System and Method for Extracting Nonlinear Trends of Carbon Sinks Based on Ensemble Empirical Mode Decomposition

Abstract

The present invention relates to the technical field of carbon sink trend extraction, and specifically relates to an extraction method performed by a carbon sink non-linear trend extraction system based on ensemble empirical mode decomposition, including the following steps: S1: Collect forest carbon sink time series data to form a forest carbon sink time series data set. The forest carbon sink time series data is the vegetation coverage rate VCR, soil coverage rate SCR, vegetation carbon content C at different seasonal times veg , soil carbon content C soil , total carbon storage C total . By monitoring the carbon storage in vegetation and soil, the present invention can more accurately evaluate the overall carbon sink capacity of forests. And considering that the carbon storage in vegetation and soil may fluctuate with seasonal changes, by monitoring these changes, it is possible to better understand how carbon sinks are affected by climate factors. At the same time, during the process of monitoring the changes in carbon storage in different seasons, and according to the law of the non-linear changes in the carbon storage in vegetation and soil affecting the forest carbon sink capacity, a long-term change trend map of the forest carbon sink is extracted.

Description

Technical Field

[0001] The present invention relates to the technical field of carbon sink trend extraction, and particularly to a carbon sink non-linear trend extraction system and method based on ensemble empirical mode decomposition. Background Art

[0002] Carbon sink refers to the absorption and storage of carbon dioxide in the atmosphere by natural ecosystems such as forests, oceans, and soils through biochemical processes such as photosynthesis. It plays a key role in the global carbon cycle and has a positive effect on inhibiting climate warming. However, the efficiency of carbon sinks is affected by various factors such as climate change, land use, and human activities. Therefore, the monitoring and analysis of its dynamic changes are crucial.

[0003] After retrieval, a forest carbon storage and carbon sink value monitoring system and dynamic evaluation method are disclosed in the Chinese patent document with the patent number CN202311214927.0. The above solution divides the forest area to be monitored according to vegetation type, soil type, and climate conditions, reducing the error in carbon storage assessment between different types of forest communities; and on the basis of remote sensing image data inversion, plot surveys and information collection are introduced to collect the vegetation conditions at different levels in the plots and the environmental data of the plots, making up for the defect that it is difficult to fully reflect the forest structure during remote sensing image data inversion, and considering environmental factors, the output results of the evaluation model that may have errors and fluctuations are corrected, improving the accuracy of carbon storage and carbon sink value assessment.

[0004] However, during the process of forest carbon sink monitoring, the carbon storage in the forest is in a static state, which describes the amount of carbon that has been accumulated in the forest carbon sink system at a certain moment. The carbon fixed by plants is not only retained in the plants but may also be transferred to the soil and become part of the soil organic carbon. With the change of seasons, it will affect the growth of plants and the state of the surrounding soil, resulting in non-linear changes and mutual correlations in vegetation carbon storage and soil carbon storage at different time scales. The existing evaluation systems cannot extract the seasonal change trend of forest carbon sinks according to the non-linear changes between vegetation carbon storage and soil carbon storage. Summary of the Invention

[0005] Technical Problems to be Solved

[0006] Aiming at the above-mentioned shortcomings of the existing technology, the present invention provides a carbon sink non-linear trend extraction system and method based on ensemble empirical mode decomposition, which can solve the problem that the existing forest carbon sink monitoring system cannot extract the carbon sink change trend line graph of the forest through the seasonal changes of vegetation carbon storage and soil carbon storage during the monitoring process.

[0007] Technical Solutions

[0008] To achieve the above object, the present invention is realized through the following technical solutions:

[0009] The present invention provides an extraction method carried out by a carbon sink non-linear trend extraction system based on ensemble empirical mode decomposition, including the following steps:

[0010] S1: Collect forest carbon sink time series data to form a forest carbon sink time series data set. The forest carbon sink time series data is the vegetation coverage rate VCR, soil coverage rate SCR, vegetation carbon content C veg , soil carbon content C soil , total carbon storage C total ;

[0011] S2: Preprocess the forest carbon sink time series data set based on the forest carbon storage theory to obtain the forest carbon sink C Carbon Sinkl (t) at different seasonal times;

[0012] S3: Use the ensemble empirical mode decomposition model to extract the data set X Carbon Sinkl (t) and use the ANN model to process the X features (t) to extract the non-linear trend of the forest carbon sink. Among them, the ensemble empirical mode decomposition model uses ensemble empirical mode decomposition EEMD to process C features (t), extracts the intrinsic mode functions IMFs in C Carbon Sinkl (t), and extracts the feature X Carbon Sinkl (t) through the IMFs. features (t).

[0013] Furthermore, in the S1, the data acquisition module collects the forest carbon sink time series data through remote sensing monitoring technology and ground measurement technology.

[0014] Furthermore, in the S2, the calculation process of C Carbon Sinkl (t) is as follows:

[0015] S201: Based on the forest carbon storage theory, obtain the total carbon storage C total in the static time, specifically:

[0016] C total =(α*VCR*C veg )+(β*SCR*C soil )+[λ*VCR*SCR*(C veg +C soil )]

[0017] Where:

[0018] α*VCR*Cveg Indicates the contribution of vegetation coverage rate to vegetation carbon storage; β*SCR*C soil Indicates the contribution of soil coverage rate to soil carbon storage; λ*VCR*SCR*(C veg +C soil ) represents the additional impact of the interaction between vegetation and soil coverage rates on total carbon storage; α is the influence coefficient of vegetation coverage rate on vegetation carbon storage; β is the influence coefficient of soil coverage rate on soil carbon storage; γ is the influence coefficient of the interaction between vegetation and soil on total carbon storage; α, β, and γ are determined through non-linear regression analysis;

[0019] S202: According to the total carbon storage C total , calculate the forest carbon sink C Carbon Sinkl (t) at different times during the seasonal change process, specifically:

[0020] C Carbon Sink (t) = C total (t) + ∫[I(t) - O(t)]dt

[0021] where I(t) is the carbon input rate varying with time, O(t) is the carbon output rate varying with time, and ∫[I(t) - O(t)]dt represents the cumulative integral of carbon input minus carbon output from the initial time to the t-th moment, that is, the total carbon increase or decrease.

[0022] Furthermore, in the step S3, use the ensemble empirical mode decomposition model to extract the C Carbon Sinkl (t)

[0023] including the following steps:

[0024] S301: Add white noise to C Carbon Sinkl (t):

[0025] C Carbon Sinkl(i) (t) = C Carbon Sinkl (t) + αn i (t)

[0026] S302: Perform empirical mode decomposition EMD on each signal C Carbon Sinkl(i) (t) with added white noise;

[0027] S303: Extraction of intrinsic mode functions:

[0028]

[0029] where IMF ij (t) indicates that this is the j-th mode function decomposed after the i-th addition of white noise;

[0030] S304: Xfeatures (t) extraction:

[0031] X features (t) = [mean(IMF ij (t)), std(IMF ij (t)), skewness(IMF ij (t)), kurtosis(IMF ij (t)), frequency(IMF ij (t))]

[0032] Wherein,

[0033]

[0034] n is the number of data points, and is the value of the i-th data point.

[0035] Furthermore, in the step S3, the ANN model is used to process the X features (t) to extract the non-linear trend of forest

[0036] carbon sink, specifically:

[0037] Construct an ANN model based on X features (t):

[0038] h t = σ(W * X features (t) + b)

[0039] h t is the output of the hidden layer, W is the weight matrix, b is the bias, and σ is the activation function;

[0040] The training formula of the ANN model is:

[0041]

[0042] ΔW is the weight update, η is the learning rate, L is the loss function, is the partial derivative of the weight with respect to the loss function;

[0043] The evaluation formula of the ANN model is:

[0044]

[0045] MSE is the mean square error, is the actual value, y i is the actual value, is the predicted value;

[0046] The parameter tuning formula is:

[0047]

[0048] Δη is the learning rate update, and α is the learning rate adjustment coefficient. is the partial derivative of the loss function with respect to the learning rate;

[0049] y predicted = h T · softmax(W O · h T + b O )

[0050] h T is a vector representing the output of the last (hidden) layer of the neural network;

[0051] W O is the output layer weight matrix that connects the hidden layer and the output layer;

[0052] b O is the bias vector of the output layer;

[0053] softmax is an activation function that converts the output of the network into a probability distribution, where the value of each element represents the probability of the corresponding class;

[0054] y predicted is the final non - linear trend extraction result.

[0055] A carbon sink non - linear trend extraction system based on ensemble empirical mode decomposition, including a data acquisition module and a data processing module;

[0056] The data acquisition module is used to collect forest carbon sink time - series data to form a forest carbon sink time - series data set, and the forest carbon sink time - series data set includes the vegetation coverage rate VCR, soil coverage rate SCR, vegetation carbon content C veg , soil carbon content C soil , total carbon storage C total ;

[0057] The data processing module is used to pre - process the forest carbon sink time - series data set to obtain the forest carbon sink C Carbon Sinkl (t) at different seasonal times and extract the data set X features (t) from the forest carbon sink time - series data set according to a pre - constructed ensemble empirical mode decomposition model, and process the X features (t) according to a pre - constructed ANN model to extract the non - linear trend of the forest carbon sink.

[0058] Beneficial effects

[0059] The technical solution provided by the present invention has the following beneficial effects compared with the known prior art:

[0060] 1. By monitoring the carbon storage in vegetation and soil, the overall carbon sink capacity of forests can be more accurately evaluated. Given that the carbon storage in vegetation and soil may fluctuate with seasons, monitoring these changes can help better understand how carbon sinks are affected by climate factors such as temperature, precipitation, and sunlight. Meanwhile, during the process of monitoring the changes in carbon storage in different seasons, and based on the law of the impact of the non-linear changes in the carbon storage of vegetation and soil on the forest carbon sink capacity, a long-term change trend graph of forest carbon sinks can be extracted.

[0061] 2. This solution extracts the forest carbon sink trend through EEMD, which can handle the non-linear and non-stationary data of forest carbon sinks, a task difficult to achieve by traditional extraction methods. This makes EEMD more effective in processing data with complex fluctuation patterns such as carbon sinks. Moreover, EEMD can adaptively decompose data into multiple IMFs, with each IMF representing a different fluctuation scale, which helps extract the multi-scale characteristics of carbon sink data and provides more details for in-depth analysis.

[0062] 3. During the process of extracting the forest carbon sink trend by EEMD in this solution, an ANN model is added for optimization. ANN can utilize the multi-scale characteristics decomposed by EEMD, comprehensively consider the carbon sink changes at different time scales, improve the comprehensiveness of prediction, and reduce the risk of overfitting in EEMD, thereby enhancing the generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0064] Figure 1 It is a schematic diagram of the working process of the extraction system in the embodiments of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0065] The present invention will be further described in detail below with reference to the drawings and embodiments. It can be understood that the specific embodiments described herein are only for explaining the present invention and not for limiting the present invention. Additionally, it should be noted that for the sake of description, only parts related to the present invention are shown in the drawings rather than all structures.

[0066] The present invention will be further described below with reference to the embodiments.

[0067] Embodiment:

[0068] Please refer to the attached Figure 1, this solution proposes a carbon sink non-linear trend extraction system based on ensemble empirical mode decomposition. According to the law of the non-linear change between vegetation carbon storage and soil carbon storage in the process of temporal change and its impact on forest carbon sink capacity, the seasonal change trend of forest carbon sink is extracted.

[0069] Specifically, the system includes a data acquisition module, a data processing module, and a construction module.

[0070] First, the data acquisition module collects forest carbon sink time series data through remote sensing monitoring technology and ground measurement technology, including: vegetation coverage rate VCR, soil coverage rate SCR, vegetation carbon content C veg , soil carbon content C soil , total carbon storage C total .

[0071] Then, the forest carbon sink time series data is transmitted to the data processing module.

[0072] The data processing module preprocesses the forest carbon sink time series data and obtains the forest carbon sink C Carbon Sinkl (t) at different time scales during the seasonal change process.

[0073] Based on the ensemble empirical mode decomposition EEMD, the forest carbon sink C Carbon Sinkl (t) is processed.

[0074] First, C Carbon Sinkl (t) is preprocessed, including:

[0075] Data cleaning: removing missing values and outliers, and performing necessary smoothing processing;

[0076] Detrending and de-meaning: performing detrending processing to ensure that the data is more suitable for EEMD analysis.

[0077] Then, white noise is added to C Carbon Sinkl (t).

[0078] Then, through EEMD, empirical mode decomposition EMD is performed on each signal C Carbon Sinkl(i) (t) with added noise, and the intrinsic mode functions IMFs and the residual r Carbon Sinkl (t) in C i (t) are extracted.

[0079] The construction module extracts the feature X features (t) from the extracted multiple IMFs, and constructs an artificial neural network structure ANN according to the extracted X features (t).

[0080] Finally, through the trained ANN model, C Carbon Sinkl(t) Trend prediction, and draw a trend line graph of the prediction results to visualize the non-linear trend of forest carbon sinks.

[0081] More specifically, the calculation formula for vegetation carbon content: C veg = A veg * B * CF

[0082] Among them, C veg is the vegetation carbon storage; A veg is the area of the vegetation-covered area determined by remote sensing technology; B is the average biomass of vegetation per unit area, calculated through the correlation between the normalized difference vegetation index and ground measurement data; CF is the carbon fraction, which refers to the proportion of carbon in the biomass, and about 50% of the dry weight of plants is usually carbon.

[0083] The calculation formula for soil carbon content: C soil = A soil * D * BD * SOC * (1 - VCR)

[0084] Among them, C soil is the soil carbon storage; A soil is the area covered by the soil; D is the depth of the soil, as carbon storage is usually only calculated to a certain soil depth; BD is the soil bulk density, indicating the mass of the soil per unit volume; SOC is the soil organic carbon content, indicating the proportion of organic carbon in the soil per unit mass, obtained by taking soil samples.

[0085] Through the above calculations of C veg and C soil , the total carbon storage C total can be calculated, and the calculation formula is as follows:

[0086] Among them:

[0087] α * VCR * C veg represents the contribution of vegetation coverage rate to vegetation carbon storage; β * SCR * C soil represents the contribution of soil coverage rate to soil carbon storage; λ * VCR * SCR * (C veg + C soil ) represents the additional impact of the interaction between vegetation and soil coverage rates on the total carbon storage; α is the influence coefficient of vegetation coverage rate on vegetation carbon storage; β is the influence coefficient of soil coverage rate on soil carbon storage; γ is the influence coefficient of the interaction between vegetation and soil on the total carbon storage; α, β, γ are determined through non-linear regression analysis.

[0088] Furthermore, based on C total during the static time, the forest carbon sink C Carbon Sinkl (t) at different times during the seasonal change process is calculated, and the calculation formula is as follows: CCarbon Sink C(t) = C total (t) + ∫[I(t) - O(t)]dt

[0089] where C total (t) is the total carbon storage varying with time; I(t) is the carbon input rate varying with time, including photosynthesis, carbon deposition, or other sources. O(t) is the carbon output rate varying with time, which may include respiration, combustion, decay, or other forms of carbon release; ∫[I(t) - O(t)]dt represents the cumulative integral of carbon input minus carbon output from the initial time to time t, that is, the total carbon increase or decrease.

[0090] The calculation formula of I(t) is: I(t) = P MAX *L(t)*T(t)*VCR(t)

[0091] where P MAX is the maximum photosynthetic rate; L(t) is the light intensity function varying with time, T(t) is the temperature function varying with time, and VCR(t) is the plant coverage rate varying with time.

[0092] The calculation formula of O(t) is: O(t) = R base *T(t)*B(t)

[0093] where: R base is the basal respiration rate; B(t) is the biomass function varying with time.

[0094] The above conventional parameters are all collected and calculated through ground measurement techniques.

[0095] However, the difference is that when adding white noise to C Carbon Sinkl (t), multiple groups of white noise n i (t) will be generated and added to the original signal. By setting different noise intensities α, the randomness of the signal is increased and the mode mixing phenomenon is reduced. The formula for C Carbon Sinkl (t) with added white noise is:

[0096] C Carbon Sinkl(i) (t) = C Carbon Sinkl (t) + αn i (t)

[0097] When EEMD performs empirical mode decomposition EMD on each signal C Carbon Sinkl(i) (t) with added noise, K mode functions will be obtained, where K represents the total number of intrinsic mode functions extracted from the signal: The calculation formula is as follows:

[0098]

[0099] where IMFij $(t)$ represents the $j$-th mode function obtained after the $i$-th addition of white noise decomposition. Each mode function represents a specific frequency component in the original signal, which helps to capture the local features and variations of the signal.

[0100] It should be noted that the extracted $X$ features $(t)$ includes the following: mean, standard deviation, skewness, kurtosis, and frequency. Through effective feature extraction, the prediction performance and generalization ability of the model can be improved. The specific extraction formula is:

[0101] $X$ features $(t)= [mean(IMF$ ij $(t)), std(IMF$ ij $(t)), skewness(IMF$ ij $(t)), kurtosis(IMF$ ij $(t)), frequency(IMF$ ij $(t))]$

[0102] where

[0103] $n$ is the number of data points, $\chi$ i is the value of the $i$-th data point;

[0104]

[0105] Based on the extracted $X$ features $(t)$, construct an ANN model:

[0106] $h$ t $=\sigma(W*X$ features $(t)+b)$

[0107] $h$ t is the output of the hidden layer, $W$ is the weight matrix, $b$ is the bias, and $\sigma$ is the activation function.

[0108] Then train the ANN model:

[0109]

[0110] $\Delta W$ is the weight update, $\eta$ is the learning rate, $L$ is the loss function, is the partial derivative of the weight with respect to the loss function.

[0111] Then evaluate the ANN model:

[0112]

[0113] $MSE$ is the mean squared error, $y$ i is the actual value, is the predicted value;

[0114] and optimize the parameters:

[0115]

[0116] Δη is the learning rate update, and α is the learning rate adjustment coefficient. is the partial derivative of the loss function with respect to the learning rate;

[0117] Application of the final result:

[0118] y predicted = h T ·softmax(W O ·h T + b O )

[0119] h T is a vector representing the output of the last layer (hidden layer) of the neural network, which contains the processing results of the network on the input data;

[0120] W O is the output layer weight matrix that connects the hidden layer and the output layer. Each element of W O determines the influence degree of the hidden layer features on the output result;

[0121] b O is the bias vector of the output layer, which has an independent bias term for each output category;

[0122] softmax is an activation function that converts the output of the network into a probability distribution, where the value of each element represents the probability of the corresponding category;

[0123] y predicted is the final prediction result, which is obtained by multiplying the output of the softmax function by the hidden layer output h T This multiplication operation can be regarded as weighting each category, and the weight is determined by the hidden layer output h T decides.

[0124] Through this integration method, the powerful feature learning and pattern recognition capabilities of the ANN can be utilized to optimize the decomposition results of EEMD, thereby improving the accuracy and efficiency of carbon sink data analysis.

[0125] According to the above content, it can be concluded that when this solution monitors and predicts the forest carbon sink trend, it has the following advantages:

[0126] 1. By monitoring the carbon storage in vegetation and soil, the overall carbon sink capacity of forests can be more accurately evaluated. Given that the carbon storage in vegetation and soil may fluctuate with seasonal changes, monitoring these changes can better understand how the carbon sink is affected by climate factors (such as temperature, precipitation, and sunlight). Meanwhile, during the process of monitoring the changes in carbon storage in different seasons, and based on the law of the impact of the non-linear changes in the carbon storage of vegetation and soil on the forest carbon sink capacity, a long-term change trend graph of the forest carbon sink can be extracted.

[0127] 2. This solution extracts the forest carbon sink trend through EEMD, which can handle the non-linear and non-stationary data of forest carbon sinks, something that is difficult to achieve with traditional extraction methods. This makes EEMD more effective in dealing with data with complex fluctuation patterns such as carbon sinks. Moreover, EEMD can adaptively decompose the data into multiple IMFs, and each IMF represents a different fluctuation scale, which helps to extract the multi-scale characteristics of carbon sink data and provide more details for in-depth analysis.

[0128] 3. During the process of extracting the forest carbon sink trend by EEMD in this solution, an ANN model is added for optimization. ANN can utilize the multi-scale characteristics decomposed by EEMD, comprehensively consider the carbon sink changes at different time scales, improve the comprehensiveness of prediction, and reduce the risk of overfitting in EEMD, thereby enhancing the generalization ability of the model.

[0129] A method for extracting the non-linear trend of carbon sink based on ensemble empirical mode decomposition includes the following steps:

[0130] Step 1: Collect carbon sink time series data through a data acquisition module;

[0131] Step 2: Process the collected carbon sink time series data through a data processing module, and calculate the forest carbon sink C total in different times during the seasonal change based on the C Carbon Sinkl (t) obtained after processing;

[0132] Step 3: Process C Carbon Sinkl (t) based on the ensemble empirical mode decomposition EEMD, and extract the intrinsic mode functions IMFs in C Carbon Sinkl (t), and extract the feature X features (t) from the multiple extracted IMFs;

[0133] Step 4: The construction module constructs an artificial neural network structure ANN according to the extracted X features (t);

[0134] Step 5: Perform C Carbon Sinkl(t) Trend prediction, and extract the non-linear trend of forest carbon sink according to the trend line graph of the prediction result.

[0135] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than limiting it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for extracting carbon sink nonlinear trend based on ensemble empirical mode decomposition, characterized in that: The following steps are involved: S1: Collect forest carbon sink time series data to form a forest carbon sink time series data set. The forest carbon sink time series data includes vegetation coverage rate VCR, soil coverage rate SCR, vegetation carbon content C in different seasons. veg , soil carbon content C soil , total carbon storage C total ; S2: Preprocess the forest carbon sink time series data set based on forest carbon storage theory to obtain forest carbon sink C in different seasons CarbonSinkl (t); S3: Using the ensemble empirical mode decomposition model to CarbonSinkl (t) Extract and obtain data set X features (t) and using the ANN model to features (t) processing to extract the nonlinear trend of forest carbon sink, wherein the ensemble empirical mode decomposition model uses ensemble empirical mode decomposition (EEMD) to calculate the C CarbonSinkl (t) Process and extract C CarbonSinkl (t) and extract the feature X through IMFs features (t).

2. The extraction method according to claim 1, characterized in that: In S1, the data acquisition module collects forest carbon sink time series data through remote sensing monitoring technology and ground measurement technology.

3. The extraction method according to claim 2, characterized in that: In S2, C CarbonSinkl The calculation process of (t) is: S201: Based on forest carbon storage theory, obtain the total carbon storage C in static time total , specifically: C total =(α*VCR*C veg )+(β*SCR*C soil )+[λ*VCR*SCR*(C veg +C soil )] in: α*VCR*C veg Indicates the contribution of vegetation coverage to vegetation carbon storage; β*SCR*C soil represents the contribution of soil coverage to soil carbon storage; λ*VCR*SCR*(C veg +C soil ) represents the additional effect of the interaction between vegetation and soil coverage on total carbon storage; α is the influence coefficient of vegetation coverage on vegetation carbon storage; β is the influence coefficient of soil coverage on soil carbon storage; γ is the influence coefficient of the interaction between vegetation and soil on total carbon storage; α, β, and γ are determined by nonlinear regression analysis; S202: Based on the total carbon storage C total Calculate the forest carbon sink C at different times during seasonal changes CarbonSinkl (t), specifically: C CarbonSink (t)=C total (t)+∫[I(t)-O(t)]dt Among them, I(t) is the carbon input rate that changes with time, O(t) is the carbon output rate that changes with time, and ∫[I(t)-O(t)]dt represents the cumulative integral of carbon input minus carbon output from the initial time to time t, that is, the total carbon increase or decrease.

4. The extraction method according to claim 3, characterized in that: In S3, the C CarbonSinkl (t) extracting, comprising the following steps: S301: Towards C CarbonSinkl Add white noise to (t): C CarbonSinkl ( i) (t)=C CarbonSinkl (t)+αn i (t) S302: For each signal C that introduces white noise CarbonSinkl(i) (t) Performing empirical mode decomposition (EMD); S303: Extraction of intrinsic mode function: Among them, IMF ij (t) indicates that this is the jth mode function obtained by decomposition after adding white noise for the i-th time; S304:X features (t) Extraction: in, n is the number of data points and is the value of the ith data point.

5. The extraction method according to claim 4, characterized in that: In S3, the ANN model is used to analyze the X features (t) Processing to extract nonlinear trends in forest carbon sinks, specifically: According to X features (t) Construct ANN model: h t =σ(W*X features (t)+b) h t is the output of the hidden layer, W is the weight matrix, b is the bias, and σ is the activation function; The training formula of the ANN model is: ΔW is the weight update, η is the learning rate, L is the loss function, is the partial derivative of the weight with respect to the loss function; The evaluation formula of the ANN model is: MSE is the mean square error, which is the actual value, y i is the actual value, is the predicted value; The parameter tuning formula is: Δη is the learning rate update, α is the learning rate adjustment factor, is the partial derivative of the loss function with respect to the learning rate; y predicted =h T ·softmax(W O ·h T +b O ) h T is a vector representing the output of the last layer (hidden layer) of the neural network; W O is the output layer weight matrix, connecting the hidden layer and the output layer; b O is the bias vector of the output layer; Softmax is an activation function that converts the output of the network into a probability distribution, where the value of each element represents the probability of the corresponding category; y predicted for the final nonlinear trend extraction results.

6. A carbon sink nonlinear trend extraction system based on ensemble empirical mode decomposition, applied to execute the extraction method performed by the nonlinear trend extraction system according to any one of claims 1 to 5, the system comprising a data acquisition module and a data processing module; The data acquisition module is used to collect forest carbon sink time series data to form a forest carbon sink time series data set. The forest carbon sink time series data set includes vegetation coverage rate VCR, soil coverage rate SCR, vegetation carbon content C in different seasons. veg , soil carbon content C soil , total carbon storage C total ; The data processing module is used to preprocess the forest carbon sink time series data set to obtain forest carbon sink C in different seasons. CarbonSinkl (t) and extracting the forest carbon sink time series data set according to the pre-constructed ensemble empirical mode decomposition model to obtain a data set X features (t), according to the pre-built ANN model, features (t) Processing to extract nonlinear trends in forest carbon sinks.

Images