A method for constructing a grain snow density profile model and a method for obtaining grain snow density

Abstract

The application provides a construction method of a granular snow density profile model and a granular snow density acquisition method, and belongs to the technical field of cold region snow and glacier observation. The application first collects multi-dimensional granular snow observation data of a target area, obtains a standardized data set through abnormal value elimination, missing data completion and normalization preprocessing; constructs a multi-physical field coupling initial density profile theoretical model fusing depth, time, temperature and pore elements based on the micro-mechanism of granular snow compaction; performs global iterative optimization on model parameters by using an improved adaptive particle swarm optimization algorithm; divides the granular snow profile into a shallow rapid compaction layer, a middle stable transition layer and a deep dense constant layer according to compaction evolution differences, matches exclusive residual correction strategies for different horizons respectively, eliminates the fitting distortion problem of the full-depth profile, and constructs a high-precision granular snow density profile model.

Description

Technical Field

[0001] This invention belongs to the field of cold region snow cover and glacier observation technology, and more specifically, it relates to a method for constructing a snow grain density profile model and a method for obtaining snow grain density. Background Technology

[0002] Frost density is a core basic parameter in glacier evolution, cold region hydrology, ice sheet mass balance, and polar remote sensing inversion research. The vertical density profile of frost depicts the evolution law of snow compaction and densification from the surface downwards, and is an indispensable input condition for glacier thickness estimation, ice core age analysis, and snowmelt runoff simulation.

[0003] Currently, there are two main methods for obtaining frit density profiles: The first is on-site ice core drilling and sampling, where density is calculated by weighing ice core samples collected in layers. This method provides intuitive data but has significant limitations: drilling operations are costly and the field conditions are harsh, resulting in only discrete depth point data and making it difficult to obtain smooth and continuous longitudinal profiles; furthermore, the number of boreholes is limited, leading to insufficient spatial representativeness and preventing full coverage monitoring of large glacier areas. The second method is empirical model simulation. Most existing mainstream frit density models use a single-exponential fitting model with depth as the only independent variable, considering only the density changes caused by self-weight compaction.

[0004] Existing technologies have several inherent flaws: The modeling of snow granulation relies on a single physical dimension and neglects the coupling effects of multiple factors. The densification process of snow granulation is the result of the synergistic effects of gravity compaction, grain recrystallization, and pore evolution. Snow accumulation duration and snow granulation temperature significantly alter the recrystallization rate, while pore structure directly regulates the upper limit of density. Traditional univariate models only incorporate depth parameters, failing to couple time, temperature, and pore conditions. This results in an incomplete understanding of the physical mechanisms of the model, leading to a significant increase in fitting errors in high-altitude glaciers with seasonal freeze-thaw cycles and in near-polar regions.

[0005] There is a lack of a differential correction mechanism for the stratified profile. The compaction rate of the granular snow vertical profile exhibits obvious stratification characteristics from top to bottom: the shallow granular snow is strongly affected by atmospheric freeze-thaw disturbances and has a fast compaction rate; the middle layer has a smooth transition in compaction; and the deep layer has gradually closed pores and its density tends to stabilize. A unified global fitting model cannot adapt to the differentiated evolution of the three layers, and generally exhibits distortions such as shallow layer density fluctuations, abrupt changes in the predicted boundary between layers, and infinitely continuous growth of the predicted density in the deep layers.

[0006] Traditional models rely on simple parameter optimization methods, which are prone to overfitting and underfitting. They depend on least squares fitting coefficients, which are insufficient for global optimization and make them difficult to adapt to multi-parameter coupled nonlinear models. At the same time, they lack hierarchical error compensation schemes, which further amplify the profile prediction bias.

[0007] The models are static and fixed, with weak adaptability. Once the parameters of existing empirical models are determined and remain unchanged for a long time, they cannot respond to changes in the compaction pattern of snow grains caused by interannual climate fluctuations and changes in snowfall conditions. With long-term use, the prediction accuracy continues to decline, requiring repeated remodeling, which is a lot of work.

[0008] Therefore, it is urgent to propose a snow density profile model construction scheme that takes into account the multi-physics coupling mechanism, hierarchical error correction, and intelligent parameter optimization, and to develop a set of efficient, continuous, and dynamically updatable snow density acquisition methods to solve the technical problems of insufficient accuracy, poor applicability, and high observation cost of existing modeling methods. Summary of the Invention

[0009] To address the technical problems of existing frit density modeling, which mostly relies on deep univariate fitting, ignores the coupling effect of multiple physical parameters, lacks a profile layering correction mechanism, and suffers from severe model fitting distortion; and the traditional method relies on discrete sampling of ice cores, which cannot obtain continuous density profiles, resulting in low calculation efficiency and poor universality, this invention provides a method for constructing a frit density profile model and a method for obtaining frit density, enabling high-precision and continuous calculation of frit vertical profile density in cold regions.

[0010] A method for constructing a snow grain density profile model includes the following steps: S1. Collect multi-dimensional measured data of the target snow / glacier area, construct the original grain snow observation dataset, and perform standardized preprocessing on the dataset to obtain a standardized grain snow sample dataset. S2. Based on the snow compaction evolution mechanism, a multi-physics coupled initial theoretical model of snow density profile is constructed, which includes depth attenuation factor, time accumulation factor, temperature correction factor, and porosity compensation factor. S3. Divide the dataset and use an improved adaptive particle swarm optimization algorithm to iteratively optimize the parameters of the initial theoretical model and determine the optimal set of basic model parameters. S4. The snow pellet profile is divided into a shallow rapid compaction layer, a middle stable transition layer, and a deep dense constant layer. For different layers, corresponding residual correction strategies are used to carry out layer compensation correction, and a high-precision snow pellet density profile model is constructed.

[0011] Preferably, the standardization preprocessing in step S1 includes using the 3σ criterion to remove outliers, distinguishing between short-term missing and continuously missing data and using different methods to complete them, and unifying the physical dimensions through extreme value normalization.

[0012] Preferably, the initial theoretical model expression for the multiphysics coupled particle snow density profile in step S2 is: .

[0013] Preferably, the improved adaptive particle swarm optimization algorithm described in step S3 introduces adaptive inertia weights and dynamic learning factors, sets dual iteration termination conditions, and takes minimizing the mean square error between the measured density and the model calculated density as the optimization objective.

[0014] Preferably, in step S4, wavelet denoising residual correction is used for the shallow layer, piecewise linear gradual residual compensation is used for the middle layer, and constant threshold residual calibration is used for the deep layer.

[0015] A method for obtaining snow grain density, implemented using the aforementioned snow grain density profile model, includes: collecting real-time basic parameters of the target location, inputting the snow grain density profile model to calculate the snow grain density result, verifying the accuracy of the output density, and periodically adding measured data to dynamically update and optimize the model.

[0016] Compared with the prior art, the present invention has the following beneficial effects: This invention constructs a multi-physics coupled initial theoretical model, comprehensively considering the coupling effects of depth, time, temperature, and pore structure, overcoming the limitations of traditional single-variable fitting, resulting in a more complete physical mechanism and stronger regional universality. This invention implements layered differential residual correction for the vertical profile of granular snow, specifically addressing defects such as shallow density fluctuations, abrupt changes in inter-layer fitting, and distortion of extreme density values ​​in deep layers, achieving high-precision fitting of the entire depth profile. This invention employs an improved adaptive particle swarm optimization algorithm to complete global parameter optimization, effectively suppressing overfitting and underfitting problems, and improving the accuracy and stability of model parameter calibration. This invention is accompanied by a complete granular snow density acquisition process, eliminating the need for extensive ice core drilling and sampling, and can output continuous longitudinal density profiles, significantly reducing the workload of field observations. This invention sets clear data accuracy judgment criteria and automatically verifies the model output results, ensuring reliable and controllable density calculation results. This invention supports periodic dynamic iterative updates of the model, adapting to changes in granular snow evolution patterns caused by freeze-thaw cycles and variations in snow cover years, maintaining long-term measurement accuracy. This invention is applicable to various scenarios such as polar glaciers, permanent snow cover in high mountains, and seasonal snow cover, and has a wide range of applications. It can provide reliable data support for hydrological and glacier evolution research in cold regions. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall process of constructing a snow density profile model according to the present invention; Figure 2 This is a schematic diagram of the layered structure of the granular snow in a vertical cross-section according to the present invention; Figure 3 This is a flowchart illustrating the parameter optimization process of the improved adaptive particle swarm optimization algorithm of this invention. Figure 4 This is a schematic diagram illustrating the implementation process of the method for obtaining snow particle density according to the present invention; Figure 5This is a graph comparing the density profile prediction results of the model of the present invention and the traditional model in an embodiment of the present invention. Detailed Implementation

[0018] The present invention will now be described in detail with reference to specific embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention. All other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of the present invention.

[0019] This embodiment selects a high-altitude glacier on the eastern edge of the Qinghai-Tibet Plateau as the target study area. This area has obvious seasonal freeze-thaw action, and the thickness of the granular snow layer can reach 32m. The vertical profile shows a gradual transition from fresh snow layer to granular snow layer and dense granular snow layer from top to bottom. It is suitable for carrying out observation of granular snow compaction evolution and is used to implement the method for constructing a granular snow density profile model and obtaining granular snow density described in this invention.

[0020] Step 1: Simultaneous multi-parameter observation of the target area and standardization preprocessing of the raw dataset Six sets of ice core drilling observation points were set up in flat, ice-flow-undisturbed areas of the target glacier, with a maximum drilling depth of 32m. Layered sampling and in-situ exploration were conducted at 0.2m intervals along the vertical direction. Simultaneously collected observation parameters included: measured density of granules at different vertical depths z, in-situ temperature T of the granule layer, cumulative snow cover duration t, granule porosity, and granule self-weight compaction stress. At the same time, long-term meteorological observation data for the region over the past 15 years, including daily temperature, snowfall, and radiation flux, were collected. All observation information was compiled to form a raw granule observation dataset.

[0021] During field observations, instrument vibration, ice core sampling disturbance, and surface meltwater infiltration can easily generate anomalous data. This method uses the 3σ coarse error discrimination criterion to identify and remove outliers; for short-term discrete missing data at a single depth point, a weighted interpolation method based on upper and lower neighborhoods is used to complete the data; for long intervals of missing data in continuous depth segments and continuous time series, a meteorological element-grain compaction correlation regression model is established for fitting and completion. Due to the significant differences in the dimensions of physical quantities such as density, temperature, depth, porosity, and duration, direct participation in model fitting would cause weight imbalance. Therefore, an extreme value normalization algorithm is used to uniformly map all dimensional observation data to the [0,1] interval, eliminating dimensional interference while fully preserving the original evolution trend and mutual coupling relationship of each physical parameter, ultimately obtaining a standardized grain snow sample dataset. The standardized grain snow sample dataset is randomly divided in a 7:2:1 ratio as the training set, validation set, and test set, respectively. The division process ensures that the shallow, middle, and deep samples are evenly distributed, avoiding model bias caused by stratified sample imbalance in the dataset.

[0022] Step 2: Constructing the initial theoretical model for the multiphysics coupled snow density profile Grain snow undergoes continuous compaction under its own weight, resulting in recrystallization between particles and gradual closure of internal pores. Density evolution is a nonlinear process influenced by depth, compaction time, temperature environment, and pore structure. Traditional empirical models only use depth as a single variable for exponential fitting, neglecting temperature-induced changes in recrystallization rate, the cumulative effect of long-term compaction, and the density compensation effect brought about by pore evolution. This lack of physical mechanism leads to significant fitting errors in areas with active freeze-thaw cycles. Based on the microscopic evolution mechanism of grain snow compaction, a multi-physics coupled initial theoretical model for grain snow density profiles is constructed, incorporating depth attenuation factors, time accumulation factors, temperature correction factors, and pore compensation factors. ; In the formula: ρ(z,t,T) is the snow density corresponding to depth z, cumulative time t, and snow temperature T; ρ0 is the initial reference density of fresh snow on the surface; α is the snow density depth attenuation coefficient; β is the cumulative correction coefficient for snow compaction time; γ is the snow temperature sensitivity correction coefficient; ΔT is the difference between the real-time snow temperature and the regional multi-year average reference temperature; η is the snow porosity dynamic compensation coefficient; all of which are unknown model coefficients to be solved, and need to be obtained by iterative optimization based on standardized snow sample datasets.

[0023] Step 3: Use an improved adaptive particle swarm optimization algorithm to perform global optimization of model parameters. The particle population size is set to 40. The value range of all coefficients to be determined is pre-defined based on the physical boundary conditions of frigid fern grains to avoid physically meaningless solutions during parameter optimization. This algorithm introduces an adaptive inertia weight update strategy: a relatively large inertia weight of 0.8 is set in the early stage of iteration to enhance global search capability and expand the parameter search range; in the later stage of iteration, the inertia weight is linearly reduced to 0.4 to improve local fine-grained optimization capability; simultaneously, the learning factor is dynamically adjusted to balance the optimal information of individual particles with the global optimal information of the population. The mean square error (MSE) between the model-predicted frigid grain density and the measured frigid grain density is used as the objective function; a dual iteration termination condition is set. Condition 1: The mean square error fluctuation is less than 0.5 kg / m3 for 10 consecutive iterations; Condition 2: The number of iterations reaches the preset maximum number of iterations, 200. The iteration stops when any condition is met. After the iteration is complete, the optimal parameter combination is output, resulting in the basic snow density prediction model without stratified correction. A preliminary test of the basic model using a test set reveals significant jump errors at the stratification boundaries, large fluctuations in predicted values ​​in shallow freeze-thaw disturbance areas, and a distortion phenomenon where the density in deeper layers continuously approaches the limit value. Therefore, stratified residual correction is necessary.

[0024] Step 4: Vertical profile division of snow grains and differential residual compensation correction Based on the compaction rate and pore evolution characteristics of the snowpack in this region, the 0–32m snowpack profile is divided into three layers: a shallow, rapidly compacted layer (0–5m), a medium-level, stable transition layer (5–20m), and a deep, dense, and constant layer (20–32m). Specific residual correction strategies are matched to the compaction patterns of each layer. ① Shallow rapid compaction layer (0-5m): Directly in contact with the atmosphere, with frequent seasonal freeze-thaw cycles, fast compaction rate, and obvious random fluctuations in density; a wavelet multi-scale noise reduction residual correction algorithm is used to separate the high-frequency noise term and compaction trend term in the model prediction results, filter out random disturbance errors, and retain the true compaction evolution trend of snow. ②Middle stable transition layer (5-20m): The temperature change range decreases, the snow compaction rate continues to slow down, and the pores shrink slowly; a piecewise linear gradual residual compensation method is adopted to achieve a smooth transition of parameters between the shallow correction model and the deep correction model, and to eliminate the prediction abrupt error at the layer boundary. ③ Deep, dense, and constant layer (20–32 m): The upper part of the snow pellet load is stable, the pores are largely closed, and the density variation is minimal. Constant threshold residual calibration is used to limit the defect of unlimited density growth in the model prediction, which conforms to the objective physical law that the density of deep snow pellets tends to be stable. After completing the layered residual compensation correction, a high-precision, layered adaptive snow pellet density profile model is constructed.

[0025] Step 5: Obtain snow pellet density at field locations based on the constructed model. An independent verification site in the glacier region, not involved in the modeling, was selected for the experiment. Input parameters at the target site were collected through ice core drilling and a distributed temperature and density sensor array. The collected parameters included vertical depth, snow accumulation duration, in-situ frit temperature, and frit porosity. Strict accuracy standards were adhered to: vertical depth accuracy better than 0.1m, frit temperature accuracy better than 0.1℃, porosity accuracy better than 0.5%, snow accumulation duration error not exceeding 15 days, and compaction stress accuracy better than 0.2kPa. Multi-device cross-calibration eliminated systematic errors from individual instruments. All collected real-time parameters were input into the trained and calibrated frit density profile model. The model automatically identified the layer at the target depth, invoked the corresponding residual correction strategy, and, through multi-factor coupling calculations, simultaneously output the frit density value at a specified depth at a single point, as well as a longitudinal continuous frit density profile distribution curve from 0 to 32m.

[0026] Step 6: Verification of model prediction accuracy and determination of validity The in-situ measured snow density from the borehole was selected as the true value. The prediction accuracy was comprehensively evaluated using three indicators: Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Coefficient of Determination (R²). Preset accuracy thresholds were: MAE ≤ 3 kg / m³, RMSE ≤ 5 kg / m³, and R² ≥ 0.98. In this embodiment, the verification point measurement results were: MAE = 2.1 kg / m³, RMSE = 3.6 kg / m³, and R² = 0.986. All three indicators met the preset thresholds, indicating the data was valid. The final snow density result and continuous profile curve were output. If the accuracy indicators did not meet the threshold requirements, the sensor calibration status and on-site sampling process were checked, and parameters were re-collected and recalculated.

[0027] Step 7: Periodic Dynamic Adaptive Update and Optimization of the Model Supplementary field observations are conducted every summer to collect new ice core samples. Valid samples are then selected and incorporated into the standardized frit sample dataset. An improved adaptive particle swarm optimization algorithm is periodically initiated to iteratively optimize various coupling coefficients of the model and the residual correction thresholds corresponding to the three frit layers. Regional climate fluctuations, interannual variations in snowfall, and changes in the strength of freeze-thaw cycles all affect the evolution of frit compaction. By periodically updating the dataset and model parameters, the model can adapt to environmental changes over the long term, maintaining high prediction accuracy without requiring model building from scratch, significantly reducing the cost of repetitive modeling.

[0028] Controlled test description: A traditional single-exponential empirical fitting model relying solely on depth was used as a control group, with identical observation datasets for modeling and prediction. Experimental results showed that the traditional univariate model had an average absolute error of 8.7 kg / m³; in the shallow, actively frozen-thawed zone (0–5 m), predicted values ​​deviated significantly from measured values; and below 20 m, density showed a continuous and unreasonable increase, with the profile curve failing to closely match the measured trend. In contrast, the method of this invention significantly reduced prediction error and achieved excellent full-depth profile fitting.

[0029] Cross-regional expansion implementation instructions: This embodiment takes alpine glaciers as the research object. When this method is extended to polar glaciers, seasonal snow cover on plateaus, and continental snow cover areas, the overall model construction process, the hierarchical residual correction framework, and the steps for obtaining snow grain density do not need to be modified. Only the standardized dataset needs to be reconstructed using the field observation data of the target area, and the model parameters need to be optimized before it can be put into use. The method has strong versatility and is easy to port.

[0030] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0031] The embodiments of the present invention are given for illustrative and descriptive purposes only, and are not intended to be exhaustive or to limit the invention to the forms disclosed. Many modifications and variations will be apparent to those skilled in the art. The embodiments were chosen and described in order to better illustrate the principles and practical application of the invention, and to enable those skilled in the art to understand the invention and to design various embodiments with various modifications suitable for a particular purpose.

Claims

1. A method for constructing a flake density profile model and a method for obtaining flake density, characterized in that, Includes the following steps: S1. Conduct multi-parameter synchronous field observations in the target high-altitude snow / glacier area, collect measured density of granules, temperature of granules layer, years of snow accumulation, porosity of granules, self-weight compaction stress, and long-series meteorological observation data at different vertical depths to construct the original granules observation dataset; perform standardization preprocessing on the original granules observation dataset, complete outlier removal, missing data completion, and dimension unification and normalization to obtain a noise-free, standardized granules sample dataset; S2. Based on the micro-evolution mechanism of snow grain self-weight compaction, recrystallization and pore closure, this paper breaks through the limitations of traditional single-variable fitting modeling that only relies on depth. It comprehensively considers the four core coupling effects of depth decay, time accumulation, temperature stress and pore structure evolution, and constructs a multi-physics field coupled initial theoretical model of snow grain density profile that includes depth decay factor, time accumulation factor, temperature correction factor and pore compensation factor. S3. Divide the standardized snow sample dataset into training set, validation set and test set according to a preset ratio. Use the improved adaptive particle swarm optimization algorithm to perform global iterative optimization on all unknown coefficients and correction factors in the initial theoretical model. Use the validation set to verify the model fitting accuracy, screen out overfitting and underfitting invalid parameter combinations, and lock in the optimal basic model parameter set. S4. Based on the differentiated compaction characteristics of the vertical profile of snow grains, the snow grain layer is divided into a shallow rapid compaction layer, a middle stable transition layer, and a deep dense constant layer. For the differences in density evolution patterns of different layers, a dedicated residual correction strategy is matched for layer-by-layer compensation and correction. This solves the technical defects of traditional models, such as shallow density fluctuation distortion and deep density tending to constant value distortion. Finally, a snow grain density profile model with high accuracy across the entire domain and layer-by-layer adaptation is constructed.

2. The method for constructing a snow grain density profile model according to claim 1, characterized in that, The specific steps of the standardization preprocessing in step S1 include: The 3σ gross error criterion was used to remove abnormal instrumental mutations and observational interference anomalies from the dataset. For short-term missing discrete data at a single time point or location, neighborhood weighted interpolation was used for accurate completion. For long-term missing data with continuous time series and continuous depth, a regional meteorological-snow compaction correlation regression model was used for fitting and completion. The extreme value normalization algorithm was used to uniformly map multi-dimensional heterogeneous data of density, temperature, stress, porosity, and time to the standard interval [0,1], eliminating the interference of dimensional differences while fully preserving the original evolution trend and coupling correlation characteristics of each physical parameter.

3. The method for constructing a snow grain density profile model according to claim 1, characterized in that, The calculation formula for the initial theoretical model of the multiphysics coupled snow density profile in step S2 is as follows: ； in, The real-time density of snow particles corresponding to depth z, cumulative time t, and ambient temperature T; The initial baseline density of fresh snow grains on the ground surface; α is the depth attenuation coefficient of snow grain density; β is the cumulative correction factor for snow compaction time; γ is the temperature sensitivity correction factor for snow pellets; ΔT is the difference between the real-time snow temperature and the regional baseline annual average temperature; η is the dynamic compensation coefficient for the porosity of snow particles. All coefficients are obtained by iterative fitting and optimization calibration of the standardized snow particle sample dataset.

4. The method for constructing a snow grain density profile model according to claim 1, characterized in that, The optimization process of the improved adaptive particle swarm optimization algorithm in step S3 includes: The model initializes reasonable value ranges for the coefficients of each factor and the particle population size. An iterative convergence adaptive weight adjustment mechanism is introduced, which increases the inertia weight in the early stage of iteration to improve global search capability and decreases the inertia weight in the later stage of iteration to improve local precision optimization capability. At the same time, the learning factors are dynamically updated to balance the iteration efficiency between individual optimality and global optimality. The core optimization objective is to minimize the mean square error between the model's calculated density and the measured density, and the optimal particle position and the global optimal parameter combination are continuously updated. A dual iteration termination condition is set, which stops iteration when the mean square error fluctuation is less than a preset threshold for 10 consecutive iterations or when the maximum number of iterations is reached, and outputs the optimal model parameter set.

5. The method for constructing a snow grain density profile model according to claim 1, characterized in that, The specific implementation method of stratified residual correction in step S4 is as follows: The 0–5m granular snow layer is designated as a shallow, rapidly compacted layer. To address its characteristics of rapid compaction, strong meteorological interference, and high-frequency density fluctuations, a wavelet multi-scale noise reduction residual correction algorithm is employed to filter out high-frequency noise errors and retain the true compaction evolution trend. The 5–20m granular snow layer is designated as a mid-level stable transition layer. A piecewise linear gradual residual compensation method is used to achieve smooth connection of model parameters between shallow and deep layers, eliminating abrupt errors in layer boundary fitting. The granular snow layer below 20m is designated as a deep, dense, and constant layer. To address its characteristics of essentially closed pores and stable density, a constant threshold residual calibration algorithm is used to correct the distortion problem of the traditional model's deep layer density infinitely approaching extreme values, achieving high-precision continuous fitting of density across the entire depth profile.

6. A method for obtaining snow pellet density, characterized in that, Performing the application of the snow grain density profile model constructed according to any one of claims 1-5 includes the following steps: S10. Real-time core parameters of the target monitoring point are obtained through ice core drilling, UAV remote sensing, and distributed temperature and density sensor array. The core parameters include vertical depth of detection, duration of snow accumulation, real-time temperature of snow layer, in-situ snow porosity and compaction stress. S20. Input all the collected real-time core parameters into the trained and calibrated snow density profile model. Through multi-factor coupling operation combined with hierarchical residual adaptive correction, output the accurate snow density value of the target single point depth and the continuous snow density longitudinal profile distribution curve of the whole domain. S30. Verify the accuracy and validity of the snow density data and profile curves output by the model. If the accuracy meets the standard, output the final result. If the accuracy does not meet the standard, re-collect the field parameters and repeat the calculation to complete the accurate acquisition of snow density.

7. The method for obtaining snow pellet density according to claim 6, characterized in that, The accuracy control standard for the acquisition of each parameter in step S10 is as follows: The vertical depth acquisition accuracy is no less than 0.1m, the snow temperature acquisition accuracy is no less than 0.1℃, the snow porosity acquisition accuracy is no less than 0.5%, the snow accumulation time traceability error is no more than 15 days, and the compaction stress acquisition accuracy is no less than 0.2kPa. Multi-device collaborative calibration eliminates single-device acquisition system errors, ensuring the accuracy and reliability of input parameters.

8. The method for obtaining snow pellet density according to claim 6, characterized in that, The specific method for accuracy verification in step S30 is as follows: The in-situ measured snow pellet density at the target location was selected as the true value. The mean absolute error, root mean square error, and coefficient of determination between the model output density and the true value were calculated. The preset acceptable accuracy threshold was: mean absolute error ≤ 3 kg / m³. 3 Root mean square error ≤ 5 kg / m 3 Coefficient of determination R 2 ≥0.98; if all three indicators are met, the accuracy is deemed acceptable and the data is valid; if any one indicator is not met, the data is deemed abnormal, and the parameters are re-collected and iterative calculation is performed.

9. The method for obtaining snow pellet density according to claim 6, characterized in that, It also includes a model dynamic adaptive iterative optimization step: Newly measured flake samples from the target area are collected periodically on a quarterly and annual basis, and the newly added valid samples are incorporated into the standardized flake sample dataset. An improved adaptive particle swarm optimization algorithm is periodically launched to re-calibrate the parameters of each coupling factor and the stratified residual correction threshold of the model. This allows the model to adapt to the flake compaction evolution patterns of different seasonal freeze-thaw cycles, different snow cover years, and different climate regions, thereby achieving long-term high-precision and stable operation.

10. The method for obtaining snow pellet density according to claim 6, characterized in that, The acquisition method is applicable to continuous monitoring of snow density at full depth in polar glaciers, permanent snow cover in high mountains, seasonal snow cover in mountains, and snow cover in plateaus. It can simultaneously output instantaneous density data at a single point, longitudinal continuous density profile, and regional snow density spatial distribution characteristic parameters, overcoming the technical shortcomings of traditional manual sampling, which is discrete, has large piecewise fitting errors, and cannot achieve continuous and accurate measurement across the entire area.

Images