Snow surface change determination method and system based on multi-source time sequence characteristics and space-time constraints

Abstract

The application provides a snow surface change determination method and system based on multi-source time sequence characteristics and space-time constraints, and relates to the fields of cryosphere environment monitoring, meteorological sensor data processing and multi-source information fusion. The method is executed by a snow surface change determination system with multiple sensor nodes distributed along the height. Each node acquires time series of at least two types of physical quantities such as illumination, temperature and pressure disturbance. After quality screening and reference normalization, multi-time window features are extracted. According to the separable degree, the reliability weight of each physical quantity is dynamically determined and fused to obtain the snow coverage probability. Through vertical monotonic constraint global solution and sub-interval probability interpolation, the snow-air interface height higher than the physical installation interval is obtained. Then, the interface migration is confirmed by the bidirectional time state machine, and the snow surface height change and change rate are output. The application does not rely on fixed absolute threshold, is adaptive to polar day and night and complex weather, effectively suppresses frost and blowing snow interference, and is suitable for long-term unattended monitoring in polar, high mountain and seasonal snow areas.

Description

Technical Field

[0001] This invention relates to the fields of cryosphere environmental monitoring, meteorological sensor data processing, and multi-source information fusion, specifically to a method for automatically identifying snow using light, temperature, and pressure disturbance sensor data distributed parallel along the altitude direction. A method and system for determining the snow or air exposure status at the air interface and calculating the snow surface rise and fall rate and change rate. This method is applicable to long-term unattended snow monitoring scenarios such as Antarctica, the Arctic, high mountains, glaciers, and seasonal snow cover areas. Background Technology

[0002] In polar and alpine snow monitoring, continuous changes in snow surface height directly reflect the accumulation and ablation processes, serving as a key parameter for assessing the Earth's surface material balance. Traditional manual snowplow readings are intuitive, but they cannot provide high-temporal-resolution long-term data under unmanned conditions. While ultrasonic, laser, and radar ranging devices can directly obtain distance values, they are prone to signal attenuation, misreading, or irreversible maintenance difficulties due to conditions such as polar frost, strong winds, tilted installation, changes in surface roughness, and long-term power supply constraints.

[0003] The prior art discloses a temperature chain device in which multiple temperature sensors are arranged at fixed intervals (typically 2 cm) along a measuring rod. This device infers snow temperature by measuring the vertical temperature profile and combining it with a heat conduction model. Location of the air-to-air interface. Alternatively, some solutions deploy optical nodes along the measuring rod to determine snow cover status by detecting changes in ambient light intensity. These sensor array structures are relatively simple and suitable for long-term deployment, but existing solutions often rely on single-shot light intensity thresholds or fixed temperature thresholds to determine whether a node is covered by snow. This reliance on absolute thresholds is highly susceptible to misjudgments under conditions of alternating polar day / night cycles, rapid changes in cloud cover, frost adhesion, snow blowing, sensor drift, and the thermal inertia of snow. Furthermore, the independent classification of each node fails to utilize the vertical structural pattern that nodes below the snow surface are generally continuously covered by snow, while nodes above the snow surface are generally continuously exposed to air. An isolated failure of a single node can lead to inferred snow cover levels. The gas interface exhibited unreasonable and sudden fluctuations.

[0004] More importantly, an inherent limitation of such sensor arrays is that, regardless of the backend algorithm used, the effective resolution of snow surface height changes is, in principle, limited by the physical mounting spacing of the sensor entities. For a device with a node spacing of 2 cm, actual snow surface displacements below this scale may only appear as a single node's state flip or remain unchanged in the sensor output, making reliable differentiation impossible. Existing methods also typically lack effective mechanisms for temporal confirmation of "true accumulation / melting" and "short-term snow blowing, frost, or signal anomalies," resulting in unstable snow surface change sequences. Therefore, under low power consumption constraints, there is an urgent need for a snow surface change determination method and system that can overcome hardware physical spacing limitations, adapt to the complex polar environment, and possess both high reliability and high resolution. Summary of the Invention

[0005] To address the technical problems in existing technologies, such as reliance on fixed absolute thresholds leading to misjudgments of snow cover status under different lighting and meteorological conditions, independent node-by-node classification failing to utilize vertical structural patterns resulting in interface jumps, lack of a time confirmation mechanism for interface migration making it difficult to distinguish between continuous accumulation and short-term disturbances, and effective resolution limited by the physical installation spacing of sensors, this invention proposes a snow surface change determination method and system based on multi-source temporal features and spatiotemporal constraints. It eliminates background drift through reference relative normalization, utilizes dynamic weighted fusion of multiple physical quantities' separability to obtain the node snow cover probability, and uses a vertical monotonic constraint for global solution to obtain a unique snow cover probability. The system simulates the snow surface interface and confirms the actual interface migration using a bidirectional time state machine. It also utilizes probabilistic gradient sub-interval interpolation to overcome the limitations of physical spacing in hardware, achieving millimeter-level resolution without increasing hardware cost or power consumption. The system outputs snow surface height, rate of change, and result confidence.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for determining snow surface changes based on multi-source temporal features and spatiotemporal constraints is executed by a snow surface change determination system based on multi-source temporal features and spatiotemporal constraints. The snow surface change determination system includes multiple sensing nodes distributed along the height direction. Each sensing node has illumination sensing, temperature sensing, and pressure disturbance sensing capabilities. Furthermore, the snow surface change determination system provides physical quantities belonging to different physical property categories on the same height profile. These physical quantities include illumination, temperature, and pressure disturbance, including: Step 1. Use low duty cycle burst sampling to acquire the time series of at least two types of physical quantities of each of the sensor nodes within multiple acquisition cycles, and form a rolling acquisition window by multiple acquisition cycles; Step 2. Perform quality screening on the time series to determine the high-confidence air reference node set and the high-confidence snow layer reference node set; Step 3. Based on the high-confidence air reference node set and / or the high-confidence snow layer reference node set, perform reference relative normalization on the time series and extract at least two dimensionless features from fluctuation amplitude, change correlation, phase lag, frequency band energy or disturbance attenuation; Step 4. Based on the signal quality of each physical quantity within the current acquisition window and the separability between the high-confidence air reference node set and the high-confidence snow layer reference node set, determine the reliability weight of each physical quantity and fuse them to obtain the fusion coverage probability of each sensor node; when the separability decreases, the reliability weight of the corresponding physical quantity is automatically reduced through a continuous nonlinear function; when a certain type of physical quantity loses its ability to be distinguished, the power supply to the corresponding sensor is automatically turned off by the snow surface change determination system. Step 5. Using the constraint that the snow cover state satisfies a monotonically distributed distribution along the height direction on both sides of the snow-air interface, the global boundary solution is performed on the snow cover probability of all the sensor nodes to obtain the candidate snow-air interface; the cost function includes a penalty term for isolated contradictory nodes to suppress local anomalies caused by frost or foreign object obstruction; after determining the boundary, the probability gradient of adjacent sensor nodes is used to interpolate on the sub-node spacing to obtain the interface height higher than the physical installation spacing. Step 6. Based on the direction of movement of the candidate snow-air interface relative to the previous confirmed interface and the consistency and continuity of the state transition of multiple physical quantities of the sensor nodes adjacent to the interface in multiple consecutive acquisition windows, confirm whether the snow surface is rising, the snow surface is falling, or the interface has not changed; the minimum number of consecutive windows for continuity is determined according to the relationship between the duration of polar snow blowing and the length of the acquisition window, so as to distinguish between continuous snow accumulation changes and short-term disturbances. Step 7. Calculate the snow surface height change and rate of change based on the confirmed snow-air interface height sequence.

[0007] On the other hand, the present invention also provides a snow surface change determination system based on multi-source temporal features and spatiotemporal constraints, including multiple sensing nodes distributed along the height direction, a memory and a processor. The memory stores a program, and when the program is executed by the processor, the system implements the above-mentioned snow surface change determination method based on multi-source temporal features and spatiotemporal constraints.

[0008] On the other hand, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when run on a processor, causes the processor to implement the above-described method for determining snow surface changes based on multi-source temporal characteristics and spatiotemporal constraints.

[0009] On the other hand, the present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements the above-mentioned method for determining snow surface changes based on multi-source temporal characteristics and spatiotemporal constraints.

[0010] Compared with the prior art, the beneficial effects of the present invention are: 1. By replacing the absolute threshold with a reference relative feature, the influence of differences in different sites, seasons, cloud cover, and device gain is eliminated, enabling stable operation under a wide range of background changes such as polar day / polar night transitions.

[0011] 2. By utilizing the differences in physical response to light, temperature, and pressure disturbances, and by continuously adjusting nonlinear weights based on the separability of the channels in the current environment, misjudgments caused by frost, snow, calm winds, and near-isothermal conditions are effectively reduced, and the fusion process is smooth without hard switching.

[0012] 3. By incorporating the coverage probability of all nodes into the global cost function solution with vertical monotonic constraints, frost or fault jumps of individual sensors can be automatically suppressed, resulting in a unique snow cover that conforms to physical laws. Gas interface.

[0013] 4. Based on the relationship between the duration of snowfall and the length of the acquisition window, a two-way time confirmation condition for the state machine is set. It is required that the interface migration is accompanied by the directional multi-physical quantity characteristic transformation between air and snow, which can accurately distinguish between continuous accumulation / melting and short-term occlusion, instantaneous wind accumulation and other illusions.

[0014] 5. By using the probability gradient of adjacent nodes for sub-spacing interpolation, the effective height resolution breaks through the limitation of the physical installation spacing of the sensor and reaches the millimeter level without increasing hardware cost and power consumption.

[0015] 6. The intermediate calculation results, such as the separability, are directly applied to the sensor power supply control. When the channel fails, the relevant circuits are automatically shut down, which significantly reduces the overall power consumption of the system and realizes the deep integration of the algorithm and the physical process of the hardware.

[0016] 7. The method mainly involves statistics, linear combinations and one-dimensional search. It has low computation and storage requirements, can run in real time on low-power microcontrollers in polar unmanned stations, and can operate sustainably for a long time.

[0017] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall process of the method of the present invention; Figure 2 The probability of global interface selection under vertical monotonic constraints Height distribution diagram; Figure 3 This is a schematic diagram illustrating the state transition between two-way time confirmation and transient disturbance rejection. Detailed Implementation

[0019] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the following embodiments are only used to explain the present invention and are not intended to limit the scope of protection of the present invention. Those skilled in the art can adapt and adjust the preset parameters and docking systems in the embodiments according to actual regulatory requirements.

[0020] like Figure 1 As shown, the snow surface change determination method based on multi-source temporal features and spatiotemporal constraints provided by this invention operates on a snow surface change determination system based on multi-source temporal features and spatiotemporal constraints. This system includes multiple sensor nodes distributed along the height direction, a controller, a memory, and a power management module. The sensor nodes are fixedly arranged along the height direction, and the known height of each node is denoted as . (i=1…N, arranged from low to high). Each node simultaneously possesses light, temperature, and pressure disturbance sensing capabilities, and the entire system provides all physical quantities on the same height profile. These different physical quantities refer to measurements belonging to different physical property categories, specifically including light, temperature, and pressure disturbance. The system performs short-term burst sampling within the acquisition period to form node time series, and completes subsequent processing within a rolling window. The entire methodology includes seven core steps: short-term acquisition, quality screening, reference normalization, multi-time-window feature extraction, adaptive fusion, monotonic boundary determination, and time series confirmation. The final output includes snow surface height and confidence level, snowfall / melting event classification, and accumulation / melting rate.

[0021] The method for determining snow surface changes based on multi-source temporal features and spatiotemporal constraints specifically includes the following steps: Step 1: Low duty cycle synchronous acquisition and scrolling window construction; When the preset base cycle is reached, the controller wakes up from sleep mode and performs synchronous or quasi-synchronous data acquisition on each altitude node. To balance power consumption and dynamic characteristics, short-time burst sampling and cross-cycle rolling statistics are adopted. The base cycle is set to 1. 6 hours, 30 minutes after each wake-up Within 180 seconds at 1 Illumination and pressure disturbance data are sampled at 10Hz, while temperature data is recorded synchronously at a lower frequency; the system immediately enters sleep mode after data acquisition. Multiple consecutive base cycles constitute a rolling acquisition window (i.e., a long-term window) used to extract features such as temperature fluctuation amplitude, phase delay, diurnal variation response, and variation trend.

[0022] The system includes an adaptive power scheduling function: based on the environmental suitability factor of a physical channel calculated in subsequent steps... When the power consumption falls below a preset wake-up threshold, such as when light separation is zero under polar night conditions or insufficient pressure disturbance under calm wind conditions, the controller generates a channel shutdown command to cut off the power supply to the corresponding sensor node, further reducing power consumption. Once the applicable factor recovers, the system performs a short-term tentative power-on verification. Furthermore, when candidate interface migration, strong wind disturbance, or a decrease in channel confidence is detected in the previous cycle, the system enters an enhanced sampling mode, shortening the acquisition cycle or extending the burst sampling duration within a limited time; the basic cycle resumes after the candidate event ends. This increases sampling only when the information gain is high, avoiding long-term high-frequency operation.

[0023] After completing the low duty cycle synchronous acquisition and rolling window construction, the raw time series data of each sensor node in multiple acquisition cycles are obtained, and then the data quality is screened in step 2.

[0024] Step 2: Data quality screening and outlier node marking; Range checks, consecutive identical value checks, abrupt jump checks, missing rate checks, and neighborhood consistency checks are performed on the time series data for illumination, temperature, and pressure, respectively. For channels identified as saturated, frozen, disconnected, or experiencing long-term drift, quality factors are generated. Where m represents the physical channel category. The quality factor is used for reliability weighting in subsequent steps and does not directly determine the snow state. One specific way to generate the quality factor is to assign a pass flag to each inspection item. When a node experiences a range exceedance, consecutive identical values, or a sudden jump within a preset number of consecutive acquisition cycles, the quality factor of the physical channel corresponding to that node is reduced to below 0.5; when multiple inspections fail simultaneously, the quality factor is further reduced. Alternatively, the quality factor can be taken as the minimum value or product of the pass rates of each inspection; the specific implementation is not limited.

[0025] Neighborhood consistency is determined using the median and median absolute deviation (MAD) of neighboring nodes at the same time. If a single node has long-term statistical inconsistencies with its upper and lower neighbors, while the remaining nodes form a continuous interface, the node is marked as an isolated anomalous node, and its contribution is reduced in subsequent global solutions.

[0026] After quality screening, abnormal nodes are identified and quality factors for each channel are generated. The selected effective time series and quality factors are then passed to step 3.

[0027] Step 3: Self-reference region construction and normalization; The system selects a high-confidence air reference set from the current profile. Snow layer reference set Upon initial startup, the topmost nodes are used as candidate air references, and the bottommost nodes are used as candidate snow layer references. After startup, only nodes with stable states across multiple consecutive windows and confidence levels above a preset threshold are used to update the reference set. Nodes adjacent to the interface, candidate migration nodes, and abnormal nodes must not be used to update the reference, to prevent interface migration from contaminating the baseline.

[0028] For any channel x, relative normalization is performed using the reference median and MAD to eliminate absolute value drift caused by seasonal variations, differences in device gain, and sensor aging, resulting in the relative eigenvalue: In the formula, This represents the original measurement value of node i at time t; For air reference set The median of the measurements at each node within the node; The median absolute deviation of the air reference set; It should be a small positive number to prevent division by zero.

[0029] When the air reference set or snow reference set is insufficient, the most recent valid reference is used, and its timeliness confidence is gradually reduced. If both types of references are insufficient, the system outputs an "uncertain" status and does not force an update of the snow surface.

[0030] After relative normalization, the normalized eigenvalues ​​of each node relative to the air background or snow background are obtained. and high-confidence air reference sets Snow layer reference set Proceed to step 4 to extract features across multiple time windows.

[0031] Step 4: Multi-time window feature extraction; Illumination, temperature, and pressure perturbation characteristics were calculated for each node. These characteristics were expressed in dimensionless forms such as ratios, correlations, energy, and phase to reduce dependence on specific device models and calibration coefficients. The main characteristics include the following: Illumination characteristics: Illumination attenuation ratio The ratio or logarithmic difference between the median nodal illumination and the median air reference illumination; illumination fluctuation ratio. The ratio of the detrended illumination MAD at the node to the MAD of the air reference illumination is taken; illumination correlation. Take the correlation coefficient or coherence coefficient between the node illumination sequence and the air reference illumination sequence.

[0032] Temperature characteristics: Temperature fluctuation suppression ratio Take the ratio of the detrended temperature fluctuation amplitude at the node to the air reference fluctuation amplitude; temperature phase lag. The time delay corresponding to the maximum cross-correlation between the node temperature and the air reference temperature is taken; the temperature band energy ratio The ratio of low-frequency energy to high-frequency energy is used to characterize the thermal buffering and low-pass effect of the snow layer.

[0033] Pressure disturbance characteristics: pressure disturbance energy After high-pass filtering the pressure sequence, the root mean square (RMS), MAD, or band energy are calculated; pressure reference correlation. The correlation or coherence between the node pressure disturbance and the upper air reference disturbance is taken to distinguish between common pressure changes and local port blockage.

[0034] Vertical local contrast Extract the differences between the node features and the features of its upper and lower neighboring nodes to highlight the change zone near the interface.

[0035] The extracted multi-time-window features include the illumination attenuation ratio. Light fluctuation ratio Light-related Temperature fluctuation suppression ratio Temperature phase lag Temperature band energy ratio Pressure disturbance energy Pressure reference correlation and vertical local contrast All features are then passed to step 5 for adaptive fusion.

[0036] Step 5: Adaptive fusion coverage probability calculation; This step is divided into two stages: single-channel evidence generation and weight determination, and multi-source fusion.

[0037] 5.1 Single-channel coverage evidence and adaptive reliability weights; By inputting multiple features of each type of physical channel into an interpretable evidence function, the single-channel snow cover probability is obtained. The evidence function can be a piecewise linear function, a logistic function, a fuzzy membership function, a lightweight decision tree, or a calibrated linear classifier. The preferred form of the logistic function is: In the formula, Let k be the feature value of the m-th channel of node i; and The parameters can be initially obtained through indoor snow burial / air exposure experiments, and adaptively corrected on-site based on a high-confidence reference set.

[0038] Calculate channel reliability At that time, the overall channel quality factor Separability between air reference and snow reference and environmental suitability factors Separability Divide the difference in the medians of the two reference sets by the combined MAD: In the formula, and These are the sets of eigenvalues ​​for the m-th channel under air reference and snow layer reference, respectively. and These are the median absolute deviations of the two values, respectively. It should be a small positive number to prevent division by zero.

[0039] Environmental suitability factors A continuous nonlinear mapping is performed based on the current environment's support for channel discrimination capabilities to avoid hard switching. For physical channel identification, , , These correspond to the environmental suitability factors for the temperature channel, light channel, and pressure channel, respectively.

[0040] Taking the applicability of temperature channels under near-isothermal conditions as an example, when air Snow layer temperature gradient When the reading is below the sensor's effective identification limit, a logarithmic decay function is used: In the formula, This refers to the equivalent temperature difference for sensor thermal noise or the equivalent temperature difference for long-wave radiation background fluctuations. When the air-snow layer temperature gradient... When it approaches 0, Continuous smoothness approaches 0, improving the reliability of the temperature channel. This effect is subsequently diminished. During polar nights or when the dynamic range of illumination is insufficient, the illumination channel... Environmental suitability factors of light channels Gradually reduce according to a similar principle. A continuous attenuation mapping can be constructed based on the ratio of the signal-to-noise ratio or effective dynamic range of the illumination signal to the sensor's lowest resolvable illuminance; the environmental suitability factor of the pressure channel when the port is blocked or in still airflow. A continuous attenuation mapping can be constructed based on the ratio of the RMS value of the pressure disturbance signal to the sensor's noise floor. The applicability factors for each channel all satisfy... When the corresponding physical quantity lacks effective discrimination capability, it smoothly approaches 0. This yields the dynamic channel reliability weight. : The aforementioned nonlinear mapping enables the fusion system to transition smoothly between different environments, rather than simply switching to a backup channel when a channel fails.

[0041] 5.2 Multi-source fusion coverage probability; Single-channel evidence is weighted and fused using channel reliability, and node quality factors and local vertical consistency are superimposed. A log-odds fusion method is employed to ensure that extreme probabilities in a single channel do not easily suppress other channels. In the formula, ; For bias terms; The weighting coefficient for local vertical consistency; Let be the vertical local contrast features of node i. After fusion, the fused coverage probability of node i at time t is obtained. When the number of effective channels is less than the set number or the fusion probability is in the preset gray zone, the node status remains unknown and is not subject to subsequent mandatory determination.

[0042] After adaptive fusion, the fusion coverage probability of each sensor node is obtained. And the reliability weights of each physical channel in the current environment. Proceed to step 6 to perform a global solution for the snow-air interface.

[0043] Step 6: Snow with vertical monotonic constraints Global solution and interpolation of the air-gas interface; For nodes arranged in ascending order of height, the ideal coverage state should satisfy the condition that nodes below the interface are covered by snow, and nodes above the interface are exposed to air. Using each possible boundary j between adjacent nodes as a candidate interface, the global cost function is calculated. : In the formula, The weights are determined by node quality and channel reliability; The fusion coverage probability obtained in step 5; The known installation height of the node immediately below candidate boundary j; The height of the interface confirmed at the previous moment; This is a function to penalize interface movement, suppressing unreasonable jumps when there is insufficient evidence. The number of isolated nodes that contradict the candidate boundary. The coefficient for conflict punishment. This constitutes a penalty item for isolated contradictory nodes.

[0044] Isolated contradiction node penalty item Directly corresponding to physical failure modes in the polar environment: When an individual node exhibits a coverage state opposite to that of surrounding nodes due to frost on the windward side, short-term shading by wild animals, or local ice crystal adhesion, the node is identified as an isolated physical anomaly rather than a true change in the snow surface, and its impact on the global interface is reduced through a penalty term.

[0045] Iterate through all candidate boundaries j and select the one that makes them equal to the boundary j. smallest This serves as the optimal candidate boundary. The computational complexity of this process is O(N), making it suitable for low-power microcontrollers.

[0046] Determine the optimal candidate boundary Later, when and At that time, sub-interval interpolation is performed between nodes on both sides of the boundary using probability gradients: In the formula, Let be the snow-air interface height after interpolation at time t. and They are nodes With nodes The known installation height and Each of the corresponding nodes With nodes The fused coverage probability after step 5 is 0.5, which is the preset snow coverage probability threshold. The interpolation result is limited to... This step utilizes the continuity of probability distribution to reconstruct the continuous position of the snow-air interface within the physical sensor spacing, enabling effective height resolution to overcome the mechanical limitations of sensor mounting spacing, reaching the millimeter level, without increasing any hardware costs.

[0047] like Figure 2 As shown, under the vertical monotonic constraint, the fused snow cover probability of each node exhibits a gradual distribution along the height, ranging from close to 1.0 (lower snow cover area) to close to 0.0 (upper air area). The global optimal boundary is determined at the position with the steepest probability gradient. The dashed line in the figure represents the estimated snow cover probability. The interface confidence level is calculated by combining the difference between the optimal and suboptimal costs, the probability interval on both sides of the boundary, the number of effective channels, and historical consistency. A higher confidence level is achieved by a larger cost difference, a more pronounced separation of probabilities on both sides, and a higher degree of channel consistency.

[0048] After global solution and sub-interval interpolation, the candidate snow-air interface height is obtained. The interface confidence level is then input into step 7 for time confirmation and final judgment.

[0049] Step 7: Confirm the migration between the two-way time state machine and the actual interface; Receive the candidate snow-air interface height output from step 6 Establish a hysteresis state machine that includes stable, rising candidate, rising confirmed, falling candidate, falling confirmed, and uncertain / fault states, such as Figure 3 As shown, bidirectional temporal confirmation is performed on the interface migration. The new candidate interface must meet the conditions for establishment in at least N windows within M consecutive acquisition windows, and at least two types of physical evidence near the boundary must show a state transition with the same direction.

[0050] The state machine parameters M and N are not based on purely empirical values, but rather on physical calculations derived from the typical duration of polar snowstorm events and the length of the acquisition window. Let the average or percentile duration of snowstorm events over many years of local observations be... The length of the complete feature extraction window is Then the following is required: Ensure that the minimum observation duration required to confirm the interface migration is greater than the lifecycle of a typical snowstorm event, thus physically eliminating wind artifacts. M can be set as an appropriate multiple of N to provide fault tolerance.

[0051] Evidence for the direction of snow rise preferably includes: attenuation or decreased fluctuation in the light response of the original air node, enhanced suppression of temperature fluctuations or increased phase lag, and a significant reduction in pressure disturbance energy. Evidence for the direction of snow fall is the reverse recovery process of the above characteristics. If only a single physical quantity changes briefly and other effective physical quantities do not support the transition of the cover state, the change is marked as a transient event caused by snow blowing, cloud shadow, frost, obstruction, or sensor anomaly, and the confirmed snow cover is not updated. Gas interface.

[0052] The state machine employs a hysteresis mechanism with different entry and exit thresholds, and sets a minimum hold time. After state confirmation, the current snow surface height and reference set are updated; when candidate evidence decays, it returns to a stable state without changing the confirmed snow surface.

[0053] After confirmation, the snow confirmed via the state machine is processed. air interface height Perform a confidence-weighted robust regression. Huber regression or Theil regression can be used. Sen slope estimation or weighted median slope estimation is used to reduce the impact of a small number of outliers. Time interval The rate of change of the snow surface within is the fitted slope. Positive values ​​indicate that the snow surface is rising, while negative values ​​indicate that the snow surface is falling.

[0054] Based on the direction and duration of interface changes, the background of pressure disturbances, and temperature characteristics, the event type can be further categorized as: surface uplift due to snowfall / wind deposition, surface subsidence due to wind erosion / subsidence / melting, or an uncertain mechanism. When external data such as wind speed, radiation, or snow density are lacking, excessive inferences about specific mechanisms are avoided.

[0055] If representative snow density is obtained simultaneously Then the rate of change of snow height can be converted into the rate of change of surface mass: Otherwise, only output the rate of change of snow surface height. The final output information should include at least the snow... Air-air interface height, snow surface rise or fall status, snow surface change rate, effective physical quantity indicators, abnormal node indicators, and interface confidence level.

[0056] The above seven steps constitute the core decision-making process of this method. To ensure long-term self-sustaining operation, the method also includes reference set self-update and parameter calibration, low power consumption, and fault degradation mechanisms, as detailed below: In a stable state, nodes that are far from the current interface and maintain consistently high confidence are added to the corresponding reference set, and the reference statistics are updated using exponential moving median or limit-based recursion. Nodes near the interface, candidate change nodes, and anomalous nodes are strictly excluded from the update. The evidence function parameters are adaptively adjusted after two types of clustering using the high-confidence upper and lower reference sets on-site. The adjustment range and step size are both subject to preset limitations, and the factory parameters are retained as fallback values.

[0057] Regarding low power consumption and fault degradation, in addition to the adaptive power supply control described in step 1, the system shuts down or enables the light channel at extremely low frequencies in polar night mode, focusing primarily on temperature and pressure disturbances; in near-isothermal mode, it automatically reduces the temperature weight, prioritizing light and pressure; in calm wind mode, it reduces the pressure channel weight and does not interpret "low disturbance" solely as snow burial. When the node failure rate exceeds the allowable threshold, the system outputs a low-confidence state or stops updating to avoid generating untraceable erroneous data. During communication, snow surface height, event flags, confidence levels, quality flags, and key features are transmitted first, while raw data is stored locally or transmitted back in batches to save communication energy.

[0058] The following are examples of applying the method proposed in this invention: Taking snow surface change monitoring at an unmanned Antarctic station as an example, N sensing nodes are densely arranged at 5mm intervals on a vertical measuring component. Each node integrates an ambient light sensor, a digital temperature sensor, and a wind pressure pulsation sensing element. Ten additional nodes, permanently exposed to air, are located at the top as initial air references. The controller is activated every 3 hours, and the light and pressure sensors operate at 1... The temperature sensor samples at 10Hz for 60 seconds, and takes 6 samples during this period. 12 times; scroll window length is 24. 72 hours.

[0059] After performing range, missing, and neighborhood checks in step 2, the quality factor for each channel is calculated. Step 3 constructs and normalizes air and snow layer references from stable nodes. Step 4 extracts illumination features including attenuation ratio, fluctuation ratio, and reference correlation; temperature features including fluctuation suppression ratio, phase lag, and band energy ratio; and pressure features including high-frequency disturbance RMS and air reference correlation. Step 5 assigns weights according to the separability and environmental suitability factors of each channel. The typical value of the equivalent temperature difference for sensor thermal noise is taken as 0.05K. Step 6 determines the optimal boundary through a global cost function and obtains the millimeter-level interface height within a 5mm spacing through probabilistic interpolation. Step 7 sets the state machine based on continuous observation of local snow blowing. Hours, window length After calculating N≥1 over several hours, considering reliability, M=5 and N=3 are selected to ensure that the true accumulation occurs continuously across multiple windows. After confirmation by the state machine, the most recent 7... 30-day high-confidence height sequence for Theil Sen slope estimation yields the average accumulation or ablation rate.

[0060] During the above operation, the light channel during the polar night... When the temperature drops to zero, the system automatically cuts off the power supply to the light sensor; when the temperature gradient is insufficient, The weights are reduced logarithmically, and the fusion weights are smoothly transferred to the pressure disturbance channel while the interface judgment remains stable. The final output snow surface height change resolution can reach the millimeter level, and the overall system average power consumption is maintained at the microwatt to milliwatt level, meeting the requirements for long-term unattended operation.

[0061] Other implementation methods: This invention also provides a snow surface change determination system based on multi-source temporal features and spatiotemporal constraints, comprising multiple sensing nodes distributed along the height direction, a memory, and a processor. The memory stores a computer program, and when the processor executes the program, it implements the steps of the snow surface change determination method based on multi-source temporal features and spatiotemporal constraints described in any of the above embodiments. The sensing nodes, memory, and processor can be integrated into the same data acquisition device or deployed distributedly via wired or wireless means. The processor can be a low-power microcontroller, and the memory can include flash memory or static random access memory for storing acquired data, intermediate features, state machine states, and output results.

[0062] The present invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the snow surface change determination method based on multi-source temporal characteristics and spatiotemporal constraints described in any of the above embodiments. The computer-readable storage medium may be a non-volatile storage medium, such as a read-only memory, flash memory, magnetic disk, or optical disk.

[0063] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the snow surface change determination method based on multi-source temporal characteristics and spatiotemporal constraints described in any of the above embodiments. The computer program product can be deployed as a software package in the data processing terminal of an unmanned polar monitoring station.

[0064] Although the present invention has been described in detail with reference to the accompanying drawings and preferred embodiments, the invention is not limited thereto. Various equivalent modifications or substitutions can be made to the embodiments of the invention by those skilled in the art without departing from the spirit and essence of the invention. Such modifications or substitutions should all fall within the scope of the invention, or any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the invention should be covered within the protection scope of the invention. Therefore, the protection scope of the invention should be determined by the scope of the claims.

Claims

1. A method for determining snow surface changes based on multi-source temporal characteristics and spatiotemporal constraints, executed by a snow surface change determination system based on multi-source temporal characteristics and spatiotemporal constraints, wherein the snow surface change determination system comprises multiple sensing nodes distributed along the height direction, each sensing node having illumination sensing, temperature sensing, and pressure disturbance sensing capabilities, and the snow surface change determination system provides physical quantities belonging to different physical property categories on the same height profile, the physical quantities including illumination, temperature, and pressure disturbance, characterized in that... include: Step 1. Use low duty cycle burst sampling to acquire the time series of at least two types of physical quantities of each of the sensor nodes within multiple acquisition cycles, and form a rolling acquisition window by multiple acquisition cycles; Step 2. Perform quality screening on the time series to determine the high-confidence air reference node set and the high-confidence snow layer reference node set; Step 3. Based on the high-confidence air reference node set and / or the high-confidence snow layer reference node set, perform reference relative normalization on the time series and extract at least two dimensionless features from fluctuation amplitude, change correlation, phase lag, frequency band energy or disturbance attenuation; Step 4. Based on the signal quality of each physical quantity within the current acquisition window and the separability between the high-confidence air reference node set and the high-confidence snow layer reference node set, determine the reliability weight of each physical quantity and fuse them to obtain the fusion coverage probability of each sensor node; when the separability decreases, the reliability weight of the corresponding physical quantity is automatically reduced through a continuous nonlinear function; when a certain type of physical quantity loses its ability to be distinguished, the power supply to the corresponding sensor is automatically turned off by the snow surface change determination system. Step 5. Using the constraint that the snow cover state satisfies a monotonically distributed distribution along the height direction on both sides of the snow-air interface, the global boundary solution is performed on the snow cover probability of all the sensor nodes to obtain the candidate snow-air interface; the cost function includes a penalty term for isolated contradictory nodes to suppress local anomalies caused by frost or foreign object obstruction; after determining the boundary, the probability gradient of adjacent sensor nodes is used to interpolate on the sub-node spacing to obtain the interface height higher than the physical installation spacing. Step 6. Based on the direction of movement of the candidate snow-air interface relative to the previous confirmed interface and the consistency and continuity of the state transition of multiple physical quantities of the sensor nodes adjacent to the interface in multiple consecutive acquisition windows, confirm whether the snow surface is rising, the snow surface is falling, or the interface has not changed; the minimum number of consecutive windows for continuity is determined according to the relationship between the duration of polar snow blowing and the length of the acquisition window, so as to distinguish between continuous snow accumulation changes and short-term disturbances. Step 7. Calculate the snow surface height change and rate of change based on the confirmed snow-air interface height sequence.

2. The method according to claim 1, characterized in that, In step 1, low duty cycle burst sampling is used in multiple acquisition cycles. The basic state is acquired according to the first cycle. When candidate interface migration or channel reliability is detected, the sampling is switched to a second cycle shorter than the first cycle for enhanced sampling. After the candidate event ends, the first cycle is restored.

3. The method according to claim 1, characterized in that, The specific manifestations of the dimensionless features in step 3 include multiple types such as illumination attenuation ratio, illumination fluctuation ratio, illumination correlation, temperature fluctuation suppression ratio, temperature phase lag, temperature frequency band energy ratio, pressure disturbance energy, pressure reference correlation, and vertical local comparison of the features of a node with its upper and lower neighboring nodes.

4. The method according to claim 1, characterized in that, In step 4, the reliability weight is determined by the channel quality factor, the feature separability between the high-confidence air reference node set and the high-confidence snow layer reference node set, and the environmental applicability factor. When the illumination is below the effective dynamic range, the temperature is close to isothermal, or the pressure disturbance is insufficient, the reliability weight of the corresponding physical quantity is reduced.

5. The method according to claim 1, characterized in that, The continuous nonlinear function in step 4 acts on the environmental applicability factor. Among them, the environmental suitability factor of the temperature channel Using a logarithmic decay function: ； In the formula, The air-snow temperature gradient, This refers to the equivalent temperature difference of sensor thermal noise or the equivalent temperature difference of long-wave radiation background fluctuations. Environmental suitability factors of light channels A continuous attenuation mapping is constructed based on the ratio of the light signal signal-to-noise ratio to the minimum resolvable illuminance, and the environmental suitability factor of the pressure channel is also considered. A continuous attenuation mapping is constructed based on the ratio of the pressure disturbance RMS value to the sensor's background noise. The applicability factor of each channel smoothly approaches 0 when the corresponding physical quantity loses its ability to distinguish.

6. The method according to claim 1, characterized in that, The cost function for global boundary solving in step 5 for: ； In the formula, For node weights, This represents the probability of snow cover. Given the known installation height of the node immediately below candidate boundary j. The interface height confirmed at the previous moment. For the move penalty function, The number of isolated contradictory nodes, The coefficient for conflict punishment. This constitutes a penalty item for isolated contradictory nodes.

7. The method according to claim 1, characterized in that, The interpolation in step 5 is calculated using the following formula: ； In the formula, Let be the snow-air interface height after interpolation at time t. and They are nodes With nodes The known installation height and Each of the corresponding nodes With nodes The merged snow cover probability is calculated, with 0.5 as the preset snow cover probability threshold; the interpolation result is limited to... Inside.

8. A snow surface change determination system based on multi-source temporal features and spatiotemporal constraints, comprising multiple sensing nodes distributed along the height direction, a memory, and a processor, characterized in that, The memory stores a program that, when executed by the processor, causes the system to implement the snow surface change determination method based on multi-source temporal characteristics and spatiotemporal constraints as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run on the processor, it causes the processor to implement the snow surface change determination method based on multi-source temporal characteristics and spatiotemporal constraints as described in any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the snow surface change determination method based on multi-source temporal characteristics and spatiotemporal constraints as described in any one of claims 1 to 7.

Images