Mangrove blue carbon ecological risk identification and early warning method and system for climate change

Abstract

The present application relates to the technical field of climate change warning, and discloses a mangrove blue carbon ecological risk identification and warning method and system for climate change, comprising the following steps: acquiring time series of global land surface characteristic satellite leaf area index data and environmental factors, performing detrending, deseasonalization and standardization processing to extract abnormal sequences; determining climate oscillation windows and neutral periods according to multivariate El Nino-Southern Oscillation index and dipole mode index, and calculating the mean value of the environmental factors in the neutral period as a reference; inputting the environmental factors into a long short-term memory network to extract attention weights, calculating the factor difference between the mean value of the climate oscillation window and the reference, mapping the factor difference by using a nonlinear saturation kernel function, combining the weights to obtain contribution amplitude, and selecting a dominant risk factor by comparing the direction; processing the absolute value of the leaf area index anomaly by using an S function to output risk intensity, and predicting future leaf area index anomaly based on the dominant risk factor to generate a warning evaluation.

Description

Technical Field

[0001] This invention relates to the field of climate change early warning technology, specifically to a method and system for identifying and warning of blue carbon ecological risks in mangroves in response to climate change. Background Technology

[0002] Mangroves, as important blue carbon ecosystems, play a vital role in carbon sequestration and emission reduction. However, due to climate change and extreme weather events, mangrove ecosystems face severe degradation risks. Accurately identifying and providing early warnings of the ecological risks posed by climate change to mangroves is of great significance for ecological protection.

[0003] Existing methods for assessing mangrove ecological risks typically rely on linear models to establish the correlation between environmental factors and ecological indicators, neglecting the nonlinearity and threshold effects of mangrove responses to environmental stresses. When processing complex environmental time-series data, current technologies struggle to effectively distinguish between long-term astronomical tidal drivers and short-term climate system anomaly drivers, easily misinterpreting normal fluctuations caused by astronomical effects as climate anomaly risks. Existing assessment systems fail to exclude residual physical effects after extreme climate events when determining benchmarks, resulting in the selected benchmark state being influenced by previous events and reducing the accuracy of risk attribution. Furthermore, existing early warning mechanisms lack effective handling of the time lag of environmental factors, failing to dynamically quantify the sensitivity distribution of different environmental factors to abnormal fluctuations in leaf area index, and struggling to scientifically map the abnormal absolute values ​​of ecological indicators to specific carbon storage losses, leading to a lack of intuitive and quantitative scientific basis for risk warning assessment results. Therefore, this invention proposes a method and system for identifying and warning of blue carbon ecological risks in mangroves in response to climate change, addressing the shortcomings of existing technologies. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for identifying and warning of blue carbon ecological risks in mangroves in response to climate change. It solves the problems of existing assessment methods relying on linear models and ignoring the nonlinear threshold effect of ecological response, the difficulty in effectively distinguishing between long-period astronomical tides and short-term climate anomaly driving components, and the residual influence of previous extreme climate events on the baseline state.

[0005] To achieve the above objectives, the present invention provides the following technical solution: The first aspect of the present invention provides a method for identifying and warning of blue carbon ecological risks in mangroves in response to climate change, comprising the following steps:

[0006] We acquired global land surface characteristic satellite leaf area index data and environmental factors time series, performed detrending, deseasonalizing and standard score standardization processing, and extracted anomalous sequences containing leaf area index anomalies.

[0007] Based on the multivariate El Niño-Southern Oscillation Index and the Dipole Model Index, the climate oscillation window and neutral period are determined, and the mean values ​​of environmental factors during the neutral period are calculated as a benchmark.

[0008] Environmental factors are input into a long short-term memory network to extract attention weights. The difference between the mean of environmental factors and the benchmark factor is calculated within the climate oscillation window. The factor difference is mapped using a nonlinear saturated kernel function and combined with the attention weights to obtain the contribution magnitude. The direction of the contribution magnitude is compared with the direction of leaf area index anomaly to select the dominant risk factor.

[0009] By processing the absolute value of leaf area index anomalies through S-shaped function risk mapping, the risk contribution and risk intensity are output. Based on the dominant risk factors, future leaf area index anomalies are predicted and early warning assessments are generated.

[0010] Preferably, the extraction of anomalous sequences containing leaf area index anomalies specifically includes: calculating the ratio of mangrove pixels to the total number of pixels to obtain the coverage rate; applying a preset coverage rate threshold to filter mixed pixels and boundary pixels; using linear regression to fit the long-term trend of global land surface feature satellite leaf area index data to extract detrended data; subtracting monthly climatological data to extract residual sequences to complete deseasonalization; applying the mean and standard deviation to the residual sequences for standard score standardization; setting the standardized value to zero when the standard deviation is equal to zero; and outputting the anomalous sequence. This invention eliminates long-term interannual variation trends and periodic seasonal fluctuations through linear regression and deseasonalization, removes background climate noise, and extracts anomalous signals caused by climate oscillations.

[0011] Preferably, determining the climate oscillation window and neutral period based on the multivariate El Niño-Southern Oscillation Index and the Dipole Model Index specifically includes: combining the multivariate El Niño-Southern Oscillation Index, the Dipole Model Index, and the duration span condition to determine El Niño events, La Niña events, and positive Indian Ocean Dipole events, and outputting a climate oscillation window including the start and end times; extracting time nodes based on a preset index range to divide them into neutral periods, setting the consecutive months after the end of the climate oscillation window as the recovery period, and removing time nodes within the recovery period from the neutral period. This invention establishes a benchmark reference unaffected by climate oscillations by eliminating the residual physical effects after the end of a climate event and filtering out interference from previous events.

[0012] Preferably, the extraction of attention weights from environmental factors by inputting them into a Long Short-Term Memory (LSTM) network specifically includes: constructing an input matrix of environmental factors; processing the input matrix through the forgetting gate, input gate, and output gate mechanisms of the LSTM network; extracting the output of the last time step of the parallel channel as a hidden feature representation; calculating the attention score of the hidden feature representation using a fully connected layer and a hyperbolic tangent activation function; and applying a normalized exponential function to perform probability normalization on the attention score to extract attention weights. This invention uses a LSTM network combined with an attention mechanism to process nonlinear mapping relationships with time lags, quantifying the sensitivity distribution of different environmental factors to abnormal fluctuations in leaf area index data.

[0013] Preferably, the time series of environmental factors includes a sea level anomaly time series. The acquisition of global land surface characteristic satellite leaf area index data and environmental factor time series also includes separating the monthly node periodic contribution of the sea level anomaly time series: a multivariate harmonic regression model is constructed by introducing trigonometric function terms containing perigee subharmonics and the monthly node period to fit the sea level anomaly time series; the amplitude coefficients in the multivariate harmonic regression model are estimated using the least squares algorithm, and the monthly node periodic contribution determined by celestial motion is calculated by combining the trigonometric function terms; the monthly node periodic contribution is subtracted from the sea level anomaly time series to separate the sea level anomaly driven by climate oscillations, and the sea level anomaly driven by climate oscillations is used as the time series of the acquired environmental factors. This invention distinguishes between long-period astronomical tidal-driven components and short-term climate system-driven components, avoiding the misclassification of long-term fluctuations caused by astronomical effects as climate system anomalies.

[0014] Preferably, the method of using a nonlinear saturated kernel function to map factor differences and combining it with attention weights to derive contribution magnitudes specifically includes: applying a nonlinear saturated kernel function containing a saturation rate parameter and a hyperbolic tangent activation function to map and transform the absolute values ​​of factor differences to calculate the magnitude of factor anomalies; multiplying the attention weights by the factor anomaly magnitudes to calculate the contribution magnitude of each environmental factor; and introducing a sign function to assign the original positive or negative sign of the factor differences to the multiplication result to calculate the attribution contribution of the environmental factors. This invention introduces a nonlinear saturated kernel function to characterize the diminishing marginal effect, correcting the calculation errors caused by directly extrapolating climate anomalies in linear models, and reflecting the threshold effect of environmental factor anomalies on mangrove growth.

[0015] Preferably, the selection of the dominant risk factor by comparing the direction of the contribution amplitude with the direction of the leaf area index anomaly specifically includes: selecting the environmental factor with the largest contribution amplitude as a candidate factor; determining that when the attribution contribution value of the candidate factor and the leaf area index anomaly value are both positive or both negative, selecting the candidate factor as the dominant risk factor; determining that when the attribution contribution value of all environmental factors is inconsistent with the leaf area index anomaly value, selecting the environmental factor with the largest attribution contribution product as the dominant risk factor and adding a direction inconsistency marker. This invention verifies the consistency between the contribution direction of physical driving factors and the observed direction of ecosystem degradation, ensuring that the attribution results conform to the physical mechanism of ecological response.

[0016] Preferably, the specific steps for processing abnormal absolute values ​​of leaf area index (LAI) using sigmoid function risk mapping to output risk contribution and risk intensity include: combining a dimensionless scaling constant, a risk response steepness parameter, and a risk trigger threshold; processing abnormal absolute values ​​of LAI using sigmoid function risk mapping to obtain risk contribution; calculating the ratio of abnormal absolute values ​​of LAI to the historical maximum abnormal absolute values ​​of LAI to obtain risk intensity; and multiplying the risk contribution, the area of ​​the risk region, and the mangrove carbon density to calculate carbon storage loss, and quantifying the risk intensity through carbon storage loss. This invention uses sigmoid function risk mapping to process abnormal absolute values ​​of LAI to correspond to the nonlinear variation law of vegetation degradation under stress.

[0017] Preferably, the specific steps for predicting future leaf area index anomalies and generating early warning assessments based on dominant risk factors include: calculating predicted values ​​of environmental factors based on currently acquired climate indices and historical response patterns; calculating predicted future leaf area index anomalies using attention weights, sign functions, and nonlinear saturated kernel functions of dominant risk factors; combining predicted future leaf area index anomalies, risk intensity, and carbon storage loss to classify risk warning levels; and generating early warning assessments using the determined risk warning levels as output results.

[0018] This invention also provides a mangrove blue carbon ecological risk identification and early warning system for climate change, used to execute the mangrove blue carbon ecological risk identification and early warning method for climate change as described in the first aspect of this invention, comprising: a data processing module, used to acquire global land surface characteristic satellite leaf area index data and environmental factor time series, perform detrending, deseasonalizing and standard score standardization processing, and extract anomalous sequences containing leaf area index anomalies; and a baseline determination module, used to determine the climate oscillation window and neutral period based on the multivariate El Niño-Southern Oscillation Index and Dipole Model Index, and calculate the environmental factors in the neutral period. The system uses the mean value of the period as a benchmark; the attribution analysis module is used to input environmental factors into a long short-term memory network to extract attention weights; it calculates the factor difference between the mean value of environmental factors and the benchmark value in the climate oscillation window, uses a nonlinear saturated kernel function to map the factor difference and combines it with attention weights to obtain the contribution magnitude, and compares the direction of the contribution magnitude with the direction of leaf area index anomaly to select the dominant risk factor; the risk warning module is used to process the absolute value of leaf area index anomaly through S-shaped function risk mapping to output the risk contribution and risk intensity, predict future leaf area index anomalies based on the dominant risk factor and generate a warning assessment.

[0019] This invention provides a method and system for identifying and providing early warning of blue carbon ecological risks in mangroves in response to climate change. It has the following beneficial effects:

[0020] 1. This invention acquires global land surface characteristic satellite leaf area index data and environmental factors time series, performs detrending, deseasonalization, and standard score standardization, and separates the monthly node periodic contribution of sea level anomaly time series. Through these processing steps, this invention eliminates long-term interannual variation trends, periodic seasonal fluctuations, and long-period astronomical tidal driving components, avoiding the misidentification of long-term fluctuations caused by astronomical effects as short-term climate system anomalies, and improving the accuracy of extracting anomalous signals caused by climate oscillations.

[0021] 2. This invention extracts attention weights by inputting environmental factors into a long short-term memory network, maps factor differences using a nonlinear saturated kernel function, and combines the attention weights to obtain the contribution amplitude. Finally, it compares the direction of the contribution amplitude with the direction of leaf area index anomalies, quantifying the sensitivity of different environmental factors to abnormal fluctuations in leaf area index, characterizing the threshold effect of abnormal variations in environmental factors on mangrove growth, ensuring that the attribution results conform to the physical mechanism of ecological response, and improving the reliability of identifying dominant risk factors.

[0022] 3. This invention uses an S-shaped function risk mapping to process the absolute value of abnormal leaf area index and output the risk contribution and risk intensity. It then multiplies the risk contribution, the area of ​​the risk region, and the carbon density of the mangrove forest to calculate the carbon storage loss. This adapts to the nonlinear variation law of vegetation degradation under stress, transforms the abnormal leaf area index into a specific carbon storage loss value, realizes the intuitive quantification of the blue carbon ecological risk of mangrove forests, and improves the scientific nature of risk early warning assessment. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the overall method flow of the present invention;

[0024] Figure 2 This is a schematic diagram of the system architecture of the present invention;

[0025] Figure 3 This is a schematic diagram comparing the risk identification accuracy of the present invention;

[0026] Figure 4 This is a schematic diagram illustrating the differentiated LAI responses in different regions during the El Niño event of this invention. Detailed Implementation

[0027] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] See attached document Figure 1 This invention provides a mangrove blue carbon ecological risk identification and early warning system for climate change. The system includes a data processing module, a baseline determination module, an attribution analysis module, and a risk early warning module.

[0029] The data processing module is connected to the baseline determination module. The data processing module transmits the extracted residual sequence and the standardization results to the baseline determination module to provide underlying data support.

[0030] The baseline determination module is connected to the attribution analysis module. The baseline determination module transmits the mean and standard deviation statistics of the neutral period obtained by calculation to the attribution analysis module as the climatological benchmark reference for the factors.

[0031] The attribution analysis module is connected to the risk warning module. The attribution analysis module transmits the dominant factor type, dominant risk factor, and dominant factor contribution to the risk warning module for ecological risk quantification calculation.

[0032] See attached documentFigure 2 This invention provides a method for identifying and warning of blue carbon ecological risks in mangroves in response to climate change, comprising the following steps:

[0033] The data processing module acquires global land surface characteristics satellite leaf area index data (GLASS LAI), with a spatial resolution of 250 meters and consisting of 8-day composite data. It also acquires Modern Retrospective Climate Research and Applications Version 2 (MERRA-2) climate data, with a spatial resolution of 0.5 degrees x 0.625 degrees. The module acquires three-dimensional ocean observation sea surface temperature data (ARMOR3D), with a spatial resolution of 0.25 degrees. It acquires global ocean physical reanalysis sea surface salinity data (GLORYS12V1), with a spatial resolution of 0.125 degrees. The module also acquires satellite altimeter sea level data, with a spatial resolution of 0.25 degrees. Finally, it acquires standardized precipitation evapotranspiration index (SPEI) data, with a spatial resolution of 0.5 degrees. Finally, it acquires global mangrove monitoring distribution data (Global Mangrove Watch). The module performs data format conversion. The data processing module unifies the spatiotemporal alignment of data from different data sources to a spatial resolution of 0.5 degrees and a monthly temporal resolution. The module also uses a 5×5 mean filter to extrapolate near-shore missing values.

[0034] The data processing module labels the data quality level. It resamples the 8-day synthetic GLASS LAI to a monthly mean. It extracts unchanged, undisturbed mangrove areas from Global Mangrove Watch and GLASS LAI from 1996 to 2020. The module calculates the coverage rate by the ratio of mangrove pixels to total pixels. It applies a 75% coverage threshold to filter mixed and boundary pixels. It performs detrending processing on the original time-series data by fitting a long-term trend using linear regression. It calculates the monthly climatology. The module subtracts the monthly climatology from the original time-series data to extract the residual series, completing the deseasoning process. The module standardizes the residual series using standard scores.

[0035] The baseline determination module identifies El Niño events based on a multivariate El Niño-Southern Oscillation Index (ENSO index) greater than 0.5 for more than 3 months. It identifies La Niña events based on an ENSO index less than -0.5 for more than 3 months. It identifies the Positive Indian Ocean Dipole (pIOD) event based on a Dipole Model Index greater than or equal to 0.52 for more than 3 months. The baseline determination module outputs a climate oscillation event time window including start time, end time, intensity, and event type. It identifies neutral periods and excludes recovery periods based on an ENSO index range of -0.5 to 0.5 and a Dipole Model Index range of -0.52 to 0.52. The baseline determination module calculates the mean and standard deviation statistics for neutral periods.

[0036] The attribution analysis module constructs an input sequence matrix containing air temperature, standardized precipitation evapotranspiration, wind speed, sea surface temperature, sea surface salinity, and sea level anomalies. The module processes the input sequence matrix using a Long Short-Term Memory (LSTM) network hidden state update mechanism. It also employs forgetting gates, input gates, and output gates to further process the input sequence matrix. Finally, the module calculates attention weights. The attribution analysis module is based on attention weights. Obtain the model's output. The formula for the model's output is:

[0037] ;

[0038] in, This represents the predicted value of the global land surface characteristic satellite leaf area index data; This represents the weight matrix of the linear output layer; This represents a weighted summation of all factors, including air temperature, standardized precipitation evapotranspiration, wind speed, sea surface temperature, sea surface salinity, and sea level anomaly. The index variable representing the factor takes values ​​from 1 to 6; Indicates the first Hidden feature representation of each factor; This represents the learnable parameters.

[0039] The attribution analysis module fits the sea level anomaly time series using a multiharmonic regression model. It extracts the monthly node periodic contribution and climate oscillation residuals. The module calculates factor differences and applies a nonlinear saturated kernel function to calculate the contribution of each factor. The formula for the nonlinear saturated kernel function is:

[0040] ;

[0041] in, This represents the abnormal amplitude of the factor after transformation by a nonlinear saturated kernel function. Indicates the first The absolute value of the differences among the factors; This represents the saturation rate parameter, used to adjust the sensitivity of the difference magnitude to the final weight. The larger the weight, the stronger the amplification effect on small differences, and the more sensitive the weight is to changes in differences. The smaller the value, the smoother the response to differences, and the more gentle the changes in weights; This represents the hyperbolic tangent activation function.

[0042] The attribution analysis module performs a directional consistency assessment. It selects the dominant risk factor. It calculates the contribution percentage. The attribution analysis module outputs the dominant factor type, attention weight, contribution percentage, and dominant factor contribution.

[0043] The risk warning module calculates risk contribution using a sigmoid risk mapping function. The module then calculates risk intensity using this sigmoid risk mapping function. The formula for risk contribution is:

[0044] ;

[0045] in, Indicates the degree of risk contribution; This represents the dimensionless scaling constant for the risk mapping, used to normalize the output of the Sigmoid function; its default value is 0.45tC / m. 2 The value ranges from 0.25 to 0.75 tC / m. 2 ; This represents the risk response steepness parameter, which is a dimensionless constant with a default value of 3.0. This represents the risk trigger threshold, which is a dimensionless constant corresponding to the outlier value of the leaf area index after standardized calculation. The default value is 0.8. This represents the absolute value of the leaf area index anomaly after standardized calculation.

[0046] The risk warning module calculates carbon storage loss. It also calculates the impact on carbon sink capacity. The module is divided into the western Pacific, eastern Pacific, western Indian Ocean, and eastern Indian Ocean regions. It identifies spatial regions with anomalies in the east-west direction of global land surface feature satellite leaf area index data, with correlation coefficients less than -0.5. It also identifies spatial regions in the western and eastern Indian Oceans with anomalies in the west-east direction of global land surface feature satellite leaf area index data, with correlation coefficients less than -0.5. The module predicts future anomalies. It overlays the monthly nodal period background to predict sea level changes. Based on risk contribution, the module classifies risks into low, medium, high, and very high levels, generating warning assessments. The module outputs a spatial response pattern map, a global land surface feature satellite leaf area index anomaly prediction map, a risk warning level map, and a risk assessment report.

[0047] See attached document Figure 1 The mangrove blue carbon ecological risk identification and early warning method for climate change includes the following sub-steps when executing the multi-source spatiotemporal data processing module:

[0048] The data processing module acquires global land surface characteristic satellite leaf area index (LIA) data with a spatial resolution of 250 meters. This data is an 8-day composite. The module also acquires Modern Retrospective Climate Study and Applications Version 2 (MRC) climate data with a spatial resolution of 0.5 degrees x 0.625 degrees. Furthermore, the module acquires 3D ocean observation sea surface temperature (SST) data with a spatial resolution of 0.25 degrees. It also acquires global ocean physical reanalysis (OPE) sea surface salinity data with a spatial resolution of 0.125 degrees. Finally, the module acquires satellite altimeter sea level data with a spatial resolution of 0.25 degrees. The module also acquires standardized precipitation evapotranspiration index (SPEI) data with a spatial resolution of 0.5 degrees. Finally, the module acquires global mangrove monitoring distribution data.

[0049] Due to the objective differences in spatial resolution and temporal acquisition cycles between multi-source satellite data and climate reanalysis data, constructing a unified analytical platform requires a data processing module for data format conversion. For the specific data reading and file rewriting involved in data format conversion, those skilled in the art can use general scientific data parsing code. The execution of the data format conversion program is well-known in the field and will not be elaborated upon here.

[0050] The data processing module unifies the spatiotemporal alignment of data from different data sources to a spatial resolution of 0.5 degrees and a monthly temporal resolution. During spatial alignment, the module applies bilinear interpolation to adjust the spatial resolution. Addressing the technical challenge of missing climate data pixels in nearshore mangrove areas due to differences in land-sea cover, the module uses a 5x5 mean filter to extrapolate missing nearshore values. The module calculates the mean of valid pixels within a 5x5 grid and fills in the spatial locations of missing values. The module also labels the data quality level. Furthermore, the module resamples the 8-day synthetic global land surface feature satellite leaf area index data to a monthly mean. Based on the time range of the month, the module calculates the average and outputs monthly time-series data. Finally, the module outputs a multi-source dataset with a spatial resolution of 0.5 degrees, a monthly temporal resolution, and a complete time series.

[0051] The data processing module extracts undisturbed mangrove areas that remain unchanged from global mangrove monitoring distribution data and global land surface characteristic satellite leaf area index data from 1996 to 2020. The module compares spatial pixel attributes across the time span to extract mangrove pixels that remain constant. Finally, the module calculates the coverage rate by dividing the number of mangrove pixels by the total number of pixels. The formula for calculating the coverage rate is as follows:

[0052] ;

[0053] in, Indicates coverage rate; This indicates the number of mangrove pixels within a given pixel. This indicates the total number of pixels.

[0054] The data processing module uses a 75% coverage threshold for filtering. This threshold is determined based on the natural distribution characteristics of mangrove patches, aiming to filter out as much interference as possible from mixed pixels (water bodies, bare land, etc.) and mangrove boundary pixels on vegetation growth signals. The module removes pixels with a coverage of 75% or less. It also filters mixed pixels and boundary pixels. Finally, the module outputs a set of high-coverage mangrove pixels.

[0055] Because climate and marine environmental factors typically exhibit significant long-term interannual trends and periodic seasonal fluctuations, the data processing module performs detrending processing on the raw time-series data (0.5-degree grid) using linear regression to fit the long-term trend, in order to extract the true abrupt anomaly signals caused by climate oscillations. The raw time-series data includes global land surface characteristic satellite leaf area index data, temperature data, wind speed data, sea surface temperature data, sea surface salinity data, and sea level anomaly data. The linear regression fitting formula is as follows: ;

[0056] in, Representing continuous time The corresponding original time series data values, Indicates time, in months; Indicates the slope of the trend; This represents the intercept.

[0057] The data processing module calculates the detrended data after removing the long-term trend, using the following formula: ;

[0058] in, This indicates detrended data.

[0059] The data processing module calculates the monthly climatology. The calculation of the monthly climatology does not include standardized precipitation evapotranspiration index data. The formula for calculating the monthly climatology is as follows: ;

[0060] in, Indicates the first Monthly climate conditions for one month; Indicates the total number of years; Indicates the first Year Monthly detrended data values.

[0061] The data processing module extracts the residual series by subtracting the monthly climatological data from the detrended data, thus completing the deseasonalization process. The formula for calculating the residual series is as follows: ;

[0062] in, This represents the residual sequence. The data processing module outputs detrended and deseasonalized outlier time series.

[0063] Because environmental factors such as temperature, wind speed, and sea level vary greatly in size and absolute amplitude of fluctuation, the data processing module performs Z-score standardization on the residual series to unify all factors to the same statistical benchmark. Standardized precipitation evapotranspiration index data, since it already possesses standardization properties, does not participate in the aforementioned detrending and deseasonalization processes; instead, it is directly used as the residual series for Z-score standardization. The formula for Z-score standardization is as follows: ;

[0064] in, Represents the standardized value; This represents the numerical values ​​corresponding to the residual sequence; This represents the mean; It represents the standard deviation.

[0065] The formula for calculating the mean is as follows: ;

[0066] in, Represents the total number of samples in the time series; Indicates the first The residual sequence data values ​​at each time point. The formula for calculating the standard deviation is as follows:

[0067] ;

[0068] in, Represents the total number of samples in the time series; Indicates the first Numerical values ​​of residual sequence data at each time point; This represents the mean.

[0069] When the standard deviation is 0, the data processing module determines that the time series within the corresponding spatial grid lacks effective variation information and cannot calculate outlier fluctuations. In this case, the data processing module sets the standardization value to 0 and marks the corresponding grid as a non-variable state. The data processing module outputs a dimensionless standardized outlier time series with a mean of 0 and a standard deviation of 1.

[0070] See attached document Figure 1 The method for identifying and warning of blue carbon ecological risks in mangroves in response to climate change includes the following sub-steps when performing the baseline determination module:

[0071] To accurately capture the signals of large-scale climate oscillations that trigger anomalies in global climate models, the baseline determination module acquires the time series of the multivariate El Niño-Southern Oscillation Index, and the baseline determination module acquires the time series of the Dipole Model Index. For methods of obtaining publicly available meteorological index data, those skilled in the art can call the data interfaces of publicly available meteorological databases for data reading and local storage. Acquiring network data is a well-known technique in this field and will not be elaborated upon here.

[0072] The baseline determination module identifies climate oscillation events based on set numerical thresholds and time spans. Specifically, it determines El Niño events based on a multivariate ENSO index greater than 0.5 for more than 3 months, and La Niña events based on a multivariate ENSO index less than -0.5 for more than 3 months. Finally, it determines pIOD events based on a dipole mode index greater than or equal to 0.52 for more than 3 months.

[0073] The threshold values ​​of 0.5, -0.5, and 0.52 are set based on the international meteorological organization's general criteria, combined with a time span requirement of more than three months. The aim is to filter out short-term atmospheric noise and random fluctuations, ensuring that identified climate oscillation events have a definite and sustained physical state in ocean thermodynamics and atmospheric circulation that can substantially impact mangrove ecosystems. The baseline determination module outputs a climate oscillation event time window including start time, end time, intensity, and event type. The start time corresponds to the first calendar month in which the index continuously exceeds the threshold, and the end time corresponds to the previous calendar month in which the index falls back to within the threshold range.

[0074] To establish a background reference state unaffected by extreme weather events, the baseline determination module determines neutral periods based on index ranges. The module extracts time points based on the multivariate ENSO index range of -0.5 to 0.5 and the dipole model index range of -0.52 to 0.52, classifying time points conforming to the corresponding index ranges as neutral periods.

[0075] After determining the neutral period, the baseline determination module performs a recovery period exclusion process. Following a climate oscillation event, the physiological characteristics of the mangrove ecosystem and surrounding marine environmental parameters require a period of time to recover from the abnormal fluctuations. Directly including adjacent time periods after the event's end in the neutral period would result in the baseline data containing residual physical effects of the preceding event. The baseline determination module filters out residual effects by excluding the recovery period after the event, extracting the pure climate neutral period. Specifically, the baseline determination module sets a recovery period window of 3 to 6 months after the end of the climate oscillation event, and removes time points within this recovery period window from the neutral period statistics.

[0076] The baseline determination module calculates the mean and standard deviation statistics for the neutral period. Using the neutral period as the baseline provides an ecological and climatological benchmark under conditions free from strong climate oscillations, allowing subsequent processes to compare environmental factor fluctuations during climate anomalies with this benchmark. The statistic is calculated on the dimensionless, standardized outlier time series output by the data processing module. The formula for calculating the mean statistic for the neutral period is as follows:

[0077] ;

[0078] in, The mean statistic representing the neutral period; This indicates the total number of months in the neutral period after the recovery period exclusion process; Indicates the first [number]th ... Dimensionless standardized values ​​of each independent environmental factor variable or ecological indicator variable for each month.

[0079] The formula for calculating the standard deviation statistic during the neutral period is as follows:

[0080] ;

[0081] in, The standard deviation statistic represents the neutral period; This indicates the total number of months in the neutral period after the recovery period exclusion process; Indicates the first [number]th ... The baseline determination module stores the calculated mean and standard deviation statistics for each independent environmental factor variable or ecological indicator variable over the months and passes them to subsequent modules.

[0082] See attached document Figure 1 The mangrove blue carbon ecological risk identification and early warning method for climate change includes the following sub-steps when performing the attribution analysis module:

[0083] The attribution analysis module constructs an input sequence matrix containing air temperature, standardized precipitation evapotranspiration, wind speed, sea surface temperature, sea surface salinity, and sea level anomalies. The purpose of constructing the input sequence matrix is ​​to input the dynamic fluctuation characteristics of environmental factors over time into the nonlinear neural network model. The attribution analysis module segments the dimensionless standardized outlier time series output by the data processing module according to a set time sliding window length. The sliding window length is typically set to 12 to 36 months. This range is determined based on covering at least one complete mangrove growth year cycle and typical short-term climate oscillation cycles to ensure that the network model can learn the lagged and cumulative effects of marine and atmospheric environmental factors on the ecosystem. The attribution analysis module constructs independent one-dimensional time series input vectors for each environmental factor from the segmented air temperature time series, standardized precipitation evapotranspiration time series, wind speed time series, sea surface temperature time series, sea surface salinity time series, and sea level anomaly time series. The number of factors in the input sequence matrix is ​​fixed at 6.

[0084] The attribution analysis module processes the input sequence matrix through the hidden state update mechanism of the Long Short-Term Memory (LSTM) network. The LSTM network, with its gating mechanism for maintaining long-term memory, can handle nonlinear mappings with time lags, making it suitable for capturing the nonlinear delayed effects of complex climate systems on mangrove growth. The attribution analysis module processes the input sequence matrix through forgetting, input, and output gates. For the specific matrix multiplication and forward propagation calculations of the forgetting, input, and output gates in the LSTM network, those skilled in the art can utilize the built-in function libraries of open-source deep learning frameworks; the forward propagation calculation of the LSTM network is a well-known technique in this field and will not be elaborated upon here. The LSTM network processes the input sequence matrix to extract the temporal dependency features of each environmental factor over time. The attribution analysis module uses the LSTM network to perform parallel encoding of the independent one-dimensional time-series input vectors of each environmental factor, extracting the temporal dependency features of each environmental factor in the input sequence. The attribution analysis module extracts the output of the last time step of each parallel channel, using this as the hidden feature representation of each environmental factor. The set of hidden feature representations contains six independent feature vectors corresponding to air temperature, standardized precipitation evapotranspiration index, wind speed, sea surface temperature, sea surface salinity, and sea level anomaly, respectively.

[0085] The attribution analysis module calculates attention weights. To quantify the contribution of different environmental factors to the nonlinear sensitivity of global land surface characteristic satellite leaf area index (LAI) data anomalies, the attribution analysis module constructs a feature attention mechanism. Since the environmental factors dominating mangrove LAI anomalies dynamically change under different spatial locations and climatic backgrounds, the feature attention mechanism can dynamically identify the dominant risk factor at the current spatiotemporal node by adaptively learning the importance distribution of the hidden feature representations of each factor. The attribution analysis module uses fully connected layers and a hyperbolic tangent activation function to calculate the attention score of the hidden feature representations of each environmental factor. The formula for calculating the attention score is as follows:

[0086] ;

[0087] in, Indicates the first Attention scores for each factor; The transpose of the parameter vector representing the attention mechanism; The weight matrix representing the attention mechanism; Indicates the first Hidden feature representation of each factor; This represents the bias vector of the attention mechanism. , as well as All of these are learnable parameters that are continuously updated and iterated through the backpropagation algorithm during model training.

[0088] The attribution analysis module applies a normalized exponential function to perform probability normalization on the attention score, and outputs the attention weights corresponding to each environmental factor. The formula for calculating the attention weights is as follows:

[0089] ;

[0090] in, Indicates the first Attention weights for each factor; Indicates the first Attention scores for each factor; It represents the total number of factors including air temperature, standardized precipitation evapotranspiration index, wind speed, sea surface temperature, sea surface salinity, and sea level anomaly.

[0091] The attribution analysis module obtains the model output based on attention weights. It performs a weighted summation of the attention weights of each environmental factor and the corresponding hidden feature representations to obtain a context representation matrix that fuses global features. This context representation matrix is ​​then input into the linear output layer to calculate the model output. The model output is a predicted value for anomalous fluctuations in global land surface feature satellite leaf area index data. .

[0092] For the training process of this neural network model, those skilled in the art can use mean squared error as the loss function and apply a gradient-based optimizer to update the network weights. The training iteration of model parameters is a well-known technique in this field and will not be elaborated here. The attribution analysis module completes the nonlinear function mapping from multiple environmental factor sequences to global land surface characteristic satellite leaf area index data through the model output formula.

[0093] See attached document Figure 1 The mangrove blue carbon ecological risk identification and early warning method for climate change includes the following sub-steps when executing the multi-timescale sea level fluctuation separation module:

[0094] A multi-timescale sea-level fluctuation separation module acquires the sea-level anomaly time series. Based on physical and marine ecological principles, mangroves grow in the intertidal zone, and the frequency of seawater inundation they experience is influenced not only by random climate oscillations but also by long-period astronomical tides, such as the 18.61-year lunar node cycle and the 4.4-year perigee harmonics. The lunar node cycle alters the maximum tidal range in the region, thus periodically changing the salinity and anoxic state of the mangrove sediment. To accurately assess the actual contribution of short-term climate change to the blue carbon ecological risk of mangroves, it is necessary to separate the long-period astronomical tidal components from the total sea-level anomaly fluctuations. The multi-timescale sea-level fluctuation separation module constructs a multivariate harmonic regression model to fit the sea-level anomaly time series for each 0.5-degree spatial resolution grid cell. The multivariate harmonic regression model quantifies the fixed-period astronomical driving components by introducing trigonometric function terms. The regression model fitting formula for the sea-level anomaly time series is as follows:

[0095] ;

[0096] in, Indicates time Sea level anomaly at the location; This represents a time variable, with the unit being years. The starting point for this time variable is set to 2001. The time corresponds to January 2001. Since the input sea level anomaly time series has a monthly time resolution, the time... The step size is one-twelfth of a year; This represents the intercept term of the multivariate harmonic regression model; and This represents the amplitude coefficient of the 4.4-year perigee subharmonic, which represents the period of low-frequency sea level fluctuations caused by the lunar perigee precession. and The amplitude coefficient represents the 18.61-year lunar node cycle, which represents the long-term astronomical tidal cycle caused by the recession of the ascending node of the moon's orbit. This represents the residual term of the multivariate harmonic regression model.

[0097] The multi-timescale sea-level wave separation module estimates the amplitude coefficients in a multivariate harmonic regression model. The module applies the least squares algorithm to estimate the amplitude coefficients. Amplitude coefficient Amplitude coefficient and amplitude coefficient In implementing the lower-level characteristics of coefficient estimation, the multi-timescale sea-level wave separation module constructs an observation design matrix with trigonometric function periodic terms as independent variables. By minimizing the sum of squared residuals between observed sea-level anomalies and model-fitted values, the amplitude coefficients of the multivariate harmonic regression model are solved. For the specific matrix operations of solving the matrix equations using the least squares method and optimizing the sum of squared residuals, those skilled in the art can use the linear fitting functions in the numerical analysis library for programming calculations. The least squares method for solving linear regression parameters is a well-known technique in this field and will not be elaborated upon here.

[0098] The multi-timescale sea-level wave separation module extracts the lunar node periodic contribution component and the sea-level anomaly component caused by climate oscillations. The module substitutes the obtained amplitude coefficients into the astronomical tidal periodic function equation to calculate the lunar node periodic contribution determined by celestial motion. The formula for calculating the lunar node periodic contribution is as follows:

[0099] ;

[0100] in, Indicates time The monthly nodal periodic contribution value is used to calculate the sea level anomaly. The multi-timescale sea level fluctuation separation module subtracts the monthly nodal periodic contribution value from the acquired sea level anomaly values ​​to obtain the residual of the sea level anomaly time series. The multi-timescale sea level fluctuation separation module defines the residual of the sea level anomaly time series as the sea level anomaly caused by climate oscillations. The calculation formula for sea level anomalies caused by climate oscillations is as follows:

[0101] ;

[0102] In the formula, This indicates sea level anomalies caused by climate oscillations.

[0103] The multi-timescale sea level fluctuation separation module outputs the monthly nodal period contribution value, sea level anomaly values ​​caused by climate oscillations, and amplitude coefficients. Amplitude coefficient Amplitude coefficient Amplitude coefficient The multi-timescale sea-level fluctuation separation module, through the above calculations, has achieved a quantitative distinction between astronomical tidal cycles and climate oscillation driving factors, avoiding the misjudgment of long-term astronomical effects as short-term climate system anomalies. This provides a data foundation for subsequent assessment of the relative importance of multi-timescale factors to the ecological risk of mangrove blue carbon.

[0104] See attached document Figure 1 The mangrove blue carbon ecological risk identification and early warning method for climate change includes the following sub-steps when calculating the contribution of each factor:

[0105] The contribution calculation module for each factor acquires environmental factor data during the event period and the neutral period, and calculates the factor differences of each environmental factor. Based on the attention weights output from the training of a long short-term memory network and attention mechanism model, and combined with the mean environmental factors during the climate oscillation event period and the neutral period, the module quantifies the degree of physical environmental deviation caused by the climate oscillation event. The formulas for calculating the factor differences of each environmental factor are as follows:

[0106] ;

[0107] in, Indicates the first Factor differences among factors; Indicates the first The mean of each factor during a climate oscillation event; Indicates the first The mean of each factor during the neutral period. Specific environmental factors include air temperature, standardized precipitation evapotranspiration index, wind speed, sea surface temperature, sea surface salinity, and sea level anomaly. For the conventional algebraic statistics and calculation process of the mean of discrete time series data, those skilled in the art can directly call the basic mean calculation functions of conventional numerical analysis software. The calculation of the mean of discrete data samples is a well-known technique in this field and will not be elaborated upon here.

[0108] The contribution calculation modules for each factor employ a nonlinear saturated kernel function to perform a nonlinear mapping transformation on the differences between environmental factors. In mangrove ecology, the physical driving force of abnormal environmental factor variations on mangrove growth exhibits a significant threshold effect. When the abnormal amplitude of a certain environmental factor exceeds the physiological tolerance limit of the mangrove species, the ecological degradation response of the mangrove reaches its limit. At this point, further increases in the abnormal amplitude of the environmental factor will not lead to a proportional, infinitely decreasing leaf area index; this ecological characteristic is characterized by diminishing marginal effects. To accurately characterize the diminishing marginal effect feature in the numerical model, the contribution calculation modules for each factor introduce a nonlinear saturated kernel function based on the hyperbolic tangent function to calculate the abnormal amplitude of the factors. .

[0109] In the principles of mangrove ecology, different environmental factors exhibit varying sensitivities to limiting mangrove growth. Temperature has a narrow physiological tolerance range for its impact on mangrove growth; once temperature anomalies exceed the optimum range, the mangrove leaf area index (LAI) will rapidly decline. Therefore, the saturation rate parameter set for temperature is typically high, close to 3.0. For wind speed or sea level anomalies, the mangrove system possesses a wider buffering capacity, and its ecological degradation response is relatively mild. Therefore, the saturation rate parameter set for wind speed or sea level anomalies is typically low, close to 0.8. When no specific factor category is specified, the default value of the saturation rate parameter in the contribution calculation module is set to 1.5. Through the mapping transformation of the nonlinear saturation kernel function, the contribution calculation module corrects the calculation errors caused by directly extrapolating extreme climate anomalies in the traditional linear regression model.

[0110] The contribution calculation module for each factor combines the attention weights to calculate the contribution magnitude of each environmental factor and the final attribution contribution of each environmental factor to the satellite leaf area index anomaly of global land surface characteristics. The contribution calculation module multiplies the attention weights output by the deep learning network by the factor anomaly magnitude after nonlinear saturated kernel function transformation. The formula for calculating the contribution magnitude of each environmental factor is as follows: ;

[0111] in, Indicates the first The contribution magnitude of each factor; Indicates the first Attention weights for each factor; This represents the abnormal magnitude of the factor after transformation by a nonlinear saturated kernel function.

[0112] The contribution calculation module for each factor retains the positive or negative driving direction of environmental factor anomalies on mangrove growth, and introduces a sign function to restore the original positive or negative sign of factor differences to the attribution contribution calculation results. The calculation formulas for the attribution contribution of each environmental factor to leaf area index anomalies are as follows:

[0113] ;

[0114] in, Indicates the first The attribution contribution of each factor to the anomaly of satellite leaf area index of global land surface features; Represents a sign function, when The sign function outputs 1 when the value is greater than zero. When the value is zero, the sign function outputs 0; when... When the value is less than zero, the sign function outputs -1. Each factor contribution calculation module performs this mathematical operation on each grid cell in the data processing matrix, independently outputting the contribution of air temperature, standardized precipitation evapotranspiration index, wind speed, sea surface temperature, sea surface salinity, and sea level anomaly. The attribution contribution set output by each factor contribution calculation module provides a data foundation for the subsequent automatic identification of dominant risk factors and the quantitative classification of blue carbon ecological risks.

[0115] See attached document Figure 1 The mangrove blue carbon ecological risk identification and early warning method for climate change includes the following sub-steps when identifying the dominant ecological risk factors:

[0116] The dominant risk factor identification module acquires the attention weights output from the multi-factor nonlinear sensitivity analysis and the environmental factor attribution contributions calculated from the contribution levels of each factor, and extracts candidate dominant factors within the grid cells. The module sorts these environmental factors in descending order of their contribution magnitude. The contribution magnitude is obtained by multiplying the attention weights output from the long short-term memory network combined with the attention mechanism model by the factor anomaly magnitude after nonlinear saturation kernel function transformation. The contribution magnitude comprehensively reflects the model sensitivity of climate driving factors and the actual impact intensity of the current anomalous event.

[0117] The dominant risk factor identification module selects the environmental factors with the largest contribution amplitude as candidate dominant factors. The set of environmental factors includes six specific categories: air temperature, standardized precipitation evapotranspiration index, wind speed, sea surface temperature, sea surface salinity, and sea level anomaly. For the computer hardware and software implementation methods of one-dimensional array numerical sorting, those skilled in the art can directly call the quicksort or mergesort functions in the standard program library. The numerical sorting processing of data arrays is a well-known technology in this field and will not be elaborated further here.

[0118] The dominant risk factor identification module performs a direction consistency judgment and determines the dominant risk factor for each spatial grid cell based on the judgment result. In the physical mechanism of ecological response, the contribution direction of the physical driving factor must be consistent with the observed direction of ecosystem degradation for the attribution result to have physical meaning. For example, if satellite observations show a decrease in the leaf area index of mangroves in a certain area, but an abnormal change in a certain environmental factor ecologically promotes mangrove growth, then even if the contribution amplitude of the environmental factor is extremely large, it cannot be regarded as the dominant risk factor causing mangrove degradation. Therefore, the dominant risk factor identification module determines whether the sign of the attribution contribution value of the candidate dominant factor is the same as the sign of the observed abnormal leaf area index value, that is, whether the attribution contribution value and the abnormal leaf area index value are both positive or both negative. The dominant factor selection formula of the dominant risk factor identification module is as follows:

[0119] ;

[0120] The conditions and constraints of the dominant factor selection formula are: ;

[0121] in, Indicates the selected dominant risk factor; This indicates the environment variable index used to find the value of the objective function. Indicates the first Attention weights for each factor; This represents the abnormal amplitude of the factor after transformation by a nonlinear saturated kernel function. This represents a sign function that outputs 1 when the input value is greater than zero, 0 when the input value is equal to zero, and -1 when the input value is less than zero. Indicates the first The attribution contribution of each factor to the anomaly of satellite leaf area index of global land surface features; This indicates anomalies in the leaf area index observed in practice.

[0122] After traversing all environmental factors, if the attribution contribution direction of all environmental factors is inconsistent with the observed abnormal direction of leaf area index, the dominant risk factor identification module indicates from an ecological mechanism perspective that the abnormal mangrove status of the current grid unit is mainly caused by non-climate external disturbances, such as human reclamation and deforestation activities or sudden outbreaks of pests and diseases. In this case, the dominant risk factor identification module removes the constraint of directional consistency and directly selects factors that... The environmental factor with the largest product value is selected as the final dominant risk factor, and data markers indicating inconsistencies in direction are added to the dominant risk factors of the corresponding spatial grid cells. These inconsistent data markers are used in subsequent risk assessment reports to indicate uncertainties in climate attribution results, requiring the investigation of human activity interference in conjunction with high-resolution remote sensing imagery.

[0123] The dominant risk factor identification module calculates the contribution percentage of the dominant factor and outputs the risk identification results for spatial grid cells. This module quantifies the relative proportion of the dominant risk factor among all environmental factors. The formula for calculating the contribution percentage of the dominant factor is as follows:

[0124] ;

[0125] in, This indicates the percentage of contribution from the dominant factor, ranging from 0% to 100%. This indicates the attention weight of the selected dominant risk factor; This indicates the abnormal magnitude of the dominant risk factor after nonlinear saturated kernel function transformation; This indicates a summation operation on the product terms of all six environmental factors.

[0126] The dominant risk factor identification module independently performs the above calculations for each grid cell in the spatial matrix, ultimately outputting the dominant factor type, attention weight, contribution percentage, and attribution contribution of the dominant factor to leaf area index anomalies for each grid cell. The dominant factor-related parameters output by the module provide numerical basis for management departments to formulate targeted blue carbon ecological risk prevention and control measures.

[0127] See attached document Figure 1 The mangrove blue carbon ecological risk identification and early warning method for climate change includes the following sub-steps when performing blue carbon ecological risk identification and quantification:

[0128] The risk quantification module establishes a quantitative relationship between abnormal leaf area index (LAI) and blue carbon ecological risk. In mangrove ecosystems, an abnormal decline in LAI directly corresponds to a decrease in the ecosystem's carbon sequestration capacity and the risk of ecosystem degradation. In ecological principles, the response of mangroves to abnormal climate stress exhibits a nonlinear process of buffering, abrupt change, and limit saturation: when the magnitude of climate anomaly is small, mangroves rely on their own physiological regulatory mechanisms to maintain growth, and degradation is not obvious; when climate anomaly exceeds a certain tolerance limit, LAI declines rapidly, and ecological risk increases sharply; when damage reaches a certain level, the risk growth trend slows down due to the limited amount of degradable vegetation. Based on ecological physical characteristics, the risk quantification module introduces a nonlinear Sigmoid risk mapping function to calculate the risk contribution. .

[0129] The risk quantification module calculates the risk intensity and duration within a spatial grid cell. Risk intensity characterizes the proportion of damage to mangroves caused by the current climate oscillation event relative to extreme degradation recorded in the region's historical observations. The formula for calculating risk intensity in the risk quantification module is as follows: ;

[0130] in, Indicates the intensity of risk; This represents the absolute value of the leaf area index anomaly after standardized calculation. This represents the absolute value of the largest leaf area index anomaly occurring in the historical time series records for the same grid cell. Extracting the historical maximum value can be achieved using conventional extreme value retrieval algorithms, which are well-known techniques in the field and will not be elaborated upon here. The risk quantification module defines the risk duration as the number of months the identified climate oscillation event lasts.

[0131] The risk quantification module identifies high-risk areas. It assesses each grid cell in the data matrix. A grid cell is identified as a high-risk area when it simultaneously meets both the following conditions: a risk contribution greater than 0.35 and a leaf area index outlier less than -0.9. This dual assessment using risk contribution and leaf area index outliers effectively filters out minor numerical fluctuations caused by climatic background noise, confirming substantial ecological degradation in the mangroves.

[0132] The risk quantification module calculates the impact of blue carbon ecological risks on mangrove carbon storage and carbon sink capacity. The formula for calculating carbon storage loss using the risk quantification module is as follows:

[0133] ;

[0134] in, This represents an estimated value for carbon storage losses; Indicates the degree of risk contribution; This indicates the area of ​​the risk zone, expressed in square kilometers. This represents the carbon density of mangroves, expressed in tC / km². 2 The carbon density of mangroves ranges from 450 tC / km². 2 Up to 1800tC / km 2 Between these values, the default value is 950 tC / km. 2 The specific value of mangrove carbon density is determined by the risk quantification module based on the mangrove community morphology within the grid. For tall tree-type mangrove communities, the carbon density is set at 1200 tC / km². 2 Up to 1800tC / km 2 Between; for low-growing shrub-type mangrove communities, the carbon density is taken as 450 tC / km. 2 Up to 800tC / km 2 between.

[0135] The formula for calculating the decline in carbon sequestration capacity using the risk quantification module is as follows:

[0136] ;

[0137] in, Indicates the rate of decline in carbon sequestration capacity; The conversion factor between leaf area index and carbon sequestration capacity, expressed in tC·m³. -2 ·a -1 The conversion factor ranges from 0.18 tC·m. -2 ·a -1 Up to 0.55tC·m -2 ·a -1 Between these values, the default value is 0.32 tC·m.-2 ·a -1 The specific values ​​of the conversion coefficients are matched to the baseline of the historical net primary productivity observations of the mangroves in that grid cell.

[0138] The risk quantification module outputs blue carbon ecosystem risk assessment indicators. The data output by this module includes a blue carbon ecosystem risk distribution map, risk contribution, risk intensity, high-risk area identification results, and a blue carbon ecosystem impact assessment. The blue carbon ecosystem impact assessment includes estimated carbon storage loss and changes in carbon sink capacity. The quantitative indicators output by the risk quantification module are used to construct a risk classification system for blue carbon ecosystems in different regions.

[0139] See attached document Figure 1 The mangrove blue carbon ecological risk identification and early warning method for climate change includes the following sub-steps when performing trans-ocean basin spatial response model analysis and generating forward-looking risk warnings:

[0140] The spatial response pattern analysis module performs geographical region division and spatial correlation analysis. This module divides the global mangrove research area into four core geographical regions: the Western Pacific region (longitude range 120...). o E to 180 o (between E and 130°), the eastern Pacific region (longitude range within 130°). o W to 70 o (between W and 25) and the western Indian Ocean region (longitude range between 25) o E to 77 o (between E and E) and the eastern Indian Ocean region (longitude range between 77° and E) o E to 130 o (between E). The geographical regions correspond to the east and west banks of ocean basins, used to capture the dipole response characteristics triggered by large-scale climate oscillations at different geographical locations. The spatial response pattern analysis module calculates the spatial correlation coefficients of leaf area index anomaly sequences between different regions. For the data processing of calculating Pearson or Spearman correlation coefficients between spatial matrix sequences, those skilled in the art can directly call the correlation calculation functions in the statistical analysis library. The calculation of spatial correlation coefficients is a well-known technique in this field and will not be elaborated here.

[0141] The spatial response pattern analysis module identifies differentiated response patterns between regions. Climate oscillation events typically trigger climate anomalies with opposite physical directions on the east and west sides of ocean basins. In meteorological and oceanographic principles, climate oscillations (such as El Niño-Southern Oscillation or the Indian Ocean Dipole) cause large-scale spatial displacements in the Walker circulation, leading to a seesaw effect between sea surface heat distribution and wind field stress. This results in opposite characteristics in sea level height, precipitation distribution, and temperature changes on the east and west sides of the ocean basin. This reverse environmental stress is ultimately reflected in the opposite trends in mangrove leaf area index.

[0142] The spatial response pattern analysis module performs logical judgments on various values ​​within the divided regions. When the module determines that the leaf area index anomalies in corresponding east-west regions (or the western Indian Ocean region and the eastern Indian Ocean region) are in opposite directions, and the spatial correlation coefficient between corresponding east-west regions is less than -0.5, the module confirms the existence of differentiated response patterns between these regions. A correlation coefficient less than -0.5 serves as a threshold, statistically representing a moderate to strong negative correlation between spatial regions. Using this threshold effectively eliminates random climate fluctuation noise.

[0143] The spatial early warning module predicts future leaf area index anomalies and sea-level physical changes. Based on currently acquired climate indices and mined historical response patterns, the module extrapolates and predicts anomalous states of environmental factors. In its specific implementation, the module acquires the current El Niño multivariate index or dipole model index, performs numerical calculations using established lag regression equations, or employs historical similar climate event matching methods to extrapolate predicted values ​​for environmental factor anomalies at future time steps. In extreme climate oscillation events, the short-term sudden risks to mangrove ecosystems are primarily determined by a single dominant stressor. To improve the computational efficiency of forward-looking early warning and highlight the main physical driving force, the spatial early warning module uses the nonlinear response component of the dominant factor to approximate the overall ecosystem's changing trend. The formula for calculating and predicting leaf area index anomalies by the spatial early warning module is as follows:

[0144] ;

[0145] In the formula, This indicates an anomaly in the predicted leaf area index; This represents the attention weight of the dominant factor invoked by the spatial early warning module. The attention weight of the dominant factor is output by a long short-term memory network combined with an attention mechanism model. This represents a sign function that outputs 1 when the input value is greater than zero, 0 when the input value is equal to zero, and -1 when the input value is less than zero. This represents the abnormal values ​​of environmental factors predicted based on current climate indices and historical models. This represents the magnitude of predicted environmental factor anomalies after transformation by a nonlinear saturated kernel function. The nonlinear saturated kernel function is used to characterize the ecological feature of diminishing marginal effects after environmental anomalies exceed a threshold.

[0146] The space-based early warning module combines previously obtained numerical values ​​of the lunar noon cycle to predict future sea level changes. The formula used by the space-based early warning module to calculate and predict sea level is as follows:

[0147] ;

[0148] in, Indicates time The predicted sea level anomaly; Indicates time The sea level anomaly component contributed by the monthly nodal period, which includes a perigee subharmonic period of 4.4 years and a main period of 18.61 years; Indicates time Sea level anomaly components driven by climate oscillations.

[0149] The spatial early warning module generates a forward-looking blue carbon ecological risk early warning and assessment report. Based on the predicted risk contribution value, the module classifies risk areas into early warning levels. The module establishes four clear risk early warning level determination criteria: when the risk contribution is determined to be less than 0.25, a low-risk early warning level is output; when the risk contribution is determined to be greater than or equal to 0.25 and less than 0.55, a medium-risk early warning level is output; when the risk contribution is determined to be greater than or equal to 0.55 and less than 0.85, a high-risk early warning level is output; and when the risk contribution is determined to be greater than or equal to 0.85, an extremely high-risk early warning level is output.

[0150] The observable decline in leaf area index (LAI) caused by abnormal ocean climate events in nearshore mangrove ecosystems exhibits a physical time lag. The spatial early warning module sets the warning window at 4 to 7 months. The length of the warning window is determined by the time required for Kelvin waves from the center of the ocean to reach the continental margin coastline, superimposed with the cumulative physiological response time of mangrove vegetation from experiencing salinity or temperature stress to exhibiting massive leaf drop. Through lag correlation analysis of historical climate oscillation events and LAI change sequences, the spatial early warning module has identified 4 to 7 months as the optimal risk prediction window.

[0151] The spatial early warning module generates a risk assessment report based on historical response patterns, current climate indices, and calculated forecast results. The data entities output by the spatial early warning module to external systems include a spatial response pattern map, a predicted leaf area index anomaly distribution map, a blue carbon ecological risk early warning level map, the risk assessment report, and regional statistical data. The risk assessment report records the coordinates of threatened mangrove risk areas, the corresponding risk level, the warning time window length, and warning recommendations. The regional statistical data includes the area percentage of clearly risky areas and the spatial distribution statistics of risk directions across different ocean basins.

[0152] To verify the effectiveness and advancement of the proposed mangrove blue carbon ecological risk identification and early warning method for climate change in practical and complex ecological assessment projects, this embodiment takes the ecological response of typical global mangrove distribution areas (western Pacific, eastern Pacific, western Indian Ocean, and eastern Indian Ocean) under extreme climate events as an example, and describes the specific implementation process and beneficial effects of the invention in detail with reference to the accompanying drawings.

[0153] Comparative test of the accuracy of risk identification using multiple methods (combined) Figure 3 ):

[0154] To objectively evaluate the overall performance and early warning reliability of the algorithm of this invention, this embodiment sets up three sets of comparative experiments: the traditional scheme, the simplified scheme, and the method of this invention. The risk identification accuracy test results of each scheme are as follows: Figure 3 As shown:

[0155] Traditional solution:

[0156] This approach relies solely on the linear relationship of a single factor (such as a single meteorological indicator) for evaluation. Limited by the model's insufficient ability to learn time-series characteristics and complex nonlinear relationships, its risk identification accuracy is low, at only 55%, failing to reach the 85% target value required for business scenarios.

[0157] Simplified solution:

[0158] This scheme employs a 3-factor linear model, comprehensively considering multiple environmental input variables. Although its accuracy is improved compared to traditional schemes, reaching 72%, it still fails to break through the warning target line when facing the highly dynamic and long-term memory characteristics of mangrove growth due to the lack of adaptive allocation of time-series weights.

[0159] Method of the present invention:

[0160] Based on the constructed LSTM-Attention network, it not only effectively mines the time-lag correlation between long-term meteorological data and leaf area index (LAI), but also accurately focuses on key disaster-causing factors through the attention mechanism. Its risk identification accuracy reaches 88%, exceeding the target baseline of 85%, and is 33% higher than traditional solutions, demonstrating that the present invention has extremely high accuracy and application value in complex and multivariate ecological early warning tasks.

[0161] Verification of spatial response patterns of mangroves under extreme climate events (combined with) Figure 4 ):

[0162] The model constructed in this invention was applied to core mangrove growing areas globally to identify and verify spatial patterns of leaf area index (LAI) anomalies under typical historical extreme climate events. For example... Figure 4 As shown:

[0163] The figure illustrates the response of the standard outlier (Z-score) of mangrove LAI in four ocean regions under the forcing of three typical extreme climate events (2015-2016 El Niño, 2010-2012 La Niña, and 2019 Positive Indian Ocean Dipole).

[0164] El Niño and La Niña seesaw simulation effect:

[0165] Driven by the 2015-2016 El Niño event, the model of this invention identified a seesaw-like negative correlation effect in the growth of mangroves on both sides of the Pacific Ocean: the western Pacific region showed an extremely negative LAI (high risk of degradation, reaching -0.75), while the eastern Pacific region showed a significant positive anomaly (0.61) during the same period. Conversely, during the 2010-2012 La Niña event, the model also perfectly captured the opposite spatial distribution pattern, with the western Pacific turning to a positive anomaly (0.65) and the eastern Pacific turning to a negative anomaly (-0.47), showing a high degree of consistency in spatial teleconnection identification.

[0166] Simulation results of the Indian Ocean Dipole space response:

[0167] During the 2019 Positive Indian Ocean Dipole (pIOD) event, the model of this invention effectively resolved the heterogeneous ecological responses on the east and west coasts of the Indian Ocean: the western Indian Ocean region showed a significant positive anomaly (0.43), while the eastern Indian Ocean region showed a severe negative anomaly (-0.56), which is consistent with the physical mechanism of precipitation and sea surface temperature anomalies in this climate model.

[0168] The results demonstrate that the early warning model obtained using this invention not only has high single-point prediction accuracy when describing large-scale, complex physical climate and ecological response processes, but also possesses a strong generalization ability and realism in capturing macroscopic seesaw teleconnection effects.

Claims

1. A method for identifying and early warning of blue carbon ecological risks in mangroves in response to climate change, characterized in that, Includes the following steps: We acquired global land surface characteristic satellite leaf area index data and environmental factors time series, performed detrending, deseasonalizing and standard score standardization processing, and extracted anomalous sequences containing leaf area index anomalies. The climate oscillation window and neutral period are determined based on the multivariate El Niño-Southern Oscillation Index and the Dipole Model Index, and the mean value of the environmental factors during the neutral period is calculated as a benchmark reference. The environmental factors are input into a long short-term memory network to extract attention weights. The difference between the mean of the environmental factors in the climate oscillation window and the factor of the benchmark is calculated. The factor difference is mapped using a nonlinear saturated kernel function and combined with the attention weights to obtain the contribution amplitude. The direction of the contribution amplitude is compared with the direction of the leaf area index anomaly to select the dominant risk factor. The absolute value of the leaf area index anomaly is processed by S-shaped function risk mapping to output the risk contribution and risk intensity. Based on the dominant risk factor, future leaf area index anomalies are predicted and early warning assessments are generated.

2. The method for identifying and early warning of blue carbon ecological risks in mangroves in response to climate change, as described in claim 1, is characterized in that... The extraction of anomalous sequences containing leaf area index anomalies specifically includes: The coverage rate is obtained by calculating the ratio of mangrove pixels to the total number of pixels, and a preset coverage threshold is applied to filter mixed pixels and boundary pixels. The global land surface characteristic satellite leaf area index data are subjected to linear regression to fit the long-term trend and extract detrended data. The monthly climatological data is then subtracted to extract the residual sequence to complete the deseasonalization process. The residual sequence is standardized using the mean and standard deviation. When the standard deviation is zero, the standardized value is set to zero, and the outlier sequence is output. The coverage threshold is set based on the natural distribution characteristics of mangrove patches.

3. The method for identifying and early warning of blue carbon ecological risks in mangroves in response to climate change, as described in claim 1, is characterized in that... The specific methods for determining climate oscillation windows and neutral periods based on the multivariate El Niño-Southern Oscillation Index and the Dipole Model Index include: By combining the multivariate El Niño-Southern Oscillation Index, the dipole model index, and the duration span condition, El Niño events, La Niña events, and positive Indian Ocean Dipole events are determined, and the climate oscillation window containing the start and end times is output. The time nodes extracted based on the preset index range are divided into the neutral period, and the consecutive months after the end of the climate oscillation window are set as the recovery period. The time nodes within the recovery period are removed from the neutral period.

4. The method for identifying and early warning of blue carbon ecological risks in mangroves in response to climate change, as described in claim 1, is characterized in that... The process of inputting the environmental factors into the long short-term memory network to extract attention weights specifically includes: The environmental factors are constructed as an input matrix. The input matrix is ​​processed through the forget gate, input gate, and output gate mechanism of the long short-term memory network. The output result of the last time step of the parallel channel is extracted as the hidden feature representation. The attention score of the hidden feature representation is calculated by applying a fully connected layer and a hyperbolic tangent activation function, and the attention weight is extracted by applying a normalized exponential function to the attention score for probability normalization.

5. The method for identifying and early warning of blue carbon ecological risks in mangroves in response to climate change, as described in claim 1, is characterized in that... The time series of environmental factors includes sea level anomaly time series; the acquisition of global land surface characteristic satellite leaf area index data and environmental factor time series also includes separating the monthly node periodic contribution of the sea level anomaly time series: A multivariate harmonic regression model was constructed by introducing trigonometric function terms that include perigee subharmonics and the lunar node period, and this model was used to fit the sea level anomaly time series. The amplitude coefficients in the multivariate harmonic regression model are estimated using the least squares algorithm, and the contribution of the lunar node period determined by celestial motion is calculated by combining the trigonometric function terms. Subtract the monthly node periodic contribution from the sea level anomaly time series to separate the sea level anomalies driven by climate oscillations, and use the sea level anomalies driven by climate oscillations as the time series of the obtained environmental factors.

6. The method for identifying and early warning of blue carbon ecological risks in mangroves in response to climate change, as described in claim 1, is characterized in that... The specific steps of using a nonlinear saturated kernel function to map the factor differences and combining them with the attention weights to derive the contribution magnitude include: The nonlinear saturation kernel function, which includes a saturation rate parameter and a hyperbolic tangent activation function, is used to map and transform the absolute value of the factor difference to calculate the factor anomaly magnitude. The contribution magnitude of each environmental factor is calculated by multiplying the attention weight by the factor anomaly magnitude. A sign function is introduced to assign the original positive or negative sign of the factor differences to the multiplication result, and the attribution contribution of the environmental factors is calculated.

7. The method for identifying and early warning of blue carbon ecological risks in mangroves in response to climate change, as described in claim 1, is characterized in that... The selection of the dominant risk factor by comparing the direction of the contribution magnitude with the direction of the leaf area index anomaly specifically includes: The environmental factor with the largest contribution amplitude is selected as the candidate factor; When the attribution contribution sign of the candidate factor and the leaf area index anomaly sign are both positive or both negative, the candidate factor is selected as the dominant risk factor. When the attribution contribution values ​​of all the environmental factors are inconsistent with the abnormal leaf area index values, the environmental factor with the largest attribution contribution product is selected as the dominant risk factor and marked with a direction inconsistency.

8. The method for identifying and early warning of blue carbon ecological risks in mangroves in response to climate change, as described in claim 1, is characterized in that... The specific steps for processing the absolute value of the leaf area index anomalies and outputting the risk contribution and risk intensity using S-shaped function risk mapping include: By combining the dimensionless scaling constant, the risk response steepness parameter, and the risk triggering threshold, the abnormal absolute value of the leaf area index is processed through the S-shaped function risk mapping to obtain the risk contribution. The risk intensity is obtained by calculating the ratio of the abnormal absolute value of the leaf area index to the historical maximum abnormal absolute value of the leaf area index. The carbon storage loss is calculated by multiplying the risk contribution, the area of ​​the risk zone, and the carbon density of the mangroves, and the risk intensity is quantified by the carbon storage loss. The risk trigger threshold is set based on the abnormal value of the leaf area index after standardized calculation.

9. The method for identifying and early warning of blue carbon ecological risks in mangroves in response to climate change, as described in claim 8, is characterized in that... The specific steps for predicting future leaf area index anomalies and generating early warning assessments based on the aforementioned dominant risk factors include: Based on the currently acquired climate indices and historical response patterns, the predicted values ​​of environmental factors are extrapolated, and the predicted future leaf area index anomalies are calculated using the attention weight, sign function, and nonlinear saturated kernel function of the dominant risk factors. The predicted future leaf area index anomaly, the risk intensity, and the carbon storage loss are combined to classify the risk warning level, and the determined risk warning level is used as the output to generate the warning assessment.

10. A mangrove blue carbon ecological risk identification and early warning system for climate change, characterized in that, The method for identifying and warning of blue carbon ecological risks in mangroves in response to climate change, as described in any one of claims 1-9, includes: The data processing module is used to acquire global land surface characteristic satellite leaf area index data and environmental factors time series, perform detrending, deseasonalization and standard score standardization processing, and extract anomalous sequences containing leaf area index anomalies. The baseline determination module is used to determine the climate oscillation window and neutral period based on the multivariate El Niño-Southern Oscillation Index and the Dipole Model Index, and to calculate the mean value of the environmental factors during the neutral period as a benchmark reference. The attribution analysis module is used to input the environmental factors into a long short-term memory network to extract attention weights; calculate the difference between the mean of the environmental factors in the climate oscillation window and the factor of the benchmark reference; map the factor difference using a nonlinear saturated kernel function and combine it with the attention weights to obtain the contribution amplitude; and compare the direction of the contribution amplitude with the direction of the leaf area index anomaly to select the dominant risk factor. The risk warning module is used to process the absolute value of the leaf area index anomaly through S-shaped function risk mapping, output the risk contribution and risk intensity, predict future leaf area index anomalies based on the dominant risk factor, and generate a warning assessment.

Images