Karst wetland vegetation nutrient element spectral mechanism analysis method

Abstract

The invention provides a karst wetland vegetation nutrient element spectral mechanism analysis method. The method comprises the steps of in-situ hyperspectral measurement of karst wetland vegetation leaves, determination of nutrient elements of the karst wetland vegetation leaves, continuous wavelet decomposition and morphological characteristic calculation of the in-situ hyperspectral, and quantification of spectral heterogeneity among different karst wetland vegetation communities. And mining potential correlation between the leaf nutrient elements and the spectral response mechanism based on a dynamically enhanced wavelet two-dimensional correlation spectrum technology, and the like. According to the method, subtle spectral heterogeneity among different karst wetland vegetation communities can be accurately captured by using improved continuous wavelet information divergence, the internal relation between leaf nutrient elements and a spectral mechanism can be excavated by using a dynamically enhanced wavelet two-dimensional correlation spectrum technology, and sensitive wavebands of different karst wetland vegetation leaf nutrient elements are obtained; and a spectral mechanism of karst wetland vegetation community nutrient elements is depicted finely.

Description

Technical Field

[0001] The present invention belongs to the field of wetland vegetation monitoring. Specifically, it relates to a method for analyzing the leaf nutrient elements, especially the contents of nitrogen, phosphorus, and potassium, of a typical karst wetland vegetation community. Background Art

[0002] Karst wetlands are a special type of wetland. They have the unique calcium-rich and slightly alkaline soil and water characteristics and the typical karst soil and water circulation evolution mechanism in karst areas. They are the main carbon sinks of terrestrial ecosystems and play a crucial role in global carbon sequestration and terrestrial carbon cycling. Vegetation, as the most sensitive element of karst wetlands, plays a crucial role in maintaining water balance, purifying water quality, and absorbing pollutants. Nutrient elements (nitrogen, phosphorus, and potassium), etc., as the core attributes and measurable characteristics of wetland plants, play an important role in the physiological regulation and overall growth of vegetation, and can also accurately capture the growth status of wetland vegetation and the changes in wetland ecosystems. At present, under the dual influence of human activities and climate change, karst wetlands are facing major ecological and environmental governance challenges, including biodiversity decline, shrinking water area, and ecosystem degradation. Therefore, estimating the contents of nitrogen, phosphorus, and potassium in leaves is of great significance for monitoring the growth of karst wetland vegetation and protecting wetland ecosystems.

[0003] Traditional measurement of the contents of nitrogen, phosphorus, and potassium in leaves relies on destructive field sampling and laboratory chemical analysis, which consumes a large amount of manpower, material resources, and financial resources. Moreover, the wetland environment is complex and changeable, and the local field sampling measurement results are difficult to represent the contents of nitrogen, phosphorus, and potassium in leaves of the entire region. However, remote sensing technology can achieve large-area synchronous observation, making up for the deficiencies of traditional ground measurement. Currently, it has been widely used in the estimation of vegetation physiological parameters. Due to limitations in spectral range, resolution, cost, etc., unmanned aerial vehicle (UAV) multi-spectral and hyperspectral images and satellite images cannot fully cover the sensitive spectral range of the contents of nitrogen, phosphorus, and potassium in leaves of karst wetland vegetation communities, and it is difficult to achieve their high-precision inversion. Compared with UAV and satellite images, ground-measured hyperspectral data at the leaf scale has higher spectral resolution and signal-to-noise ratio, laying an important foundation for the fine analysis of the spectral characteristics of the contents of nitrogen, phosphorus, and potassium and their high-precision inversion. However, ground-measured hyperspectral data often has problems such as baseline shift, spectral overlapping peaks, and spectral noise, which restrict our accurate analysis of the spectral reflectance characteristics of vegetation. Some scholars have used methods such as derivative transformation and Pearson correlation coefficient to screen sensitive bands of vegetation functional trait parameters. However, the above studies only screen based on the characteristics of vegetation spectral reflectance, ignoring the importance of the morphological characteristics of vegetation spectral curves, and the spectral heterogeneity among different wetland vegetation communities remains to be carefully analyzed.

[0004] Most of the studies on the functional trait parameters of karst wetland vegetation focus on chlorophyll content, and there is little research on the spectral response mechanism of nutrient elements. At the same time, karst wetlands have the characteristics of calcium-rich and slightly alkaline soil and water, and a typical karst soil and water cycle evolution mechanism. The spectral characteristics of karst wetland vegetation are different from those of other vegetation. Therefore, the spectral response mechanism of other vegetation cannot be simply transferred to the inversion of nutrient elements in karst wetland vegetation.

[0005] Therefore, there is an urgent need for a method based on ground-measured hyperspectral data to capture the sensitive mechanisms of different nutrient elements, realize the fine inversion of nutrient elements in karst wetland vegetation, and lay a foundation for the large-scale, stable and high-precision monitoring of nutrient elements in karst wetland vegetation communities. Summary of the Invention

[0006] To solve the above problems, the primary object of the present invention is to provide a method for analyzing the spectral mechanism of nutrient elements in karst wetland vegetation. This method can accurately capture the subtle spectral heterogeneity among different karst wetland vegetation communities by using the improved continuous wavelet information divergence. The dynamically enhanced wavelet two-dimensional correlation spectroscopy technology can explore the internal relationship between leaf nutrient elements and spectral mechanisms, and obtain the sensitive bands of leaf nutrient elements in different karst wetland vegetation. The two finely depict the spectral mechanism of nutrient elements in karst wetland vegetation communities.

[0007] Another object of the present invention is to provide a method for analyzing the spectral mechanism of nutrient elements in karst wetland vegetation. By combining the nutrient element spectral mechanisms excavated by the improved continuous wavelet information divergence and the dynamically enhanced wavelet two-dimensional correlation spectroscopy technology with the adaptive ensemble learning algorithm, a mechanism-guided adaptive ensemble learning regression model is constructed, which can effectively overcome the limitations of traditional inversion models, make full use of the information in spectral data, and realize the accurate inversion of nutrient elements in various karst wetland vegetation, providing reliable data support for evaluating the growth status of vegetation, monitoring the health of the ecosystem, and formulating scientific and reasonable ecological protection strategies, and contributing to the sustainable development of karst wetland ecosystems.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] A method for analyzing the spectral mechanism of nutrient elements in karst wetland vegetation, the method comprising the following steps:

[0010] Step (1): In-situ hyperspectral measurement of karst wetland vegetation leaves;

[0011] Use a portable ASD ground object spectrometer to obtain in-situ hyperspectral data (350 - 2500 nm) of leaves of different karst wetland vegetation communities. 70 spectral curves are collected for each vegetation sample, and their mean value is taken as the final spectral data of the karst wetland leaf sample.

[0012] Step (2): Determination of the nutrient elements in the leaves of karst wetland vegetation;

[0013] The determination of the nutrient elements in the leaves includes: measuring the potassium content in the leaves of karst wetland vegetation, measuring the nitrogen content and phosphorus content in the leaves of karst wetland vegetation.

[0014] Step (3): Continuous wavelet decomposition and morphological feature calculation of in-situ hyperspectral data;

[0015] Perform continuous wavelet decomposition on the in-situ hyperspectral data of the leaves of karst wetland vegetation at 10 scales (2 1 , 2 2 ,...., 2 10 ), and calculate the angle between the extension direction of the leaf spectral curve and the horizontal direction within adjacent wavelengths of the continuously wavelet decomposed spectrum. This angle finely depicts the morphological information such as the bending degree and change trend of the wavelet decomposed spectral curve. The angle calculation formula is as follows:

[0016]

[0017] where θ is the angle between the extension direction of the leaf wavelet decomposed spectral curve and the horizontal direction within adjacent wavelengths, λ k+1 and λ k are the wavelength numbers, r k+1 and r k are the reflectances of the leaf wavelet decomposed spectrum corresponding to λ k+1 and λ k wavelengths.

[0018] Step (4): Quantify the spectral heterogeneity among different karst wetland vegetation communities;

[0019] Use the improved continuous wavelet information divergence to finely quantify the spectral heterogeneity among different karst wetland vegetation communities from two aspects: spectral reflectance and spectral curve morphological features. The larger the value of the spectral information divergence, the greater the spectral heterogeneity among different karst wetland vegetation communities.

[0020] Step (5): Scale optimization and dimension transformation based on multi-scale continuous wavelet decomposed spectra;

[0021] According to the reflection peaks, absorption valleys and the offset of spectral differences of the multi-scale continuous wavelet decomposed spectra of the leaves of karst wetland vegetation, remove the wavelet decomposition scales with larger offset situations. Rearrange and combine the reflectance features and spectral morphological features (sample, scale, band) of each band of each leaf sample of karst wetland vegetation at the optimized scale respectively.

[0022] For each leaf sample of karst wetland vegetation, taking the scale as the unit, concatenate all the band reflectance data / morphological feature data at the same scale in sequence to form a long sequence containing all the band information at that scale. Then, connect these long sequences in order of scale to construct a new two-dimensional matrix (samples, scale × bands).

[0023] Step (6): Construction of multi-modal feature fusion dataset;

[0024] Fuse the reflectance features and spectral morphological features of each leaf sample of karst wetland vegetation to construct a multi-modal feature fusion dataset. During the fusion process, adopt a strategy based on element-wise addition. This dataset is more robust in complex environments such as karst wetlands.

[0025] Step (7): Spectral response clustering of vegetation leaf spectra based on nutrient element gradients;

[0026] Use the K-means method to perform spectral response clustering on the multi-modal feature fusion dataset based on the above-mentioned leaf nutrient element contents to distinguish the spectral information corresponding to different nutrient elements. Finally, cluster into 6 groups with significant spectral response differences, denoted as clustering spectra.

[0027] Step (8): Mining the potential association between leaf nutrient elements and spectral response mechanism based on dynamically enhanced wavelet two-dimensional correlation spectroscopy;

[0028] The autocorrelation peaks on the synchronous spectrum diagonal of the dynamically enhanced wavelet two-dimensional correlation spectroscopy represent the overall sensitivity of the corresponding spectral bands to the spectral intensity change under the perturbation of nutrient elements. Based on this, we analyze the potential association between the leaf nutrient elements of karst wetland vegetation and the spectral response mechanism, and obtain the sensitive bands of different karst wetland vegetation leaf nutrient elements.

[0029] Step (9): Construction of an adaptive ensemble learning regression model;

[0030] Based on the spectral heterogeneity screened by the improved continuous wavelet information divergence and the internal association between the leaf nutrient elements and the spectral response mechanism mined by the dynamically enhanced wavelet two-dimensional correlation spectroscopy, construct a mechanism-guided adaptive ensemble learning regression model to quantitatively evaluate the inversion performance of the above inversion scheme for karst wetland nutrient elements. Among them, the model includes five base models: PLSR, Random Forest, XGBoost, CatBoost, and Support Vector Machine.

[0031] Use the coefficient of determination (R 2 ) and root mean square error (RMSE) to evaluate the accuracy of the inversion model for each nutrient element of the karst wetland vegetation community.

[0032] Furthermore, in the step (4), the improved continuous wavelet information divergence is used to calculate the spectral differences of each band of different karst wetland vegetations, and the formula is as follows:

[0033]

[0034] R k leaf，i and R k leaf，j are the leaf spectral reflectances / spectral morphological characteristics of the karst wetland vegetation communities i and j at the k wavelength, and S k ij is the corresponding continuous wavelet information divergence value, and n represents the total number of wavelengths; S k represents the continuous wavelet information divergence value of the leaf spectra of any two vegetation communities at the k wavelength, and SID all k is the total value of the continuous wavelet information divergence of the leaf spectra of multiple karst wetland vegetation communities at the k wavelength, and m represents the total number of vegetation communities.

[0035] Furthermore, in the step (8), the dynamic enhanced wavelet two-dimensional correlation spectroscopy technology is used to explore the potential association between nutrient elements and spectral response mechanisms, and the formula is as follows:

[0036]

[0037] Among them, w group is the clustering spectrum, is the dynamic clustering spectrum under the change of nutrient elements, is the average spectrum of the leaves of a certain vegetation community, and we regard it as the reference leaf spectrum of the karst wetland vegetation community.

[0038] X(w group，i , w group，j ) = φ(w group，i , w group，j ) + iψ(w group，i , w group，j )

[0039]

[0040] Among them, w group，i and w group，j are the clustering spectra of the leaves of the karst wetland vegetation corresponding to the i and j wavelengths, and X(w group，i , w group，j ) is the correlation of the dynamic leaf clustering spectra of the two leaf clustering spectral variables w group，i , w group，j , and φ(w group，i , w group，j) is the synchronous spectral intensity of the leaf cluster spectrum, ψ(w group，i , w group，j ) is the asynchronous spectral intensity of the leaf cluster spectrum.

[0041] Compared with the prior art, the present invention is beneficial in that:

[0042] The present invention can deeply analyze the spectral mechanism of leaf nutrient elements of karst wetland vegetation communities from multiple scales and dimensions through improved continuous wavelet information divergence and dynamically enhanced wavelet two-dimensional correlation spectroscopy technology. Among them, the improved continuous wavelet information divergence can accurately quantify the spectral heterogeneity between different karst wetland vegetation, and the dynamically enhanced wavelet two-dimensional correlation spectroscopy technology can accurately mine the intrinsic connection between nutrient elements and spectral mechanism, and obtain the sensitive bands of nutrient elements in the leaves of different karst wetland vegetation, which greatly improves the reliability of spectral analysis and sensitive band selection. The spectral mechanism of karst wetland vegetation mined by the two lays a solid foundation for the subsequent inversion of nutrient element content.

[0043] At the same time, obtaining the spectral mechanism of nitrogen, phosphorus and potassium content in karst wetlands can provide a scientific basis for customized multi-temporal and large-scale fine inversion of karst wetlands.

[0044] The present invention makes up for the defects of unclear causal relationship and underlying mechanism between spectrum and indicators in traditional inversion models, and improves the accuracy and stability of model prediction; the prediction results of nutrient element content provide a scientific basis for monitoring the health of karst wetland vegetation and maintaining wetland ecosystems, and has important application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 Flow chart of the implementation of the present invention.

[0046] Figure 2 Spectral heterogeneity of different karst wetland vegetation communities.

[0047] Figure 3 It is a schematic diagram of the adaptive ensemble learning model implemented in the present invention.

[0048] Figure 4 This is the inversion result diagram of the nutrient element content of different karst wetland vegetation communities. DETAILED DESCRIPTION

[0049] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention. The principles and features of the present invention will be described below in conjunction with the accompanying drawings. The examples given are only for explaining the present invention and are not used to limit the scope of the present invention.

[0050] The following will specifically describe the implementation of the present invention in detail in conjunction with the accompanying drawings.

[0051] Figure 1 As shown, it is a flowchart of the method for analyzing the spectral mechanism of nutrient elements of karst wetland vegetation implemented by the present invention. As shown in the figure, the method includes the following steps:

[0052] Step (1): In-situ hyperspectral measurement of karst wetland vegetation leaves.

[0053] Generally, the in-situ hyperspectral data (350 - 2500 nm) of leaves of different karst wetland vegetation communities are obtained by using the American portable ASD ground object spectrometer. 70 spectral curves are collected for each vegetation sample, and their mean value is taken as the final spectral data of the karst wetland leaf sample.

[0054] Step (2): Determination of nutrient elements in karst wetland vegetation leaves.

[0055] Collect the leaves of karst wetland vegetation, place the leaves of karst wetland vegetation in a dryer, and dry them at 75 °C until they are in an easily grindable state; grind the dried leaves into powder (<0.150 mm), take 0.3000 g of the powder and place it at the bottom of the digestion tube; add 5 ml of concentrated sulfuric acid to the digestion tube, shake well, and place it in a digestion furnace and let it stand for 12 h. Then, gradually add hydrogen peroxide during the heating process until the solution is colorless or clear; remove the digestion tube, and after cooling to room temperature, transfer the digestion solution without loss into a 100 ml volumetric flask, make up the volume with plasma water, shake well, and perform dry filtration for later measurement.

[0056] Then, use a flame photometer to measure the potassium content in the leaves of karst wetland vegetation, and use an automatic discrete chemical analyzer Clever Chem 380 (De ChemTech Company, Germany) to measure the nitrogen and phosphorus contents in the leaves of karst wetland vegetation.

[0057] Step (3): Continuous wavelet decomposition and morphological feature calculation of in-situ hyperspectrum.

[0058] Perform 10-scale (2 1 ,2 2,..., 2 10 ) continuous wavelet decomposition, and calculate the angle between the extension direction of the leaf spectral curve within adjacent wavelengths of the continuous wavelet decomposition spectrum and the horizontal direction. This angle finely depicts morphological information such as the bending degree and change trend of the wavelet decomposition spectral curve. The angle calculation formula 1.1 is as follows:

[0059]

[0060] where θ is the angle between the extension direction of the leaf wavelet decomposition spectral curve within adjacent wavelengths and the horizontal direction, λ k+1 and λ k are the number of wavelengths, r k+1 and r k are the leaf wavelet decomposition spectral reflectances corresponding to λ k+1 and λ k wavelengths.

[0061] Step (4): Quantify the spectral heterogeneity between different karst wetland vegetation communities.

[0062] Use the improved continuous wavelet information divergence to finely quantify the spectral heterogeneity between different karst wetland vegetation communities from two aspects: spectral reflectance and spectral curve morphological characteristics. The larger the value of the spectral information divergence, the greater the spectral heterogeneity between different vegetation communities. The formulas 1.2 - 1.3 are as follows:

[0063]

[0064]

[0065] R k leaf，i and R k leaf，j are the leaf spectral reflectances / spectral morphological characteristics of karst wetland vegetation communities i and j at wavelength k, S k ij is its corresponding continuous wavelet information divergence value, n represents the total number of wavelengths; S 寿 represents the continuous wavelet information divergence value of the leaf spectra of any two vegetation communities at wavelength k, SID all k is the total value of the continuous wavelet information divergence of the leaf spectra of multiple karst wetland vegetation communities at wavelength k, and m represents the total number of vegetation communities.

[0066] Step (3) and Step (4) constitute the improved continuous wavelet information divergence, which can accurately capture the subtle spectral heterogeneity between different karst wetland vegetation communities.

[0067] Step (5): Scale optimization and dimension transformation based on multi-scale continuous wavelet decomposition spectra.

[0068] According to the reflection peaks, absorption valleys, and spectral difference offsets of the multi-scale continuous wavelet decomposition spectra of karst wetland vegetation leaves, remove the wavelet decomposition scales with larger offsets. Rearrange and combine the reflectance characteristics and spectral morphological characteristics (samples, scales, bands) of each band of each karst wetland vegetation leaf sample at the optimized scale respectively.

[0069] For each karst wetland vegetation leaf sample, taking the scale as the unit, concatenate the reflectance data / morphological characteristic data of all bands at the same scale in sequence to form a long sequence containing all band information at this scale. Then, connect these long sequences in sequence according to the scale order to construct a new two-dimensional matrix (samples, scale × bands).

[0070] Step (6): Construction of a multi-modal feature fusion dataset.

[0071] Fuse the reflectance characteristics and spectral morphological characteristics of each karst wetland vegetation leaf sample to construct a multi-modal feature fusion dataset. During the fusion process, adopt a strategy based on element-wise addition. This dataset is more robust in complex environments such as karst wetlands.

[0072] Step (7): Spectral response clustering of vegetation leaves based on nutrient element gradients.

[0073] Based on the above-mentioned leaf nutrient element (nitrogen, phosphorus, and potassium) contents, use the K-means method to perform spectral response clustering on the multi-modal feature fusion dataset to distinguish the spectral information corresponding to different nutrient elements. Finally, cluster into 6 groups with significant spectral response differences, denoted as clustering spectra.

[0074] Step (8): Mining the potential association between leaf nutrient elements and spectral response mechanism based on dynamically enhanced wavelet two-dimensional correlation spectroscopy.

[0075] The autocorrelation peaks on the synchronous spectrum diagonal of the dynamically enhanced wavelet two-dimensional correlation spectrum represent the overall sensitivity of the corresponding spectral bands to spectral intensity changes under the perturbation of nutrient elements. We analyze the potential association between karst wetland vegetation leaf nutrient elements and spectral response mechanism based on this to obtain the sensitive bands of different karst wetland vegetation leaf nutrient elements. The formulas 1.4 - 1.6 of the dynamically enhanced wavelet two-dimensional correlation spectroscopy are as follows:

[0076]

[0077] where, w group is the clustering spectrum, It is the dynamic clustering spectrum under the change of nutrient elements. It is the average spectrum of the leaves of a certain vegetation community, which we regard as the reference leaf spectrum of the karst wetland vegetation community.

[0078] X(w group，i , w group，j ) = φ(w group，i , w group，j ) + iψ(w group，i , w group，j ) (1.5)

[0079]

[0080] Among them, w group，i and w group，j are the clustering spectra of the karst wetland vegetation leaves corresponding to the i-th and j-th wavelengths. X(w group，i , w group，j ) is the correlation of the dynamic leaf clustering spectrum of two leaf clustering spectrum variables w group，i , w group，j . φ(w group，i , w group，j ) is the synchronous spectral intensity of the leaf clustering spectrum, and ψ(w group，i , w group，j ) is the asynchronous spectral intensity of the leaf clustering spectrum.

[0081] Steps (5) to (8) constitute the dynamic enhanced wavelet two-dimensional correlation spectroscopy technology, which can explore the internal relationship between leaf nutrient elements and spectral mechanisms and obtain the sensitive bands of leaf nutrient elements of different karst wetland vegetation.

[0082] Step (9): Construct an adaptive ensemble learning regression model.

[0083] Based on the spectral heterogeneity screened by the improved continuous wavelet information divergence and the internal association between the leaf nutrient elements and spectral response mechanisms mined by the dynamic enhanced wavelet two-dimensional correlation spectroscopy technology, a mechanism-guided adaptive ensemble learning regression model is constructed to quantitatively evaluate the inversion performance of the above inversion scheme for the nutrient elements of the karst wetland. Among them, the model includes five base models: PLSR, Random Forest, XGBoost, CatBoost, and Support Vector Machine.

[0084] Use the coefficient of determination (R 2 ) and root mean square error (RMSE) to evaluate the accuracy of the inversion model of each nutrient element of the karst wetland vegetation community.

[0085] In summary, the present invention explores the spectral mechanism of nutrient elements in leaves of karst wetland vegetation based on improved continuous wavelet information divergence and dynamically enhanced wavelet two-dimensional correlation spectroscopy technology, and develops a method for analyzing the spectral mechanism of nutrient elements in karst wetland vegetation. The improved continuous wavelet information divergence can identify the spectral heterogeneity between different karst wetland vegetation communities, and the dynamically enhanced wavelet two-dimensional correlation spectroscopy technology can explore the potential connection between nutrient element content and spectral mechanism, and obtain the sensitive bands of nutrient elements in leaves of different karst wetland vegetation; on the basis of exploring the spectral mechanism of nutrient elements, a mechanism-guided adaptive integrated learning regression model is constructed, thereby breaking through the defects of unclear causal relationship and underlying mechanism between spectrum and target variable in traditional inversion model, and providing a basic scientific basis for monitoring the healthy growth status of karst wetland vegetation and changes in wetland ecosystems.

[0086] At the same time, the present invention makes up for the defects of unclear causal relationship and underlying mechanism between spectrum and indicators in traditional inversion models, and improves the accuracy and stability of model prediction; the prediction results of nutrient element content provide a scientific basis for monitoring the health of karst wetland vegetation and maintaining wetland ecosystems.

[0087] In addition, the spectral mechanism analysis method proposed in the present invention, including improved continuous wavelet information divergence and dynamically enhanced wavelet two-dimensional correlation spectroscopy technology, is not limited to the spectral analysis of karst wetland vegetation communities, but can also be expanded to other vegetation, water bodies and other spectral analysis fields.

[0088] The specific embodiments described above are exemplary and non-limiting. Various solutions and all changes that can be thought of by those skilled in the art under the inspiration of the disclosure of the present invention are included in the present invention.

Claims

1. A method for analyzing the spectral mechanism of nutrient elements in karst wetland vegetation, characterized in that, It includes the following steps: Step (1): In-situ hyperspectral measurement of karst wetland vegetation leaves; Step (2): Determination of nutrient elements in karst wetland vegetation leaves; Step (3): Continuous wavelet decomposition and morphological feature calculation of in-situ hyperspectrum; In-situ hyperspectral data of karst wetland vegetation leaves were subjected to continuous wavelet decomposition at 10 scales (2 1 , 2 2 ,..., 2 10 ). The angle between the extension direction of the leaf spectral curve and the horizontal direction within adjacent wavelengths of the continuous wavelet decomposition spectrum was calculated. This angle precisely characterized morphological information such as the bending degree and change trend of the continuous wavelet decomposition spectral curve. The angle calculation formula is as follows: where θ is the angle between the extension direction of the continuous wavelet decomposition spectral curve of the blade within adjacent wavelengths and the horizontal direction, λ k+1 and λ k are the number of wavelengths, r k+1 and r k are the continuous wavelet decomposition spectral reflectances of the blade corresponding to λ k+1 and λ k wavelengths; Step (4): Quantifying spectral heterogeneity among different karst wetland vegetation communities; Using the improved continuous wavelet information divergence, finely quantify the spectral heterogeneity among different karst wetland vegetation communities from two aspects of spectral reflectance and spectral curve morphological features, at the in-situ hyperspectrum and multi-scale wavelet decomposition scales; Step (5): Scale optimization and dimension conversion based on multi-scale continuous wavelet decomposition spectra of karst wetland vegetation leaves; According to the offset of the reflection peaks, absorption valleys and spectral differences of the multi-scale continuous wavelet decomposition spectra of karst wetland vegetation leaves, remove the wavelet decomposition scales with larger offset situations, and re-arrange and combine the reflectance features and spectral morphological features (sample, scale, band) of each band of each karst wetland vegetation leaf sample at the optimized scale; Step (6): Construction of a multi-modal feature fusion data set; Fuse the reflectance features and spectral morphological features of each karst wetland vegetation leaf sample to construct a multi-modal feature fusion data set; during the fusion process, adopt a strategy based on element-wise addition; Step (7): Spectral response clustering of vegetation leaves based on nutrient element gradients; Use the K-means method to perform spectral response clustering on the multi-modal feature fusion data set based on the nutrient element content to distinguish the spectral information corresponding to different nutrient elements; Step (8): Mining the potential association between leaf nutrient elements and spectral response mechanism based on the dynamic enhanced wavelet two-dimensional correlation spectroscopy technique; The autocorrelation peaks on the synchronous spectral diagonal line of the dynamically enhanced wavelet two-dimensional correlation spectrum represent the overall sensitivity of the corresponding spectral bands to spectral intensity changes under the perturbation of nutrient elements. Based on this, we analyze the potential association between leaf nutrient elements and spectral response mechanism of karst wetland vegetation, and obtain the sensitive bands of nutrient elements of different karst wetland vegetation leaves; Step (9): Construction of an adaptive ensemble learning regression model; Based on the spectral heterogeneity screened by the improved continuous wavelet information divergence and the internal association between leaf nutrient elements and spectral response mechanism mined by the dynamically enhanced wavelet two-dimensional correlation spectroscopy technique, construct a mechanism-guided adaptive ensemble learning regression model to quantitatively evaluate the inversion performance of the above inversion scheme for karst wetland nutrient elements.

2. The method for analyzing the spectral mechanism of nutrient elements in the vegetation of karst wetlands according to claim 1, wherein In the said step (1), use a portable ASD ground object spectrometer to obtain 350 - 2500 nm in-situ hyperspectral data of different karst wetland vegetation communities. 70 spectral curves are collected for each vegetation sample, and their mean value is taken as the final spectral data of the karst wetland leaf sample.

3. The method for analyzing the spectral mechanism of nutrient elements in karst wetland vegetation according to claim 1, characterized in that In the said step (2), the determination of leaf nutrient elements refers to measuring the potassium content of karst wetland vegetation leaves, measuring the nitrogen content and phosphorus content of karst wetland vegetation leaves.

4. The method for analyzing the spectral mechanism of nutrient elements in karst wetland vegetation according to claim 3, characterized in that For the determination of the nutrient elements in the leaves, the leaves of the karst wetland vegetation are placed in a dryer and dried at 75 °C until they are in an easily ground state; the dried leaves are ground into powder, and 0.3000 g of the powder is placed at the bottom of a digestion tube; 5 ml of concentrated sulfuric acid is added to the digestion tube, shaken well, and placed in a digestion furnace and left standing for 12 h, and then hydrogen peroxide is gradually added during heating until the solution becomes colorless or clear; the digestion tube is taken down, and after cooling to room temperature, the digestion solution is transferred without loss into a 100-ml volumetric flask, fixed volume with plasma water, shaken well, and dry-filtered for later measurement; a flame photometer is used to measure the potassium content in the leaves of the karst wetland vegetation, and an automatic discrete chemical analyzer is used to measure the nitrogen content and phosphorus content in the leaves of the karst wetland vegetation.

5. The method for analyzing the spectral mechanism of nutrient elements in karst wetland vegetation according to claim 1, characterized in that In the step (4), an improved continuous wavelet information divergence is used to calculate the spectral differences of each waveband of different karst wetland vegetations, and the formula is as follows: R k leaf,i and R k leaf,j are the leaf spectral reflectance / spectral morphological characteristics of karst wetland vegetation communities i and j at wavelength k, S k ij is its corresponding continuous wavelet information divergence value, and n represents the total number of wavelengths; S k represents the continuous wavelet information divergence value of the leaf spectra of any two vegetation communities at wavelength k, SID all k is the total continuous wavelet information divergence value of the leaf spectra of multiple karst wetland vegetation communities at wavelength k, and m represents the total number of vegetation communities.

6. The method for analyzing the spectral mechanism of nutrient elements of karst wetland vegetation according to claim 1, wherein In the step (5), for each leaf sample of the karst wetland vegetation, taking the scale as the unit, the reflectance data / morphological feature data of all wavebands at the same scale are concatenated in sequence to form a long sequence containing all waveband information at this scale; then, in the order of the scales, these long sequences are connected in sequence to construct a new two-dimensional matrix (sample, scale × waveband).

7. The method for analyzing the spectral mechanism of nutrient elements of karst wetland vegetation according to claim 1, characterized in that In the step (8), a dynamic enhanced wavelet two-dimensional correlation spectroscopy technology is used to mine the potential correlation between the nutrient elements and the spectral response mechanism, and the formula is as follows: Among them, w group is the clustering spectrum, is the dynamic clustering spectrum under the change of nutrient elements, is the average spectrum of the leaves of a certain vegetation community, which we regard as the reference leaf spectrum of the karst wetland vegetation community. X(w group，i , w group，j ) = φ(w group，i , w group,j ) + iψ(w group，i , w group,j ) where, w group，i and w group,j are the clustering spectra of karst wetland vegetation leaves corresponding to the i-th and j-th wavelengths, X(w group，i , w group，j ) is the correlation of the dynamic leaf clustering spectra of two leaf clustering spectral variables w group，i , w group，j , φ(w group，i , w group，j ) is the synchronous spectral intensity of the leaf clustering spectra, and ψ(w group，i , w group，j ) is the asynchronous spectral intensity of the leaf clustering spectra.

8. The method for analyzing the spectral mechanism of nutrient elements in karst wetland vegetation according to claim 1, characterized in that In the step (9), the model includes five base models: PLSR, Random Forest, XGBoost, CatBoost, and Support Vector Machine; Using the coefficient of determination (R 2 ), the root mean square error (RMSE) to evaluate the inversion model accuracy of the nutrient elements of each karst wetland vegetation community.

Images