per unit volume of soil. This represents the volumetric heat capacity during the freezing period. For pixels In the Average soil temperature during the annual freezing period This refers to the volumetric heat capacity during the melting period. For pixels In the Average soil temperature during the annual melting period.

[0013] in, The calculation formula is: ; In the formula, The types of solid components in soil, It is a solid component of soil. For the first Specific heat capacity of soil-like solids For density, It is the volume fraction. The specific heat capacity of water by mass. This refers to the total volumetric water content. The calculation formula is: ; In the formula, The specific heat capacity of ice is given by mass. This represents the volume fraction of unfrozen water in a frozen state. The calculation formula is: ; In the formula, For the density of water, The latent heat of water's melting. This is the function for finding the maximum value.

[0014] In the above technical solution, step 4 involves defining a frozen region. Warm Frozen Zone ,exist Internal calculation of high quantile threshold for active layer thickness of permafrost and lower quantile threshold ,exist Calculate the threshold of deep active layers ,according to , , , and The basic state category of permafrost is determined by the following criteria: when At that time, and when At that time, the basic state category of permafrost was determined to be stable, among which, This is the first dividing temperature threshold. when At that time, and when At that time, or when At that time, and when At that time, the basic state category of permafrost was determined to be basically stable, among which, This is the second dividing temperature threshold. when At that time, and when At that time, or when At that time, the basic state category of permafrost was determined to be high-temperature unstable, among which, This is the third dividing temperature threshold; when At that time, and when At that time, the basic condition category of permafrost was determined to be transitional. This is the fourth dividing temperature threshold. when At that time, and when At that time, the basic condition category of permafrost was determined to be marginal. when At that time, the basic state category of permafrost was determined to be non-permafrost.

[0015] In the above technical solution, The calculation formula is: ; In the formula, Target area grid The effective set of pixels that passes quality control and is included in the calculation; The calculation formula is: ; The calculation formula is: ; In the formula, The function is for calculating the lower quantiles; The calculation formula is: ; In the formula, The function for calculating higher quantiles; The calculation formula is: ; In the formula, This is the threshold calculation function for deep active layers.

[0016] In the above technical solution, in step 5, The calculation formula is: ; In the formula, For indicator functions, The interannual standard deviation of the annual thermal state. The interannual standard deviation threshold. For the proportion of warm years, The threshold for the proportion of warm years, This represents the temperature change over a period of time. The threshold for the staged temperature increase, The slope of the long-term trend in the thermal state. This is the temperature trend threshold. This represents the change in the thickness of the active layer during the phase. The threshold for the change in the thickness of the active layer during a phase. The slope of the long-term trend of the active layer thickness. The activity layer trend threshold, Indicates the logical "OR"; in, The calculation formula is: ; In the formula, This represents the average annual thermal state after the disturbance. For the year, This represents the total number of years. The calculation formula is: ; In the formula, The temperature threshold for a warm year; The calculation formula is: ; In the formula, This is the early stage of the research period. This refers to the later part of the research period; The calculation formula is: ; The calculation formula is: ; In the formula, The average of the time coordinates during the study period; The calculation formula is: ; The The calculation formula is: ; In the formula, The threshold for the proportion of strongly unstable warm years. The temperature rise threshold for the highly unstable phase. The threshold for thickening the highly unstable activity layer. This indicates the logical "AND"; and The process of dynamically correcting the basic state category of permafrost is as follows: When the basic state category of permafrost is stable and At that time, the basic state of permafrost was adjusted from stable to basically stable. When the basic state category of permafrost is basically stable and At that time, the basic state of permafrost was adjusted from basically stable to high-temperature unstable. When the basic state category of permafrost is high-temperature unstable and At that time, the basic condition of permafrost was adjusted from high-temperature unstable type to transitional type; When the basic condition category of permafrost is marginal or transitional and the subsequent multi-year average ground temperature reaches a non-negative temperature, at the same time... At that time, the basic condition of permafrost was adjusted to non-permafrost; If the above conditions are not met, maintain the basic state category of permafrost.

[0017] In the above technical solution, The calculation formula is: ; In the formula, To be based on pixels Number of probability downgrade steps For time-corrected categories Operators that perform one-way degradation This is the time-adjusted category of the basic condition of permafrost. For pixels The probability of downgrading steps; in, Represented as: ; In the formula, For high probability threshold, The threshold is a medium probability threshold. Low probability threshold; Permafrost state gradient map Represented as: ; In the formula, For neighboring pixels The final category after time and probability adjustments. It is the mode function. For post-classification processing of the neighborhood, For the first A series of interconnected patches of the same type, For the first The area of ​​a connected patch of the same type. This is the preset threshold for the area of ​​the smallest connected patch of the same type.

[0018] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention first utilizes multi-source remote sensing and reanalysis data to obtain the multi-year average ground temperature and its uncertainty of permafrost through ensemble simulation, thereby constructing a prior probability of thermal state. Based on this, it integrates multi-dimensional environmental information such as topography, vegetation, snow cover, and soil moisture to establish an annual prior probability field, and generates the permafrost occurrence probability for the target time window through long-term bias learning and correction. Simultaneously, it uses a physical model to simulate the annual active layer thickness as evidence of structural response. Finally, it uses the multi-year average ground temperature and multi-year average active layer thickness to determine the basic state and construct a general dynamic instability indicator. and strong dynamic instability indicators By introducing dynamic corrections over time, this method effectively identifies the intermediate gradients in the transition from a stable to an unstable state in permafrost, overcoming the limitations of traditional methods' binary judgments of permafrost degradation processes. Furthermore, it employs a unidirectional conservative correction based on the probability of permafrost occurrence, elevating permafrost state identification from qualitative description to quantitative probabilistic output, ultimately identifying six states: stable, basically stable, high-temperature unstable, transitional, marginal, and non-permafrost. This method integrates thermal state, structural response, and occurrence probability into a unified discriminant framework, providing more refined spatial evidence for site selection and foundation stability evaluation in cold-region engineering projects.

[0019] 2. This invention incorporates three types of evidence—multi-year average ground temperature, multi-year average active layer thickness, and permafrost occurrence probability—into a unified judgment framework, overcoming the limitations of existing methods where thermal state simulation, probability mapping, and active layer thickness calculation are independent and difficult to jointly serve engineering judgments. In traditional methods, ground temperature fields, probability maps, and active layer thickness typically exist as separate maps, making it difficult to merge these three types of information into a unified classification standard in engineering practice. This invention establishes a systematic process where thermal state dominates foundation judgment, active layer thickness serves as a structural response constraint, and probability information serves as a conservative correction, enabling the three types of evidence to work synergistically within a unified framework, effectively improving the reliability and engineering applicability of the judgment results.

[0020] 3. This invention employs a strategy of thermal state dominance and unidirectional conservative correction of probability, avoiding excessive or reverse interference from uncertainties in probabilistic information on the judgment results. Existing probabilistic mapping methods are significantly affected by training samples, environmental variable selection, and model parameter settings, especially in boundary regions and areas with scarce data, where probability estimation often involves considerable uncertainty. If probabilistic information is used as the basis for improving the stability level, it may lead to judgments that are biased towards danger in areas with high probability values ​​but actually warmer thermal states. This invention limits the role of probabilistic information to unidirectional conservative correction, that is, when the probability is below a threshold, the level is only reduced, and when the probability is sufficiently high, the original level is maintained, which conforms to the basic principle of safety first in engineering practice.

[0021] 4. This invention introduces active layer thickness as evidence of structural response and employs an adaptive quantile threshold for classification within permafrost regions, rather than relying on a fixed absolute thickness threshold. Due to differences in climate and lithology, the structural stability reflected by the same active layer thickness varies across different cold regions, making it difficult to apply a fixed threshold across regions. This invention calculates the quantile value of active layer thickness within permafrost regions as the basis for classifying shallow, intermediate, and deep active layers. This allows the discrimination rule to be dynamically adjusted according to the actual thickness distribution in the region, avoiding the limitations of empirical fixed thresholds while maintaining the comparability of the discrimination rule across different regions, and possessing the ability to be extended to other cold regions.

[0022] 5. This invention outputs six engineering stability levels for permafrost: stable, basically stable, high-temperature unstable, transitional, marginal, and non-permafrost. This overcomes the shortcomings of traditional permafrost maps, which primarily focus on distribution continuity or binary freeze-thaw states and fail to reflect internal stability differences. Traditional classifications reflect the spatial proportion and fragmentation degree of permafrost, rather than its thermal stability and engineering performance as a foundation medium. This invention outputs six states based on ground temperature, active layer thickness, and probabilistic constraints. Each state has distinguishable statistical characteristics, and different levels correspond to differentiated recommendations for foundation treatment intensity, foundation type selection, and monitoring frequency. This shifts the foundation design of cold-region buildings from relying on regional experience to differentiated design based on state classification. Attached Figure Description

[0023] Figure 1 The flowchart shown is a method for identifying the state gradient of permafrost based on the prior probability of thermal state according to the present invention.

[0024] Figure 2 The figure shows the multi-year average ground temperature distribution.

[0025] Figure 3 The figure shows the multi-year average active layer thickness distribution calculated using the Kudryavtsev model.

[0026] Figure 4The image shown is an example of basic condition identification for permafrost. Detailed Implementation

[0027] The present invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0028] Example 1 A method for identifying the state gradient of permafrost based on prior probability of thermal state, referring to... Figure 1 This includes the following steps: Step 1, Multi-source spatiotemporal data acquisition, unified gridding and physical preprocessing: Acquire multi-year multi-source spatiotemporal data of the target region and establish a grid for the target region. The collected multi-source spatiotemporal data are sequentially subjected to unified gridding, resampling, and quality control to obtain a unified multi-source spatiotemporal dataset. And calculate the freezing days and thawing days.

[0029] The multi-source spatiotemporal data includes surface temperature, near-surface air temperature, snow cover, soil moisture, vegetation index, soil properties, land cover category, and topographic data.

[0030] Specifically, by unifying the gridding of multi-source spatiotemporal data, the spatial reference, cell size, range, and cell origin of the multi-source spatiotemporal data are kept consistent. Continuous variables in the unified gridded multi-source spatiotemporal data are resampled using bilinear or area-conserving methods, while categorical variables are resampled using nearest-neighbor methods. The quality control includes at least invalid value masking, physical range filtering, cell offset correction from different data sources, and intersection of common effective domains.

[0031] The unified multi-source spatiotemporal dataset is represented as follows: ; In the formula, For passing through the target area grid Inner The first pixel Year Multi-source spatiotemporal data after unifying the grid for each data category For the target area grid, For the spatial location index of the effective cell set, To be the set of valid cells that pass quality control and enter the computation, For the year, This represents the total number of years. For multi-source spatiotemporal data category indexing, This represents the total number of categories of multi-source spatiotemporal data.

[0032] in, The calculation formula is: ; In the formula, The original multi-source spatiotemporal data, For resampling operators corresponding to multi-source spatiotemporal data types; set up The daily average temperature driving force used to calculate the freeze-thaw index, in degrees Celsius, is given by the formula for calculating the daily freeze index: ; In the formula, For the first Frozen in a pixel-by-pixel life For the first One pixel is used to calculate the daily average temperature driving force for the freeze-thaw index. For freezing season days or thawing season days, To gather for the freezing season, This is the function for finding the maximum value.

[0033] Formula for calculating melting days: ; In the formula, For the first The melting of individual pixels to survive the day Gathering for the melting season.

[0034] Step 2, Snow-corrected thermal state ensemble simulation and thermal state prior probability generation: Freeze-thaw parameters are assigned values, corrected based on snow parameters, to obtain the effective freeze-thaw factor, and then combined with… and Calculate the annual thermal state ,right The parameters in the data are perturbed to calculate the multi-year average ground temperature. and thermal state prior probability .

[0035] Furthermore, the freeze-thaw parameters include a baseline value for the freezing period factor. Melting period factor benchmark value And the baseline values ​​of thermal conductivity in frozen and thawed states. .

[0036] The effective freeze-thaw factor includes a freezing period factor and a thawing period factor; To express the insulating effect of snow cover, the freezing period factor is calculated using a monotonically decreasing first-order correction form based on snow cover parameters: ; In the formula, For the freeze period factor, This is the baseline value for the freeze period factor. This represents the average snow depth during the freezing season. For the proportion of snow cover, among which, , This is the snow depth attenuation coefficient, whose dimension is the reciprocal of the snow depth unit. is the snow cover continuous proportional decay coefficient, and is the dimensionless correction coefficient.

[0037] The formula for calculating the melting period factor is: ; In the formula, For the first Annual Image The melting period factor is used to characterize the conversion relationship between the air melting index and the surface melting index. For pixels In the The annual surface melting degree days calculated from surface temperature For pixels In the Annual air melting degree days calculated from near-surface air temperature; according to , and The formula for calculating the heat discrimination term is as follows: ; In the formula, For the first The heat discrimination item for each pixel.

[0038] The annual thermal state is calculated in TTOP form and is expressed as follows: ; In the formula, For the first The annual heat status of each pixel, This is the ratio of thermal conductivity in the frozen state to that in the thawed state. The annual cycle length is usually taken as... .

[0039] To characterize parameter and input uncertainties, this invention employs an ensemble simulation approach. Multiple random perturbations are applied to parameters such as the freezing period factor, thawing period factor, the ratio of thermal conductivity in the frozen to thawing states, the average snow depth during the freezing season, and the snow cover duration ratio, generating multiple thermal state simulation results. Each perturbation corresponds to a ensemble member. All set members together constitute the thermal state simulation set. The perturbation parameters of each set member are expressed as: ; In the formula, For the first The perturbation parameters of each set member, For pixels In the Annual baseline parameter values, Parameters to be disturbed In the The zero-mean perturbation in each set member can be a uniform distribution, a truncated normal distribution, or a distribution determined by observation error. For the first Cells in a set In the The ratio of thermal conductivity in the frozen state to thermal conductivity in the thawed state. For the first Cells in a set In the Effective freezing period factor corrected for annual snow cover insulation effect For the first Cells in a set In the The melting period factor of the year; The annual thermal state after the disturbance is calculated by each set member. Annual thermal state after perturbation of each set member for: ; In the formula, The annual thermal state after the disturbance. This is the annual thermal state mapping function, whose inputs are perturbation parameters, freezing degree days, and thawing degree days, and whose output is the annual thermal state of the corresponding set member.

[0040] The formula for calculating the mean of the annual thermal state after disturbance is: ; The multi-year average ground temperature is further calculated from the mean of the annual thermal state after disturbance. The calculation formula is as follows: ; The formula for calculating the standard deviation of the annual thermal state after disturbance is: ; In the formula, This represents the standard deviation of the annual thermal state after the disturbance.

[0041] Assuming the annual thermal state after the disturbance Approximated by a normal distribution, the probability that a pixel is in a negative temperature hot state, i.e., the prior probability of the hot state, is calculated using the following formula: ; In the formula, Let be the prior probability of the hot state. The cumulative distribution function of the standard normal distribution. For numerically stable terms, where, .

[0042] Reference Figure 2 The graph shows the distribution of multi-year average ground temperature in the study area. Colors change from cool to warm, corresponding to increasing multi-year average ground temperature. A multi-year average ground temperature below zero degrees Celsius indicates that the corresponding pixel possesses the thermal conditions for forming or maintaining permafrost. The closer the multi-year average ground temperature is to zero degrees Celsius, the weaker the thermal stability of permafrost and the more sensitive it is to climate warming and changes in the surface environment. A multi-year average ground temperature above zero degrees Celsius indicates that the corresponding pixel has relatively insufficient thermal conditions for maintaining permafrost.

[0043] Step 3, Annual Prior Probability Fusion, Long-Term Bias Learning, and Target Time Window Probability Reconstruction: Based on and Calculate the environmental suitability score ,based on Calculate the annual prior fusion probability The target region is divided into non-overlapping or low-overlapping spatial blocks. The non-overlapping spatial blocks are used to train the ridge regression model, and the low-overlapping spatial blocks are used to train the gradient boosting regression tree model, thus obtaining the long-term bias prediction field. , combined and Calculate the average probability of permafrost occurrence within the target time window. ; For step 1 Robust standardization is performed to obtain robustly standardized multi-source spatiotemporal data. The calculation formula is as follows: ; In the formula, The median. Interquartile range, It is a numerically stable term.

[0044] To unify the physical contribution directions of different variables, direction coefficients are set. When the increase of variables favors the existence of permafrost, ;on the contrary, .based on Further calculate the annual environmental suitability score for: ; In the formula, For environment variable indexing, The total number of environmental variables involved in the fusion. Let be the variable weights, where And satisfy .

[0045] The prior probability of the thermal state Clipping to an open interval yields the prior probability of the clipped hot state. The calculation formula is as follows: ; In the formula, For probability-based pruning stability parameters, To find the minimum value function, This is the function for finding the maximum value.

[0046] Again Perform a logit transformation to obtain the thermal state term. The calculation formula is as follows: ; In the formula, For the thermal state term, It is the natural logarithm function.

[0047] Combination and Calculate the annual fusion score The calculation formula is as follows: ; In the formula, For the weight of the thermal state term, and , For the environmental modulation term weights, and , This is the intercept.

[0048] based on Calculate the prior probability of annual pixels The calculation formula is as follows: ; In the formula, It is a natural exponential function.

[0049] To suppress pixel noise and preserve local spatial continuity, neighborhood fusion is used. The annual prior fusion probability is obtained. The calculation formula is as follows: ; In the formula, For the first Annual Image The prior probability, For pixels The local neighborhood, The number of neighboring pixels, The weight of the center cell, and , For any pixel in the neighborhood, For the first Annual neighborhood pixels The annual pixel prior probability.

[0050] based on Calculate the annual average probability and the standard deviation of the annual average probability ,in: The calculation formula is: ; The calculation formula is: ; Based on the unified multi-source spatiotemporal dataset and frozen daily data in step 1 Living by melting , , Constructing bias prediction feature vectors It is represented as: ; In the formula, This represents the average prior probability over many years. The standard deviation of the average probability. The average ground temperature over many years, Living on the freeze To survive by melting away For elevation, Slope direction, Northwards, Normalized Difference Vegetation Index (NDVI) This represents the average snow depth over many years. This represents the volumetric water content of the root zone. Soil organic matter content, The interaction features are probability-temperature, probability-elevation, probability-snow cover, probability-soil moisture, and probability mean-interannual variability.

[0051] For pixels In its first annual root zone volumetric water content The calculation formula is: ; In the formula, For the first Volumetric water content of each soil layer For the first The thickness of each soil layer, For soil layer indexing; This refers to the total number of soil layers included in the root zone moisture content weighted calculation.

[0052] For pixels Convert its slope aspect to north. The The calculation formula is: ; In the formula, For pixels Slope aspect expressed in radians It is a cosine function. The larger the value, the further north it is.

[0053] The target region grid is divided into multiple spatial blocks based on spatial location. Each spatial block consists of several spatially continuous and adjacent grid cells, with no overlap between blocks; alternatively, adjacent spatial blocks are allowed only a boundary overlap of less than a predetermined proportion to form low-overlap spatial blocks. Then, using the spatial block as the smallest unit of division, rather than a single cell, all spatial blocks are divided into training, validation, and test sets, ensuring that all samples within the same spatial block belong to only one of these sets. The ridge regression model is trained using the training set of non-overlapping spatial blocks, validated using the validation set, and tested using the test set, yielding the pixel-level long-term bias prediction values ​​of the ridge regression model. The The calculation formula is: ; In the formula, For ridge regression model.

[0054] The gradient boosting regression tree model was trained using a training set with low-overlapping spatial blocks, validated using a validation set, and tested using a test set to obtain the pixel-level long-term bias prediction values ​​of the gradient boosting regression tree model. The The calculation formula is: ; In the formula, This is a gradient boosting regression tree model.

[0055] The spatial cross-validation errors of the Ruling regression model and the gradient boosting regression tree model are respectively and The ensemble weights for the ridge regression model are: ; In the formula, For the ensemble weights of the ridge regression model, This is the numerically stable term in the calculation of the reciprocal of the error, used to prevent the denominator from diverging when the cross-validation error is zero.

[0056] based on , and Calculate the long-term deviation prediction field The calculation formula is as follows: ; use right After correction, the first... Probability of permafrost occurrence per year The calculation formula is as follows: ; In the formula, Given an interval clipping function, the corrected probability is always located within the specified range. .

[0057] based on Calculate the average probability of permafrost occurrence within the target time window. for: ; Step 4, Simulation of Active Layer Structural Response and Basic Judgment of Thermal State-Structural Response: Using soil thermal parameters as input, the annual active layer thickness is simulated using an active layer thickness model. ,based on Calculate the multi-year average active layer thickness ,according to and Determine the basic condition category of permafrost; The active layer thickness model is the Kudryavtsev model, the Stefan model, or the GIPL (Geophysical Institute Permafrost Laboratory Model). The inputs to the active layer thickness model include at least air temperature, surface temperature amplitude, snow cover, soil moisture, vegetation thermal resistance, and soil thermal parameters. Since the active layer thickness is determined by the soil's thermal conductivity and the heat absorbed or released during the freeze-thaw process, the volumetric heat capacity during the thawing period, the volumetric heat capacity during the freezing period, and the latent heat of phase change per unit volume of soil are calculated to characterize the soil's heat storage capacity, the energy exchange capacity during the freeze-thaw process, and the propulsion capacity of the frozen interface.

[0058] Assuming the soil is composed of If the composition is a solid-like component, then the volumetric heat capacity during the melting period is: ; In the formula, This refers to the volumetric heat capacity during the melting period. For the first Specific heat capacity of soil-like solids For density, It is the volume fraction. The specific heat capacity of water by mass. This represents the total volumetric water content.

[0059] The volumetric heat capacity during the freezing period is: ; In the formula, This represents the volumetric heat capacity during the freezing period. The specific heat capacity of ice is given by mass. This represents the volume fraction of unfrozen water in a frozen state.

[0060] The formula for calculating the latent heat of phase change per unit volume of soil is: ; In the formula, The latent heat of phase change per unit volume of soil. For the density of water, The latent heat of water's melting. This is the function for finding the maximum value.

[0061] When using an equivalent Stefan parsing implementation, the annual activity layer thickness The calculation formula is: ; In the formula, Thermal conductivity in the molten state This refers to the volumetric heat capacity during the melting period. This represents the volumetric heat capacity during the freezing period. The latent heat of phase change per unit volume of soil. For pixels In the Average soil temperature during the annual freezing period For pixels In the Average soil temperature during the annual melting period, a constant Used to convert "days" to "seconds".

[0062] The multi-year average active layer thickness is: ; In the formula, The average active layer thickness over many years. For the year, This represents the total number of years. Reference Figure 3The figure shows the distribution of the multi-year average active layer thickness in the study area, calculated using the Kudryavtsev model. Cooler colored areas correspond to smaller multi-year average active layer thicknesses, while warmer colored areas correspond to larger multi-year average active layer thicknesses. A larger multi-year average active layer thickness generally indicates stronger seasonal thawing and a greater depth of heat transfer from the surface to the subsurface; a smaller multi-year average active layer thickness generally indicates a relatively colder subsurface thermal state and a relatively weaker structural response of the permafrost.

[0063] Figure 2 and Figure 3 A comparison reveals a clear correlation between the multi-year average active layer thickness and the multi-year average ground temperature. Generally, areas with higher multi-year average ground temperatures allow surface heat to penetrate deeper in summer, resulting in a thicker active layer; conversely, areas with lower multi-year average ground temperatures have limited seasonal melting depth, leading to a thinner active layer. However, this correlation is not absolute. Besides ground temperature, factors such as snow cover insulation, vegetation cover, soil moisture content, soil thermal conductivity, and latent heat of phase change also influence the active layer. Therefore, even regions with similar multi-year average ground temperatures may exhibit different active layer thicknesses. Overall, Figure 2 and Figure 3 The data reflects the thermal state of permafrost in the study area from two aspects: the multi-year average ground temperature reflects the background ground temperature, and the multi-year average active layer thickness reflects the seasonal thawing response under the influence of surface heat. Using both in combination allows for a more comprehensive assessment of the stability of permafrost and its spatial variations.

[0064] Define the frozen domain for: ; In the formula, Target area grid The effective set of pixels is entered into the calculation after quality control.

[0065] Warm Freezing Zone for: ; Calculate the high quantile threshold for active layer thickness of permafrost within the frozen zone. and lower quantile threshold In this embodiment, the lower quantile threshold is set to the 35th percentile threshold, and the higher quantile threshold is set to the 70th percentile threshold. The calculation formula is: ; In the formula, The low quantile threshold for the active layer thickness of the frozen domain. This is a function for calculating lower quantiles.

[0066] The calculation formula is: ; In the formula, The high quantile threshold for the active layer thickness of the frozen domain. This is a function for calculating higher quantiles.

[0067] Calculating the threshold of the deep active layer within the warm frozen zone In this embodiment, the deep activity layer threshold is set to the 65th percentile threshold. The calculation formula is: ; In the formula, For the threshold of the thickness of the deep active layer in the warm freezing zone, This is the threshold calculation function for deep active layers.

[0068] Reference Figure 4 ,according to and The basic state category of permafrost is determined by the following criteria: when At that time, and when When the foundation state of permafrost is classified as stable (ST), this type represents permafrost with low temperature and weak structural response. The first dividing temperature threshold is used in this embodiment. .

[0069] when At that time, and when At that time, or when At that time, and when When the basic state category of permafrost is determined to be basically stable (denoted as BS), this category maintains a frozen background, but its thermal sensitivity or active layer response is higher than ST. The second dividing temperature threshold is used in this embodiment. .

[0070] when At that time, and when At that time, or when When the foundation state of permafrost is classified as high-temperature unstable (HU), this type indicates warm permafrost exhibiting a strong structural response. The third dividing temperature threshold is used in this embodiment. .

[0071] when At that time, and when At that time, or when At that time, the basic state category of permafrost was determined to be transitional (denoted as TR). This type is close to the thawing threshold and exhibits a significant deepening of the active layer. The fourth dividing temperature threshold is used in this embodiment. .

[0072] when At that time, and when At that time, the basic state category of permafrost was determined to be marginal (denoted as MG). This category is in the near-melt threshold zone, but the thermal state and structural response evidence are mixed.

[0073] when When the basic state category of permafrost is determined to be non-permafrost (denoted as NP), the thickness of the active layer is no longer a necessary criterion.

[0074] Step 5, Time-aware dynamic correction, probabilistic one-way conservative correction, and spatial post-processing: Constructing a general dynamically unstable indicator. and strong dynamic instability indicators ,use and The basic state category of permafrost is dynamically corrected by perception to obtain the time-corrected category. Introducing the results from step 3 right After applying a probabilistic conservative correction, the final state category of permafrost is obtained. ,for By performing neighborhood mode replacement on isolated connected patches, the state gradient map of permafrost is obtained. .

[0075] Interannual standard deviation of annual thermal state for: ; Let the temperature threshold for a warm year be... The proportion of warm years The calculation formula is: ; In the formula, For indicator functions, In this embodiment, the temperature threshold is the temperature for a warm year. .

[0076] The time set consisting of multiple consecutive or discrete years in step 1 is taken as the study period, and the study period is divided into the earlier set of the study period according to the time sequence. and the later stages of the research period Then the temperature change during the stage is: ; In the formula, This represents the temperature change during the stage.

[0077] The change in the thickness of the active layer during each stage is: ; In the formula, This represents the change in the thickness of the active layer during a given stage.

[0078] The slope of the long-term trend of thermal state is: ; In the formula, The slope of the long-term trend in the thermal state. The mean of the time coordinates during the study period.

[0079] The long-term trend slope of the active layer thickness is: ; In the formula, The slope represents the long-term trend of the active layer thickness.

[0080] in accordance with , , , , and Constructing general dynamically unstable indicators The calculation formula is as follows: ; In the formula, The interannual standard deviation of the annual thermal state. The interannual standard deviation threshold. For the proportion of warm years, The threshold for the proportion of warm years, This represents the temperature change over a period of time. The threshold for the staged temperature increase, The threshold for the change in the thickness of the active layer during a phase. The slope of the long-term trend in the thermal state. This is the temperature trend threshold. This represents the change in the thickness of the active layer during the phase. The trend threshold for the activity layer. The slope represents the long-term trend of the active layer thickness. It represents the logical "OR".

[0081] based on , , Define strongly dynamically unstable indicators The calculation formula is as follows: ; In the formula, The threshold for the proportion of strongly unstable warm years. The temperature rise threshold for the highly unstable phase. The threshold for thickening the highly unstable activity layer. It represents the logical "AND".

[0082] according to and The basic state category of permafrost is dynamically corrected based on time perception. The correction process is as follows: When the basic condition category of permafrost is ST and At that time, the basic condition of permafrost was adjusted from ST to BS; When the basic condition category of permafrost is BS and At that time, the basic condition of permafrost was adjusted from BS to HU; When the basic state category of permafrost is HU and At that time, the basic condition of permafrost was adjusted from HU to TR; When the basic condition category of permafrost is MG or TR and the subsequent multi-year average ground temperature reaches a non-negative temperature, at the same time At that time, the basic state of permafrost was adjusted to NP; If the above conditions are not met, maintain the basic state category.

[0083] The time-corrected basic condition category of permafrost is denoted as: .

[0084] Set strictly ordered probability thresholds: ; In the formula, For high probability threshold, The threshold is a medium probability threshold. This is a low probability threshold.

[0085] In this embodiment , , .

[0086] Definition reduction Operators of a stable level It is represented as: ; In the formula, This is the inverse mapping from ordinal to category. To find the maximum value function, For category Stability order, The number of degradation steps is a non-negative integer. For time-corrected categories, In this embodiment, , , , , .

[0087] according to Determine the number of downgrade steps : ; based on Calculate the final foundation condition category of permafrost The calculation formula is as follows: ; In the formula, To be based on pixels Number of probability downgrade steps For time-corrected categories Operators that perform one-way degradation.

[0088] The time-corrected permafrost category is TR or MG and Pixels close to 0°C are defined as near-fusion threshold pixels, if simultaneously satisfying... and You can directly adjust TR or MG to NP.

[0089] The probability correction must satisfy: ; That is, probabilistic evidence cannot elevate a pixel to a more stable category on its own, which constitutes the asymmetric conservative correction feature of this method.

[0090] After classification, for patches with areas smaller than the minimum threshold for connected patches of the same type... The isolated connected patches are replaced with neighborhood mode.

[0091] Let the first The connected regions contain a number of pixels. The area of ​​a single pixel is Then the area of ​​connected patches of the same type The calculation is as follows: ; when It was determined to be an isolated connected patch.

[0092] set up For the first There are 10 connected patches of the same type, with an area of ​​1 / 2 * 1 / 3 ... The permafrost state gradient diagram Represented as: ; In the formula, For neighboring pixels The final category after time and probability adjustments. For post-classification processing neighborhood Neighborhood pixels within, It is the mode function. For post-classification processing of the neighborhood, For the first A series of interconnected patches of the same type, For the first The area of ​​a connected patch of the same type. This is the preset threshold for the area of ​​the smallest connected patch of the same type.

[0093] The above description is only a preferred embodiment of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for identifying the state gradient of permafrost based on prior probability of thermal state, characterized in that, Includes the following steps: Step 1: Obtain multi-source spatiotemporal data for multiple years in the target region. Perform unified gridding, resampling, and quality control on the multi-source spatiotemporal data sequentially to obtain a unified multi-source spatiotemporal dataset. And calculate the frozen living days Living by melting ; Step 2: Assign values ​​to the freeze-thaw parameters, correct them based on the snow accumulation parameters, and obtain the effective freeze-thaw factor. and Calculate the annual thermal state The parameters are perturbed to calculate the multi-year average ground temperature. and thermal state prior probability ; Step 3, based on and Calculate the environmental suitability score Calculate the annual prior fusion probability Long-term bias prediction field is obtained based on spatial bias learning. , combined and Calculate the average probability of permafrost occurrence within the target time window. ; Step 4: Using soil thermal parameters as input, simulate the annual active layer thickness. Calculate the multi-year average active layer thickness ,according to and Determine the basic condition category of permafrost; Step 5, Construct a general dynamically unstable indicator and strong dynamic instability indicators The basic state category of permafrost is dynamically corrected by perception to obtain the time-corrected category. Introducing the result from step 3 right After applying a probabilistic conservative correction, the final state category of permafrost is obtained. ,for By performing neighborhood mode replacement on isolated connected patches, the state gradient map of permafrost is obtained. .

2. The method for identifying permafrost state gradients based on prior thermal state probability as described in claim 1, characterized in that, In step 1, the multi-source spatiotemporal data includes surface temperature, near-surface air temperature, snow cover, soil moisture, vegetation index, soil properties, land cover category and topographic data; The Represented as: ； In the formula, For passing through the target area grid Inner One pixel Year Multi-source spatiotemporal data after unifying the grid for each data category For the target area grid, For the spatial location index of the effective cell set, To be the set of valid cells that pass quality control and enter the computation, For the year, This represents the total number of years. For multi-source spatiotemporal data category indexing, This represents the total number of categories in multi-source spatiotemporal data. in, The calculation formula is: ； In the formula, The original multi-source spatiotemporal data, For resampling operators corresponding to multi-source spatiotemporal data types; The The calculation formula is: ； In the formula, For the first Frozen in a pixel-by-pixel life For the first One pixel is used to calculate the daily average temperature driving force for the freeze-thaw index. For freezing season days or thawing season days, To gather for the freezing season, This is a function to find the maximum value. The The calculation formula is as follows: ； In the formula, For the first The melting of individual pixels to survive the day Gathering for the melting season.

3. The method for identifying permafrost state gradients based on prior thermal state probability as described in claim 1, characterized in that, In step 2, the The calculation formula is: ； In the formula, The cumulative distribution function of the standard normal distribution. This represents the average annual thermal state after the disturbance. The standard deviation of the annual thermal state after the disturbance. It is a numerically stable term; in, The calculation formula is: ； In the formula, The annual thermal state after the disturbance. For collection members, The total number of members in the set; The calculation formula is: ； in, The calculation formula is: ； In the formula, For the annual thermal state mapping function, Parameters to be disturbed The A set of members; The calculation formula is: ； In the formula, As the baseline parameter, Parameters to be disturbed In the Zero-mean perturbations in each set member; The The calculation formula is: ； In the formula, For the first The annual heat status of each pixel, This is the ratio of thermal conductivity in the frozen state to that in the thawed state. The length of the annual cycle, For the freeze period factor, For the first Annual Image Melting period factor, For the first The heat determination item for each pixel; in, The calculation formula is: ； The calculation formula is: ； In the formula, For pixels In the The annual surface melting degree days calculated from surface temperature For pixels In the Annual air melting degree days calculated from near-surface air temperature; The calculation formula is: ； In the formula, This is the baseline value for the freeze period factor. For land cover categories, It is a natural exponential function. This is the snow depth attenuation coefficient. This represents the average snow depth during the freezing season. For the continuous proportion of snow cover, The snow cover duration proportional decay coefficient; The calculation formula is: 。 4. The method for identifying permafrost state gradients based on prior thermal state probability as described in claim 1, characterized in that, In step 3 The calculation formula is: ； In the formula, For the first The probability of permafrost occurrence per year For the year, This represents the total number of years. in, The calculation formula is: ； In the formula, This is an interval clipping function. For long-term deviation prediction field, The annual prior fusion probability; The calculation formula is: ； In the formula, For the ensemble weights of the ridge regression model, Ridge regression model for pixels The predicted value of the systematic bias, For gradient boosting regression tree models, the pixels The predicted value of the system bias; The calculation formula is: ； In the formula, For ridge regression model, This is the bias prediction feature vector; The calculation formula is: ； In the formula, For gradient boosting regression tree models; in, The calculation formula is: ； In the formula, This represents the annual average probability. The standard deviation of the annual average probability. For elevation, Slope direction, Northwards, Normalized Difference Vegetation Index (NDVI) This represents the average snow depth over many years. This represents the volumetric water content of the root zone. Soil organic matter content, The interaction features are probability-temperature, probability-elevation, probability-snow cover, probability-soil moisture, and probability mean-interannual variability.

5. The method for identifying permafrost state gradients based on prior thermal state probability as described in claim 4, characterized in that, The calculation formula is: ； In the formula, For the annual prior fusion probability, For the year, This represents the total number of years. The calculation formula is: ； in, The calculation formula is: ； In the formula, For the first Annual Image The prior probability, For pixels The local neighborhood, The number of neighboring pixels, Weights for the center pixel, For any pixel in the neighborhood, For the first Annual neighborhood pixels The annual pixel prior probability; in, The calculation formula is: ； In the formula, It is a natural exponential function. For annual integration score; in, The calculation formula is: ； In the formula, For the weight of the thermal state term, For the thermal state term, For the environmental modulation term weight, The intercept; in, The calculation formula is: ； In the formula, It is the natural logarithm function. The prior probability of the thermal state after trimming; in, The calculation formula is: ； In the formula, For probability-based pruning stability parameters, To find the minimum value function, This is a function to find the maximum value. The calculation formula is: ； In the formula, The total number of environmental variables involved in the fusion. For variable weights, For the first Multi-source spatiotemporal data of environmental variables after robust standardization Direction coefficient; in, The calculation formula is: ； In the formula, The median. Interquartile range, For numerically stable terms, For passing through the target area grid Inner One pixel Year Multi-source spatiotemporal data after unifying the grid for each data category; The calculation formula is: ； In the formula, For soil layer indexing, The total number of soil layers included in the root zone moisture content weighted calculation. For the first Layer thickness, For the first The layer in the first Annual Image Volumetric water content; The calculation formula is: ； In the formula, For pixels Slope aspect expressed in radians It is a cosine function.

6. The method for identifying permafrost state gradients based on prior thermal state probability as described in claim 1, characterized in that, In step 4, the annual activity layer thickness is simulated using an activity layer thickness model. The active layer thickness model is the Kudryavtsev model, the Stefan model, or the GIPL model; The The calculation formula is: ； In the formula, For the year, This represents the total number of years. The thickness of the annual activity layer; in, The calculation formula is: ； In the formula, Thermal conductivity in the molten state The latent heat of phase change per unit volume of soil. This refers to the volumetric heat capacity during the freezing period. For pixels In the Average soil temperature during the annual freezing period This refers to the volumetric heat capacity during the melting period. For pixels In the Average soil temperature during the annual melting period; in, The calculation formula is: ； In the formula, The types of solid components in soil, It is a solid component of soil. For the first Specific heat capacity of soil-like solids For density, It is the volume fraction. The specific heat capacity of water by mass. This refers to the total volumetric water content. The calculation formula is: ； In the formula, The specific heat capacity of ice is given by mass. This represents the volume fraction of unfrozen water in a frozen state. The calculation formula is: ； In the formula, For the density of water, The latent heat of water's melting. This is the function for finding the maximum value.

7. The method for identifying permafrost state gradients based on prior thermal state probability as described in claim 6, characterized in that, In step 4, the freeze area is defined. Warm Frozen Zone ,exist Internal calculation of high quantile threshold for active layer thickness of permafrost and lower quantile threshold ,exist Calculate the threshold of deep active layers ,according to , , , and The basic state category of permafrost is determined by the following criteria: when At that time, and when At that time, the basic state category of permafrost was determined to be stable, among which, This is the first dividing temperature threshold. when At that time, and when At that time, or when At that time, and when At that time, the basic state category of permafrost was determined to be basically stable, among which, This is the second dividing temperature threshold. when At that time, and when At that time, or when At that time, the basic state category of permafrost was determined to be high-temperature unstable, among which, This is the third dividing temperature threshold; when At that time, and when At that time, the basic condition category of permafrost was determined to be transitional. This is the fourth dividing temperature threshold. when At that time, and when At that time, the basic condition category of permafrost was determined to be marginal. when At that time, the basic state category of permafrost was determined to be non-permafrost.

8. The method for identifying permafrost state gradients based on prior thermal state probability as described in claim 7, characterized in that, The calculation formula is: ； In the formula, Target area grid The effective set of pixels that passes quality control and is included in the calculation; The calculation formula is: ； The calculation formula is: ； In the formula, The function is for calculating the lower quantiles; The calculation formula is: ； In the formula, The function for calculating higher quantiles; The calculation formula is: ； In the formula, This is the threshold calculation function for deep active layers.

9. The method for identifying permafrost state gradients based on prior thermal state probability as described in claim 1, characterized in that, In step 5 The calculation formula is: ； In the formula, For indicator functions, The interannual standard deviation of the annual thermal state. The interannual standard deviation threshold. For the proportion of warm years, The threshold for the proportion of warm years, This represents the temperature change over a period of time. The threshold for the staged temperature increase, The slope of the long-term trend in the thermal state. This is the temperature trend threshold. This represents the change in the thickness of the active layer during the phase. The threshold for the change in the thickness of the active layer during a phase. The slope represents the long-term trend of the active layer thickness. The trend threshold for the activity layer. This represents the logical "OR" operator. in, The calculation formula is: ； In the formula, This represents the average annual thermal state after the disturbance. For the year, This represents the total number of years. The calculation formula is: ； In the formula, The temperature threshold for a warm year; The calculation formula is: ； In the formula, This is the early stage of the research period. This refers to the later part of the research period; The calculation formula is: ； The calculation formula is: ； In the formula, The average of the time coordinates during the study period; The calculation formula is: ； The The calculation formula is: ； In the formula, The threshold for the proportion of strongly unstable warm years. The temperature rise threshold for the highly unstable phase. The threshold for thickening the highly unstable activity layer. This represents the logical AND; and The process of dynamically correcting the basic state category of permafrost is as follows: When the basic state category of permafrost is stable and At that time, the basic state of permafrost was adjusted from stable to basically stable. When the basic state category of permafrost is basically stable and At that time, the basic state of permafrost was adjusted from basically stable to high-temperature unstable. When the basic state category of permafrost is high-temperature unstable and At that time, the basic condition of permafrost was adjusted from high-temperature unstable type to transitional type; When the basic condition category of permafrost is marginal or transitional and the subsequent multi-year average ground temperature reaches a non-negative temperature, at the same time... At that time, the basic condition of permafrost was adjusted to non-permafrost; If the above conditions are not met, maintain the basic state category of permafrost.

10. The method for identifying permafrost state gradients based on prior thermal state probability as described in claim 9, characterized in that, The calculation formula is: ； In the formula, To be based on pixels Number of probability downgrade steps For time-corrected categories Operators that perform one-way degradation This is the time-adjusted category of the basic condition of permafrost. For pixels The probability of downgrading steps; in, Represented as: ； In the formula, For high probability threshold, The threshold is a medium probability threshold. Low probability threshold; Permafrost state gradient map Represented as: ； In the formula, For neighboring pixels The final category after time and probability adjustments. It is the mode function. For post-classification processing of the neighborhood, For the first A series of interconnected patches of the same type, For the first The area of ​​a connected patch of the same type. This is the preset threshold for the area of ​​the smallest connected patch of the same type.

Images