Multi-component greenhouse gas flux observation data self-adaptive correction and completion method

Abstract

The invention discloses a multi-component greenhouse gas flux observation data adaptive correction and completion method. The method comprises the following steps: acquiring data by using edge computing nodes deployed in a vortex correlation observation tower and uploading the data to a cloud platform; the cloud platform calculates a Richardson number according to the received data, and constructs a weight function dynamically changing along with the Richardson number to perform weighted adjustment on a sensible heat flux correction term and a latent heat flux correction term in a WPL density correction formula to obtain a correction formula; and correcting the currently received data by using the correction formula. According to the method, the technical problems that cold event interference is difficult to eliminate, density correction errors are large in an unsteady state, sudden emission in a freeze-thaw period is difficult to complement and the like in existing high and cold greenhouse gas flux data observation are solved, so that carbon income and expenditure estimation better conforms to a biological objective law, and the credibility of a scientific conclusion is improved.

Description

Technical Field

[0001] This invention belongs to the field of ecological environment monitoring technology, specifically involving a method for adaptive correction and intelligent completion of greenhouse gases (carbon dioxide CO2 and methane CH4) collected by an eddy covariance observation system deployed in high-altitude permafrost regions, using edge computing real-time diagnostic technology and cloud-based physical constraint artificial intelligence collaborative technology. Background Technology

[0002] Vast areas of permafrost exist in high-altitude and cold regions worldwide, including the Tibetan Plateau. These areas are not only sensitive to global climate change but also vast terrestrial carbon sinks. Under the backdrop of global warming, the decomposition of soil organic carbon caused by permafrost degradation and the resulting release of greenhouse gases (especially methane and carbon dioxide) have a significant positive feedback effect on the global climate system. To accurately quantify this biogeochemical process, eddy covariance techniques based on micrometeorology have been established as the international standard method for flux observation. Eddy covariance techniques measure the covariance between vertical wind speed fluctuations and gas concentration fluctuations at high frequencies (typically 10 Hz), enabling direct determination of greenhouse gas fluxes at the ecosystem scale.

[0003] However, although eddy covariance technology has been successfully applied in temperate and tropical regions, its practical application in high-altitude permafrost areas faces severe technical challenges due to the extreme natural environment and unique physical processes. This results in low data efficiency, poor accuracy, and even scientific conclusions that contradict physical facts. Specifically, existing technologies suffer from three major bottlenecks that are difficult to overcome: First, the challenge of sensor physical failure and false signal identification caused by extreme cold environments.

[0004] In high-altitude and frigid regions, nighttime temperatures often drop below -30°C during winter, and the relative humidity near the ground is frequently close to saturation. The optical window of the open-circuit gas analyzer (such as the LI-7500 series or LI-7700), the core component of the eddy covariance system, is directly exposed to the environment. On extremely cold nights, the surface of the window is highly susceptible to physical frost, condensation, and even snow accumulation.

[0005] Unable to actively eliminate interference: Existing observation systems typically record passively. Once frost forms, the infrared light path is blocked, the signal strength index (RSSI) decreases, and gas concentration readings exhibit non-physical, violent fluctuations (e.g., CO2 concentration readings spike or drop into negative values ​​instantly). This failure state often persists for hours or even days until natural heating melts the frost or manual on-site maintenance is required.

[0006] Inadequate data cleaning: Currently, commonly used data processing software (such as EddyPro) mainly relies on signal strength thresholds (e.g., RSSI < 70%) for hard filtering. This processing method can lead to a data loss rate of 40%-60% during winter nights in cold regions, severely disrupting data continuity.

[0007] High false positive rate: The simple threshold method cannot distinguish between "signal attenuation caused by heavy precipitation" and "signal attenuation caused by frost", nor can it identify the weak characteristics in the early stage of frost. This results in a large amount of contaminated data being mixed into subsequent calculations, causing an overestimation of throughput.

[0008] The classic WPL density correction formula fails theoretically in cold, unsteady atmospheric environments.

[0009] A gas analyzer measures the molar density of a gas. Ecological research requires molar mixing ratios (), while ecological research requires molar mixing ratios (). Because air density varies with temperature and water vapor, density fluctuation correction is necessary, namely the well-known Webb-Pearman-Leuning (WPL) correction.

[0010] The main drawback of existing technology is that: The steady-state assumption is invalid: the WPL formula is derived under the premise that "the atmosphere is in a stable state" and "the vertical average wind speed is zero." However, uneven surface heating in high-altitude and cold regions leads to extreme diurnal temperature variations. Especially during winter nights, strong radiative cooling of the surface forms a thick inversion layer, resulting in extremely stable atmospheric stratification and almost complete disappearance of turbulence (intermittent turbulence). At this time, the atmosphere is under typical "unsteady-state" and "non-stationary" conditions.

[0011] False fluxes: Under strong temperature inversion conditions, although vertical wind speed fluctuations are minimal, the calculated sensible heat flux may still contain a value. If the classic WPL formula is applied directly, a large sensible heat correction term will be imposed on the weak original flux, causing the calculated CO2 flux to show a strong negative value (i.e., carbon absorption). This is biologically absurd (permafrost plants stop photosynthesis in winter, and can only emit weak amounts of respiration, not absorb carbon), thus misleading the assessment of the carbon source / sink function of alpine ecosystems.

[0012] The challenge of nonlinear completion of flux data during the critical freeze-thaw period.

[0013] The spring freeze-thaw cycle is a period of explosive greenhouse gas emissions in high-altitude permafrost regions. As the freezing front moves downwards, frozen methane and carbon dioxide are released in a burst. However, this period is often accompanied by severe weather such as strong winds and blizzards, leading to malfunctions in solar power systems or damage to instruments, making data loss highly likely.

[0014] The main drawback of existing technology is that: Limitations of linear interpolation: Traditional gap-filling methods, such as linear interpolation or the average daily variation method, assume that flux changes are smooth and periodic. They cannot simulate abrupt processes like freeze-thaw bursts.

[0015] Lack of mechanistic constraints: Existing completion methods based on random forests or neural networks are mostly purely data-driven. Under extreme conditions where training samples are scarce, these models cannot understand the physical causal relationship between "soil thawing depth" and "gas release," often predicting smooth curves, leading to a serious underestimation of annual total emissions.

[0016] In summary, there is an urgent need for an intelligent processing method that can actively sense and eliminate physical interference from sensors, adapt to the characteristics of unstable atmospheres in high-altitude and cold regions, and reconstruct data based on the physical mechanisms of permafrost, in order to enhance the scientific value of ecological monitoring data in my country's high-altitude and cold regions. Summary of the Invention

[0017] The purpose of this invention is to solve the technical problems existing in the observation of greenhouse gas flux in high-altitude and cold regions, such as the difficulty in eliminating cold event interference, large density correction errors under unsteady conditions, and difficulty in supplementing sudden emissions during freeze-thaw periods. The invention provides an adaptive correction and supplementation method for multi-component greenhouse gas flux observation data under unsteady conditions in high-altitude and cold permafrost regions.

[0018] This invention adopts a closed-loop architecture of "edge perception - cloud computing - mechanism constraint". At the edge, multi-dimensional signal fingerprints are used to identify frost and trigger active heating; at the algorithm layer, Richardson number is introduced to construct dynamic weights to improve the WPL formula; in the cloud, a large model intelligent agent with embedded physical equations is used to repair data gaps.

[0019] The technical solution of the present invention is as follows: An adaptive correction and completion method for multi-component greenhouse gas flux observation data, the steps of which include: Data is collected using edge computing nodes deployed in the eddy coherence observation tower and uploaded to the cloud platform; The cloud platform calculates the Richardson number based on the received data, constructs a weighting function that dynamically changes with the Richardson number, and adjusts the sensible heat flux correction term and latent heat flux correction term in the WPL density correction formula to obtain the correction formula. The received data is corrected using the aforementioned correction formula.

[0020] Preferably, when there is missing data in the corrected data, the large model agent based on the physical constraints of the freeze-thaw front calculates the depth of the freeze-thaw front based on the multi-layer soil hydrothermal sequence and meteorological driving factors that are contemporaneous with the missing data; and generates missing data based on the depth of the freeze-thaw front to complete the data.

[0021] Preferably, the frost discrimination index is calculated in real time; the frost discrimination index is used to detect whether the window of the gas analyzer is blocked; when the blockage occurs, an active defrosting command is triggered through the edge computing node to drive the heating device of the gas analyzer to remove the blockage, and the data output during the blockage removal process is marked as contaminated data.

[0022] Preferably, the method for weighting the sensible heat flux correction term and latent heat flux correction term in the WPL density correction formula using a weighting function is as follows: Under strongly stable stratification, the weighting function automatically attenuates the correction magnitude of the WPL density correction formula, eliminating spurious carbon absorption flux introduced by incorrect turbulence assumptions; under unstable stratification, the weighting function approaches 1, restoring the standard WPL density correction formula; the method for constructing the large model intelligent agent based on the physical constraints of the freeze-thaw front includes: using a Transformer deep neural network as the time series prediction base and training it; the loss function used in the training process is a composite loss function containing a data fitting error term and a physical mechanism penalty term; the physical mechanism penalty term is used to constrain the output of the time series prediction base to conform to the nonlinear laws of permafrost biogeochemical cycles.

[0023] Preferably, the method for training the time series prediction base is as follows: acquiring greenhouse gas flux data and labeling it as training samples; extracting multi-layer soil hydrothermal sequences and meteorological driving factors contemporaneous with the training samples as feature data; and calculating the multi-layer soil temperature data of the training samples through interpolation. The vertical burial depth of isotherms is used to obtain the freeze-thaw front depth; after incorporating the freeze-thaw front depth into the feature data, it is input into the time series prediction base to output the predicted value of greenhouse gas flux for the currently missing period. According to the predicted value and corresponding training labels Calculate the data fitting error According to the predicted value The physical mechanism penalty term is calculated from the physical mechanism equation of the forced injection of gas into permafrost. Then, based on the data fitting error... and physical mechanism penalty items Calculate the total loss value Optimize the time series prediction base.

[0024] Preferably, the weighting function is an inverse S-shaped function; the data collected by the edge computing node includes high-frequency statistical characteristics of optical path signal strength and original gas concentration, as well as synchronized environmental meteorological parameters; the Richardson number is calculated based on wind speed, wind direction, and temperature data at different altitudes.

[0025] An adaptive correction system for multi-component greenhouse gas flux observation data, characterized in that it includes: The edge-aware control module, deployed on edge computing nodes within the eddy coherence observation tower, is used to collect data and upload it to the cloud platform. The cloud-based algorithm engine module, deployed on a cloud platform, is used to calculate the Richardson number based on the received data, construct a weight function that dynamically changes with the Richardson number, and adjust the sensible heat flux correction term and latent heat flux correction term in the WPL density correction formula to obtain the correction formula. The data correction module uses the correction formula to correct the currently received data.

[0026] Preferably, it also includes an intelligent data completion service module, deployed on a cloud platform, which is used to calculate the depth of the freeze-thaw front based on the multi-layer soil hydrothermal sequence and meteorological driving factors contemporaneous with the missing data when there is missing data in the corrected data; and to generate the missing data based on the depth of the freeze-thaw front to realize the data incompleteness.

[0027] A server is characterized by comprising a memory and a processor, the memory storing a computer program configured to be executed by the processor, the computer program including instructions for performing the methods described above.

[0028] A computer-readable storage medium having a computer program stored thereon, characterized in that the computer program implements the above-described method when executed by a processor.

[0029] Compared with the prior art, the beneficial effects of the present invention are as follows: 1) Significantly improved data recovery rate in cold regions: Traditional methods passively remove frost-covered data, resulting in less than 40% of valid data in winter. This invention improves the data recovery rate to over 75% in winter through active edge heating defrosting, greatly reducing manual maintenance costs.

[0030] 2) Corrected misjudgments of carbon source / sink properties: Dynamic WPL correction effectively solved the industry problem of false "carbon absorption" at night in high-altitude and cold regions during winter, making carbon budget estimation more in line with objective biological laws and improving the credibility of scientific conclusions.

[0031] 3) It restored the real emission process: AI completion based on physical constraints overcame the smoothing effect of traditional interpolation methods and successfully captured the emission peak during the spring freeze-thaw outbreak period, which corrected the annual total methane emission estimate by about 15%, providing more accurate basic data for global change research. Attached Figure Description

[0032] Figure 1 This is a flowchart of the method of the present invention.

[0033] Figure 2 This is a flowchart of the overall technical architecture of the method of the present invention, which shows the data flow process from edge perception, algorithm correction to cloud completion.

[0034] Figure 3 This is a detailed logic flowchart of the verification of the real-time identification and active intervention mechanism for edge-side frost signals in Embodiment 1 of the present invention.

[0035] Figure 4 This is a flowchart of the adaptive correction of unsteady atmospheric flux in Embodiment 2 of the present invention.

[0036] Figure 5 This is a schematic diagram illustrating the data completion principle of a large-scale intelligent agent based on the physical constraints of freeze-thaw fronts in Embodiment 3 of the present invention.

[0037] Figure 6 This is a system diagram of the present invention. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific engineering implementation examples. This section will use the actual application of an eddy flux observation station in a high-altitude wetland in the hinterland of the Qinghai-Tibet Plateau as an example to illustrate the implementation details of this invention.

[0039] The method flow of this invention is as follows: Figure 1 As shown, the steps include: Data is collected using edge computing nodes deployed in the eddy coherence observation tower and uploaded to the cloud platform; The cloud platform calculates the Richardson number based on the received data, constructs a weighting function that dynamically changes with the Richardson number, and adjusts the sensible heat flux correction term and latent heat flux correction term in the WPL density correction formula to obtain the correction formula. The received data is corrected using the aforementioned correction formula.

[0040] An optional embodiment of the present invention provides an adaptive correction and completion method for multi-component greenhouse gas flux observation data under unsteady conditions in high-altitude permafrost regions, such as... Figure 2 As shown, it includes the following steps: Step S1: As Figure 3 As shown, an adaptive cleaning and proactive defense mechanism for cold event interference is constructed at the edge. In the edge computing nodes deployed on the eddy covariance observation tower, high-frequency data communication with the gas analyzer is established to collect gas concentration data. The edge computing nodes continuously collect the optical path signal intensity (RSSI) output by the analyzer, the high-frequency (10Hz) statistical characteristics of the original gas concentration, and synchronous environmental meteorological parameters (air temperature, dew point).

[0041] A physical frost recognition model based on multi-dimensional signal fingerprint feature fusion is constructed. This model calculates the frost discrimination index in real time. When physical frost, condensation, or snow cover is detected in the gas analyzer window, the edge computing node immediately triggers an active defrosting command through the hardware reverse control link.

[0042] The active defrosting command drives the gas analyzer's built-in or external heating device to perform pulsed heating until the signal characteristics return to normal. Simultaneously, the edge computing node marks the data during heating and the initial frosting phase with a specific quality flag (Quality Flag = Bad), completing the cleaning of the source data and preventing contaminated data from being uploaded to the cloud for throughput calculations.

[0043] Step S2: As Figure 4 As shown, the unsteady atmospheric density adaptive correction mechanism of the algorithm layer is constructed. The cloud platform receives the raw flux data and meteorological gradient data uploaded in step S1. Adaptive correction is performed to address the frequent strong temperature inversions and resulting unsteady turbulence characteristics that occur at night in high-altitude cold regions during winter.

[0044] First, sensors were used to collect wind speed, direction, and temperature data at different heights of the observation tower. Then, the Richardson number, a physical quantity characterizing atmospheric turbulence stability, was calculated. This quantifies the vertical mixing capacity of the atmosphere.

[0045] Secondly, constructing Dynamically changing correction weight function This function is used to perform weighted adjustments on the sensible heat flux correction term and latent heat flux correction term in the classic Webb-Pearman-Leuning (WPL) density correction formula.

[0046] In strongly stable stratifications (such as...) Under these conditions, the weighting function automatically decays the correction magnitude of WPL, eliminating spurious carbon absorption flux introduced by incorrect turbulence assumptions; in unstable stratifications (such as... Under these conditions, the weighting function approaches 1, restoring the standard WPL correction. Finally, the original gas concentration data is corrected point by point using the improved WPL density correction formula to generate a corrected greenhouse gas flux time series that eliminates spurious flux interference, which is then used as the input for step S3.

[0047] Step S3: As Figure 5 As shown, a cloud-based intelligent data completion mechanism for physical mechanism constraints is constructed. To address the data marked as "Bad" during step S1 and the temporal gaps in the corrected flux sequence generated in step S2, a large-scale model agent based on the physical constraints of the freeze-thaw front in the cloud is initiated to perform data reconstruction. The specific construction, training, and processing flow is as follows: 3.1 Construction of Training Sample Set and Integration of Physical Features The training of the large model agent based on the physical constraints of the freeze-thaw front depends on the construction of a multi-source heterogeneous sample set: Training labels (denoted as parameters) ): Select the corrected greenhouse gas flux data produced in step S2 that are not marked as Bad and have an excellent quality rating as the physical truth criterion for supervised model learning.

[0048] Input features (denoted as parameters) ): Integrates multi-layer soil hydrothermal sequences (including soil temperature and moisture content data at different depths) and meteorological driving factors collected concurrently.

[0049] Physical state quantity calculation: Specifically calculates the "freeze-thaw front depth" (denoted as a parameter). This is used as the core physical constraint feature. This feature is calculated by spline interpolation of multi-layer soil temperature data. The isotherm was obtained from its vertical burial depth.

[0050] 3.2 Processing flow of time series forecasting base Using a Transformer deep neural network as the basis for time series prediction, it adapts to input features Perform the following processing in sequence: Temporal segmentation and feature extraction: using the sliding window technique to segment input features (including physical state quantities) The time series is divided into fixed-length segments, and the nonlinear mapping relationship and long-range time dependence between environmental driving factors and flux fluctuations are captured by a multi-head self-attention mechanism.

[0051] Generate prediction results: After calculation by the nonlinear mapping layer inside the prediction base, the predicted greenhouse gas flux value for the current missing period is output (denoted as parameter). This parameter It is the core estimator used for subsequent loss calculation and parameter optimization.

[0052] 3.3 Construction of Composite Loss Function and Acquisition of Loss Value During the model training phase, the parameters are... Substituting into the composite loss function yields the total loss value. The total loss value consists of the following two parts: Data fitting error : By calculating the predicted value With training labels The deviation between the two is used to constrain the accuracy of the model in the statistical dimension.

[0053] Physical mechanism penalty item The physical mechanism equations for the forced injection of gases into permafrost (introducing a freeze-thaw burst mechanism equation for methane; introducing a photosynthesis-respiration coupling equation for carbon dioxide). When input features are displayed... When rapidly descending through the soil organic matter layer, if the parameters If no high-intensity gas emission peak is observed, then a very large value is generated through the equation. The penalty value is used to force the model output to conform to the physical laws of frozen soil.

[0054] 3.4 Iterative Optimization and Agent Generation Using the obtained total loss value The gradient of the predicted base's internal weights is calculated using the backpropagation algorithm, and the optimized unit continuously updates the predicted base's internal parameters. By minimizing the physical penalty term, the prediction base is forced to learn the nonlinear physical laws of freeze-thaw outbreaks.

[0055] When the total loss value converges to a preset threshold, the training and knowledge implantation of the prediction base are completed, thereby obtaining a large model agent based on the physical constraints of the freeze-thaw front.

[0056] 3.5 Agent Reasoning and Completion Task The environmental features corresponding to the missing or marked "Bad" periods in steps S1 and S2 are input into the trained agent. Based on the learned physical mechanism-constrained patterns, the agent accurately reproduces the explosive emission characteristics of the freeze-thaw period, generating high-intensity gas emission peaks instead of smooth interpolation, and finally outputs a complete spatiotemporal flux dataset.

[0057] In an optional embodiment, the construction and execution logic of the physical frosting recognition model in step S1 is as follows: 1. Feature vector extraction: Feature 1: The first derivative of optical signal intensity with time Physical frost exhibits a gradual signal attenuation (negative slope), unlike the instantaneous step caused by insect blockage.

[0058] Feature 2: Variance of raw gas concentration data within a short time window (e.g., 1 minute) The ice crystals formed by frost will cause diffuse reflection and refraction of infrared light, resulting in high-frequency "glitches" in the concentration reading at the receiving end, and a significant increase in variance.

[0059] Feature 3: Environmental thermodynamic conditions The physical conditions for frost formation only exist when the temperature is below freezing and close to dew point.

[0060] 1) Calculation of the discriminant index: in, These are the weighting coefficients. The first derivative of the optical path signal intensity; The variance of the original concentration data; This is an environmental condition indicator function used for logical judgments to determine whether the current meteorological environment meets the physical conditions for frost formation. (Environmental condition weights) The weighting coefficients are adaptively adjusted based on the average wind speed of the monitoring stations to compensate for the impact of wind-induced vibrations on the signal variance.

[0061] 2) Closed-loop control: When Exceeding the preset threshold At this time, the system enters "defrost mode". The heating duration is set to... (e.g., 300 seconds). After heating, it enters a cooling and stabilization period. If the temperature drops back to the normal range, then defrosting is considered successful.

[0062] In an optional embodiment, in step S2, the mathematical model of the unsteady atmospheric density adaptive correction mechanism is as follows: 1. Richardson's number calculation: in It is the acceleration due to gravity. For potential temperature, For horizontal wind speed, For height.

[0063] 2. Construct a dynamic weight function: Inverse S-shaped function: in, The critical Richardson number ( The value ranges from 0.2 to 0.3, and is usually taken as 0.25; calibration is performed by comparing the observation residuals of open-circuit gas analyzers and closed-circuit gas analyzers under different stability conditions. The slope of the curve. This function guarantees performance in the turbulent region ( ) In the laminar flow region ( ) .

[0064] 3. Correction formula: in: This is the corrected scalar flux; This represents the original turbulent flux; This is a temperature-density correction factor; The covariance of vertical wind speed fluctuations and temperature fluctuations; The covariance between vertical wind speed fluctuations and water vapor density fluctuations; The total density of air, including dry air and water vapor; This represents the mass of the target gas per unit volume of air. This represents the mass of water vapor per unit volume of air. This formula physically achieves automatic suppression of "spurious corrections caused by unsteady states".

[0065] In an optional embodiment, in step S3, the specific form of the physical mechanism penalty term is: For methane emissions from high-altitude wetlands, physical losses are defined. : in: The flux value predicted by the agent; The temperature sensitivity coefficient for microbial respiration; Soil temperature; , which is the freeze-thaw burst coefficient, characterizing the multiplier effect of gas release when the frozen layer thaws; This is a freeze-thaw state function. When the calculated velocity of the freeze-thaw front exceeds the threshold and is within the depth of the organic matter layer, this value is 1; otherwise, it is 0.

[0066] This constraint forces the AI ​​model to learn that when a freeze-thaw front moves rapidly downwards, it must predict a peak flux value, rather than a smoothed value.

[0067] Example 1: Verification of Real-time Identification and Active Intervention Mechanism for Edge-Side Frost Signals 1. Implementation Scenarios and Problems The experiment was conducted at a high-altitude, frigid observation station at 4800m, equipped with an open-circuit eddy covariance system (LI-7500A + CSAT3). In January, the lowest nighttime temperature reached -28℃, making the sensor's optical window highly susceptible to radiative cooling and frost formation.

[0068] In traditional observation mode, from 03:00 to 08:00 every day, the optical path signal intensity (RSSI) of the LI-7500A often drops slowly from 98% to below 30% due to frost, causing the CO2 concentration readings to fluctuate wildly without any physical meaning (a sharp increase in variance). These data must be completely discarded in post-processing, resulting in an average of 5 hours of missing data per day.

[0069] 2. Implementation steps of the present invention The edge computing gateway (model: Advantech ARK-1123) loaded the algorithm described in step S1: 1) Fingerprint Extraction: The gateway reads raw data 10 times per second. At 03:15, the algorithm detected: Environmental conditions: And the dew point difference (Conditions for frosting are met).

[0070] RSSI characteristics: RSSI decreased from 98.5% to 95.0%, and showed a continuous negative slope. .

[0071] Noise characteristics: Variance of the original CO2 concentration It suddenly increased threefold.

[0072] 2) Logical determination: Substitute into the formula Calculate the frost index Exceeding the set threshold The system determined: "Physical frosting is occurring."

[0073] 3) Active intervention: The gateway immediately closes the heating circuit through the relay to heat the sapphire window of the LI-7500A with a power of 10W.

[0074] 4) Feedback on results: After heating for 5 minutes, the viewing window temperature increased and ice crystals sublimated. The RSSI quickly recovered to 99%, and the concentration variance returned to the normal background value.

[0075] 5) Data cleaning: The gateway automatically marked the data between 03:15 and 03:25 as "Bad", but successfully preserved the high-quality data for the 5 hours from 03:25 until dawn.

[0076] 3. Comparison of Results Compared to a neighboring control site that did not have the system installed (which lost data overnight), the data recovery rate at this site during winter nights increased from 35% to 82%.

[0077] Example 2: Verification of Adaptive Correction for Unsteady Flux Based on Atmospheric Stability 1. Implementation Scenarios and Problems Data segments from typical clear, calm winter nights were selected for analysis. During this period, surface radiative cooling is intense, resulting in a very strong temperature inversion stratification.

[0078] Data characteristics: Vertical wind speed fluctuations Extremely small, but due to the obvious temperature stratification, the sensible heat flux is... The calculated value is relatively large (negative).

[0079] Existing technical bias: If the classic WPL formula is applied, a huge sensible heat correction term will be imposed on the original flux, resulting in a calculated CO2 flux of [value missing]. This appears to be a false carbon sink effect, inconsistent with the biophysical mechanisms of that period, and is a typical false signal.

[0080] 2. Implementation steps of the present invention Stability calculation: The system uses wind speed and temperature data at 2m and 10m on the observation tower to calculate the Richardson number. The results show The value is much greater than the critical value of 0.25, indicating that the atmosphere is in a strongly stable stratification and the turbulence has almost been extinguished.

[0081] Weight adjustment: This occurs during system call step S2. Dynamically changing correction weight function : Correction execution: Multiply the classic WPL correction by a weight of 0.018 (i.e., retain less than 2% of the correction).

[0082] Final result: The corrected flux is This represents minimal soil respiration emissions, perfectly consistent with the actual biogeochemical state of permafrost in winter.

[0083] Example 3: Validation of Intelligent Data Completion for Freeze-Thaw Period Based on Physical Mechanism Constraints of Freeze-Thaw Fronts 1. Experimental Background and Data Selection Observational data from a high-altitude monitoring station in early April 2025 were selected as the verification object. This period coincided with the thawing of the permafrost surface. The original observational data fully recorded a typical methane (CH4) burst emission process, which was used as the baseline true value. To verify the effectiveness of the method of this invention under extreme data missing conditions, a data masking experiment was used to remove eddy flux observational data for 72 hours from April 5th to April 8th, creating data gaps. At the same time, soil temperature, humidity, and meteorological driving data from the same period were retained as input features of the model.

[0084] 2. Implementation steps of the present invention Feature engineering and state recognition: The agent reads multi-layer soil hydrothermal sequence data from the same period. Through calculation and analysis, the depth of the freeze-thaw front at noon on April 6th was identified. It rapidly traversed the organic-rich layer 15-25cm underground, a physical process that corresponds to a dramatic increase in the release potential of the trapped gas.

[0085] Mechanistic constraint reasoning: Agent startup, in the Loss function, The item includes the burst coefficient. (Set to 10 times the background value). The model detected... Due to the drastic changes in physics, the model is forced to predict a high value in order to minimize the physical loss.

[0086] Completed result: The agent generated a Gundam at the data gap on April 6th. Methane emission pulses.

[0087] Scientific verification: Comparison with data from neighboring observations using artificial static chambers confirms the existence of the burst peak. The completion error of this invention is only 8%, while the error of traditional methods is as high as 80%.

[0088] like Figure 6 As shown, an optional embodiment of the present invention provides an adaptive correction system for multi-component greenhouse gas flux observation data, characterized in that it includes... The edge-aware control module, deployed on edge computing nodes within the eddy coherence observation tower, is used to collect data and upload it to the cloud platform. The cloud-based algorithm engine module, deployed on a cloud platform, is used to calculate the Richardson number based on the received data, construct a weight function that dynamically changes with the Richardson number, and adjust the sensible heat flux correction term and latent heat flux correction term in the WPL density correction formula to obtain the correction formula. The data correction module uses the correction formula to correct the currently received data.

[0089] Preferably, it also includes an intelligent data completion service module, deployed on a cloud platform, which is used to calculate the depth of the freeze-thaw front based on the multi-layer soil hydrothermal sequence and meteorological driving factors contemporaneous with the missing data when there is missing data in the corrected data; and to generate the missing data based on the depth of the freeze-thaw front to realize the data incompleteness.

[0090] An optional embodiment of the present invention provides a server, characterized in that it includes a memory and a processor, the memory storing a computer program configured to be executed by the processor, the computer program including instructions for performing the above-described method.

[0091] An optional embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program implements the above-described method when executed by a processor.

[0092] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any simple modifications, equivalent changes, and alterations made by those skilled in the art to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An adaptive correction and completion method for multi-component greenhouse gas flux observation data, comprising the following steps: Data is collected using edge computing nodes deployed in the eddy coherence observation tower and uploaded to the cloud platform; The cloud platform calculates the Richardson number based on the received data, constructs a weighting function that dynamically changes with the Richardson number, and adjusts the sensible heat flux correction term and latent heat flux correction term in the WPL density correction formula to obtain the correction formula. The received data is corrected using the aforementioned correction formula.

2. The method according to claim 1, characterized in that, When there are missing data in the corrected data, the large model agent based on the physical constraints of the freeze-thaw front calculates the depth of the freeze-thaw front based on the multi-layer soil hydrothermal sequence and meteorological driving factors that are contemporaneous with the missing data; and generates missing data based on the depth of the freeze-thaw front to complete the data.

3. The method according to claim 1 or 2, characterized in that, The system calculates the frost discrimination index in real time; it detects whether the window of the gas analyzer is blocked based on the frost discrimination index; when blockage occurs, it triggers an active defrosting command through the edge computing node to drive the heating device of the gas analyzer to remove the blockage, and marks the data output during the blockage removal process as contaminated data.

4. The method according to claim 1, characterized in that, The method for weighting the sensible heat flux correction term and latent heat flux correction term in the WPL density correction formula using a weighting function is as follows: Under strongly stable stratification, the weighting function automatically attenuates the correction magnitude of the WPL density correction formula, eliminating spurious carbon absorption flux introduced by incorrect turbulence assumptions; under unstable stratification, the weighting function approaches 1, restoring the standard WPL density correction formula. The method for constructing the large-scale intelligent agent based on the physical constraints of the freeze-thaw front includes: using a Transformer deep neural network as the time series prediction base and training it; the loss function used in the training process is a composite loss function containing a data fitting error term and a physical mechanism penalty term; the physical mechanism penalty term is used to constrain the output of the time series prediction base to conform to the nonlinear laws of permafrost biogeochemical cycles.

5. The method according to claim 4, characterized in that, The method for training the time series prediction base is as follows: Greenhouse gas flux data are acquired and labeled as training samples; multi-layer soil hydrothermal sequences and meteorological driving factors contemporaneous with the training samples are extracted as feature data; and multi-layer soil temperature data from the training samples are interpolated to calculate… The vertical burial depth of isotherms is used to obtain the freeze-thaw front depth; after incorporating the freeze-thaw front depth into the feature data, it is input into the time series prediction base to output the predicted value of greenhouse gas flux for the currently missing period. According to the predicted value and corresponding training labels Calculate the data fitting error According to the predicted value The physical mechanism penalty term is calculated from the physical mechanism equation of the forced injection of gas into permafrost. Then, based on the data fitting error... and physical mechanism penalty items Calculate the total loss value Optimize the time series prediction base.

6. The method according to claim 1, characterized in that, The weighting function is an inverse S-shaped function; the data collected by the edge computing node includes high-frequency statistical characteristics of optical path signal strength and original gas concentration, as well as synchronized environmental meteorological parameters; the Richardson number is calculated based on wind speed, wind direction and temperature data at different altitudes.

7. An adaptive correction system for multi-component greenhouse gas flux observation data, characterized in that, include The edge-aware control module, deployed on edge computing nodes within the eddy coherence observation tower, is used to collect data and upload it to the cloud platform. The cloud-based algorithm engine module, deployed on a cloud platform, is used to calculate the Richardson number based on the received data, construct a weight function that dynamically changes with the Richardson number, and adjust the sensible heat flux correction term and latent heat flux correction term in the WPL density correction formula to obtain the correction formula. The data correction module uses the correction formula to correct the currently received data.

8. The system as described in claim 7, characterized in that, It also includes an intelligent data completion service module, deployed on a cloud platform. When there is missing data in the corrected data, the large model agent based on the physical constraints of the freeze-thaw front calculates the depth of the freeze-thaw front based on the multi-layer soil hydrothermal sequence and meteorological driving factors that are contemporaneous with the missing data; and generates the missing data based on the depth of the freeze-thaw front to resolve the data incompleteness.

9. A server, characterized in that, It includes a memory and a processor, the memory storing a computer program configured to be executed by the processor, the computer program including instructions for performing the method of any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 6.

Images