CN-122017194-A - Supervision type method and system for identifying nitrous oxide Birch effect and pulse emission amount

CN122017194ACN 122017194 ACN122017194 ACN 122017194ACN-122017194-A

Abstract

The invention relates to a supervision type method and a system for identifying nitrous oxide Birch effect and pulse emission amount, wherein the method comprises the steps of obtaining continuous N2O flux observation data and related covariates, organizing the continuous N2O flux observation data and the covariates on a time axis to form a time sequence, carrying out rolling window judgment and event judgment scoring on the time sequence to generate an effective pulse event section, applying morphological constraint of Birch pulses, determining event start-stop boundaries in the pulse event section subjected to the morphological constraint, estimating background baseline levels of fluxes, finely adjusting the event start-stop boundaries to generate time intervals of pulses from the beginning to fall back to the baseline, carrying out unequal interval time integration on fluxes exceeding the baseline on the time intervals to obtain pulse emission amount of the pulse event section and contribution rate of total annual or seasonal emission, carrying out uncertainty propagation through Monte Carlo sampling, and outputting event levels and annual level confidence intervals.

Inventors

WEI QI
WEI QI
XU JUNZENG
LIU XIAOYIN
CHEN PENG
ZHOU XUE

Assignees

河海大学

Dates

Publication Date: 20260512
Application Date: 20260202

Claims (9)

1. A supervision type method for identifying the nitrous oxide Birch effect and pulse emission amount of soil is characterized by comprising the following steps: acquiring continuous N2O flux observation data and related covariates, and uniformly organizing the continuous N2O flux observation data and the related covariates on the same set of time axis to form a time sequence; performing rolling window judgment and event judgment scoring on the time sequence to generate an effective pulse event segment, and applying morphological constraint of Birch pulse for eliminating the pulse event segment which does not accord with typical pulse characteristics; determining event start-stop boundaries in morphologically constrained pulse event segments, and estimating background baseline levels of flux for fine tuning the event start-stop boundaries, generating time intervals for pulses to start and fall back to baseline; and carrying out time integration at unequal intervals on the flux exceeding the baseline in the time interval to obtain the pulse emission of the pulse event section and the contribution rate of the pulse event section to the total annual or seasonal emission, carrying out uncertainty propagation through Monte Carlo sampling, and outputting event level and annual level confidence intervals.
2. The supervised identification soil nitrous oxide Birch effect and pulse emission calculation method of claim 1, wherein generating the valid pulse event segments comprises: Setting a rolling window with a target length, calculating the median and the quartile range of a flux value in each rolling window, and adaptively determining a threshold coefficient according to a target false alarm rate to generate a rolling threshold value, wherein T=A+k·IQR; when the numerical value of a plurality of continuous sampling points in the time sequence exceeds the rolling threshold value in the rolling window, marking the continuous sampling points as candidate pulse event segments; Extracting multidimensional feature vectors representing the amplitude, the morphology and the time sequence background of the event from the candidate pulse event segments, inputting the multidimensional feature vectors into a supervised time sequence event scoring model, obtaining event scores, and taking the candidate pulse event segments with the event scores exceeding a screening threshold as effective pulse events; In the training process, the historical data in the training set is adopted to conduct weak supervision pre-training, then manually marked pulse samples are used for conducting fine adjustment correction on the model, and a focus loss function or a grading threshold strategy is introduced to reduce the bias of the model to non-events.
3. The supervised identification soil nitrous oxide Birch effect and pulse emission calculation method of claim 1, wherein applying morphological constraints of the Birch pulse comprises: Limiting the duration of the pulse event segment to a target range, taking an event flux curve as a target, splitting the pulse event segment into a plurality of single-peak sub-events according to the rule of local extremum and valley depth if a secondary peak exists in the pulse event segment in the target range, restricting the ascending and descending slope characteristics and the lag time length of the pulse event segment, namely controlling the ratio of the maximum ascending rate to the maximum descending rate to the target interval, and controlling the time lag between the pulse occurrence and the triggering event to be not more than the preset maximum time length.
4. The supervised identification soil nitrous oxide Birch effect and pulse emission calculation method of claim 1, wherein determining the event start-stop boundaries comprises: And (3) carrying out change point analysis on the flux sequence in the pulse event section subjected to morphological constraint by adopting a parallel pruning accurate linear time algorithm and a Bayesian online change point detection method, if the difference between the time of the start point or the end point of any pulse event section given by the pruning accurate linear time algorithm and the time of the start point or the end point of any pulse event section given by the Bayesian online change point detection method is not more than a preset tolerance, and the confidence coefficient of the Bayesian online change point detection method at the corresponding moment is not lower than a target threshold, determining the time point as an event start-stop boundary of the pulse event section, and if the difference between the time of the start point or the end point of any pulse event section given by the pruning accurate linear time algorithm and the time of any pulse event section given by the Bayesian online change point detection method is more than the time tolerance or the confidence coefficient is insufficient, triggering local fine adjustment in the adjacent area of the boundary, and carrying out secondary positioning on the boundary position with divergence to obtain the event start-stop boundary of the pulse event section.
5. The supervised identification soil nitrous oxide Birch effect and pulse emission calculation method of claim 1, wherein generating the time interval comprises: Fitting a baseline of the soil N2O flux observation data by using the observation data of a non-pulse period, namely fitting the baseline after masking by using the current identified pulse, reevaluating whether residual pulse influence which is not masked exists according to the baseline, repeatedly updating a mask and fitting again until the baseline is converged to a stable state; And constructing a local interpolation curve based on original observed data points in the vicinity of the beginning and ending of the event in the event beginning and ending boundary, and determining the intersection point position of the pulse segment flux curve and the baseline curve to serve as the time interval from the beginning and the falling of the pulse to the baseline.
6. The method for monitoring and identifying nitrous oxide Birch effect and pulse emission amount calculation according to claim 1, wherein in the process of unequal interval time integration of flux exceeding a base line in the time interval, if a short data missing segment exists in an event time window, a ground fault state space model is introduced to perform multiple interpolation on the data missing segment, namely, adjacent observed values are utilized to perform predictive interpolation on the data missing segment, a plurality of interpolation sequences are generated through random sampling of the ground fault state space model, the multiple interpolation sequences are used for obtaining multiple interpolation scenes of a complete flux curve of the event, and pulse emission amount is calculated for each interpolation scene respectively to perform multiple interpolation integration.
7. The supervised identification soil nitrous oxide Birch effect and pulse emission calculation method of claim 1, wherein outputting the event level and grade level confidence interval comprises: And sampling the threshold coefficient, the base line and the start-stop boundary of the event by using a joint Monte Carlo, obtaining a group of different threshold coefficients, base line and start-stop boundary of the event by each simulation sampling, recalculating the pulse emission and the contribution rate, repeatedly simulating the target times, summarizing all simulation results, obtaining probability distribution of the event emission and the contribution rate, and finally outputting the credible range of the emission and the contribution rate of each pulse event in a confidence interval mode.
8. The supervised identification soil nitrous oxide Birch effect and pulse emission calculation method of claim 7, wherein joint monte carlo sampling of said threshold coefficients, said baseline, and said event start-stop boundaries comprises: Sampling and taking values according to the estimated error range of the threshold coefficient, generating a plurality of baseline curves through residual distribution of the baselines, and recalculating pulse flux by using the baselines with slight deviation for each simulation, and sampling and disturbing the start and stop time of each pulse event segment in the neighborhood of the start and stop time based on the boundary posterior probability.
9. A supervised identification soil nitrous oxide Birch effect and pulse emission calculation system implemented by the method of any of claims 1-8, comprising: The data management module is used for acquiring continuous N2O flux observation data and related covariates, and organizing the continuous N2O flux observation data and the related covariates on the same time axis uniformly to form a time sequence; the rolling window judging and event scoring module is used for scoring the time sequence through rolling window judgment and event judgment, and generating an effective pulse event segment; The morphological constraint module is used for applying morphological constraint of the Birch pulse and removing pulse event segments which do not accord with typical pulse characteristics; The baseline estimation and integration module is used for determining event start-stop boundaries in a pulse event section subjected to morphological constraint, estimating the background baseline level of flux, fine-adjusting the event start-stop boundaries, generating a time interval from the beginning to the falling of a pulse to the baseline, and carrying out time integration at unequal intervals on the flux exceeding the baseline on the time interval to obtain the pulse emission of the pulse event section and the contribution rate of the pulse emission to the total annual or seasonal emission; and the uncertainty evaluation module is used for carrying out uncertainty propagation through Monte Carlo sampling and outputting event-level and grade-level confidence intervals.

Description

Supervision type method and system for identifying nitrous oxide Birch effect and pulse emission amount Technical Field The invention relates to the technical field of ecological environment monitoring and greenhouse gas emission evaluation, in particular to a method and a system for supervising and identifying the nitrous oxide Birch effect and the pulse emission amount of soil. Background Soil nitrous oxide (N2O) often appears short-time high-throughput "hot periods/pulses" (Birch effect) after triggering events such as rainfall, fertilization, irrigation and freeze thawing, which are relatively rare in time, have higher amplitude than background flux and contribute disproportionately to annual emission, high-time-frequency automatic observation reveals that such events are significantly increased, but currently quantitative definition and unified identification flow of "hot periods" are not yet agreed, and low-frequency manual observation is also prone to peak value missing, thereby affecting accuracy of regional to global balance assessment and inventory compilation. The existing threshold class method (mean value +/-standard deviation, 1.5 x IQR and the like) and unsupervised anomaly detection (such as IsolationForest, IF) have significant fluctuation of threshold values along with background variance and season segmentation under different seasons and data distribution conditions, SD and IQR are easy to be missed due to the influence of bias heavy tail distribution on threshold value setting, IF may cause excessive marking due to super parameters and distribution assumption, and individual scenes even approximately totally attribute the season accumulated flux to a thermal period. Comprehensive evaluation also shows that a single statistical method is difficult to stably apply across places and across management situations. Therefore, it is desirable to design an automatic identification and pulse emission calculation scheme with both statistical robustness and event morphology determination capability, which is helpful to reduce the missing detection/false detection of the conventional fixed threshold method and the excessive marking problem of the single unsupervised method, and provides a basis for the uniform identification and accounting across seasons and regions. Disclosure of Invention Aiming at the short-time high-flux 'Birch effect' of soil nitrous oxide (N2O) after triggering such as rainfall, fertilization, irrigation, freeze thawing and the like, the existing method taking fixed threshold value or single unsupervised anomaly detection as a core has the problems of insufficient robustness across seasons and management situations, high missing detection and false detection rate, difficult automatic refinement of event starting and stopping boundaries, lack of overall process uncertainty quantification and standardized accounting links and the like. The invention aims to provide a supervision type method and a system for identifying the nitrous oxide Birch effect and the pulse emission amount, which can automatically and robustly identify Birch pulses, accurately define event start and stop boundaries, calculate peak values, duration, pulse emission amount and contribution rate according to uniform caliber and output event level and grade confidence intervals on the premise of not changing an original flux sequence sampling grid. In order to achieve the above object, the present invention provides the following solutions: A supervision type method for identifying the nitrous oxide Birch effect and pulse emission amount of soil comprises the following steps: acquiring continuous N2O flux observation data and related covariates, and uniformly organizing the continuous N2O flux observation data and the related covariates on the same set of time axis to form a time sequence; performing rolling window judgment and event judgment scoring on the time sequence to generate an effective pulse event segment, and applying morphological constraint of Birch pulse for eliminating the pulse event segment which does not accord with typical pulse characteristics; determining event start-stop boundaries in morphologically constrained pulse event segments, and estimating background baseline levels of flux for fine tuning the event start-stop boundaries, generating time intervals for pulses to start and fall back to baseline; and carrying out time integration at unequal intervals on the flux exceeding the baseline in the time interval to obtain the pulse emission of the pulse event section and the contribution rate of the pulse event section to the total annual or seasonal emission, carrying out uncertainty propagation through Monte Carlo sampling, and outputting event level and annual level confidence intervals. Optionally, generating the valid pulse event segment includes: Setting a rolling window with a target length, calculating the median and the quartile range of a flux value in each rolling window, and ad