Forest crown vertical structure feature inversion method and large-area forest classification method and system

Abstract

The invention provides a crown vertical structure feature inversion method and a large-area forest classification method and system, and belongs to the field of forest ecological environment monitoring. The inversion method comprises the steps of collecting data of two different years in a long time sequence, speculating canopy height spatial distribution diagrams respectively, extracting pixels with increased canopy height as a height sample set, taking the rest as a feature sample set, and obtaining forest canopy height data; spectral features and spatial texture features are obtained; taking the year with forest disturbance as a growth and development starting point, and obtaining the canopy height, spectral reflectivity and spatial texture of each pixel from the growth and development starting point to the last year; respectively extracting annual average growth speed characteristics in the last year or when the spectral reflectivity of the canopy is saturated, and representing the vertical structure characteristics of the canopy by using the height characteristics and the growth speed characteristics of the canopy; according to the classification method, the horizontal space-time spectrum features and the vertical structure features are fused for forest classification. According to the invention, the accuracy and precision of large-area forest classification are improved.

Description

Technical Field

[0001] The present invention belongs to the field of forest ecological environment monitoring, and specifically relates to a method for inverting vertical structural characteristics of a canopy, and a method and system for classifying large-area forests. Background Art

[0002] Forests are a vital component of terrestrial ecosystems, playing a key role in preventing soil erosion, maintaining biodiversity, maintaining global carbon and oxygen balance, and mitigating global warming. However, different forest types exhibit significant differences in their ecological processes and service functions, such as regional soil and water conservation, watershed hydrothermal cycles, and local microclimate regulation. Their ecological benefits and carbon sequestration capacities also vary. For example, broadleaf forests have superior photosynthesis, soil and water conservation, and nutrient retention capabilities than coniferous forests. Coniferous forests also have a higher carbon sequestration capacity than broadleaf forests, but their ecological stability is slightly lower. Mixed forests, on the other hand, have superior soil nutrient content, in-forest temperature and humidity, vegetation cover, and soil and water loss prevention capabilities compared to pure forests. Therefore, large-scale, high-precision, and detailed spatial distribution of forest types has become essential data for modern applications such as optimizing forest resource allocation, sustainable management, and ecosystem service evaluation. Research on high-precision forest type classification methods has become a research hotspot for forest resource monitoring and sustainable management.

[0003] Forest classification generally involves methods such as spectral remote sensing, LiDAR, GPS measurements, and area sampling. Spectral remote sensing data can be obtained through satellites, aerial vehicles, and drones. Satellite remote sensing technology, with its wide coverage, short revisit cycles, and low data acquisition costs, has become a key tool for remote sensing classification of forest types at regional, national, and global scales.

[0004] Existing forest classification methods based on satellite remote sensing data use different classification features, classifying forest types based on differences in canopy spectral response characteristics, phenological characteristics, canopy spatiotemporal characteristics, and canopy vertical structure characteristics derived from airborne LiDAR data. However, these classification methods suffer from insufficient refinement, incomplete considerations, inaccurate classification, and a limited classification scope, making them unable to meet the needs of refined forest resource monitoring and sustainable management. Summary of the Invention

[0005] In view of the above-mentioned defects or deficiencies in the prior art, the present invention aims to provide a method for inverting the vertical structural characteristics of a canopy, a large-area forest classification method and system provided in an embodiment of the present invention, which integrates the horizontal spatiotemporal spectral characteristics and vertical structural characteristics of the canopy to improve the accuracy and precision of classification.

[0006] In order to achieve the above objectives, the embodiments of the present invention adopt the following technical solutions: In a first aspect, an embodiment of the present invention provides a method for inverting vertical structural characteristics of a forest canopy, comprising the following steps: Step S101, determining the monitoring area, collecting sample point data and long-term optical remote sensing data from two different years within a long time series, and inferring the spatial distribution maps of forest canopy height in the two years respectively; Step S102, based on the canopy height spatial distribution map, extracting pixels whose forest canopy height increases in the second year compared to the first year as a height sample set, and extracting pixels whose forest canopy height does not increase as a feature sample set; Step S103, respectively calculating the height difference and characteristic difference corresponding to the second year and the first year, constructing a relationship model between the height difference and the characteristic difference, and obtaining forest canopy height data within a long time series; Step S104: obtaining remote sensing images of the monitoring area in a long time series, and using the Normalized Difference Vegetation Index (NDVI) to characterize the canopy spectral characteristics and using the Spectral Variation Vegetation Index (SVVI_svar) to characterize the spatial texture characteristics after preprocessing; Step S105, filling in missing values ​​to obtain a complete forest canopy height dataset, NDVI dataset, and SVVI_svar dataset; Step S106, obtaining forest disturbance detection data in a long time series within the monitoring area, and taking the year in which the forest disturbance occurred as the starting point of growth and development; Step S107, based on the forest canopy height dataset, NDVI dataset, and SVVI_svar dataset, obtain the canopy height, spectral reflectance, and spatial texture of each pixel from the starting point of growth and development to the last year in the long time series to quantitatively describe the growth trajectory of the forest type; Step S108 , extracting the forest canopy spectrum, spatial texture, and canopy height characteristics of each pixel from the year of forest disturbance to the last year in the long time series or when the forest canopy spectral reflectance reaches saturation, to characterize the canopy growth rate characteristics; Step S109: Characterize the vertical structure characteristics of the canopy by combining the canopy height characteristics and the growth rate characteristics.

[0007] As a preferred embodiment of the present invention, in step S103, the height difference and feature difference corresponding to the second year and the first year are calculated based on the height sample set and the feature sample set, respectively.

[0008] In a second aspect, an embodiment of the present invention further provides a large-area forest classification method, the method comprising: Step S1, determining the monitoring area and long time series interval, and obtaining multi-source remote sensing data of the monitoring area; Step S2, extracting the spatiotemporal spectral characteristics of the canopy level using multi-source remote sensing data; Step S3: Using the above-mentioned canopy vertical structure feature inversion method, obtain the vertical structure features between forest types.

[0009] Step S4: integrating the horizontal spatiotemporal spectral characteristics and vertical structural characteristics of the forest canopy, and using field surveyed forest type samples and machine learning methods to perform large-scale forest type remote sensing classification; Step S5: quantitatively evaluate the mapping accuracy using forestry field survey sample data of the target year.

[0010] As a preferred embodiment of the present invention, the multi-source remote sensing data in step S1 includes SRTM DEM data, Landsat series data in a long time series interval, satellite-borne LiDAR data, Sentinel-1 / 2 data, meteorological data, soil data and forest type field survey data.

[0011] As a preferred embodiment of the present invention, step S4 specifically includes: Step S41, fusing the extracted spatiotemporal spectral characteristics of the forest canopy level, the forest canopy height characteristics, and the growth rate characteristics to form a feature space for remote sensing classification of forest types; Step S42: Based on the horizontal spatiotemporal spectral characteristics and vertical structural characteristics of the forest canopy, the correlation hierarchical clustering method and the random forest algorithm (RF) are used to calculate feature correlation and feature importance, so as to select classification features that are beneficial to the remote sensing classification of forest types in the monitoring area, and all the beneficial classification feature sets are used as the optimal classification feature combination; Step S43: Based on the optimal classification feature combination, the RF classifier is used to perform remote sensing classification of forest types.

[0012] As a preferred embodiment of the present invention, the canopy-level spatiotemporal spectral features extracted in step S3 include spectrum, phenology, texture, polarization, and geographical environment.

[0013] As a preferred embodiment of the present invention, when extracting the texture features, 18 texture features calculated using the RedEdge1 band and the gray-level co-occurrence matrix are used to comprehensively characterize the spatial texture feature differences of the forest vegetation canopy.

[0014] As a preferred embodiment of the present invention, when extracting phenological characteristics, statistical values ​​of 14 annual time series data are selected; the statistical values ​​include: 10-day interval red edge position index REP, comprehensive statistical characteristics of time series data, annual vertical polarization VV, vertical horizontal polarization VH, standard deviation of modified vegetation index mRVI, maximum value difference of normalized vegetation index between annual and winter NDVI_maxsummer, and synthetic image difference of mRVI maximum value between summer and winter mRVI_summerwinter; wherein, for the extraction of comprehensive statistical characteristics of time series data, linear interpolation method and Savitzky-Golay filtering algorithm are used to generate REP time series data set with 10-day intervals within the year, and all REP time series data are used to calculate 12 characteristics including mean, standard deviation, maximum value, minimum value, median, range, coefficient of variation, 0.25 quantile, 0.75 quantile, interquartile range, interquartile range, and interquartile range to characterize the differences in phenological characteristics of different forest types.

[0015] As a preferred embodiment of the present invention, when extracting the geographic environmental characteristics, three terrain characteristics, namely altitude, slope and aspect, are extracted; four climate characteristics, namely annual mean temperature, seasonal mean temperature, annual mean precipitation and seasonal mean precipitation, are extracted based on the global 1 km resolution and monthly scale climate products over a long period of time; based on the soil moisture data products with 1 km resolution, daily scale and 10-100 cm depth within the year, the soil moisture response index (SMRI) is calculated using a water loss rate model to quantify the ability of soil to provide the water required for the growth and development of forest types; and all geographic environmental covariates are resampled to a spatial resolution of 30 m using a bilinear interpolation method.

[0016] In a third aspect, an embodiment of the present invention provides a large-area forest classification system, the system comprising: a data acquisition module, a horizontal spatiotemporal spectrum feature extraction module, a vertical structure feature extraction module, a remote sensing classification module, and an evaluation module; wherein, The data acquisition module is used to determine the monitoring area and the long time series interval, and obtain multi-source remote sensing data of the monitoring area; The horizontal spatiotemporal spectrum feature extraction module is used to extract the horizontal spatiotemporal spectrum features of the canopy using multi-source remote sensing data; The vertical structure feature extraction module is used to obtain the vertical structure features between forest types using the canopy vertical structure feature inversion method described above; The remote sensing classification module is used to integrate the horizontal spatiotemporal spectral characteristics and vertical structural characteristics of the forest canopy, and use forest type samples from field surveys and machine learning methods to perform large-scale remote sensing classification of forest types; The evaluation module is used to quantitatively evaluate mapping accuracy using forestry field survey sample data of the target year.

[0017] The technical solution provided by the embodiment of the present invention has the following beneficial effects: The canopy vertical structural feature inversion method and large-area forest classification method and system provided by the embodiments of the present invention improve the classification effect of large-area forest types, enhance the accuracy and precision of classification, and address the core pain point of excessive cost in remote sensing classification of large-area forest types. This invention, for the first time, integrates canopy horizontal spatiotemporal spectral features (spectral, phenological, texture, etc.) with vertical structural features (canopy height, growth rate), addressing the limitation of traditional methods that rely solely on single-dimensional features in large-scale remote sensing classification of forest types. By fusing spaceborne LiDAR (GEDI / ICESat-2) and optical remote sensing (Landsat / Sentinel-2) data to extract forest canopy vertical structural features, it replaces the traditional method that relies on airborne LiDAR, effectively reducing the cost of large-area forest monitoring. Furthermore, it proposes a "growth rate feature" (annual average height / spectral / texture feature change rate) to characterize the differences in canopy development among different forest types from a temporal perspective, providing basic data support for the optimal allocation of forest resources.

[0018] Of course, it is not necessary to achieve all of the advantages described above simultaneously in order to implement any product or method of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0020] Figure 1 This is a flow chart of the method for inverting the vertical structural characteristics of a forest canopy according to an embodiment of the present invention; Figure 2 is a flow chart of the large-area forest classification method according to an embodiment of the present invention; Figure 3 It is an overall accuracy map of remote sensing classification of forest types based on different classification feature combinations in an application example of the present invention; Figure 4 This is a statistical graph of the classification accuracy F1 scores of different forest types based on different classification feature combinations in the application example of the present invention; Figure 5 This is a statistical graph of classification accuracy PA scores for different forest types based on different classification feature combinations in an application example of the present invention; Figure 6 It is a statistical graph of classification accuracy UA scores of different forest types based on different classification feature combinations in the application example of the present invention; Figure 7It is a comparison diagram of remote sensing classification details of forest types based on different classification feature combinations in the application example of the present invention. DETAILED DESCRIPTION

[0021] After discovering the above problems, the inventors of this application conducted a detailed study of existing remote sensing classification methods for large-scale forest types. The study found that remote sensing classification methods for forest types based on satellite remote sensing data can be roughly divided into four categories: The first type of method involves remote sensing classification of forest types based on differences in the spectral response characteristics (spectral signatures) of forest canopies. This classification is based on differences in the spectral reflectance of the forest canopy over time, as it varies with electromagnetic wavelength. For example, some researchers have used differences in the spectral signatures of forest types in QuickBird imagery to map and classify coniferous, broadleaved, and mixed coniferous and broadleaved forests. Alternatively, researchers have used subtle differences in the spectral signatures of different forest types in Earth Observing-1 (EO-1) hyperspectral imagery to perform a fine-scale classification of forest types in the western Himalayas. This type of method relies solely on spectral signature differences in a single-temporal remote sensing image, placing extremely high demands on image quality, spatial resolution, and spectral resolution. However, due to the complex forest stand structure and the limitations of sensor spectral resolution, significant spectral aliasing occurs between forest types, making fine-scale classification of forest types difficult to achieve using single-temporal optical remote sensing data.

[0022] The second category involves remote sensing classification of forest types based on phenological characteristics. This involves using multi-temporal remote sensing data to extract seasonal variations (phenological characteristics) in forest vegetation that develop over the long term during adaptation to environmental changes. Forest types are then classified based on differences in these phenological characteristics. For example, some researchers have used Sentinel-2 data from spring, summer, and autumn to extract phenological characteristics of forest types and classify them based on differences in these characteristics. Alternatively, they have combined Sentinel-2 data from autumn and spring to extract phenological characteristics and classify them based on differences in these characteristics. Compared to forest classification methods that rely solely on single-temporal spectral features, phenological characteristics effectively characterize the annual growth and development rhythms of different forest vegetation types that develop during long-term adaptation to environmental changes, providing an important perspective for remote sensing classification of forest types over large areas. However, this approach overlooks the influence of spectral and spatial texture features on remote sensing classification of forest types.

[0023] The third category of methods involves remote sensing classification of forest types based on differences in the spatiotemporal spectral characteristics of the forest canopy. This involves using hyperspectral remote sensing data to enhance the spectral resolution of multispectral remote sensing data, thereby obtaining optical remote sensing data with high spatial, temporal, and spectral resolution. This allows for the extraction of fine-scale phenological, spatial texture, and spectral characteristics of the forest canopy, and then classifies forest types based on differences in these characteristics. For example, some researchers have used airborne hyperspectral imagery to enhance the spectral resolution of Sentinel-2 imagery to obtain remote sensing images with high temporal and spectral resolution. These methods then classify forest types based on differences in the spectral, phenological, and spatial texture characteristics of the forest canopy. Alternatively, researchers have used the high spectral resolution of CHRIS-Proba data to enhance the spectral resolution of Landsat 8 OLI data to obtain remote sensing images with both high spatial and spectral resolution. These methods then use the Spectral Angle Mapper (SAM) to classify forest types based on spectral, phenological, and spatial texture characteristics. Compared to remote sensing classification methods that rely solely on phenological characteristics, the integration of spatiotemporal and spectral features reduces the impact of spectral aliasing on remote sensing classification of forest types. However, such methods ignore the potential impact of canopy vertical structural characteristics on the accuracy of remote sensing classification of forest types.

[0024] The fourth method involves remote sensing classification of forest types by integrating horizontal spatiotemporal spectral characteristics and vertical structural characteristics of the forest canopy. This involves first extracting spatiotemporal spectral characteristics (spectral, spatial texture, and phenology) of the forest canopy using optical remote sensing imagery, then extracting vertical structural characteristics of the forest canopy, such as canopy height, shape, and density, using lidar remote sensing data. Finally, these horizontal spatiotemporal spectral characteristics and vertical structural characteristics are integrated for remote sensing classification of forest types. For example, some researchers first extract vertical structural characteristics of the forest canopy, such as canopy height, canopy shape, and echo intensity, using airborne LiDAR data, then extract spectral characteristics of the forest canopy using satellite-borne multispectral data, and finally classify tree species based on differences in canopy spectral and vertical structural characteristics. Alternatively, researchers first extract phenological, spatial texture, and spectral characteristics of forest types using multi-temporal (spring, summer, and autumn) satellite-borne optical remote sensing imagery (GeoEye, Pleiades, and WorldView2), then accurately extract three-dimensional spatial structural characteristics of the forest canopy using airborne LiDAR data during the leaf-out period, and finally classify forest types based on differences in phenological, spectral, texture, and three-dimensional spatial structural characteristics of the forest canopy. Compared with remote sensing forest classification methods that only use horizontal spatiotemporal spectral features of the canopy, incorporating vertical canopy structure features significantly improves the accuracy of regional forest classification. However, these methods rely on high-density airborne LiDAR data to extract accurate vertical canopy structure features. Due to data acquisition costs, these methods are difficult to meet the needs of large-scale remote sensing forest classification.

[0025] The above analysis reveals that the integration of spatiotemporal spectral characteristics (phenology, spatial texture, and spectrum) of forest canopies significantly improves the accuracy of forest type classification based on satellite remote sensing imagery. However, currently used spatiotemporal spectral features only characterize differences in the horizontal spatiotemporal spectral characteristics of the forest canopy, overlooking the impact of vertical structural characteristics on remote sensing classification of forest types. In fact, the comprehensive adaptation of different forest vegetation types to their ecological environment over a long phylogenetic process, particularly in response to competition from surrounding trees, leads to significant differences in vertical structural parameters (such as tree height, crown shape, and growth rate). These can, to a certain extent, complement the horizontal spatiotemporal spectral characteristics of the canopy in terms of feature classification capabilities. This could theoretically further enhance the separability between forest types and, therefore, improve classification accuracy. Although previous studies have demonstrated that the integration of three-dimensional canopy structural characteristics extracted by airborne LiDAR with horizontal spatiotemporal spectral features can achieve high forest classification accuracy at small scales, the high cost of acquiring airborne LiDAR data makes this approach difficult to apply to large-scale forest classification. Therefore, exploring methods to extract vertical structural features of forest canopies based on satellite remote sensing data and integrating horizontal spatiotemporal spectra and vertical structural features of forest canopies to perform remote sensing classification of forest types in large areas remains a scientific problem that needs to be solved urgently.

[0026] It should be noted that the defects existing in the solutions in the above-mentioned prior art are the results obtained by the inventors after practice and careful research. Therefore, the discovery process of the above-mentioned problems and the solutions proposed in the embodiments of the present invention below for the above-mentioned problems should all be the contributions made by the inventors to the present invention in the process of the invention.

[0027] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations. It should be noted that the embodiments of the present invention and the features in the embodiments can also be combined with each other in the absence of conflict.

[0028] It should be noted that similar reference numerals and letters represent similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined or explained in subsequent figures. In the description of the present invention, the terms "first," "second," "third," "fourth," etc. are used only to distinguish the description and are not to be understood as indicating or implying relative importance.

[0029] Based on the above in-depth analysis, the embodiments of the present invention provide a method for inverting the vertical structural characteristics of a forest canopy, a method for classifying forests in large areas, and a system. The comprehensive adaptation of different forest types to the ecological environment during the long process of phylogeny, especially the competitive pressure from surrounding trees, leads to significant differences in the canopy height and growth rate characteristics of different forest types. For example, the growth rate of oak forests is generally lower than that of birch forests; while the growth rate of Simiao pine is generally higher than that of Yunnan pine and Armand pine, and the canopy height of cloud fir is significantly higher than that of Yunnan pine and Armand pine. The forest canopy height describes the differences in the three-dimensional spatial structural characteristics of the forest canopy from the dimension of the canopy spatial structure, while the growth rate characteristics describe the differences in the growth and development characteristics of forest types from the time dimension. The fusion of these two spatiotemporal structural characteristics of the forest canopy can, to a certain extent, complement the classification capabilities of the spectrum, spatial texture, and phenological characteristics extracted from optical remote sensing data, thereby helping to improve the classification accuracy of plateau mountain forest types. In response to the problem that existing research lacks the characterization of large-scale forest vertical structure parameters, the present invention integrates satellite-borne optical and lidar data to extract canopy height and growth rate characteristics to comprehensively characterize the vertical structure characteristics of the canopy, and integrates the horizontal characteristics of the canopy to form a comprehensive spatiotemporal feature, so as to perform remote sensing classification of large-area forest types in complex environments.

[0030] like Figure 1 As shown, the method for inverting the vertical structural characteristics of a canopy provided by an embodiment of the present invention includes the following steps: Step S101, determining the monitoring area, collecting sample point data and long-term optical remote sensing data from two different years within a long time series, and inferring the spatial distribution maps of forest canopy height in the two years respectively; Step S102, based on the canopy height spatial distribution map, extracting pixels whose forest canopy height increases in the second year compared to the first year as a height sample set, and extracting pixels whose forest canopy height does not increase as a feature sample set; Step S103, calculating the height difference and feature difference corresponding to the second year and the first year based on the height sample set and the feature sample set, constructing a relationship model between the height difference and the feature difference, and obtaining forest canopy height data within a long time series; In step S104, remote sensing images of the monitoring area in a long time series are obtained from Google Earth Engine (GEE). After preprocessing, the spectral characteristics of the forest canopy are characterized by the Normalized Difference Vegetation Index (NDVI), and the spatial texture characteristics of the forest canopy are characterized by the Spectral Variability Vegetation Index (SVVI) and the Sum of Squares: Variance (svar) calculated using the Gray-Level Co-occurrence Matrix (GLCM).

[0031] In this step, the SVVI_svar is calculated using the SVVI index and the GLCM method.

[0032] Step S105, using an interpolation method to fill in the forest canopy height data, NDVI, and SVVI_svar corresponding to the years in which the pixels are missing, to obtain a complete forest canopy height dataset, NDVI set, and SVVI_svar set; Step S106, obtaining forest disturbance detection data in a long time series within the monitoring area, and taking the year in which forest disturbance occurs in each pixel as the growth and development starting point of the pixel; Step S107, based on the forest canopy height dataset, NDVI dataset, and SVVI_svar dataset, obtain the canopy height, spectral reflectance, and spatial texture of each pixel from the starting point of growth and development to the last year in the long time series to quantitatively describe the growth trajectory of the forest type; In step S108, based on the forest type growth trajectory of each pixel, the annual average growth rate characteristics of the forest canopy spectrum, spatial texture, and canopy height of each pixel are extracted from the year of forest disturbance to the last year in the long time series or when the forest canopy spectral reflectance reaches saturation, and the canopy growth rate characteristics of different forest types are characterized by this.

[0033] In this step, the canopy's growth rate is characterized by inverting the canopy's spectrum, spatial texture, and canopy height. Forest growth is manifested not only in the increase in tree height but also in the occupancy and distribution heterogeneity of canopy elements such as leaves and branches in three-dimensional space—in other words, changes in the canopy's spectral and spatial texture characteristics.

[0034] Step S109: Characterize the vertical structure characteristics of the canopy by combining the canopy height characteristics and the growth rate characteristics.

[0035] Based on the canopy vertical structure characteristics obtained by the above canopy vertical structure characteristics inversion method, the embodiment of the present invention also provides a large area forest classification method. Figure 2 As shown, the method includes the following steps: Step S1, determine the monitoring area and long time series interval, and obtain multi-source remote sensing data of the monitoring area, wherein the multi-source remote sensing data includes SRTM DEM data, Landsat series data in the long time series interval, space-borne LiDAR data (GEDI and ICESat-2 ATLAS), Sentinel-1 / 2 data, meteorological data, soil data, and forest type field survey data.

[0036] In this step, multi-source remote sensing data basically comes from satellite remote sensing data; the long time series interval is a time period, such as 1986-2023.

[0037] Step S2: extracting the spatiotemporal spectral characteristics of the canopy using multi-source remote sensing data.

[0038] In this step, the extracted canopy-level spatiotemporal spectral features include spectrum, phenology, texture, polarization, and geographic environment. Each horizontal spatiotemporal spectral feature contains several specific characteristics or indicators. The specific characteristics of each feature are listed in Table 1, and the vegetation indices used in Table 1 and their corresponding calculation formulas are listed in Table 2.

[0039] Table 1 Extraction of spatiotemporal spectral characteristics of forest canopy level

[0040] Table 2 Vegetation index and corresponding calculation formula used in the present invention

[0041] In Table 2, They represent green, blue, red, near infrared, shortwave infrared 1, shortwave infrared 2, red edge band 1, red edge band 2, red edge band 3 and red edge band 4 respectively; Indicates the standard deviation of the blue, green, red, near-infrared, SWIR1, and SWIR2 bands; Indicates the standard deviation of the NIR, SWIR1, and SWIR2 bands. VV and VH represent the backscatter coefficients for vertical polarization (single polarization, both transmission and reception are vertical) and horizontal polarization (dual polarization, transmission direction is vertical, reception direction is horizontal), respectively.

[0042] In this step, when extracting the texture features, 18 texture features calculated using the RedEdge1 band and the Gray Level Co-occurrence Matrix (GLCM) are used to comprehensively characterize the spatial texture feature differences of the forest vegetation canopy.

[0043] When extracting the phenological characteristics, the statistical values ​​of 14 annual time series data were selected, including the 10-day interval red edge position index (REP), the comprehensive statistical characteristics of the time series data, the annual vertical polarization (VV), the vertical horizontal polarization (VH), the standard deviation of the modified radar vegetation index (mRVI), the normalized difference vegetation index (NDI) in the year and in winter (December to March), and the annual and winter (December to March) of the year. The difference in the maximum value of the National Forest Reservoir Index (NDVI) (NDVI_maxsummer) and the difference in the composite image of the maximum value of mRVI in summer and winter (mRVI_summerwinter) were used to jointly characterize the differences in phenological characteristics among different forest types. In particular, the linear interpolation method and Savitzky-Golay filtering algorithm were used to extract the comprehensive statistical characteristics of the REP time series data. The REP time series data set with 10-day intervals within the year was generated. The mean, standard deviation, maximum, minimum, median, range (i.e., the difference between the maximum and minimum values, Range of maximum and minimum, RMM), coefficient of variation (cv), 0.25 quantile (First Quartile, Q25), 0.75 quantile (Third Quartile, Q75), the interquartile range 1 (The interquartile range of Q75 and Q25, IRQ3Q1), and the interquartile range 2 (The interquartile range of Q25 and A total of 12 characteristics were used to characterize the differences in phenological characteristics of different forest types, including the minimum value (IRQ25Qmin), the interquartile range of maximum and Q75 (IRQmaxQ75).

[0044] To extract these geographic environmental characteristics, three topographic features (altitude, slope, and aspect) were extracted based on SRTMDEM data. Four climate characteristics (annual mean temperature, seasonal mean temperature, annual mean precipitation, and seasonal mean precipitation) were extracted based on a global, 1-km-resolution, monthly-scale climate product (WorldClim_version 2.1) over a long period of time. The Soil Moisture Response Index (SMRI) was calculated using a water loss rate model based on the 1-km-resolution, daily-scale, 10-100-cm-deep soil moisture data product (Soil Moisture of China by the in situ data, version 1.0, SMCI1.0) for the entire year. This was used to quantify the ability of soil to provide the water required for forest growth and development. All geographic environmental covariates were resampled to a spatial resolution of 30 m using bilinear interpolation.

[0045] The seasonal average temperature (precipitation) is calculated as follows: (1) In formula (1), MS represents the seasonal mean temperature (precipitation), and represent the standard deviation and mean of monthly temperature (precipitation), respectively.

[0046] Step S3: Using the above-mentioned canopy vertical structure feature inversion method, obtain the vertical structure features between forest types.

[0047] Step S4: Integrate the horizontal spatiotemporal spectral characteristics and vertical structural characteristics of the forest canopy, and use forest type samples from field surveys and machine learning methods to perform large-scale forest type remote sensing classification.

[0048] Furthermore, this step specifically includes: Step S41 , fusing the extracted horizontal spatiotemporal spectral features and vertical structural features of the forest canopy to form a feature space for remote sensing classification of forest types.

[0049] In step S42, based on the horizontal spatiotemporal spectral characteristics and vertical structural characteristics of the forest canopy, the correlation hierarchical clustering method and the random forest algorithm (RF) are used to calculate the feature correlation and feature importance, so as to select classification features that are beneficial to the remote sensing classification of forest types in the monitoring area, and all the beneficial classification feature sets are used as the optimal classification feature combination.

[0050] In this step, the beneficial classification features are taken as examples of coniferous forests and broad-leaved forests. For example, when the forest types include coniferous forests and broad-leaved forests, the beneficial classification features of the horizontal structure of the forest canopy corresponding to the two types are shown in Tables 3 and 4.

[0051] Table 3 Preferred classification characteristics of coniferous forest areas

[0052] Table 4 Preferred classification characteristics of broad-leaved forest areas

[0053] Step S43: Based on the optimal classification feature combination, the RF classifier is used to perform remote sensing classification of forest types.

[0054] Step S5: quantitatively evaluate the accuracy of remote sensing classification of forest types using forestry field survey sample data of the target year.

[0055] In this step, based on the validation samples from the field survey, five commonly used indicators, including Overall Accuracy (OA), Kappa coefficient, Producer's Accuracy (PA), User's Accuracy (UA), and F1 score (F1score), were selected to quantitatively evaluate the effectiveness of the forest canopy vertical structure characteristics extracted by the present invention for remote sensing classification of plateau mountain forest types. The calculation formulas for these indicators are as follows: (2) (3) (4) (5) (6) In formulas (2)-(6), is the total number of forest types, The total number of validation samples, Represents the elements on the diagonal of the confusion matrix, that is, the total number of correctly classified samples in each forest type classification, and They represent the total column and row sums of each forest type, that is, the total number of misclassified and missed samples; the larger the OA and Kappa values ​​are, the higher the overall classification accuracy of regional forest types; the larger the PA, UA and F1score values ​​are, the higher the classification accuracy of each forest type is.

[0056] Based on the same idea, an embodiment of the present invention also provides a large-area forest classification system, which includes: a data acquisition module, a horizontal spatiotemporal spectrum feature extraction module, a vertical structure feature extraction module, a remote sensing classification module and an evaluation module.

[0057] The data acquisition module is used to determine the monitoring area and long time series interval, and obtain multi-source remote sensing data of the monitoring area, wherein the multi-source remote sensing data includes SRTM DEM data, Landsat series data in the long time series interval, space-borne LiDAR data, Sentinel-1 / 2 data, meteorological data, soil data and forest type field survey data; The horizontal spatiotemporal spectrum feature extraction module is used to extract the horizontal spatiotemporal spectrum features of the canopy using multi-source remote sensing data; The vertical structure feature extraction module is used to obtain the vertical structure features between forest types using the canopy vertical structure feature inversion method described above; The remote sensing classification module is used to integrate the horizontal spatiotemporal spectral characteristics and vertical structural characteristics of the forest canopy, and use forest type samples from field surveys and machine learning methods to perform large-scale remote sensing classification of forest types; The evaluation module is used to quantitatively evaluate the accuracy of remote sensing classification of forest types using forestry field survey sample data of the target year.

[0058] In this embodiment, each module is implemented by a processor, and a memory is appropriately added when storage is required. The processor may be, but is not limited to, a microprocessor MPU, a central processing unit (CPU), a network processor (NP), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), other programmable logic devices, discrete gates, transistor logic devices, discrete hardware components, etc. The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0059] In the above embodiments, all or part of the embodiments may be implemented using software, hardware, firmware, or any combination thereof. When implemented using software, all or part of the embodiments may be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in accordance with the embodiments of the present invention are generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0060] It should also be noted that the large-area forest classification system described in this embodiment corresponds to the large-area forest classification method described above. The description and limitation of the method also apply to the system and will not be repeated here.

[0061] The large-area forest classification method and system described in the embodiments of the present invention are applied to the forest type classification in a monitoring area in Yunnan Province. Training samples and verification samples are generated based on the monitoring area, and the remote sensing classification accuracy and results of forest types based on different classification feature combinations are compared and analyzed.

[0062] For classification accuracy, the preferred features shown in Tables 3 and 4 were selected based on the training samples, and the forest type classification was performed using the random forest classifier. The validation samples were used to quantitatively evaluate the effectiveness of different classification feature combinations for remote sensing classification of forest types. The evaluation results are shown in Figure 2. Figure 3As shown in the figure. Overall, the fusion of spatiotemporal, spectral, and polarimetric features of forest canopies significantly improved the classification accuracy of plateau and mountainous forest types, but the classification contributions of different features varied. Because the spectral reflectance curves of different forest types are relatively similar in waveform, but their backscatter coefficients exhibit certain differences, the OA and Kappa coefficients for remote sensing classification of forest types in Yunnan Province using the fusion of Sentinel-1 and Sentinel-2 data were 65.38% and 0.5770, respectively. Texture features contributed little to the accuracy of remote sensing classification of forest types. Compared with using only spectral and polarimetric features, adding texture features increased OA and Kappa by 0.14% and 0.0005, respectively. The comprehensive adaptation of forest types to their growing environments results in significant differences in the growth and development rhythms of different forest types. In forest type remote sensing classification, the incorporation of phenological features can significantly improve tree species classification accuracy. Compared with using only spectral, polarimetric, and texture features, adding phenological features increased OA and Kappa by 1.13% and 0.0227, respectively. During their growth and development, forest vegetation constantly exchanges matter and energy with the surrounding environment. Geographical factors such as topography, climate, hydrology, and soils significantly influence the spatial distribution of forest types. Consequently, different forest types exhibit significant differences in these factors. For example, rubber trees are primarily distributed in low-altitude areas with adequate water and heat, while spruce trees are primarily found at higher altitudes. Incorporating geographic environmental features significantly improves the accuracy of remote sensing classification of forest types in mountainous plateau regions, with increases in OA and Kappa of 12.55% and 0.1442 compared to methods without these features. The spatiotemporal structure of forest canopies can effectively enhance remote sensing classification accuracy in mountainous plateau regions. Compared to using only spectral, polarimetric, texture, phenological, and geographic environmental features of the forest canopy, integrating spatiotemporal, spectral, and polarimetric features significantly improves classification accuracy, with increases in OA and Kappa of 3.29% and 0.0393, respectively.

[0063] In order to verify the contribution of different classification feature combination methods to remote sensing classification of different forest types, this example compares and analyzes the changing trends of classification accuracy of different forest types under different classification feature combination methods. Figure 4-Figure 6As shown, overall, the integration of forest canopy temporal, spatial, spectral, and polarimetric features significantly improved the classification accuracy of various forest types. However, due to differences in tree species composition and the complexity of the plant communities within different forest types, there were significant differences in the remote sensing classification accuracy of various forest types using different feature combination methods. In general, the classification accuracy of forest types was negatively correlated with the complexity of tree species composition and community structure. For example, the classification accuracy of coniferous forest types was significantly higher than that of broad-leaved forest types. Furthermore, the classification accuracy of pure forests such as Yunnan pine and Simiao pine was significantly higher than that of other coniferous forests. Furthermore, rubber, a typical economic forest in Yunnan Province with a simple tree species composition, had significantly higher classification accuracy than species with more complex community compositions such as oak, birch, and other broad-leaved forests. However, different classification feature combinations contributed differently to the classification performance of different tree species.

[0064] First, in the classification of coniferous forest types, the classification accuracy of Yunnan pine generally showed a trend of first decreasing and then increasing with the increase in the number of classification features; among them, the mapping accuracy of Yunnan pine that integrated spectral, polarization and texture features was the lowest, with an F1score of 0.7534. Compared with the use of only spectral and polarization features, the addition of texture features caused a certain degree of feature redundancy, and the classification accuracy of Yunnan pine decreased slightly, with its F1score and UA decreasing by 0.0024 and 1.87%, but its PA increased by 1.54%; this proves that the addition of texture features slightly increased the misclassification rate of Yunnan pine, but significantly reduced its omission rate; and the integration of forest canopy spatiotemporal-spectral-polarization features significantly improved the classification accuracy of Yunnan pine, with its F1score and PA increasing by 0.0121 and 2.3%, respectively. The classification accuracy of Pinus koraiensis generally increased with the addition of classification features, then decreased and then increased again. The highest classification accuracy was achieved after adding the spatiotemporal structure of the forest canopy, with F1score, UA, and PA reaching 0.9596, 94.41%, and 97.57%, respectively. Compared to the pre-introduction results, the F1score and UA increased by 0.0081 and 2.06%, respectively. The classification accuracy of Abies styracifolia continued to increase with the addition of classification features. The addition of spatiotemporal structure significantly improved the classification accuracy of Abies styracifolia. Compared to the pre-introduction results, the F1score, UA, and PA of Abies styracifolia, which incorporated the spatiotemporal, spectral, and polarizational features of the forest canopy, increased by 0.0377, 3.57%, and 2.81%, respectively. The classification accuracy of other coniferous forest types showed a continuous upward trend with the increase of classification features. After integrating the spatiotemporal, spectral, and polarimetric characteristics of the forest canopy, the F1score, UA, and PA of other coniferous forest types increased by 0.0262, 3.6%, and 1.54%, respectively.

[0065] Secondly, in the remote sensing classification of broadleaved forest types, oaks, as a typical hard broadleaved forest type, grow slowly and encompass a wide range of forest types. The classification accuracy of oaks in plateau and mountainous areas generally showed an upward and then downward trend with increasing classification accuracy. The highest classification accuracy for oaks was achieved when integrating forest canopy spatiotemporal, spectral, and polarizational features, with F1score, UA, and PA of 0.7232, 84.55%, and 63.18%, respectively. Compared to the classification without incorporating forest canopy spatiotemporal structural features, these improved by 0.0618, 10.79%, and 3.23%, respectively. Birches, as a typical soft broadleaved tree species (group) in Yunnan Province, grow rapidly. Incorporating forest canopy spatiotemporal structural features significantly improved classification accuracy for birch species, with F1score, UA, and PA increasing by 0.1209, 9.86%, and 15.34%, respectively, compared to the classification without incorporating forest canopy spatiotemporal, spectral, and polarizational features. The classification accuracy of rubber showed an overall trend of first decreasing and then increasing with the addition of classification features. The highest classification accuracy was achieved when integrating forest canopy spatiotemporal, spectral, and polarization features, with an F1score, UA, and PA of 0.9011, 82.00%, and 100%, respectively. Compared to the method without forest canopy spatiotemporal structural features, the PA increased by 4.65%. The classification accuracy of other broadleaf forests showed an initial increase, then a decrease, and then an increase. The highest classification accuracy was achieved when integrating forest canopy spatiotemporal, spectral, and polarization features, with an F1score, UA, and PA of 0.3648, 30.37%, and 45.67%, respectively. Compared to the method without forest canopy spatiotemporal structural features, the PA increased by 7.00%.

[0066] In addition, this example also conducted a comparative analysis of remote sensing classification results of plateau mountain forest types based on different feature combinations. The integration of forest canopy temporal-spatial-spectral-polarization features significantly improved the classification accuracy of plateau mountain forest types.

[0067] In addition, the present invention uses field photos to further compare the effectiveness of integrating forest canopy temporal-spatial-spectral-polarization characteristics for remote sensing classification of plateau mountain forest types, such as Figure 7 As shown. Figure 7 It can be seen that the results of remote sensing classification mapping of forest types in Yunnan Province based on different classification feature combinations have great differences in the detail scale. Figure 7The integration of spectral, polarimetric, textural, and phenological features in the classification of Pinus koraiensis resulted in significant confusion with Pinus yunnanensis. However, the integration of geographic environmental characteristics and the spatiotemporal structure of the forest canopy significantly improved the misclassification error for Pinus koraiensis. Compared to using only spectral, polarimetric, textural, and phenological features, the integration of these features with geographic environmental characteristics reduced the misclassification error for Pinus koraiensis by 47.94%. Compared to not incorporating spatiotemporal structure, the integration of spatiotemporal, spectral, and polarimetric forest canopy features increased the UA by 2.06%, significantly reducing the misclassification error. Rubber forest extraction based on spectral, polarimetric, textural, and phenological features showed some misclassification as oak and other broadleaf forests. Since rubber forests are primarily distributed in low-altitude areas with sufficient water and heat, the addition of geographic environmental characteristics significantly reduced the misclassification rate for rubber forests, improving the UA by 46% compared to the pre-incorporation of geographic environmental characteristics. The forest canopy-level characteristics of oak, such as spectrum, polarization, texture, phenology and geographical environment, are highly similar to those of birch and other broad-leaved forests. However, due to the slower growth rate of oak, there are significant differences in the spatiotemporal structural characteristics of the forest canopy between them.

[0068] Compared with the results before adding the spatiotemporal structural characteristics of the forest canopy, the UA after integrating the spatiotemporal, spectral and polarization characteristics of the forest canopy increased by 3.50%. Figure 7 It can be seen that other pine forests in high altitude areas were misclassified as Yunnan pine and cloud fir. After adding the spatiotemporal structure characteristics of the forest canopy, the misclassification rate and omission rate of other coniferous forests were significantly reduced, and their UA and PA increased by 3.60% and 1.54% respectively; cloud fir was largely misclassified as Yunnan pine and other coniferous forests. Considering the unique growth environment of cloud fir, the classification accuracy of cloud fir that integrated spectral, polarization, texture, phenology and geographical environment characteristics was significantly improved. Compared with before adding geographical environment characteristics, its UA and PA increased by 22.29% and 13.10%; In addition, the growth of cloud fir The speed is slower, and the forest canopy height is significantly higher than that of other coniferous forests. Therefore, compared with the situation before adding the spatiotemporal structure of the forest canopy, the UA and PA of the spruce fir increased by 17.62% and 2.81%, respectively, after fusing the horizontal spatiotemporal spectrum and vertical structure characteristics. Before adding the spatiotemporal structure of the forest canopy, the classification and mapping of birch forests showed a lot of confusion with oak and other broadleaf forests. Because birch forests are typical soft broadleaf forests with a fast growth rate, the integration of the spatiotemporal, spectral, and polarization characteristics of the forest canopy significantly improved the classification accuracy of birch, with its UA and PA increasing by 9.66% and 15.34%, respectively. Therefore, the integration of the spatiotemporal, spectral, and polarization characteristics of the forest canopy significantly improved the classification accuracy of each forest type in the plateau mountains, with the most significant improvements in the mapping accuracy of birch, oak, and spruce fir.

[0069] From the above, it can be seen that the canopy vertical structure feature inversion method, large-area forest classification method and system provided by the embodiments of the present invention improve the classification effect of large-area forest types and improve the accuracy and precision of classification.

[0070] The above description is only a preferred embodiment of the present invention and an explanation of the technical principles used. It is not intended to limit the scope of the invention to be protected, but merely represents a preferred embodiment of the present invention. Those skilled in the art should understand that the scope of the invention involved in the present invention is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the inventive concept. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative work shall fall within the scope of protection of the present invention.

Claims

1. A method for inverting vertical structural characteristics of a forest canopy, characterized in that: The steps include: Step S101, determining the monitoring area, collecting sample point data and long-term optical remote sensing data from two different years within a long time series, and inferring the spatial distribution maps of forest canopy height in the two years respectively; Step S102, based on the canopy height spatial distribution map, extracting pixels whose forest canopy height increases in the second year compared to the first year as a height sample set, and extracting pixels whose forest canopy height does not increase as a feature sample set; Step S103, respectively calculating the height difference and characteristic difference corresponding to the second year and the first year, constructing a relationship model between the height difference and the characteristic difference, and obtaining forest canopy height data within a long time series; Step S104: obtaining remote sensing images of the monitoring area in a long time series, and using the Normalized Difference Vegetation Index (NDVI) to characterize the canopy spectral characteristics and using the Spectral Variation Vegetation Index (SVVI_svar) to characterize the spatial texture characteristics after preprocessing; Step S105, filling in missing values ​​to obtain a complete forest canopy height dataset, NDVI dataset, and SVVI_svar dataset; Step S106, obtaining forest disturbance detection data in a long time series within the monitoring area, and taking the year in which the forest disturbance occurred as the starting point of growth and development; Step S107, based on the forest canopy height dataset, NDVI dataset, and SVVI_svar dataset, obtain the canopy height, spectral reflectance, and spatial texture of each pixel from the starting point of growth and development to the last year in the long time series to quantitatively describe the growth trajectory of the forest type; Step S108 , extracting the forest canopy spectrum, spatial texture, and canopy height characteristics of each pixel from the year of forest disturbance to the last year in the long time series or when the forest canopy spectral reflectance reaches saturation, to characterize the canopy growth rate characteristics; Step S109: Characterize the vertical structure characteristics of the canopy by combining the canopy height characteristics and the growth rate characteristics.

2. The method for inverting the vertical structural characteristics of a forest canopy according to claim 1, characterized in that: In step S103, the height difference and feature difference corresponding to the second year and the first year are calculated based on the height sample set and the feature sample set respectively.

3. A large-area forest classification method, characterized in that: The method comprises: Step S1, determining the monitoring area and long time series interval, and obtaining multi-source remote sensing data of the monitoring area; Step S2, extracting the spatiotemporal spectral characteristics of the canopy level using multi-source remote sensing data; Step S3, using the canopy vertical structure feature inversion method according to any one of claims 1 to 2 to obtain vertical structure features between forest types; Step S4: integrating the horizontal spatiotemporal spectral characteristics and vertical structural characteristics of the forest canopy, and using field surveyed forest type samples and machine learning methods to perform large-scale forest type remote sensing classification; Step S5: quantitatively evaluate the mapping accuracy using forestry field survey sample data of the target year.

4. The large-area forest classification method according to claim 3, characterized in that: The multi-source remote sensing data in step S1 includes SRTM DEM data, Landsat series data in a long time series interval, space-borne LiDAR data, Sentinel-1 / 2 data, meteorological data, soil data, and forest type field survey data.

5. The large-area forest classification method according to claim 3, characterized in that: Step S4 specifically includes: Step S41, fusing the extracted spatiotemporal spectral characteristics of the forest canopy level, the forest canopy height characteristics, and the growth rate characteristics to form a feature space for remote sensing classification of forest types; Step S42: Based on the horizontal spatiotemporal spectral characteristics and vertical structural characteristics of the forest canopy, the correlation hierarchical clustering method and the random forest algorithm (RF) are used to calculate feature correlation and feature importance, so as to select classification features that are beneficial to the remote sensing classification of forest types in the monitoring area, and all the beneficial classification feature sets are used as the optimal classification feature combination; Step S43: Based on the optimal classification feature combination, the RF classifier is used to perform remote sensing classification of forest types.

6. The large-area forest classification method according to claim 3, characterized in that: The spatiotemporal spectral features of the canopy level extracted in step S3 include spectrum, phenology, texture, polarization, and geographical environment.

7. The large-area forest classification method according to claim 6, characterized in that: When extracting the texture features, 18 texture features calculated using the RedEdge1 band and the gray-level co-occurrence matrix are used to comprehensively characterize the spatial texture feature differences of the forest vegetation canopy.

8. The large-area forest classification method according to claim 6, characterized in that: When extracting phenological characteristics, the statistical values ​​of 14 annual time series data were selected; the statistical values ​​included: 10-day interval red edge position index REP, comprehensive statistical characteristics of time series data, annual vertical polarization VV, vertical horizontal polarization VH, standard deviation of modified vegetation index mRVI, difference in maximum value of normalized vegetation index between the year and winter NDVI_maxsummer, and synthetic image difference of maximum value of mRVI between summer and winter mRVI_summerwinter; among them, for the extraction of comprehensive statistical characteristics of time series data, linear interpolation method and Savitzky-Golay filtering algorithm were used to generate REP time series data set with 10-day interval within the year, and all REP time series data were used to calculate 12 features including mean, standard deviation, maximum value, minimum value, median, range, coefficient of variation, 0.25 quantile, 0.75 quantile, interquartile range, interquartile range, and interquartile range to characterize the differences in phenological characteristics of different forest types.

9. The large-area forest classification method according to claim 6, characterized in that: When extracting the above-mentioned geographical environmental characteristics, three terrain characteristics, namely altitude, slope, and aspect, were extracted. Four climate characteristics, namely annual mean temperature, seasonal mean temperature, annual mean precipitation, and seasonal mean precipitation, were extracted based on global 1 km resolution and monthly scale climate products over a long period of time. Based on the soil moisture data products with 1 km resolution, daily scale, and 10-100 cm depth within the year, the soil moisture response index (SMRI) was calculated using a water loss rate model to quantify the ability of soil to provide the water required for the growth and development of forest types. All geographical environmental covariates were resampled to a spatial resolution of 30 m using the bilinear interpolation method.

10. A large-area forest classification system, characterized in that: The system includes: a data acquisition module, a horizontal spatiotemporal spectrum feature extraction module, a vertical structure feature extraction module, a remote sensing classification module and an evaluation module; wherein, The data acquisition module is used to determine the monitoring area and the long time series interval, and obtain multi-source remote sensing data of the monitoring area; The horizontal spatiotemporal spectrum feature extraction module is used to extract the horizontal spatiotemporal spectrum features of the canopy using multi-source remote sensing data; The vertical structure feature extraction module is used to obtain the vertical structure features between forest types using the canopy vertical structure feature inversion method described above; The remote sensing classification module is used to integrate the horizontal spatiotemporal spectral characteristics and vertical structural characteristics of the forest canopy, and use forest type samples from field surveys and machine learning methods to perform large-scale remote sensing classification of forest types; The evaluation module is used to quantitatively evaluate mapping accuracy using forestry field survey sample data of the target year.

Images