Optically fingerprinting sediment sources using a multi-grain-size combination optimization approach

Abstract

The application discloses a multi-particle size combination optimized optical fingerprint silt source tracing method, which comprises the following steps: dividing silt source types into S type and G type, collecting corresponding type samples, wet screening the samples into three particle size ranges respectively, and measuring the spectra; performing PCA-LDA discriminant analysis on the samples obtained after wet screening based on the spectra; preparing a series of mixed samples by different combination modes and different gradient proportions according to the samples through discriminant analysis, and measuring the spectral data for modeling; performing modeling by using a partial least squares regression method, fitting the original spectral data and the spectral data after spectral pretreatment of the modeling set samples with the source sample proportion values of the corresponding samples respectively, establishing a spectral tracing model, and screening the best tracing model from the spectral tracing model; and predicting and evaluating silt sources by using the best tracing model. The application can improve the accuracy of optical fingerprint tracing results.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of soil erosion sediment source identification, and particularly relates to a multi-particle size combination optimized optical fingerprint sediment source tracing method. BACKGROUND

[0002] Identifying the source of erosion sediment is the key to understanding the occurrence, development and prevention of soil erosion and soil and water loss, and has an important reference function for targeted management of the source area and development of soil and water conservation measures. Because the optical characteristics can comprehensively reflect the physical and chemical properties of the sample, different erosion source areas will have differences in material composition, organic matter content, particle size and aggregate characteristics, and the corresponding optical characteristics will also show heterogeneity, such as different shapes, positions, intensities and numbers of characteristic peaks. Therefore, different types of soil have different spectral characteristics, and the spectral characteristics of soil in different particle size ranges of the same type are also different. The optical fingerprint tracing technology takes the optical characteristics of sediment as the fingerprint factor, uses the significant differences in the optical characteristics of sediment from different sources and the conservation of the optical characteristics in the process of erosion and migration to determine the source of sediment and its relative contribution. The optical fingerprint tracing technology has great potential in sediment source identification due to its high identification degree, fast speed, low cost and non-destructive sample. However, during the process of erosion and transportation, the particles are often sorted, that is, coarse particles are deposited first and fine particles are enriched, and the content of organic matter and other components in fine particles is usually higher than that in coarse particles, and the corresponding components are also enriched. These effects will affect the optical characteristics, and cannot ensure the conservation of the optical characteristics from the source to the sink, thereby affecting the accuracy of the optical fingerprint tracing results of sediment sources. The traditional optical fingerprint tracing method of sediment sources does not consider the influence of particle sorting and organic matter enrichment on the accuracy of the tracing results, and usually assumes that the optical characteristics of sediment will not change during the process of erosion and transportation, and only uses the combination of soil types in the source area for modeling without considering the particle size combination, resulting in low matching degree of the optical characteristics of the source sample and the sediment sample, and therefore it is difficult to ensure the accuracy of the tracing results. SUMMARY

[0003] The application provides a multi-particle size combination optimized optical fingerprint sediment source tracing method, which further classifies the source sample based on the particle size, and creates a more comprehensive modeling set sample in the form of multi-particle size combination to solve the problem of inaccurate optical fingerprint tracing results caused by particle sorting and organic matter enrichment.

[0004] The technical scheme adopted by the application is as follows:

[0005] The application discloses a multi-particle size combination optimized optical fingerprint sediment source tracing method, which comprises the following steps:

[0006] Dividing source type: the silt source of the sample to be collected is divided into S-type source and G-type source by the boundary line; wherein, the S-type source is a new loess source, and the G-type source is an old loess source;

[0007] Collecting samples: five sampling areas are selected in the S-type source, and samples are collected in each sampling area to obtain five S-type samples, represented by S1, S2, S3, S4 and S5; five sampling areas are selected in the G-type source, and samples are collected in each sampling area to obtain five G-type samples, represented by G1, G2, G3, G4 and G5; the five S-type samples (S1-S5) and the five G-type samples (G1-G5) are subjected to wet screening treatment to obtain thirty samples S1-31, S1-63, S1-125, S2-31, S2-63, S2-125…S5-31, S5-63, S5-125, G1-31, G1-63, G1-125, G2-31, G2-63, G2-125…G5-31, G5-63, G5-125, wherein, S1-31, S1-63, S1-125 represent the samples with particle size range of 0-31 μm, 31-63 μm and 63-125 μm obtained by wet screening of the S1-type sample, S2-31, S2-63, S2-125 represent the samples with particle size range of 0-31 μm, 31-63 μm and 63-125 μm obtained by wet screening of the S2-type sample, and so on; G1-31, G1-63, G1-125 represent the samples with particle size range of 0-31 μm, 31-63 μm and 63-125 μm obtained by wet screening of the G1-type sample, and so on;

[0008] The thirty samples obtained by wet screening are used for spectral-based silt source sample discrimination analysis, the spectral data of each sample is determined, and the obtained spectral data is subjected to PCA-LDA discrimination analysis to determine whether the samples of different source types and different particle size ranges can be effectively distinguished; if the discrimination passes, the next step is entered, otherwise, the samples are reclassified and the discrimination analysis is performed again until the discrimination analysis passes; wherein, the spectral data includes mid-infrared spectral data, visible spectral data or near-infrared spectral data;

[0009] After passing the judgment analysis, six source samples of the same particle size from the same source area were mixed in the same mass ratio to obtain six source samples: S31, S63, S125, G31, G63, and G125. These six source samples were then combined in different ways and with different gradient ratios to prepare a series of mixed samples, and their spectral data were measured. Two-thirds of the samples were randomly selected from the mixed samples as the modeling set samples, and the remaining one-third were selected as the prediction set samples. Among them, sample S31 is composed of S1-31, S2-31, S3-31, and S4-31 after passing the judgment analysis. Sample S5-31 was prepared by mixing samples S1-63, S2-63, S3-63, S4-63, and S5-63 in a mass ratio of 1:1:1:1:1 after passing the judgment analysis. Sample S125 was prepared by mixing samples G1-31, G2-31, G3-31, G4-31, and G5-31 in a mass ratio of 1:1:1:1:1 after passing the judgment analysis. Samples G63 and G125 were prepared by mixing samples G1-31, G2-31, G3-31, G4-31, and G5-31 in a mass ratio of 1:1:1:1:1.

[0010] The spectral data of the modeling set samples are preprocessed, and partial least squares regression is used for modeling. The original spectral data and the spectral data after spectral preprocessing of the modeling set samples are fitted with the source sample ratio values ​​of the corresponding samples to establish a spectral tracer model, and the best tracer model is selected from them.

[0011] The optimal tracer model is used to predict and assess sediment sources.

[0012] As an optional technical solution, the collection of sample S1 includes: in the first sampling area of ​​the five sampling areas divided into S-type source areas, the top 0-20cm thick soil sample is peeled off as the sampling layer. Within the sampling layer, c subsamples are randomly collected at a grid area of ​​am×bm and mixed evenly, where a and b are both 4-6 and c is 8-12. The mixed sample is then air-dried after removing impurities and then screened through a 125μm sieve to obtain the S1 type sample.

[0013] As an optional technical solution, the wet sieving process for sample S1 includes: adding ultrapure water to sample S1 and mixing it evenly, then placing it in an ultrasonic disperser for dispersion, and wet sieving to three particle size ranges: 0-31μm, 31-63μm, and 63-125μm. Finally, after wet sieving of the S1 sample, three types of samples S1-31, S1-63, and S1-125 are obtained.

[0014] As an optional technical solution, before measuring the spectral data of each sample, the method further includes drying the thirty samples obtained after wet sieving in an oven at 40°C.

[0015] As an optional technical solution, the combination mode includes at least one sample in six source samples S31, S63, S125, G31, G63 and G125.

[0016] As an optional technical solution, the pretreatment includes at least one of convolution smoothing (SG-Smoothing), standard normalization (SNV) and baseline correction (Baseline).

[0017] As an optional technical solution, the optimal tracing model is determined by the coefficient of determination R 2 The two indicators of root mean square error RMSE are screened.

[0018] As an optional technical solution, before the prediction and evaluation of sediment source by the optimal tracing model, the method further includes: the target sediment sample is prepared to obtain mixed samples with different degrees of sorting, and the mixed samples are used to evaluate the prediction accuracy of the optimal tracing model under different particle sorting conditions.

[0019] The beneficial effects of the application are: in the application, two different types of source samples, new loess and old loess, are selected, and the three particle size ranges (0-31 μm, 31-63 μm, 63-125 μm) are further wet screened respectively, and the samples after wet screening are subjected to discriminant analysis; the samples subjected to discriminant analysis are mixed according to the same source, the same particle size range and the same mass ratio to obtain six source samples S31, S63, S125, G31, G63 and G125, the six source samples are prepared into a series of mixed samples according to different combination modes and different gradient proportions, and the spectral data is measured; about 2 / 3 of the samples are randomly selected from the mixed samples as a modeling set, and the remaining about 1 / 3 of the samples are used as a prediction set, and the spectral data of the modeling set samples is subjected to different pretreatment; the partial least squares regression method is used for modeling, the original spectral data and the spectral data after spectral pretreatment of the modeling set samples are fitted with the source sample proportion value of the corresponding sample respectively, the spectral tracing model is established, and the optimal tracing model is screened out; and the mixed samples with different particle size sorting degrees are used to verify the optimal spectral tracing model, and the verification result shows that the maximum value of the mean absolute error MAE between the actual value and the predicted value is 9.2%, and the average value is 4.4%, while the maximum value of the mean absolute error MAE predicted by the traditional spectral modeling method is 35.2%, and the average value is 23.2%, so the prediction accuracy of the application is greatly improved. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 Flowchart of the optical fingerprint sediment source tracing method for multi-particle size combination optimization.

[0021] Figure 2 The images show the mid-infrared spectra of the six source samples.

[0022] Figure 3 This is the result of linear discrimination.

[0023] Figure 4 This is a schematic diagram showing the mean absolute error value of the source type prediction for mixed samples H1-H6 using the method of this embodiment.

[0024] Figure 5 This is a schematic diagram showing the mean absolute error values ​​for predicting the source types of mixed samples H1-H6 using traditional methods.

[0025] Figure 6 This is a schematic diagram showing the predicted mean absolute error (MAE) of the contribution of different particle size ranges in the mixed samples H1-H6 using the method of this embodiment.

[0026] Figure 7 A schematic diagram of the spectral modeling method. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0028] Before describing this embodiment, the background and overall inventive concept of this application will be explained first.

[0029] Because optical characteristics comprehensively reflect the physicochemical properties of a sample, different erosion source areas will exhibit heterogeneity in their optical characteristics due to differences in material composition, organic matter content, particle size, and aggregate characteristics. For example, the shape, position, intensity, and number of characteristic peaks will vary. Therefore, different types of soil have different spectral characteristics, and even soils of the same type but with different particle size ranges will have different spectral characteristics. Figure 2 The data consists of mid-infrared wavenumber-absorbance spectra of six source samples from two different sediment source types and three particle size ranges. These six source samples exhibit different spectral characteristics. PCA-LDA discriminant analysis was used to differentiate their spectral data, and the results are as follows: Figure 3As shown, the six source samples are all aggregated, and there is no cross-over or overlapping part between different source samples, indicating that the six sources can be effectively distinguished by using the spectrum.

[0030] Optical fingerprint tracing of sediment sources is based on the optical characteristics of stable properties, which are significantly different in different source sediments and remain stable during transport, to determine the source of sediment and its relative contribution. However, in reality, sediment is often accompanied by particle sorting during erosion and transport, that is, coarse particles are deposited during transport, and fine particles are preferentially transported to a long distance, and the content of fine-grained organic matter and other components is usually higher than that of coarse-grained particles, and the corresponding components are also enriched. There is a significant difference between the particle size of the collected sediment sample and the source sample, and the greater the difference, the greater the difference in optical characteristics. If only the optical characteristics of the source sample are used for modeling without considering the effects of particle size changes, the matching degree of the optical characteristics of the source sample and the sediment sample will be low, that is, the optical characteristics are not stable during erosion and transport, thereby affecting the accuracy of the optical fingerprint tracing results of the source of sediment.

[0031] The present application further classifies the traditional sediment source sample on the basis of particle size, and a series of mixed samples are prepared in the form of multi-particle size combination to create a more comprehensive modeling sample set that can represent the sediment samples that may be formed under different erosion conditions, thereby establishing a comprehensive spectrum database for subsequent modeling and prediction. Compared with the traditional spectrum tracing method, the multi-particle size combination modeling method avoids the problem of low matching degree of the optical characteristics of the source sample and the sediment sample caused by particle sorting, and provides the possibility for realizing the accuracy of model prediction.

[0032] Taking two kinds of sediment sources of new loess and old loess as examples, on the basis of classification of two kinds of sediment sources of new loess and old loess, the two kinds of sediment sources are further subdivided into three particle size ranges (0-31 μm, 31-63 μm, 63-125 μm), and a total of six categories are obtained. A series of mixed samples (74) are prepared according to different combinations and proportions of the six categories, wherein the mixed samples contain sample combinations of different particle size ranges that can be formed. A model is established by using a partial least squares regression method, 2 / 3 of the prepared mixed samples (74) are randomly selected as a modeling set, a model is established by using a partial least squares regression method, the original spectral data and the spectral data after spectral pretreatment of the samples in the modeling set are fitted with the source sample proportion values of the corresponding samples, a spectral tracing model is established, and the best tracing model is screened from the model. The remaining 1 / 3 samples are used as a prediction set for verification of the model. Whether different spectral pretreatments can improve the prediction ability of the model is explored, and the spectral data after different pretreatments can be used to establish the model. In order to simulate the sediment samples that can be formed under different erosion conditions, six mixed sample sediments H1, H2, H3, H4, H5 and H6 with different soil particle sorting degrees are prepared according to the original proportions of the particle size ranges in the source soil, and the spectral data of the six mixed sample sediments are measured. The spectral data of the six mixed sample sediments are input into the established model, and the source proportions of the six mixed sample sediments are obtained. The source proportions include not only the proportions of the soil types, but also the proportions of the different particle size ranges corresponding to the different soil types. The data statistical analysis is completed by using Unscrambler X 10.4 software (CAMO Software AS., Oslo, Norway).

[0033] However, the traditional spectral tracing method only considers the combination of the source soil types and does not consider the combination of the particle sizes. The new loess and the old loess are only mixed in different proportions, and a model is established by using a partial least squares regression method. However, the particle sizes of the soil types in the modeling set samples are the original proportions, and the particle combinations formed by particle sorting are not included. When the particle sorting degree of the sediment sample is higher, that is, the optical characteristics matching between the source sample and the sediment sample is lower, the tracing result is less accurate.

[0034] Embodiment

[0035] The embodiment discloses a multi-particle size combination optimized optical fingerprint sediment source tracing method, which comprises the following steps:

[0036] a. Dividing source types: dividing the sediment source of the sample to be collected into S-type source and G-type source according to the gully edge line; wherein the S-type source is a new loess source, and the G-type source is an old loess source;

[0037] b, collecting samples: selecting five sampling areas in the S-type source area, collecting samples in each sampling area to obtain five S-type samples, denoted as S1, S2, S3, S4, S5; selecting five sampling areas in the G-type source area, collecting samples in each sampling area to obtain five G-type samples, denoted as G1, G2, G3, G4, G5; wet screening the five S-type samples (S1-S5) and five G-type samples (G1-G5) collected, to obtain thirty samples S1-31, S1-63, S1-125, S2-31, S2-63, S2-125…S5-31, S5-63, S5-125, G1-31, G1-63, G1-125, G2-31, G2-63, G2-125…G5-31, G5-63, G5-125, wherein S1-31, S1-63, S1-125 represent the samples with particle size range of 0-31 μm, 31-63 μm, 63-125 μm obtained by wet screening the S1-type sample, S2-31, S2-63, S2-125 represent the samples with particle size range of 0-31 μm, 31-63 μm, 63-125 μm obtained by wet screening the S2-type sample, and so on; G1-31, G1-63, G1-125 represent the samples with particle size range of 0-31 μm, 31-63 μm, 63-125 μm obtained by wet screening the G1-type sample, and so on;

[0038] c, using the thirty samples obtained by wet screening to perform spectral-based sediment source sample discrimination analysis, measuring the spectral data of each sample, and performing PCA-LDA discrimination analysis on the obtained spectral data to determine whether samples of different source types and different particle size ranges can be effectively distinguished, if the discrimination analysis passes, then proceed to the next step, otherwise, reclassify and perform discrimination analysis again until the discrimination analysis passes; wherein the spectral data includes mid-infrared spectral data, visible spectral data or near-infrared spectral data; reclassification means reclassifying the S-type samples and G-type samples that cannot pass the discrimination analysis according to particle size until the PCA-LDA analysis can completely distinguish samples from different source areas and with different particle sizes;

[0039] d. After the judgment analysis is passed, samples of the same particle size from the same source area are mixed in the same mass ratio to obtain six source samples: S31, S63, S125, G31, G63, and G125. These six source samples are then combined in different ways and with different gradient ratios to prepare a series of mixed samples, and their spectral data are measured. Two-thirds of the samples are randomly selected from the mixed samples as the modeling set samples, and the remaining one-third are selected as the prediction set samples. Among them, sample S31 is the mixture of S1-31, S2-31, S3-31, and S4-31 after the judgment analysis is passed. Sample S5-31 was prepared by mixing samples S1-63, S2-63, S3-63, S4-63, and S5-63 in a mass ratio of 1:1:1:1:1 after passing the judgment analysis. Sample S125 was prepared by mixing samples G1-31, G2-31, G3-31, G4-31, and G5-31 in a mass ratio of 1:1:1:1:1 after passing the judgment analysis. Samples G63 and G125 were prepared by mixing samples G1-31, G2-31, G3-31, G4-31, and G5-31 in a mass ratio of 1:1:1:1:1.

[0040] e. Preprocess the spectral data of the modeling set samples, use partial least squares regression to model, fit the original spectral data and the spectral data after spectral preprocessing of the modeling set samples with the source sample ratio values ​​of the corresponding samples, establish a spectral tracer model, and select the best tracer model from it.

[0041] f. Use the optimal tracer model to predict and assess sediment sources.

[0042] As an optional implementation method, the collection of sample S1 includes: in the first sampling area of ​​the five sampling zones divided into S-type source areas, a 0-20cm thick top layer of soil is peeled off as a sampling layer. Within a grid area of ​​am × bm in the sampling layer, c subsamples are randomly collected and uniformly mixed, where a and b are both 4-6, and c is 8-12. In this embodiment, a and b are both 5, and c is 10. The mixed sample is then cleaned of impurities, air-dried, and pre-sieved through a 125μm sieve to obtain the S1 type sample. The collection methods for samples S2-S5 and samples G1-G5 are the same as for sample S1, and samples are collected from their respective designated sampling areas.

[0043] As an optional implementation, the wet screening treatment of the sample S1 includes: adding ultrapure water in the sample S1 and mixing uniformly, and then placing in an ultrasonic disperser for dispersion, wet screening to three particle size ranges: 0-31 μm, 31-63 μm, and 63-125 μm, and finally obtaining three samples S1-31, S1-63, and S1-125 after wet screening of the sample S1. The wet screening treatment processes of the samples S2-S5 and the samples G1-G5 are the same as the wet screening treatment process of the sample S1.

[0044] As an optional implementation, before the determination of the spectral data of each sample, the thirty samples obtained after wet screening are placed in an oven at 40℃ for drying.

[0045] As an optional implementation, the combination mode includes at least one sample in the six source samples S31, S63, S125, G31, G63, and G125.

[0046] As an optional implementation, the pretreatment includes at least one of a convolution smoothing method SG-Smoothing (SG is Savitzky-Golay), a standard normalization SNV (SNV is Standard Normalized Variate), and baseline correction Baseline.

[0047] As an optional implementation, the optimal tracing model is determined by a coefficient R 2 and a root mean square error RMSE.

[0048] As an optional implementation, before the prediction and evaluation of the sediment source by using the optimal tracing model, the target deposition sample is prepared to obtain mixed samples with different degrees of sorting, and the mixed samples are used to evaluate the prediction accuracy of the optimal tracing model under different particle sorting conditions.

[0049] In order to better understand the present application, the present application is further described below.

[0050] A typical small watershed Shayan Gully watershed in the loess plateau hilly and gully region with serious soil erosion is taken as an actual application region.

[0051] 1. Divide source area types and collect samples. Divide the source area into two types of new loess (S) and old loess (G) according to the boundary line of the gully rim, and collect source sample soil. In the S-type source area, divide five sampling areas, and collect one sample in each sampling area to obtain samples S1-S5. Similarly, in the G-type source area, also divide five sampling areas, and collect one sample in each sampling area to obtain samples G1-G5. The way of collecting samples in each sampling area is consistent. Taking sample S1 as an example, when collecting, select a representative and easily eroded soil sample in the corresponding sampling area, and use a stainless steel shovel to peel off the 0-20 cm thick upper layer soil sample as the sampling layer. Each sample is composed of 5×5 m 2 The 10 subsamples collected in the grid area are uniformly mixed. After removing animal and plant residues, gravel and other impurities, the collected samples are naturally air-dried, and uniformly passed through a 125 μm aperture sieve, i.e. sample S1 is obtained.

[0052] 2、Sample wet screening treatment. Add ultrapure water to the sieved samples (S1-S5 and G1-G5, a total of ten samples) respectively and mix uniformly, then put into an ultrasonic disperser for dispersion. According to the particle composition characteristics of the new loess, old loess and eroded and deposited sediment, the ten samples are wet sieved into three particle size ranges: 0-31 μm, 31-63 μm, 63-125 μm, obtaining thirty samples, represented as: S1-31, S1-63, S1-125, S2-31, S2-63, S2-125, S3-31, S3-63, S3-125, S4-31, S4-63, S4-125, S5-31, S5-63, S5-125; G1-31, G1-63, G1-125, G2-31, G2-63, G2-125, G3-31, G3-63, G3-125, G4-31, G4-63, G4-125, G5-31, G5-63, G5-125; wherein S1-31, S1-63, S1-125 respectively represent the samples of particle size range 0-31 μm, 31-63 μm, 63-125 μm obtained after wet sieving of S1 type samples, S2-31, S2-63, S2-125 respectively represent the samples of particle size range 0-31 μm, 31-63 μm, 63-125 μm obtained after wet sieving of S2 type samples, S3-31, S3-63, S3-125 respectively represent the samples of particle size range 0-31 μm, 31-63 μm, 63-125 μm obtained after wet sieving of S3 type samples, S4-31, S4-63, S4-125 respectively represent the samples of particle size range 0-31 μm, 31-63 μm, 63-125 μm obtained after wet sieving of S4 type samples, S5-31, S5-63, S5-125 respectively represent the samples of particle size range 0-31 μm, 31-63 μm, 63-125 μm obtained after wet sieving of S5 type samples; G1-31, G1-63, G1-125 respectively represent the samples of particle size range 0-31 μm, 31-63 μm, 63-125 μm obtained after wet sieving of G1 type samples, G2-31, G2-63, G2-125 respectively represent the samples of particle size range 0-31 μm, 31-63 μm, 63-125 μm obtained after wet sieving of G2 type samples, G3-31, G3-63, G3-125 respectively represent the samples of particle size range 0-31 μm, 31-63 μm, 63-125 μm obtained after wet sieving of G3 type samples, G4-31, G4-63, G4-125 respectively represent the samples of particle size range 0-31 μm, 31-63 μm, 63-125 μm obtained after wet sieving of G4 type samples, G5-31, G5-63, G5-125 respectively represent the samples of particle size range 0-31 μm, 31-63 μm, 63-125 μm obtained after wet sieving of G5 type samples.

[0053] It should be noted that the sample particles in the particle size range of 0-31μm in this application do not include 0μm particles, but do include 31μm sample particles; all sample particles in the 31-63μm range in this application do not include 31μm sample particles, but do include 63μm sample particles; and all sample particles in the 63-125μm range in this application do not include 63μm sample particles, but do include 125μm sample particles.

[0054] 3. Discriminant Analysis. Thirty samples obtained after wet sieving were dried in an oven at 40°C, and their spectral data were then measured. The spectral data could be mid-infrared, visible, or near-infrared. In this embodiment, mid-infrared spectroscopy was used as an example. A ThermoNicolet iS50 infrared spectrometer (ThermoNicolet, USA) was used to measure the mid-infrared spectral data. The mid-infrared spectra of the six source samples are shown below. Figure 2 As shown. A total of thirty samples from six different sources with varying particle sizes were included (S1-31, S1-63, S1-125, S2-31, S2-63, S2-125, S3-31, S3-63, S3-125, S4-31, S4-63, S4-125, S5-31, S5-63, S5-125, G1-31, G1-63, G1-125, G2-31, G2-63, G2-125, G3-31, G3-63, G3-125, G4-3...). 1. PCA-LDA analysis (PCA is principal component analysis, LDA is linear discriminant analysis) was performed on the spectral data of G4-63, G4-125, G5-31, G5-63, and G5-125. The results showed that the six source samples aggregated independently, with no overlap or intersection between different source samples. This indicates that spectral analysis can effectively distinguish these six source samples (samples of the same particle size from the same type of source are considered to be of the same source, such as S1-31, S2-31, S3-31, S4-31, and S5-31). Specific results are as follows... Figure 3After the determination analysis, the same size samples from the same source were mixed according to the same mass ratio to obtain six source samples S31, S63, S125, G31, G63, and G125. Among them, sample S31 is a mixture of S1-31, S2-31, S3-31, S4-31, and S5-31 samples after the determination analysis, mixed according to a mass ratio of 1:1:1:1:1, sample S63 is a mixture of S1-63, S2-63, S3-63, S4-63, and S5-63 samples after the determination analysis, mixed according to a mass ratio of 1:1:1:1:1, and sample S125 is mixed in the same way; sample G31 is a mixture of G1-31, G2-31, G3-31, G4-31, and G5-31 samples after the determination analysis, mixed according to a mass ratio of 1:1:1:1:1, and samples G63 and G125 are mixed in the same way.

[0055] 4. Preparation of modeling set samples. Six source samples S31, S63, S125, G31, G63, and G125 of three particle size ranges from two sources are prepared into a series of mixed samples in different combinations, and a gradient ratio is set under each combination condition. The purpose of this step is to create a comprehensive modeling sample set that can represent the sediment samples that may be formed under different erosion conditions, thereby establishing a comprehensive spectral database for subsequent modeling and prediction. This multi-particle size combination modeling method provides a basis for achieving model accuracy. The specific proportions are shown in Table 1 below.

[0056] Table 1, preparation ratio of modeling samples

[0057]

[0058]

[0059]

[0060] 5. Model establishment: a spectral tracing model is established using a regression method, and the modeling process is referred to Figure 7 . In the modeling process, the spectral data of the modeling samples are first preprocessed, and in this application, data after various preprocessing (SG-Smoothing, SNV, and Baseline) and raw data (Raw) are used to establish a spectral data tracing model in combination with a partial least squares regression method (PLSR). The best tracing model is selected for subsequent analysis through two indicators, the determination coefficient (R 2 ) and the root mean square error (RMSE). The higher the R 2 , the better the fitting effect of the model, the smaller the RMSE, and the higher the stability of the model. Among them, the raw data Raw refers to the spectral data without preprocessing.

[0061] Specifically, the root mean square error (RMSE) is calculated according to the following formula:

[0062]

[0063] In the formula, y i These are measured values. This is the estimated value, and n is the number of samples.

[0064] Specifically, the coefficient of determination R 2 Calculate according to the following formula:

[0065]

[0066] In the formula, y i These are measured values. This is an estimated value. is the average value, and n is the number of samples.

[0067] Specifically, 50 samples (approximately 2 / 3 of the total samples) from the 74 total samples in Table 1 were randomly selected as the modeling set for model building, and leave-one-out cross-validation was used to validate the model. The remaining 24 samples (approximately 1 / 3 of the total samples), with known true proportions, were used as the prediction set to verify the accuracy of the tracer results. After screening the established tracer models, SNV combined with partial least squares regression (PLSR) was found to be the optimal tracer model, as shown in Table 2 below.

[0068] Table 2. Prediction accuracy of different preprocessing methods combined with partial least squares regression model

[0069]

[0070] Among them, RMSE C R represents the root mean square error calculated using the modeling set. C 2 Represents the coefficient of determination calculated using the modeling set; RMSE p R represents the root mean square error calculated using the prediction set. p 2 This represents the coefficient of determination calculated using the prediction set.

[0071] 6、Target sediment sample preparation: To simulate the sediment samples that could be formed under different erosion conditions. Six soil mixtures with different degrees of particle size sorting were prepared, denoted as H1, H2, H3, H4, H5, and H6. The proportion of the 0-31 μm fraction was calculated based on the original proportions of each particle size range in the source soil (Table 3) and the enrichment ratios set for the 31-63 μm and 63-125 μm fractions (Table 4). The final proportions of the six soil mixtures with different degrees of particle size sorting were obtained (Table 5). These six mixtures were used to evaluate the prediction accuracy of the model under different particle sorting conditions. The preparation of the mixtures not only considered soil from a single source but also considered combinations of different sources and particle sizes, thus more comprehensively simulating the mixing process of sediment under natural conditions. In this way, researchers could evaluate the robustness and accuracy of the model in dealing with complex and variable sediment samples. The specific preparation is as follows.

[0072]

[0073] where ER is the enrichment ratio of particles in a certain particle size range; PS is the mass percentage content of particles in a certain particle size range in the target sediment (%); and PS is the mass percentage content of particles in a certain particle size range in the original soil (%). PS sediment source

[0074] Table 3, Mass percentage of each particle size range in new and old loess

[0075]

[0076] Table 4, Enrichment ratio of each particle size range in different mixtures

[0077]

[0078] Table 5, Preparation ratio of mixtures

[0079]

[0080]

[0081] 7、Sediment source prediction and evaluation: The optimal tracer model (SNV source prediction and evaluation) was used to predict the proportion of source types (i.e., S type or G type) of mixtures H1-H6, as follows: Figure 4 ​​​As shown, the results show the average absolute error value (MAE) of the prediction of the source type of mixed samples H1-H6 is 1.19%, 2.12%, 8.33%, 1.11%, 9.17%, 4.40% respectively, and the average MAE value is 4.4%. In addition, the traditional spectral modeling method is used for prediction of the source type: the traditional spectral tracking method only divides the source samples into two categories S and G, and performs PCA-LDA discriminant analysis, under the condition of passing the discrimination, the S and G are mixed according to different proportions to prepare mixed samples, the spectrum of the mixed samples is obtained as the modeling set, then the spectrum is preprocessed and combined with the PLSR method for modeling, and each category is not modeled after particle size division. The prediction results of the traditional spectral tracking method are as follows Figure 5 As shown, the average absolute error value (MAE) of the prediction of the mixed samples H1-H6 is 13.20%, 12.44%, 24.72%, 30.54%, 22.94%, 35.22% respectively, and the average MAE value is 23.2%. With the increase of the particle size sorting degree (simulated by the soil particle enrichment degree), the recognition accuracy gradually decreases, and the MAE value is as high as 35.2%. The results show that using the modeling method of this embodiment based on the classification of the source samples of the silt can accurately identify the source contribution of the two source types of new and old loess, and the recognition accuracy remains at a high level without obvious change with the increase of the particle size sorting degree, indicating that the modeling method of silt source tracking by multi-particle size combination can maintain high accuracy under different particle size sorting degrees. In addition, the modeling method optimized by multi-particle size combination can also predict the source contribution of different particle size ranges (i.e. S31, S63, S125, G31, G63, G125, a total of 6 sources) in the sample, as shown in Figure 6 As shown, the average absolute error value (MAE) of the prediction of the mixed samples H1-H6 is 13.20%, 12.44%, 24.72%, 30.54%, 22.94%, 35.22% respectively, and the average MAE value is 23.2%. With the increase of the particle size sorting degree (simulated by the soil particle enrichment degree), the recognition accuracy gradually decreases, and the MAE value is as high as 35.2%. The results show that using the modeling method of this embodiment based on the classification of the source samples of the silt can accurately identify the source contribution of the two source types of new and old loess, and the recognition accuracy remains at a high level without obvious change with the increase of the particle size sorting degree, indicating that the modeling method of silt source tracking by multi-particle size combination can maintain high accuracy under different particle size sorting degrees. In addition, the modeling method optimized by multi-particle size combination can also predict the source contribution of different particle size ranges (i.e. S31, S63, S125, G31, G63, G125, a total of 6 sources) in the sample, as shown in

[0082] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which shall be covered within the protection scope of the present application, and any technical solution falling within the scope defined by the claims of the present application shall fall within the protection scope of the present application.

Claims

1. A method for tracing the source of mud and sand in optical fingerprints using multi-particle-size combination optimization, characterized in that, include: Source area classification: The sediment source areas of the samples to be collected are divided into S-type source areas and G-type source areas, with the gully edge line as the boundary; among them, the S-type source area is the new loess source area, and the G-type source area is the old loess source area; Sample collection: Five sampling areas were selected within the S-type source area, and five S-type samples were collected from each area, denoted as S1, S2, S3, S4, and S5. Similarly, five sampling areas were selected within the G-type source area, and five G-type samples were collected from each area, denoted as G1, G2, G3, G4, and G5. The collected S-type samples (S1-S5) and G-type samples (G1-G5) were subjected to wet sieving to obtain thirty samples across three particle size ranges: S1-31, S1-63, S1-125, S2-31, S2-63, S2-125…S5-31, S5-63, S5-125, G1-31, G1-63, G1-125. G2-31, G2-63, G2-125…G5-31, G5-63, G5-125, where S1-31, S1-63, and S1-125 represent samples with particle sizes ranging from 0-31μm, 31-63μm, and 63-125μm obtained after wet sieving of S1 type samples, respectively; S2-31, S2-63, and S2-125 represent samples with particle sizes ranging from 0-31μm, 31-63μm, and 63-125μm obtained after wet sieving of S2 type samples, respectively; and so on. G1-31, G1-63, and G1-125 represent samples with particle sizes ranging from 0-31μm, 31-63μm, and 63-125μm obtained after wet sieving of G1 type samples, respectively; and so on. Thirty samples obtained after wet sieving were used for spectral-based sediment source discrimination analysis. The spectral data of each sample were measured, and PCA-LDA discrimination analysis was performed on the obtained spectral data to determine whether samples of different source types and different particle size ranges could be effectively distinguished. If the discrimination analysis passed, the process proceeded to the next step; otherwise, the samples were reclassified and the discrimination analysis was performed again until the discrimination analysis passed. The spectral data included mid-infrared spectral data, visible light spectral data, or near-infrared spectral data. After passing the judgment analysis, six source samples of the same particle size from the same source area were mixed in the same mass ratio to obtain six source samples: S31, S63, S125, G31, G63, and G125. These six source samples were then combined in different ways and with different gradient ratios to prepare a series of mixed samples, and their spectral data were measured. Two-thirds of the samples were randomly selected from the mixed samples as the modeling set samples, and the remaining one-third were selected as the prediction set samples. Among them, sample S31 is composed of S1-31, S2-31, S3-31, and S4-31 after passing the judgment analysis. Sample S5-31 was prepared by mixing samples S1-63, S2-63, S3-63, S4-63, and S5-63 in a mass ratio of 1:1:1:1:1 after passing the judgment analysis. Sample S125 was prepared by mixing samples G1-31, G2-31, G3-31, G4-31, and G5-31 in a mass ratio of 1:1:1:1:1 after passing the judgment analysis. Samples G63 and G125 were prepared by mixing samples G1-31, G2-31, G3-31, G4-31, and G5-31 in a mass ratio of 1:1:1:1:

1. The spectral data of the modeling set samples are preprocessed, and partial least squares regression is used for modeling. The original spectral data and the spectral data after spectral preprocessing of the modeling set samples are fitted with the source sample ratio values ​​of the corresponding samples to establish a spectral tracer model, and the best tracer model is selected from them. The optimal tracer model is used to predict and assess sediment sources.

2. The optical fingerprint mud source tracing method with multi-particle-size combination optimization according to claim 1, characterized in that, The collection of sample S1 includes: In the first of the five sampling areas in the S-shaped source area, the top 0-20cm thick soil layer was stripped off as the sampling layer. Within the sampling layer, c subsamples were randomly collected and uniformly mixed within a grid area of ​​am×bm. The values ​​of a and b were both 4-6, and the value of c was 8-12. The mixed sample was decontaminated and air-dried, then sieved through a 125μm sieve to obtain the S1 type sample.

3. The optical fingerprint mud source tracing method with multi-particle-size combination optimization according to claim 1, characterized in that, The wet sieving process for sample S1 includes: adding ultrapure water to sample S1 and mixing it evenly, then placing it in an ultrasonic disperser for dispersion, and wet sieving to three particle size ranges: 0-31μm, 31-63μm, and 63-125μm. Finally, after wet sieving of the S1 sample, three samples S1-31, S1-63, and S1-125 were obtained.

4. The optical fingerprint mud source tracing method with multi-particle size combination optimization according to claim 1, characterized in that: Before determining the spectral data for each sample, the process also includes drying the thirty samples obtained after wet sieving in an oven at 40°C.

5. The optical fingerprint mud and sand source tracing method with multi-particle-size combination optimization according to claim 1, characterized in that: The combination method includes at least one of the six source samples: S31, S63, S125, G31, G63, and G125.

6. The optical fingerprint mud and sand source tracing method with multi-particle-size combination optimization according to claim 1, characterized in that: The preprocessing includes at least one of convolutional smoothing (SG-Smoothing), standard normalization (SNV), and baseline correction.

7. The optical fingerprint mud source tracing method with multi-particle-size combination optimization according to claim 1, characterized in that: The optimal tracer model is determined by the coefficient of determination R. 2 The selection was based on two metrics: root mean square error (RMSE).

8. The optical fingerprint mud source tracing method with multi-particle-size combination optimization according to claim 1, characterized in that, Before using the optimal tracer model to predict and assess sediment sources, the method further includes: preparing target sediment samples to obtain mixtures with different degrees of sorting, wherein the mixtures are used to evaluate the prediction accuracy of the optimal tracer model under different particle sorting conditions.

Images