Permafrost organic carbon content inversion method, medium and equipment
The construction of a deep convolutional neural network through multi-source remote sensing data and deep learning methods has solved the problem of soil organic carbon estimation in the permafrost area of the Qinghai-Tibet Plateau, and achieved higher-precision spatial distribution and reserve estimation of organic carbon.
Patent Information
- Application Number
- CN202411377714.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-30
- Publication Date
- 2025-03-07
AI Technical Summary
The prior art is difficult to accurately estimate the soil organic carbon content and spatial distribution of the permafrost areas of the Qinghai-Tibet Plateau under large-scale and multi-environmental factors, especially when the organic carbon density point data is scarce and the characteristics of the permafrost are complex.
Multi-source remote sensing data combined with deep learning methods are used to build a dual-input deep convolutional neural network, and feature sets are constructed through various factors such as spectral angle, climate, vegetation, soil and topographic characteristics, rich spatial context information is extracted, and the most important channel and spatial information is screened to achieve high-precision estimation of organic carbon density.
A higher accuracy estimation of the spatial distribution and reserves of soil organic carbon in the permafrost area of the Qinghai-Tibet Plateau has been achieved, and the problems of poor fitting of the model and incomplete consideration of the characteristics of the permafrost in the prior art are overcome, and a high-precision and high-resolution data set of long-time series permafrost organic carbon density is provided.
Smart Images
Figure CN119577353A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of frozen soil carbon reserve estimation, and more specifically to a method, medium and equipment for inverting the organic carbon content of permafrost. Background Art
[0002] As an important part of the cryosphere, frozen soil is widely distributed in high latitudes and high altitudes in the Northern Hemisphere, of which permafrost accounts for about 24% of the land surface in the Northern Hemisphere and seasonal frozen soil accounts for about 50.5%. The soil carbon pool in the frozen soil area accounts for more than 50% of the global soil carbon pool, which is the largest carbon pool in the terrestrial ecosystem. Its content is about twice the carbon content in the atmosphere. The slight change in the soil organic carbon (SOC) reserves will have an important impact on the atmospheric CO2 concentration, so it plays an important role in the global carbon cycle. The Qinghai-Tibet Plateau is known as the "Roof of the World" and the "Third Pole of the World". It is the highest and largest plateau in the world. The high altitude and harsh climate have formed the highest and largest permafrost area in the world's mid- and low-latitude regions. Compared with other permafrost areas at the same latitude in the Northern Hemisphere, the permafrost in the Qinghai-Tibet Plateau has the characteristics of high temperature, thin thickness, poor thermal stability, limited and uneven distribution, etc., which makes it more sensitive to climate change and is one of the most sensitive regions to global climate change. Studies have shown that the total amount of soil organic carbon in the Qinghai-Tibet Plateau is about 50.43 Pg, of which about 37.21 Pg is stored in permafrost. With climate warming, the permafrost areas of the Qinghai-Tibet Plateau have undergone significant degradation in the past few decades. The degradation of permafrost has reduced the diversity and stability of soil microorganisms, which will not only affect infrastructure such as the Qinghai-Tibet Railway, but also cause the release of organic carbon in the permafrost area in the form of carbon dioxide or methane, triggering a strong permafrost carbon-climate feedback and exacerbating the greenhouse effect. Therefore, it is important to accurately estimate the SOC content and spatial distribution in the permafrost areas of the Qinghai-Tibet Plateau.
[0003] Most existing SOC estimation methods are based on spatial statistical methods (such as spatial kriging, geographically weighted regression) or machine learning methods (such as random forests, support vector machines). However, under the interaction of large-scale and multi-environmental factors, soil properties and environmental factors have a nonlinear relationship. As the number of features increases, such methods are difficult to apply to large-scale multi-parameter nonlinear fitting; and the complex nonlinear properties of SOC and its relationship with the environment often show high variability, which is more significant at large scales. Compared with statistical models and machine learning models, deep learning models can extract relevant nonlinear and complex attributes from multi-source data to improve the learning process, identify complex spatial patterns related to big data, and learn rich spatial context information, thereby overcoming the nonlinear problem between SOC and multi-source data, and achieving better fitting effects. At present, many studies have used deep learning methods combined with remote sensing data to predict SOC, such as Liu et al. combined with time-space-spectral data fusion features and used convolutional neural networks to calculate SOCS in Northeast China; Odebiri et al. used CAE-DNN hybrid neural networks to simulate the spatial distribution of SOC in South Africa. However, considering the complex properties of permafrost, the limitations of the number and distribution of SOC points, and the spatiotemporal heterogeneity of data, further research is still needed to estimate the soil organic carbon content in the permafrost areas of the Qinghai-Tibet Plateau using deep learning methods. Summary of the invention
[0004] The purpose of the present invention is to provide a method, medium and equipment for inversion of permafrost organic carbon content, which can obtain the spatial distribution and reserves of soil organic carbon in permafrost areas with higher accuracy and analyze the dynamic changes of organic carbon more comprehensively and effectively.
[0005] The present invention provides a method for inverting the organic carbon content in permafrost, comprising the following steps: S01: acquiring organic carbon density point data and regional characteristic data in a study area, preprocessing the regional characteristic data, and obtaining a feature set; S02: normalizing the organic carbon density point data according to the feature set, and obtaining normalized organic carbon density point data; S03: sampling the feature set according to the normalized organic carbon density point data, and obtaining a sample set, and dividing the sample set to obtain a training set, a test set, and a validation set; S04: training different deep learning models using the training set, the test set, and the validation set, and obtaining different soil organic carbon estimation models; performing comparative tests and screening using a machine learning algorithm according to different soil organic carbon estimation models, and obtaining an optimal estimation model; obtaining the spatial distribution of organic carbon density according to the optimal estimation model and the organic carbon density point data; S05: obtaining the organic carbon reserve according to the spatial distribution of organic carbon density.
[0006] Furthermore, the organic carbon density point data of the above-mentioned permafrost organic carbon content inversion method includes organic carbon density point data at depths of 0-30cm, 0-50cm, 0-100cm, 0-200cm, and 0-300cm.
[0007] Furthermore, the regional characteristic data of the above-mentioned permafrost organic carbon content inversion method include remote sensing images, public data sets, surface temperature data, soil data sets and space shuttle geomorphic radar mission data; the regional characteristic data are preprocessed, specifically including: S011: declouding the remote sensing image to obtain spectral index data; S012: according to the public data set, the annual average rainfall data is obtained, the surface temperature data is converted into 0cm real surface temperature data, and the annual average surface temperature data is obtained. According to the annual average rainfall data and the annual average surface temperature data, the climate index data is obtained; S013: according to the spectral index data, the ecological index data is obtained, as shown in the formula: , , , , , , , , , , Among them, NDVI is the normalized difference vegetation index, GNDVI is the green normalized index, SAVI is the soil adjusted vegetation index, OSAVI is the optimized soil adjusted vegetation index, EVI is the enhanced vegetation index, RVI is the ratio vegetation index, TVI is the triangular vegetation index, DVI is the vegetation difference index, MSAVI is the adjusted soil brightness vegetation index, and WET is the wetness index; S014: Filter the soil data set to obtain soil attribute data; S015: Obtain terrain feature data based on the space shuttle landform radar mission data; S016: Normalize the spectral index data, annual average rainfall data, annual average surface temperature data, normalized vegetation index, green normalized index, soil adjusted vegetation index, optimized soil adjusted vegetation index, enhanced vegetation index, ratio vegetation index, triangle vegetation index, vegetation difference index, adjusted soil brightness vegetation index, moisture index, soil attribute data and terrain feature data, sample them to a resolution of 1 km, and project them to the same geographic coordinate system to obtain a feature set.
[0008] Furthermore, the feature set of the above-mentioned permafrost organic carbon content inversion method includes spectral index data, climate index data, ecological index data, soil property data and terrain feature data; the spectral index data includes red light band data, green light band data, blue light band data, near infrared band data, first mid-infrared band data and second mid-infrared band data; the climate index data includes annual average rainfall data and annual average surface temperature data; the ecological index data includes normalized vegetation index, green normalized index, soil adjusted vegetation index, optimized soil adjusted vegetation index, enhanced vegetation index, ratio vegetation index, triangle vegetation index, vegetation difference index, adjusted soil brightness vegetation index and humidity index; soil property data includes sand content, gravel content, soil moisture content, soil pH value, soil bulk density, soil texture, soil components and soil biological communities; terrain feature data includes altitude, slope, slope aspect and terrain shadow; and the feature set is preprocessed.
[0009] Furthermore, step S02 of the above-mentioned permafrost organic carbon content inversion method specifically includes: S021: based on the organic carbon density point data and feature set, the spectral vector of each organic carbon density point in its corresponding sampling year is composed of red, green, blue, near infrared and two mid-infrared bands to obtain a first spectral vector; S022: based on the organic carbon density point data and feature set, the spectral vector of each organic carbon density point during the baseline period is composed of red, green, blue, near infrared and two mid-infrared bands to obtain a second spectral vector; S023: based on the first spectral vector and the second spectral vector, the spectral angle and spectral distance of each organic carbon density point are obtained, as shown in the formula: , , X:[R, G, B, NIR, SWIR1, SWIR2], Y:[R, G, B, NIR, SWIR1, SWIR2], Wherein, D is the spectral distance, θ is the spectral angle, X is the first spectral vector, Y is the second spectral vector, R is the red light band data, G is the green light band data, B is the blue light band data, NIR is the near infrared band data, SWIR1 is the first mid-infrared band data, and SWIR2 is the second mid-infrared band data; S024: When the spectral distance is less than a preset spectral distance value and the spectral angle is greater than a preset spectral angle value, the organic carbon density point has time series consistency and the organic carbon density point is retained; otherwise, the organic carbon density point is deleted to obtain normalized organic carbon density point data.
[0010] Furthermore, step S03 of the above-mentioned permafrost organic carbon content inversion method specifically includes: S031: constructing 9×9 grids with each organic carbon density point in the normalized organic carbon density point data as the center to obtain a grid; S032: using each grid in the grid, sampling on the feature set to obtain a sample set; S033: dividing the sample set according to a preset ratio to obtain a training set, a test set and a validation set; the training set is used for model training, the test set is used to adjust the training model hyperparameters, and the validation set is used to test the model accuracy.
[0011] Furthermore, step S04 of the above-mentioned permafrost organic carbon content inversion method specifically includes: S041: using the attention mechanism and void convolution to construct different deep learning models, using the training set, the test set and the validation set to train the different deep learning models, and obtain different soil organic carbon estimation models; according to the different soil organic carbon estimation models, using the machine learning algorithm to conduct comparative tests and screening to obtain the optimal estimation model; S042: using the optimal estimation model to simulate the permafrost area in the study area to obtain the first organic carbon density spatial distribution; S043: according to the organic carbon density point data, using the extrapolation model to calculate, to obtain the second organic carbon density spatial distribution, according to the first organic carbon density spatial distribution and the second organic carbon density spatial distribution, to obtain the organic carbon density spatial distribution.
[0012] Furthermore, step S05 of the above-mentioned permafrost organic carbon content inversion method specifically includes: obtaining organic carbon reserves according to the spatial distribution of organic carbon density, such as the formula: , in, represents the organic carbon stock, Indicates the organic carbon density corresponding to each point, It represents the area corresponding to each point, width represents the width of the study area, and height represents the length of the study area.
[0013] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned permafrost organic carbon content inversion method are implemented.
[0014] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned permafrost organic carbon content inversion method when executing the program.
[0015] The implementation of the method, medium and device for inverting the organic carbon content of permafrost provided by the present invention has the following beneficial effects: Based on the fact that the characteristics of frozen soil are relatively complex, the number of organic carbon density point data is small and widely distributed, and it has large spatial heterogeneity, the present invention considers multiple influencing factors such as spectral characteristics, climate characteristics, ecological indexes, soil properties and terrain characteristics, combines the field sampling data of organic carbon density with multi-source remote sensing data, and selects thirty characteristic factors from the perspectives of spectrum, climate, vegetation, soil and terrain based on multi-source remote sensing data to construct a feature set. The dilated convolution is used to extract rich spatial context information, and the hybrid attention mechanism is used to screen the most important channel and spatial information. Taking into account the vertical layered structure of the soil, the dilated convolution is used to extract rich spatial context information, and the hybrid attention mechanism is used to screen the most important channel and spatial information. The method takes into account the soil vertical layered structure coupled with multi-source remote sensing data, and constructs a dual-input deep convolutional neural network as the organic carbon density estimation network architecture to simulate the relationship between organic carbon density and factors. The present invention can effectively solve the problems of inversion misfit or overfitting in some areas of the prior art, especially the technical problems of poor model fitting and incomplete consideration of complex frozen soil characteristics due to the scarcity of organic carbon density point data in the Qinghai-Tibet Plateau. It provides a high-precision and high-resolution long-term series frozen soil organic carbon density data set, which can obtain higher-precision spatial distribution and reserves of soil organic carbon in frozen soil areas of the Qinghai-Tibet Plateau, and can more comprehensively estimate the permafrost carbon reserves of the Qinghai-Tibet Plateau and analyze its dynamic changes and stability. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which: Figure 1 It is a flow chart of the method for inverting the organic carbon content of permafrost provided by the present invention; Figure 2 It is a flow chart of a method for inverting organic carbon content in permafrost based on deep learning and attention mechanism provided by the present invention; Figure 3 It is an overview of the study area provided by the present invention, including DEM data and spatial distribution of organic carbon density point data; Figure 4 is the spatial distribution of different characteristic factors provided by the present invention; Figure 5 It is a shallow organic carbon density estimation network framework provided by the present invention; Figure 6 It is a deep organic carbon density estimation network framework provided by the present invention; Figure 7 It is a Taylor diagram for comparing and verifying the accuracy of different models provided by the present invention; Figure 8 The spatial distribution results of organic carbon density at different depths in the permafrost area of the Qinghai-Tibet Plateau in 2020 provided by the present invention; Fig. 9The present invention provides the results of organic carbon density and organic carbon storage at different soil depths in the permafrost region of the Qinghai-Tibet Plateau in 2005 and 2020; Fig.10 It is a structural block diagram of the computer device provided by the present invention. DETAILED DESCRIPTION
[0017] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described in detail with reference to the accompanying drawings.
[0018] Figure 1 A schematic diagram of the permafrost organic carbon content inversion method of this embodiment is shown. In this embodiment, the permafrost organic carbon content inversion method includes the following steps: S01: Obtain the organic carbon density point data and regional characteristic data in the study area, pre-process the regional characteristic data, and obtain the feature set; Specifically, the organic carbon density point data includes organic carbon density point data at depths of 0-30cm, 0-50cm, 0-100cm, 0-200cm, and 0-300cm; Specifically, the regional characteristic data include remote sensing images, public data sets, surface temperature data, soil data sets and space shuttle geomorphic radar mission data; preprocessing of the regional characteristic data includes: S011: declouding the remote sensing image to obtain spectral index data; S012: according to the public data set, the annual average rainfall data is obtained, the surface temperature data is converted into 0cm real surface temperature data, and the annual average surface temperature data is obtained; according to the annual average rainfall data and the annual average surface temperature data, the climate index data is obtained; S013: according to the spectral index data, the ecological index data is obtained, as shown in the formula: , , , , , , , , , , Among them, NDVI is the normalized vegetation index, GNDVI is the green normalized index, SAVI is the soil adjusted vegetation index, OSAVI is the optimized soil adjusted vegetation index, EVI is the enhanced vegetation index, RVI is the ratio vegetation index, TVI is the triangle vegetation index, DVI is the vegetation difference index, MSAVI is the adjusted soil brightness vegetation index, and WET is the moisture index; S014: the soil data set is screened to obtain soil attribute data; S015: the terrain feature data is obtained according to the space shuttle landform radar mission data; S016: the spectral index data, the annual average rainfall data, the annual average surface temperature data, the normalized vegetation index, the green normalized index, the soil adjusted vegetation index, the optimized soil adjusted vegetation index, the enhanced vegetation index, the ratio vegetation index, the triangle vegetation index, the vegetation difference index, the adjusted soil brightness vegetation index, the moisture index, the soil attribute data and the terrain feature data are normalized, sampled to a resolution of 1 km, and projected to the same geographic coordinate system to obtain a feature set; Specifically, the feature set includes spectral index data, climate index data, ecological index data, soil property data and terrain feature data; the spectral index data includes red light band data, green light band data, blue light band data, near infrared band data, first mid-infrared band data and second mid-infrared band data; the climate index data includes annual average rainfall data and annual average surface temperature data; the ecological index data includes normalized vegetation index, green normalized index, soil adjusted vegetation index, optimized soil adjusted vegetation index, enhanced vegetation index, ratio vegetation index, triangle vegetation index, vegetation difference index, adjusted soil brightness vegetation index and humidity index; the soil property data includes sand content, gravel content, soil water content, soil pH value, soil bulk density, soil texture, soil components and soil biological community; the terrain feature data includes altitude, slope, slope aspect and terrain shadow; and the feature set is preprocessed; S02: According to the feature set, the organic carbon density point data is normalized to obtain normalized organic carbon density point data; Specifically, step S02 includes: S021: according to the spectral index data of the organic carbon density point data and the feature set, the spectral vector of each organic carbon density point in its corresponding sampling year is composed of red, green, blue, near infrared and two mid-infrared bands to obtain a first spectral vector; S022: according to the spectral index data of the organic carbon density point data and the feature set, the spectral vector of each organic carbon density point during the baseline period is composed of red, green, blue, near infrared and two mid-infrared bands to obtain a second spectral vector; S023: according to the first spectral vector and the second spectral vector, the spectral angle and spectral distance of each organic carbon density point are obtained, as shown in the formula: , , X:[R, G, B, NIR, SWIR1, SWIR2], Y:[R, G, B, NIR, SWIR1, SWIR2], Wherein, D is the spectral distance, θ is the spectral angle, X is the first spectral vector, Y is the second spectral vector, R is the red light band data, G is the green light band data, B is the blue light band data, NIR is the near infrared band data, SWIR1 is the first mid-infrared band data, and SWIR2 is the second mid-infrared band data; S024: when the spectral distance is less than the preset spectral distance value and the spectral angle is greater than the preset spectral angle value, the organic carbon density point has time series consistency and thus the organic carbon density point is retained, otherwise the organic carbon density point is deleted to obtain normalized organic carbon density point data; Specifically, the preset spectral distance value is 0.05, and the preset spectral angle value is 0.95; S03: sampling the feature set according to the normalized organic carbon density point data to obtain a sample set, and dividing the sample set to obtain a training set, a test set, and a validation set; Specifically, step S03 includes: S031: taking each organic carbon density point in the normalized organic carbon density point data as the center, constructing 9×9 grids respectively to obtain grids; S032: using each grid in the grid, sampling on the feature set respectively to obtain a sample set; S033: dividing the sample set according to a preset ratio to obtain a training set, a test set and a validation set; the training set is used for model training, the test set is used to adjust the training model hyperparameters, and the validation set is used to test the model accuracy; Specifically, the preset ratio is 6:2:2; S04: Use the training set, test set and validation set to train different deep learning models to obtain different soil organic carbon estimation models; use machine learning algorithms to conduct comparative tests and screening based on different soil organic carbon estimation models to obtain the optimal estimation model; obtain the spatial distribution of organic carbon density based on the optimal estimation model and organic carbon density point data; Specifically, step S04 includes: S041: Use the attention mechanism and dilated convolution to build different deep learning models, and use the training set, test set and validation set to train different deep learning models to obtain different soil organic carbon estimation models; Based on different soil organic carbon estimation models, use machine learning algorithms to conduct comparative tests and screening to obtain the optimal estimation model; Specifically, the machine learning algorithm is a random forest algorithm; Specifically, the machine learning algorithm may also be a support vector machine algorithm; Specifically, step S041 includes: using the attention mechanism and the hole convolution to construct a shallow soil organic carbon estimation model and a deep soil organic carbon estimation model; wherein, different deep learning models are constructed based on the attention mechanism modules such as the SE attention mechanism, the ECA attention mechanism and the CBAM attention mechanism; for the shallow soil organic carbon (0-30cm), the feature set is used as the input layer, and the convolutional neural network model is constructed by using the hole convolution and the attention mechanism; for the deeper soil organic carbon (0-50cm, 0-100cm), the vertical layered structure of the soil is considered, and the shallow organic carbon density distribution is used as the input of the second layer network to construct a dual-input convolutional neural network; and a comparative test is conducted using traditional machine learning models (such as RF, SVM, etc.) to screen out the optimal estimation model; S042: Use the optimal estimation model to simulate the frozen soil area in the study area and obtain the spatial distribution of the first organic carbon density; Specifically, step S042 includes: using the optimal estimation model to simulate the spatial distribution of organic carbon density in the frozen soil areas of 0-30 cm, 0-50 cm, and 0-100 cm in the study area; S043: According to the organic carbon density point data, the extrapolation model is used to calculate and obtain the second organic carbon density spatial distribution, and according to the first organic carbon density spatial distribution and the second organic carbon density spatial distribution, the organic carbon density spatial distribution is obtained; Specifically, step 043 includes: for soil organic carbon with a depth > 1m (0-200cm, 0-300cm), an extrapolation model is constructed through the collected soil organic carbon point data for calculation, and the calculation formula is as follows: , , in, is the organic carbon density in the 0-200cm permafrost zone, is the organic carbon density in the 0-100cm permafrost zone, is the organic carbon density in the 0-300cm permafrost zone; S05: According to the spatial distribution of organic carbon density, the organic carbon storage is obtained; Specifically, step S05 includes: obtaining organic carbon reserves according to the spatial distribution of organic carbon density, such as the formula: , in, represents the organic carbon stock, Indicates the organic carbon density corresponding to each point, It represents the area corresponding to each point, width represents the width of the study area, and height represents the length of the study area.
[0019] In some embodiments, the above-mentioned permafrost organic carbon content inversion method can also be implemented in the following manner.
[0020] This example uses the Qinghai-Tibet Plateau as the research area. Figure 2 As shown, in this embodiment, the method for inverting the organic carbon content of permafrost includes the following steps: S1: Obtain the organic carbon density point data in the study area, such as Figure 3 The figure shows the overview of the study area provided in this embodiment, including DEM data and spatial distribution of organic carbon density point data; specifically, organic carbon density point data of different depths such as 0-30cm, 0-50cm, 0-100cm, 0-200cm, and 0-300cm in the permafrost area of the Qinghai-Tibet Plateau in published literature are collected; S2: Get the feature set in the study area, such as Figure 4 The figure shows the spatial distribution of different characteristic factors provided by the present embodiment; the characteristic set includes: six spectral indexes (red band R, green band G, blue band B, near infrared band NIR, two mid-infrared bands SWIR1 and SWIR2), two climate indexes (annual average rainfall, annual average surface temperature), ten ecological indexes (normalized difference vegetation index NDVI, green normalized index GNDVI, soil adjusted vegetation index SAVI, optimized soil adjusted vegetation index OSAVI, enhanced vegetation index EVI, ratio vegetation index RVI, triangle vegetation index TVI, vegetation difference index DVI, adjusted soil brightness vegetation index MSAVI and moisture index WET), eight soil attribute data (sand content, gravel content, soil water content, soil pH value, soil bulk density, soil texture, soil components, soil biological community), four terrain characteristics (altitude, slope, slope direction, terrain shadow), and the characteristic set is preprocessed, and the specific preprocessing method is as follows: S2.1: On the Google Earth Engine cloud platform, select the Landsat5 / 7 / 8 images of the Qinghai-Tibet Plateau and perform cloud removal to obtain the red band R, green band G, blue band B, near infrared band NIR and two mid-infrared bands SWIR1 and SWIR2; S2.2: Based on the public dataset, obtain the annual average rainfall dataset in the Qinghai-Tibet Plateau; S2.3: Obtain the MODIS surface temperature dataset for the Qinghai-Tibet Plateau and convert it into 0 cm true surface temperature data using the following formula: and The average surface temperature from September to April of the following year ( ) and the average surface temperature from May to August ( ), and They are the daytime and nighttime observation values of MOD11A2 product data ( ), and are the daytime and nighttime observation values of MYD11A2 product data respectively ( ): , , S2.3: Based on the six spectral bands that have been obtained, ten ecological indices are calculated: Normalized Difference Vegetation Index NDVI, Green Normalized Difference Index GNDVI, Soil Adjusted Vegetation Index SAVI, Optimized Soil Adjusted Vegetation Index OSAVI, Enhanced Vegetation Index EVI, Ratio Vegetation Index RVI, Triangular Vegetation Index TVI, Vegetation Difference Index DVI, Adjusted Soil Brightness Vegetation Index MSAVI and Wetness Index WET. The calculation formula is as follows: , , , , , , , , , , S2.4: Obtain the soil dataset of the Qinghai-Tibet Plateau and screen out eight soil characteristics: sand content, gravel content, soil water content, soil pH, soil bulk density, soil texture, soil components, and soil biological communities; S2.5: Four terrain features, elevation, slope, aspect, and terrain shadow, are calculated based on the 30 m SRTM data; S2.6: Normalize the above features, sample to 1km resolution, and project to the same geographic coordinate system; S3: Use spectral angle and spectral vector to normalize the organic carbon density point data to the same time baseline. The specific steps are as follows: S3.1: Use red, green, blue, near infrared and two mid-infrared bands to form the spectral vector X: [R, G, B, NIR, SWIR1, SWIR2] of each organic carbon density point in its corresponding sampling year; S3.2: Compose the spectral vector Y: [R, G, B, NIR, SWIR1, SWIR2] for each organic carbon density point during the baseline period (2014) using the red, green, blue, near-infrared, and two mid-infrared bands; S3.3: Calculate the spectral angle and spectral distance of each organic carbon density point based on the spectral vectors X and Y. The calculation formula is as follows: , , S3.4: When the spectral distance And the spectral angle When , the organic carbon density point is considered to have time series consistency, otherwise the point will be deleted; S4: Construct a 9×9 grid with the normalized organic carbon density point data as the center, sample the feature set to construct a sample set, and divide it into training set, test set, and validation set. The specific steps are as follows: S4.1: Construct a 9×9 grid with each organic carbon density point after S3 as the center; S4.2: Use each grid obtained in S4.1 to sample the feature set obtained in S2 to construct a sample set; S4.3: Divide the sample set into training set, test set, and validation set in a ratio of 6:2:2, where the training set is used for model training, the test set is used to adjust the training model hyperparameters, and the test set is used to verify the model accuracy; S5: Use machine learning / deep learning models for training and select the model with the best training effect. Finally, based on the best model, the spatial distribution of organic carbon density in permafrost on the Qinghai-Tibet Plateau is obtained. The specific steps are as follows: S5.1: Construct different deep learning models based on the attention mechanism modules (SE attention mechanism, ECA attention mechanism, CBAM attention mechanism). For shallow soil organic carbon (0-30cm), use the feature set as the input layer, and use the dilated convolution and attention mechanism to construct a convolutional neural network model. For deeper soil organic carbon (0-50cm, 0-100cm), consider the vertical layered structure of the soil, use the shallow organic carbon density distribution as the input of the second layer network, and construct a dual-input convolutional neural network. Compare with traditional machine learning models (such as RF, SVM), and select the optimal model. S5.2: Use the optimal model to simulate the spatial distribution of organic carbon density in the 0-30cm, 0-50cm, and 0-100cm permafrost zones of the Qinghai-Tibet Plateau; S5.3: For soil organic carbon at a depth > 1 m (0-200 cm, 0-300 cm), an extrapolation model is constructed based on the collected soil organic carbon point data for calculation. The calculation formula is as follows: , , S6: Calculate organic carbon storage based on the spatial distribution of organic carbon density. The specific steps are as follows: S6.1: Calculate the organic carbon stock based on the spatial distribution of organic carbon density in the permafrost region of the Qinghai-Tibet Plateau obtained in S5. The formula is as follows; , in represents organic carbon stock (Pg), Indicates the organic carbon density corresponding to each point ( ), Indicates the area corresponding to each point ( ); Please refer to Figure 5 , Figure 5 This is the network structure diagram for simulating the organic carbon density in the shallow layer (0-30cm). The network is a single-layer feature input structure, and the input is a feature set. Please refer to Figure 6 , Figure 6 The network structure diagram of deep (0-50cm, 0-100cm) organic carbon density simulation. The network has a double-layer feature input structure: the first layer input is the feature set; when estimating the organic carbon density at a depth of 0-50cm (0-100cm), the second layer feature input is the organic carbon density at a depth of 0-30cm (0-50cm); Please refer to Figure 7 , Figure 7 Taylor diagrams comparing the accuracy of different models. a) The results of simulating 0-30cm organic carbon density using the shallow estimation network in Figure 4, and b) The results of simulating 0-50cm organic carbon density using the deep estimation network in Figure 5. The results show the superiority of the models in Figure 4 (SSCNN) and Figure 5 (DDSCNN). Please refer to Figure 8 , Figure 8 This is the spatial distribution map of soil organic carbon density at 0-30cm, 0-50cm, 0-100cm, 0-200cm and 0-300cm in the permafrost area of the Qinghai-Tibet Plateau in 2020; Please refer to Fig. 9 , Fig. 9 This is a bar chart of organic carbon density and organic carbon storage at different soil depths in the permafrost area of the Qinghai-Tibet Plateau in 2005 and 2020.
[0021] This embodiment provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned permafrost organic carbon content inversion method are implemented. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory (Flash Memory), a hard disk (Hard Disk Drive, abbreviated: HDD) or a solid-state drive (SSD), etc.; the storage medium can also include a combination of the above-mentioned types of memory.
[0022] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the above-mentioned permafrost organic carbon content inversion method are implemented.
[0023] like Fig.10 As shown, the computer device may include: at least one processor 121, such as a CPU (Central Processing Unit), at least one communication interface 123, a memory 124, and at least one communication bus 122. Among them, the communication bus 122 is used to realize the connection and communication between these components. Among them, the communication interface 123 may include a display screen (Display), a keyboard (Keyboard), and the optional communication interface 123 may also include a standard wired interface and a wireless interface. The memory 124 may be a high-speed RAM memory (Random Access Memory, volatile random access memory), or a non-volatile memory (non-volatile memory), such as at least one disk memory. The memory 124 may also be at least one storage device located away from the aforementioned processor 121. Among them, the memory 124 stores an application program, and the processor 121 calls the program code stored in the memory 124 to perform any of the above method steps. Among them, the communication bus 122 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The communication bus 122 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Fig.10Only one line is used to represent it, but it does not mean that there is only one bus or one type of bus. Among them, the memory 124 may include a volatile memory (English: volatile memory), such as a random access memory (English: random-access memory, abbreviated: RAM); the memory may also include a non-volatile memory (English: non-volatile memory), such as a flash memory (English: flash memory), a hard disk drive (English: hard disk drive, abbreviated: HDD) or a solid-state drive (English: solid-state drive, abbreviated: SSD); the memory 124 may also include a combination of the above-mentioned types of memory. Among them, the processor 121 may be a central processing unit (English: central processing unit, abbreviated: CPU), a network processor (English: network processor, abbreviated: NP) or a combination of CPU and NP. Among them, the processor 121 may further include a hardware chip. The above-mentioned hardware chip may be an application-specific integrated circuit (English: application-specific integrated circuit, abbreviated: ASIC), a programmable logic device (English: programmable logic device, abbreviated: PLD) or a combination thereof. The PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof. Optionally, the memory 124 is also used to store program instructions. The processor 121 may call program instructions to implement the permafrost organic carbon content inversion method of the present embodiment.
[0024] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.
Claims
1. A method for inverting the organic carbon content of permafrost, characterized in that: The following steps are involved: S01: Acquire organic carbon density point data and regional characteristic data in the study area, pre-process the regional characteristic data, and obtain a feature set; S02: normalizing the organic carbon density point data according to the feature set to obtain normalized organic carbon density point data; S03: sampling the feature set according to the normalized organic carbon density point data to obtain a sample set, and dividing the sample set to obtain a training set, a test set, and a validation set; S04: using the training set, the test set and the validation set to train different deep learning models to obtain different soil organic carbon estimation models; using a machine learning algorithm to perform comparative tests and screening based on the different soil organic carbon estimation models to obtain an optimal estimation model; and obtaining the spatial distribution of organic carbon density based on the optimal estimation model and the organic carbon density point data; S05: Obtaining organic carbon reserves according to the spatial distribution of organic carbon density.
2. The method for inverting the organic carbon content of permafrost according to claim 1, characterized in that: The organic carbon density point data include organic carbon density point data at depths of 0-30cm, 0-50cm, 0-100cm, 0-200cm, and 0-300cm.
3. The method for inversion of permafrost organic carbon content according to claim 1, characterized in that: The regional characteristic data includes remote sensing images, public data sets, surface temperature data, soil data sets and space shuttle geomorphic radar mission data; the preprocessing of the regional characteristic data specifically includes: S011: performing cloud removal processing on the remote sensing image to obtain spectral index data; S012: According to the public data set, annual average rainfall data is obtained, the surface temperature data is converted into 0 cm real surface temperature data to obtain annual average surface temperature data, and climate index data is obtained according to the annual average rainfall data and the annual average surface temperature data; S013: According to the spectral index data, obtain ecological index data, such as formula: ， ， ， ， ， ， ， ， ， ， Among them, NDVI is the normalized difference vegetation index, GNDVI is the green normalized index, SAVI is the soil adjusted vegetation index, OSAVI is the optimized soil adjusted vegetation index, EVI is the enhanced vegetation index, RVI is the ratio vegetation index, TVI is the triangular vegetation index, DVI is the vegetation difference index, MSAVI is the adjusted soil brightness vegetation index, and WET is the wetness index; S014: Screening the soil data set to obtain soil attribute data; S015: Obtaining terrain feature data according to the space shuttle terrain radar mission data; S016: The spectral index data, the annual average rainfall data, the annual average surface temperature data, the normalized vegetation index, the green normalized index, the soil-adjusted vegetation index, the optimized soil-adjusted vegetation index, the enhanced vegetation index, the ratio vegetation index, the triangle vegetation index, the vegetation difference index, the adjusted soil brightness vegetation index, the moisture index, the soil attribute data and the terrain feature data are normalized, sampled to a resolution of 1 km, and projected to the same geographic coordinate system to obtain a feature set.
4. The method for inverting the organic carbon content of permafrost according to claim 1, characterized in that: The feature set includes spectral index data, climate index data, ecological index data, soil property data and terrain feature data; the spectral index data includes red light band data, green light band data, blue light band data, near infrared band data, first mid-infrared band data and second mid-infrared band data; the climate index data includes annual average rainfall data and annual average surface temperature data; the ecological index data includes normalized vegetation index, green normalized index, soil adjusted vegetation index, optimized soil adjusted vegetation index, enhanced vegetation index, ratio vegetation index, triangle vegetation index, vegetation difference index, adjusted soil brightness vegetation index and humidity index; the soil property data includes sand content, gravel content, soil water content, soil pH value, soil bulk density, soil texture, soil components and soil biological community; the terrain feature data includes altitude, slope, slope direction, terrain shadow; and the feature set is preprocessed.
5. The method for inverting the organic carbon content of permafrost according to claim 1, characterized in that: Step S02 specifically includes: S021: Based on the organic carbon density point data and the feature set, a spectral vector of each organic carbon density point in its corresponding sampling year is formed using red, green, blue, near infrared and two mid-infrared bands to obtain a first spectral vector; S022: According to the organic carbon density point data and the feature set, a spectral vector of each organic carbon density point during the baseline period is formed using red, green, blue, near infrared and two mid-infrared bands to obtain a second spectral vector; S023: According to the first spectral vector and the second spectral vector, the spectral angle and spectral distance of each organic carbon density point are obtained, such as the formula: ， ， X:[R, G, B, NIR, SWIR1, SWIR2], Y:[R, G, B, NIR, SWIR1, SWIR2], Wherein, D is the spectral distance, θ is the spectral angle, X is the first spectral vector, Y is the second spectral vector, R is the red light band data, G is the green light band data, B is the blue light band data, NIR is the near infrared band data, SWIR1 is the first mid-infrared band data, and SWIR2 is the second mid-infrared band data; S024: When the spectral distance is less than a preset spectral distance value and the spectral angle is greater than a preset spectral angle value, the organic carbon density point has time series consistency and the organic carbon density point is retained; otherwise, the organic carbon density point is deleted to obtain normalized organic carbon density point data.
6. The method for inverting the organic carbon content of permafrost according to claim 1, characterized in that: Step S03 specifically includes: S031: taking each organic carbon density point in the normalized organic carbon density point data as the center, respectively constructing a 9×9 grid to obtain a grid; S032: using each grid in the grid, respectively sampling on the feature set to obtain a sample set; S033: Divide the sample set according to a preset ratio to obtain a training set, a test set and a validation set; the training set is used for model training, the test set is used to adjust the training model hyperparameters, and the validation set is used to verify the model accuracy.
7. The method for inverting the organic carbon content of permafrost according to claim 1, characterized in that: Step S04 specifically includes: S041: Using the attention mechanism and the dilated convolution, construct different deep learning models, and use the training set, the test set and the validation set to train the different deep learning models to obtain different soil organic carbon estimation models; according to the different soil organic carbon estimation models, use the machine learning algorithm to conduct comparative tests and screening to obtain the optimal estimation model; S042: simulating the frozen soil area in the study area using the optimal estimation model to obtain a first spatial distribution of organic carbon density; S043: Calculate using an extrapolation model based on the organic carbon density point data to obtain a second organic carbon density spatial distribution, and obtain an organic carbon density spatial distribution based on the first organic carbon density spatial distribution and the second organic carbon density spatial distribution.
8. The method for inverting the organic carbon content of permafrost according to claim 1, characterized in that: Step S05 specifically includes: obtaining the organic carbon reserve according to the spatial distribution of the organic carbon density, such as the formula: ， in, represents the organic carbon stock, Indicates the organic carbon density corresponding to each point, It represents the area corresponding to each point, width represents the width of the study area, and height represents the length of the study area.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the permafrost organic carbon content inversion method as described in any one of claims 1 to 8 are implemented.
10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the permafrost organic carbon content inversion method as described in any one of claims 1-8 are implemented.