Method for high-precision inversion and verification of leaf area index in alpine and high-drought mountain region

Abstract

In consideration of a high-precision inversion and verification method of a leaf area index in a high-cold and high-drought mountain region, data fusion is performed through a fusion model of space-time non-local filtering based on Sentinel-2 remote sensing data and Landsat series remote sensing data images, and fine-resolution remote sensing reflectivity data is constructed; performing remote sensing image preprocessing including the steps of cutting, atmospheric correction, terrain correction, resampling and the like; in combination with indexes such as TVDI, VTCI, NDMVI and SAVI, a new mountain vegetation index THSAMVI is constructed by using a principal component analysis method, and the inversion precision is improved in combination with texture features; then, a machine learning method is utilized to invert LAI; and finally, constructing a test model taking # imgabs0 #, RMSE and MAE as evaluation indexes, and carrying out precision verification on an LAI data set, so that the LAI inversion precision of the region is remarkably improved, and powerful support is provided for researches in the aspects of climate change, ecological quality, ecological evaluation and the like.

Description

Technical Field

[0001] The present invention relates to the application field of remote sensing technology, and in particular to a method for high-precision inversion and verification of leaf area index in high-cold and high-drought mountainous areas. Background Art

[0002] Leaf Area Index (LAI) is an important vegetation parameter that characterizes leaf density and canopy structure. It plays a vital role in climate change research, crop growth model establishment, and carbon and water cycle analysis. Accurate measurement and estimation of LAI is of great significance for assessing the health of ecosystems, monitoring dynamic changes in vegetation, and predicting environmental responses.

[0003] In high-altitude and arid mountainous areas, due to low temperatures, less precipitation and complex terrain, the vegetation types are diverse and the spatial distribution is complex. The LAI inversion in these areas faces many challenges. On the one hand, the special climatic conditions in high-altitude and arid mountainous areas have a significant impact on the growth and distribution of vegetation, making the spatiotemporal changes of LAI more complex; on the other hand, the changes in terrain undulations and surface features will cause changes in the geometric and spectral characteristics of remote sensing images, thereby interfering with the inversion accuracy of LAI.

[0004] At present, the method of inverting LAI based on remote sensing data has achieved relatively remarkable results in flat areas, but in areas with high terrain undulation and high altitude, the temporal and spatial integrity and accuracy of LAI inversion products are still obviously insufficient. This is mainly due to the complex mountain terrain. Remote sensing images are interfered by many factors during the acquisition and processing process, such as terrain shadows, atmospheric influences, surface reflections, etc. These factors will cause the quality of remote sensing images to deteriorate, thereby affecting the inversion accuracy of LAI.

[0005] In order to overcome the shortcomings of existing technologies, domestic and foreign scholars have conducted a lot of research. However, most of these studies are aimed at flat areas or specific types of vegetation. For the LAI inversion problem in high-cold and high-drought mountainous areas, a complete and effective technical solution has not yet been formed. The existing LAI inversion methods often have problems such as unreasonable parameter selection, poor model adaptability, and low inversion accuracy when applied in high-cold and high-drought mountainous areas.

[0006] Specifically, the existing LAI inversion methods often fail to adjust the temperature, hydrological and topographic conditions of high-cold and high-drought mountainous areas in parameter selection, resulting in poor adaptability of the model in high-cold and high-drought mountainous areas. In addition, most of the existing LAI inversion methods are based on a single vegetation index or texture feature for inversion, without fully considering the complexity and diversity of vegetation in high-cold and high-drought mountainous areas, resulting in inaccurate inversion results.

[0007] Therefore, it is urgent to develop a new technical system to improve the accuracy and reliability of LAI inversion in order to solve the LAI inversion problem in high-cold and high-drought mountainous areas. This new technical system needs to fully consider factors such as climate conditions, terrain characteristics and vegetation distribution in high-cold and high-drought mountainous areas, and improve the accuracy and adaptability of LAI inversion by optimizing parameter selection, constructing new vegetation indices and texture features, and other methods. Summary of the invention

[0008] The technical problem to be solved by the present invention is to provide a method for high-precision inversion and verification of leaf area index in high-cold and high-drought mountainous areas, to solve the technical problems faced when performing leaf area index LAI inversion in high-cold and high-drought mountainous areas, especially under these special climatic and terrain conditions, the problem that the existing technical solutions are difficult to provide high-precision vegetation index and high-precision leaf area index technical routes, to overcome the limitation that the existing technology lacks parameter selection optimized for temperature, hydrology and terrain conditions in high-cold and high-drought mountainous areas, resulting in insufficient accuracy of LAI inversion results.

[0009] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for high-precision inversion and verification of leaf area index in high-cold and high-drought mountainous areas, comprising the following steps: Step 1: Obtain remote sensing images, mainly based on Sentinel-2 remote sensing data and Landsat series remote sensing data images; Step 2: Data preprocessing, cropping, atmospheric correction, terrain correction and resampling of remote sensing images to improve image quality and usability; Step 3: Data fusion, through the fusion model of spatiotemporal non-local filtering, data fusion is performed to construct fine-resolution remote sensing reflectance data; Step 4: Obtain vegetation index and texture features. Based on TVDI (Temperature Vegetation Dryness Index), VTCI (Vegetation Temperature Condition Index), NDMVI (Normalized Difference Mountain Vegetation Index), SAVI (Soil Adjusted Vegetation Index), and combined with eight texture features including contrast, difference, entropy, correlation, angular second moment, variance, homogeneity and mean, construct a vegetation index suitable for high-cold and high-drought mountainous areas, the Temperature-Hydrological-Soil Adjusted Mountain Vegetation Index (THSAMVI). Step 5: Leaf area index inversion, using machine learning methods, including support vector machine, random forest, and artificial neural network models, to invert the leaf area index; Step 6: Accuracy verification, constructing the coefficient of determination The accuracy of the leaf area index dataset was verified by using the test model with root mean square error (RMSE) and mean absolute error (MAE) as evaluation indicators and the confusion matrix method.

[0010] In a preferred solution, the data preprocessing in Step 2 also includes using Landsat surface reflectance data to perform atmospheric correction and identification of clouds, cloud shadows and saturated pixels.

[0011] In a preferred solution, the data preprocessing in Step 2 includes the following specific steps: Step 2.1: Landsat surface reflectance data, collect Landsat-8 surface reflectance SR (Spatial Resolution) images, six optical bands, namely blue, green, red, near infrared NIR (Near Infrared), SWIR (Short Wavelength Infrared) 1 and SWIR2 for land cover classification; Step 2.2: Auxiliary remote sensing data products Sentinel series satellite program provides a 6-component Earth observation suite. The first two components, namely Sentinel-1A and Sentinel-2A, are in operation. Sentinel-1 has a synthetic aperture radar SAR (Synthetic Aperture Radar) sensor operating in the C band, and Sentinel-2 is equipped with a multispectral instrument MSI (Multispectral Imaging system) sensor, which collects data in 13 spectral bands in the visible and infrared parts of the electromagnetic spectrum; Step 2.3: Atmospheric correction: by using the radiation transfer model and related parameters, the image is corrected to obtain data closer to the actual surface reflectivity and improve the accuracy of subsequent analysis; Step 2.4: Terrain correction, by eliminating image distortion caused by terrain changes, ensure that the image can accurately represent the true position and characteristics of surface objects, including using the digital elevation model DEM (Digital Elevation Model) to correct the geometric distortion in the image, ensure that each pixel corresponds to the actual position of the surface, and at the same time reduce the spectral changes in the image and improve the spectral consistency of the image, making subsequent analysis more accurate and reliable.

[0012] In the preferred scheme, the spatial and temporal nonlocal filter fusion model STNLFFM (Spatial and Temporal Nonlocal Filter based data Fusion Method) in Step 3 provides a new transformation relationship between multi-temporal fine-resolution reflectivity images with the help of multi-temporal coarse-resolution reflectivity data, thereby making full use of the high spatiotemporal redundancy in the remote sensing image sequence to obtain the final fused image result for unified auxiliary verification of remote sensing data products.

[0013] In a preferred embodiment, the fusion model of spatiotemporal non-local filtering in Step 3 includes searching for similar pixels, calculating weights and regression coefficients, and applying the weights and regression coefficients to the high-resolution reflectivity image to generate the high-resolution reflectivity of the predicted date.

[0014] In a preferred embodiment, the implementation of the STNLFFM model comprises the following steps: Step 3.1: For each pixel in the reference high-resolution image, search for similar pixels in the image; Step 3.2: Weight W is used to calculate all similar pixels; Step 3.3: Calculate the regression coefficient for each similar pixel based on the available coarse-resolution reflectance image; Step 3.4: Apply the weights and regression coefficients to the two high-resolution reflectance images collected on the reference date to produce the high-resolution reflectance for the desired prediction date.

[0015] In a preferred solution, in Step 4, a new mountain vegetation index THSAMVI is constructed using the principal component analysis method PCA (Principal Component Analysis), which can fully reflect the impact of temperature, hydrology, topography and soil on the vegetation index.

[0016] In the preferred scheme, the texture features in the Step 4 include contrast CON (Contrast), dissimilarity DIS (Dissimilarity), entropy ENT (Entropy), correlation COR (Correlation), angular second moment SEC (AngularSecond Moment), variance VAR (Variance), homogeneity HOM (Homogeneity) and mean MEA (Mean). The gray level co-occurrence matrix is ​​used to select the above eight texture parameters to characterize the texture spatial structure of the study area. The inversion accuracy of remote sensing images is improved by combining vegetation index with texture features.

[0017] In a preferred solution, the machine learning method in Step 5 uses a non-parametric regression method to learn and establish a regression model from a training set constructed from multiple vegetation indices for inverting the leaf area index.

[0018] In a preferred solution, the accuracy verification in Step 6 also includes using a confusion matrix to evaluate the classification model performance to further verify the accuracy of the leaf area index inversion.

[0019] The method for high-precision inversion and verification of leaf area index in high-cold and high-drought mountainous areas provided by the present invention has the following beneficial effects: The present invention solves the technical problems faced when performing leaf area index LAI inversion in high-cold and high-drought mountainous areas, especially under these special climate and terrain conditions, the existing technical solutions are difficult to provide high-precision vegetation index and high-precision leaf area index technical route problems, overcomes the limitation of the existing technology that lacks parameter selection optimized for temperature, hydrology and terrain conditions in high-cold and high-drought mountainous areas, resulting in insufficient accuracy of LAI inversion results; the present invention constructs model parameters for high-cold and high-drought climate conditions, solves the problem of high-precision LAI inversion in high-cold and high-drought mountainous areas, improves the accuracy and reliability of inversion results, and makes the results of the LAI inversion model more accurate; the present invention adjusts the parameters when inverting the LAI model to make it more suitable for places with rugged mountainous terrain; the present invention reduces the interference of terrain factors on remote sensing inversion LAI results by adjusting model parameters and introducing new vegetation indices; the present invention constructs a new vegetation index THSAMVI, combined with texture Characteristics, and the use of machine learning models for inversion significantly improves the inversion accuracy of LAI; the present invention develops a specialized LAI inversion method for the special climate and hydrological conditions in high-cold and high-mountainous areas. The technical solution is suitable for high-cold and high-drought mountainous areas, and provides strong support for vegetation monitoring and ecological research in these areas; the proposal and implementation of the inversion and verification method of the present invention will help promote scientific research on climate change, ecological quality and ecological evaluation, and provide a scientific basis for ecological protection and environmental management; the inversion and verification method of the present invention effectively reduces overfitting and improves the generalization ability of the model, and is suitable for processing large-scale data and high-dimensional features. Machine learning methods such as random forests also have the advantages of processing missing values ​​and evaluating feature importance; the technical solution of the present invention shows strong flexibility and adaptability in practical applications, and can be fine-tuned according to the specific environmental characteristics of different regions to ensure the accuracy and practicality of LAI inversion, and provide technical support for regional ecological management and sustainable development. BRIEF DESCRIPTION OF THE DRAWINGS The present invention will be further described below in conjunction with the accompanying drawings and implementation examples: Figure 1 It is a schematic diagram of the technical process of the present invention. DETAILED DESCRIPTION

[0020] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments: Example 1 like Figure 1 As shown in the figure, the method of high-precision inversion and verification of leaf area index in high-cold and high-drought mountainous areas includes the following steps: Step 1: Obtain remote sensing images, mainly based on Sentinel-2 remote sensing data and Landsat series remote sensing data images; Step 2: Data preprocessing, cropping, atmospheric correction, terrain correction and resampling of remote sensing images to improve image quality and usability; Step 3: Data fusion, through the fusion model of spatiotemporal non-local filtering, data fusion is performed to construct fine-resolution remote sensing reflectance data; Step 4: Obtain vegetation index and texture features. Based on TVDI, VTCI, NDMVI, SAVI index, and combined with eight texture features including contrast CON, difference DIS, entropy ENT, correlation COR, angular second moment SEC, variance VAR, homogeneity HOM and mean MEA, construct the vegetation index THSAMVI suitable for high-cold and high-drought mountainous areas. Step 5: Leaf area index inversion, using machine learning methods, including support vector machine, random forest, and artificial neural network models, to invert the leaf area index; Step 6: Accuracy verification, build a determination coefficient , root mean square error RMSE and mean absolute error MAE as the test model evaluation indicators, and the accuracy of the leaf area index dataset was verified.

[0021] In this embodiment, the data preprocessing in Step 2 also includes using Landsat surface reflectance data to perform atmospheric correction and identification of clouds, cloud shadows, and saturated pixels.

[0022] Furthermore, the data preprocessing of Step 2 includes the following specific steps: Step 2.1: Landsat surface reflectance data, collect Landsat-8 surface reflectance SR images, and six optical bands, namely blue, green, red, near infrared NIR, SWIR1 and SWIR2, are used to extract vegetation indices; Step 2.2: Auxiliary remote sensing data products Sentinel series satellite program provides a 6-component Earth observation suite. The first two components, namely Sentinel-1A and Sentinel-2A, are in operation. Sentinel-1 has a synthetic aperture radar SAR sensor operating in the C-band, and Sentinel-2 is equipped with a multi-spectral instrument MSI sensor, which collects data in 13 spectral bands in the visible and infrared parts of the electromagnetic spectrum. Step 2.3: Atmospheric correction: by using the radiation transfer model and related parameters, the image is corrected to obtain data closer to the actual surface reflectivity and improve the accuracy of subsequent analysis; Step 2.4: Terrain correction, by eliminating image distortion caused by terrain changes, ensure that the image can accurately represent the true position and characteristics of surface objects, including using the digital elevation model (DEM) to correct the geometric distortion in the image, ensuring that each pixel corresponds to the actual position on the surface, while reducing spectral changes in the image and improving the spectral consistency of the image, making subsequent analysis more accurate and reliable.

[0023] Furthermore, the spatiotemporal non-local filtering fusion model STNLFFM in Step 3 provides a new transformation relationship between multi-temporal fine-resolution reflectivity images with the help of multi-temporal coarse-resolution reflectivity data, thereby making full use of the high spatiotemporal redundancy in the remote sensing image sequence to obtain the final fused image result for unified auxiliary verification of remote sensing data products.

[0024] Furthermore, the fusion model of spatiotemporal non-local filtering in Step 3 includes searching for similar pixels, calculating weights and regression coefficients, and applying the weights and regression coefficients to the high-resolution reflectance image to generate a high-resolution reflectance of the predicted date.

[0025] Furthermore, the implementation of the STNLFFM model includes the following steps: Step 3.1: For each pixel in the reference high-resolution image, search for similar pixels in the image; Step 3.2: Weight W is used to calculate all similar pixels; Step 3.3: Calculate the regression coefficient for each similar pixel based on the available coarse-resolution reflectance image; Step 3.4: Apply the weights and regression coefficients to the two high-resolution reflectance images collected on the reference date to produce the high-resolution reflectance for the desired prediction date.

[0026] Furthermore, in Step 4, a new vegetation index THSAMVI is constructed using the principal component analysis method PCA, which can fully reflect the impact of temperature, hydrology, topography and soil on the vegetation index.

[0027] Furthermore, the texture features in Step 4 include contrast CON, difference DIS, entropy ENT, correlation COR, angular second moment SEC, variance VAR, homogeneity HOM and mean MEA. The gray level co-occurrence matrix is ​​used to select the above eight texture parameters to characterize the texture spatial structure of the study area. The inversion accuracy of remote sensing images is improved by combining vegetation index with texture features.

[0028] Furthermore, the machine learning method in Step 5 uses a non-parametric regression method to learn and establish a regression model from a training set constructed from multiple vegetation indices for inverting the leaf area index.

[0029] Furthermore, the accuracy verification in Step 6 is The test model with RMSE and MAE as evaluation indicators further verifies the accuracy of leaf area index inversion.

[0030] Example 2 In another preferred embodiment, based on Example 1, refer to Figure 1 This embodiment describes in detail the specific implementation steps of the method for high-precision inversion and verification of leaf area index in high-cold and high-drought mountainous areas as follows: 1. Remote sensing image acquisition and preprocessing The present invention mainly performs leaf area index inversion based on Sentinel-2 remote sensing data and Landsat series remote sensing data images. First, remote sensing images need to be preprocessed to improve image quality and usability. The main steps of preprocessing include geometric correction, radiation correction and cloud removal: (1) Geometric correction: Use accurate geographic reference information to perform geometric correction on remote sensing images to ensure that each pixel in the image accurately corresponds to its actual location on the surface; (2) Radiation correction: The image is corrected through radiation transfer models and related parameters to eliminate the influence of the atmosphere on the surface reflectance spectrum, thereby obtaining data that is closer to the actual surface reflectance; (3) Cloud removal: Use advanced technologies such as the CF-Mask (Collaborative Filtering-Mask) algorithm to identify and remove clouds, cloud shadows, and saturated pixels in images to ensure image integrity and availability; In addition, the present invention also collects Landsat-8 surface reflectance (SR) images, which have been atmospherically corrected by the Landsat surface reflectance code atmospheric correction method; the images include six optical bands: blue, green, red, near infrared (NIR), SWIR1 and SWIR2, which are used for subsequent vegetation index calculations.

[0031] 2. Fusion model of spatiotemporal non-local filtering In order to construct fine-resolution remote sensing reflectance data, the present invention adopts a spatiotemporal non-local filtering fusion model (STNLFFM). This method provides a new transformation relationship between multi-temporal fine-resolution reflectance images with the help of multi-phase coarse-resolution reflectance data, making full use of the spatiotemporal redundancy in the remote sensing image sequence to obtain the final fused image result.

[0032] The implementation of the STNLFFM algorithm consists of three main steps: first, for each pixel in the reference high-resolution image, a search is made for pixels similar to it in the image; second, weights are calculated for all similar pixels; and finally, a regression coefficient is calculated for each similar pixel based on the available coarse-resolution reflectance image. The weights and regression coefficients are applied to the two high-resolution reflectance images acquired on the reference date to produce high-resolution reflectance for the desired forecast date.

[0033] 3. Obtain vegetation index and texture characteristics suitable for high-cold and high-drought mountainous areas In high-cold and high-drought mountainous areas, the selection of vegetation index and texture features is crucial for the inversion of leaf area index. The present invention fully considers the influence of moisture, temperature, topography and soil on vegetation index, and constructs the vegetation index THSAMVI suitable for high-cold and high-drought mountainous areas using principal component analysis.

[0034] THSAMVI is a new index constructed based on TVDI (Temperature Vegetation Drought Index), VTCI (Temperature Condition Index of Vegetation), NDMVI (Normalized Difference Mountain Vegetation Index) and SAVI (Soil Adjusted Vegetation Index), further considering temperature, hydrological and topographic conditions; it can fully reflect the impact of these environmental factors on vegetation indices and improve the accuracy of leaf area index inversion.

[0035] In addition, the present invention also selects eight texture features, including contrast (CON), difference (DIS), entropy (ENT), correlation (COR), angular second moment (SEC), variance (VAR), homogeneity (HOM) and mean (MEA), and extracts them through gray-level co-occurrence matrix. These texture features are helpful to improve the inversion accuracy of remote sensing images, especially under complex terrain and climate conditions.

[0036] 4. Using machine learning to invert leaf area index The present invention uses machine learning methods such as support vector machine, random forest and artificial neural network model to invert the leaf area index; these methods have strong nonlinear fitting capabilities and can learn the complex relationship between the leaf area index and multiple vegetation indices and texture features: (1) Support Vector Machine (SVM): It separates data points of different categories by finding a hyperplane and maximizes the interval between two categories. SVM can process nonlinear data and map data in low-dimensional space to high-dimensional space, thereby finding the best separation hyperplane. (2) Random Forest: Build multiple decision trees and combine their results to improve the accuracy and robustness of the model; Random Forest randomly selects samples and features during training and builds multiple decision trees for independent predictions. The final output is the average of the prediction results of all trees; (3) Artificial neural network model: A typical feedforward artificial neural network consisting of an input layer, one or more hidden layers, and an output layer; the weights are adjusted through the back-propagation algorithm to minimize the prediction error, thereby optimizing the model performance.

[0037] 5. Accuracy verification In order to verify the accuracy and reliability of the technical system proposed by the present invention, the present invention constructs a coefficient of determination ( ), root mean square error (RMSE) and mean absolute error (MAE) as evaluation indicators to verify the accuracy of the data set: (1) : Measures the degree of variation of the model's explanatory variables. The closer the value is to 1, the stronger the explanatory power of the model; (2) RMSE: It represents the square root of the average of the squared errors between the predicted value and the actual value, and is more sensitive to large errors. (3) MAE: The average of the absolute values ​​of all prediction errors, which directly reflects the average deviation between the predicted value and the actual value.

[0038] In the preferred solution, the data preprocessing in Step 2 also includes using Landsat surface reflectance data for atmospheric correction and identification of clouds, cloud shadows and saturated pixels; the above settings are intended to improve data quality and ensure the accuracy of subsequent analysis; at the same time, advanced algorithms are introduced to automate the processing flow, improve processing efficiency, and lay a solid foundation for subsequent vegetation index extraction.

[0039] In the preferred scheme, the spatiotemporal non-local filtering fusion model STNLFFM in Step 3 provides a new transformation relationship between multi-temporal fine-resolution reflectance images with the help of multi-phase coarse-resolution reflectance data, thereby making full use of the high spatiotemporal redundancy in the remote sensing image sequence to obtain the final fused image result for unified auxiliary verification of remote sensing data products; the above settings not only significantly improve the fusion accuracy and efficiency, but also enhance the adaptability of the model to complex scenes; the fused images perform well in detail retention, edge clarity and color consistency, providing reliable support for the subsequent application of remote sensing data.

[0040] In the preferred scheme, the fusion model of spatiotemporal non-local filtering in Step 3 includes searching for similar pixels, calculating weights and regression coefficients, and applying the weights and regression coefficients to the high-resolution reflectance image to generate the high-resolution reflectance of the predicted date; the above settings can significantly improve the spatial resolution and temporal continuity of the image, effectively reduce noise interference, make the generated predicted image closer to the actual situation, and provide more reliable data support for fields such as remote sensing monitoring and environmental assessment.

[0041] In the preferred scheme, the principal component analysis method PCA is used in Step 4 to construct a new vegetation index THSAMVI, which can fully reflect the impact of temperature, hydrology, topography and soil on the vegetation index; the above settings not only improve the comprehensiveness and accuracy of the vegetation index, but also provide strong data support for the realization of precision agricultural management, environmental monitoring and ecological assessment, and significantly enhance the scientificity and efficiency of decision-making.

[0042] In the preferred scheme, the texture features in Step 4 include contrast CON, difference DIS, entropy ENT, correlation COR, angular second moment SEC, variance VAR, homogeneity HOM and mean MEA. The gray level co-occurrence matrix is ​​used to select the above eight texture parameters to characterize the texture spatial structure of the study area. The remote sensing image inversion accuracy is improved by combining the vegetation index with the texture features. The above settings enable the model to show higher accuracy in identifying different surface cover types, especially in distinguishing vegetation types and their health status, which significantly improves the parsing ability of remote sensing data and provides strong technical support for environmental monitoring and resource assessment.

[0043] In summary, the method proposed in the present invention for high-precision inversion and verification of leaf area index in high-cold and high-drought mountainous areas solves the technical problems faced in performing leaf area index LAI inversion in high-cold and high-drought mountainous areas, especially under these special climatic and topographical conditions, the problem that the existing technical solutions are difficult to provide high-precision vegetation index and high-precision leaf area index technical routes, and overcomes the limitation of the existing technology that lacks parameter selection optimized for temperature, hydrology and topographical conditions in high-cold and high-drought mountainous areas, resulting in insufficient accuracy of LAI inversion results; the present invention specifically performs LAI inversion in high-cold and high-drought mountainous areas. Due to the special climatic and topographical conditions in this area, the traditional LAI inversion This method is often difficult to achieve high accuracy, and it is novel in the selection of target areas; it combines Sentinel-2 and Landsat series remote sensing data, and uses the spatiotemporal non-local filtering fusion model (STNLFFM) for data fusion. This method can significantly improve the resolution and availability of data when processing remote sensing data under complex terrain and climatic conditions, showing the novelty of data processing technology; based on existing vegetation indices (such as TVDI, VTCI, NDMVI, SAVI), the present invention constructs a new mountain vegetation index THSAMVI through principal component analysis (PCA), which fully considers the effects of temperature, hydrology and terrain conditions on the vegetation index. The influence of vegetation provides a more comprehensive information source for LAI inversion; and texture features (such as contrast, difference, entropy, etc.) are introduced in the inversion process. These features can capture subtle differences in remote sensing images, thereby improving the recognition accuracy of LAI inversion. This approach is relatively rare in similar studies; the present invention proposes a complete technical route from remote sensing image acquisition, preprocessing, vegetation index and texture feature extraction, machine learning inversion to accuracy verification, which provides a systematic solution for LAI inversion in high-altitude and high-drought mountainous areas; in view of the special conditions of high-altitude and high-drought mountainous areas, the scheme has made innovative adjustments to the parameters of the inversion model, making the model more suitable for this area and significantly improving the accuracy of the inversion. ; In view of the particularity of high-altitude cold and high-arid mountainous areas, the scheme has developed a specialized LAI inversion method, filling the technical gap in this field and providing strong technical support for related scientific research; the present invention creatively improves the effect of LAI inversion by combining multi-source remote sensing data, advanced data processing technology, new vegetation index and texture features and other technical means, and provides new ideas for vegetation monitoring and ecological research under similar complex terrain and climatic conditions; it not only achieves technological breakthroughs, but also demonstrates strong adaptability and accuracy in practical applications, providing a scientific basis for ecological protection and resource management, and is expected to be promoted globally to promote the further development of ecological environment monitoring technology.

Claims

1. Considering the method of high-precision inversion and verification of leaf area index in high-cold and high-drought mountainous areas, it is characterized by: The method comprises the following steps: Step 1: Obtain remote sensing images, mainly based on Sentinel-2 remote sensing data and Landsat series remote sensing data images; Step 2: Data preprocessing, cropping, atmospheric correction, terrain correction and resampling of remote sensing images to improve image quality and usability; Step 3: Data fusion, through the fusion model of spatiotemporal non-local filtering, data fusion is performed to construct fine-resolution remote sensing reflectance data; Step 4: Obtain vegetation index and texture features. Based on TVDI, VTCI, NDMVI, SAVI index, and combined with eight texture features including contrast, difference, entropy, correlation, angular second moment, variance, homogeneity and mean, construct the vegetation index THSAMVI suitable for high-cold and high-drought mountainous areas. Step 5: Leaf area index inversion, using machine learning methods, including support vector machine, random forest, and artificial neural network models, to invert the leaf area index; Step 6: Accuracy verification, build a determination coefficient , root mean square error RMSE and mean absolute error MAE as the test model for evaluation indicators, and the accuracy of the leaf area index dataset was verified.

2. The method for high-precision inversion and verification of leaf area index in high-cold and high-drought mountainous areas according to claim 1 is characterized by: The data preprocessing in Step 2 also includes using Landsat surface reflectance data to perform atmospheric correction and identification of clouds, cloud shadows, and saturated pixels.

3. The method for high-precision inversion and verification of leaf area index in high-cold and high-drought mountainous areas according to claim 2 is characterized in that: The data preprocessing of Step 2 includes the following specific steps: Step 2.1: Landsat surface reflectance data, collect Landsat-8 surface reflectance SR images, and six optical bands, namely blue, green, red, near infrared NIR, SWIR1 and SWIR2, are used to extract vegetation indices; Step 2.2: Auxiliary remote sensing data products Sentinel series satellite program provides a 6-component Earth observation suite. The first two components, namely Sentinel-1A and Sentinel-2A, are in operation. Sentinel-1 has a synthetic aperture radar SAR sensor operating in the C-band, and Sentinel-2 is equipped with a multi-spectral instrument MSI sensor, which collects data in 13 spectral bands in the visible and infrared parts of the electromagnetic spectrum. Step 2.3: Atmospheric correction: by using the radiation transfer model and related parameters, the image is corrected to obtain data closer to the actual surface reflectivity and improve the accuracy of subsequent analysis; Step 2.4: Terrain correction, by eliminating image distortion caused by terrain changes, ensure that the image can accurately represent the true position and characteristics of surface objects, including using the digital elevation model (DEM) to correct the geometric distortion in the image, ensuring that each pixel corresponds to the actual position on the surface, while reducing spectral changes in the image and improving the spectral consistency of the image, making subsequent analysis more accurate and reliable.

4. The method for high-precision inversion and verification of leaf area index in high-cold and high-drought mountainous areas according to claim 1, characterized in that: The spatiotemporal non-local filtering fusion model STNLFFM in the Step 3 provides a new transformation relationship between multi-temporal fine-resolution reflectivity images with the help of multi-temporal coarse-resolution reflectivity data, thereby making full use of the high spatiotemporal redundancy in the remote sensing image sequence to obtain the final fused image result for unified auxiliary verification of remote sensing data products.

5. The method for high-precision inversion and verification of leaf area index in high-cold and high-drought mountainous areas according to claim 4 is characterized by: The fusion model of spatiotemporal non-local filtering in Step 3 includes searching for similar pixels, calculating weights and regression coefficients, and applying the weights and regression coefficients to the high-resolution reflectance image to generate the high-resolution reflectance of the predicted date.

6. The method for high-precision inversion and verification of leaf area index in high-cold and high-drought mountainous areas according to claim 4 is characterized in that: The implementation of the STNLFFM model includes the following steps: Step 3.1: For each pixel in the reference high-resolution image, search for similar pixels in the image; Step 3.2: Weight W is used to calculate all similar pixels; Step 3.3: Calculate the regression coefficient for each similar pixel based on the available coarse-resolution reflectance image; Step 3.4: Apply the weights and regression coefficients to the two high-resolution reflectance images collected on the reference date to produce the high-resolution reflectance for the desired prediction date.

7. The method for high-precision inversion and verification of leaf area index in high-cold and high-drought mountainous areas according to claim 1, characterized in that: In the Step 4, the principal component analysis method PCA is used to construct a new vegetation index THSAMVI, which can fully reflect the impact of temperature, hydrology, topography and soil on the vegetation index.

8. The method for high-precision inversion and verification of leaf area index in high-cold and high-drought mountainous areas according to claim 7, characterized in that: The texture features in Step 4 include contrast CON, difference DIS, entropy ENT, correlation COR, angular second moment SEC, variance VAR, homogeneity HOM and mean MEA. The gray level co-occurrence matrix is ​​used to select the above eight texture parameters to characterize the texture spatial structure of the study area. The inversion accuracy of remote sensing images is improved by combining vegetation index with texture features.

9. The method for high-precision inversion and verification of leaf area index in high-cold and high-drought mountainous areas according to claim 1, characterized in that: The machine learning method in Step 5 uses a non-parametric regression method to learn and establish a regression model from a training set constructed from multiple vegetation indices for inverting the leaf area index.

10. The method for high-precision inversion and verification of leaf area index in high-cold and high-drought mountainous areas according to claim 1, characterized in that: The accuracy verification in Step 6 is performed using , RMSE and MAE were used as the test model for evaluation indicators to further verify the accuracy of leaf area index inversion.

Images