An ecological carbon sink potential prediction method and system based on a fusion attention mechanism

Abstract

The application discloses a kind of ecological carbon sink potential prediction method and system based on fusion attention mechanism, it is related to ecological carbon sink monitoring and image processing technical field, the method includes: obtaining the reflectivity image of the scene to be evaluated, and the meteorological variable data of the time period to be evaluated;Image local feature is extracted from reflectivity image by CNN, meteorological variable feature is extracted from meteorological variable data by BiLSTM;Meteorological time series vector is calculated based on attention mechanism;The dynamic weight of image local feature and meteorological time series vector feature is calculated, and the weight fusion feature is obtained by the respective dynamic weight;Weight fusion feature is input into fully connected layer and is processed, feature variable is mapped to prediction space to obtain the predicted ecological carbon sink data of the scene to be evaluated in the time period to be evaluated;Solve the technical difficulties that current terrestrial ecosystem carbon sink estimation cannot accurately predict carbon sink change under the real demand of large range and high precision.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ecological carbon sink monitoring and image processing, more specifically, it relates to an ecological carbon sink potential prediction method and system based on a fusion attention mechanism. BACKGROUND

[0002] Traditional methods are mainly divided into "top-down" and "bottom-up" according to data sources and model methods; among them, the "bottom-up" method refers to the generalization of sample or grid scale ground observations and simulation results to regional scale, mainly relying on ground sampling, and the commonly used methods include: inventory method, eddy correlation method and ecological process simulation method. The "top-down" method mainly refers to carbon sink inversion technology, which is based on observation data from different remote sensing platforms, and uses a radiation transfer model to calculate the carbon sink and storage of the terrestrial ecosystem. This method of using model simulation has clear technical solutions and high algorithm execution efficiency, and can realize the ecological system carbon sink assessment of a large range of regions, becoming an effective technical means that many ecologists are keen to study. At the same time, with the rapid development of remote sensing technology and computer vision technology, using high-resolution satellite remote sensing data to estimate the carbon sink of the terrestrial ecosystem has become a research hotspot. However, due to the complex environment of the terrestrial ecosystem, there are still great challenges in accurately describing the quantitative relationship between past and future terrestrial ecosystem carbon sinks and ecological environmental factors; the existing "top-down" based ecological carbon sink estimation technical solutions mainly include:

[0003] Patent No. CN117575359B discloses an ecological system carbon sink evaluation method for unified regulation; this scheme uses land patches as unified spatial units, and determines the amount of carbon dioxide that can be absorbed and fixed by the ecological system of the plot by inputting ecological system carbon sink evaluation data into an ecological system carbon sink evaluation model.

[0004] Patent No. CN115310757B discloses a high-precision dynamic variable resolution ecological carbon sink evaluation method and system; this scheme is realized based on the YIBs vegetation model, and modifies the spatial resolution and format of the data on multiple spatial scales to adapt to data of different formats and spatial resolutions, and inputs the converted ecological carbon sink related variables into the YIBs vegetation model for carbon sink carbon fixation calculation; this method effectively solves the problem of mismatching of data format and resolution caused by different data sources through spatial interpolation and upsizing and downsizing methods.

[0005] The patent with publication number CN119003679B discloses a method for monitoring carbon sink of an ecological system based on remote sensing; the scheme collects real-time remote sensing data of a wetland protection area through a remote sensing satellite, analyzes the remote sensing data in a ground station, and obtains and calculates wetland carbon sink data, then builds a carbon sink monitoring model to calculate and output regional carbon sink and extend to overall carbon sink.

[0006] The paper "Estimation and Protection Zone Identification of Terrestrial Ecosystem Carbon Sink in Zhejiang Province under Future Climate Background" proposes to use the Integrated Biosphere Simulator (IBIS) to simulate the terrestrial ecosystem carbon sink at a kilometer grid scale, and identifies the spatial range of carbon sink protection zones based on different climate backgrounds. The article discusses the level and spatial distribution characteristics of the terrestrial ecosystem carbon sink at the regional scale through quantitative modeling, and evaluates the degree of action of optimization strategies on maintaining the terrestrial ecosystem carbon sink in a quantitative way.

[0007] The above-mentioned existing "top-down" ecological carbon sink estimation technical scheme mainly has the following problems:

[0008] The patent with publication number CN117575359B evaluates the total amount of ecological carbon sink by building a unified evaluation unit, but the proposal lacks technical elaboration of carbon sink estimation and prediction models, and does not establish and clarify the quantitative relationship between the total amount of ecological carbon sink and climate environmental factors.

[0009] The patent with publication number CN115310757B builds a YIBs vegetation model and uses multi-scale spatial resolution and format dynamic correction methods to realize a high-precision dynamic variable resolution ecological carbon sink evaluation system based on physical processes; however, this system relies on the YIBs model to achieve high-precision ecological carbon sink simulation, and requires multiple parameters such as input data demand control information, calculation parameters, driving data, initial field data, and environmental data, each of which is further subdivided into multiple data sources, which has great limitations in the application of long-term, large-scale ecological carbon sink evaluation.

[0010] The patent with publication number CN119003679B mentions using real-time data from remote sensing satellites to build a carbon sink estimation model for ecological system carbon sink monitoring, but this scheme only estimates carbon sink for regional real-time data and cannot realize long-term, large-scale quantitative evaluation and prediction of ecological carbon sink for the past and future.

[0011] The paper "Estimation and Protection Zone Identification of Terrestrial Ecosystem Carbon Sink in Zhejiang Province under Future Climate Background" uses the Integrated Biosphere Simulator (IBIS) to simulate the terrestrial ecosystem carbon sink at a kilometer grid scale, and discusses the level and spatial distribution characteristics of the terrestrial ecosystem carbon sink at the regional scale; but in the context of the application demand of fine ecological carbon sink evaluation, the results at the kilometer scale cannot meet the actual business demand.

[0012] In view of this, the present application is proposed. SUMMARY

[0013] The present application aims to provide a fusion attention mechanism-based ecological carbon sink potential prediction method and system, which is suitable for high-resolution ecological carbon sink evaluation scenarios, and realizes long-time series and high-spatial resolution land ecosystem carbon sink estimation and prediction at regional and even global scales by combining remote sensing reflectivity images and future climate scenario data through ecological carbon sink estimation and ecological carbon sink potential prediction models, and solves the technical difficulties that current land ecosystem carbon sink estimation cannot accurately predict carbon sink changes under the realistic demand of large-scale and high-precision.

[0014] The above technical purposes of the present application are achieved by the following technical solutions.

[0015] In a first aspect, the present application provides a fusion attention mechanism-based ecological carbon sink potential prediction method, comprising the following specific steps:

[0016] S1, obtaining reflectivity images of a scene to be evaluated and meteorological variable data of the scene to be evaluated in a future evaluation period to be evaluated;

[0017] S2, extracting image local features from the reflectivity images by CNN and extracting meteorological variable features from the meteorological variable data by BiLSTM;

[0018] S3, performing weighted processing on the meteorological variable features based on an attention mechanism, and calculating a meteorological time series vector corresponding to the fusion attention weight of the meteorological variable features;

[0019] S4, calculating the dynamic weights of the image local features and the meteorological time series vector features based on a dynamic weight distribution mechanism, and obtaining weight fusion features through the respective dynamic weights;

[0020] S5, inputting the weight fusion features into a fully connected layer for processing, mapping the feature variables to a prediction space to obtain predicted ecological carbon sink data of the scene to be evaluated in the evaluation period to be evaluated.

[0021] On the basis of the above technical solutions, the present application can also be improved as follows.

[0022] Further, the meteorological time series vector corresponding to the fusion attention weight of the meteorological variable features is calculated as follows:

[0023] wherein:

[0024] , ;

[0025] In the formula, represents the meteorological time series vector feature, is the calculated attention weight, represents the attention score of the t-th time step, and T represents the total time step, , , respectively represent learnable parameters, represents the hidden state of the BiLSTM at time step t, represents the context vector.

[0026] Further, the above dynamic weight distribution mechanism is specifically:

[0027] ;

[0028] ;

[0029] In the formula, represents the dynamic weight corresponding to the image local feature, represents the dynamic weight of the meteorological time series vector feature, respectively represent learnable parameters, is the reflectivity image remote sensing image quality, represented by the cloud cover ratio, and ; is the reliability of the meteorological variable data, represented by the model uncertainty in the CMIP6 dataset, and ; is the time series integrity, represented by the data missing rate, and .

[0030] Further, the above weight fusion feature is specifically:

[0031] ;

[0032] In the formula, is the weight fusion feature, represents the image local feature, represents the meteorological time series vector feature, represents the dynamic weight corresponding to the image local feature, represents the dynamic weight of the meteorological time series vector feature.

[0033] Further, the above steps S2-S5 are executed by the constructed CNN-BiLSTM-Attention hybrid model, and the CNN-BiLSTM-Attention hybrid model is obtained by the following way:

[0034] ​constructing a training sample, the training sample comprising spatio-temporally aligned historical meteorological data, historical reflectivity images and historical carbon sink data;

[0035] inputting the historical time-series reflectivity images and the historical meteorological data of the specified period into a preset hybrid network model for model training to obtain predicted carbon sink data of the specified period;

[0036] calculating a loss function of the hybrid network model based on the predicted carbon sink data and the historical carbon sink data of the corresponding period until a training end condition is reached, and determining the hybrid network model that reaches the training end condition as a CNN-BiLSTM-Attention hybrid model.

[0037] Further, the loss function is specifically:

[0038] ;

[0039] In the formula, represents the historical carbon sink data, represents the predicted carbon sink data, represents the number of samples.

[0040] Further, the historical carbon sink data is obtained by the following method:

[0041] Optical reflectivity images are obtained by Sentinel-2A, leaf area index is retrieved by a radiation transfer model, and historical terrestrial ecosystem carbon sink is estimated by a dual-leaf light energy utilization rate model to obtain the historical carbon sink data.

[0042] In a second aspect, the present application provides an ecological carbon sink potential prediction system based on a fusion attention mechanism, which is applied to the ecological carbon sink potential prediction method based on the fusion attention mechanism in any one of the first aspect, and comprises:

[0043] A sample data acquisition module is configured to acquire reflectivity images of a scene to be evaluated and meteorological variable data of a future evaluation period of the scene to be evaluated.

[0044] A data feature extraction module is configured to extract image local features from the reflectivity images by CNN and extract meteorological variable features from the meteorological variable data by BiLSTM.

[0045] A meteorological feature weighting module is configured to perform weighting processing on the meteorological variable features based on an attention mechanism and calculate a meteorological time-series vector corresponding to the fusion attention weight of the meteorological variable features.

[0046] The fusion feature acquisition module is configured to calculate dynamic weights of the image local features and the meteorological time series vector features based on a dynamic weight distribution mechanism, and obtain the weight fusion features through the respective dynamic weights.

[0047] The ecological carbon sink calculation module is configured to input the weight fusion features into a full connection layer for processing, map the feature variables to a prediction space, and obtain predicted ecological carbon sink data of the to-be-evaluated scene in the to-be-evaluated period.

[0048] In a third aspect, the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method of any one of the first aspect when executing the computer program.

[0049] In a fourth aspect, the present application provides a non-transitory computer readable storage medium, which stores computer instructions, wherein the computer instructions cause a computer to execute the method of any one of the first aspect.

[0050] Compared with the prior art, the present application has at least the following beneficial effects:

[0051] 1. Compared with the current technology of realizing kilometer grid scale terrestrial ecosystem carbon sink simulation by using the integrated biosphere model (IBIS), the present application has the advantages of fusing CNN to extract local features, BiLSTM to capture long-term dependence of long time series data, and attention mechanism to dynamically adjust the attention degree to the input sequence, and can realize ten-meter-level terrestrial ecological carbon sink estimation and predict the terrestrial ecological carbon sink potential level and change trend of the same order of magnitude in the future to 2300.

[0052] 2. The present application breaks through the dependence of traditional models on specific data formats and resolutions, and adapts multi-source remote sensing and climate data through a dynamic weight mechanism and a multi-modal data preprocessing module, and is suitable for carbon sink prediction in different regions and time scales. BRIEF DESCRIPTION OF DRAWINGS

[0053] The accompanying drawings, which are included to provide a further understanding of the embodiments of the present application and constitute a part of the present application, do not constitute a limitation to the embodiments of the present application. In the drawings:

[0054] Figure 1 The method flowchart of the prediction method in the embodiments of the present application;

[0055] Figure 2 The connection schematic diagram of the prediction system in the embodiments of the present application;

[0056] Figure 3 The connection schematic diagram of the electronic device in the embodiments of the present application. DETAILED DESCRIPTION

[0057] So that the purposes, technical solutions and advantages of the embodiments of the present application are more apparent, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations.

[0058] Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative labor are within the scope of protection of the present application.

[0059] It should be noted that: similar reference numbers and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0060] In the description of the embodiments of the present application, "a plurality of" represents at least 2.

[0061] Embodiment 1: In order to solve the technical difficulty that the current carbon sink estimation of terrestrial ecosystem cannot accurately predict the change of carbon sink under the real demand of large range and high precision, the embodiment provides a method for predicting ecological carbon sink potential based on fusion attention mechanism, as shown in Figure 1 The specific steps include:

[0062] S1, obtaining the reflectivity image of the scene to be evaluated and the meteorological variable data of the future to be evaluated period of the scene to be evaluated.

[0063] Wherein, Sentinel-2A can be used to obtain optical reflectivity image, and for meteorological variable data of the future to be evaluated period, based on the integrated Coupled Model Intercomparison Project Phase 6 (CMIP6) historical and future climate scenario data, which mainly includes 4 kinds of climate scenarios of SSP1-2.6, SSP2-4.5, SSP3-7.0 and SSP5-8.5, the longest can obtain each meteorological variable data up to 2300 years, and the meteorological variable data can include temperature, precipitation, radiation, etc.

[0064] S2, extracting image local features from reflectivity image by CNN, and extracting meteorological variable features from meteorological variable data by BiLSTM.

[0065] Wherein, the image local feature can be realized by one-dimensional convolution layer (CNN-1d), which can support input of multiple features and adapt to the multi-feature complexity scene in land ecosystem carbon sink prediction; the main calculation process can be abstracted as the following formula:

[0066] ;

[0067] In the formula, is the index in Batch, is the index of output channel, is the position index in the convolution kernel, is the dilation coefficient, is the size of left padding, represents the position index of the input sequence, c is is the index of input channel, in channels represents the length of the input sequence; wherein the length of the output sequence (output channels) can be calculated according to the length of the input sequence (input channels), the size of the convolution kernel (kernel size), the step (stride), the padding (padding) and other parameters, and the calculation formula is as follows:

[0068] In the formula, represents the number of input channels, represents the number of output channels of the convolution layer, represents the size of the convolution kernel, represents the step of convolution, represents the padding amount.

[0069] Further, the meteorological variable feature is executed by BiLSTM, which is a special recurrent neural network (RNN) that captures the context information at each time step in the sequence by simultaneously processing the forward and backward information of the sequence data, effectively solving the problems of RNN gradient disappearance and gradient explosion; BiLSTM is often used to capture long-term dependencies in long time series data, and the information transmission is controlled through the gating mechanism (input gate, forget gate, output gate), which mainly consists of two LSTM networks: a forward LSTM and a backward LSTM, and the mathematical expression can be abstracted as follows:

[0070] For the input sequence X={x1,x2,x3,…x t}, the forward LSTM processes the input in order from the first element to the last element of the sequence and generates the forward hidden state .

[0071] Forget gate: ;

[0072] Input gate: ;

[0073] Candidate update cell state: ;

[0074] Cell state update: ;

[0075] Output gate: ;

[0076] Hidden state update: .

[0077] Further, the backward LSTM processes the output in reverse from the last element of the sequence to the first element and generates a backward hidden state .

[0078] Forget gate: ;

[0079] Input gate: ;

[0080] Candidate update cell state: ;

[0081] Cell state update:

[0082] Output gate: ;

[0083] Hidden state update: ;

[0084] Hidden state fusion, at each time step t, the forward and backward hidden are concatenated together to form the final bidirectional hidden state: ;

[0085] In the above formula, , , respectively represent the forget gate, input gate, output gate, is the neuron state, is the hidden state. is the weight matrix, is the bias term, is the sigmoid function. The superscripts f and b correspond to the forward and backward LSTM respectively. Among them, represents pixel-by-pixel multiplication, represents the input sequence.

[0086] Specifically, BiLSTM utilizes the context information before and after each time step in the time series in this way, providing a more comprehensive feature representation when processing sequence data, thus obtaining the above-mentioned meteorological variable features.

[0087] S3, based on the attention mechanism, the meteorological variable features are weighted and processed, and the meteorological time series vector corresponding to the fusion attention weight of the meteorological variable features is calculated.

[0088] Optionally, the meteorological time series vector corresponding to the fusion attention weight of the meteorological variable features is calculated, specifically:

[0089] , wherein:

[0090] ,

[0091] In the formula, indicates the meteorological time series vector feature, is the calculated attention weight, indicates the attention score of the t-th time step, and T indicates the total time step, , , respectively indicate learnable parameters, indicates the hidden state of BiLSTM at time step t, indicates the context vector.

[0092] S4, based on the dynamic weight distribution mechanism, the dynamic weights of the image local features and the meteorological time series vector features are calculated respectively, and the weight fusion features are obtained through the corresponding dynamic weights.

[0093] Wherein, based on the dynamic weight distribution mechanism, the contribution of each network (CNN / BiLSTM) is dynamically adjusted according to real-time data quality (such as remote sensing image cloud cover, climate data uncertainty); The mechanism maximizes the prediction accuracy as the goal, updates the weight parameters using the policy gradient algorithm, and improves the robustness of the model in complex climate environment, and the specific steps are as follows:

[0094] 1) Data quality index quantization: real-time extraction of quality features of input data, mainly including:

[0095] Remote sensing image quality: represented by cloud cover ratio ( ), the smaller the value, the higher the image quality.

[0096] Climate data reliability: represented by model uncertainty in CMIP6 data set ( ), the smaller the value, the higher the reliability.

[0097] Temporal integrity: characterized by data missing rate ( The smaller the value, the better the data integrity.

[0098] 2) Dynamic Weight Function: A dynamic weight function is designed based on data quality metrics to output the CNN network weights. and BiLSTM weights When the quality of remote sensing data is high ( (smaller values), increase the weights of the CNN network. The importance of enhancing local features. When the reliability of climate time-series data is high ( , (If the value is small), increase the BiLSTM weights. This enhances the importance of time-series features; the dynamic weighting function is calculated as follows: , ;in, , Set initial values ​​for learnable parameters. = =0.5.

[0099] 3) Weight fusion and prediction output: The local features extracted by the CNN are fused together... ) and BiLSTM extraction of temporal features ( Based on dynamic weight fusion: The fused features are processed through an attention mechanism and a fully connected layer to output a predicted value of carbon sequestration potential.

[0100] 4) Weight parameter optimization: Using the mean squared error (MSE) between the predicted and actual values ​​as the loss function, the parameters of the weight function are optimized synchronously during model training. , ).

[0101] S5 inputs the weighted fusion features into the fully connected layer for processing, and maps the feature variables to the prediction space to obtain the predicted ecological carbon sink data of the scenario to be evaluated during the evaluation period.

[0102] Among them, the fully connected layer (FC) will be obtained through weight fusion. Mapped to the prediction space, future carbon sequestration data is output through a fully connected layer; the output layer uses a linear activation function, and the loss function can be the mean squared error (MSE); its mathematical expression can be: In the formula, For the predicted results, This is the weight matrix. Bias term.

[0103] Optionally, the above steps S2-S5 are executed by a constructed CNN-BiLSTM-Attention hybrid model, and the CNN-BiLSTM-Attention hybrid model is obtained by the following method:

[0104] S51, constructing a training sample, the training sample including spatio-temporally aligned historical meteorological data, historical reflectivity image and historical carbon sink data.

[0105] Wherein, the obtained ecological carbon sink net primary productivity (NPP) and the temperature, precipitation, radiation and other climate variable data of historical (2017-2025) different scenario modes (SSP1-2.6, SSP2-4.5, SSP3-7.0, SSP5-8.5) in the sixth international coupling mode comparison project (CMIP6) are sampled to the same resolution by the spatio-temporal alignment method (such as bilinear interpolation, spatio-temporal kriging interpolation), and are time series aligned by using the time series label DOY, and the data are placed in the same folder as the training data set in the sample-label mode.

[0106] S52, inputting the historical time series reflectivity image and the historical meteorological data of the specified period into the preset hybrid network model for processing to obtain the predicted carbon sink data of the specified period.

[0107] S53, calculating the loss function of the hybrid network model based on the predicted carbon sink data and the historical carbon sink data of the corresponding period until the training end condition is reached, and determining the hybrid network model reaching the training end condition as the CNN-BiLSTM-Attention hybrid model.

[0108] Wherein, the CNN-BiLSTM-Attention hybrid model is constructed based on the Pytorch deep learning framework, and the time series segmentation method is used for division, and the training, verification and test data sets are divided according to the ratio of 8:1:1; during training, the loss function selects mean square error (MSE) for measuring the difference between the predicted value and the true value of the model; the loss function calculation formula is as follows:

[0109] ;

[0110] In the formula, represents the historical carbon sink data, is the predicted carbon sink data, The number of samples; the optimizer can select the Adam optimizer, the initial learning rate is 0.001, the learning rate strategy uses cosine annealing (Cosine Annealing) to dynamically adjust the learning rate, and the batch size can be set to 8; After training, the model performance can be evaluated using the test dataset, and evaluation indicators such as root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²) can be calculated.

[0111] The carbon storage estimation technology route of PROSAIL+TL-LUE and CMIP6 historical and future multi-scenario climate data are integrated, the climate variables are spatiotemporally aligned with the ecological carbon sink estimation results using interpolation methods, a multi-source multi-modal dataset is formed, and the whole-chain modeling of historical carbon sink-future climate-potential prediction is realized; At the same time, a deep fusion model of one-dimensional convolutional neural network (CNN-1D) and bidirectional long short-term memory network (BiLSTM) with fusion attention mechanism is applied to realize the triple functions of local feature extraction, long-term sequence dependence capture and key period focusing of carbon sink data, adapt to the characteristics of ecological carbon sink "local climate response + long-term cumulative effect", and improve the accuracy of prediction.

[0112] In the above, the historical carbon sink data is obtained by the following method:

[0113] Optical reflectance images are obtained through Sentinel-2A, leaf area index (LAI) is retrieved using a radiation transfer model (PROSAIL), and historical terrestrial ecosystem carbon sinks are estimated using a dual-leaf light use efficiency model (TL-LUE) to calculate net primary productivity (NPP); Then, using sliding window statistics and seasonal decomposition trend components, combined with climate factors (such as temperature, precipitation, radiation, etc.), a multi-climate variable input is formed, and these climate variables are mapped to the corresponding date of carbon sink data through time sequence labels, and a carbon sink dataset is integrated, which contains historical meteorological data, historical reflectance images and historical carbon sink data; Wherein:

[0114] 1. About PROSAIL radiative transfer model, PROSAIL radiative transfer model is derived by coupling PROSPECT leaf model and SAIL canopy structure model, which explains vegetation reflectance from biophysical properties, thus expressing vegetation optical properties. PROSPECT model (leaf scale) simulates the reflectance and transmittance of vegetation leaves at 400-2500 nm, and the model has two types of input parameters, leaf reflectance and transmittance, which change with leaf structure parameters, leaf chlorophyll content, water thickness, dry matter content, brown pigment and leaf carotenoid content, and other structural and biochemical parameters. SAIL model (canopy scale) calculates the bidirectional reflectance at the top of the canopy, which is a function of leaf reflectance and transmittance derived from PROSPECT and other input parameters (LAI, average leaf angle, hotspot size, solar zenith angle, viewing zenith angle, relative azimuth angle, albedo coefficient and diffuse / solar ratio). The PROSAIL model can be formally expressed as:

[0115] (Formula 8-1)

[0116] In the formula, is the canopy reflectance; N is the leaf structure parameter; is the chlorophyll content; is the water content; is the dry matter content; LAI is the leaf area index; ALA is the average leaf angle; is the hotspot parameter; is the soil moisture, is the soil albedo; is the observation zenith angle; is the solar zenith angle; is the solar and observation relative azimuth angle.

[0117] Through the PROSAIL model, the bidirectional reflectance at the top of the canopy at 400-2500 nm can be simulated . Then, LAI is retrieved by look-up table method. The main steps are as follows:

[0118] 1) Generate a dataset of LAI and other parameters, calculate the corresponding ;

[0119] 2) Calculate the mean square error of the measured reflectance of the image and in the LUT;

[0120] 3) Select the LAI corresponding to the LUT record with the smallest RMSE as the retrieval result, as follows:

[0121] (Formula 8-2)

[0122] 2. Calculate the historical terrestrial ecosystem carbon sink net primary productivity based on the TL-LUE model:

[0123] 1) Inversion of gross primary productivity

[0124] The two-leaf light use efficiency (TL-LUE) model is derived from the MOD17 algorithm, which divides the canopy into sunlit leaves and shaded leaves, and improves the calculation of absorbed photosynthetically active radiation (APAR) and net primary productivity (GPP) in the BEPS model. The formula for calculating the net primary productivity (GPP) of the vegetation canopy is as follows:

[0125] (Formula 8-3)

[0126] In the formula, and are the maximum light use efficiency of sunlit leaves and shaded leaves, respectively; and are the amount of incident photosynthetically active radiation (PAR) absorbed by sunlit and shaded leaves, and the calculation formula is as follows:

[0127] (Formula 8-4)

[0128] (Formula 8-5)

[0129] where, is the canopy albedo related to vegetation type; is the average leaf angle, which is set to 60 degrees for a canopy with spherical distribution of leaf angles; is the solar zenith angle; , and are the diffuse, direct components of incident PAR and the diffuse PAR under the canopy, respectively. represents multiple scattering of direct radiation, and the calculation formula is as follows:

[0130] (Formula 8-6)

[0131] and are the LAI of sunlit and shaded leaves, which are divided according to the canopy LAI, clumping index (C ) and solar zenith angle; The calculation formula is as follows:

[0132] (Formula 8-7)

[0133] In formula (8-3), and are the scalar of VPD minimum temperature, and the calculation method is as follows:

[0134] (Formula 8-8)

[0135] (Formula 8-9)

[0136] In the above formula, VPD is the average VPD during the day, is the daily minimum temperature, which can be calculated from half-hour data. and , and is the minimum maximum parameter value of the saturated water vapor pressure difference and temperature determined according to the vegetation type.

[0137] 2) NPP calculation

[0138] Net primary productivity is the gross primary productivity minus plant autotrophic respiration, and its calculation formula is as follows:

[0139] (Formula 8-10)

[0140] In the formula, NPP is the net primary productivity of vegetation, GPP is the gross primary productivity, and Ra is the plant autotrophic respiration.

[0141] Further, the spatio-temporal interpolation method in the above is realized by the following steps:

[0142] 1) Data preprocessing and spatio-temporal reference unification

[0143] Spatial reference: The resolution of the vegetation net primary productivity data is resampled to 1km resolution, which is convenient for adaptation with CMIP6 climate scenario data, and the bilinear interpolation method is used for processing.

[0144] Temporal reference: The climate variables in the CMIP6 daily scale climate scenario data are aggregated to monthly mean values, and the processing method can be abstracted as:

[0145] (Formula 8-11)

[0146] where, is the mean value of the climate variable in the tth month, is the number of days in the tth month, is the meteorological data of the dth day in the tth month. In addition, the carbon sink data is also synthesized to the monthly scale by the mean synthesis method, which is convenient for completely aligning with the time sequence label of the meteorological data.

[0147] 2) Spatial interpolation, spatial interpolation adopts the spatio-temporal Kriging interpolation method, which is used to interpolate the low-resolution CMIP6 grid data to 1km*1km grid size. The spatio-temporal Kriging extends the traditional Kriging interpolation to the time dimension, and its prediction value is the weighted sum of known points The calculation formula is as follows:

[0148] (Formula 8-12)

[0149] where the weight is solved by minimizing the prediction variance and satisfies the unbiasedness constraint .

[0150] The most critical component of spatiotemporal Kriging interpolation is the spatiotemporal variogram , which describes the difference in data under spatial distance and time interval :

[0151] (Formula 8-13)

[0152] 3) Spatiotemporal co-Kriging interpolation method

[0153] Calculate the spatial variogram and the temporal variogram , which represent the spatiotemporal autocorrelation of the data, and the calculation formula is as follows:

[0154] (Formula 8-14)

[0155] (Formula 8-15)

[0156] where is the spatial lag distance, is the time lag, and is the lag sample size, and Z is the climate variable.

[0157] where the spatiotemporal covariance function is:

[0158] (Formula 8-16)

[0159] where , where a is the spatial range, b is the time range, is the base value.

[0160] Interpolation prediction: for the target point , the predicted value of the climate variable is:

[0161] (Formula 8-17)

[0162] where is an interpolation weight, is a Lagrange multiplier satisfying the constraint condition:

[0163] , (Formula 8-18)

[0164] wherein, is a spatial distance between the target point and the sample point k, is a time difference.

[0165] Embodiment 2: The embodiment of the present application provides an ecological carbon sink potential prediction system based on a fusion attention mechanism, which is applied to the ecological carbon sink potential prediction method based on the fusion attention mechanism in embodiment 1, as shown in Figure 2 , comprising:

[0166] A sample data acquisition module is configured to acquire reflectivity images of a scene to be evaluated and meteorological variable data of the scene to be evaluated in a future to-be-evaluated period.

[0167] A data feature extraction module is configured to extract image local features from the reflectivity images by using a CNN and extract meteorological variable features from the meteorological variable data by using a BiLSTM.

[0168] A meteorological feature weighting module is configured to perform weighting processing on the meteorological variable features based on an attention mechanism and calculate a meteorological time sequence vector corresponding to a fusion attention weight of the meteorological variable features.

[0169] A fusion feature acquisition module is configured to calculate dynamic weights of the image local features and the meteorological time sequence vector features based on a dynamic weight distribution mechanism, and obtain weight fusion features through the respective dynamic weights.

[0170] An ecological carbon sink calculation module is configured to input the weight fusion features into a fully connected layer for processing, map the feature variables to a prediction space to obtain predicted ecological carbon sink data of the scene to be evaluated in the to-be-evaluated period.

[0171] Embodiment 3: The embodiment of the present application provides an electronic device, as shown in Figure 3 , comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the method of embodiment 1 when executing the computer program.

[0172] Embodiment 4: The embodiment of the present application provides a non-transitory computer readable storage medium, which stores computer instructions, and the computer instructions make the computer execute the method of embodiment 1.

[0173] Those skilled in the art will appreciate that embodiments of the application can be readily used as software, hardware, or a combination of software and hardware. In a software embodiment, various routines according to embodiments of the application can be stored in the memories of a general purpose computer, special purpose computer, or microprocessor. Such data can he downloaded to the memories 5 from computer program product available over the wired or wireless network, from computer system or from a floppy disk or from any other computer program product. Software routines can be implemented with assembly or machine language, or with higher level languages such as C.

[0174] The present application is described in reference to the flowchart and / or block diagrams of the methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each block of the flowchart and / or block diagrams, and combinations of blocks in the flowchart and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks.

[0175] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks.

[0176] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks.

[0177] Those skilled in the art will appreciate that all or portions of the methods and apparatuses described herein can be embodied in a program or other processor-readable storage media. Program code can be applied to input data to perform the functions described and to generate output information. The output information, similarly, can be applied to one or more output devices, in real time or in batch processing. It will be understood that the processes described can be implemented in software, firmware, hardware, or any combination thereof.

[0178] The above detailed description of the specific embodiments of the present application has been given to understand the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not used to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for predicting ecological carbon sink potential based on a fusion attention mechanism, characterized in that, The specific steps include the following: S1, acquire reflectance images of the scene to be evaluated, as well as meteorological variable data of the scene to be evaluated in the future evaluation period; S2, local image features are extracted from the reflectance image using CNN, and meteorological variable features are extracted from the meteorological variable data using BiLSTM; S3, the meteorological variable features are weighted based on the attention mechanism, and the meteorological time series vector corresponding to the meteorological variable features with fused attention weights is calculated; S4, based on a dynamic weight allocation mechanism, calculate the dynamic weights of the local image features and the meteorological time-series vector features respectively, and obtain the weighted fusion features through the corresponding dynamic weights; the dynamic weight allocation mechanism is specifically as follows: ； ； In the formula, This represents the dynamic weights corresponding to local features of the image. Dynamic weights representing the characteristics of meteorological time-series vectors. , They represent the learnable parameters, The quality of remote sensing images is represented by the proportion of cloud cover, and the reflectance image is the reflectance image. ; To assess the reliability of meteorological variable data, model uncertainty in the CMIP6 dataset is used as a representation. ; For temporal integrity, it is characterized by the data missing rate, and ; S5, the weighted fusion features are input into the fully connected layer for processing, and the feature variables are mapped to the prediction space to obtain the predicted ecological carbon sink data of the scenario to be evaluated during the evaluation period.

2. The method for predicting ecological carbon sink potential based on a fusion attention mechanism according to claim 1, characterized in that, The calculation yields a meteorological time-series vector with fused attention weights corresponding to the meteorological variable features, specifically: ,in: ， ； In the formula, Representing the characteristics of meteorological time series vectors, For calculating attention weights, Let T represent the attention score at time step t, and T represent the total number of time steps. , , They represent the learnable parameters, This represents the hidden state of the BiLSTM at time step t. This represents the context vector.

3. The method for predicting ecological carbon sink potential based on a fusion attention mechanism according to claim 1, characterized in that, The weighted fusion feature is specifically as follows: ； In the formula, For weighted fusion features, Indicates local features of the image. Representing the characteristics of meteorological time series vectors, This represents the dynamic weights corresponding to local features of the image. The dynamic weights represent the characteristics of meteorological time series vectors.

4. A method for predicting ecological carbon sink potential based on a fusion attention mechanism according to any one of claims 1-3, characterized in that, Steps S2-S5 are executed using the constructed CNN-BiLSTM-Attention hybrid model, which is obtained through the following method: Construct training samples, which include spatiotemporally aligned historical meteorological data, historical reflectance images, and historical carbon sink data; The historical time-series reflectance images and the historical meteorological data for a specified period are input into a preset hybrid network model for model training to obtain the predicted carbon sink data for the specified period. The loss function of the hybrid network model is calculated based on the predicted carbon sink data and the historical carbon sink data for the corresponding period until the training termination condition is met. The hybrid network model that meets the training termination condition is determined as the CNN-BiLSTM-Attention hybrid model.

5. The method for predicting ecological carbon sink potential based on a fusion attention mechanism according to claim 4, characterized in that, The loss function is specifically as follows: ； In the formula, Represents historical carbon sequestration data. To predict carbon sink data, This represents the number of samples.

6. The method for predicting ecological carbon sink potential based on a fusion attention mechanism according to claim 5, characterized in that, The historical carbon sequestration data was obtained through the following methods: Optical reflectance images were acquired using Sentinel-2A, leaf area index was inverted using a radiative transfer model, and historical carbon sink data were estimated using a two-leaf light energy utilization model.

7. An ecological carbon sink potential prediction system based on a fusion attention mechanism, characterized in that, include: The sample data acquisition module is used to acquire reflectance images of the scene to be evaluated, as well as meteorological variable data of the scene to be evaluated in the future evaluation period. The data feature extraction module is used to extract local image features from the reflectance image using CNN and to extract meteorological variable features from the meteorological variable data using BiLSTM. The meteorological feature weighting module is used to perform weighted processing on the meteorological variable features based on the attention mechanism, and to calculate the meteorological time series vector corresponding to the meteorological variable features with fused attention weights; The feature fusion module is used to calculate the dynamic weights of the local image features and the meteorological time-series vector features based on a dynamic weight allocation mechanism, and obtain the weighted fusion features through the corresponding dynamic weights; the dynamic weight allocation mechanism is specifically as follows: ； ； In the formula, This represents the dynamic weights corresponding to local features of the image. Dynamic weights representing the characteristics of meteorological time-series vectors. , They represent the learnable parameters, The quality of remote sensing images is represented by the proportion of cloud cover, and the reflectance image is the reflectance image. ; To assess the reliability of meteorological variable data, model uncertainty in the CMIP6 dataset is used as a representation. ; For temporal integrity, it is characterized by the data missing rate, and ; The ecological carbon sink calculation module is used to input the weighted fusion features into the fully connected layer for processing, and to map the feature variables to the prediction space to obtain the predicted ecological carbon sink data of the scenario to be evaluated during the evaluation period.

8. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the ecological carbon sink potential prediction method based on the fusion attention mechanism as described in any one of claims 1-6.

9. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions that cause the computer to execute the ecological carbon sink potential prediction method based on the fusion attention mechanism according to any one of claims 1-6.

Images