CN115797790A

CN115797790A - Satellite remote sensing partition modeling method for salinity of whole soil profile of regional scale

Info

Publication number: CN115797790A
Application number: CN202211173144.8A
Authority: CN
Inventors: 彭杰; 冯春晖; 王明玥; 石帅帅; 翟家祥; 韩建文
Original assignee: Tarim University
Current assignee: Tarim University
Priority date: 2022-09-26
Filing date: 2022-09-26
Publication date: 2023-03-14

Abstract

The invention discloses a satellite remote sensing partition modeling method for salinity of a regional scale whole-segment soil profile, which comprises the following steps: acquiring a vegetation index and a soil salinity index of a remote sensing image; determining partition factors of the partitions according to the soil profiles at different depths, and classifying the vegetation index and the soil salinity index of the remote sensing image into corresponding partitions according to the partition factors; and performing multiple linear regression on the vegetation index and the soil salinity index of the remote sensing image in each partition to obtain a piecewise linear function as a final model. The method can acquire the salinization information of the soil profile at high precision, provides an advanced technical means for the development and utilization of salinized soil resources, and has positive significance on the background of rapid increase of global population.

Description

Satellite remote sensing partition modeling method for salinity of whole soil profile of regional scale

Technical Field

The invention relates to the technical field of satellite remote sensing, in particular to a satellite remote sensing partition modeling method for salinity of a regional scale whole-segment soil profile.

Background

Salinized soil is a soil resource widely distributed in arid and semi-arid regions, although the soil has high salt content, low fertility, monotonous biodiversity, low plant productivity (Bui, 2013, cassel et al,2015 et al,2018, wang et al, 2018), but with the rapid expansion of population number, the number of existing cultivated lands hardly meets the living needs of rapidly growing population and the needs of social sustainable development, salinized soil has been reclaimed in a large scale as a backup cultivated land resource (Peng et al,2019, gutierrez and johnson, 2010. In the process of improving salinized soil, a large amount of fresh water resources are consumed, and the consumption of the fresh water resources is increased along with the increase of the salt content in the soil (Corwin et al, 2007). The water consumption required for improving the salinized soil is calculated according to a washing quota, wherein the washing quota refers to the water consumption of leaching salt washing required by the unit area of the soil for reaching a desalination standard, the desalination standard is related to the thickness of a desalinization soil layer, and the thickness of the desalinization soil layer is generally 0-60cm,0-80cm or 0-100cm (Corwin et al,2007, letey et al, 2011. Salinized soil needs to be reasonably reclaimed according to washing quota data, so that the utilization efficiency of limited water resources can be improved, and the requirements of ecological water and agricultural water can be balanced. Therefore, the profile soil salinity data is an important parameter for calculating the saline soil irrigation quota and is also an important reference basis in saline soil resource reclamation planning.

The traditional profile soil salinity survey depends on digging a soil profile, which is time-consuming, labor-consuming and low in survey efficiency, and for large-area regional survey, as soil salinization has strong spatial variation characteristics, only depending on limited sample points of manual survey, dynamic information of spatial distribution of soil salinity is difficult to accurately reflect (Allled et al,2014;barbouchi et al,2015; harti et al, 2016). Satellite remote sensing technology is a means capable of rapidly, cheaply, real-timely and dynamically monitoring soil properties (Farifteh et al,2006, peng et al, 2019), and has been widely applied to monitoring soil salinization. Currently, most reports mainly carry out research work on quantitative monitoring and spatio-temporal variation characteristics of surface soil salinity (Fern & dez-places et al,2006 ² Between 0.05 and 0.74. To date, very few reports have been made of remote satellite monitoring of soil salination for sections of depths equal to or greater than 40cm (0-40cm, 0-60cm,0-80cm and 0-100 cm).

Surface soil salinization has strong space-time variability, and meanwhile, soil salinization also has strong differentiation in the vertical direction of a section, the monitoring result of surface soil salinization often has strong uncertainty, and even the monitoring result among different months in the year can generate larger difference (Cho et al,2018

2015; harti et al, 2016; gutierrez and johnson, 2010), it is difficult to provide a decision basis for the national soil resource management department to stably and reliably develop and utilize the saline soil resources. However, in case of arid regions, due to the rare precipitation and the long time interval between two adjacent precipitations, especially in case of arid regions where the single precipitation is usually less than 20mm, the migration depth of salt in the profile is very limited, and the migration phenomenon usually only occurs in surface soil. Therefore, salination information with a certain depth of soil profile is relative to the surface soil when the salination information is relative to the surface soilThe space variant is relatively weak and the monitoring results are also relatively more stable (Zhang et al, 2014). Therefore, how to acquire the salinization information of the soil profile with high precision is a key scientific problem which needs to be solved in the process of developing and utilizing the salinization soil resources under the background of rapid increase of global population.

Disclosure of Invention

The invention provides a satellite remote sensing partition modeling method for salinity of an entire regional scale soil section, aiming at the defect that the acquisition precision of salinization information of the soil section is to be improved, and the technical problem in the background technology is effectively solved.

The specific technical scheme of the invention is as follows:

according to a first technical scheme of the invention, a satellite remote sensing partition modeling method for regional scale whole-segment soil profile salinity is provided, and the method comprises the following steps:

acquiring a vegetation index and a soil salinity index of a remote sensing image;

determining partition factors of the partitions according to the soil profiles at different depths, and classifying the vegetation index and the soil salinity index of the remote sensing image into corresponding partitions according to the partition factors;

and performing multiple linear regression on the vegetation index and the soil salinity index of the remote sensing image in each partition to obtain a piecewise linear function as a final model.

Further, the vegetation index includes:

green normalized index NG, NG = G/(NIR + R + G);

red light normalization index NR, NR = R/(NIR + R + G);

near infrared normalized index NNIR, NNIR = NIR/(NIR + R + G);

ratio vegetation index RVI, RVI = NIR/R;

greenness ratio vegetation index GRVI, GRVI = NIR/G;

differential vegetation index DVI, DVI = NIR-R;

greenness difference vegetation index GDVI, GDVI = NIR-G;

normalized difference vegetation index NDVI, NDVI = (NIR-R)/(NIR + R);

green normalized difference vegetation index GNDVI, GNDVI = (NIR-G)/(NIR + G);

soil adjusted vegetation index SAVI, SAVI = (1+L) × (NIR-R)/(NIR + R + L) ];

greenness soil regulating vegetation index GSAVI, GSAVI = (1+L) = [ (NIR-G)/(NIR + G + L) ];

optimizing soil adjusted vegetation index OSAVI, OSAVI = (NIR-R)/(NIR + R + 0.16);

green light optimized soil adjusted vegetation index GOSAVI, GOSAVI = (NIR-G)/(NIR + G + 0.16);

modified soil is used for adjusting vegetation index 2MSAVI2,

MSAVI2＝0.5*[2*(NIR+1)-SQRT((2*NIR+1)2-8*(NIR-R))]；

renormalized vegetation index RDVI, RDVI = SQRT (NDVI DVI);

three-band maximum gradient difference vegetation index TGDVI, TGDVI = (NIR-R)/(lambda) _NIR -λR)-(SWIR–NIR)/(λ _SWIR –λ _NIR )；

Integrated spectral response index COSRI, COSRI = (B + G)/(R + NIR) × (NIR-R)/(NIR + R);

normalized differential moisture index NDWI, NDWI = (NIR-SWIR)/(NIR + SWIR);

greenness vegetation index GVI,

GVI＝-0.2848*B-0.2435*G+0.5436*R+0.7243*NIR+0.084*SWIR1-0.18*SWIR2；

canopy response salinity index CRSI, CRSI = SQRT [ (NIR R-G B)/(NIR R + G B) ];

enhanced vegetation index EVI, EVI = G [ (NIR-R)/(NIR + C1R-C2B + L) ];

a green wave atmospheric impedance index GARI, GARI = { NIR- [ G + γ (B-R) ] }/{ NIR + [ G + γ (B-R) ] };

generalized differential vegetation index GDVI, GDVI = (NIR) ⁿ –R ⁿ )/(NIR ⁿ +R ⁿ )；

Nonlinear vegetation index NLI, NLI = (NIR) ² -R)/(NIR ² +R)；

The soil salinity index comprises:

salting index SI-T, SI-T = (R/NIR) × 100;

luminance index BI, BI = SQRT (R) ² +NIR ² )；

Normalized difference salt index NDSI, NDSI = (R-NIR)/(R + NIR);

salinity index SI, SI = SQRT (B × R);

salinity index 1si1, si1=sqrt (G ×);

salinity index 2si2, si2= sqrt (G2 + R2+ NIR 2);

salinity index 3si3, si3= sqrt (G2 + R2);

salinization index 1S1, S1= B/R;

salination index 2S2, S2= (B-R)/(B + R);

salination index 3S3, S3= (G × R)/B;

salination index 5S5, S5= (B × R)/G;

salination index 6S6, S6= (R NIR)/G;

intensity index 1int1, int1= (G + R)/2;

intensity index 2int2, int2= (G + R + NIR)/2;

in the following formulas, the first and second groups,

g represents the reflectivity received by the satellite multispectral sensor in the green band,

NIR represents the reflectivity received by the satellite multispectral sensor in the near infrared band,

r represents the reflectivity received by the satellite multi-spectral sensor in the red band,

SQRT represents the mathematical operation square-on,

λ _SWIR represents the central wavelength of the satellite multispectral sensor in the short-wave infrared band,

λ _NIR represents the central wavelength of the satellite multispectral sensor in the near infrared band,

SWIR1 represents the reflectivity received by the satellite multispectral sensor in the short wave infrared 1 band,

SWIR2 represents the reflectivity received by the satellite multispectral sensor in the short wave infrared 2 band,

and B represents the reflectivity received by the satellite multispectral sensor in a red wave band.

Further, according to the soil section of different depths, confirm the subregion factor of subregion to according to the subregion factor with the vegetation index and the soil salinity index of remote sensing image classify to corresponding subregion, include:

presetting four partitions which are a first partition, a second partition, a third partition and a fourth partition respectively;

when the soil profile depth is 0-60 cm:

the partition factor of the first partition is a normalized difference vegetation index NDVI, and under the condition that the NDVI is greater than 0.224361, the vegetation index and the soil salinity index of the remote sensing image are classified into the first partition;

the partition factors of the second partition are satellite remote sensing ground temperature LST and digital elevation data DEM, and under the condition that LST is greater than 48.61&DEM < =1035, the vegetation index and the soil salinity index of the remote sensing image are classified into the second partition;

the partition factors of the third partition are satellite remote sensing ground temperature LST and digital elevation data DEM, and under the condition that LST < =48.61 and DEM < =1035, the vegetation index and the soil salinity index of the remote sensing image are classified into the third partition;

the partition factors of the fourth partition are digital elevation data DEM and normalized difference vegetation indexes NDVI, and under the condition that DEM is more than 1035&NDVI < =0.224361, the vegetation indexes and the soil salinity indexes of the remote sensing images are classified into the fourth partition.

when the soil profile depth is 0-80 cm:

the partition factor of the first partition is a normalized difference vegetation index NDVI, and under the condition that the NDVI is greater than 0.177036, the vegetation index and the soil salinity index of the remote sensing image are classified into the first partition;

the partition factors of the second partition are satellite remote sensing ground temperature LST, digital elevation data DEM and normalized difference vegetation index NDVI, and under the condition that LST is greater than 52.7477 and DEM >, 1022 and NDVI < =0.177036, the vegetation index and the soil salinity index of the remote sensing image are classified into the second partition;

the partition factor of the third partition is digital elevation data DEM, and under the condition that DEM < =1022, the vegetation index and the soil salinity index of the remote sensing image are classified into the third partition;

the partition factors of the fourth partition are satellite remote sensing ground temperature LST and normalized difference vegetation index NDVI, and under the condition that LST < =52.7477 and NDVI < =0.177036, the vegetation index and the soil salinity index of the remote sensing image are classified into the fourth partition.

when the soil profile depth is 0-100 cm:

the partition factor of the first partition is a normalized difference vegetation index NDVI, and under the condition that the NDVI is greater than 0.214, the vegetation index and the soil salinity index of the remote sensing image are classified into the first partition;

the partition factors of the second partition are satellite remote sensing ground temperature LST, digital elevation data DEM and normalized difference vegetation index NDVI, and under the condition that LST is greater than 52.7477 and DEM > & 1022 and NDVI < =0.214, the vegetation index and the soil salinity index of the remote sensing image are classified into the second partition;

the partition factors of the fourth partition are satellite remote sensing ground temperature LST and normalized difference vegetation index NDVI, and under the condition that LST < =52.7477 and NDVI < =0.214, the vegetation index and the soil salinity index of the remote sensing image are classified into the fourth partition.

Further, when the soil section depth is 0-60 cm:

the regression model for the first partition is represented as: EC =34.51461-12.6 _13NDVI-33 _14NR-0.013DEM +2.1 _11BI-0.201518 TGDVI +0.04 15_13LST;

the regression model for the second partition is represented as: EC = -208.734453+366 15_14NR +26.3 _11BI +0.72 _13LST +0.058DEM-13.5 _13NDVI;

the regression model for the third partition is represented as: EC =75.630928-39.2 _11bi-7.8 _13ndvi-0.006DEM-4 _14nr;

the regression model for the fourth partition is represented as: EC =101.650755+228 15_14NR +3.76 2015TGDVI-0.145DEM-25.5 _00GARI.

Further, when the soil profile depth is 0-80 cm:

the regression model for the first partition is represented as: EC = -109.575883+443 20151026NR-0.022DEM +3 15 _1BI-3.9 _00GARI;

the regression model for the second partition is represented as: EC =71.659178-0.81 _07LST +20 20151026NR-0.003DEM;

the regression model for the third partition is represented as: EC =55.34712-178.1 \ u 14NDVI +3.46 2015TGDVI-0.36 15 \ u 07LST-0.008DEM +0.8 \ u 11BI +18 20151026NR;

the regression model for the fourth partition is represented as: EC =77.466334-91 15_14NDVI-2.89 2015TGDVI +20.7 _00GARI-0.044DEM.

Further, when the soil profile depth is 0-60 cm:

the regression model for the first partition is represented as: EC = -101.20565 377 20151026NR-0.011DEM-2.1 15 u 00GARI +1 15 u 11BI;

the regression model for the second partition is represented as: EC =234.33144-0.153DEM-0.9 \ u 07lst;

the regression model for the third partition is represented as: EC =36.85798-91.1 15\u14NDVI-3.45 15_02TGDVI-0.008DEM +21 20151026NR;

the regression model for the fourth partition is represented as: EC =154.16555-70.7 _14NDVI-0.12 DEM-0.02 _07LST.

According to a second technical scheme of the invention, the satellite remote sensing partition modeling device for regional scale whole soil profile salinity comprises a processor, wherein the processor is configured to:

According to a third aspect of the present invention, there is provided a computer-readable storage medium having stored thereon computer-readable instructions which, when executed by a processor of a computer, cause the computer to perform the method as described above.

According to the satellite remote sensing partition modeling method for regional scale whole-section soil profile salinity, disclosed by the embodiment of the invention, the salinization information of the soil profile can be obtained at high precision, an advanced technical means is provided for the development and utilization of the salinized soil resources, and the method has positive significance on the background of rapid global population growth.

Drawings

In order to more clearly illustrate the detailed description of the invention or the technical solutions in the prior art, the drawings that are needed in the detailed description of the invention or the prior art will be briefly described below. Throughout the drawings, like elements or portions are generally identified by like reference numerals. In the drawings, elements or portions are not necessarily drawn to scale.

Fig.1 shows a flowchart of a method for satellite remote sensing zonal modeling of regional-scale whole-soil profile salinity according to an embodiment of the present invention.

FIG. 2 is a diagram illustrating a geographic location of a research area and a distribution of sampling points according to an embodiment of the present invention.

FIG. 3a shows a salinity profile for a soil profile at a depth of 0-60cm, according to an embodiment of the present invention.

FIG. 3b shows a salinity profile for a soil profile at a depth of 0-60cm according to an embodiment of the present invention.

FIG. 3c shows a salinity profile for a soil profile at a depth of 0-60cm according to an embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be described clearly and completely below, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

The invention will now be further described with reference to the accompanying drawings.

The embodiment of the invention provides a satellite remote sensing partition modeling method for salinity of an integral soil section of a regional scale. The method adopts a Cubist method for modeling. Cubist originates from a machine learning method of a predictive model tool, which is similar to a classification regression tree, and builds linear regression models on different leaves. Therefore, the regression model it constructs is a piecewise linear model. The Cubist model tree is converted into a series of rules, each of which generates an associated easy-to-interpret linear model, which involves a simplification of the rule set derived from the path from the root node to each leaf. Each Cubist rule is based on a linear model of the If condition. The linear model is used to predict the observation dataset if the predictor variables associated with the observations satisfy the set of conditions. When any one of the observed values and the predictive variable associated therewith satisfy a plurality of rules, in this case, the average value of the predictions is taken as the final predicted value.

Referring to fig.1, it is a flow chart of the method of the present invention, which includes steps S101-103, specifically as follows:

step S101, acquiring a vegetation index and a soil salinity index of a remote sensing image;

step S102, determining partition factors of partitions according to soil profiles of different depths, and classifying vegetation indexes and soil salinity indexes of the remote sensing images into corresponding partitions according to the partition factors;

and S103, performing multiple linear regression on the vegetation index and the soil salinity index of the remote sensing image in each partition to obtain a piecewise linear function as a final model.

In step S101, the original band information, the vegetation index, and the soil salinity index are classified by calculating a detailed calculation formula of the vegetation index and the soil salinity index of the remote sensing image of the current day, which is shown in table 1.

Table 1.

In each formula, G represents the reflectivity received by the satellite multispectral sensor in the green wave band,

r represents the reflectivity received by the satellite multispectral sensor in the red band,

SQRT represents the mathematical operation square-on,

and B represents the reflectivity received by the satellite multispectral sensor in the red band.

In step S102, three different depth profiles, which are 0-60cm,0-80cm and 0-100cm, are selected, and corresponding to the different depth profiles, related partition factors are determined according to the different depth profiles, and partition processing is performed on the vegetation index and the soil salinity index of the remote sensing image according to preset conditions, which is specifically referred to table 2.

Table 2.

In step S103, referring to table 3, multiple linear regression is performed based on the data in the partitions formed in table 2. And finally, combining the rule and the multiple linear regression model to obtain a piecewise linear function as a final model.

Table 3.

The following examples of the present invention will further illustrate the feasibility and advancement of the present invention in conjunction with specific experimental data.

The research area selected in this example is located in Xinjiang, china, and the specific locations are sky Tai rike alluvial fan, beijing Tianshan mountain range, and south-adjacent Alar City (FIG. 1). The research area is a desert-oasis desert transition zone typical of Xinjiang, belongs to a prefecture flood alluvial fan, is positioned at 40-41-32 'N, 80-36-81-41' E, has an altitude of 1000-1400m, is high in the north, south, low in the south, west and east, is 110km long in the south and north, is 80km wide in the east and west, is fan-shaped, and has an area of 5600km ² The solar energy-saving greenhouse belongs to the arid climate in the continental warm temperature zone, the annual average precipitation is 46.4-64.5mm and is concentrated in 6, 7 and 8 months, the annual average evaporation rate is 1992.0-2863.4mm, the evaporation-reduction ratio is about 40Rich, annual average solar total radiant quantity 544-590kJ cm ^-2 Annual sunshine 2855-2 967h, and frost-free period up to 205-219d (Peng et al, 2019). The research area is provided with a south-north S215 province, and the south-north S215 province only has a plurality of agricultural tractor-ploughed roads in the east-west direction through the whole research area. The land utilization types comprise newly-reclaimed lands and deserts, most of the newly-reclaimed lands are in a salinization improvement stage, only a small part of the newly-reclaimed lands begin to be cultivated, and the planted crops are mainly cotton. Most of the natural vegetation growing in desert is halophytes, such as tamarisk, halostachys caspica, haloxylon, reed, alhagi, etc., and the vegetation coverage is 0-100% (Jiang et al, 2019). Vegetation coverage has an increasing trend from south to north. Ground investigation results for many years show that the salinization degree of soil shows a trend of increasing first and then decreasing from south to north, and the salinization trend of soil shows a trend of increasing first and then decreasing then increasing from west to east (Peng et al, 2019). In recent years, due to the rapid increase of population, large-area salinized soil in the south of the research area is blindly reclaimed into cultivated land, a large amount of fresh water resources are consumed for improvement and irrigation, the contradiction between ecological water and agricultural water is increased, and the sustainable development of agriculture and ecological environment in the area is endangered.

And (3) measuring the apparent conductivity by adopting an EM38-MK2 geodetic conductivity meter, wherein the specific measurement time is 26 days in 10 months in 2015 to 28 days in 10 months in 2015, and the weather is sunny and has no rainfall. Comprehensively considering the vegetation coverage and the salinization degree, 30 survey samples with different vegetation coverage and salinization degrees are arranged in a research area. When the survey sample is selected, the area with uniform vegetation coverage and salinization degree is selected as much as possible, and the problem of heterogeneity is avoided to cause the overall data quality to be reduced. According to the spatial resolution of Landsat8, the area of each survey sample is set to be 100m × 100m, which is approximately the range of 3 pixels × 3 pixels. By adopting a grid sampling mode, each survey sample side respectively collects 5 straight lines in the south-north and east-west directions, each survey sample side respectively collects about 1000 measuring points, and the measuring point distance is about 1.0m (see figure 2).

After the apparent conductivity is collected in each survey sample, the sample is divided into 3 areas with high, medium and low values of the apparent conductivity, and 1 each of the section soil columns of 0-60cm,0-80cm and 0-100cm is collected in each area, so that the condition that the sample point data is concentrated in a certain narrow interval to influence the measuring range of the model in the modeling process and cause poor universality of the model is avoided. The sampling tool is an earth drill with the caliber of 38mm, and the whole research area collects 270 profile samples with different depths for constructing an inversion model between the conductivity and the apparent conductivity. And after the apparent conductivity is converted into the conductivity by using the inversion model, the average value of the conductivities of all measuring points in the same pixel is taken to represent the final conductivity of the pixel.

The collected soil sample is taken back to a laboratory to remove plant litters, roots and stones, is naturally air-dried in the laboratory and ground by a 2mm sieve, is dried at 105 ℃ and is naturally cooled, and then the conductivity is measured by adopting a soil-water ratio 1:5 (Peng et al, 2019).

In order to establish the grading standard of the salinization of the desert soil, the salinization condition of the soil and the growth condition of salt-tolerant vegetation in a research area are investigated, the main investigation content of the salinization condition of the soil is the conductivity, and the main investigation content of the growth condition of the salt-tolerant vegetation is the vegetation coverage and the diversity of plants. According to the investigation result, the grading standards of natural soil salinization of Zhang et al (2015) and Fern-ndez-Buces et al (2006) are used for reference, and the grading standards of the research area are defined as 5 grades of non-salinization, light salinization, medium salinization, heavy salinization and saline soil. In the investigation, 5 typical sample plots are investigated for each salinization grade, 5 surface soil samples of 0-20cm are collected from each sample plot, and the vegetation coverage and the plant species are investigated simultaneously. The soil sample is air-dried, ground and sieved indoors, and then the conductivity is measured by adopting a soil-water ratio 1:5. And (3) synthesizing the growth condition of the salt-tolerant vegetation and the conductivity data of the surface soil to prepare a salinization grading standard of the desert soil, which is shown in a table 1.

TABLE 1 grading Standard for salinization of desert soil

Grade	EC(dSm ^-1 )	Vegetation coverage (%)
			Non-salinization	≤7.5	80-100
Slight salinization	7.5-15	40-80
			Moderate salinization	15-30	10-40
Severe salinization	30-60	5-10
			Extreme salinization	≥60	0-5

The remote sensing data selected by the embodiment are satellite image data and DEM data of three sensors, namely Landsat5, landsat7 and Landsat8, a research area needs two images to be completely covered, the row numbers and the column numbers of the images are 146/31 and 146/32, 163 images with cloud coverage lower than 10% and no ice and snow coverage in 1990-2015 are collected, and 1990.8-1999.3 and 2003.6-2013.1 select data, namely Landsat5, and 92 scenes are collected; 1999.4-2003.5 data selected is Landsat7, 28 scenes in total; 2013.2-2015.10 data selected is Landsat8, 43 scenes total. Landsat5, landsat7, and Landsat8 images were downloaded from a website provided by the United States Geological Survey (USGS) at https:// glovis. USGS/, at 30m DEM data resolution, and at https:// earth x plorer. USGS/. The satellite image processing flow comprises radiation correction, atmospheric correction, geometric correction and mosaic processing, and the DEM is filled with depression. The processed satellite images are processed by vegetation indexes, salinity index calculation, surface Wen Fanyan, synthesis of perennial maximum values, minimum values and average values of the parameters and the like.

Based on the remote sensing data of the research area acquired by the method, partition modeling is carried out according to the flow method shown in fig.1, and the performance of the obtained regression model is detected. In this example, 4 indices of R2, RMSE, MAE and RPD were used for evaluation. Generally, the smaller the RMSE and MAE, the higher the R2 and RPD, which indicates that the higher the model precision and the more reliable the performance; conversely, the lower the model accuracy, the less reliable the performance (Peng et al, 2016). RPD >2, indicating that the model has high-precision prediction capability, 1.4 of RPD <2, indicating that the model has only the capability of distinguishing high values from low values, RPD <1.4, indicating that the model has no prediction capability (Nocita et al, 2013, gomez et al, 2013, chang and Laird, 2002.

Using apparent conductivity data of profiles of 450 different depths in 30 sample areas (see fig. 2) collected on days 26-28 of 10 months 10 in 2015 and measured conductivity data in a room as data sources, 3 regression models of measured conductivity were constructed using 1m coil (ECav 1 m) or 0.5m coil (ECav 0.5 m) of the apparent conductivity in vertical mode and the apparent conductivities of both coils as independent variables (see table 2). In the test, 5 section soil samples with different depths are respectively collected in each of 30 samples, and 90 section soil samples with different depths are collected in each sample. In the modeling process, 90 samples of each section are sorted according to the conductivity, and then are sampled at equal intervals, and the ratio of the number of the samples of the modeling set to the number of the samples of the prediction set is 2:1, namely, 60 modeling sets and 30 prediction sets are obtained.

Through significance test, all models of the modeling set reach an extremely significant level, but through test, the R of soil models with different depth profiles is found ² Has certain difference, 0.68-0.79,0-40cm in the soil section of 0-20cm, 0.75-0.80,0-60cm, 0.79-0.86,0-80cm0.81-0.89,0-100cm is 0.82-0.90. It can also be seen that as the profile depth increases, R ² Showing an increasing trend, R of 0-20cm ² Relatively low and relatively high in the range of 0-100 cm. This is because the ECav1m and ECav0.5m data respectively reflect the total average information of the salinity of the soil section of 0-100cm and 0-75cm, and the salinity information in the section is higher in coincidence with the ECav and ECah information along with the increase of the depth, so that R is presented ² The trend toward an increase. In the three models constructed by different independent variables, wherein the ECav0.5m can only detect the salinity information of 0-75cm and is not independently used for constructing the section conductivity model of 0-100cm, compared with the modeling precision of the three models, the precision of the two-variable model of ECav and ECah is obviously superior to that of the single-variable model, and the R of the two-variable model is obviously superior to that of the single-variable model ² 0.79-0.93, and R for the ECav and ECah models ² Respectively 0.68-0.82 and 0.70-0.92. Therefore, the subsequent inversion of the conductivity of the soil profile is carried out by adopting an ECav and ECah two-variable model.

TABLE 2 regression model of apparent conductivity versus measured conductivity (dS m) ^-1 )(n＝60)

The results of the model test using 30 independent samples are shown in Table 3, and the predicted results show that R is compared against the modeled set ² There was only a small drop, indicating that the model was very stable. Meanwhile, the RMSE is between 2.31 and 10.70dS m ^-1 And the RMSE shows a trend of obvious reduction along with the increase of the section depth, which is mainly attributed to the obvious apparent aggregation phenomenon of the saline soil in Xinjiang, most of salt is aggregated in 0-20cm, and the salt content is obviously reduced along with the increase of the section depth. Among the three models, the RMSE of the bivariate model is obviously lower than that of the univariate model, the second is ECav0.5m, and the RMSE of the ECav1m model is the highest, which is similar to that of the modeling set R ² The change rules of the above-mentioned two methods are completely identical.

The application of electromagnetic induction technology in the field of soil salination monitoring began in the 60's of the 20 th century (Doolittle and Brevik, 2014) at the earliest. To date, a great deal of correlation has been developed around the worldResearch, but most have focused on surface soils or soil profiles in a particular soil layer (Lesch et al, 2005 yao and Yang,2010, taghizadeh-Mehrjardi et al,2014 and Yu,2014 triantafils et al, 2013 triantafils and Monteiro Santos,2013, scudiero et al,2014 scudiero et al,2015 corwin and Lesch,2014 wu et al, 2014), while research for different depth profiles is less. Compared with the similar research, firstly, the research starts from the basic principle of electromagnetic induction, the salinity information of the soil profile is highly matched with the apparent conductivity, the model precision is effectively improved, particularly the model precision is improved by 0-60cm and 0-80cm, the matching degree of the salinity information of the soil profile and ECav0.5m is higher, R is ² Can reach more than 0.89. And secondly, the ECav1m and ECav0.5m information is simultaneously adopted in the research, so that the model precision is obviously improved. Finally, the investigation time is concentrated on 26-28 days of 10 months in 2015, the period is short, and the investigation is in a dry season period, so that the interference of cooling water and air temperature is effectively avoided. However, the method has certain defects, because the workload of ground investigation is large, the task is heavy, the number of single depth profile samples collected by each sample area is small, and the model of the sample area scale cannot be built, so that a global modeling mode is adopted. Corwin and Lesch (2014) studies show that the field scale model accuracy is significantly higher than the zone scale model. Therefore, if the number of samples of the sample area is increased and the sample area is used as a basic unit for modeling, higher model accuracy can be obtained.

TABLE 3 comparison of measured and predicted conductivities (dS m) ^-1 )(n＝30)

Conductivity data of different depth profiles of 30 sample areas are obtained by using an inversion model, and specific statistical conditions are shown in table 4. The conductivity of the profiles with different depths is 4.14-68.07dS m ^-1 The average value is 22.19-41.55dS m ^-1 According to the grading standard of salinization, depth sections of 0-60cm,0-80cm and 0-100cm belong to moderate salinization, and depth sections of 0-20cm and 0-40cm belong to severe salinization; the apparent conductivity of ECav1m is 134.69-1120.23dS m ^-1 ECav0.5m is 143.35-1446.54dS m ^-1 . ComparisonThe conductivity of the sections with different depths can be known, the conductivity shows a trend of obvious reduction along with the increase of the depth of the sections, and the conductivity of 0-20cm is 9.54-68.07dS m ^-1 Average value of 41.55dS m ^-1 And 0-100cm are 4.14-39.02 and 22.19dS m respectively ^-1 (ii) a Meanwhile, the apparent conductivities of the ECav0.5m and the ECav1m are counted, the statistical result shows that the apparent conductivity of the ECav0.5m is higher than that of the ECav1m, and mutual evidence shows that the soil salinization in the research area has an obvious apparent aggregation phenomenon. In addition, the coefficient of variation also shows a remarkable increasing trend along with the increase of the depth, the coefficient of variation is increased from 33.27% of 0-20cm to 37.28% of 0-100cm, which indicates that the surface conductivity of the soil section is similar, and the conductivity of the soil deeper than the surface conductivity of the soil section can be greatly differentiated along with the increase of the depth. Therefore, uncertainty exists in the relationship between the conductivity data of the surface soil and the conductivity of the deep soil, the conductivity of the deep soil of some soils with high surface conductivity is relatively high, and the conductivity of the deep soil of some soils with high surface conductivity is relatively low.

TABLE 4 descriptive statistics of soil profile characteristics (dS m) ^-1 )

The method comprises the steps of extracting 55 surface parameters such as NDVI, EVI, SI and the like by utilizing a synchronous satellite image, analyzing the correlation between the parameters and the soil conductivity, and simultaneously carrying out autocorrelation inspection, wherein only 6 surface parameters such as NR, TGDVI, NDVI, LST, GARI and TC1 with the best correlation are selected for carrying out time sequence analysis on the basis of the size of statistical workload, wherein the TGDVI, NDVI, NR and GARI are mainly used for reflecting the dynamic change of vegetation coverage, among the four parameters, the TGDVI, NDVI and GARI are in positive correlation with the vegetation coverage, the interference of accidental factors such as grazing, plant diseases and insect pests, drought and the like on the vegetation coverage is considered, so the perennial maximum value of the vegetation coverage is solved, the NR and the negative correlation with the vegetation coverage are solved, the perennial minimum value of the TGDVI, NDVI and GARI are also used for reflecting the dynamic change of the brightness and the temperature of a bare soil area, and the perennial maximum value of the vegetation is solved. Table 5 shows the correlation of the conductivity of the different depth profiles to the optimal time scale of the surface parameters. As can be seen from Table 5, the time period corresponding to the highest correlation coefficient between conductivity and the maximum or minimum of the surface parameters showed a significantly increasing trend with increasing depth of the soil profile, 5-10 years for 0-60cm, 7-18 years for 0-80cm and 0-100 cm. As the depth of the section increases, the correlation between the surface parameters and the conductivity has a trend of being obviously weakened, which is mainly related to that the satellite remote sensing can only detect the soil surface layer information. Among these 6 parameters, although the conductivity-dependent property reaches a very significant level, NR, TGDVI and NDVI are superior to LST, GARI and TC 1in comparison, where NR is the most relevant and TC1 is the worst.

TABLE 5 maximum correlation coefficient between extreme parameter values synthesized at different time intervals and conductivity of soil profile

Table 6 is the Cubist subset partitioning rule. As can be seen from Table 7, the divisions of the 0-60cm,0-80cm and 0-100cm subsets were mainly made according to the values of NDVI, DEM and LST, and each section was divided into 4 subsets, but the division rules were different. It was also found that NR, TGDVI and TC1 did not participate in the partitioning of the subsets, indicating that these 3 factors did not differ significantly in the study area or differenced similarly to the enrolled partition factors, but not as significantly as the enrolled partition factors. In addition, comparing the subset partition factors of the soil profiles with different depths, it can be known that as the depth of the soil profile increases, the number of the subset partition factors increases, and the subset partition of the deeper soil profile is often limited by multiple factors. Among the selected partition factors, NDVI participates in all the partitions of the soil profile, DEM participates in the partitions except for the soil profile of 0-20cm, and LST only participates in the partitions of the soil profile of 0-60cm,0-80cm and 0-100 cm.

TABLE 6 partition factors and rules for different depth soil profiles

Table 7 lists the contribution rates of the participating partition factors. In a 0-60cm soil section, the DEM contribution rate is highest, and then LST and NDVI contribution rates are lowest; the two soil sections of 0-80cm and 0-100cm have the same law, and both have the highest NDVI contribution rate, and secondly have the LST and the DEM minimum. In all the 3 factors participating in zoning, the contribution rate does not show a trend of increasing or decreasing along with the increase of the depth of the soil profile, which mainly comes from two factors, namely that the change of soil salinity in the profile has strong spatial variability, and the time scales of the same factor selected by the profiles of different depths are not completely consistent.

TABLE 7 partition contribution rate of surface parameters of different depth soil profiles

Soil profile	NDVI	LST	DEM
				0-60cm	44％	56％	84％
0-80cm	70％	50％	44％
				0-100cm	68％	49％	46％

Table 8 shows the results of the Cubist model run for different depth profile conductivities. R of different-depth profile Cubist model in modeling set ² At 0.88-0.90, there was no significant difference, RMSE was 2.65-4.28dS m ^-1 MAE at 2.03-3.33dS m ^-1 RMSE and MAE have a tendency to decrease with increasing profile depth, which is mainly related to the cause of the decrease in conductivity with increasing profile depth. In the prediction set, R for soil sections of 0-20cm and 0-40cm and 0-100cm compared to the modeling set ² The reduction does not appear to be obvious, but the reduction of 0-60cm and 0-80cm has slight amplitude, the reduction is respectively reduced to 0.83 and 0.80 from the original 0.90 and 0.89, the RMSE is 2.97-5.08dS m ^-1 MAE at 2.28-3.92dS m ^-1 . The RPD for all sections was greater than 2.0, compared to 0-40cm, the highest, reaching 3.00, and 0-80cm, the lowest, 2.10. According to the evaluation standard of Williams (2001), the soil sections of 0-20cm, 0-40cm,0-60cm and 0-100cm all satisfy RPD ≥ 2.0 and R ≤ 0.82 ² The condition of less than or equal to 0.90 indicates that the corresponding section model has good prediction capability, and the soil section of 0-80cm can only meet the conditions that RPD is more than or equal to 1.5 and R is more than or equal to 0.66 ² And the condition is less than or equal to 0.81, and the model only has the capability of rough prediction. In order to verify the prediction effect of the model, the average values of profile conductivity distribution maps of different depths of a research area run by using the model are specially counted, and the average values of 5 profile conductivities of 0-20-0-100 cm are 40.91, 31.96, 26.75, 24.36 and 21.92dS m ^-1 And the method is very close to the average value of the actual sampling points, so that the model has good reliability.

TABLE 8 cubic model Performance of different depth soil profiles

And (3) according to the grading standard of soil salinization, the conductivity of the sections at different depths in the research area is divided into 5 grades of non-salinization, light salinization, medium salinization, heavy salinization and saline soil, and the salinization conditions of the sections at different depths are shown in figures 3a-c. From fig. 3a-c, it can be seen that the degree of salinization of the soil has a significantly reduced tendency as the depth of the soil profile increases, for example, the heavily salinized or salinized soil area in the 0-20cm profile is reduced to moderately or slightly salinized soil area in the 0-100cm profile, which is consistent with the strong clustering characteristic of the soil salinization in the area, and the results are also illustrated from the side face to have high reliability. The salinization is accelerated from north to south and from west to east, which is mainly related to the landforms with high north, low south, high west and low east in the research area, salt is gathered in relatively low regions of the landforms under the drive of surface runoff, and the salinization degree of the southeast part is heavy after long-time salt accumulation. The degree of salinization in the vegetation-covered area and the periphery of the water system is remarkably low compared with other areas, and the degree of salinization is completely consistent with the results of multiple field investigations. The field investigation result shows that the high vegetation coverage area is generally located in a non-salinized or slightly salinized area, the sparse vegetation is generally located in a moderately salinized area, and the vegetation in the severely salinized and saline soil areas is difficult to grow and mostly is bare soil; the periphery of the water system is usually a light or medium salinization area, but the influence range is limited to be within 50 meters, and high-coverage shrubs and herbaceous plants such as tamarix chinensis, saline spike wood, salt-knot wood, reed and the like are common around the water system. Newly-reclaimed farmlands distributed in large area in the south of a research area are improved in different degrees, the salinization degree of 0-20cm surface soil of some newly-reclaimed farmlands is obviously higher than that of other surrounding areas, the newly-reclaimed farmlands belong to the salinization grade, most of the newly-reclaimed farmlands are cultivated in 1-2 year old, the salinization improvement is carried out by utilizing flood in summer, the evaporation capacity of the newly-reclaimed farmlands is extremely high, water-soluble salts are accumulated in the surface soil layer along with water evaporation in autumn and winter to form obvious secondary salinization, and white salt crusts of 5-10mm can be seen in ground investigation.

FIGS. 3a-c showThe area and percentage of different salination grades of different depth soil profiles are shown. As can be seen from figures 3a-c, in the soil sections of 3 different depths, the sections of 0-60cm,0-80cm and 0-100cm are mainly moderately salted, and the areas and the percentages are 1099.19km respectively ² (49.09％)、1422.5km ² (63.54%) and 1902.18km ² (84.96%). The percentage of the non-salinized soil is extremely low, only 0.1-3.04%, and the area and percentage of the non-salinized soil tend to obviously increase along with the increase of the depth of a section, but the area and percentage of the non-salinized soil are obviously reduced from 1.83% to 0.13% when the 0-80cm is compared with the 0-60cm soil section, according to the ground investigation result, a vegetation coverage area has an obvious salt segregation layer at 60-80cm, and the conductivity is obviously higher than that of an adjacent soil layer. The percentage of lightly salinized soil in the sections of 0-20 and 0-40cm is very low, only 1.03% and 2.13%, and the percentage in the sections of 0-60, 0-80 and 0-100cm is approximately equivalent, and is about 11%. Moderately saline soils have a tendency to increase significantly in percentage from 13.77% to 84.96% with increasing profile depth, while heavily saline soils exhibit the opposite tendency from 84.27% down to 0.13%.

The above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; such modifications and substitutions do not depart from the spirit and scope of the present invention, and they should be construed as being included in the following claims and description.

Claims

1. A satellite remote sensing partition modeling method for salinity of an area scale whole-segment soil profile is characterized by comprising the following steps:

2. The method of claim 1, wherein the vegetation index comprises:

green normalized index NG, NG = G/(NIR + R + G);

red light normalization index NR, NR = R/(NIR + R + G);

near infrared normalized index NNIR, NNIR = NIR/(NIR + R + G);

ratio vegetation index RVI, RVI = NIR/R;

greenness ratio vegetation index GRVI, GRVI = NIR/G;

differential vegetation index DVI, DVI = NIR-R;

greenness difference vegetation index GDVI, GDVI = NIR-G;

normalized difference vegetation index NDVI, NDVI = (NIR-R)/(NIR + R);

green normalized difference vegetation index GNDVI, GNDVI = (NIR-G)/(NIR + G);

soil adjusted vegetation index SAVI, SAVI =1.5 (NIR-R)/(NIR + R + 0.5) ];

greenness soil regulating vegetation index GSAVI, GSAVI = (1.5) = [ (NIR-G)/(NIR + G + 0.5) ];

modified soil is used for adjusting vegetation index 2MSAVI2,

MSAVI2＝0.5*[2*(NIR+1)-SQRT((2*NIR+1)2-8*(NIR-R))]；

renormalized vegetation index RDVI, RDVI = SQRT (NDVI DVI);

three-band maximum gradient difference vegetation index TGDVI, TGDVI = (NIR-R)/(λ) _NIR -λR)-(SWIR–NIR)/(λ _SWIR –λ _NIR )；

normalized differential moisture index NDWI, NDWI = (NIR-SWIR)/(NIR + SWIR);

greenness vegetation index GVI,

GVI＝-0.2848*B-0.2435*G+0.5436*R+0.7243*NIR+0.084*SWIR1-0.18*SWIR2；

canopy response salinity index CRSI, CRSI = SQRT [ (NIR R-G B)/(NIR R + G B) ];

enhanced vegetation index EVI, EVI = G [ (NIR-R)/(NIR +6*R-7.5 × b + 1) ];

a green wave atmospheric impedance index GARI, GARI = { NIR- [ G +1.7 (B-R) ] }/{ NIR + [ G +1.7 (B-R) ] };

generalized differential vegetation index GDVI, GDVI = NIR-R;

nonlinear vegetation index NLI, NLI = (NIR) ² -R)/(NIR ² +R)；

The soil salinity index comprises:

salting index SI-T, SI-T = (R/NIR) × 100;

luminance index BI, BI = SQRT (R) ² +NIR ² )；

Normalized difference salt index NDSI, NDSI = (R-NIR)/(R + NIR);

salinity index SI, SI = SQRT (B × R);

salinity index 1si1, si1= sqrt (G ×);

salinity index 2si2, si2= sqrt (G2 + R2+ NIR 2);

salinity index 3si3, si3= sqrt (G2 + R2);

salination index 1S1, S1= B/R;

salination index 2S2, S2= (B-R)/(B + R);

salination index 3S3, S3= (G × R)/B;

salination index 5S5, S5= (B × R)/G;

salination index 6S6, S6= (R NIR)/G;

intensity index 1int1, int1= (G + R)/2;

intensity index 2int2, int2= (G + R + NIR)/2;

SQRT represents the mathematical operation square-on,

λ _NIR indicating the center wavelength of the satellite multi-spectral sensor in the near infrared band,

3. The method of claim 1, wherein determining partition factors for the partitions according to the soil profiles at different depths, and classifying the vegetation index and the soil salinity index of the remote sensing image into corresponding partitions according to the partition factors comprises:

when the soil profile depth is 0-60 cm:

4. The method of claim 1, wherein determining partition factors for the partitions according to the soil profiles at different depths, and classifying the vegetation index and the soil salinity index of the remote sensing image into corresponding partitions according to the partition factors comprises:

when the soil profile depth is 0-80 cm:

5. The method of claim 1, wherein determining partition factors for the partitions according to the soil profiles at different depths, and classifying the vegetation index and the soil salinity index of the remote sensing image into corresponding partitions according to the partition factors comprises:

when the soil profile depth is 0-100 cm:

6. A method according to claim 3, wherein, when the soil profile depth is 0-60 cm:

7. The method of claim 4, wherein when the soil profile depth is 0-80 cm:

8. The method of claim 5, wherein when the soil profile depth is 0-60 cm:

the regression model for the second partition is represented as: EC =234.33144-0.153DEM-0.9_07lst;

the regression model for the fourth partition is represented as: EC = 154.16555-70.7_14NDVI-0.12 DEM-0.02 _07LST.

9. A device for remotely zoning and modeling satellite for salinity of an area-scale whole soil profile, comprising a processor configured to:

10. A computer-readable storage medium having computer-readable instructions stored thereon, which, when executed by a processor of a computer, cause the computer to perform the method of any one of claims 1-8.