A forestry ecological monitoring and evaluation method based on data analysis
By combining adaptive Kalman filtering and spatiotemporal attention LSTM network models, the problems of multi-source data fusion and ecosystem coupling relationship identification are solved, achieving high-precision forestry ecological monitoring and assessment and future trend prediction, and supporting proactive prevention and control of forestry ecosystems.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SICHUAN FORESTRY RES INST (SICHUAN FORESTRY IND RES & DESIGN INST)
- Filing Date
- 2026-05-12
- Publication Date
- 2026-06-09
AI Technical Summary
Existing technologies cannot achieve high-precision, dynamic, and forward-looking forestry ecological monitoring and assessment. They cannot effectively integrate multi-source heterogeneous data, ignore the inherent coupling relationships of the ecosystem, cannot identify key weaknesses that restrict the health of the system, and are difficult to predict future ecological health trends.
Using a data analysis-based approach, an adaptive Kalman filter is used to optimally fuse multi-source data, constructing a four-dimensional coupled assessment system of ecological structure, function, stress, and resilience. A spatiotemporal attention LSTM network model is then used for prediction, outputting a comprehensive health index of the forestry ecosystem.
It has achieved high-precision forestry ecological monitoring and assessment, which can identify key shortcomings, predict ecological trends in advance, and realize the transformation from passive management to proactive prevention and control.
Smart Images
Figure CN122173848A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of forestry ecological management, and specifically to a forestry ecological monitoring and assessment method based on data analysis. Background Technology
[0002] Forest ecosystems are a core component of terrestrial ecosystems, undertaking key ecological service functions such as carbon sequestration and oxygen release, water conservation, soil retention, and biodiversity maintenance, and serving as an important barrier to ecological security. With the dual impacts of climate change and human activities, forest ecosystems face multiple challenges, including forest degradation, biodiversity loss, and frequent disasters. Therefore, accurate, efficient, and dynamic monitoring and assessment of forest ecosystems have become a core requirement for forest ecosystem protection and management.
[0003] Currently, there is no publicly available literature on a comprehensive monitoring and analysis scheme for regional forestry ecology. Furthermore, most existing technologies assess forestry ecological conditions using a single indicator, resulting in the following significant technical shortcomings: 1. Existing forestry ecological monitoring data includes multi-source heterogeneous data such as remote sensing images, ground plots, meteorological and hydrological data, and soil monitoring data. These data suffer from significant noise interference and insufficient utilization of data complementarity. Traditional methods for fusing heterogeneous data have low accuracy and are difficult to support high-precision assessment.
[0004] 2. It ignores the inherent coupling relationship of "structure-function-stress-resilience" in forestry ecosystems, which fails to fully reflect the health status of the ecosystem and makes it even more difficult to identify the key weaknesses that restrict the health of the system.
[0005] 3. Existing assessments are mostly static and retrospective, which cannot effectively capture the nonlinear and non-stationary evolution patterns of forestry ecosystems, cannot predict future ecological health trends, and are difficult to achieve forward-looking risk warnings and proactive prevention and control. Summary of the Invention
[0006] To address the shortcomings of existing technologies, this invention provides a data analysis-based method for forestry ecological monitoring and assessment, enabling high-precision, dynamic, and forward-looking monitoring and assessment of forestry ecosystems.
[0007] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: A data analysis-based method for forestry ecological monitoring and assessment is provided, comprising: Step S1: Collect remote sensing image data, ground monitoring data, meteorological and hydrological data, and stress disturbance data of the forestry monitoring area. Construct a historical continuous sample data with an annual scale as the sample length to obtain... N One sample data; Step S2: Remove outliers from each sample data, fill in missing values, standardize the samples, and perform optimal fusion of multi-source data based on adaptive Kalman filtering to output standardized multi-source data fusion samples. Step S3: Calculate the ecological structure dimension index, ecological function dimension index, ecological stress dimension index, and ecological resilience dimension index of the forestry monitoring area based on remote sensing image data, ground monitoring data, meteorological and hydrological data, and stress disturbance data; Step S4: Standardize the ecological structure dimension indicators, ecological function dimension indicators, ecological stress dimension indicators and ecological resilience dimension indicators to obtain a standardized indicator matrix, calculate the objective weight of each standardized indicator, and calculate the comprehensive health index of the forestry ecosystem on an annual scale based on the objective weight. Step S5: Construct a continuous standardized multi-source data fusion time series vector using standardized multi-source data fusion samples at the annual scale. Use the standardized multi-source data fusion time series vector as input and the comprehensive health index of the forestry ecosystem at the annual scale as output to train the spatiotemporal attention LSTM network model and output the converged spatiotemporal attention LSTM network model. Step S6: Input the time-series vector of standardized multi-source data from consecutive historical year scales prior to the current year scale into a converged spatiotemporal attention LSTM network model, and output the comprehensive health index of the forest ecosystem predicted for several future year scales to assess the future forest ecological risks in the forest monitoring area.
[0008] Further, step S2 includes: Step S21: Identify and remove outliers in each sample data; fill in missing values in the sample data using an improved linear interpolation method with periodic correction; and then standardize the sample data to obtain standardized multi-source data samples. Step S22: Construct standardized state equations and observation equations for multi-source data samples, perform optimal fusion of multi-source data based on adaptive Kalman filtering, and output standardized multi-source data fusion samples.
[0009] Furthermore, the method for identifying outliers in the sample data is as follows: ; in, i For data types, n For sample number, For the first n The first sample i Such data values, For the first in the sample i The mean of the data. For the first in the sample i The standard deviation of the data For the first i Grubbs' statistic for this type of data; Set the Grubbs threshold ,like Then determine the data value If it is an outlier, then determine the data value. This is a normal value; The improved linear interpolation method with periodic correction for filling in missing values in sample data is as follows: ; in, To analyze the time in the sample data t Data i interpolation, For time t The nearest non-missing data to the missing data i ; The time steps before and after interpolation of missing data. Data for the same period on an annual scale i The mean correction term, These are historical data for the same period on an annual scale. i The mean.
[0010] Furthermore, the method for optimal fusion of multi-source data based on adaptive Kalman filtering is as follows: Equations of state: ; Observation equation: ; Adaptive weight calculation: ; Adaptive update of process noise covariance: ; in, For time t The state vector, A Here is the state transition matrix. For time t -1 state vector For process noise, j The data source numbering in the standardized multi-source data sample. For data source j Data in time t The standardized value, For data source j The observation matrix For data source j Observation noise, For data source j Adaptive weights, For data sourcej The variance of observation noise, The process noise covariance matrix is... Forgetting factor, For Kalman gain, For the new information sequence, J This represents the number of data sources.
[0011] Furthermore, the ecological structure dimension indicators include the tree species richness index. forest stand diameter class uniformity and landscape fragmentation index FN ; Based on the number of tree species in each quadrat from the ground monitoring data Total number of trees Calculate the tree species richness index of the quadrats Then calculate the tree species richness index of the forestry monitoring area. ; ; in, m Number the sample plots. M The number of quadrats in the forestry monitoring area; Based on the diameter at breast height (DBH) data of individual trees within each quadrat from the ground monitoring data, the evenness of stand diameter class structure in the forestry monitoring area was calculated. ; ; in, u The set of diameter at breast height (DBH) grade numbers, U The number of diaphragm size grades set, For the first sample plot u The number of individual trees of each diameter at breast height (DBH) class. For the first m Evenness of stand diameter class structure in each quadrat; Based on the total number of forest landscape patches in remote sensing image data Calculate the landscape fragmentation index of the forestry monitoring area FN ; ; in, This represents the minimum total number of grid cells in the forestry monitoring area.
[0012] Furthermore, the ecological function dimension indicators include net primary productivity of vegetation. NPP Water conservation capacity Q wr and soil retention Q sr Net primary productivity of vegetation NPP Calculations based on remote sensing image data: ; in, For pixels x In time t The corresponding net primary productivity of vegetation, For pixels x In time t Absorbed photosynthetically active radiation, For pixels x In time t Actual light energy utilization rate Total solar radiation. The proportion of photosynthetically active radiation absorbed. To maximize the utilization of light energy by trees, These are the first temperature stress coefficient and the second temperature stress coefficient, respectively. This is the water stress coefficient. This is the soil nutrient stress coefficient; Soil nutrient stress coefficient Calculated based on soil organic matter and total nitrogen content; ; in, These are the single-factor nutrient stress coefficients for nitrogen and phosphorus in soil. Assuming nitrogen and phosphorus stress weights, TN Soil total nitrogen content, TP The total phosphorus content of the soil, , The slope coefficients are related to nitrogen and phosphorus. , For the midpoint parameters related to nitrogen and phosphorus, These are the minimum and maximum values of total nitrogen content, respectively. These represent the minimum and maximum values of total phosphorus content, respectively.
[0013] Furthermore, the indicators for the ecological stress dimension include the natural stress index, the anthropogenic disturbance index, and the comprehensive stress index; Natural stress index , k The type of natural stress, The standardized stress intensity of natural stress. Weights for natural stress types; Human interference index , l This is a type of human interference. Standardized interference intensity for human-induced interference. Weights for types of human interference; Comprehensive Stress Index , The weights of the natural stress index.
[0014] Furthermore, the indicators for ecological resilience include the vegetation resilience index and the ecosystem resilience index; Vegetation resilience index , These are the recovery period after disturbance, the peak value of disturbance, and the normalized vegetation index before disturbance, respectively. Ecosystem resilience index , These are the weights for vegetation resilience index and stand diameter class uniformity, respectively.
[0015] Further, step S4 includes: Step S41: Standardize the ecological structure dimension indicators, ecological function dimension indicators, ecological stress dimension indicators, and ecological resilience dimension indicators to obtain a standardized indicator matrix; Step S42: Calculate the objective weight for each standardized indicator in the standardized indicator matrix. ; ; in, z The numbering of the standardized indicator. Numbered on a historical year scale. The quantity is on a historical year scale. For the first The first historical year scale z A standardized indicator, For the first The first historical year scale z The proportion of each standardized indicator For the first z The entropy value of a standardized indicator, For the first z The coefficient of variation of each standardized indicator Z The number of standardized indicators; Step S43: Utilize objective weights Calculate the comprehensive health index of forest ecosystems on an annual scale ; .
[0016] Furthermore, the data processing method for the spatiotemporal attention LSTM network model is as follows: Forgotten Gate: ; Input Gate: ; Cell status update: ; Output gate: ; Hidden state output: ; in, The outputs of the forget gate, input gate, and output gate are respectively. It is the sigmoid activation function. In cellular state, These are the weight matrices for the forget gate, input gate, output gate, and cell state update, respectively. These are the bias terms for the forget gate, input gate, output gate, and cell state update, respectively. It is the hyperbolic tangent function. This is the hidden state based on the previous year's scale. To standardize the time-series vector of multi-source data fusion, This represents the cell state on a previous year's timeline. Higher weights are given to key time points and key sample data that influence ecosystem evolution, highlighting the contribution of key information; Spatiotemporal attention weights ; in, For activation function, For the weight vector, This represents the decoded hidden state at the previous year's scale. This represents the hidden state on an annual scale. These are the weight matrices for the hidden state and the decoded hidden state, respectively. This is the bias term for the spatiotemporal attention weights; Spatiotemporal attention weights Hidden state at the annual scale Generate context vectors ; ; context vector The input layer is a fully connected layer, and the output is the predicted comprehensive health index of the forestry ecosystem. ; ; in, For fully connected layer activation functions, These are the weight matrix and bias terms of the activation function of the fully connected layer.
[0017] The beneficial effects of this invention are as follows: It solves the problems of mismatched multi-source data and high noise interference, providing high-quality basic data for high-precision assessment. A four-dimensional coupled assessment system of "structure-function-stress-resilience" is constructed, comprehensively covering the core characteristics of forestry ecosystems. By quantifying the interactions within the system through coupling, it can accurately identify key weaknesses restricting ecological health. A spatiotemporal attention LSTM prediction model is proposed, which can predict ecological trends and identify potential risks in advance, realizing a shift from passive management to proactive prevention and control. Attached Figure Description
[0018] Figure 1 This is a flowchart of a data analysis-based forestry ecological monitoring and assessment method. Detailed Implementation
[0019] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.
[0020] like Figure 1 As shown, a data analysis-based forestry ecological monitoring and assessment method includes: Step S1: Collect remote sensing image data, ground monitoring data, meteorological and hydrological data, and stress disturbance data of the forestry monitoring area. Construct a historical continuous sample data with an annual scale as the sample length to obtain... N One sample data; The remote sensing image data in this embodiment includes the Normalized Difference Vegetation Index (NDVI). NDVI Leaf area index LAI Data includes land cover data, surface temperature data, etc., with a spatial resolution of 10m to 1km and a temporal resolution of daily to annual scales.
[0021] The ground monitoring data in this embodiment were obtained through a survey of fixed vegetation quadrats in the forestry monitoring area, including data such as tree species composition, diameter at breast height (DBH), tree height, canopy closure, biomass, understory vegetation coverage, and soil physicochemical properties (organic matter, pH, water content, total nitrogen / phosphorus / potassium).
[0022] The meteorological and hydrological data in this embodiment include time-series data such as precipitation, temperature, relative humidity, wind speed, and sunshine duration from meteorological stations in the forestry monitoring area and surrounding areas, and time-series data such as runoff and water level from hydrological stations.
[0023] The stress interference data in this embodiment includes natural and human interference data such as forest fires, pests and diseases, logging, afforestation, engineering construction, and grazing. For example, the area of forest damage caused by forest fires, pests and diseases, logging, afforestation, and engineering construction, the duration of grazing in the forestry monitoring area, and the area involved by the grazing tracks.
[0024] Step S2: Remove outliers from each sample data, fill in missing values, standardize the samples, and perform optimal fusion of multi-source data based on adaptive Kalman filtering to output standardized multi-source data fusion samples.
[0025] Step S2 specifically includes: Step S21: Identify and remove outliers in each sample data; fill in missing values in the sample data using an improved linear interpolation method with periodic correction; and then standardize the sample data to obtain standardized multi-source data samples. The method for identifying outliers in sample data is as follows: ; in, i For data types, n For sample number, For the first n The first sample i Such data values, For the first in the sample i The mean of the data. For the first in the sample i The standard deviation of the data For the first i Grubbs' statistic for this type of data; Set the Grubbs threshold ,like Then determine the data value If it is an outlier, then determine the data value. This is a normal value; This invention uses the Grubbs criterion to screen outliers, which is more applicable to small sample data. It can accurately remove outliers while retaining effective ecological characteristic information, thus avoiding the interference of bad data on subsequent fusion and evaluation results.
[0026] The improved linear interpolation method with periodic correction for filling in missing values in sample data is as follows: ; in, To analyze the time in the sample data t Data i interpolation, For time tThe nearest non-missing data to the missing data i ; The time steps before and after interpolation of missing data. Data for the same period on an annual scale i The mean correction term, These are historical data for the same period on an annual scale. i The mean.
[0027] Traditional linear interpolation only considers the linear changes in data before and after the data point, without taking into account the seasonal periodicity of vegetation growth and meteorological factors. This invention addresses the seasonal and periodic characteristics of forestry ecological data, filling in missing values during the data collection process to ensure the integrity of time-series data. At the annual scale, it incorporates a correction term based on the average value of the same period over many years, which can accurately fit the periodic variation patterns of relevant data and effectively improve interpolation accuracy.
[0028] The method for standardizing sample data is as follows: For the positive data in the sample data: Positive data refers to data with higher values indicating better forestry ecological conditions, such as canopy closure, biomass, and understory vegetation coverage. For negative data in the sample data: Negative data refers to data where the larger the value, the worse the forestry ecological condition, such as data related to stress and disturbance.
[0029] For the first n Data from each sample i The standardized value, n The sample data number, For the data in the sample data i The maximum and minimum values; By performing dimensionless processing on the sample data, standardized values are obtained, eliminating differences in dimensions and orders of magnitude. After data standardization, the value range is mapped to... Inside.
[0030] Step S22: Construct standardized state equations and observation equations for multi-source data samples, perform optimal fusion of multi-source data based on adaptive Kalman filtering, and output standardized multi-source data fusion samples.
[0031] Equations of state: ; Observation equation: ; Adaptive weight calculation: ; Adaptive update of process noise covariance: ; in, For timet The state vector, A Here is the state transition matrix. For time t -1 state vector For process noise, j The data source numbering in the standardized multi-source data sample. For data source j Data in time t The standardized value, For data source j The observation matrix For data source j Observation noise, For data source j Adaptive weights, For data source j The variance of observation noise, The process noise covariance matrix is... The forgetting factor (values range from 0.9 to 0.95). For Kalman gain, For the new information sequence, J Number of data sources; This method calculates adaptive weights by observing noise variance, assigning higher weights to high-precision data, and adaptively updates process noise covariance to track the nonlinear and non-stationary state changes of the forestry ecosystem in real time, which greatly improves the accuracy of data fusion from different data sources.
[0032] Step S3: Calculate the ecological structure dimension index, ecological function dimension index, ecological stress dimension index, and ecological resilience dimension index of the forestry monitoring area based on remote sensing image data, ground monitoring data, meteorological and hydrological data, and stress disturbance data; The ecological structure dimension indicators in this embodiment include the tree species richness index. forest stand diameter class uniformity and landscape fragmentation index FN .
[0033] Based on the number of tree species in each quadrat from the ground monitoring data Total number of trees Calculate the tree species richness index of the quadrats Then calculate the tree species richness index of the forestry monitoring area. ; ; in, m Number the sample plots. M The number of quadrats in the forestry monitoring area; the species diversity of the forest stand is quantified by the tree species richness index, reflecting the community's resistance to disturbance.
[0034] Based on the diameter at breast height (DBH) data of individual trees within each quadrat from the ground monitoring data, the evenness of stand diameter class structure in the forestry monitoring area was calculated. ; ; in, u The set of diameter at breast height (DBH) grade numbers, U The number of diaphragm size grades set, For the first sample plot u The number of individual trees of each diameter at breast height (DBH) class. For the first m The evenness of stand diameter class structure in each quadrat; the evenness of stand diameter class structure is used to quantify the rationality of stand age structure and reflect the natural regeneration capacity of the stand.
[0035] Based on the total number of forest landscape patches in remote sensing image data Calculate the landscape fragmentation index of the forestry monitoring area FN ; ; in, The minimum total number of grid cells in the forestry monitoring area; the spatial connectivity of the forest landscape is quantified by the landscape fragmentation index, reflecting the degree of damage to the landscape structure caused by human disturbance.
[0036] The ecological function dimension indicators in this embodiment include vegetation net primary productivity. NPP Water conservation capacity Q wr and soil retention Q sr .
[0037] Water conservation capacity Q wr Soil retention capacity was calculated using the water balance equation and the RUSLE model. Q sr .
[0038] Net primary productivity of vegetation NPP By improving the CASA model calculation and incorporating a soil nutrient stress coefficient into the traditional CASA model, the productivity and carbon sequestration capacity of forest vegetation are accurately simulated, and the net primary productivity of vegetation is obtained. NPP Calculations based on remote sensing image data: Net primary productivity of vegetation NPP Take the net primary productivity of vegetation on an annual scale. The average value.
[0039] ; in, For pixelsx In time t The corresponding net primary productivity of vegetation, For pixels x In time t Absorbed photosynthetically active radiation, For pixels x In time t Actual light energy utilization rate Total solar radiation. The proportion of photosynthetically active radiation absorbed. To maximize the utilization of light energy by trees, These are the first temperature stress coefficient and the second temperature stress coefficient, respectively. This is the water stress coefficient. This is the soil nutrient stress coefficient; 0.5 represents the percentage of photosynthetically active radiation available to vegetation and the soil nutrient stress coefficient. Calculated based on soil organic matter and total nitrogen content; ; in, These are the single-factor nutrient stress coefficients for nitrogen and phosphorus in soil. Nitrogen and phosphorus stress weights are commonly used in forestry. , TN The total nitrogen content of the soil (0~20cm topsoil). TP The total phosphorus content of the soil, , The slope coefficients related to nitrogen and phosphorus are obtained through sigmoid membership functions, which are commonly used in forestry. , , , For the midpoint parameters related to nitrogen and phosphorus, These are the minimum and maximum values of total nitrogen content, respectively. These are the minimum and maximum values of total phosphorus content, respectively. Soil nutrient stress coefficient The value range is [0,1], which represents the degree to which insufficient supply of soil nutrients such as nitrogen and phosphorus inhibits the light energy utilization rate of vegetation. Sufficient nutrients, no stress. Nutrient deficiency severely inhibits growth. Nitrogen and phosphorus single-factor nutrient stress coefficients. The membership function is obtained by using a positive S-shaped fuzzy membership function (which conforms to plant growth: rapid rise in low nutrients, stable growth in medium nutrients, and saturation in high nutrients).
[0040] Indicates the lower limit of nitrogen deficiency and the upper limit of nitrogen saturation, typical of forestry: =0.5g / kg, =1.8g / kg; This indicates the lower limit of phosphorus deficiency and the upper limit of phosphorus saturation, typical of forestry: =0.2g / kg, =0.8g / kg; slope coefficient , Used to control the curve shape of the positive S-shaped fuzzy membership function, controlling the steepness of the curve's ascent; the larger the value, the steeper the curve. Midpoint parameter... , The nutrient value is the value when the single-factor nutrient stress coefficient is equal to 0.5. In this example, the average value of the lower limit of nitrogen deficiency and the upper limit of saturation (lower limit of phosphorus deficiency and upper limit of saturation) is taken.
[0041] This invention introduces nutrients such as nitrogen and phosphorus, which is more in line with the actual growth of forests (especially plantations, secondary forests, and barren mountainous areas). The S-shaped membership function is reasonable: low nutrients: stress increases sharply (growth is restricted); medium nutrients: slow saturation (approaching the optimum); high nutrients: no longer respond (luxury absorption). It conforms to the Michaelis-Menten growth curve of trees and is more scientific than linear and piecewise functions.
[0042] This invention incorporates a soil nutrient stress coefficient into the traditional CASA model, which better reflects the actual growth patterns of forest ecosystems, and the average relative error of vegetation net primary productivity simulation is lower than that of the traditional model.
[0043] The ecological stress dimension indicators in this embodiment include the natural stress index, the anthropogenic disturbance index, and the comprehensive stress index; Natural stress index , k The type of natural stress, The standardized stress intensity of natural stress. Weights are assigned to the types of natural stresses, including fire, pests and diseases, and meteorological disasters; the negative impacts of natural disturbances on ecosystems are quantified; and stress intensity is standardized. The percentage of the forestry monitoring area affected by fire, pests and diseases, and meteorological disasters is taken. In this embodiment, the weight of fire stress is 0.3, the weight of pests and diseases stress is 0.3, and the weight of meteorological disaster stress is 0.4.
[0044] Human interference index , l This is a type of human interference. Standardized interference intensity for human-induced interference. The weights represent the types of human-induced disturbances; human-induced disturbances include logging, construction, grazing, and farmland expansion. This embodiment standardizes the disturbance intensity. The area proportions of logging, engineering construction, grazing, and farmland expansion within the forestry monitoring area are taken, with each of these factors having a weight of 0.25.
[0045] Comprehensive Stress Index , The weights of the natural stress index comprehensively reflect the total external stresses experienced by the ecosystem. In this embodiment, we take... .
[0046] The ecological resilience dimension indicators in this embodiment include the vegetation resilience index and the ecosystem resilience index; Vegetation resilience index , These are the recovery period after disturbance, the peak value of disturbance, and the normalized vegetation index before disturbance, respectively. Ecosystem resilience index , These are the weights for the vegetation resilience index and the evenness of stand diameter class structure, respectively. In this embodiment, we take... .
[0047] Step S4: Standardize the ecological structure dimension indicators, ecological function dimension indicators, ecological stress dimension indicators, and ecological resilience dimension indicators to obtain a standardized indicator matrix. Calculate the objective weight of each standardized indicator and calculate the comprehensive health index of the forestry ecosystem on an annual scale based on the objective weight.
[0048] Step S4 specifically includes: Step S41: Standardize the ecological structure dimension indicators, ecological function dimension indicators, ecological stress dimension indicators, and ecological resilience dimension indicators to obtain a standardized indicator matrix; Step S42: Calculate the objective weight for each standardized indicator in the standardized indicator matrix. ; ; in, z The numbering of the standardized indicator. Numbered on a historical year scale. The quantity is on a historical year scale. For the first The first historical year scale z A standardized indicator, For the first The first historical year scale z The proportion of each standardized indicator For the first z The entropy value of a standardized indicator, For the first z The coefficient of variation of each standardized indicator Z The number of standardized indicators; The objective weights calculated in this invention incorporate correction coefficients to avoid the influence of zero and extreme values, fully utilize the information inherent in the indicator data, avoid human bias in purely subjective weighting, and ensure the objectivity of the weights.
[0049] Step S43: Utilize objective weights Calculate the comprehensive health index of forest ecosystems on an annual scale ; ; Forest ecosystems are complex coupled systems, with four dimensions—structure, function, stress, and resilience—interacting and constraining each other. Only when these four dimensions develop in a coordinated manner can forest ecosystems maintain health and stability. Traditional assessments only calculate single indicators, neglecting the coupling and coordination relationships within the system and failing to identify key weaknesses that hinder health. This new approach breaks through the limitations of traditional single-indicator assessments, enabling quantitative analysis of the coupling relationships within the system. It can accurately identify key dimensions and weak indicators that restrict ecosystem health, providing targeted objectives for ecological restoration.
[0050] Step S5: Construct a continuous standardized multi-source data fusion time series vector using standardized multi-source data fusion samples at the annual scale. Standardized multi-source data fusion time series vector As an input, the comprehensive health index of forestry ecosystems on an annual scale As output, the spatiotemporal attention LSTM network model is trained, and the converged spatiotemporal attention LSTM network model is output. The data processing method for the spatiotemporal attention LSTM network model is as follows: Forgotten Gate: ; Input Gate: ; Cell status update: ; Output gate: ; Hidden state output: ; in, The outputs of the forget gate, input gate, and output gate are respectively. It is the sigmoid activation function. In cellular state, These are the weight matrices for the forget gate, input gate, output gate, and cell state update, respectively. These are the bias terms for the forget gate, input gate, output gate, and cell state update, respectively. It is the hyperbolic tangent function. This is the hidden state based on the previous year's scale. This represents the cell state on a previous year's timeline. The LSTM network model effectively captures the dependencies of long-term time-series data through a gating mechanism, solving the gradient vanishing problem of traditional prediction models and making it suitable for the long-term, non-linear time-series characteristics of forestry ecosystems.
[0051] We assign higher weights to key time points and key sample data that influence ecosystem evolution, highlight the contribution of key information, and introduce spatiotemporal attention weights. Spatiotemporal attention weights ; in, For activation function, For the weight vector, This represents the decoded hidden state at the previous year's scale. This represents the hidden state on an annual scale. These are the weight matrices for the hidden state and the decoded hidden state, respectively. This is the bias term for the spatiotemporal attention weights; Spatiotemporal attention weights Hidden state at the annual scale Generate context vectors ; ; context vector The input layer is a fully connected layer, and the output is the predicted comprehensive health index of the forestry ecosystem. ; ; in, For fully connected layer activation functions, These are the weight matrix and bias terms of the activation function of the fully connected layer.
[0052] During the training of the spatiotemporal attention LSTM network model, the comprehensive health index of the forestry ecosystem is compared and predicted. and the actual input of the comprehensive health index of the forestry ecosystem The error between the two is used to determine whether the spatiotemporal attention LSTM network model has converged. When the error reaches the convergence requirement, the converged spatiotemporal attention LSTM network model is output.
[0053] This invention solves the problem of low fitting accuracy of traditional prediction models for nonlinear, nonstationary, and heterogeneous data. It can achieve high-precision prediction of the comprehensive health index of forest ecosystems in forestry monitoring areas for at least 1 to 5 years in the future. It integrates time-series dependence and key feature weighting to predict the future evolution trend of ecosystems in advance, realizing the transformation from passive management to active prevention. Step S6: Input the fused time-series vectors of standardized multi-source data from consecutive historical year scales prior to the current year scale into a converged spatiotemporal attention LSTM network model, and output the future... Annual-scale prediction of the comprehensive health index of forestry ecosystems To assess future forestry ecological risks in forestry monitoring areas.
[0054] The comprehensive health index of the forest ecosystem ranges from [0,1], with a higher value indicating a better forest ecosystem health. This embodiment can be based on the predicted comprehensive health index of the forest ecosystem. The forestry ecological health status of the forestry monitoring area is divided into five levels: Excellent [0.8~1.0], Good [0.6~0.8], Average [0.4~0.6], Poor [0.2~0.4], and Very Poor [0~0.2]. If in the future... Annual-scale prediction of the comprehensive health index of forestry ecosystems A decrease year by year indicates that the forestry ecology in the forestry monitoring area is deteriorating year by year.
Claims
1. A forestry ecological monitoring and assessment method based on data analysis, characterized in that, include: Step S1: Collect remote sensing image data, ground monitoring data, meteorological and hydrological data, and stress disturbance data of the forestry monitoring area. Construct a historical continuous sample data with an annual scale as the sample length to obtain... N One sample data; Step S2: Remove outliers from each sample data, fill in missing values, standardize the samples, and perform optimal fusion of multi-source data based on adaptive Kalman filtering to output standardized multi-source data fusion samples. Step S3: Calculate the ecological structure dimension index, ecological function dimension index, ecological stress dimension index, and ecological resilience dimension index of the forestry monitoring area based on remote sensing image data, ground monitoring data, meteorological and hydrological data, and stress disturbance data; Step S4: Standardize the ecological structure dimension indicators, ecological function dimension indicators, ecological stress dimension indicators and ecological resilience dimension indicators to obtain a standardized indicator matrix, calculate the objective weight of each standardized indicator, and calculate the comprehensive health index of the forestry ecosystem on an annual scale based on the objective weight. Step S5: Construct a continuous standardized multi-source data fusion time series vector using standardized multi-source data fusion samples at the annual scale. Use the standardized multi-source data fusion time series vector as input and the comprehensive health index of the forestry ecosystem at the annual scale as output to train the spatiotemporal attention LSTM network model and output the converged spatiotemporal attention LSTM network model. Step S6: Input the time-series vector of standardized multi-source data from consecutive historical year scales prior to the current year scale into a converged spatiotemporal attention LSTM network model, and output the comprehensive health index of the forest ecosystem predicted for several future year scales to assess the future forest ecological risks in the forest monitoring area.
2. The forestry ecological monitoring and assessment method based on data analysis according to claim 1, characterized in that, Step S2 includes: Step S21: Identify and remove outliers in each sample data; fill in missing values in the sample data using an improved linear interpolation method with periodic correction; and then standardize the sample data to obtain standardized multi-source data samples. Step S22: Construct standardized state equations and observation equations for multi-source data samples, perform optimal fusion of multi-source data based on adaptive Kalman filtering, and output standardized multi-source data fusion samples.
3. The forestry ecological monitoring and assessment method based on data analysis according to claim 2, characterized in that, The method for identifying outliers in the sample data is as follows: ； in, i For data types, n For sample number, For the first n The first sample i Such data values, For the first in the sample i The mean of the data. For the first in the sample i The standard deviation of the data For the first i Grubbs' statistic for this type of data; Set the Grubbs threshold ,like Then determine the data value If it is an outlier, then determine the data value. This is a normal value; The improved linear interpolation method with periodic correction for filling missing values in sample data is as follows: ； in, To analyze the time in the sample data t Data i interpolation, For time t The nearest non-missing data to the missing data i ; The time steps before and after interpolation of missing data. Data for the same period on an annual scale i Mean correction term, These are historical data for the same period on an annual scale. i The mean.
4. The forestry ecological monitoring and assessment method based on data analysis according to claim 2, characterized in that, The method for optimal fusion of multi-source data based on adaptive Kalman filtering is as follows: Equations of state: ; Observation equation: ; Adaptive weight calculation: ; Adaptive update of process noise covariance: ; in, For time t The state vector, A Here is the state transition matrix. For time t -1 state vector For process noise, j The data source numbering in the standardized multi-source data sample. For data source j Data in time t The standardized value, For data source j The observation matrix For data source j Observation noise, For data source j Adaptive weights, For data source j The variance of observation noise, The process noise covariance matrix is... Forgetting factor, For Kalman gain, For the new information sequence, J This represents the number of data sources.
5. The forestry ecological monitoring and assessment method based on data analysis according to claim 1, characterized in that, The ecological structure dimension indicators include the tree species richness index. forest stand diameter class uniformity and landscape fragmentation index FN ; Based on the number of tree species in each quadrat from the ground monitoring data Total number of trees Calculate the tree species richness index of the quadrats Then calculate the tree species richness index of the forestry monitoring area. ; ； in, m Number the sample plots. M The number of quadrats in the forestry monitoring area; Based on the diameter at breast height (DBH) data of individual trees within each quadrat from the ground monitoring data, the evenness of stand diameter class structure in the forestry monitoring area was calculated. ; ； in, u The set of diameter at breast height (DBH) grade numbers, U The number of diaphragm size grades set, For the first sample plot u The number of individual trees in each diameter-at-breast-length class. For the first m Evenness of stand diameter class structure in each quadrat; Based on the total number of forest landscape patches in remote sensing image data Calculate the landscape fragmentation index of the forestry monitoring area FN ; ； in, This represents the minimum total number of grid cells in the forestry monitoring area.
6. The forestry ecological monitoring and assessment method based on data analysis according to claim 1, characterized in that, The ecological function dimension indicators include vegetation net primary productivity. NPP Water conservation capacity Q wr and soil retention Q sr Net primary productivity of vegetation NPP Calculations based on remote sensing image data: ； in, For pixels x In time t The corresponding net primary productivity of vegetation, For pixels x In time t Absorbed photosynthetically active radiation, For pixels x In time t Actual light energy utilization rate Total solar radiation. The proportion of photosynthetically active radiation absorbed. To maximize the utilization of light energy by trees, These are the first temperature stress coefficient and the second temperature stress coefficient, respectively. This is the water stress coefficient. This is the soil nutrient stress coefficient; Soil nutrient stress coefficient Calculated based on soil organic matter and total nitrogen content; ； in, These are the single-factor nutrient stress coefficients for nitrogen and phosphorus in soil. Assuming nitrogen and phosphorus stress weights, TN Soil total nitrogen content, TP The total phosphorus content of the soil, , The slope coefficients are related to nitrogen and phosphorus. , For the midpoint parameters related to nitrogen and phosphorus, These are the minimum and maximum values of total nitrogen content, respectively. These represent the minimum and maximum values of total phosphorus content, respectively.
7. The forestry ecological monitoring and assessment method based on data analysis according to claim 1, characterized in that, The ecological stress dimension indicators include the natural stress index, the anthropogenic disturbance index, and the comprehensive stress index; Natural stress index , k Types of natural stress, The standardized stress intensity of natural stress. Weights for natural stress types; Human interference index , l This is a type of human interference. Standardized interference intensity for human-induced interference. Weights for types of human interference; Comprehensive Stress Index , The weights of the natural stress index.
8. The forestry ecological monitoring and assessment method based on data analysis according to claim 1, characterized in that, The ecological resilience dimension indicators include the vegetation resilience index and the ecosystem resilience index; Vegetation resilience index , These are the recovery period after disturbance, the peak value of disturbance, and the normalized vegetation index before disturbance, respectively. Ecosystem resilience index , These are the weights for vegetation resilience index and stand diameter class uniformity, respectively.
9. The forestry ecological monitoring and assessment method based on data analysis according to claim 1, characterized in that, Step S4 includes: Step S41: Standardize the ecological structure dimension indicators, ecological function dimension indicators, ecological stress dimension indicators, and ecological resilience dimension indicators to obtain a standardized indicator matrix; Step S42: Calculate the objective weight for each standardized indicator in the standardized indicator matrix. ; ； in, z The numbering of the standardized indicator. Numbered on a historical year scale. The quantity is on a historical year scale. For the first The first historical year scale z A standardized indicator, For the first The first historical year scale z The proportion of each standardized indicator For the first z The entropy value of a standardized indicator, For the first z The coefficient of variation of each standardized indicator Z The number of standardized indicators; Step S43: Utilize objective weights Calculate the comprehensive health index of forest ecosystems on an annual scale ; 。 10. The forestry ecological monitoring and assessment method based on data analysis according to claim 1, characterized in that, The data processing method for the spatiotemporal attention LSTM network model is as follows: Forgotten Gate: ; Input Gate: ; Cell status update: ; Output gate: ; Hidden state output: ; in, The outputs of the forget gate, input gate, and output gate are respectively... It is the sigmoid activation function. In cellular state, These are the weight matrices for the forget gate, input gate, output gate, and cell state update, respectively. These are the bias terms for the forget gate, input gate, output gate, and cell state update, respectively. It is the hyperbolic tangent function. This is the hidden state based on the previous year's scale. To standardize the time-series vector of multi-source data fusion, This represents the cell state on a previous year's timeline. Higher weights are given to key time points and key sample data that influence ecosystem evolution, highlighting the contribution of key information; Spatiotemporal attention weights ; in, For activation function, For the weight vector, This represents the decoded hidden state at the previous year's scale. This represents the hidden state on an annual scale. These are the weight matrices for the hidden state and the decoded hidden state, respectively. This is the bias term for the spatiotemporal attention weights; Spatiotemporal attention weights Hidden state at the annual scale Generate context vectors ; ； context vector The input layer is a fully connected layer, and the output is the predicted comprehensive health index of the forestry ecosystem. ; ； in, For fully connected layer activation functions, This represents the weight matrix and bias terms of the activation function of the fully connected layer.