CN115656416A - Prediction Method of Soil Organic Matter Based on Spectral Shape Features - Google Patents

Abstract

Soil organic matter prediction method based on spectral shape features. The invention belongs to the field of spectrum prediction of soil organic matter, and in particular relates to a method for estimating soil organic matter. The method is as follows: 1. Soil sample collection and processing; 2. Soil sample spectral testing; 3. Soil spectral processing and spectral characteristic parameter extraction; 4. SOM prediction model construction to estimate SOM content. The model constructed by the invention has high prediction accuracy, and the spectral characteristic parameters based on shape features have great potential in SOM prediction.

Description

Prediction Method of Soil Organic Matter Based on Spectral Shape Features

technical field

The invention belongs to the field of spectrum prediction of soil organic matter, and in particular relates to a method for estimating soil organic matter.

Background technique

Soil organic matter (SOM) is an important physical and chemical parameter, and SOM is an important index for soil safety and fertility evaluation. Fast and accurate SOM estimation is required in precision agriculture and sustainable land use. People use spectral technology to understand the differences in physical and chemical properties of different ground objects. The specific spectral differences are mainly reflected in the reflection and absorption characteristics of different spectral wavelengths. Studies have shown that SOM has strong absorption characteristics in the visible and near-infrared region, which can be used to quickly and accurately estimate the SOM content. With the increase of SOM content, the spectral reflectance decreases, therefore, SOM has a good negative correlation with spectral reflectance. In order to improve the correlation between SOM and spectral reflectance, mathematical transformation methods such as differential, logarithm, and reciprocal are applied to SOM prediction. Among them, differential transformation is the most commonly used spectral transformation method, which can not only improve the correlation with SOM, but also eliminate the influence of vegetation. In the current study, the fractional order differential is applied to SOM prediction, which makes up for the fact that the spectral information cannot be fully captured by the integer order differential in SOM prediction. However, for soil spectral SOM prediction, the reflectance or mathematical transformation values of many wavelengths are generally used as predictor variables, which increases the computational intensity.

The spectral characteristic parameters can characterize the shape characteristics of the spectral curve. Generally, the spectral characteristic parameters are extracted based on the spectral de-envelope curve, mainly including absorption area, absorption position, absorption depth, and symmetry. Spectral characteristic parameters are mostly used in the study of soil spectral classification. Different soil types have large differences in the spectral characteristics of visible light. Therefore, using spectral characteristic parameters for soil classification can obtain high classification accuracy. There are few studies on the use of spectral characteristic parameters for SOM prediction, but as the SOM content increases in visible light, when the reflectivity decreases, the shape of the spectral curve also changes. In order to study the potential of spectral characteristic parameters in SOM prediction, a SOM prediction method based on spectral curve shape characteristics is proposed. It is necessary to develop a fast, accurate and less predictive variable SOM prediction method, which can greatly improve the prediction efficiency. While using spectral feature parameter prediction, local modeling methods are also applied to improve the predictive ability of the model.

Contents of the invention

The purpose of the present invention is to solve the technical problem of complex computation due to too many predictor variables used in current SOM prediction, and to provide a soil organic matter prediction method based on spectral shape features.

The prediction method of soil organic matter based on spectral shape features is as follows:

1. Soil sample collection and processing:

Collect surface soil samples of cultivated land in Northeast China in Northeast China, air-dry, grind and sieve the soil samples, use potassium dichromate heating method, and obtain SOM content through the conversion coefficient of 1.724;

The operating steps of the potassium dichromate heating method are as follows:

1.1. Weigh 0.1000g-0.5000g air-dried soil sample with the subtraction method, accurate to 0.0001g, put the air-dried soil sample into a hard test tube through a 0.149mm sieve, add 0.1g silver sulfate, add 5.00mL concentration 0.8000mol/L potassium dichromate standard solution, then inject 5mL sulfuric acid with a syringe, rotate and shake well;

1.2. First heat the oil bath to 185°C-190°C, insert the hard test tube containing the soil sample into the wire cage in the oil bath to heat, control the temperature in the oil bath to 170°C-180°C, and Keep the solution boiling for 5 minutes, then take out the wire cage, and after the hard test tube is slightly cooled, wipe off the oil on the outside of the test tube with clean paper;

1.3. If the boiled solution turns green, it means that the amount of potassium dichromate standard solution is insufficient, and the soil sample should be weighed again; if the boiled solution turns orange-yellow or yellow-green, inhale 250mL of the mixture in the test tube after cooling In the Erlenmeyer flask, the inner volume of the bottle is controlled at 60mL-80mL, add 3-4 drops of o-phenanthroline indicator, and titrate with 0.2mol/L ferrous ammonium sulfate standard solution until the solution changes from orange-yellow to blue-green to brown-red as the end point; if N-phenylanthranilic acid indicator is used, the discoloration process is from brownish red to purple to blue green as the end point;

1.4. When analyzing each batch of soil samples, do 2-3 blank tests. No soil samples are added to the blank tests, but 0.1g-0.5g of quartz sand is added. The other operating steps are exactly the same as the soil sample analysis.

2. Spectral test of soil samples:

Use the ASD ground object spectrometer to carry out spectral testing in the dark room. The spectral range is 350-2500nm. Put the soil sample into the container to scrape the surface. Use a 50W halogen lamp as the light source to irradiate the surface of the soil sample. Use the probe of the ASD ground object spectrometer Collect the soil spectrum, collect 10 spectral curves each time, and calculate the average value as the spectral reflectance curve of a soil sample point;

3. Soil spectral processing:

Nine-point smoothing, 10nm resampling, and envelope removal are performed on the spectral reflectance curve, and the absorption positions of the two absorption valleys in the visible light are determined through the envelope removal curve. Based on the two absorption positions, the spectral reflectance curve is extracted Spectral characteristic parameters length L and area A;

The formula for the nine-point smoothing is as follows:

R′ _i ＝0.04R _i-4 +0·08R _i-3 +0.12R _i- 2 +0.16R _i-1 +0.2R _i +0.16R _i+1 +0.12R _i+2 +0.08R _i+3 +0.04R _i+4

_R'i represents the reflectance of a certain wavelength after smoothing, R represents the measured reflectance, and i represents the wavelength;

The formula of the characteristic parameter length L is:

n=180°-arctan(α)

In the length L formula, λ represents the wavelength, λ _a represents the wavelength a, R represents the reflectivity, β represents the radius, and π=3.14;

The characteristic parameter area A formula is:

A _c ＝A _abc -A _▲abc

A _▲abc = [(P ₂ -P ₁ )×(R _c -R _a )]×0.5

Among them, λ represents the wavelength, λ _a represents the wavelength a, d represents the resampling interval, R represents the reflectivity, P represents the absorption position, n and α represent the included angle (see Figure 2 of the specification), β represents the radius, π=3.14 ;

4. Clustering of soil samples: use k-mean clustering to cluster surface soil samples of cultivated land in Northeast China, and determine the optimal number of clusters;

5. Construction of SOM prediction model:

The prediction model uses the random forest model, and the spectral characteristic parameter length L and area A are used as predictor variables. The random forest model is constructed in the software R, and the SOM content is estimated to obtain the predicted soil organic matter.

In step 1, soil samples of 0.5-20 cm of the cultivated land surface are collected.

In the second step, the ASD surface object spectrometer is used for spectral testing, and the spectral range is 350-2500nm.

In step five, the random forest model is constructed using the RandomForest package. When the number of trees is set to 500 and the optimal split node is set to one-third of the predictor variable, the out-of-bag error of the model is stable.

Step 5 After the SOM prediction model is constructed, the SOM prediction accuracy is evaluated using the coefficient of determination and the root mean square error. The RMSE calculation formula is as follows:

(此公式中的yp_i和yo_i是y_pi和y_oi)

(yp _i and yo _i in this formula are y _pi and y _oi )

where y _p is the predicted value of SOC, y _o is the measured value of SOC, and m is the number of soil samples.

The model constructed by the invention has high prediction accuracy, and the spectral characteristic parameters based on shape features have great potential in SOM prediction.

Description of drawings

Fig. 1 is a distribution diagram of soil sampling points in experiment one of the present invention;

Fig. 2 is a schematic diagram of extracting spectral feature parameters in Experiment 1 of the present invention, among which (a) is a schematic diagram of length, and (b) is a schematic diagram of area;

Fig. 3 is the training result of Nong'an County in experiment one of the present invention;

Fig. 4 is the verification result of Nong'an County in experiment one of the present invention;

Fig. 5 is the training result of the Sanjiang Plain in Experiment 1 of the present invention;

Fig. 6 is the verification result of the Sanjiang Plain in Experiment 1 of the present invention.

Detailed ways

The technical solution of the present invention is not limited to the specific embodiments listed below, but also includes any combination of the specific embodiments.

Specific implementation mode one: the method for predicting soil organic matter based on spectral shape features in this implementation mode is as follows:

1. Soil sample collection and processing:

Collect surface soil samples of cultivated land in Northeast China in Northeast China, air-dry, grind and sieve the soil samples, use potassium dichromate heating method, and obtain SOM content through the conversion coefficient of 1.724;

The operating steps of the potassium dichromate heating method are as follows:

1.1. Weigh 0.1000g-0.5000g air-dried soil sample with the subtraction method, accurate to 0.0001g, put the air-dried soil sample into a hard test tube through a 0.149mm sieve, add 0.1g silver sulfate, add 5.00mL concentration 0.8000mol/L potassium dichromate standard solution, then inject 5mL sulfuric acid with a syringe, rotate and shake well;

1.2. First heat the oil bath to 185°C-190°C, insert the hard test tube containing the soil sample into the wire cage in the oil bath to heat, control the temperature in the oil bath to 170°C-180°C, and Keep the solution boiling for 5 minutes, then take out the wire cage, and after the hard test tube is slightly cooled, wipe off the oil on the outside of the test tube with clean paper;

1.3. If the boiled solution turns green, it means that the amount of potassium dichromate standard solution is insufficient, and the soil sample should be weighed again; if the boiled solution turns orange-yellow or yellow-green, inhale 250mL of the mixture in the test tube after cooling In the Erlenmeyer flask, the inner volume of the bottle is controlled at 60mL-80mL, add 3-4 drops of o-phenanthroline indicator, and titrate with 0.2mol/L ferrous ammonium sulfate standard solution until the solution changes from orange-yellow to blue-green to brown-red as the end point; if N-phenylanthranilic acid indicator is used, the discoloration process is from brownish red to purple to blue green as the end point;

1.4. When analyzing each batch of soil samples, do 2-3 blank tests. No soil samples are added to the blank tests, but 0.1g-0.5g of quartz sand is added. The other operating steps are exactly the same as the soil sample analysis.

2. Spectral test of soil samples:

Use the ASD ground object spectrometer to carry out spectral testing in the dark room. The spectral range is 350-2500nm. Put the soil sample into the container to scrape the surface. Use a 50W halogen lamp as the light source to irradiate the surface of the soil sample. Use the probe of the ASD ground object spectrometer Collect the soil spectrum, collect 10 spectral curves each time, and calculate the average value as the spectral reflectance curve of a soil sample point;

3. Soil spectral processing:

Nine-point smoothing, 10nm resampling, and envelope removal are performed on the spectral reflectance curve, and the absorption positions of the two absorption valleys in the visible light are determined through the envelope removal curve. Based on the two absorption positions, the spectral reflectance curve is extracted Spectral characteristic parameters length L and area A;

The formula for the nine-point smoothing is as follows:

R′ _i ＝0.04R _i-4 +0.08R _i-3 +0.12R i-2 +0.16R _{i-1 +} 0.2R _i +0.16R _i+1 +0.12R _i+2 +0.08R _i+3 + 0.04R _i+4

_R'i represents the reflectance of a certain wavelength after smoothing, R represents the measured reflectance, and i represents the wavelength;

The formula of the characteristic parameter length L is:

n=180°-arctan(θ)

In the length L formula, λ represents the wavelength, λ _a represents the wavelength a, R represents the reflectivity, β represents the radius, and π=3.14;

The characteristic parameter area A formula is:

A _c ＝A _abc -A _▲abc

A _▲abc = [(P ₂ -P ₁ )×(R _c -R _a )]×0.5

Among them, λ represents the wavelength, λ _a represents the wavelength a, d represents the resampling interval, R represents the reflectivity, P represents the absorption position, n and α represent the included angle (see Figure 2 of the specification), β represents the radius, π=3.14 ;

4. Clustering of soil samples: use k-mean clustering to cluster surface soil samples of cultivated land in Northeast China, and determine the optimal number of clusters;

5. Construction of SOM prediction model:

The prediction model uses the random forest model, and the spectral characteristic parameter length L and area A are used as predictor variables. The random forest model is constructed in the software R, and the SOM content is estimated to obtain the predicted soil organic matter.

Embodiment 2: This embodiment is different from Embodiment 1 in that the soil sample of 0.5-20 cm of the cultivated land surface is collected in Step 1. Others are the same as the first embodiment.

Embodiment 3: The difference between this embodiment and Embodiment 1 or 2 is that in step 2, an ASD surface feature spectrometer is used for spectral testing, and the spectral range is 350-2500nm. Others are the same as those in Embodiment 1 or 2.

Embodiment 4: The difference between this embodiment and Embodiments 1 to 3 is that the random forest model is constructed in step 5 using the RandomForest package. When the number of trees is set to 500, the optimal split node is set to 1/3 of the predictor variable At one time, the out-of-bag error of the model is stable. Others are the same as one of the specific embodiments 1 to 3.

Specific embodiment five: the difference between this embodiment and one of the specific embodiments one to four is that after the step five SOM prediction model is constructed, the SOM prediction accuracy is evaluated using the coefficient of determination and the root mean square error, and the RMSE calculation formula is as follows:

where y _p is the predicted value of SOC, y _o is the measured value of SOC, i represents the number, and m is the number of soil samples. Others are the same as one of the specific embodiments 1 to 4.

Adopt following experiment verification effect of the present invention:

experiment one:

The soil organic matter prediction method based on the spectral shape feature is characterized in that the soil organic matter prediction method based on the spectral shape feature is as follows:

1. Soil sample collection and processing:

Collect surface soil samples of cultivated land in Northeast China (Sanjiang Plain, Heilongjiang Province and Nong’an County, Changchun City, Jilin Province collect surface soil samples of cultivated land), air-dry, grind and sieve the soil samples, use potassium dichromate heating method, pass The conversion coefficient of 1.724 obtains SOM content;

2. Spectral test of soil samples:

Use the ASD ground object spectrometer to carry out spectral testing in the dark room. The spectral range is 350-2500nm. Put the soil sample into the container to scrape the surface. Use a 50W halogen lamp as the light source to irradiate the surface of the soil sample. Use the probe of the ASD ground object spectrometer Collect the soil spectrum, collect 10 spectral curves each time, and calculate the average value as the spectral reflectance curve of a soil sample point;

3. Soil spectral processing:

Nine-point smoothing, 10nm resampling, and envelope removal are performed on the spectral reflectance curve, and the absorption positions of the two absorption valleys in the visible light are determined through the envelope removal curve. Based on the two absorption positions, the spectral reflectance curve is extracted Spectral characteristic parameters length L and area A;

The formula for the nine-point smoothing is as follows:

R′ _i ＝0.04R _i-4 +0.08R _i-3 +0.12R i-2 +0.16R _{i-1 +} 0.2R _i +0.16R _i+1 +0.12R _i+2 +0.08R _i+3 + 0.04 _R i ₊₄

_R'i represents the reflectance of a certain wavelength after smoothing, R represents the measured reflectance, and i represents the wavelength;

The formula of the characteristic parameter length L is:

n=180°-arctan(θ)

In the length L formula, λ represents the wavelength, λ _a represents the wavelength a, R represents the reflectivity, β represents the radius, and π=3.14;

The characteristic parameter area A formula is:

A _c ＝A _abc -A _▲abc

A _▲abc = [(P ₂ -P ₁ )×(R _c -R _a )]×0.5

Among them, λ represents the wavelength, λ _a represents the wavelength a, d represents the resampling interval, R represents the reflectivity, P represents the absorption position, n and α represent the included angle (see Figure 2 of the specification), β represents the radius, π=3.14 ;

4. Clustering of soil samples: use k-mean clustering to cluster surface soil samples of cultivated land in Northeast China (soil samples from Sanjiang Plain and Nong’an County) respectively, and determine the optimal number of clusters;

5. Construction of SOM prediction model:

The prediction model uses the random forest model, and the spectral characteristic parameter length L and area A are used as predictor variables. The random forest model is constructed in the software R, and the SOM content is estimated to obtain the predicted soil organic matter.

In step 1, soil samples of 0.5-20 cm of the cultivated land surface are collected.

In the second step, the ASD surface object spectrometer is used for spectral testing, and the spectral range is 350-2500nm.

In step five, the random forest model is constructed using the RandomForest package. When the number of trees is set to 500 and the optimal split node is set to one-third of the predictor variable, the out-of-bag error of the model is stable.

Step 5 After the SOM prediction model is constructed, the SOM prediction accuracy is evaluated using the coefficient of determination (R ² ) and the root mean square error (RMSE). The RMSE calculation formula is as follows:

where y _p is the predicted value of SOC, y _o is the measured value of SOC, i represents the number, and m is the number of soil samples.

From Figures 1 to 6, it can be seen that for the soil data of the Sanjiang Plain in Heilongjiang, the verification accuracy of the model is R ² =0.69; RMSE=3.76g kg ^-1 ; for Nong'an County, Jilin Province, the verification accuracy of the model is R ² =0.76; RMSE = 7.43 gkg ^-1 . The model has high prediction accuracy, and the spectral characteristic parameters based on shape features have great potential in SOM prediction.

Claims (5)

1. The soil organic matter prediction method based on the spectral shape characteristics is characterized by comprising the following steps of:

1. collecting and processing a soil sample:

collecting a soil sample on the surface layer of the northeast China cropland in the northeast China, carrying out air drying, grinding and sieving treatment on the soil sample, and obtaining the SOM content by using a potassium dichromate heating method and a conversion coefficient of 1.724;

2. and (3) soil sample spectrum test:

performing spectrum test in a darkroom by using an ASD (automatic sampling device) surface feature spectrometer, wherein the spectrum range is 350-2500nm, placing a soil sample into a vessel to scrape the surface, adopting a 50W halogen lamp as a light source to irradiate the surface of the soil sample, collecting soil spectra by using a probe of the ASD surface feature spectrometer, collecting 10 spectrum curves each time, and calculating the average value as the spectrum reflectivity curve of a soil sample point;

3. soil spectrum treatment:

carrying out nine-point smoothing, 10nm resampling and envelope removal processing on the spectral reflectance curve, determining absorption positions of two absorption valleys in visible light through the envelope removal curve, and extracting spectral characteristic parameter length L and area A on the spectral reflectance curve based on the two absorption positions;

the formula of the nine-point smoothing is as follows:

R′ _i ＝0.04R _i-4 +0.08R _i-3 +0.12R _i-2 +0.16R _i-1 +0.2R _i +0.16R _i+1 +0.12R _i+2 +0.08R _i+3 +0.04R _i+4

R′ _i expressing the reflectivity of a certain wavelength after smoothing, R expressing the actually measured reflectivity, and i expressing the wavelength;

the characteristic parameter length L formula is as follows:

n＝180°-arctan(θ)

length L formula where λ represents wavelength, λ _a Denotes wavelength a, R denotes reflectance, β denotes radius, pi =3.14;

the characteristic parameter area A formula is as follows:

A _c ＝A _abc -A _▲abc

A _▲abc ＝[(P ₂ -P ₁ )×(R _c -R _a )]×0.5

wherein λ represents the wavelength, λ _a Denotes wavelength a, d denotes resampling interval, R denotes reflectance, P denotes absorption position, n and α denote included angle, β denotes radius, and pi =3.14, respectively;

4. soil sample clustering: respectively clustering surface soil samples of northeast China cropland by using k-mean clustering, and determining an optimal clustering number;

5. constructing an SOM prediction model:

the prediction model uses a random forest model, the spectral characteristic parameter length L and the area A are used as prediction variables, the random forest model is built in software R, and the SOM content is estimated, so that the predicted soil organic matter is obtained.

2. The method for predicting soil organic matter based on spectral shape characteristics according to claim 1, wherein a soil sample of 0.5-20cm on the surface of the cultivated land is collected in the first step.

3. The method for predicting soil organic matter based on spectral shape characteristics according to claim 1, wherein in the second step, an ASD surface texture spectrometer is used for performing spectral test, and the spectral range is 350-2500nm.

4. The soil organic matter prediction method based on spectral shape characteristics according to claim 1, characterized in that a random forest model is constructed in the fifth step by using a random forest bag, and when the number of trees is set to be 500 and the optimal splitting node is set to be one third of a prediction variable, the out-of-bag error of the model is stable.

5. The method for predicting soil organic matter based on spectral shape characteristics according to claim 1, wherein after the five-step SOM prediction model is constructed, the SOM prediction accuracy is estimated by using a decision coefficient and a root mean square error, and an RMSE calculation formula is as follows:

wherein y is _p Is the SOC prediction value, y _o Is measured SOC value, i represents quantity, and m is soil sample number.

Images