Forest resource health state evaluation method and system based on machine learning

Abstract

The application discloses a forest resource health state evaluation method and system based on machine learning, and relates to the technical field of machine learning.The method comprises the following steps: acquiring remote sensing vegetation index time series data and contemporaneous multi-source influence factor time series data of each forest patch in a target area in the same phase of consecutive years; calculating vegetation index time series stability, sustained degradation index and vegetation index recovery elasticity as degradation risk characteristics based on the vegetation index time series data; inputting the degradation risk characteristics into a degradation evaluation model to obtain a comprehensive degradation index; calculating the factor time lag matching degree of multi-source influence factors and evaluating important influence factor types for forest patches with a comprehensive degradation index exceeding a preset risk threshold; and analyzing and obtaining forest resource health state evaluation results of each forest patch according to the comprehensive degradation index and the important influence factor types.The application realizes comprehensive quantitative evaluation of forest degradation risk and accurate positioning of degradation causes.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, and specifically to a method and system for assessing the health status of forest resources based on machine learning. Background Technology

[0002] The health status of forest resources directly affects regional ecological security and carbon sequestration capacity. In recent years, due to climate change, pests and diseases, and human activities, some forest areas have experienced vegetation degradation. Accurately assessing the health status of forest land and identifying the causes of degradation are prerequisites for formulating scientific management strategies.

[0003] Traditional methods rely on ground-based sample plot surveys and manual experience-based judgment, resulting in limited coverage, long cycles, and high costs, making them unsuitable for large-scale dynamic monitoring. Existing methods based on time-series remote sensing vegetation index data assess forest degradation by analyzing interannual trends, but they still have the following shortcomings: First, they often use single trend indicators, such as linear regression slopes, ignoring multidimensional health characteristics such as vegetation fluctuation stability and resilience, leading to one-sided assessment results. Second, they fail to integrate multiple influencing factors such as meteorology, soil, and pests for synergistic attribution analysis, making it impossible to distinguish whether degradation is caused by climate fluctuations, soil degradation, or pests, thus limiting the guiding value of the assessment results for management decisions. Third, the impact of meteorological factors on vegetation growth has a lag effect; existing methods do not consider this time lag, leading to distorted assessments of the contribution of influencing factors. Therefore, there is an urgent need for a forest health assessment method that comprehensively assesses degradation risk from multiple dimensions of time-series characteristics and can accurately pinpoint the causes of degradation. Summary of the Invention

[0004] This invention addresses the technical problems in existing technologies, such as forest degradation assessment relying on a single trend indicator while ignoring multi-dimensional health characteristics, failing to integrate multi-source influencing factors for synergistic attribution analysis, and failing to consider the lag effect of meteorological factors, which leads to distorted assessment of the contribution of influencing factors. It provides a method and system for assessing the health status of forest resources based on machine learning.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows:

[0006] In a first aspect, the present invention provides a method for assessing the health status of forest resources based on machine learning, including:

[0007] The remote sensing vegetation index time series data of each forest patch in the target area in the same seasonal phase over several consecutive years are obtained, as well as the time series data of multi-source influencing factors during the same period. The multi-source influencing factors include at least meteorological factors, soil factors and pest and disease factors.

[0008] Based on the time series data of vegetation index of each forest patch, the time series stability of vegetation index, continuous degradation index and vegetation index recovery elasticity of each forest patch are calculated as degradation risk characteristics.

[0009] The degradation risk characteristics of each forest patch are input into a pre-trained degradation assessment model to obtain a comprehensive degradation index;

[0010] For forest patches whose comprehensive degradation index exceeds a preset risk threshold, the factor time lag matching degree of the multi-source influencing factors is calculated respectively, and the types of important influencing factors are evaluated and obtained based on the factor time lag matching degree and the vegetation index time series data.

[0011] Based on the comprehensive degradation index and the types of important influencing factors, the forest resource health status assessment results of each forest patch are obtained through analysis.

[0012] Secondly, this invention provides a forest resource health status assessment system based on machine learning, comprising:

[0013] The data acquisition module is used to acquire remote sensing vegetation index time series data of each forest patch in the target area in the same seasonal phase over multiple consecutive years, as well as time series data of multi-source influencing factors during the same period. The multi-source influencing factors include at least meteorological factors, soil factors, and pest and disease factors.

[0014] The risk feature module is used to calculate the vegetation index time-series stability, continuous degradation index and vegetation index recovery elasticity of each forest patch based on the vegetation index time-series data of each forest patch, as degradation risk features.

[0015] The degradation assessment module is used to input the degradation risk characteristics of each forest patch into a pre-trained degradation assessment model to obtain a comprehensive degradation index.

[0016] The factor analysis module is used to calculate the factor time lag matching degree of the multi-source influencing factors for the forest patches where the comprehensive degradation index exceeds the preset risk threshold, and to evaluate and obtain the types of important influencing factors based on the factor time lag matching degree and the vegetation index time series data.

[0017] The results output module is used to analyze and obtain the forest resource health status assessment results of each forest patch based on the comprehensive degradation index and the types of important influencing factors.

[0018] The beneficial effects of this invention are:

[0019] Compared to existing technologies, this invention first obtains time-series vegetation index data for various forest patches over several consecutive years in the same seasonal phase. It then calculates degradation risk characteristics across three dimensions: vegetation index time-series stability, persistent degradation index, and vegetation index recovery elasticity. This comprehensively characterizes the health status of forest land from three perspectives: interannual fluctuation amplitude, degradation persistence, and post-disaster recovery capacity, overcoming the limitations of single-trend indicator assessments. Secondly, a pre-trained degradation assessment model is used to perform nonlinear mapping and cross-feature fusion on degradation risk characteristics, outputting a comprehensive degradation index, thus achieving a comprehensive quantitative assessment of forest land degradation risk. Thirdly, for forest patches where the comprehensive degradation index exceeds the risk threshold, a time-lag matching mechanism is introduced. By shifting the time-series data of influencing factors, the maximum correlation coefficient with the time-series data of vegetation indices is calculated. This quantitatively assesses the time-lag matching degree of meteorological, soil, and pest / disease factors, and identifies important influencing factor types. This solves the problem of distorted assessment of influencing factor contribution caused by the lag effect of meteorological factors, achieving accurate localization of degradation causes. Finally, based on the comprehensive degradation index, degradation risk levels were classified, and degradation attribution information and management recommendations were output in combination with the types of important influencing factors, providing a scientific basis for the precise management of forest resources. Attached Figure Description

[0020] Figure 1 A flowchart illustrating the machine learning-based forest resource health status assessment method provided by this invention;

[0021] Figure 2 A logical schematic diagram of the forest land resource health status assessment method based on machine learning provided by the present invention;

[0022] Figure 3 This is a schematic diagram of the structure of the forest land resource health status assessment system based on machine learning provided by the present invention.

[0023] In the attached diagram, the components represented by each number are as follows:

[0024] Data acquisition module 11, risk characteristics module 12, degradation assessment module 13, factor analysis module 14, and results output module 15. Detailed Implementation

[0025] Example 1, as Figure 1 , Figure 2 As shown, embodiments of the present invention provide a method for assessing the health status of forest resources based on machine learning, including:

[0026] S10: Obtain remote sensing vegetation index time series data for each forest patch in the target area in the same seasonal phase over multiple consecutive years, as well as time series data of multi-source influencing factors during the same period, wherein the multi-source influencing factors include at least meteorological factors, soil factors and pest and disease factors.

[0027] In forest resource health assessment, it is necessary to obtain basic data that reflects the growth status and changing trends of forest vegetation. The Normalized Difference Vegetation Index (NDV) is a commonly used remote sensing indicator reflecting vegetation growth; a higher NAV indicates better vegetation cover and stronger growth vitality. However, vegetation growth varies significantly across seasons. Comparing image data from different seasonal phases may misinterpret seasonal factors as vegetation degradation. Therefore, it is necessary to select remote sensing images from the same seasonal phase across multiple consecutive years for vegetation index extraction to ensure data comparability across different years.

[0028] The target area refers to the administrative region or ecological functional zone where the forest health status assessment needs to be conducted. Within this target area, the forest land is divided into multiple forest patches according to its spatial distribution and boundary characteristics, with each forest patch serving as an independent assessment unit. The number of consecutive years can be set based on data availability and the assessment time span, for example, the most recent 10 years.

[0029] Meanwhile, changes in forest vegetation growth status are driven not only by the vegetation's own succession patterns but also by various external factors such as meteorology, soil, and pests and diseases. Precipitation and temperature directly affect tree growth, soil organic matter content reflects soil fertility levels, and the area affected by pests and diseases directly impacts vegetation health. These factors exhibit complex time-lag relationships and interaction effects with vegetation indices, thus requiring the simultaneous acquisition of multi-source influencing factors from the same period. Data acquisition in this step provides a data foundation for subsequent extraction of degradation risk characteristics, calculation of the comprehensive degradation index, and attribution analysis of influencing factors.

[0030] Specifically, in this embodiment, acquiring remote sensing vegetation index time-series data for each forest patch within the target area in the same seasonal phase over multiple consecutive years, as well as concurrent multi-source influencing factor time-series data, includes:

[0031] Extract the boundary range of each forest patch from the target area and obtain the vegetation type information of each forest patch;

[0032] For each forest patch, image data of the same seasonal phase in each year are selected from remote sensing images of multiple consecutive years. The mean normalized vegetation index within the boundary of the forest patch is extracted and arranged in chronological order of year to form the time series data of vegetation index of the forest patch.

[0033] The annual precipitation and average annual temperature data of each forest patch were obtained to form time series data of meteorological factors;

[0034] The annual soil organic matter content data of each forest patch were obtained to form soil factor time series data;

[0035] The annual pest and disease occurrence area data of each forest patch are obtained to form time series data of pest and disease factors.

[0036] First, the boundary extent of each forest patch is extracted from the target area, and vegetation type information for each patch is obtained. The target area refers to the administrative region or ecological functional zone where forest health status assessment is required. A forest patch refers to an independent forest unit within the target area, separated by natural features such as roads, rivers, and topography, or by man-made boundaries. Each forest patch has a relatively uniform vegetation cover type and growth conditions. Specifically, the boundary extent can be obtained through forest resource survey data, geographic information system data, or interpretation of high-resolution remote sensing imagery. Vegetation type information includes coniferous forests, broad-leaved forests, mixed coniferous and broad-leaved forests, and shrublands. Different vegetation types exhibit varying sensitivities to hydrothermal conditions and disturbance factors, serving as the basis for distinguishing the background status of different forest patches in subsequent assessments.

[0037] Secondly, for each forest patch, image data from the same seasonal phase each year is selected from remote sensing images spanning multiple consecutive years. The mean normalized vegetation index (NDI) within the boundary of the forest patch is extracted and arranged chronologically to form the time-series vegetation index data for that forest patch. Specifically, the same seasonal phase refers to selecting a period each year when vegetation growth is at a similar phenological stage, such as the peak vegetation growth period from July to August, to ensure the comparability of vegetation indices across different years. Since leaf area, chlorophyll content, and canopy structure vary significantly across different seasons, comparing images from different seasonal phases would introduce seasonal variations into the interannual variation of vegetation indices, failing to accurately reflect the long-term health status of the forest. By fixing the data extraction to the same seasonal phase each year, the influence of seasonal factors on the time-series analysis of vegetation indices can be eliminated.

[0038] The Normalized Difference Vegetation Index (NDC) is an indicator of vegetation growth status calculated based on the reflectance of near-infrared and red bands in remote sensing imagery. Healthy vegetation exhibits strong reflectivity in the near-infrared band and strong absorption in the red band. The NDC converts the spectral characteristics of vegetation into a dimensionless numerical value, ranging from -1 to 1, by calculating the difference between the two values ​​and dividing by their sum. For example, the NDC for bare soil or water bodies is close to 0 or negative, the NDC for sparse vegetation is between 0.1 and 0.3, and the NDC for dense forests is typically between 0.6 and 0.9.

[0039] For each forest patch, the normalized vegetation index (NDI) values ​​of all pixels within its boundary are extracted and the average value is calculated as the representative vegetation index value for that patch in that year. The NDI values ​​of all years are arranged chronologically to form the time-series vegetation index data for that forest patch.

[0040] Furthermore, annual precipitation and average annual temperature data for each forest patch are obtained to form time-series meteorological factor data. The meteorological factor data originates from meteorological observation stations within the target area or from meteorological grid data generated by interpolation based on meteorological station data. Annual precipitation is a crucial water supply indicator affecting tree growth and vegetation cover, while average annual temperature reflects regional heat conditions; together, they determine the vegetation growth rate and distribution pattern.

[0041] Simultaneously, annual soil organic matter content data for each forest patch were obtained to construct time-series soil factor data. Soil organic matter content is a core indicator for evaluating soil fertility, and its changes directly affect the soil's water and fertilizer retention capacity and the plant root growth environment. A decline in soil organic matter content is often one of the important indicators of forest degradation.

[0042] In addition, annual data on the area affected by pests and diseases in each forest patch were obtained to construct time-series data on pest and disease factors. The data on the area affected by pests and diseases originated from forest pest monitoring and survey data, recording information such as the types, affected areas, and severity of forest pests and diseases occurring in the target area each year. The occurrence of pests and diseases directly leads to a reduction in vegetation leaf area and a decrease in photosynthetic capacity, thereby causing a decrease in the normalized difference vegetation index (NDI).

[0043] The data obtained above can provide multi-source data support for the subsequent extraction of degradation risk characteristics and the attribution analysis of influencing factors.

[0044] S20: Based on the time series data of vegetation index of each forest patch, calculate the time series stability of vegetation index, continuous degradation index and vegetation index recovery elasticity of each forest patch as degradation risk characteristics.

[0045] Secondly, after obtaining multi-year time-series vegetation index data for each forest patch, it is necessary to extract key characteristics that can characterize the health status of the forest from this data. The annual trend of the normalized vegetation index (NVI), such as its increase or decrease, cannot fully reflect the degradation risk of the forest. Healthy forests should simultaneously possess three characteristics: small interannual fluctuations in the vegetation index (i.e., high stability); no sustained downward trend in the vegetation index (i.e., no persistent degradation); and a relatively quick recovery to normal levels after exposure to climatic stresses such as drought or low temperatures (i.e., good resilience). Therefore, this step calculates three dimensions of degradation risk characteristics from the time-series vegetation index data: vegetation index time-series stability, persistent degradation index, and vegetation index resilience.

[0046] Specifically, the temporal stability of vegetation indices reflects the ability of forest vegetation to resist external disturbances and maintain its stable state between years. Higher temporal stability indicates smaller fluctuations in vegetation indices across different years and a healthier forest ecosystem. The persistence degradation index reflects the degree of continuous decline in forest vegetation indices over a long time series; a higher persistence degradation index indicates a greater risk of continuous degradation of forest vegetation. Vegetation index resilience reflects the ability of forest vegetation to recover to pre-disturbance levels after a decline caused by external disturbances; higher resilience indicates a stronger self-repair capacity of the forest.

[0047] By using the degradation risk characteristics of the above three dimensions, the health status of forest patches can be comprehensively characterized from the perspectives of stability, degradation trend and recovery capacity. The degradation risk characteristics obtained will be used as input for subsequent degradation assessment models to calculate the comprehensive degradation index of each forest patch, and at the same time provide characteristic basis for distinguishing forest patches with different degradation patterns.

[0048] Specifically, in this embodiment, the calculation of the vegetation index time-series stability, persistent degradation index, and vegetation index recovery resilience of each forest patch based on the vegetation index time-series data of each forest patch, as degradation risk characteristics, includes:

[0049] The absolute difference between the mean normalized vegetation index of each two adjacent years of the forest patch is obtained to obtain the vegetation index change sequence.

[0050] The interannual volatility of the vegetation index is obtained by dividing the standard deviation of the vegetation index change series by the arithmetic mean of the vegetation index change series.

[0051] Based on the interannual volatility of the vegetation index, the temporal stability of the vegetation index is calculated, wherein the temporal stability of the vegetation index and the interannual volatility of the vegetation index are negatively correlated.

[0052] The number of years in which the vegetation index decreased compared to the previous year is counted in the time series data of the vegetation index, and the percentage of years with a decrease is calculated.

[0053] The longest consecutive number of years in which the vegetation index has continuously declined is statistically obtained as the longest consecutive decline year, and the proportion of the longest consecutive decline year in the vegetation index change sequence is calculated to obtain the continuous decline proportion.

[0054] The continuous degradation index is obtained by weighted summing of the percentage of consecutive declines and the percentage of years of decline.

[0055] For each year in the vegetation index time series data where the vegetation index decreased compared to the previous year, check whether the vegetation index in the following year recovers to or exceeds the preset ratio of the vegetation index of the previous year in the year of decline. If so, it is recorded as a successful recovery.

[0056] The total number of successful recoverys is counted, and the total number of successful recoverys is divided by the number of years of decline to obtain the vegetation index recovery elasticity.

[0057] The vegetation index temporal stability, the continuous degradation index, and the vegetation index recovery elasticity are combined to form the degradation risk characteristics of the forest patch.

[0058] First, the absolute difference between the mean normalized vegetation index (NDI) of forest patches in each pair of adjacent years is obtained to acquire a vegetation index variation sequence. Specifically, the mean NDI of forest patches over N consecutive years is obtained, where N is the total number of years. For each pair of adjacent years, the absolute value of the difference in the NDI is calculated, resulting in N-1 absolute differences, which are then arranged in chronological order to form a vegetation index variation sequence. This vegetation index variation sequence reflects the fluctuation range of the vegetation index between adjacent years and serves as the basis for subsequent calculations of interannual variability.

[0059] Secondly, the interannual volatility of vegetation indices is obtained by dividing the standard deviation of the vegetation index change series by the arithmetic mean of the vegetation index change series. Specifically, the formula for calculating the interannual volatility of vegetation indices is: Interannual volatility = Standard deviation of vegetation index change series / Arithmetic mean of vegetation index change series. This interannual volatility of vegetation indices reflects the relative amplitude of interannual fluctuations in vegetation indices. When the interannual volatility of vegetation indices is large, it indicates that the vegetation indices change drastically between adjacent years, and the stability of the forest ecosystem is poor.

[0060] Next, based on the interannual volatility of the vegetation index, the temporal stability of the vegetation index is calculated, where the temporal stability of the vegetation index and the interannual volatility of the vegetation index are negatively correlated. Preferably, the formula for calculating the temporal stability of the vegetation index is: Temporal stability of vegetation index = 1 / (1 + interannual volatility of vegetation index). When the interannual volatility of the vegetation index approaches 0, the temporal stability of the vegetation index approaches 1, indicating that the interannual volatility of the vegetation index is minimal and the forest land condition is stable; when the interannual volatility of the vegetation index increases, the temporal stability of the vegetation index decreases accordingly, indicating that the forest land stability deteriorates.

[0061] Furthermore, the number of years in which the vegetation index decreased compared to the previous year is counted in the time-series data of vegetation index, and the percentage of declining years is calculated. Specifically, the percentage of declining years = number of declining years / (total number of years - 1). This percentage reflects the frequency of forest vegetation decline in the overall time series; a higher percentage indicates more widespread degradation.

[0062] Next, the longest consecutive number of years of continuous decline in vegetation index is obtained as the longest consecutive decline year, and the proportion of the longest consecutive decline year in the vegetation index change series is calculated to obtain the continuous decline percentage. Continuous decline percentage = longest consecutive decline year / (total number of years - 1). This continuous decline percentage reflects whether the degradation is persistent; a higher continuous decline percentage indicates that the degradation is cumulative.

[0063] Secondly, a weighted sum is calculated from the percentage of consecutive declines and the percentage of decline years to obtain the sustained degradation index. Specifically, the sustained degradation index is calculated as follows: Sustained Degradation Index = Percentage of Declining Years × First Weighting Coefficient + Percentage of Continuous Decline × Second Weighting Coefficient, where the sum of the first and second weighting coefficients is 1. The first and second weighting coefficients are set based on the contribution of decline frequency and persistence to forest degradation risk. For example, the weighting coefficient for the percentage of decline years can be set to 0.4, and the weighting coefficient for the percentage of consecutive declines can be set to 0.6 to highlight the importance of a sustained degradation trend. It should be noted that the above weighting is the default configuration and can be adjusted according to the different characteristics of forest degradation in different regions. For example, in arid and semi-arid regions, vegetation is greatly affected by interannual fluctuations in precipitation, with more isolated decline years but faster recovery; in such cases, the weight of the percentage of consecutive declines can be appropriately reduced. The higher the final sustained degradation index, the greater the risk of forest vegetation being in a state of sustained degradation.

[0064] Furthermore, for each year in the vegetation index time series data where the vegetation index decreased compared to the previous year, it is checked whether the vegetation index in the following year recovers to or exceeds a preset percentage of the vegetation index of the previous year. If so, it is recorded as a successful recovery. The preset percentage can be set according to the natural fluctuation range of vegetation, for example, it can be set to 90%, that is, if the vegetation index in the following year recovers to more than 90% of the vegetation index of the year before the decline, it is considered a successful recovery.

[0065] Next, the total number of successful recovery attempts is counted, and this total number is divided by the number of years of decline to obtain the vegetation index resilience. Specifically, the vegetation index resilience is calculated as follows: Vegetation Index Resilience = Total Number of Successful Recoverys / Number of Years of Decline. When the number of years of decline is zero, the resilience value is 1. The higher the vegetation index resilience, the stronger the forest vegetation's ability to recover after disturbance, and the better the ecosystem's self-repair capacity.

[0066] Finally, the temporal stability of vegetation indices, the persistent degradation index, and the resilience of vegetation indices are combined to form the degradation risk characteristic of the forest patch. This degradation risk characteristic is a three-dimensional vector containing these three dimensions, which serve as input to the subsequent degradation assessment model. By combining these three dimensions, the health status of the forest patch in terms of stability, degradation trend, and recovery capacity can be comprehensively characterized.

[0067] S30: Input the degradation risk characteristics of each forest patch into a pre-trained degradation assessment model to obtain a comprehensive degradation index;

[0068] Furthermore, after obtaining the degradation risk characteristics of each forest patch, it is necessary to integrate the three dimensions of features into a comprehensive quantitative index that can comprehensively reflect the forest degradation risk. The three features—temporal stability of vegetation index, persistent degradation index, and vegetation index resilience—characterize the health status of forest land from different perspectives, but complex nonlinear interactions exist between them. For example, forest land with a high persistent degradation index but also high resilience may have a lower degradation risk than forest land with a moderate persistent degradation index but low resilience; forest land with low temporal stability but a low persistent degradation index may be affected by interannual climate fluctuations rather than truly degraded. A simple linear weighted fusion method cannot capture these nonlinear interaction effects. Therefore, this step uses a pre-trained degradation assessment model to perform nonlinear mapping and feature interaction fusion of the three degradation risk features, outputting a comprehensive degradation index.

[0069] The degradation assessment model is a machine learning model pre-trained based on historical forest patch sample data. This model takes vegetation index time-series stability, persistent degradation index, and vegetation index resilience as inputs, and outputs a comprehensive degradation index. Internally, it uses a feature cross-product layer to perform pairwise cross-products on the input features, generating stability-persistent degradation cross-terms, persistent degradation-recovery resilience cross-terms, and resilience-stability cross-terms. The original features and cross-terms are then concatenated and passed through a gated fusion layer to learn the differentiated contribution weights of each feature and cross-term to the comprehensive degradation index. Finally, the output layer maps this to the comprehensive degradation index. The final output comprehensive degradation index is a continuous numerical value ranging from 0 to 1; a higher value indicates a higher risk of forest degradation.

[0070] By using this degradation assessment model, the degradation risk characteristics of the three dimensions can be integrated into a quantifiable and rankable assessment index, providing a unified basis for subsequent risk level classification and attribution analysis of influencing factors.

[0071] Specifically, in this embodiment, the pre-training of the degradation assessment model includes:

[0072] Collect and acquire historical forest patch sample data, wherein each set of historical forest patch sample data includes the historical vegetation index time-series stability, historical vegetation index continuous degradation index and historical vegetation index recovery elasticity of a historical forest patch as sample degradation risk characteristics, and the historical comprehensive degradation index of the historical forest patch confirmed by manual investigation, which constitutes a training sample set;

[0073] A degradation assessment model is constructed, which takes the temporal stability of vegetation index, the continuous degradation index of vegetation index, and the recovery elasticity of vegetation index as inputs, and the comprehensive degradation index as output.

[0074] The degradation assessment model is trained using the training sample set, with the goal of minimizing the deviation between the output comprehensive degradation index and the historical comprehensive degradation index, until the preset convergence condition is met, thus obtaining the pre-trained degradation assessment model.

[0075] The degradation assessment model further includes:

[0076] The feature cross layer is used to receive the vegetation index time-series stability, the continuous degradation index, and the vegetation index recovery elasticity, and perform pairwise cross-products to generate stability-continuous degradation cross term, continuous degradation-recovery elasticity cross term, and recovery elasticity-stability cross term. The three original risk features and the three cross terms are concatenated into a six-dimensional cross feature vector.

[0077] The gated fusion layer is used to receive the six-dimensional cross feature vector and learn the differentiated contribution weights of each cross term to the comprehensive degradation index through a gating mechanism.

[0078] The output layer is used to receive the output of the gated fusion layer, and outputs the comprehensive degradation index after full connection mapping.

[0079] Specifically, the degradation assessment model consists of a feature crossover layer, a gated fusion layer, and an output layer.

[0080] The feature cross-layer receives three input features: vegetation index time-series stability, persistent degradation index, and vegetation index resilience, and performs pairwise cross-products. Specifically, the feature cross-layer generates three cross terms: a stability-persistent degradation cross term, which equals vegetation index time-series stability multiplied by the persistent degradation index, reflecting the joint effect of stability and degradation trend; a persistent degradation-recovery resilience cross term, which equals persistent degradation index multiplied by vegetation index resilience, reflecting the interaction between degradation trend and recovery capacity; and a resilience-stability cross term, which equals vegetation index resilience multiplied by vegetation index time-series stability, reflecting the synergistic relationship between recovery capacity and stability. The three original risk features are concatenated with the three cross terms to form a six-dimensional cross-feature vector. This concatenation operation allows the model to simultaneously utilize the independent information of the original features as well as the interaction information between features.

[0081] The gated fusion layer receives a six-dimensional cross-feature vector and learns the differentiated contribution weights of each cross-term to the comprehensive degradation index through a gating mechanism. Specifically, the core of the gating mechanism is a learnable weight allocation network, whose output is the weight coefficients of each of the six features, with the sum of the six weight coefficients being 1. During training, the gated fusion layer automatically adjusts the weight coefficients based on historical sample data, enabling the model to adaptively highlight features that contribute more to degradation risk assessment. When the vegetation index has high resilience, the gated fusion layer automatically reduces the contribution weight of the persistent degradation index to reflect the inhibitory effect of resilience on degradation risk; when the persistent degradation index is high and the resilience is low, the gated fusion layer automatically amplifies the contribution weight of the persistent degradation-recovery resilience cross-term to reflect the high-risk pattern of continuous decline and irrecoverability. Through this gated fusion mechanism, the degradation assessment model can dynamically adjust the contribution level of each feature according to the feature combinations of different forest patches.

[0082] The output layer receives the output of the gated fusion layer and outputs the comprehensive degradation index after fully connected mapping. The fully connected mapping transforms the weighted feature vector output by the gated fusion layer into a scalar value between 0 and 1 through a linear transformation and activation function; this value is the comprehensive degradation index. A comprehensive degradation index closer to 1 indicates a higher risk of degradation, while a value closer to 0 indicates a more stable health state.

[0083] Through layer-by-layer processing of feature cross-layer, gated fusion layer and output layer, the degradation assessment model can capture the nonlinear interaction relationship between the three degradation risk features, adaptively evaluate the contribution of each feature under different feature combinations, and output a comprehensive degradation index.

[0084] Specifically, the pre-training of the degradation assessment model includes the following steps:

[0085] First, historical forest patch sample data were collected. This data originated from completed forest resource surveys and ecological monitoring projects. Each set of historical forest patch sample data comprises two parts: the first part is the sample degradation risk characteristics, including the historical vegetation index time-series stability, historical vegetation index continuous degradation index, and historical vegetation index recovery elasticity. These three characteristics are calculated from the historical vegetation index time-series data using the same methods described above. The second part is the historical comprehensive degradation index of the historical forest patch, confirmed through manual investigation. Manual investigation refers to forestry technicians or ecological experts comprehensively judging the degree of degradation of the patch through methods such as ground plot surveys, tree growth status assessments, and stand structure analysis, and quantifying it as a value between 0 and 1. All historical forest patch sample data together constitute the training sample set.

[0086] Secondly, a degradation assessment model is constructed. The degradation assessment model takes the temporal stability of vegetation indices, the continuous degradation index of vegetation indices, and the resilience of vegetation indices as inputs, and the comprehensive degradation index as output. The model structure includes a feature cross-layer, a gated fusion layer, and an output layer, constructed according to the aforementioned network structure. The model parameters are randomly initialized before training begins.

[0087] Next, the degradation assessment model is trained using a training sample set, aiming to minimize the deviation between the output comprehensive degradation index and the historical comprehensive degradation index, until a preset convergence condition is met, resulting in a pre-trained degradation assessment model. Specifically, during training, the historical vegetation index time-series stability, historical vegetation index continuous degradation index, and historical vegetation index recovery elasticity of each training sample are input into the model, and the model outputs a predicted value of the comprehensive degradation index. Mean squared error is used as the loss function to measure the deviation between the predicted value and the historical comprehensive degradation index. The model parameters are iteratively updated through backpropagation and an optimizer, causing the loss function value to gradually decrease. The preset convergence condition is set based on the trend of the model's loss value on the validation set and the training efficiency requirements. For example, it can be set to ensure that the loss function value on the validation set no longer decreases after 10 consecutive training epochs or reaches a preset number of training epochs, such as 200 epochs. When the convergence condition is met, training stops, the model parameters are saved, and the pre-trained degradation assessment model is obtained. Through the aforementioned pre-training process, the degradation assessment model can learn the mapping relationship between degradation risk characteristics and comprehensive degradation index in historical samples, which can be used to automatically assess the comprehensive degradation index of new forest patches.

[0088] For example, in an optional embodiment, the degradation assessment model may be constructed using the following network structure:

[0089] The input layer receives a three-dimensional degradation risk feature vector, namely, vegetation index time-series stability, persistent degradation index, and vegetation index resilience. Before entering the feature cross-layer, the three input features are standardized by a batch normalization layer to accelerate model convergence. The feature cross-layer receives the three input features and calculates pairwise cross-products: stability-persistent degradation cross-term = vegetation index time-series stability × persistent degradation index; persistent degradation-recovery resilience cross-term = persistent degradation index × vegetation index resilience; resilience-stability cross-term = vegetation index resilience × vegetation index time-series stability. The three original features are concatenated with the three cross-terms to form a six-dimensional cross-feature vector.

[0090] The gated fusion layer receives a six-dimensional cross-feature vector and learns the differentiated contribution weights of each feature to the overall degradation index through a gating mechanism. The gated fusion layer contains a fully connected sub-network that maps the six-dimensional input to six weight coefficients. These weight coefficients are normalized using Softmax and then multiplied element-wise with the six-dimensional features to output a weighted fusion feature vector. The fully connected sub-network contains a hidden layer with 16 neurons and ReLU activation.

[0091] The output layer receives the weighted fusion feature vector from the gated fusion layer and maps it to a comprehensive degradation index through two fully connected layers. The first fully connected layer contains 8 neurons with ReLU activation function and is followed by a Dropout layer with a dropout rate of 0.2 to prevent overfitting. The second fully connected layer contains 1 neuron with Sigmoid activation function, compressing the output to the range of 0 to 1 as the comprehensive degradation index.

[0092] During training, key hyperparameters of the degradation assessment model include a learning rate of 0.001, a training epoch count of 200, a batch size of 32, an Adam optimizer, and mean squared error as the loss function. Historical forest patch sample data are divided into training, validation, and test sets in a 7:2:1 ratio. After each training epoch, the mean squared error loss value on the validation set is calculated. When the validation set loss value no longer decreases for 10 consecutive training epochs, the degradation assessment model is considered converged, training is stopped, and the model parameters with the minimum validation set loss are saved. Through this pre-training process, the degradation assessment model can learn the nonlinear mapping relationship between degradation risk characteristics and the comprehensive degradation index in historical samples.

[0093] Specifically, the execution steps of this embodiment, namely, inputting the degradation risk characteristics of each forest patch into a pre-trained degradation assessment model to obtain a comprehensive degradation index, include:

[0094] The degradation risk feature vector of the forest patch is input into a pre-trained degradation assessment model, wherein the degradation assessment model includes a feature cross layer, a gating fusion layer and an output layer;

[0095] The degradation assessment model is used to perform layer-by-layer feature transformation and nonlinear mapping on the degradation risk feature vector to output the comprehensive degradation index of the forest patch.

[0096] Specifically, in practice, this step inputs the degradation risk feature vector of forest patches into the trained degradation assessment model. The degradation risk feature vector consists of three dimensions: vegetation index temporal stability, persistent degradation index, and vegetation index resilience. The degradation assessment model captures the nonlinear interaction between these three degradation risk features through layer-by-layer feature transformation and nonlinear mapping, outputting a comprehensive degradation index for forest patches.

[0097] The obtained comprehensive degradation index is a dimensionless continuous value ranging from 0 to 1. The closer the comprehensive degradation index is to 1, the higher the risk of vegetation growth sequence degradation in the forest patch, that is, the worse the forest health status, and the problem of continuous degradation or insufficient recovery capacity. The closer the comprehensive degradation index is to 0, the more stable the vegetation health status of the forest patch, that is, the smaller the interannual fluctuation of the vegetation index, the no continuous downward trend, and the strong recovery capacity, indicating that it is in a healthy state.

[0098] S40: For the forest patches whose comprehensive degradation index exceeds the preset risk threshold, calculate the factor time lag matching degree of the multi-source influencing factors respectively, and evaluate and obtain the types of important influencing factors based on the factor time lag matching degree and the vegetation index time series data.

[0099] Furthermore, after obtaining the comprehensive degradation index of each forest patch, it is necessary to identify patches with high degradation risk and analyze the causes of their degradation. Specifically, not all forest patches require attribution analysis of influencing factors. For patches with low comprehensive degradation indices, their vegetation is generally healthy, and there is no need to further trace the causes of degradation. Therefore, this step first sets a preset risk threshold and screens forest patches with comprehensive degradation indices exceeding this threshold as high-risk patches that require attribution analysis.

[0100] The preset risk threshold can be set based on the statistical distribution of historical degradation samples or management objectives. In one embodiment, samples of historical forest patches within the target area or similar ecological zones that have been confirmed to have significant degradation through ground surveys are collected. The minimum value of the patch's comprehensive degradation index is calculated, and this minimum value is used as the preset risk threshold. The lowest comprehensive degradation index among historically confirmed degraded patches represents the minimum risk level required for actual degradation to occur. Patches with a comprehensive degradation index below this value have not experienced actual degradation in historical experience and do not require attribution analysis. For example, if the lowest comprehensive degradation index among the 50 collected historical degraded patches is 0.6, the preset risk threshold can be set to 0.6.

[0101] Furthermore, for the selected high-risk patches, the main factors leading to the degradation of these patches are identified from multiple influencing factors, including meteorological factors, soil factors, and pest and disease factors. The impact of meteorological factors on vegetation growth has a lag effect; for example, the impact of insufficient rainfall in a given year on forest growth may not become apparent until the following or even third year. Directly calculating the correlation coefficient between contemporaneous influencing factors and vegetation indices would severely underestimate the contribution of meteorological factors. Therefore, this step introduces a time-lag matching mechanism. By shifting the time-series data of influencing factors backward by different numbers of years and calculating the correlation coefficient with the time-series data of vegetation indices, the time shift that maximizes the absolute value of the correlation coefficient is identified. This maximum correlation coefficient represents the factor time-lag matching degree. A higher factor time-lag matching degree indicates a stronger correlation between the influencing factor and changes in the vegetation index, leading to a greater likelihood of degradation.

[0102] Specifically, in this embodiment, the step of calculating the factor lag matching degree of the multi-source influencing factors for the forest patches whose comprehensive degradation index exceeds a preset risk threshold includes:

[0103] For forest patches whose comprehensive degradation index exceeds a preset risk threshold, a time shift range is set between 0 and a preset maximum time shift, where the preset maximum time shift is the longest delay in years that the influencing factor may have an observable impact on vegetation health.

[0104] For each time-shifted value, the time-series data of the influencing factor is shifted backward by the number of years corresponding to the time-shifted value, and the Pearson correlation coefficient between the time-series data of the influencing factor and the time-series data of the vegetation index is calculated.

[0105] The maximum absolute value of the Pearson correlation coefficient corresponding to all the time shift values ​​is taken as the factor time lag matching degree of the influencing factor.

[0106] First, for forest patches whose comprehensive degradation index exceeds a preset risk threshold, multiple time shift values ​​are set between 0 and a preset maximum time shift. The preset maximum time shift is the longest delay in which influencing factors may have an observable impact on vegetation health, determined based on the response cycle of vegetation to climate factors. For example, it can be set to 3 years. According to ecological studies on forest vegetation responses to climate factors, the response delay of most tree growth to meteorological factors such as precipitation and temperature is typically between 1 and 3 years; delays exceeding 3 years lack ecological mechanistic support. The time shift values ​​are integers between 0 and the preset maximum time shift, i.e., 0, 1, 2, and 3, representing the impact of the current year, a lag of one year, a lag of two years, and a lag of three years, respectively.

[0107] Secondly, for each time-shift value, the time-series data of the influencing factor is shifted backward by the number of years corresponding to that time-shift value, and the Pearson correlation coefficient between the time-series data of the influencing factor and the time-series data of the vegetation index is calculated. Specifically, taking annual precipitation in meteorological factors as an example, the annual precipitation data series is as follows: The vegetation index data series is When the shift value is 1, the precipitation sequence is shifted forward by one year, making... and correspond, and Similarly, the Pearson correlation coefficient between the shifted precipitation data and the corresponding vegetation index data is calculated. The Pearson correlation coefficient ranges from [value missing]. A value up to 1 indicates a positive correlation between the influencing factor and the vegetation index, while a negative value indicates a negative correlation. The larger the absolute value of the Pearson correlation coefficient, the stronger the linear association between the two.

[0108] Next, the time-shift calculations described above were performed on the three influencing factors: meteorological factors, soil factors, and pest and disease factors. The maximum absolute value of the Pearson correlation coefficient corresponding to all time-shift values ​​was taken as the factor time-lag fit degree for that influencing factor. The closer the factor time-lag fit degree is to 1, the stronger the correlation between this type of influencing factor and changes in vegetation health, and the higher the probability that it has a significant impact on forest degradation. For example, if the absolute value of the Pearson correlation coefficient for annual precipitation is the largest when the time-shift value is 1, it indicates that the impact of precipitation on vegetation growth has a one-year lag, and the two are highly correlated, thus the time-lag fit degree of the meteorological factor is relatively high.

[0109] Through the above calculations, the time-lag matching degree of each of the three influencing factors—meteorological factors, soil factors, and pest and disease factors—can be obtained. Specifically, the core of this step lies in eliminating the lag effect through time-shift matching, so that the influencing factors and vegetation indices achieve optimal alignment in time, thereby accurately assessing the contribution of each factor to forest degradation.

[0110] Furthermore, after obtaining the time-lag matching degree of each influencing factor, the magnitudes of the time-lag matching degrees of different factors are compared to identify the types of important influencing factors that contribute the most to degradation. Specifically, in this embodiment, the step of evaluating and obtaining the types of important influencing factors based on the factor time-lag matching degree and the vegetation index time-series data includes:

[0111] Obtain the factor time lag matching degree of the meteorological factor, the soil factor, and the pest and disease factor;

[0112] The factors with the highest time lag matching degree are sorted from largest to smallest, and the influencing factor with the highest time lag matching degree is selected as the dominant influencing factor.

[0113] The difference between the time-delay matching degree of the factor ranked first and the time-delay matching degree of the factor ranked second is calculated and used as the matching degree discrimination.

[0114] When the matching degree discrimination is greater than or equal to the preset discrimination threshold, the dominant influencing factor is regarded as an important influencing factor type.

[0115] When the matching degree discrimination is less than the discrimination threshold, the important influencing factor type is determined to be multi-factor coupled type.

[0116] First, the time-lag matching degrees of meteorological factors, soil factors, and pest and disease factors were obtained. Through the aforementioned time-lag matching calculations, the time-lag matching degree corresponding to each of the three influencing factors was obtained. Each time-lag matching degree is a value between 0 and 1, reflecting the degree of correlation between this type of influencing factor and changes in vegetation health.

[0117] Secondly, the factors were ranked from largest to smallest time lag matching degree, and the influencing factor with the largest time lag matching degree was selected as the dominant influencing factor. The larger the time lag matching degree, the stronger the correlation between this type of influencing factor and the change in vegetation index, and the higher the probability of its contribution to forest degradation. The factor ranked first is the dominant influencing factor of forest patches.

[0118] Furthermore, the difference between the time-lag matching degree of the first-ranked factor and the time-lag matching degree of the second-ranked factor is calculated as the matching degree discrimination score. The matching degree discrimination score is used to measure whether the dominant factor is statistically significantly better than other factors. The larger the matching degree discrimination score, the higher the attribution reliability of the dominant factor.

[0119] When the matching degree discrimination is greater than or equal to a preset discrimination threshold, the dominant influencing factor is considered as an important influencing factor type. The discrimination threshold is a critical value for determining whether a single factor can independently explain the causes of degradation. It is set based on the statistical distribution of matching degree discrimination in known single-factor degradation cases in historical sample data and the attribution reliability requirements; for example, it can be set to 0.15. When the matching degree difference between the dominant factor and the second largest factor reaches this discrimination threshold, it indicates that the dominant factor has a significantly stronger impact on vegetation health than other factors, degradation is mainly driven by this single factor, and the attribution result is reliable.

[0120] Furthermore, when the matching degree of discrimination is less than the discrimination threshold, the important influencing factor type is identified as multi-factor coupling type. Specifically, this multi-factor coupling type indicates that the matching degree between the dominant influencing factor and the second largest influencing factor is close, a single factor cannot explain the cause of degradation, and degradation is more likely to be driven by multiple factors such as meteorology, soil, and pests and diseases.

[0121] The aforementioned discrimination mechanism can prevent misattribution to a single factor when multiple influencing factors have similar contributions, thereby improving the reliability of attribution analysis.

[0122] S50: Based on the comprehensive degradation index and the types of important influencing factors, analyze and obtain the forest resource health status assessment results for each of the forest patches.

[0123] Finally, after obtaining the comprehensive degradation index and key influencing factor types for each forest patch, this information needs to be integrated into a structured forest resource health status assessment result to provide actionable decision-making basis for forestry management departments. The comprehensive degradation index reflects the risk level of forest vegetation degradation and is a core quantitative indicator for assessing forest health status; the key influencing factor types reveal the main driving factors leading to forest degradation and are crucial for formulating targeted management measures. Combining these two aspects can provide differentiated management recommendations for forest patches with different degrees of degradation.

[0124] Specifically, in this embodiment, the step of analyzing and obtaining the forest resource health status assessment results of each forest patch based on the comprehensive degradation index and the types of important influencing factors includes:

[0125] Based on the comprehensive degradation index, each of the forest patches is divided into high-risk degradation patches, medium-risk degradation patches, and low-risk degradation patches;

[0126] For the high-risk and medium-risk patches, corresponding degradation attribution information and management recommendations are output according to the types of important influencing factors.

[0127] The comprehensive degradation index, degradation risk level, types of important influencing factors, and management recommendations for each forest patch are combined to output the forest resource health status assessment results for that forest patch.

[0128] First, based on the comprehensive degradation index, the forest patches were divided into high-risk, medium-risk, and low-risk patches. The comprehensive degradation index is a continuous value between 0 and 1, with a higher value indicating a higher risk of degradation.

[0129] Optionally, grading thresholds can be set based on the distribution of historical degradation samples or management objectives. In one embodiment, historical forest patch samples whose degradation levels have been confirmed through ground surveys are collected within the target area or similar ecological zones. These samples are then sorted from low to high according to the comprehensive degradation index obtained from manual surveys, and the value at the third-to-last position after sorting is used as the grading threshold. Specifically, the comprehensive degradation indices of all historical samples are arranged in ascending order, and the value at the top third position is used as the boundary threshold between low and moderate degradation, while the value at the bottom third position is used as the boundary threshold between moderate and high degradation. For example, if the third-to-last values ​​of the comprehensive degradation index in the historical samples are 0.4 and 0.7, then patches with a comprehensive degradation index greater than or equal to 0.7 can be defined as high-risk degradation patches, those between 0.4 and 0.7 as moderate-risk degradation patches, and those less than 0.4 as low-risk degradation patches. High-risk degradation patches indicate that the forest vegetation has shown a significant trend of degradation and insufficient recovery capacity, requiring priority intervention; medium-risk degradation patches indicate that there are signs of degradation but have not yet deteriorated severely, requiring continuous monitoring and timely preventive measures; low-risk degradation patches indicate that the vegetation is in a stable state, and routine monitoring is sufficient.

[0130] Secondly, for high-risk and medium-risk degradation patches, corresponding degradation attribution information and management recommendations are output based on the type of important influencing factor. Preferably, when the important influencing factor is a meteorological factor, the degradation attribution information indicates that degradation is mainly driven by climatic factors such as reduced precipitation or abnormal temperatures. Management recommendations include strengthening drought monitoring and emergency irrigation, such as adopting water-collecting afforestation and soil mulching to conserve moisture, to improve the forest's ability to resist drought stress. When the important influencing factor is a soil factor, the degradation attribution information indicates that degradation is mainly driven by soil degradation factors such as decreased soil organic matter content. Management recommendations include soil improvement and organic matter replenishment, such as applying organic fertilizers, planting green manure plants, and implementing no-till or reduced-tillage soil conservation measures to restore soil fertility. When the important influencing factor is a pest or disease factor, the degradation attribution information indicates that degradation is mainly driven by pest or disease outbreaks. Management recommendations include strengthening pest and disease monitoring and biological control, such as setting up insect-attracting lamps, releasing natural enemy insects, and promptly removing diseased plants to control the spread of pests and diseases.

[0131] In addition, when the important influencing factor type is multi-factor coupled, the degradation attribution information is that degradation is driven by multiple factors such as meteorology, soil, and pests and diseases. The management recommendation is to develop an integrated management plan that takes into account water management, soil improvement and pest and disease control, and implement a systematic restoration strategy with multiple measures.

[0132] Finally, the comprehensive degradation index, degradation risk level, types of important influencing factors, and management recommendations for each forest patch are combined to output the forest resource health status assessment result for that forest patch. Optionally, the output format can be a structured table, with forest patch identifiers as row indexes, including fields such as comprehensive degradation index value, degradation risk level label, types of important influencing factors, degradation attribution information, and management recommendations, or the assessment results of each patch can be presented in a spatial visualization on a forest resource distribution map.

[0133] In summary, the output of the above assessment results can provide a basis for forestry management departments to formulate differentiated and precise management strategies for forest patches with different degrees of degradation and different causes of degradation.

[0134] In summary, the embodiments of this application have at least the following technical effects:

[0135] This invention first constructs three degradation risk characteristics—temporal stability, persistent degradation index, and resilience—based on vegetation index time-series data from the same seasonal phase over several consecutive years. These characteristics characterize forest health from three dimensions: volatility, degradation trend, and recovery capacity, overcoming the limitations of relying solely on a single trend indicator. Second, a pre-trained degradation assessment model nonlinearly fuses and enhances the cross-features of these three characteristics, outputting a comprehensive degradation index to quantify degradation risk. Third, a time-lag matching mechanism is introduced for high-risk patches. The time-series data of meteorological, soil, and pest / disease factors are shifted backward by different numbers of years and their correlation coefficients are calculated sequentially with the vegetation index sequence. The maximum value is taken as the time-lag matching degree for each factor, resolving the distortion in the assessment of the contribution of influencing factors caused by lag effects. Finally, risk levels are classified according to the comprehensive degradation index, and degradation attribution information and management recommendations are output based on the type of influencing factor with the highest matching degree, providing a quantifiable decision-making basis for precise forest resource management.

[0136] Example 2, as Figure 3 As shown, based on the same inventive concept as the machine learning-based forest land resource health status assessment method provided in Embodiment 1, this embodiment of the invention also provides a machine learning-based forest land resource health status assessment system, including:

[0137] The data acquisition module 11 is used to acquire remote sensing vegetation index time series data of each forest patch in the target area in the same seasonal phase over multiple consecutive years, as well as time series data of multi-source influencing factors during the same period. The multi-source influencing factors include at least meteorological factors, soil factors, and pest and disease factors.

[0138] The risk feature module 12 is used to calculate the vegetation index time-series stability, continuous degradation index and vegetation index recovery elasticity of each forest patch based on the vegetation index time-series data of each forest patch, as degradation risk features.

[0139] The degradation assessment module 13 is used to input the degradation risk characteristics of each forest patch into a pre-trained degradation assessment model to obtain a comprehensive degradation index.

[0140] Factor analysis module 14 is used to calculate the factor time lag matching degree of the multi-source influencing factors for the forest patches whose comprehensive degradation index exceeds the preset risk threshold, and to evaluate and obtain the types of important influencing factors based on the factor time lag matching degree and the vegetation index time series data.

[0141] The result output module 15 is used to analyze and obtain the forest resource health status assessment results of each forest patch based on the comprehensive degradation index and the types of important influencing factors.

[0142] Specifically, the data acquisition module 11 is used for:

[0143] Obtain remote sensing vegetation index time-series data for each forest patch within the target area in the same seasonal phase over multiple consecutive years, as well as concurrent time-series data for multiple source influencing factors, including:

[0144] Extract the boundary range of each forest patch from the target area and obtain the vegetation type information of each forest patch;

[0145] For each forest patch, image data of the same seasonal phase in each year are selected from remote sensing images of multiple consecutive years. The mean normalized vegetation index within the boundary of the forest patch is extracted and arranged in chronological order of year to form the time series data of vegetation index of the forest patch.

[0146] The annual precipitation and average annual temperature data of each forest patch were obtained to form time series data of meteorological factors;

[0147] The annual soil organic matter content data of each forest patch were obtained to form soil factor time series data;

[0148] The annual pest and disease occurrence area data of each forest patch are obtained to form time series data of pest and disease factors.

[0149] Specifically, the risk feature module 12 is used for:

[0150] Based on the time-series vegetation index data of each forest patch, the time-series stability of the vegetation index, the persistent degradation index, and the vegetation index resilience of each forest patch are calculated as degradation risk characteristics, including:

[0151] The absolute difference between the mean normalized vegetation index of each two adjacent years of the forest patch is obtained to obtain the vegetation index change sequence.

[0152] The interannual volatility of the vegetation index is obtained by dividing the standard deviation of the vegetation index change series by the arithmetic mean of the vegetation index change series.

[0153] Based on the interannual volatility of the vegetation index, the temporal stability of the vegetation index is calculated, wherein the temporal stability of the vegetation index and the interannual volatility of the vegetation index are negatively correlated.

[0154] The number of years in which the vegetation index decreased compared to the previous year is counted in the time series data of the vegetation index, and the percentage of years with a decrease is calculated.

[0155] The longest consecutive number of years in which the vegetation index has continuously declined is statistically obtained as the longest consecutive decline year, and the proportion of the longest consecutive decline year in the vegetation index change sequence is calculated to obtain the continuous decline proportion.

[0156] The continuous degradation index is obtained by weighted summing of the percentage of consecutive declines and the percentage of years of decline.

[0157] For each year in the vegetation index time series data where the vegetation index decreased compared to the previous year, check whether the vegetation index in the following year recovers to or exceeds the preset ratio of the vegetation index of the previous year in the year of decline. If so, it is recorded as a successful recovery.

[0158] The total number of successful recoverys is counted, and the total number of successful recoverys is divided by the number of years of decline to obtain the vegetation index recovery elasticity.

[0159] The vegetation index temporal stability, the continuous degradation index, and the vegetation index recovery elasticity are combined to form the degradation risk characteristics of the forest patch.

[0160] Specifically, the degradation assessment module 13 is used for:

[0161] The degradation risk characteristics of each of the aforementioned forest patches are input into a pre-trained degradation assessment model to obtain a comprehensive degradation index, including:

[0162] The degradation risk feature vector of the forest patch is input into a pre-trained degradation assessment model, wherein the degradation assessment model includes a feature cross layer, a gating fusion layer and an output layer;

[0163] The degradation assessment model is used to perform layer-by-layer feature transformation and nonlinear mapping on the degradation risk feature vector to output the comprehensive degradation index of the forest patch.

[0164] Specifically, the degradation assessment model also includes:

[0165] The feature cross layer is used to receive the vegetation index time-series stability, the continuous degradation index, and the vegetation index recovery elasticity, and perform pairwise cross-products to generate stability-continuous degradation cross term, continuous degradation-recovery elasticity cross term, and recovery elasticity-stability cross term. The three original risk features and the three cross terms are concatenated into a six-dimensional cross feature vector.

[0166] The gated fusion layer is used to receive the six-dimensional cross feature vector and learn the differentiated contribution weights of each cross term to the comprehensive degradation index through a gating mechanism.

[0167] The output layer is used to receive the output of the gated fusion layer, and outputs the comprehensive degradation index after full connection mapping.

[0168] The pre-training of the degradation assessment model includes:

[0169] Collect and acquire historical forest patch sample data, wherein each set of historical forest patch sample data includes the historical vegetation index time-series stability, historical vegetation index continuous degradation index and historical vegetation index recovery elasticity of a historical forest patch as sample degradation risk characteristics, and the historical comprehensive degradation index of the historical forest patch confirmed by manual investigation, which constitutes a training sample set;

[0170] A degradation assessment model is constructed, which takes the temporal stability of vegetation index, the continuous degradation index of vegetation index, and the recovery elasticity of vegetation index as inputs, and the comprehensive degradation index as output.

[0171] The degradation assessment model is trained using the training sample set, with the goal of minimizing the deviation between the output comprehensive degradation index and the historical comprehensive degradation index, until the preset convergence condition is met, thus obtaining the pre-trained degradation assessment model.

[0172] The factor analysis module 14 is specifically used for:

[0173] For forest patches whose comprehensive degradation index exceeds a preset risk threshold, the factor lag matching degree of the multi-source influencing factors is calculated, including:

[0174] For forest patches whose comprehensive degradation index exceeds a preset risk threshold, a time shift range is set between 0 and a preset maximum time shift, where the preset maximum time shift is the longest delay in years that the influencing factor may have an observable impact on vegetation health.

[0175] For each time-shifted value, the time-series data of the influencing factor is shifted backward by the number of years corresponding to the time-shifted value, and the Pearson correlation coefficient between the time-series data of the influencing factor and the time-series data of the vegetation index is calculated.

[0176] The maximum absolute value of the Pearson correlation coefficient corresponding to all the time shift values ​​is taken as the factor time lag matching degree of the influencing factor.

[0177] Furthermore, based on the factor time lag matching degree and the vegetation index time series data, the types of important influencing factors are assessed and obtained, including:

[0178] Obtain the factor time lag matching degree of the meteorological factor, the soil factor, and the pest and disease factor;

[0179] The factors with the highest time lag matching degree are sorted from largest to smallest, and the influencing factor with the highest time lag matching degree is selected as the dominant influencing factor.

[0180] The difference between the time-delay matching degree of the factor ranked first and the time-delay matching degree of the factor ranked second is calculated and used as the matching degree discrimination.

[0181] When the matching degree discrimination is greater than or equal to the preset discrimination threshold, the dominant influencing factor is regarded as an important influencing factor type.

[0182] When the matching degree discrimination is less than the discrimination threshold, the important influencing factor type is determined to be multi-factor coupled type.

[0183] Specifically, the result output module 15 is used for:

[0184] Based on the comprehensive degradation index and the types of important influencing factors, the forest resource health status assessment results for each of the aforementioned forest patches are analyzed and obtained, including:

[0185] Based on the comprehensive degradation index, each of the forest patches is divided into high-risk degradation patches, medium-risk degradation patches, and low-risk degradation patches;

[0186] For the high-risk and medium-risk patches, corresponding degradation attribution information and management recommendations are output according to the types of important influencing factors.

[0187] The comprehensive degradation index, degradation risk level, types of important influencing factors, and management recommendations for each forest patch are combined to output the forest resource health status assessment results for that forest patch.

Claims

1. A method for assessing the health status of forest resources based on machine learning, characterized in that, include: The remote sensing vegetation index time series data of each forest patch in the target area in the same seasonal phase over several consecutive years were obtained, as well as the time series data of multi-source influencing factors during the same period. The multi-source influencing factors include meteorological factors, soil factors and pest and disease factors. Based on the time-series vegetation index data of each forest patch, the time-series stability of the vegetation index, the persistent degradation index, and the vegetation index resilience of each forest patch are calculated as degradation risk characteristics, including: The absolute difference between the mean normalized vegetation index of each two adjacent years of the forest patch is obtained to obtain the vegetation index change sequence. The interannual volatility of the vegetation index is obtained by dividing the standard deviation of the vegetation index change series by the arithmetic mean of the vegetation index change series. Based on the interannual volatility of the vegetation index, the temporal stability of the vegetation index is calculated, wherein the temporal stability of the vegetation index and the interannual volatility of the vegetation index are negatively correlated. The number of years in which the vegetation index decreased compared to the previous year is counted in the time series data of the vegetation index, and the percentage of years with a decrease is calculated. The longest consecutive number of years in which the vegetation index has continuously declined is statistically obtained as the longest consecutive decline year, and the proportion of the longest consecutive decline year in the vegetation index change sequence is calculated to obtain the continuous decline proportion. The continuous degradation index is obtained by weighted summing of the percentage of consecutive declines and the percentage of years of decline. For each year in the vegetation index time series data where the vegetation index decreased compared to the previous year, check whether the vegetation index in the following year recovers to or exceeds the preset ratio of the vegetation index of the previous year in the year of decline. If so, it is recorded as a successful recovery. The total number of successful recoverys is counted, and the total number of successful recoverys is divided by the number of years of decline to obtain the vegetation index recovery elasticity. The vegetation index temporal stability, the continuous degradation index, and the vegetation index recovery elasticity are combined to form the degradation risk characteristics of the forest patch; The degradation risk characteristics of each forest patch are input into a pre-trained degradation assessment model to obtain a comprehensive degradation index; For forest patches where the comprehensive degradation index exceeds a preset risk threshold, the factor time-lag matching degree of the multi-source influencing factors is calculated respectively. Based on the factor time-lag matching degree and the vegetation index time-series data, the types of important influencing factors are assessed and obtained, including: For forest patches whose comprehensive degradation index exceeds a preset risk threshold, a time shift range is set between 0 and a preset maximum time shift, where the preset maximum time shift is the longest delay in years that the influencing factor may have an observable impact on vegetation health. For each time-shifted value, the time-series data of the influencing factor is shifted backward by the number of years corresponding to the time-shifted value, and the Pearson correlation coefficient between the time-series data of the influencing factor and the time-series data of the vegetation index is calculated. The maximum absolute value of the Pearson correlation coefficient corresponding to all the time shift values ​​is taken as the factor time lag matching degree of the influencing factor; Obtain the factor time lag matching degree of the meteorological factor, the soil factor, and the pest and disease factor; The factors with the highest time lag matching degree are sorted from largest to smallest, and the influencing factor with the highest time lag matching degree is selected as the dominant influencing factor. The difference between the time-delay matching degree of the factor ranked first and the time-delay matching degree of the factor ranked second is calculated and used as the matching degree discrimination. When the matching degree discrimination is greater than or equal to the preset discrimination threshold, the dominant influencing factor is regarded as an important influencing factor type. When the matching degree discrimination is less than the discrimination threshold, the important influencing factor type is determined to be multi-factor coupled type; Based on the comprehensive degradation index and the types of important influencing factors, the forest resource health status assessment results of each forest patch are obtained through analysis.

2. The method for assessing the health status of forest resources based on machine learning according to claim 1, characterized in that, The acquisition of remote sensing vegetation index time-series data for each forest patch within the target area in the same seasonal phase over multiple consecutive years, as well as concurrent multi-source influencing factor time-series data, includes: Extract the boundary range of each forest patch from the target area and obtain the vegetation type information of each forest patch; For each forest patch, image data of the same seasonal phase in each year are selected from remote sensing images of multiple consecutive years. The mean normalized vegetation index within the boundary of the forest patch is extracted and arranged in chronological order of year to form the time series data of vegetation index of the forest patch. The annual precipitation and average annual temperature data of each forest patch were obtained to form time series data of meteorological factors; The annual soil organic matter content data of each forest patch were obtained to form soil factor time series data; The annual pest and disease occurrence area data of each forest patch are obtained to form time series data of pest and disease factors.

3. The method for assessing the health status of forest resources based on machine learning according to claim 1, characterized in that, The step of inputting the degradation risk characteristics of each of the forest patches into a pre-trained degradation assessment model to obtain a comprehensive degradation index includes: The degradation risk feature vector of the forest patch is input into a pre-trained degradation assessment model, wherein the degradation assessment model includes a feature cross layer, a gating fusion layer and an output layer; The degradation assessment model is used to perform layer-by-layer feature transformation and nonlinear mapping on the degradation risk feature vector to output the comprehensive degradation index of the forest patch.

4. The method for assessing the health status of forest resources based on machine learning according to claim 3, characterized in that, The degradation assessment model also includes: The feature cross layer is used to receive the vegetation index time-series stability, the continuous degradation index, and the vegetation index recovery elasticity, and perform pairwise cross-products to generate stability-continuous degradation cross term, continuous degradation-recovery elasticity cross term, and recovery elasticity-stability cross term. The three original risk features and the three cross terms are concatenated into a six-dimensional cross feature vector. The gated fusion layer is used to receive the six-dimensional cross feature vector and learn the differentiated contribution weights of each cross term to the comprehensive degradation index through a gating mechanism. The output layer is used to receive the output of the gated fusion layer, and outputs the comprehensive degradation index after full connection mapping.

5. The method for assessing the health status of forest resources based on machine learning according to claim 3, characterized in that, The pre-training of the degradation assessment model includes: Collect and acquire historical forest patch sample data, wherein each set of historical forest patch sample data includes the historical vegetation index time-series stability, historical vegetation index continuous degradation index and historical vegetation index recovery elasticity of a historical forest patch as sample degradation risk characteristics, and the historical comprehensive degradation index of the historical forest patch confirmed by manual investigation, which constitutes a training sample set; A degradation assessment model is constructed, which takes the temporal stability of vegetation index, the continuous degradation index of vegetation index, and the recovery elasticity of vegetation index as inputs, and the comprehensive degradation index as output. The degradation assessment model is trained using the training sample set, with the goal of minimizing the deviation between the output comprehensive degradation index and the historical comprehensive degradation index, until the preset convergence condition is met, thus obtaining the pre-trained degradation assessment model.

6. The method for assessing the health status of forest resources based on machine learning according to claim 1, characterized in that, The step of analyzing and obtaining the forest resource health status assessment results for each forest patch based on the comprehensive degradation index and the types of important influencing factors includes: Based on the comprehensive degradation index, each of the forest patches is divided into high-risk degradation patches, medium-risk degradation patches, and low-risk degradation patches; For the high-risk and medium-risk patches, corresponding degradation attribution information and management recommendations are output according to the types of important influencing factors. The comprehensive degradation index, degradation risk level, types of important influencing factors, and management recommendations for each forest patch are combined to output the forest resource health status assessment results for that forest patch.

7. A forest land resource health status assessment system based on machine learning, characterized in that, The method for performing the forest resource health status assessment based on machine learning as described in any one of claims 1 to 6 includes: The data acquisition module is used to acquire remote sensing vegetation index time series data of each forest patch in the target area in the same seasonal phase over multiple consecutive years, as well as time series data of multi-source influencing factors during the same period. The multi-source influencing factors include meteorological factors, soil factors, and pest and disease factors. The risk characteristic module is used to calculate the temporal stability, persistent degradation index, and vegetation index resilience of each forest patch based on the time-series vegetation index data of each forest patch, as degradation risk characteristics, including: The absolute difference between the mean normalized vegetation index of each two adjacent years of the forest patch is obtained to obtain the vegetation index change sequence. The interannual volatility of the vegetation index is obtained by dividing the standard deviation of the vegetation index change series by the arithmetic mean of the vegetation index change series. Based on the interannual volatility of the vegetation index, the temporal stability of the vegetation index is calculated, wherein the temporal stability of the vegetation index and the interannual volatility of the vegetation index are negatively correlated. The number of years in which the vegetation index decreased compared to the previous year is counted in the time series data of the vegetation index, and the percentage of years with a decrease is calculated. The longest consecutive number of years in which the vegetation index has continuously declined is statistically obtained as the longest consecutive decline year, and the proportion of the longest consecutive decline year in the vegetation index change sequence is calculated to obtain the continuous decline proportion. The continuous degradation index is obtained by weighted summing of the percentage of consecutive declines and the percentage of years of decline. For each year in the vegetation index time series data where the vegetation index decreased compared to the previous year, check whether the vegetation index in the following year recovers to or exceeds the preset ratio of the vegetation index of the previous year in the year of decline. If so, it is recorded as a successful recovery. The total number of successful recoverys is counted, and the total number of successful recoverys is divided by the number of years of decline to obtain the vegetation index recovery elasticity. The vegetation index temporal stability, the continuous degradation index, and the vegetation index recovery elasticity are combined to form the degradation risk characteristics of the forest patch; The degradation assessment module is used to input the degradation risk characteristics of each forest patch into a pre-trained degradation assessment model to obtain a comprehensive degradation index. The factor analysis module is used to calculate the factor time-lag matching degree of the multi-source influencing factors for forest patches where the comprehensive degradation index exceeds a preset risk threshold, and to evaluate and obtain the types of important influencing factors based on the factor time-lag matching degree and the vegetation index time-series data, including: For forest patches whose comprehensive degradation index exceeds a preset risk threshold, a time shift range is set between 0 and a preset maximum time shift, where the preset maximum time shift is the longest delay in years that the influencing factor may have an observable impact on vegetation health. For each time-shifted value, the time-series data of the influencing factor is shifted backward by the number of years corresponding to the time-shifted value, and the Pearson correlation coefficient between the time-series data of the influencing factor and the time-series data of the vegetation index is calculated. The maximum absolute value of the Pearson correlation coefficient corresponding to all the time shift values ​​is taken as the factor time lag matching degree of the influencing factor; Obtain the factor time lag matching degree of the meteorological factor, the soil factor, and the pest and disease factor; The factors with the highest time lag matching degree are sorted from largest to smallest, and the influencing factor with the highest time lag matching degree is selected as the dominant influencing factor. The difference between the time-delay matching degree of the factor ranked first and the time-delay matching degree of the factor ranked second is calculated and used as the matching degree discrimination. When the matching degree discrimination is greater than or equal to the preset discrimination threshold, the dominant influencing factor is regarded as an important influencing factor type. When the matching degree discrimination is less than the discrimination threshold, the important influencing factor type is determined to be multi-factor coupled type; The results output module is used to analyze and obtain the forest resource health status assessment results of each forest patch based on the comprehensive degradation index and the types of important influencing factors.

Images