Soil-sediment organic carbon enrichment coefficient estimation method based on reflection spectrum

Abstract

The invention relates to a soil-sediment organic carbon enrichment coefficient estimation method based on a reflection spectrum, which comprises the following steps: respectively collecting a river sediment sample and a soil sample, measuring the content of organic carbon, and calculating an actually measured organic carbon enrichment coefficient; determining the reflection spectrum of the sample to obtain the reflectivity, and respectively screening the organic carbon key response wave bands of the reflection spectrum of the sediment and the soil; constructing an organic carbon enrichment coefficient spectral index based on the key response wave band, and constructing a soil-sediment organic carbon enrichment coefficient spectral estimation model by combining the actually measured organic carbon enrichment coefficient and the spectral index; a sediment sample to be detected and a soil sample are collected, reflection spectrums are measured respectively, the organic carbon enrichment coefficient spectral index is calculated and input into the estimation model, and the soil-sediment organic carbon enrichment coefficient is obtained. By adopting the method disclosed by the invention, the organic carbon enrichment coefficient of the soil-sediment in the drainage basin can be quickly estimated, the enrichment characteristics of local nutrient substances are obtained, and technical support is provided for water and soil conservation, eutrophication and dual-carbon target realization.

Description

Technical Field

[0001] The invention belongs to the field of remote sensing technology, and in particular relates to a method for estimating soil-sediment organic carbon enrichment coefficient based on reflectance spectrum. Background Art

[0002] Organic carbon in soil and sediment is an important component of ecosystem carbon storage, and plays a key role in maintaining soil fertility, promoting plant growth, and enhancing the carbon sink function of the ecosystem. The enrichment effect of soil organic carbon is mainly driven by a variety of environmental factors, including topography, land use patterns, hydrological characteristics, and human activities. Since sediment is an important carrier of organic carbon in a watershed, it will be enriched with organic matter during transportation and further deposited in riverbeds, lakes, or coastal areas. Therefore, the enrichment behavior of sediment has become an important factor in understanding changes in watershed carbon storage.

[0003] The enrichment degree of organic carbon is usually measured by the enrichment coefficient (EC), which refers to the ratio of the organic carbon concentration in sediment or soil particles to the organic carbon concentration in the surface soil. The traditional calculation of sediment enrichment coefficient mainly relies on outdoor collection or laboratory analysis of soil samples, which requires a lot of manpower and material resources. With the rapid development of spectral technology, technologies such as Raman spectroscopy, near-infrared spectroscopy and laser-induced spectroscopy can achieve rapid monitoring of soil nutrient content by extracting spectral information related to soil nutrients. Indoor hyperspectral data has been widely used by scholars at home and abroad in the study of soil physical and chemical property estimation. However, there is no research report on hyperspectral estimation of enrichment coefficient. Therefore, using measured hyperspectral technology to quickly estimate the organic carbon enrichment coefficient of watershed soil-sediment and obtain the enrichment characteristics of local nutrients can provide technical support for soil and water conservation, eutrophication and the realization of dual carbon goals. Summary of the invention

[0004] The purpose of the present invention is to provide a method for estimating soil-sediment organic carbon enrichment coefficient based on reflectance spectroscopy.

[0005] To achieve the above purpose, the technical solution adopted by the present invention is as follows: A method for estimating soil-sediment organic carbon enrichment coefficient based on reflectance spectroscopy, the method comprising: Collect river sediment samples and soil samples from the corresponding riverbanks, measure the organic carbon content, and calculate the soil-sediment organic carbon enrichment factor as the measured organic carbon enrichment factor; Measure the reflectance spectra of the sediment samples and soil samples to obtain the reflectance R rs , and screen the key response bands of sediment reflectance spectrum and soil reflectance spectrum respectively; Constructing an organic carbon enrichment coefficient spectral index based on the key response bands of the sediment reflectance spectrum and the soil reflectance spectrum; Constructing a soil-sediment organic carbon enrichment coefficient spectral estimation model based on the measured organic carbon enrichment coefficient and spectral index; Collect sediment samples and soil samples to be tested, measure the reflectance spectra, calculate the spectral index of organic carbon enrichment coefficient and input it into the estimation model to obtain the soil-sediment organic carbon enrichment coefficient.

[0006] As a preferred implementation, the key response bands are screened using an importance analysis method.

[0007] Furthermore, the key response bands are screened using a variety of different importance analysis methods, including: The organic carbon characteristic bands of sediment samples and soil samples were analyzed using a variety of different importance analysis methods and sorted by importance. Each method obtained several organic carbon characteristic bands of sediment samples and soil samples. After performing different mathematical permutations and combinations on the organic carbon characteristic bands of the sediment samples and soil samples obtained by each method, the key response band combinations of organic carbon of the sediment samples and soil samples are screened out respectively by using the multiple importance analysis methods; Different machine learning methods were used in combination with the band combination for testing, and the final key response bands of organic carbon in sediment samples and soil samples were determined based on the model accuracy.

[0008] Furthermore, the importance analysis method includes SKlearn importance analysis, Eli5 importance analysis, and SHAP method.

[0009] As a preferred implementation, the organic carbon enrichment factor spectral index is a normalized spectral index constructed based on the key response bands of the sediment reflectance spectrum and the key response bands of the soil reflectance spectrum.

[0010] As a preferred implementation, when calculating the spectral index, the calculation is based on the distance range and land use type classification, and models are constructed for soils of different distance ranges and different land use types respectively; the distance is the distance between the sediment sample and the corresponding soil sample.

[0011] As a preferred implementation, the step of constructing soil models for different distance ranges and different land use types includes: The distances between the sediment samples and the corresponding soil samples are divided into different distance range groups according to the ranges of 0-1km, 1-5km, 5-10km and 10-20km, and the estimation models are respectively constructed for the sample data of the different distance range groups; The sediment samples and the corresponding soil samples are divided into different land use type groups according to the land use type to which the soil samples belong, and the estimation models are respectively constructed for the different land use type groups.

[0012] As a preferred implementation, the model is constructed using random forest, SVM or XGBoost methods.

[0013] As a preferred embodiment, based on the average R 2 , RMSE or MRE to evaluate the model performance, and the model with the best performance was obtained as the final soil-sediment organic carbon enrichment coefficient spectral estimation model.

[0014] As a preferred embodiment, the collection point of the sediment sample is arranged in the river, and the collected sediment sample is river suspended sediment or bed sediment; The soil sample collection points are arranged on both sides of the river channel, and the collected soil samples are surface soil.

[0015] As a preferred implementation, the sediment samples and soil samples are air-dried and ground before measuring the reflectance spectra.

[0016] As a preferred embodiment, the reflectance is obtained by preprocessing the reflectance spectrum. R rs , the preprocessing comprises: Perform breakpoint correction on reflectance spectrum data; Calculate the average value of several spectral reflectances measured for each sample as the spectral reflectance of the sample; The reflectance spectrum curve is smoothed and denoised, and the noise edge bands at wavelengths of 350-399nm and 2451-2500nm are removed.

[0017] The method of the present invention can use indoor measured hyperspectral data to quickly obtain the soil-sediment organic carbon enrichment coefficient. By formulating standard sediment samples and testing their organic carbon and hyperspectral data, the soil organic carbon quality can be reversed, providing technical support for soil and water conservation, eutrophication and the realization of dual carbon goals.

[0018] It should be understood that all combinations of the aforementioned concepts and the additional concepts described in more detail below can be considered as part of the inventive subject matter of the present disclosure as long as such concepts are not mutually inconsistent. In addition, all combinations of the claimed subject matter are considered as part of the inventive subject matter of the present disclosure.

[0019] The foregoing and other aspects, embodiments and features of the present invention can be more fully understood from the following description in conjunction with the accompanying drawings. Other additional aspects of the present invention, such as the features and / or beneficial effects of the exemplary embodiments, will be apparent from the following description or learned from the practice of the specific embodiments according to the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The drawings are not intended to be drawn to scale. In the drawings, each identical or nearly identical component shown in various figures may be represented by the same reference numeral. For clarity, not every component is labeled in every figure. Embodiments of various aspects of the present invention will now be described by way of example and with reference to the accompanying drawings, in which: Figure 1 This is a flow chart of the soil-sediment organic carbon enrichment factor estimation method based on reflectance spectroscopy.

[0021] Figure 2 It is a box plot of the test results of sediment and soil organic matter (organic carbon in this application), where ac is the organic carbon content of three different types of soil covered on the banks of four rivers (Caoqiao River, Shaogang, Taige Canal, Yincun Port); d is the organic carbon content of sediment in the four rivers.

[0022] Figure 3 It is a box plot of measured organic carbon enrichment coefficient.

[0023] Figure 4 are the results of reflectance spectrum measurement, where a: sediment reflectance spectrum measurement result, b: soil reflectance spectrum measurement result.

[0024] Figure 5 These are the screening results of the key response bands (characteristic bands) of sediment and soil organic carbon, where a: the bands screened by the SHAP method, b: the bands screened by the Sklearn method, and c: the bands screened by the Eli5 method.

[0025] Figure 6 These are the screening results of the key response bands (characteristic bands) of sediment and soil organic carbon combinations, where a: the optimal band combination screened by the SHAP method, b: the optimal band combination screened by the Sklearn method, and c: the optimal band combination screened by the Eli5 method.

[0026] Figure 7 It is the accuracy of the soil-sediment organic carbon enrichment coefficient spectral estimation model for different rivers.

[0027] Figure 8 It is the accuracy of the soil-sediment organic carbon enrichment coefficient spectral estimation model for soil source samples under different land use.

[0028] Fig. 9is the accuracy of the soil-sediment organic carbon enrichment coefficient spectral estimation model at different distances.

[0029] In the above-mentioned Figures 1-9, the coordinates, symbols or other expressions expressed in English are all well known in the art and will not be described in detail in this example. DETAILED DESCRIPTION

[0030] In order to better understand the technical content of the present invention, specific embodiments are described below in conjunction with the accompanying drawings.

[0031] Various aspects of the present invention are described with reference to the accompanying drawings in the embodiments disclosed herein, in which many illustrative embodiments are shown. The embodiments disclosed herein are not necessarily intended to include all aspects of the present invention. It should be understood that the various concepts and embodiments introduced above, as well as those described in more detail below, can be implemented in any of many ways, because the concepts and embodiments disclosed by the present invention are not limited to any implementation. In addition, some aspects disclosed by the present invention can be used alone, or in any appropriate combination with other aspects disclosed by the present invention.

[0032] According to the general definition of organic carbon enrichment coefficient (the ratio of organic carbon concentration in sediment or soil particles to organic carbon concentration in surface soil), the soil-sediment organic carbon enrichment coefficient referred to in this application refers to the ratio of soil organic carbon content to sediment organic carbon content.

[0033] The key response band described in this application refers to the band in the reflectance spectrum that is sensitive to changes in organic carbon content.

[0034] Example 1 This embodiment takes the river sediment and surrounding soil of Caoqiao River, Shaogang Port, Taige Canal and Yincun Port in the Taihu Lake Basin as examples to further describe the technical solution of the present invention.

[0035] The soil-sediment organic carbon enrichment factor estimation method shown in the embodiment is as follows Figure 1 As shown, the specific steps include: Step 1: Place a number of sediment samples in the river and set up soil sample collection points on both sides of the river, measure the organic carbon content of the sediment samples and soil samples, and calculate the soil-sediment organic carbon enrichment factor as the measured organic carbon enrichment factor; Traditional studies on sediment source tracing in mountainous areas are based on collecting sediment source samples in the catchment area. However, it is impossible to accurately divide the catchment area in plain river network areas. Most studies use buffer zones as boundaries to study the impact of watersheds on water bodies, and the land use types for source sample collection are cultivated land, forest land, and grassland on both sides of the river.

[0036] This embodiment uses samples collected in previous studies (CN117030627A) for calculation. Specifically, sediment and source samples were collected after the rainstorm in July 2022. Among them, (1) the sediment sample is river bedload. The river sediment column is collected using a sediment column sampler, and the surface flocculent sediment is peeled off. The bedload on 4 columns is collected at each sampling point and mixed together as the sediment sample of the sampling point. A total of 36 sediment samples were collected; (2) The source sample is a soil sample. According to the land use classification of the Taihu Lake Basin, the surface 0-2cm soil samples were collected using the quartering method. A total of 88 samples were collected, including 30 cultivated soils, 28 forest soils, and 30 grassland soils. All samples were packed in self-sealing bags, kept cool and dark, and brought back to the laboratory.

[0037] Sample pretreatment includes air drying, grinding, sieving and other steps to ensure sample homogeneity and provide quality assurance for subsequent organic matter testing. The dry burning method is a classic method for determining soil organic matter content. It uses high-temperature roasting to directly oxidize organic matter in the soil and calculates the organic matter content by the mass difference before and after. Through the dry burning method, the study can quickly obtain basic data on the organic matter content of the sample, providing basic information for subsequent spectral model construction and enrichment analysis.

[0038] Spectroscopy-assisted testing: Use near-infrared spectroscopy (NIR) to quickly predict and verify the organic matter content of samples.

[0039] The test results of the organic carbon content of sediment and soil in the sediment samples and source samples in this example are as follows: Figure 2 As shown in the figure, the organic carbon enrichment factor was calculated based on the measured organic carbon content of sediment and soil as the measured organic carbon enrichment factor. The results are shown in Figure 3 shown.

[0040] The calculation method of measured organic carbon enrichment factor is:

[0041] In the formula, Indicates the concentration of organic carbon in sediment (unit: g / kg ), Indicates the concentration of organic carbon in the soil (unit: g / kg). The larger the value, the higher the enrichment of particulate organic carbon. This indicator reflects the enrichment of particulate organic carbon in a soil source sample. EC The larger the value, the higher the degree of organic carbon enrichment in the soil source sample.

[0042] Step 2: After the collected soil and sediment samples are air-dried and ground, the reflectance spectrum is measured indoors, and the reflectance R is obtained after pretreatment. rs, respectively screen the key response bands and the spectral index of soil-sediment organic carbon enrichment coefficient; Soil spectrum measurement was carried out in a closed dark laboratory. The spectrometer used was a FieldSpec 3 portable spectrometer. The probe was perpendicular to the sample with a 5-degree field of view (B=2.5°), 15 cm away from the sample (H), and a detection radius (R2) of 0.655 cm. The light source was a 50W halogen lamp, with an angle (A) of 15° to the vertical direction and 30 cm away from the sample (L). The sample container was a small aluminum round box with a radius (R1) of 2.5 cm and a depth of 1 cm. The small round box was placed on a black automatic rotating platform. Before collecting the sample spectrum, the dark current was first removed, and then the absolute reflectance was obtained using a 40×40 cm diffuse reflectance standard reference white board. During the spectrum collection process, the sample surface was first scraped flat with a ruler, and the small round box containing the soil sample was placed on the black automatic rotating platform. As the small round box rotated at a constant speed for one week, the spectrometer collected 20 spectral curves at equal time intervals, and after collecting 5 sample spectrum data, it was re-calibrated with a reference white board to minimize the error of the spectral data.

[0043] The measured original spectrum needs to go through a series of preprocessing before it can be used for model construction: 1) The spectrometer consists of three sub-spectrometers, which receive the reflection spectra in the three bands of 350~1100nm, 1000~1800nm, and 1700~2500nm respectively. Therefore, there will be breakpoints near 1000nm and 1800nm ​​due to instrument conversion. The original spectrum data is corrected for breakpoints using the spectral data processing software ViewSpecPro that comes with the ASD spectrometer ( Figure 4 ); 2) 20 spectra were measured for each sample, and the average reflectance of the 20 spectra was calculated; 3) At this time, the spectral curve still had some noise and steps, which needed to be smoothed and denoised by preprocessing to reduce noise interference. The convolution smoothing method, which is currently the most widely used denoising method, was used to resample the spectral curve at intervals of 5 nm. 4) The noise edge bands (350-399nm and 2451-2500nm) were removed to obtain the processed reflectance spectrum R rs The reflectance spectrum measurement results are as follows Figure 4 shown.

[0044] The key response bands were screened, and the SKlearn importance analysis, Eli5 importance analysis and SHAP method were used to screen the key response bands of organic carbon content in sediment and soil, respectively, and the screening results were obtained, such as Figure 5As shown, the screening results (top five in importance) obtained by the SHAP method are 1920nm, 2420nm, 400nm, 1150nm, and 565nm for sediment, and 1890nm, 1750nm, 2400nm, 2225nm, and 1915nm for soil; the screening results (top five in importance) obtained by the SKlearn importance analysis method are 2225nm, 1750nm, 1890nm, 2330nm, and 1915nm for sediment, and 2240nm, 2375nm, 1150nm, 2200nm, and 1920nm for soil; the screening results (top five in importance) obtained by the Eli5 importance analysis method are 1920nm, 400nm, 2420nm, 1150nm, and 2200nm for sediment, and 2400nm, 2225nm, 1915nm, 1890nm, and 1885nm for soil.

[0045] Based on the independent single band screening results, the soil single band screening results obtained by the above three methods (the top five bands in importance) were paired with the top five sediment single bands in importance that were also screened out. New band combinations were constructed using different mathematical forms such as addition, subtraction, multiplication, division, and normalization. The band combinations were tested using the three methods, and the results are as follows: Figure 6 As shown in the figure, the optimal band combination of the three methods is determined as follows: SHAP method: Soil 2400 +Sed 400 (i.e. the addition of the 2400nm band for soil and 400nm band for sediment); Eli5 method: Soil 1915 -Sed 2420 ;Sklearn method: Soil 1920 -Sed 565 .

[0046] Finally, the optimal band combinations selected by the three methods were input into the three machine learning models respectively, and the optimal results were determined to be RF model and Soil model. 1920 -Sed 565 The key response bands of sediment and soil organic carbon are finally determined to be 575nm and 1920nm, respectively, and the constructed spectral index is Soil 1920 -Sed 565 .

[0047] Step 3: Construct a soil-sediment organic carbon enrichment factor spectral estimation model based on the measured organic carbon enrichment factor and the organic carbon enrichment factor spectral index.

[0048] RF, XGBoost and SVM models were used to construct soil-sediment organic carbon enrichment coefficient spectral estimation models respectively.

[0049] First, all the data were mixed together for modeling, that is, the distance range of soil land types, soil samples, and sediment samples was not distinguished. From the overall performance, RF performed best in all three land use types, with R² values ​​around 0.86. The R² values ​​of SVM were generally low, ranging from 0.17-0.21, which indicates that the SVM model did not fit well and could not fully capture the relationship between variables, and parameters or methods may need to be adjusted. XGBoost had a moderate R² value of 0.32-0.50. Although the overall effect was not as good as RF, it was better than SVM, especially in grassland and woodland.

[0050] The data were divided into different groups according to soil types. In this embodiment, the data were divided into cultivated land, forest land and grassland according to soil types. Estimation models were constructed for the cultivated land group data (cultivated land soil samples and corresponding sediment samples), forest land group data and grassland group data respectively. The results are shown in Table 1 and Figure 7 As shown in the figure, from the performance of the models of different soil types, the RF model of cultivated land has the best effect, with an R² of 0.87. XGBoost's R² is only 0.32, which is poor. SVM performs the worst, with an R² of 0.17. The RF model of grassland is still the best, with an R² of 0.86. XGBoost has improved, with an R² of 0.50. SVM performs the weakest, with an R² of only 0.17. RF of woodland once again achieves the highest value, with an R² of 0.87. XGBoost's R² is 0.46, second only to RF. SVM performs slightly better than grassland and cultivated land, but still weak, with an R² of 0.21. RF is the best choice, with an R² value of close to 0.86 in all land use types, stable and excellent performance, suitable for the fitting needs of the current data. XGBoost is the second best choice. Although its overall performance is not as good as RF, its performance is improved on grassland and woodland.

[0051] Table 1 Accuracy of the three models under different land use types <![CDATA[R 2 ]]> arable land grassland woodland RF 0.87 0.86 0.87 SVM 0.17 0.17 0.21 XGBoost 0.32 0.50 0.46 The sample data is divided into different groups according to the river. In this embodiment, the sample data is divided into Caoqiao River group (using the sediment sample of Caoqiao River, i.e., the corresponding riverbank soil sample modeling), Yincun Port, Taige Canal and Shaogang Port group according to the river to which the sampling point belongs. The results are as follows: Figure 8As shown in Table 2, from the overall performance, the R² values ​​of RF in the four river categories are close to 0.86-0.87, showing a very stable fitting ability. This is the best of the three models and can well explain the changes in the data. The R² values ​​of SVM in all categories are very low, only 0.17-0.19, which can hardly explain the relationship between variables and may not be suitable for the current data set. The R² value of XGBoost is 0.41-0.50, which is between RF and SVM. Although the overall performance is not as good as RF, it performs better (close to 0.50) on Yincun Port and Taige Canal, showing certain potential. From the performance of models in different categories, the RF model of Caoqiao River has the best effect, with an R² of 0.86. XGBoost performs moderately, with an R² of 0.47. SVM performs the worst, with an R² of 0.19. RF is still the best in Yincun Port, with an R² of 0.87. XGBoost performs best in this category, with an R² of 0.50. SVM is the weakest, with an R² of 0.18. The RF model of Taige Canal performs best, with an R² of 0.86. XGBoost is second, with an R² of 0.50. SVM is the worst, with an R² of 0.17. RF of Hong Kong performs best, with an R² of 0.87, the highest among the four categories. XGBoost degrades significantly, with an R² of 0.41, slightly worse than the other categories. SVM performs the weakest, with an R² of 0.17.

[0052] Table 2 Accuracy of the three models under different river sediment conditions <![CDATA[R 2 ]]> Caoqiao River Yincun Port Taige Canal Burning Hong Kong RF 0.86 0.87 0.86 0.87 SVM 0.19 0.18 0.17 0.17 XGBoost 0.47 0.50 0.50 0.41 The samples were classified according to the distance between the sediment sample and the corresponding soil sample, and the distance ranges were 0-1km, 1-5km, 5-10km, and 10-20km, and the models were constructed separately (only the distance was considered, and the soil type and the river mentioned in the sediment were not classified). The results are as follows Fig. 9As shown in Table 3, from the overall performance point of view, RF performs stably in all distance ranges, with R² values ​​between 0.86-0.88, which is suitable for fitting the data in the whole range. The R² value of SVM is generally low, only between 0.13-0.21, indicating that SVM has weak fitting ability for the current data and cannot well explain the relationship between variables. XGBoost performs best in the range of 0-1km, with R² reaching 0.99, far exceeding other models. However, in the longer range (1-20km), the performance drops significantly, with R² values ​​between 0.34-0.50, and the stability is insufficient. From the performance of models in different distance ranges, XGBoost performs well in the range of 0-1km, with R² of 0.99, almost achieving perfect fit. RF is second, with R² of 0.88, and the performance is still excellent. SVM performs the worst, with R² of 0.13. RF still performs best in the range of 1-5km, with R² of 0.86 and strong stability. XGBoost is second, with R² of 0.50. SVM performed poorly with an R² of 0.21. RF performed best again at 5-10km with an R² of 0.8667. XGBoost performed moderately with an R² of 0.34. SVM performed the worst with an R² of 0.17. RF performed the best at 10-20km with an R² of 0.86, showing significant stability. XGBoost performed second with an R² of 0.40, showing improvement but still not enough. SVM performed the worst with an R² of 0.17.

[0053] Table 3 Different accuracy of three models at different distances <![CDATA[R 2 ]]> 0-1km 1-5km 5-10km 10-20km RF 0.88 0.86 0.87 0.86 SVM 0.13 0.21 0.17 0.17 XGBoost 0.99 0.50 0.34 0.40 Although the present invention has been disclosed as above with preferred embodiments, it is not intended to limit the present invention. A person with ordinary knowledge in the technical field to which the present invention belongs may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention shall be determined by the definition of the claims.

Claims

1. A method for estimating soil-sediment organic carbon enrichment coefficient based on reflectance spectroscopy, characterized in that: The method comprises: Collect river sediment samples and soil samples from the corresponding riverbanks, measure the organic carbon content, and calculate the soil-sediment organic carbon enrichment factor as the measured organic carbon enrichment factor; Measure the reflectance spectra of the sediment samples and soil samples to obtain the reflectance R rs , and screened the key response bands of organic carbon in sediment reflectance spectrum and soil reflectance spectrum respectively; Constructing an organic carbon enrichment coefficient spectral index based on the key response bands of the sediment reflectance spectrum and the soil reflectance spectrum; Constructing a soil-sediment organic carbon enrichment coefficient spectral estimation model based on the measured organic carbon enrichment coefficient and spectral index; Collect sediment samples and soil samples to be tested, measure the reflectance spectra, calculate the spectral index of organic carbon enrichment coefficient and input it into the estimation model to obtain the soil-sediment organic carbon enrichment coefficient.

2. The estimation method according to claim 1, characterized in that: The key response bands were screened using importance analysis methods.

3. The estimation method according to claim 1 or 2, characterized in that: The key response bands were screened using a variety of importance analysis methods, including: The organic carbon characteristic bands of sediment samples and soil samples were analyzed using a variety of different importance analysis methods and sorted by importance. Each method obtained several organic carbon characteristic bands of sediment samples and soil samples. After performing different mathematical permutations and combinations on the organic carbon characteristic bands of the sediment samples and soil samples obtained by each method, the key response band combinations of organic carbon of the sediment samples and soil samples are screened out respectively by using the multiple importance analysis methods; Different machine learning methods were used in combination with the band combination for testing, and the final key response bands of organic carbon in sediment samples and soil samples were determined based on the model accuracy.

4. The estimation method according to claim 1, characterized in that: The organic carbon enrichment coefficient spectral index is a spectral index in different mathematical forms constructed based on the key response bands of the sediment reflectance spectrum and the key response bands of the soil reflectance spectrum; the different mathematical forms include addition, subtraction, multiplication, division and normalization forms.

5. The estimation method according to claim 1, characterized in that: When calculating the spectral index, the distance range and land use type classification calculation are based on the distance range and land use type classification calculation, and estimation models are constructed for soils of different distance ranges and different land use types respectively; the distance is the distance between the sediment sample and the corresponding soil sample.

6. The estimation method according to claim 4, characterized in that: The soil estimation models for different distance ranges and different land use types are constructed separately, including: The distances between the sediment samples and the corresponding soil samples are divided into different distance range groups according to the ranges of 0-1km, 1-5km, 5-10km and 10-20km, and the estimation models are respectively constructed for the sample data of the different distance range groups; The sediment samples and the corresponding soil samples are divided into different land use type groups according to the land use type to which the soil samples belong, and the estimation models are respectively constructed for the different land use type groups.

7. The estimation method according to claim 1, characterized in that: The estimation model is constructed using random forest, SVM or XGBoost methods.

8. The estimation method according to claim 1, characterized in that: Based on the average R 2 , RMSE or MRE to evaluate the model performance, and the model with the best performance was obtained as the final soil-sediment organic carbon enrichment coefficient spectral estimation model.

9. The estimation method according to claim 1, characterized in that: The collection points of the sediment samples are arranged in the river, and the collected sediment samples are river suspended sediment or bedload; The soil sample collection points are arranged on both sides of the river channel, and the collected soil samples are surface soil.

10. The estimation method according to claim 1, characterized in that: The sediment samples and soil samples were air-dried and ground, and then the reflectance spectra were measured.

Images