A forest drought death remote sensing evaluation method coupling variable relationship and data learning model

Abstract

The present application provides a forest drought death remote sensing evaluation method coupling variable relationship and data learning model, relates to the field of ecological disasters and remote sensing technology. The method firstly constructs a structural equation model based on sample point data, quantifies the direct and indirect relationship between forest drought mortality and key meteorological conditions, forest internal physiological parameters and external greenness characteristics. Subsequently, the random forest regression technology is integrated to predict the selected key variables in space, realize the wide coverage from sample observation to global scale, and finally generate the forest drought mortality in the global range. Compared with the traditional single evaluation method, the present application integrates multiple influence factors and multi-dimensional prediction model, improves the evaluation accuracy and spatial coverage ability of global forest drought mortality, and provides more direct and practical technical support for global forest health monitoring and ecosystem management.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of ecological disasters and remote sensing technology, in particular to a method for evaluating the mortality rate of global forests caused by drought disasters using statistical analysis combined with artificial intelligence technology. BACKGROUND

[0002] Drought is one of the most common and costly natural disasters worldwide, and its frequency and intensity are further increasing due to global warming. Drought poses a significant threat to forest ecosystems, potentially exacerbating the stress on forest growth and survival. Therefore, revealing the spatiotemporal pattern of drought and its impact on forests, and assessing the damage to global forests caused by drought, have become key tasks in addressing and mitigating global forest drought risks. Understanding the response mechanisms of forests to drought, especially through remote sensing technology, has become an urgent need for developing global water resource planning and ecological drought mitigation measures.

[0003] Monitoring and assessment of drought usually rely on meteorological and hydrological indicators such as precipitation, soil moisture, vapor pressure deficit, and Palmer Drought Severity Index. These indicators have some effect on characterizing meteorological drought, but often overlook the complexity of forest ecosystems and the dynamic changes of drought processes. Forests respond differently to drought conditions due to their ecological characteristics, with significant differences in the adaptability of tropical forests and northern forests to drought, which have been confirmed in multiple studies.

[0004] Vegetation indices such as the normalized difference vegetation index, enhanced vegetation index, and leaf area index are commonly used to assess forest growth and carbon sink capacity, and are sensitive to water response. However, the changes in these vegetation indices are not only influenced by drought, but also driven by factors such as fire, pests, or human intervention. Therefore, a single vegetation index may not accurately reflect the comprehensive impact of forest drought.

[0005] In recent years, research on vegetation water content and solar-induced chlorophyll fluorescence has provided new methods for large-scale monitoring of forest drought. Vegetation optical depth, which monitors vegetation water content through microwave remote sensing technology, can effectively characterize the total water content of aboveground biomass and is less affected by atmospheric conditions. SIF, as a direct indicator of photosynthesis, responds immediately to light and water stress, making it a new tool for measuring plant functional status. Studies have shown that water stress can cause stomata to close, thereby reducing the SIF signal.

[0006] While existing studies on meteorological and vegetation responses provide an important foundation for forest drought monitoring, significant challenges remain. Although the relationships between meteorological indicators and soil moisture and tree growth are direct, they do not entirely equate to the actual situation of vegetation drought. Changes in vegetation greenness index may also be influenced by other factors. The application of microwave remote sensing and SIF requires further exploration and validation. To effectively monitor forest drought, it is urgent to construct a comprehensive drought monitoring framework that integrates the relationships between vegetation, meteorological indicators, and forest drought mortality rates to improve the robustness and accuracy of drought monitoring. Summary of the Invention

[0007] Traditional methods for assessing forest drought mortality often rely on a single data source or model, which limits a comprehensive understanding of the relationships between complex environmental variables, thus affecting the accuracy and reliability of the assessment. To overcome this limitation, this invention proposes a method that couples structural equation modeling with a data learning model. This method first clarifies the relationship between forest drought mortality and environmental variables through structural equation modeling, and then combines these relationships with machine learning techniques. This integrated approach can significantly improve the accuracy of global forest drought mortality assessment, providing more accurate and reliable assessment results.

[0008] The method for assessing global forest mortality due to drought using a coupled structural equation model and a data learning model, as described in this invention, includes the following steps:

[0009] Step S1: Data Collection and Preprocessing: Collect various meteorological, hydrological, and vegetation data, and preprocess the data to ensure data consistency and usability;

[0010] Step S2: Calculate canopy mortality sample data: Collect geographical location and time data of global forest mortality due to drought, and use high-resolution global forest cover change data as background data to calculate canopy mortality;

[0011] Step S3: Training the structural equation model: Combining the data processed in step S1 and the canopy mortality rate calculated in S2, a structural equation model is constructed using the partial least squares path model PLS-PM as the implementation tool.

[0012] Step S4: Establish a mortality prediction function: Combine the structural equation model constructed in step S3, use mortality sample data to train and validate the structural equation model to identify significant environmental features, and construct a drought-related mortality prediction function based on the optimal structural equation model. The environmental features include vegetation and meteorological features.

[0013] Step S5: Spatial extrapolation of environmental features: Spatial extrapolation of the environmental features obtained in step S4 is performed using RF technology to generate a global spatial distribution map of latent variables;

[0014] Step S6: Calculate the global forest mortality rate due to drought in each region: Substitute the potential variable distribution obtained in step S5 into the prediction function obtained in step S4 to calculate the global forest mortality rate due to drought in each region;

[0015] Step S7: Draw a global forest drought mortality map: According to the mortality rate obtained in step S6, draw a visualization map of the global forest mortality rate due to drought, which is used to evaluate the impact of drought in each region.

[0016] A storage device for storing instructions and data for implementing the forest drought death remote sensing evaluation method based on the coupling of variable relationship and data learning model.

[0017] A forest drought death remote sensing evaluation device based on the coupling of variable relationship and data learning model, comprising: a processor and a storage device; the processor loads and executes the instructions and data in the storage device to implement the forest drought death remote sensing evaluation method based on the coupling of variable relationship and data learning model.

[0018] The beneficial effects provided by the present application are:

[0019] (1) Deepen the understanding of the response mechanism of forest drought: Through the analysis of the structural equation model, the present application clearly shows the influence of meteorological environment, vegetation water content change and photosynthesis activity on the complex process of forest death due to drought, and quantifies the specific response mechanism of forest under drought conditions. This research result not only improves the accuracy of understanding of forest drought phenomenon, but also lays a foundation for accurate evaluation of global forest death due to drought.

[0020] (2) Improve the accuracy of forest drought death remote sensing evaluation based on the coupling of variable relationship and data learning model: The present application combines the physiological mechanism of forest drought with machine learning technology to construct an efficient and accurate hybrid model for evaluating the distribution and severity of forest death caused by drought disasters. This method not only significantly improves the accuracy of evaluation, but also draws a detailed global forest drought death spatial distribution map, enabling forest managers to more intuitively grasp the distribution characteristics of global drought risk. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is a technical flowchart for estimating the global forest mortality rate due to drought using a hybrid model;

[0022] Figure 2 is the PLS-PM model result based on sample information;

[0023] Figure 3 is the overall influence of potential variables calculated from meteorological hydrology, vegetation water content, chlorophyll fluorescence (i.e. photosynthesis ability) and vegetation greenness on forest mortality due to drought.

[0024] Figure 4 is the latent variable space distribution obtained based on random forest regression space extrapolation;

[0025] Figure 5 is the scatter plot of reference mortality and model estimated mortality: (a) prediction model directly using random forest only, and (b) hybrid model combining PLS-PM quantified information and RF extrapolated prediction;

[0026] Figure 6 is the global forest drought mortality map from 2014 to 2018 estimated using the hybrid model, (a) to (e) corresponding to five different years (2014-2018) respectively;

[0027] Figure 7 is a schematic diagram of the operation of the hardware device. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solutions and advantages of the present application clearer, the embodiments of the present application will be further described below with reference to the drawings.

[0029] The present application first explains the related basic concepts and the core points of the present application as follows, and then the technical solutions of the present application are described in detail.

[0030] Please refer to Figure 1 , Figure 1 is a schematic diagram of the method flow of the present application; a forest drought death remote sensing evaluation method based on coupling structural equation model and data learning model of coupling variable relationship and data learning model, comprising the following:

[0031] Step S1: data collection and preprocessing; a variety of meteorological and hydrological data and vegetation data are collected, and unified spatial resolution, average value and annual percentage change are calculated and preprocessed to ensure the consistency and availability of the data;

[0032] Step S2: calculate the crown mortality sample data; collect the geographical position and time data of global forest drought death, and use the high-resolution global forest cover change data provided by Hansen team as background data to calculate the crown mortality;

[0033] Step S3: train the structural equation model: combine the data collected and calculated in steps S1 and S2, use PLS-PM as an implementation tool, and build a theoretical model to describe and quantify the direct and indirect effects of meteorological and hydrological drought on forest mortality;

[0034] Step S4: Establishing the mortality prediction function; combining the theoretical model constructed in step S3, using the death sample data to train and verify the model to identify significant vegetation and meteorological features, and to construct the drought-induced mortality prediction function;

[0035] Step S5: Spatial extrapolation of environmental characteristics; at the same time, the RF technology is used to perform spatial extrapolation on the latent variable scores (environmental characteristics) obtained in step S3 to generate a global spatial distribution map of the latent variables;

[0036] Step S6: Constructing a global forest drought-induced mortality mixed model; substituting the latent variable distribution results obtained in step S5 into the prediction function obtained in step S4 to calculate the drought-induced mortality rate of forests in each region of the world;

[0037] Step S7: Drawing a global forest drought-induced mortality map; according to the mortality results obtained in step S6, the mortality of global forests caused by drought can be visualized.

[0038] In the method of the present application, the meteorological and hydrological variables in step S1 include precipitation, temperature, SPEI, PDSI, VPD and SM, which are obtained from TerraClimate and SPEIbase. The vegetation response variables include VOD, SWIR, NDWI, SIF, SIFyield, NDVI and LAI, which are obtained from VODCA, PROBA-V, MODIS and other data sets. Data preprocessing includes converting different resolution data to consistent spatial grids by interpolation method to ensure comparability between different data sets for subsequent analysis. The average values of all meteorological and hydrological variables in the summer three months are calculated, and the annual percentage change is calculated for each vegetation response variable.

[0039] In the step S2, calculating the canopy mortality sample data includes three steps: ① Data collection: Extract the global forest mortality geographic location and time data from existing research literature. In this embodiment, the data comes from the research of Allen et al. (2010), Caudullo and Barredo (2019), Gazol and Camarero (2022), and Hammond et al. (2022). These records include specific geographic coordinates and time information of the occurrence of death, providing the basis data for the research. And use the high-resolution global forest cover change data provided by Hansen et al. (2013) as background data. This dataset provides global forest cover, increase, and decrease, with high spatial resolution; ② Spatial overlay the collected forest mortality location data caused by drought with the high-resolution forest cover change data, and divide them into grid cells according to the unified spatial resolution size. In this embodiment, the spatial resolution is 0.25°. In this way, each death record can be classified into the corresponding grid cell, thereby realizing the structured processing of the data; ③ For each grid cell, calculate the ratio of the dead canopy area to the total forest cover area in the grid to determine the forest mortality rate. Specifically, first identify the dead canopy area in each grid, then compare it with the total forest cover area in the same grid, and calculate the corresponding mortality rate, which is expressed in percentage.

[0040] In the step S3, the training of the structural equation model includes four steps: ① constructing a theoretical model, according to the data processed in step S1 and the canopy mortality calculated in step S2, a partial least squares path model PLS-PM is used as an implementation tool to construct a theoretical model, i.e., a structural equation model, which aims to describe the direct impact of meteorological and hydrological drought on forest mortality and the indirect impact through intermediate variables (such as soil moisture and plant physiological state). In the model, the independent variables include various meteorological and hydrological variables, and the dependent variable is the forest mortality; ② standardization and stability check, in order to ensure the stability and comparability of the model, all variables are standardized to convert them into standard deviation units. This process helps to eliminate the interference of different dimensions and units on the results, and ensures the effectiveness of the path coefficients and intercepts; ③ inspection of the singularity of the index: Cronbach's alpha index is used to check the singularity between the indexes in the model to evaluate the reliability of the measurement. By calculating the loading value of each latent variable (i.e., the correlation between them), those variables with loading values lower than the preset loading value are identified, and these low-loading-value variables are deleted or adjusted, and the preset loading value in this embodiment is 0.5; ④ model training and verification, after the processing and adjustment of the above steps, the model is trained and verified. Part of the sample data is used for model training, and the other part is used for model verification to evaluate the fitting degree and prediction ability of the model to select the optimal model.

[0041] In the step S4, the formula of the drought-induced mortality prediction function established is as follows:

[0042]

[0043] In the formula, Y represents the estimated forest mortality due to drought, LV i represents a significant feature, β i is the path coefficient of LV i , n is the number of significant features detected by PLS-PM, and ε i represents an error.

[0044] In the step S5, the spatial extrapolation of environmental characteristics includes three steps: ① obtaining latent variable scores, latent variable scores related to forest mortality are obtained by PLS-PM analysis. These latent variables represent different environmental characteristics and reflect the key factors affecting forest drought-induced mortality; ② application of RF algorithm, the latent variable scores obtained by the structural equation model are used as response variables, and the corresponding external manifest variables (such as soil data, NDVI, etc.) are used as prediction factors to train the RF model; ③ spatial extrapolation, the RF model is used to predict unobserved regions, and by inputting the variable data affecting forest drought-induced mortality in the global range, the RF model can estimate the latent variable scores of these regions.

[0045] In step S6, the global forest drought-induced mortality mixing model is derived by inputting the latent variable spatial distribution generated in step S5 into the drought mortality prediction function established in step S4, to obtain the complete distribution of forest drought-induced mortality on a global scale. The latent variable spatial distribution includes hydro-meteorological, vegetation water content, vegetation photosynthesis capacity and vegetation greenness, and the calculation provides a quantitative assessment of forest drought-induced mortality under drought conditions on a global scale.

[0046] In step S7, a global forest drought-induced mortality map is drawn. According to the calculation and evaluation results, the distribution map of forest drought-induced mortality on a global scale is visualized. These maps clearly show the predicted mortality in different regions, enabling users to identify high-risk areas and potential forest mortality hotspots. The mortality rate of the six main forest types in the world and the mortality rate distribution at different latitudes are also shown.

[0047] The embodiment takes the global high-temperature drought years from 2014 to 2018 as an example to further describe the technical solution of the global forest drought-induced mortality evaluation method based on microwave remote sensing of the present application. The embodiment is used to illustrate the present application, but is not used to limit the application range of the present application, which is also applicable to different regions or different time periods.

[0048] The implementation flowchart of the forest drought-induced mortality remote sensing evaluation method based on the coupling variable relationship of the coupled structural equation model and the data learning model and the data learning model of the method of the present application is shown in Figure 1 The specific steps are as follows:

[0049] (1) Collection and preprocessing of basic data;

[0050] In this embodiment, in order to further study the influence of meteorological and hydrological environmental variables on forest water content, growth conditions and drought-induced mortality, a SEM is constructed. A plurality of key driving variables are included, including precipitation, temperature, SPEI, PDSI, VPD and SM. SPEI combines precipitation and evapotranspiration to assess the impact of precipitation on the hydrological system, and is particularly suitable for analyzing the characteristics of drought events; PDSI further integrates precipitation, evaporation, temperature and soil moisture; VPD measures the water content in the atmosphere, reflecting the degree of air saturation; and SM directly reflects the water status in the soil, which is an important indicator for monitoring vegetation water stress. Except for SPEI, all these data are derived from the TerraClimate dataset, with a spatial resolution of 1 / 24 degree and a time scale of monthly, and the SPEI data is from the SPEIbase database, with a resolution of 0.5 degree, also monthly. The average value of the three summer months is selected as the model input, because summer is the season when forests are most under drought stress.

[0051] A number of vegetation response variables are also introduced, including VOD, SWIR, NDWI, SIF, SIFyield, NDVI and LAI. These variables reflect the water content, biomass, photosynthetic activity and health status of vegetation from different perspectives, respectively. For example, VOD reflects the water and biomass of vegetation by measuring the absorption and scattering intensity of microwave radiation by plant canopy; SWIR is sensitive to the water content of vegetation and can be used to monitor the water status of vegetation; SIF, as the chlorophyll re-emitted fluorescence signal, is an effective tool for assessing the health and efficiency of plant photosynthesis; NDVI and LAI represent the growth status and density of vegetation through optical properties and leaf area, respectively. These response variables are derived from VODCA, PROBA-V, LST SIFc and MODIS data sources, respectively. The data are pre-processed to reflect the changes in forest under summer water stress in the form of annual percentage change, and the calculation formula is as follows:

[0052]

[0053] where VI after represents the vegetation index of the current summer observation year, VI before represents the vegetation index of the previous summer.

[0054] In addition, two derived indicators, relative water content (RWC) and SIFyield, are also calculated.

[0055] RWC is based on VOD data, and is calculated by a normalization method using the 5th and 95th percentiles of VOD from 2002 to 2017 (the available period of selected VOD) to represent dry and wet conditions, respectively, and then calculating the median relative water content of each summer season, with the calculation formula as follows:

[0056]

[0057] where t represents each summer season, s represents a 0.25° grid cell, median t，s (VOD) represents the median VOD of each summer season; 5 th percentile s (VOD) represents the minimum VOD in the preset period; 95 th percentile s (VOD) represents the maximum VOD in the preset period.

[0058] SIFyield is then calculated by the ratio of SIF to photosynthetically active radiation (PAR) and the fraction of absorbed PAR (fPAR) by the vegetation canopy, which eliminates the effects of different solar radiation conditions and canopy structure on SIF data, and more accurately reflects the utilization efficiency of fluorescence in the process of photosynthesis. The calculation formula is as follows:

[0059]

[0060] In the formula, PAR represents photosynthetically active radiation reaching the vegetation canopy, and fPAR represents the proportion of all absorbed PAR by the vegetation canopy.

[0061] Table 1 Selected data information

[0062] Table 1. Main data information

[0063]

[0064] (2) Calculate the sample data of canopy mortality rate;

[0065] The present application aims to build a forest mortality prediction model caused by drought, and systematically carries out data collection and processing work to generate grid-based sample data of canopy mortality rate, which will serve as the basis for model training and verification. The specific calculation process is as follows:

[0066] Firstly, 464 records of geographical location and time information about forest mortality events from 2014 to 2018 were extracted from the research results of Allen et al. (2010), Caudullo and Barredo (2019), Gazol and Camarero (2022), and Hammond et al. (2022). The choice of this time period not only because it includes one of the ten hottest years recorded by the National Oceanic and Atmospheric Administration (NOAA), but also fully considers the availability and representativeness of the required data for the study.

[0067] Subsequently, high-resolution global forest cover change data published by Hansen et al. (2013) was used as background information, which is grid data of forest cover and loss, still choosing data from 2014 to 2018. By superimposing the location data of forest mortality caused by drought on the global forest grid data, the satellite-monitored canopy mortality rate was accurately calculated within each 0.25° grid cell. This calculation process is achieved by comparing the proportion of dead canopy area to total forest cover area within each grid.

[0068] Finally, a global dataset of forest mortality caused by drought was constructed, containing 464 grid sample points. The dataset was divided into two parts: 325 sample points were used for model training to ensure that the model could learn the complex relationship between drought and forest mortality; the remaining 139 sample points were used for model validation to evaluate the prediction ability and generalization performance of the model on unknown data.

[0069] (3) Training the structural equation model;

[0070] The present application focuses on training and applying structural equation models, particularly using the statistical method of PLS-PM, to analyze the complex causal relationships behind the phenomenon of forest death caused by drought. A detailed theoretical framework was constructed to systematically reveal how these key features, alone or in combination, affect the forest ecosystem and ultimately lead to an increase in forest mortality rate by considering a variety of meteorological driving variables (such as drought indicators, precipitation, and soil moisture) and vegetation response variables (such as forest water content, photosynthetic capacity, and vegetation greenness).

[0071] To construct this model, a path analysis diagram was first designed that includes the direct effects of meteorological and hydrological drought and the indirect effects of forest mortality rate through intermediate variables such as soil moisture and plant physiological state. The theoretical model was trained using death sample records collected from the literature. The model identifies meteorological features and vegetation features that are significantly related to forest mortality rate by fitting these sample data.

[0072] The overall goodness-of-fit of the optimal theoretical model is 0.79, as shown in Figure 2 The arrows represent the one-way relationship between latent variables. Solid and dashed arrows represent positive and negative relationships, respectively. The numbers on the arrows represent the path coefficients, indicating the size and direction of the direct causal effect between two variables. Non-significant coefficients (a = 0.05) in the arrows are not drawn. The numbers inside the latent variables are the R-squared of the latent variables, which measure the explanatory or predictive power of the latent variables on their observed variables. The right bar chart shows the overall impact (direct + indirect) of the calculated latent variables of each environmental factor on forest mortality rate caused by drought. It can be seen that vegetation water content, vegetation greenness, and coverage show more significant direct relationships than dry heat conditions and photosynthetic capacity, but the overall impact of hydro-meteorological conditions is greatly improved when combined with indirect relationships. This total impact is used to construct the prediction function of mortality rate.

[0073] Table 2. PLS-PM model results based on sample information

[0074]

[0075] (4) Establishing a mortality prediction function;

[0076] The present application utilizes sample data of forest death due to drought, through a rigorous statistical analysis and model construction process, quantifies the overall influence of key characteristic variables such as vegetation and meteorology on forest mortality due to drought, Figure 3 is the overall influence of potential variables calculated from meteorological hydrology, vegetation water content, chlorophyll fluorescence (i.e. photosynthetic capacity) and vegetation greenness on forest mortality due to drought. Specifically, PLS-PM is used as an analysis tool to deeply mine and accurately identify potential variables that significantly affect forest mortality due to drought. These potential variables mainly include specific vegetation response characteristics and meteorological drought conditions.

[0077] In the model training and verification phase, the core role of these potential variables in explaining and predicting forest mortality is verified. They indirectly or directly affect mortality through complex interaction mechanisms within the model. Based on the quantitative results of the model, a mortality prediction function is further constructed, which closely integrates the core results of PLS-PM analysis, reflecting the influence of key potential variables and ensuring the high accuracy and reliability of the prediction results. The formula of the established mortality prediction function is as follows:

[0078]

[0079] In the formula, Y represents the estimated forest mortality due to drought, LV i represents significant characteristics, βi is the path coefficient of LV i , n is the number of significant characteristics detected by PLS-PM, and ε i represents the error.

[0080] (5) Spatial extrapolation of environmental characteristics;

[0081] The present application not only quantifies the complex influence relationship between potential variables through PLS-PM, but also accurately extracts the scores of these potential variables. These potential variable scores, as a quantitative expression of the intrinsic properties of key characteristics, reveal the interaction between different characteristic elements and their contribution to forest mortality due to drought.

[0082] The present application adopts a random forest regression algorithm for spatial extrapolation, and the input variables (external manifest variables) of the latent hydro-meteorological variables include cumulative precipitation, soil moisture, PDSI and SPEI, the input variables (external manifest variables) of the latent vegetation water content include AVOD, RWC and SWIR, the input variables (external manifest variables) of the latent photosynthetic capacity include SIF and SIFyield, and the input variables (external manifest variables) of the latent vegetation greenness and coverage include EVI, NDVI and LAI. Finally, the latent variable scores based on limited sample observations are successfully mapped to the continuous grid on a global scale, realizing the spatial expansion from point to plane, and generating four groups of spatially continuous global distribution maps of latent variables, providing data for subsequent calculations, such as Figure 4 .

[0083] (6) Constructing a global forest drought-induced mortality mixed model;

[0084] The present application inputs the latent variable distribution results obtained by RF extrapolation into the previously constructed mortality prediction function. This process realizes the leap from sample observation to global prediction, enabling the prediction model to output corresponding forest drought-induced mortality prediction values for different regions of the world under specific environmental conditions.

[0085] The results are verified and evaluated by a special verification set, which accounts for 30% of the total data set and covers a five-year time span to ensure the representativeness of the samples. The prediction accuracy of the mixed model and the RF model alone in predicting global forest drought-induced mortality is compared. Through in-depth analysis of the results shown in (a) and (b) of Figure 5 , it is found that the mixed model performs better in predicting global forest drought-induced mortality, with a determination coefficient (R 2 ) of 0.74, which is significantly higher than that of the RF model alone, which is 0.63, with an increase of up to 18%. This result not only indicates that the mixed model can more effectively capture the multi-dimensional and complex factors affecting forest drought-induced mortality, but also further verifies that integrating multiple algorithms or data sources can significantly improve the accuracy and robustness of the prediction model.

[0086] (7) Drawing a global forest drought-induced mortality map from 2014 to 2018;

[0087] Finally, a global forest drought-induced mortality map from 2014 to 2018 is successfully drawn, Figure 6 , where (a) is the global forest drought-induced mortality in 2014, Figure 6 , where (b) is the global forest drought-induced mortality in 2015, Figure 6 , where (c) is the global forest drought-induced mortality in 2016, Figure 6(c) in the figure is the mortality rate of global forests caused by drought in 2016, Figure 6 (d) in the figure is the mortality rate of global forests caused by drought in 2017, Figure 6 (e) in the figure is the mortality rate of global forests caused by drought in 2018. The figure reveals the geographical distribution and changes of forest canopy mortality caused by drought in different years in detail. It is worth noting that the vast majority (more than 98%) of forest regions recorded a mortality rate below 15% in the five years, while the regions with a mortality rate exceeding 15% were considered to be severely affected by drought worldwide, including but not limited to Yunnan, China in 2014, Nigeria in 2015, eastern Brazil in 2016, Chile in 2017, and southeastern Australia in 2018.

[0088] In the lower left corner of Figure 6 , the proportion of the severity of mortality of the six main forest types (evergreen needle leaf forest ENF, evergreen broadleaf forest EBF, deciduous needle leaf forest DNF, deciduous broadleaf forest DBF, mixed forest MF, and savanna forest) under the influence of drought is further counted and displayed. The analysis results show that the mortality rate of savanna is the highest, followed by evergreen broadleaf forest, mixed forest, and other types of forest. This ranking reveals the differences in sensitivity and resistance of different forest types to drought stress. At the same time, Figure 6 , the right side of the figure shows the mortality rate statistics of different latitude regions, clearly indicating that the tropical, southern temperate, and northern coniferous forest regions are the frequent areas of forest mortality caused by drought.

[0089] To verify the prediction effect of the mixed model, its results are compared with the maps drawn using only the random forest algorithm year by year. The comparison results show that although the mixed model predicts a wider range of mortality distribution, the newly added part is mainly concentrated in the regions with a mortality rate below 10%, while the severely drought-affected regions with a mortality rate exceeding 15% are relatively reduced.

[0090] Please refer to Figure 7 , Figure 7 is a hardware device working schematic diagram of an embodiment of the present application, which specifically comprises: a forest drought death remote sensing evaluation device 401 coupled with a variable relationship and data learning model, a processor 402, and a storage device 403.

[0091] The forest drought death remote sensing evaluation device 401 coupled with a variable relationship and data learning model: the forest drought death remote sensing evaluation device 401 coupled with a variable relationship and data learning model realizes the forest drought death remote sensing evaluation method coupled with a variable relationship and data learning model.

[0092] Processor 402: the processor 402 loads and executes instructions and data in the storage device 403 for implementing the forest drought death remote sensing evaluation method of the coupled variable relationship and data learning model.

[0093] Storage device 403: the storage device 403 stores instructions and data; the storage device 403 is used to implement the forest drought death remote sensing evaluation method of the coupled variable relationship and data learning model.

[0094] The beneficial effects of the present application are:

[0095] (1) Deepen the cognition of forest drought response mechanism: the present application determines the influence of meteorological environment, vegetation water content change and photosynthesis activity on the complex process of forest death due to drought through structural equation model analysis, and quantifies the specific response mechanism of forest under drought conditions. This research result not only improves the recognition accuracy of forest drought phenomenon, but also lays a foundation for accurate evaluation of forest death due to drought in the global range.

[0096] (2) Improve the precision of forest drought death remote sensing evaluation of coupled variable relationship and data learning model: the present application fuses the physiological mechanism of forest drought and machine learning technology, and constructs an efficient and accurate hybrid model for evaluating the distribution and severity of forest death caused by drought disaster in the global range. This method not only significantly improves the evaluation accuracy, but also draws a detailed global forest drought death spatial distribution map, so that forest managers can more intuitively grasp the distribution characteristics of global drought risk.

[0097] The above is only the preferred embodiment of the present application, and is not used to limit the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A remote sensing assessment method for forest drought mortality based on coupled variable relationships and a data learning model, characterized in that: Includes the following steps: Step S1: Data Collection and Preprocessing: Collect various meteorological, hydrological, and vegetation data, and preprocess the data to ensure data consistency and usability; Step S2: Calculate canopy mortality sample data: Collect geographical location and time data of global forest mortality due to drought, and use high-resolution global forest cover change data as background data to calculate canopy mortality; Step S3: Training the structural equation model: Combining the data processed in step S1 and the canopy mortality rate calculated in S2, a structural equation model is constructed using the partial least squares path model PLS-PM as the implementation tool. Step S3, training the structural equation model, includes four steps: Step S3.1: Using PLS-PM as the implementation tool for SEM, a structural equation model was constructed to describe the process by which meteorological and hydrological drought directly affects or influences forest mortality through mediating variables, including soil moisture and plant physiological state. Step S3.2: Standardize the variables to standard deviation units to normalize the path coefficients and intercepts, ensuring the stability of the structural equation model; Step S3.3: Use Cronbach's alpha index to check for uniformity among indices and adjust the structural equation model according to the loading value. Variables with loading values ​​lower than the preset loading value are deleted to improve the accuracy of the structural equation model. Step S3.4: After the processing and adjustment in the above steps, the structural equation model is trained and validated to select the optimal model; Step S4: Establish a mortality prediction function: Combine the structural equation model constructed in step S3, use mortality sample data to train and validate the structural equation model to identify significant environmental features, and construct a drought-related mortality prediction function based on the optimal structural equation model. The environmental features include vegetation and meteorological features. In step S4, the drought-induced mortality prediction function is established using an optimal structural equation model, which is used to transform environmental characteristics into predicted forest mortality rates. The formula for the established drought-induced mortality prediction function is as follows: Y = * + In the formula, Y represents the forest mortality rate estimated due to drought. Indicates significant environmental characteristics, yes Path coefficients, It is the number of salient features detected by PLS-PM. Indicates error; Step S5: Spatial extrapolation of environmental features: Spatial extrapolation of the environmental features obtained in step S4 is performed using RF technology to generate a global spatial distribution map of latent variables; Step S6: Calculate forest drought mortality rate in each region of the world: Substitute the latent variable distribution results obtained in step S5 into the prediction function obtained in step S4 to calculate the forest drought mortality rate in each region of the world. Step S7: Create a global map of forest mortality due to drought: Based on the mortality rate results obtained in Step S6, create a visualization of global forest mortality due to drought to show the extent of drought impact in various regions.

2. The remote sensing assessment method for forest drought mortality based on coupled variable relationships and a data learning model as described in claim 1, characterized in that: In step S1, the meteorological and hydrological variables include precipitation, temperature, standardized precipitation evapotranspiration index SPEI, Palmer drought severity index PDSI, vapor pressure deficit VPD, and soil moisture SM. The vegetation response variables include vegetation optical depth (VOD), shortwave infrared reflectance (SWIR), normalized water index (NDWI), solar-induced chlorophyll fluorescence (SIF), normalized vegetation index (NDVI), and leaf area index (LAI); relative water content is calculated based on VOD. : In the formula, t represents each summer season, and s represents the grid cell of uniform spatial resolution size; This represents the median VOD value for each summer. This indicates the lowest VOD value within a preset time period; This indicates the highest VOD value within a preset time period; Fluorescence utilization efficiency (SIFyield) calculated based on SIF: In the formula, PAR represents the photosynthetically active radiation reaching the vegetation canopy, and fPAR represents the proportion of all PAR absorbed by the vegetation canopy. The preprocessing includes standardizing the spatial resolution, averaging, and calculating the annual percentage change.

3. The remote sensing assessment method for forest drought mortality based on coupled variable relationships and a data learning model as described in claim 1, characterized in that: In step S2, calculating the canopy mortality rate sample data includes three parts: Step S2.1: Data collection, extracting geographic location and time data of global forest drought-related mortality events from several studies, and collecting high-resolution global forest cover change data as background data; Step S2.2: Data overlay and gridding, specifically: These forest death location data are overlaid with forest grid data and divided globally according to grid units of a unified spatial resolution; Step S2.3: Calculate the canopy mortality rate, specifically: For each grid cell obtained in step S2.1, the canopy mortality rate is calculated as the ratio of the area of ​​the dead forest canopy within that grid cell to the total forest cover area of ​​that region.

4. The remote sensing assessment method for forest drought mortality based on coupled variable relationships and a data learning model as described in claim 1, characterized in that: In step S5, the main process of spatial extrapolation of environmental variables includes extracting latent variable scores from the structural equation model, which reflect the environmental characteristics and vegetation status of each sample point; and using the RF algorithm to spatially extrapolate the extracted latent variable scores, i.e. the environmental characteristics, to unobserved areas worldwide.

5. A storage device, characterized in that: The storage device stores instructions and data for implementing the remote sensing assessment method for forest drought mortality according to any one of claims 1 to 4, which involves coupled variable relationships and data learning models.

6. A remote sensing assessment device for forest drought mortality based on coupled variable relationships and a data learning model, characterized in that: include: A processor and a storage device; the processor loads and executes instructions and data in the storage device to implement the remote sensing assessment method for forest drought mortality according to any one of claims 1 to 4, which is based on the coupled variable relationship and data learning model.

Images