Method for estimating total nitrogen of soil in different plough layers based on field in-situ spectrum

Abstract

The invention discloses a method for estimating total nitrogen of soil of different plough layers based on a field in-situ spectrum, and relates to the technical field of crop growth monitoring. Collecting soil samples of different plough layers in a field, and obtaining in-situ spectral data of soil; performing indoor analysis on the collected soil sample, and determining the total nitrogen content, the soil texture type and the porosity; classifying the soil samples according to the soil texture types, and respectively establishing correlation models between the total nitrogen contents of the soil with different textures and the in-situ spectral data; introducing a northern eagle optimization algorithm to improve a random forest and a generalized regression neural network to improve the prediction capability of the correlation model on the total nitrogen content of the deep soil; and constructing a comprehensive monitoring model by combining a surface spectrum inversion result and a soil vertical variation equation, so as to realize indirect high-precision estimation of the total nitrogen content of the soil in different plough layers. And a theoretical basis is provided for future soil total nitrogen deep monitoring model research, and a technical support and a theoretical decision are provided for rapid acquisition of deep soil nutrients in precision agriculture.

Description

Technical Field

[0001] The present invention relates to the technical field of crop growth monitoring, and more particularly to a method for estimating total nitrogen in soil of different arable layers based on field in-situ spectroscopy. Background Art

[0002] The traditional method for determining total nitrogen (TN) in soil relies on laboratory chemical analysis, which is time-consuming, costly, and difficult to achieve large-scale real-time monitoring. Currently, spectral technology provides the possibility for rapid monitoring, but is limited by the penetration of light and interference from the field environment (such as soil roughness, moisture, light conditions, etc.). The existing spectral technology has low monitoring accuracy and poor universality for total nitrogen content in deep soil (such as deep plow layer 40-60cm). At present, the traditional method for determining total nitrogen in soil is still the laboratory chemical analysis method, which is time-consuming and labor-intensive. The existing spectral technology estimation often has low accuracy in deep TN monitoring based directly on spectroscopy because the spectral signal is difficult to penetrate deep soil (R 2 <0.6), and field in-situ spectra are easily affected by factors such as soil moisture, texture, and light, resulting in high data noise and disordered correlations. Existing models (such as PLSR and SVR) do not consider soil texture classification and vertical variation, making them difficult to adapt to the differences in TN distribution under different soil types. Therefore, a more effective method is needed to quickly and non-destructively measure soil total nitrogen content. Summary of the Invention

[0003] In view of this, the present invention provides a method for estimating total nitrogen in soil of different cultivated layers based on field in situ spectroscopy, compares and analyzes the differences in pretreatment methods of different spectra, screens the characteristic bands of soil total nitrogen content, and determines the optimal soil total nitrogen deep monitoring model based on the optimized machine learning algorithm. Finally, total nitrogen monitoring models of soil of different cultivated layers under different soil textures are constructed by two methods, which provide a theoretical basis for future research on soil total nitrogen deep monitoring models and provide technical support and theoretical decision-making for precision agriculture to quickly obtain deep soil nutrients.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions:

[0005] A method for estimating total nitrogen in soil of different cultivated layers based on field in-situ spectroscopy comprises the following steps:

[0006] Collect soil samples from different tillage layers in the field to obtain in-situ spectral data of the soil;

[0007] The collected soil samples were analyzed indoors to determine the total nitrogen content, soil texture type, and porosity;

[0008] Soil samples were classified according to soil texture type, and correlation models between total nitrogen content and in situ spectral data of soils with different textures were established.

[0009] The Northern Goshawk optimization algorithm was introduced to improve the random forest and generalized regression neural network to enhance the prediction ability of the correlation model for deep soil total nitrogen content;

[0010] Combining the surface spectral inversion results with the soil vertical variation equation, a comprehensive monitoring model was constructed to achieve indirect and high-precision estimation of the total nitrogen content in soils of different cultivated layers.

[0011] Optionally, a combined preprocessing method is used to process the in situ spectral data, including convolution smoothing, derivative transformation, and normalization, and a continuous projection algorithm is used to screen sensitive bands to reduce environmental interference.

[0012] Optionally, the model is verified by cross-validation. The accuracy evaluation index uses the coefficient of determination and root mean square error to evaluate and compare the model. 2 The value is close to 1, the lower the RMSE value is, the higher the model accuracy is. The formula is described as follows:

[0013]

[0014] Where y i represents the measured total nitrogen content of cotton field soil sample i, represents the monitored total nitrogen content of cotton field soil sample i, represents the average total nitrogen content of cotton field soil samples, n is the number of cotton field soil samples, and SD represents the standard deviation of the monitored total nitrogen content and the measured total nitrogen content.

[0015] Optionally, the integrated monitoring model can be optimized using the NGO algorithm. This model includes both surface TN prediction and deep TN estimation. Because the relationship between surface spectrum and TN is susceptible to environmental interference (such as humidity and light), the existing model can be improved and optimized using the NGO algorithm. While finding the optimal model parameters, the NGO algorithm can also transform the original linear model into a nonlinear one.

[0016] By comparison, it was found that the optimal soil surface total nitrogen content monitoring model based on field in situ spectroscopy is a model constructed based on the NGO-GRNN regression algorithm. Combining it with the fitting equation of the total nitrogen content changes at different depths, a complete comprehensive monitoring model was established to monitor the total nitrogen content at different depths of the soil.

[0017] Optionally, use the SR3500 full-spectrum portable ground feature spectrometer for in-situ spectral data acquisition. During the whiteboard calibration and spectrum measurement process, the fiber optic probe is 15 cm away from the soil sample and perpendicular to the sample.

[0018] Optionally, after the soil samples are air-dried, they are ground and sieved using a 1 mm sieve, mixed thoroughly, and then divided into two parts using a quartering method, respectively, for indoor spectral collection and chemical analysis.

[0019] Optionally, the porosity calculation includes: total soil porosity (%) = 120 (1-bulk density / specific gravity),

[0020] And soil capillary porosity:

[0021] Take a magnetic disk, place a petri dish upside down in the disk, place a piece of filter paper on the petri dish, and place the ring knife and the soil column on it;

[0022] Add water to the disk and make the edge of the filter paper touch the water surface, covering the Petri dish;

[0023] Allow the soil column to absorb water through the filter paper until all the soil capillaries are filled with water;

[0024] Take out the ring knife and cut off the wet soil that has expanded due to water absorption and exceeds the ring knife with a knife. Weigh it together with the wet soil column. Then subtract the weight of the ring knife to get the weight of the wet soil filled with capillary water.

[0025] Take 10-20g of soil sample from the top of the ring cutter, place it in an aluminum box and burn it, measure its moisture content, and calculate the dry soil weight in the ring cutter;

[0026] Calculation: Capillary porosity (%) = (capillary water volume / soil volume) 120;

[0027] Capillary water volume = weight of wet soil filled with capillary water - dry weight of the same volume of soil.

[0028] Optionally, the total nitrogen content of soil samples can be determined by crushing and turning the collected soil samples to remove foreign matter, spreading them into a thin layer of 2cm to 3cm, placing them in a dry and ventilated environment to air-dry, sieving them to 1mm, conducting chemical experiments on the soil samples, and using the Kjeldahl total nitrogen method to determine nitrogen treatment.

[0029] It can be seen from the above technical solution that compared with the existing technology, the present invention provides a method for estimating total nitrogen in soil of different cultivated layers based on in-situ field spectra. It uses a machine learning optimization algorithm combined with surface spectral inversion results and vertical variation equations to achieve indirect and high-precision estimation of deep TN, thereby achieving high-precision and rapid estimation of total nitrogen content in soil of different cultivated layers (especially deep layers). Based on the relationship between in-situ field spectra and total nitrogen in deep soil, it provides a pilot study for exploring the monitoring of TN content deep in soil based on surface spectroscopy. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0031] Figure 1 Schematic diagram of the method principle of the present invention;

[0032] Figure 2 This is the distribution map of field sampling points;

[0033] Figure 3 This is the vertical distribution characteristic diagram of the total content of different soil textures;

[0034] Figure 4 This is the vertical variation trend diagram of porosity of different soil types in the study area;

[0035] Figure 5 This is the correlation diagram between porosity of different soil textures and soil total nitrogen content in the study area;

[0036] Figure 6 This is the nonlinear fitting diagram of soil depth and soil total nitrogen content in the study area;

[0037] Figure 7a This is the indoor spectrum of cotton field soil samples;

[0038] Figure 7b This is the correlation analysis chart between indoor spectra of field soil samples and nitrogen content;

[0039] Figure 8 Indoor spectrum diagrams of various indoor spectrum preprocessing methods;

[0040] Figure 9 This is the correlation analysis diagram between the indoor spectrum of each indoor spectrum pretreatment method and the total nitrogen in the soil;

[0041] Figure 10 This is the indoor spectral characteristic band screening diagram of cotton field soil samples;

[0042] Figure 11 Schematic diagram of VIF value of characteristic band;

[0043] Figure 12 Histograms monitored for the four regression models and three pretreatment combinations;

[0044] Figure 13a The in-situ spectra of 120 cotton field soil samples and the corresponding indoor spectra;

[0045] Figure 13bThe in-situ spectra and difference spectra of 120 cotton field soil samples are shown in the figure.

[0046] Figure 14 This is the correlation diagram between the total nitrogen content of each tillage layer in the cotton field and each pretreatment of the vis-NIR field spectrum;

[0047] Figure 15 This is a screening diagram of the characteristic bands of the in-situ spectrum of cotton field soil samples;

[0048] Figure 16 This is the accuracy test chart of the cotton field total nitrogen content monitoring model based on sensitive bands;

[0049] Figure 17 This is the accuracy test chart of the total nitrogen content monitoring model in cotton fields optimized based on the NGO algorithm;

[0050] Figure 18 This is a test chart of the accuracy of monitoring total nitrogen content in each tillage layer of soil using NGO-GNRR and the fitting equation. DETAILED DESCRIPTION

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0052] The embodiment of the present invention discloses a method for estimating total nitrogen in soil of different cultivated layers based on field in-situ spectroscopy. Figure 1 As shown, the following steps are included:

[0053] Collect soil samples from different tillage layers in the field to obtain in-situ spectral data of the soil;

[0054] The collected soil samples were analyzed indoors to determine the total nitrogen content, soil texture type, and porosity;

[0055] Soil samples were classified according to soil texture type, and correlation models between total nitrogen content and in situ spectral data of soils with different textures were established.

[0056] The Northern Goshawk optimization algorithm was introduced to improve the random forest and generalized regression neural network to enhance the prediction ability of the correlation model for deep soil total nitrogen content;

[0057] Combining the surface spectral inversion results with the soil vertical variation equation, a comprehensive monitoring model was constructed to achieve indirect and high-precision estimation of the total nitrogen content in soils of different cultivated layers.

[0058] Among them, the Northern Goshawk Optimization Algorithm (NGO) is a meta-heuristic optimization algorithm that simulates the hunting behavior of the goshawk (search, tracking, attack), balances global search and local convergence, can optimize the key parameters and input features of the model, and improve the prediction ability of deep soil TN.

[0059] NGO optimizes the parameters and feature selection of RF and GRNN models, enhancing the model's ability to predict total nitrogen (TN) in deep soils. Because deep soil spectral signals are weak and noisy, traditional models are prone to overfitting or underfitting. NGO uses a global search to find the optimal parameter combination, improving model robustness.

[0060] Key RF parameters include the number of decision trees (n_estimators), maximum depth (max_depth), and feature subset size (max_features). NGO simulates the search behavior of a goshawk, traversing parameter combinations to find parameters that minimize the model's prediction error. Similarly, a key GRNN parameter is the smoothing parameter (spread), which controls the width of the kernel function and affects the model's sensitivity to noise. The NGO algorithm can dynamically adjust the spread to balance the model's fitting and generalization capabilities in complex data.

[0061] The soil sample data used in the embodiments of the present invention are soil samples collected from cotton fields before sowing in April 2022 and 2023. One cotton field was selected in each of Areas A, B, C, D, and E, for a total of six cotton fields as the study areas.

[0062] In the bare soil state of the experimental area, 4×5 sampling points were evenly distributed using the grid distribution method, and a total of 120 sampling points were arranged in the six experimental areas. In order to reduce the impact of moisture on the experiment, sampling was carried out before cotton sowing and irrigation, and each sampling point was careful to avoid interference such as straw white film. Each sampling point in the experimental area was accurately located and marked using GPS, such as Figure 2 As shown (field sampling point distribution map).

[0063] Soil spectral data were measured using the SR 3500 full-spectrum portable ground feature spectrometer produced by Spectral Evolution, USA. The instrument has a measurement range of 350nm-2500nm, a spectral sampling interval of 3.5nm in the 350-1000nm band and 10nm in the 1000-1900nm band; and a spectral resolution of 7nm in the 1900-2500nm range.

[0064] Before collecting in situ spectra, the vertical surface of the soil was carefully leveled, taking care to avoid any debris that might be present in the soil sample, such as stones, crop roots, and macropores. Within each study plot, three measurements were taken at each sampling point during in situ spectra collection, and the average was taken, resulting in a total of 60 spectral data points for modeling.

[0065] The SR 3500 full-spectrum portable ground feature spectrometer was used for spectrum acquisition. During the whiteboard calibration and spectrum measurement process, the fiber optic probe was 15 cm away from the soil sample and perpendicular to the sample.

[0066] After measuring the in situ spectra of the soil, soil profiles (60-120 cm long, 50 cm wide, and 60-70 cm high) were excavated at each sampling point. After the profiles were excavated, their vertical surfaces (observation and sampling surfaces) were cleaned and leveled. Horizontal vertical soil cubes were divided into 3 cm thick sections, and soil data was collected in layers from 0 to 60 cm. The 0-3 cm layer was defined as the surface layer, the 3-20 cm layer as the shallow tillage layer, the 20-40 cm layer as the medium tillage layer, and the 40-60 cm layer as the deep tillage layer.

[0067] When collecting soil samples, a thin shovel was inserted perpendicular to the sampling surface from top to bottom at a scale of 3 cm from the excavated soil profile. The depth and number of samples at each sampling point were kept consistent. The soil column was evenly divided into 20 layers using the 3 cm standard. About 500 g of soil was collected from each portion using the quartering method. The excess soil was discarded, mixed evenly, and placed in a ziplock bag for return. The soil samples were naturally dried, cleaned of impurities, ground with a grinding rod, and passed through a 1 mm sieve to prepare the test samples for later use. A total of 2400 soil samples (20 × 20 × 6) were collected in this experiment, including 6 study areas. Each study area was grid-sampled at 20 points, and each point was vertically divided into 20 layers at a scale of 3 cm. 360 spectra (20 × 3 × 6) were collected, including 6 study areas. Each study area was grid-sampled at 20 points, and each point was vertically divided into 3 layers.

[0068] After air-drying, the soil samples were ground and sieved through a 1mm mesh. After thorough mixing, they were divided into two parts using the quartering method for indoor spectral acquisition and chemical analysis. The soil samples were placed in a 15cm diameter, 2cm deep aluminum box container with a black interior. After filling it with soil samples, the surface was scraped flat. An SR-3500 (Spectral Evolution, USA) was used to acquire indoor spectral data. During spectral data acquisition, the fiber optic probe was 15cm away from the soil sample, and the fiber optic field of view was filled with soil samples. The fiber optic probe was fixed after ensuring vertical alignment with the object being measured. For each soil sample, three locations were randomly selected for measurement. 20 spectra were measured at each measurement point, and the 60 spectra were averaged to form the indoor spectrum of the soil sampling point.

[0069] Soil texture is determined using the hydrometer method. First, soil particles are classified into several grades based on their size. The soil particulate matter content is then determined for each grade, and the soil texture is classified according to the soil texture classification standard. Sand particles larger than 0.25 mm are separated by sieving, while finer soil particles are separated by sedimentation using a suspension preparation method to ensure accurate classification.

[0070] Soil porosity determination:

[0071] (1) Total soil porosity (%) = 120 (1-bulk density / specific gravity)

[0072] (2) Soil capillary porosity:

[0073] 1) Take a magnetic disk, place a culture dish upside down on the disk, and place a piece of filter paper on the culture dish, which is slightly larger than the culture dish. Place the ring knife and the soil column on it.

[0074] 2) Add water to the disk and make the edge of the filter paper touch the water surface and cover the Petri dish.

[0075] 3) Allow the soil column to absorb water through the filter paper until all the soil capillaries are filled with water.

[0076] 4) Take out the ring knife and cut off the wet soil that has expanded due to water absorption and exceeds the ring knife with a knife. Weigh it together with the wet soil column. Then subtract the weight of the ring knife to get the weight of the wet soil filled with capillary water.

[0077] 5) Take out 10-20g of soil sample from the top of the ring cutter, place it in an aluminum box and burn it, measure its moisture content, and calculate the dry soil weight in the ring cutter.

[0078] 6) Calculation

[0079] Capillary porosity (%) = (capillary water volume / soil volume) × 120

[0080] Capillary water volume = weight of wet soil filled with capillary water - dry weight of the same volume of soil

[0081] The collected soil samples were crushed and turned to remove foreign matter such as gravel, sand, remnants of shading film, and crop debris. The samples were spread into a thin layer of 2-3 cm and placed in a dry and ventilated environment to air-dry. The samples were then sieved to 1 mm and subjected to chemical testing. Nitrogen treatment was performed using the Kjeldahl method.

[0082] According to the Technical Regulations of the Second National Soil Survey, we divided the soil total nitrogen content in the study area before cotton field sowing into the following levels from low to high, as shown in Table 1: high total nitrogen content (>0.9); relatively high total nitrogen content (0.75%-0.9%); medium total nitrogen content (0.55%-0.75%); and low total nitrogen content (<0.55%).

[0083] Table 1 Classification of total nitrogen in cotton field surface soil

[0084]

[0085] The preprocessing methods used in this experiment are: multi-scatter correction (MSC), standard normal variation (SNV),

[0086] Savitzky-Golay (AVITZKY-GOLA convolution), continuous spectrum removal (CR) and derivative change (first-order derivative FD, second-order derivative SD), and the best preprocessing method was selected through comparative analysis.

[0087] Partial least squares regression, random forest, GRNN generalized regression neural network, support vector machine regression and Northern Goshawk optimization algorithm (NGO) are used for modeling. Several common machine learning algorithms will not be described in detail. The Northern Goshawk optimization algorithm is introduced as follows.

[0088] The Northern Goshawk Optimization (NGO) algorithm was proposed by Mohammad Dehghani et al. in 2022. This algorithm simulates the behavior of the Northern Goshawk during hunting, including prey recognition, attack, pursuit, and escape. The formula is briefly described as follows:

[0089] First, initialize the data matrix:

[0090]

[0091] Then comes prey identification (reconnaissance phase):

[0092] P i =X k i,k∈[1,N]#(2-32)

[0093] P i is the position of the i-th northern goshawk’s prey; is the objective function value of the location of the i-th northern goshawk's prey; k is a random integer in the range [1, N]; is the new position of the i-th northern goshawk; is the new position of the i-th northern goshawk; is the new position of the i-th northern goshawk in the j-th dimension; is the objective function value of the i-th northern goshawk after the update based on the first stage; r is a random number in the range of [0,1]; I is a random integer of 1 or 2.

[0094] Finally, the chase and escape (development stage):

[0095]

[0096] Where: t is the current number of iterations; T is the maximum number of iterations; is the new position of the i-th northern goshawk; is the new position of the i-th northern goshawk in the j-th dimension; is the j-th dimension position of the i-th pelican after the update based on the second stage; is the objective function value of the i-th northern goshawk after updating based on the second stage.

[0097] This paper uses cross-validation to verify the model. The accuracy evaluation indicators use the determination coefficient (R2) and root mean square error (RMSE) to evaluate and compare the model. The R2 value is close to 1. The lower the RMSE value, the higher the model accuracy. The formula is described as follows:

[0098]

[0099] Where y i represents the measured total nitrogen content of cotton field soil sample i, represents the monitored total nitrogen content of cotton field soil sample i, represents the average total nitrogen content of cotton field soil samples, n is the number of cotton field soil samples, and SD represents the standard deviation of the monitored total nitrogen content and the measured total nitrogen content.

[0100] In this invention, the evaluation grades given by various scholars are comprehensively adopted. 2 Divided into four categories: Best (R 2 ≥0.90), good (R 2 ∈[0.81,0.90]), qualified(R 2 ∈[0.60,0.80]), unqualified (R 2 <0.60).

[0101] Soil nutrients are influenced by factors such as soil texture and vegetation type, and have distinct spatial and vertical distribution patterns. Accurately monitoring total nitrogen content in cotton fields requires understanding the soil texture types and spatial variation trends of total nitrogen in different tillage layers before sowing. This allows for the development of a more accurate soil total nitrogen content monitoring model.

[0102] As shown in Table 2-3, the overall coefficient of variation (CV) for the five experimental plots in the study area ranged from 24.1% to 42.3%, all reaching a moderate level of variation. Soil total nitrogen content was mostly high, accounting for 53% of the total study area and fairly evenly distributed across the six experimental plots. High and moderate levels followed, accounting for 19% and 26% of the total sampling points in the study area, respectively. High levels were concentrated in 147C (sandy clay loam), while moderate levels were more concentrated in 2 Lian and 150A (loamy sand and sandy loam). Furthermore, the CVs within the same soil showed stability and consistency, with sandy loam > loamy sand > sandy clay loam. The CVs for loamy sand ranged from 31.56% to 35.33%, for sandy clay loam from 24.11% to 27.58%, and for sandy loam from 38.45% to 42.32%.

[0103] Table 2 Classification of nitrogen in cotton field surface soil

[0104]

[0105] Table 3 Statistics of total nitrogen content in soil surface layer at different levels

[0106]

[0107] Figure 3 The selected data were averaged at the same depth for each study plot to plot soil TN variation with depth. Overall, TN content decreased with increasing soil depth, forming an S-shaped downward curve. Within the shallow tillage layer (0-20 cm), TN content in the 150C sandy loam soil decreased rapidly. Under the 2nd and 147C loamy sandy soils, TN content remained relatively stable, with a slight decrease. Under the 147A and 150A sandy clay loam soils, TN content increased slightly. Within the 20-40 cm tillage layer, TN content in all plots decreased rapidly.

[0108] Overall, there was no significant difference between the surface layer and the shallow tillage layer, but there was a highly significant difference between the various tillage layers in the cotton field. In loam and clay soils, there was no significant difference between the total nitrogen in the surface layer and the shallow tillage layer, while in sandy clay loam, there was a highly significant difference between the total nitrogen in the surface layer and the shallow tillage layer. With the exception of 147A, which showed no significant difference, there were highly significant differences in total nitrogen content between the shallow tillage layer and the middle tillage layer in the cotton field. With the exception of the second case, there were significant differences in total nitrogen content between the middle tillage layer and the deep tillage layer in the cotton field.

[0109] Depend on Figure 4 It can be seen that the porosity of soils below the soil surface shows a decreasing trend with increasing vertical depth across all soil textures, but the decreasing trends vary slightly across soil textures. The total nitrogen content in the tillage layer of sandy clay loam soils decreases at a slower rate, while the opposite is true for sandy loam soils. The porosity of loamy sand soils shows a steady downward trend with increasing vertical depth. Furthermore, the vertical variation trend of soil porosity within the same soil type appears stable.

[0110] like Figure 5 As shown in the figure, the soil porosity has a relatively stable change trend in the vertical depth of the soil with the same soil texture. Therefore, this paper established a scatter plot of soil total nitrogen content and soil porosity at 0-60 cm with different soil textures. The results showed that there is a good correlation between soil total nitrogen content and soil porosity. Figure 3-4 As shown in the data, total nitrogen content and soil porosity of different soil textures showed a strong negative correlation in vertical depth, and the correlation coefficient of loamy sand soil (0.863) > sandy loam soil (0.8443) > sandy clay loam soil (0.7846).

[0111] There is a significant correlation between soil total nitrogen content and soil depth, and the total nitrogen content is similar under the same soil texture, and shows a strong correlation with soil porosity. Therefore, the present invention establishes nonlinear fitting models of soil total nitrogen content and each arable layer according to different soil texture types, in preparation for establishing a comprehensive monitoring model of soil total nitrogen in different arable layers. The research results show that the R2 of the fitting models under various soil types are all greater than 0.7, showing a strong correlation. Figure 6 As shown in the figure, the fitting models of sandy loam, loamy sand and sandy clay loam are

[0112] y=0.8609+0.0116x-7.7846×10 -6 x 2 ,y=0.0149x+1.0242×10 -4 x 2

[0113] and y=-0.08461-0.00314x-1.6331×10 -4 x 2 .

[0114] The following conclusions were drawn:

[0115] (1) The spatial distribution of soil total nitrogen content under different soil texture types has a strong variability, but the distribution of soil total nitrogen content under the same soil texture is relatively uniform; in terms of vertical depth distribution, the soil total nitrogen content shows a trend of decreasing with increasing soil depth, and the decreasing trend under different soil textures is different, so it is necessary to conduct classification research according to soil texture type.

[0116] (2) Under different soil texture types, the total nitrogen content in the soil surface layer is strongly correlated with the total nitrogen content in the tillage layer at different depths, providing a theoretical basis for using the total nitrogen content in the surface soil to monitor the total nitrogen content in the deep soil. Secondly, the correlation analysis results between soil porosity and soil total nitrogen content at vertical depth show that soil porosity and soil total nitrogen content show a strong correlation.

[0117] First, using indoor and outdoor spectra collected from 120 cotton field surfaces in 2022, we analyzed the factors influencing differences between indoor and outdoor spectra. Second, we used eight different preprocessing and modeling combinations to identify characteristic bands and establish a soil total nitrogen monitoring model. This model demonstrated excellent accuracy and stability for monitoring total nitrogen content in all soil layers. 75% of the samples served as the modeling set, and total nitrogen content was inverted using sensitive bands extracted using different preprocessing methods. A further 25% of the samples served as the validation set to verify the model's accuracy.

[0118] Indoor spectra of cotton field soil samples before tillage and their correlation with corresponding nitrogen content Figure 7a and Figure 7b As shown. Figure 7a It can be seen that the higher the nitrogen content, the lower the reflectivity of the corresponding indoor spectral band. The performance becomes more obvious after the 950nm band. Figure 7b It was found that the indoor spectrum was negatively correlated with the corresponding nitrogen content. The absolute values of the correlation coefficients were all above 0.5 near 1550nm-1900nm, 2000nm-2200nm, and 2300nm.

[0119] Figure 8 Spectra obtained from processing the same dataset using eight preprocessing methods. Direct observation of the spectral signal characteristics reveals that maximum normalization yields better modeling results than average and peak normalization (TN) methods, compared to other preprocessing methods. Analysis of different variations in spectral reflectance reveals that first-order derivative transformations offer better overall results, while second-order derivatives increase noise and reduce model quality. The optimal preprocessing methods for each step of indoor spectral preprocessing of the soil dataset are SG convolution smoothing, first-order derivatives, and multivariate scattering correction. Scaling alone does not allow for intuitive differentiation between the methods.

[0120] Figure 9 The indoor spectra of cotton fields treated with 8 pretreatment methods and their correlation with the total nitrogen content of each tillage layer in the cotton fields were analyzed. Direct observation of the characteristics of the spectral signals showed that the spectrum after processing with the first-order derivative and the second-order derivative significantly eliminated the interference of the baseline and background; it can be seen that after the spectrum was processed by SNV and MSC respectively, the fitting degree of the spectrum became significantly higher, and the soil total nitrogen content was sensitive.

[0121] The fitting effect of band position b is better than that of band position e, which reduces the influence of scattering on the original spectrum; the spectral curve is smoother after SG convolution smoothing, which reduces the noise interference of the original spectrum; the indoor soil spectra after scaling, normalization and standardization are limited to a specific size range, eliminating the influence of different weights caused by size differences on the monitoring results.

[0122] In order to solve the problem of full spectrum collinearity, the present invention uses the successive projection algorithm (SPA) to iteratively search forward, starting from one wavelength and adding a new variable in each iteration, and finally obtains 30 characteristic bands, which pass the collinearity test, such as Figure 10 and Figure 11 The 30 sensitive indoor spectrum bands are: 627nm, 954nm, 971nm, 1042nm, 1129nm, 1133nm, 1383nm, 1451nm, 1687nm, 1693nm, 1806nm, 1856nm, 1873nm, 1883nm, 1906nm, 1919nm, 1923nm, 1930nm, 1952nm, 2053nm, 2225nm, 2237nm, 2283nm, 2306nm, 2327nm, 2341nm, 2384nm, 2451nm, 2473nm, and 2478nm. A monitoring model for total nitrogen content in indoor soil was constructed based on the selected bands.

[0123] The specific preprocessing steps are shown in Table 4. In this invention, four for loops are performed in the order of baseline removal, smoothing, scaling, and normalization for no preprocessing and eight preprocessing methods, and modeling and evaluation are performed to find the optimal combination of preprocessing methods. These four groups of preprocessing methods include:

[0124] ① Unprocessed, the spectral dataset is not preprocessed;

[0125] ② Combination 1: The spectral dataset is preprocessed by SG-FD-SNV-MinMax scaling in sequence;

[0126] ③ Combination 2: The spectral dataset was preprocessed with SG-FD-SNV-Z-score standardization in sequence;

[0127] ④ Combination 3: The spectral dataset is preprocessed with SG-FD-SNV-CR normalization in sequence.

[0128] Table 4 Preprocessing methods included in the four preprocessing steps

[0129]

[0130] Based on different combination methods, the sensitive bands and total nitrogen content after screening by the continuous projection algorithm were used to establish models and their accuracy was evaluated and the models were screened. The results showed that the accuracy of the PLSR (combination 2), RFR (combination 2), GRNN (combination 3) and SVR (combination 4) models constructed based on the modeling set (optimal preprocessing combination method) was PLSR>GNRR>RFR>SVR( Figure 12 ). The training set R of the four monitoring models 2The total nitrogen monitoring models in cotton fields constructed by generalized propagation neural network (GNRR) and partial least squares (PLSR) were the most sensitive, with R 2 The R values of the four models constructed based on the best preprocessing combination were 0.89 and RMSE were 0.07 and 0.06 respectively. 2 All of them are greater than 0.60, indicating that the model has a good monitoring effect.

[0131] Conclusion:

[0132] The optimal soil total nitrogen content monitoring model based on indoor spectral sensitivity bands is RFR, followed by GNRR. The indoor environment is simpler than the outdoor environment, and indoor soil spectral interference is less, allowing for accurate monitoring of total nitrogen content in cotton fields.

[0133] Compared to laboratory spectral acquisition, in situ spectral measurement offers advantages such as speed, real-time, non-destructiveness, and pollution-free. However, this measurement is subject to interference from external factors such as soil texture, soil moisture content, particle size, and light during sampling, making field measurement conditions more demanding. The study first analyzed the differences between in situ and indoor spectra. Secondly, seven different noise reduction methods were used to analyze the correlation between indoor spectra and soil total nitrogen content, identifying the optimal noise reduction method and sensitive phase. Two methods were used to construct the model.

[0134] The first approach is to directly construct a monitoring model based on the correlation between the surface spectral characteristics and the total nitrogen content of each cultivated layer of soil. The second approach is to model soil texture types separately. The main steps of the modeling are as follows:

[0135] (1) Construct a surface soil total nitrogen content monitoring model based on the correlation between surface spectra and surface soil total nitrogen content;

[0136] (2) Construct a monitoring model based on the correlation between total nitrogen content in the soil surface layer and each cultivated layer;

[0137] (3) Combine the two models constructed in steps 1 and 2 to establish a model for monitoring total nitrogen content in topsoil based on field in situ spectroscopy. Finally, a new soil total nitrogen dataset collected in 2023 was used to verify and evaluate the two methods for monitoring total nitrogen content in topsoil, and the optimal model was selected.

[0138] like Figure 13a and Figure 13b As shown in the figure, both the indoor spectrum and the in-situ spectrum of the soil surface have obvious absorption valleys near 1350nm, 1900nm, 2050nm and 2150nm. Compared with the indoor spectrum, the absorption valley of the in-situ spectrum is wider and deeper. However, the reflectance of the indoor spectrum is significantly higher than that of the in-situ spectrum, so further preprocessing and denoising are required.

[0139] Before spectral preprocessing, in situ spectral data from 120 cotton fields' soil surfaces were removed from the spectral ranges (350-399 nm and 2450-2500 nm) to eliminate artifacts caused by the spectrometer. Savitzky-Golay smoothing was then performed. Seven spectral preprocessing methods, including FD, SD, MSC, SNV scaling, Min-Max scaling, Z-Score normalization, and CR normalization, were used to remove non-component-related influences from the spectral data. Characteristic bands were identified by correlation analysis with the TN content of each tillage layer.

[0140] Through the correlation analysis between the in-situ spectra of cotton fields after different pretreatments and the total nitrogen content of each tillage layer in the cotton fields, it can be seen that under various pretreatments of the field vis-NIR spectra, the in-situ spectra after FD baseline correction eliminated the interference of baseline and background; the spectra under SD pretreatment increased the noise, and the correlation coefficient between the in-situ spectra after SD treatment and the total nitrogen content of each tillage layer was lower than that of FD; the correlation coefficient between the sensitive bands of the in-situ spectra after SNV scattering correction and the total nitrogen of each tillage layer of the soil was higher than that after MSC treatment; among the three scale scaling, the correlation coefficient of the surface layer was improved under Min-Max scaling, and the effects of other tillage layers became worse. The correlation coefficients of the in-situ spectra after CR normalization and Z-score standardization with the total nitrogen content of each tillage layer were improved overall. The correlation coefficient between the spectral data of the total nitrogen content of the surface layer after Z-score standardization was higher than that of CR normalization. The correlation coefficient between the spectral data of the total nitrogen content of the middle tillage layer and the deep tillage layer after CR normalization was higher than that of Z-score standardization. The results of the two treatments for the shallow tillage layer were not much different ( Figure 14 ).

[0141] The results show that the spectral processing effects of each tillage layer are different through preprocessing methods. Before building the model with total nitrogen in each tillage layer, different preprocessing combinations were applied to the in-situ spectra. The surface layer was selected: four preprocessing methods of FD-SNV-Z-score-SG were used in sequence; other tillage layers were selected: four preprocessing methods of FD-SNV-CR-SG were used in sequence. Among the 30 sensitive bands with good correlation, 10 characteristic bands were selected using the continuum removal method for model construction, namely 501nm, 693nm, 1494nm, 1507nm, 1526nm, 1555nm, 1671nm, 1816nm, 1876nm, 2096nm ( Figure 15 ).

[0142] Table 5 shows that the accuracy of cotton field total nitrogen monitoring models using different modeling methods for the same tillage layer varies. The random forest regression model, which is more tolerant to outliers and noise, is less prone to overfitting, while the generalized propagation neural network can self-learn to find the maximum likelihood estimate. Across the entire soil vertical scale, the accuracy of the RFR, PLSR, GRNN, and SVR models constructed based on the modeling set follows the order of GNRR > RFR > SVR > PLSR. Across different tillage layers, the accuracy of the total nitrogen monitoring models for each cotton field layer follows the order of shallow tillage layer > medium tillage layer > deep tillage layer.

[0143] Table 5 Regression model of total nitrogen content and sensitive bands in cotton fields (n=75)

[0144]

[0145] The results showed that from the perspective of single tillage layer, the monitoring ability of the four models in the shallow tillage layer was GRNN>RFR>SVR>PLSR, and GRNN and RFR were better than other models in monitoring total nitrogen in the soil surface layer; in the middle tillage layer, the accuracy of the four regression models was GRNN>RFR>SVR>PLSR, and in the deep tillage layer, the accuracy of the four regression models was GRNN>SVR>RFR>PLSR. Different models had different monitoring abilities in different tillage layers, but the R of the GRNN model was better than that of the other models. 2 It is greater than 0.6 in all tillage layers, showing stable monitoring ability and is the best model for monitoring total nitrogen in soil in different tillage layers.

[0146] from Figure 16 It can be seen that the R values of the PLSR, RFR, GRNN and SVR cotton field total nitrogen content monitoring models constructed based on the validation set are 2 Compared with the validation set, they have decreased.

[0147] The monitoring accuracy of the model constructed by shallow cultivation and generalized propagation neural network is better than that of the models constructed by the other three regression algorithms. The model with the highest monitoring accuracy is the model constructed by RFR. 2 The model constructed by GRNN is 0.78 and the RMSE is 0.16. 2 The accuracy and stability of the other two methods are improved compared with the validation set, but GRNN and RFR are significantly better than other methods, and the validation evaluation effect is better;

[0148] In the middle cultivation layer, the models constructed by the four regression algorithms are verified by the validation set. The accuracy of the model constructed by the GRNN algorithm is relatively good. 2 The model constructed by RFR algorithm is 0.67 and RMSE is 0.14. 2 The R of PLSR and SVR methods is 0.34 and RMSE is 0.79.2 The values were 0.05 and 0.02 respectively, which were both unable to effectively monitor the total nitrogen content in the tillage layer;

[0149] The model established by the deep-cultivation GRNN regression algorithm has the best effect. 2 The R values of the three methods, PLSR, RFR, and SVR, were 0.63 and 0.10, respectively. 2 They are 0.05, 0.34 and 0.04 respectively.

[0150] As shown in Table 6, the R values of the NGO-RFR, NGO-GRNN, and NGO-SVR models after NGO optimization are 2 The results showed that NGO-GRNN>NGO-RFR>NGO-SVR in different tillage layers.

[0151] R of NGO-GRNN monitoring model for shallow, medium and deep tillage layers 2 The results are 0.84, 0.79 and 0.82 respectively, and the RMSE are all less than 0.1. Compared with the optimal GRNN monitoring model, the optimized NGRO-GRANN model has better monitoring accuracy in the shallow, medium and deep tillage layers. 2 The increases were 0.6, 0.12 and 0.19 respectively.

[0152] Table 6 Regression model of total nitrogen content and sensitive bands in cotton fields after optimization based on the NGO algorithm (n=75)

[0153]

[0154] The results of the NGO-RFR, NGO-GRNN and NGO-SVR models built based on the validation set are as follows: Figure 17 As shown in the figure, the total nitrogen content monitoring model of each tillage layer of the three soil types is R 2 The total nitrogen monitoring model constructed by the NGO-GRNN regression algorithm has the highest accuracy and the best effect. 2 is 0.75, and the RMSE is 0.04. In the shallow and medium tillage layers, the model accuracy R 2 The RMSEs are 0.02 and 0.04 respectively. 2 and RMSE were 0.77 and 0.12, respectively, reaching the level of quantitative monitoring of total nitrogen content.

[0155] After optimizing the model algorithm, the present invention has improved the accuracy of some models to a certain extent, achieving a high level of monitoring. Separate modeling and precision stability screening are performed between field in-situ spectra and total nitrogen in the soil surface layer, and between total nitrogen in the soil surface layer and the soil top layer, further improving the accuracy of total nitrogen content in each top layer of cotton field soil based on field in-situ spectra.

[0156] Based on the 10 selected characteristic bands and the total nitrogen content in the surface layer, the present invention established PLSR, RFR, GRNN and SVR models respectively. The research results showed that the overall accuracy of each model was in the order of GNRR>RFR>SVR>PLSR.

[0157] From the perspective of model training, the R 2 The total nitrogen monitoring model for cotton fields constructed based on RFR was the most sensitive and had a higher accuracy. The total nitrogen monitoring model for cotton fields constructed based on PLSR had a lower accuracy. 2 and RMSE are 0.58 and 0.16 respectively, indicating that the algorithm has poor monitoring capabilities.

[0158] From the verification results, (Table 7) the accuracy of the other three total nitrogen content monitoring models except GNRR decreased by 3.1%, 7.4%, and 11.2%, respectively. Among them, the monitoring accuracy of the GNRR regression model was better, with R 2 The model monitoring capability was relatively stable with a RMSR of 0.05. Through the horizontal comparison of the four models, it was found that the optimal model for monitoring the total nitrogen content in the cotton field soil surface based on the field in-situ spectral sensitive band was GNRR. At the same time, each model also achieved the level of quantitative monitoring of soil total nitrogen content (R 2 >0.6).

[0159] Table 7 Regression model of total nitrogen content in cotton field surface layer and sensitive band

[0160]

[0161]

[0162] The present invention used 10 characteristic bands from the 30 characteristic bands screened using the Successive Projection Algorithm (SPA) from indoor spectra. These bands, as shown in Table 8, were used to establish nonlinear regression models optimized by the Northern Goshawk algorithm (NGO-RFR, NGO-GRNN, and NGO-SVR) based on surface total nitrogen content. The results showed that the overall accuracy of the models was in the order of NGO-GNRR > NGO-SVR > NGO-RFR.

[0163] Table 8 Regression model of total nitrogen content and sensitive bands in cotton fields after optimization based on NGO algorithm

[0164]

[0165] After optimization by the NGO algorithm, the model with the highest accuracy is the one constructed by NGO-GNRR. 2 The RMSE and RFR were 0.89 and 0.03, respectively. The results show that, except for the RFR of the built-in learning machine, the monitoring accuracy of each model after optimization by the NGO algorithm has been improved. Optimizing the model with the NGO algorithm has certain positive significance in improving the monitoring accuracy of total nitrogen content in deep soil.

[0166] The present invention is based on the optimal soil surface total nitrogen content monitoring model based on field in situ spectroscopy, which is a model constructed based on the NGO-GNRR regression algorithm. It monitors the monitoring value of the total nitrogen in the soil surface, obtains the fitting equation of the total nitrogen content change at different depths, and monitors the total nitrogen content at different depths of the soil.

[0167] like Figure 18 As shown in the data, except for the deep plowing layer of loamy sand, where the monitoring accuracy is lower than 0.6, the monitoring accuracy of the model under different soil types and different plowing layers is above 0.6, indicating that the monitoring model optimized based on NGO has strong monitoring capabilities for the new dataset.

[0168] Specifically, for loamy sandy soil, the model's monitoring accuracies were 0.74, 0.62, and 0.55 (in the shallow, medium, and deep tillage layers, respectively); for sandy loam, the model's monitoring accuracies were 0.78, 0.67, and 0.74 (in the same order); and for sandy clay loam, the model's monitoring accuracies were 0.77, 0.67, and 0.66 (in the same order). After soil texture classification, the comprehensive monitoring model had the highest monitoring accuracy for sandy loam, with the following order: sandy loam > sandy clay loam > loamy sand.

[0169] The study used a new dataset to validate the best direct and integrated monitoring models, analyzing the differences in their accuracy across different soil layers. The results showed that integrated monitoring significantly improved overall accuracy compared to unclassified direct monitoring. This improvement was most evident in sandy loam soils, where the accuracy improved by 0.25 and 0.08, respectively, compared to direct monitoring in the middle and deep tillage layers. This improvement in accuracy was also evident after soil texture classification. The integrated model for total nitrogen in the middle and deep tillage layers demonstrated higher validation accuracy than the model for direct monitoring of total nitrogen content. This model is more conducive to monitoring total nitrogen in deep layers than the model for direct monitoring of total nitrogen content.

[0170] Table 8 Comparison of monitoring accuracy between direct model and comprehensive model on new dataset

[0171]

[0172] This study used different models combined with field in-situ spectroscopy to directly and comprehensively monitor the total nitrogen content of each soil layer and compared the direct and comprehensive monitoring results. Finally, field validation was performed using total nitrogen data from soil samples collected before cotton sowing in 2023.

[0173] (1) For different sample data sets, even if the spectra are the same, the optimal pretreatment for different tillage layers is different. There is no completely universal pretreatment method. The method of eliminating environmental factors is not only related to the spectrum, but also to the components that need to be monitored.

[0174] (2) The NGO algorithm can improve the accuracy of some soil total nitrogen monitoring models

[0175] Based on field in-situ spectral monitoring of total nitrogen content in various soil layers, a comparative analysis of the RFR, GNRR, and SVR algorithms concluded that the NGO algorithm significantly improved the monitoring accuracy of GNRR and SVR, while having a negative impact on RFR. The RFR and GNRR algorithms provide a rough estimate of total nitrogen content in the surface and shallow layers of soil, but are unable to monitor the middle and deep layers. However, the NGO-GNRR algorithm further improves the accuracy of total nitrogen content monitoring in various soil layers, enabling the construction of highly accurate and stable quantitative monitoring models for both the middle and deep layers.

[0176] (3) Compared with directly monitoring the total nitrogen content of each tillage layer of soil through in situ spectroscopy in the field, the comprehensive use of total nitrogen in the soil surface layer as an intermediate medium can improve the monitoring accuracy.

[0177] Comparing the two methods for monitoring total nitrogen content in various soil layers, the use of field in-situ spectroscopy to determine total nitrogen in the surface layer, followed by monitoring of total nitrogen in each layer, yielded higher accuracy. Based on the model's construction principles, this may be due to unresolved interference between field in-situ spectroscopy and total nitrogen content in the surface layer, while the trend of total nitrogen in soil varies relatively steadily with soil depth. However, the study also showed that the integrated monitoring model for loamy sandy soils had lower accuracy than direct monitoring in the deep tillage layer. This may be because the variation in total nitrogen in each layer of loamy sandy soils is more complex than in other soil types, and further research is needed.

[0178] (4) The direct monitoring model and the comprehensive monitoring model have certain universality

[0179] To address the problem of different direct monitoring models having different prediction accuracy performance in each tillage layer, the study screened out the optimal direct monitoring model based on the NGO-GRNN algorithm. This model can monitor the total nitrogen content of soil in each tillage layer without distinguishing between soil types. The monitoring accuracy of total nitrogen in each tillage layer of soil reaches above 0.6, and it has certain universality in monitoring total nitrogen in soil in each tillage layer of three soil types: sandy loam, loamy sand, and sandy clay loam. Soil total nitrogen content is closely related to porosity, and the porosity change trend is relatively stable under the same soil type. This study proves that soil total nitrogen content and soil porosity also have a high correlation. Therefore, it can be assumed that the soil total nitrogen content does not vary much under the same soil type. The comprehensive monitoring model has certain universality in monitoring soil total nitrogen content under the same soil texture type. In addition, the verification results of the comprehensive model show that, except for the poor prediction accuracy of total nitrogen content in the deep tillage layer of loamy sand (0.55), the comprehensive monitoring model based on soil texture classification is generally superior to the direct monitoring model. The results showed that the integrated model improved the prediction accuracy of the direct monitoring model and was universally applicable in monitoring total nitrogen in topsoil under different soil textures.

[0180] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0181] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for estimating total nitrogen in soil of different cultivated layers based on field in-situ spectroscopy, characterized in that: The following steps are involved: Collect soil samples from different tillage layers in the field and obtain in-situ spectral data of the soil; The collected soil samples were analyzed indoors to determine the total nitrogen content, soil texture type, and porosity; Soil samples were classified according to soil texture type, and correlation models between total nitrogen content and in situ spectral data of soils with different textures were established. The Northern Goshawk optimization algorithm was introduced to improve the random forest and generalized regression neural network to enhance the prediction ability of the correlation model for deep soil total nitrogen content; Combining the surface spectral inversion results with the soil vertical variation equation, a comprehensive monitoring model was constructed to achieve indirect and high-precision estimation of the total nitrogen content in soils of different cultivated layers.

2. The method for estimating total nitrogen in different cultivated soil layers based on field in-situ spectroscopy according to claim 1, characterized in that: A combined preprocessing method was used to process the in situ spectral data, including convolution smoothing, derivative transformation, and normalization. The continuous projection algorithm was then used to screen sensitive bands to reduce environmental interference.

3. The method for estimating total nitrogen in different cultivated soil layers based on field in-situ spectroscopy according to claim 1, characterized in that: It also includes the use of cross-validation to verify the model, and the accuracy evaluation indicators use the determination coefficient and root mean square error to evaluate and compare the model. 2 The value is close to 1, the lower the RMSE value is, the higher the model accuracy is. The formula is described as follows: Where y i represents the measured total nitrogen content of cotton field soil sample i, represents the monitored total nitrogen content of cotton field soil sample i, represents the average total nitrogen content of cotton field soil samples, n is the number of cotton field soil samples, and SD represents the standard deviation of the monitored total nitrogen content and the measured total nitrogen content.

4. The method for estimating total nitrogen in different cultivated soil layers based on field in-situ spectroscopy according to claim 1, characterized in that: The comprehensive monitoring model is optimized based on the NGO algorithm.

5. The method for estimating total nitrogen in different cultivated soil layers based on field in-situ spectroscopy according to claim 1, characterized in that: The SR3500 full-spectrum portable ground feature spectrometer was used for in-situ spectral data acquisition. During the whiteboard calibration and spectrum measurement process, the fiber optic probe was 15 cm away from the soil sample and perpendicular to the sample.

6. The method for estimating total nitrogen in different cultivated soil layers based on field in-situ spectroscopy according to claim 1, characterized in that: The collected soil samples were air-dried, ground, sieved using a 1 mm sieve, and then thoroughly mixed and divided into two parts using the quartering method for indoor spectral collection and chemical analysis, respectively.

7. The method for estimating total nitrogen in different cultivated soil layers based on field in-situ spectroscopy according to claim 1, characterized in that: The calculation of porosity includes: total soil porosity (%) = 120 (1-bulk density / specific gravity), And soil capillary porosity: Take a magnetic disk, place a petri dish upside down in the disk, place a piece of filter paper on the petri dish, and place the ring knife and the soil column on it; Add water to the disk and make the edge of the filter paper touch the water surface, covering the Petri dish; Allow the soil column to absorb water through the filter paper until all the soil capillaries are filled with water; Take out the ring knife and cut off the wet soil that has expanded due to water absorption and exceeds the ring knife with a knife. Weigh it together with the wet soil column. Then subtract the weight of the ring knife to get the weight of the wet soil filled with capillary water. Take 10-20g of soil sample from the top of the ring cutter, place it in an aluminum box and burn it, measure its moisture content, and calculate the dry soil weight in the ring cutter; Calculation: Capillary porosity (%) = (capillary water volume / soil volume) 120; Capillary water volume = weight of wet soil filled with capillary water - dry weight of the same volume of soil.

8. The method for estimating total nitrogen in different cultivated soil layers based on field in-situ spectroscopy according to claim 1, characterized in that: The total nitrogen content of soil samples was determined as follows: the collected soil samples were crushed and turned over to remove foreign matter, spread into a thin layer of 2 cm to 3 cm, placed in a dry and ventilated environment for air drying, sieved to 1 mm, and subjected to chemical experiments on the soil samples, and nitrogen treatment was performed using the Kjeldahl total nitrogen method.

Images