Extraction and Analysis of Soil Salinization Information in an Alar Reclamation Area Based on Spectral Index Modeling

Abstract

In order to explore the optimal remote sensing salinity monitoring index model for the inversion of soil salinization in the Alar reclamation area, based on the Sentinel-2 images and field measured data, the salinity index 1 (SI1), the normalized difference vegetation index in a green–red band (GRNDVI), the normalized vegetation index of greenness (GNDVI), and the normalized difference vegetation index (NDVI) were selected to construct the remote sensing-based salinization 1 detection index (S1DI) model. Next, the cotton field soil salinization information in the Alar reclamation area was extracted, and the accuracy of the model was verified to obtain the optimal remote sensing salinity monitoring index model. The results show that the overall classification accuracy of the S1DI1 (SI1-GRNDVI), S1DI2 (SI1-GNDVI), S1DI3 (SI1-NDVI), and S1DI4 (SI1-DVI) models for salinity monitoring is 83.35%, 83.10%, 82.96%, and 80.25%, respectively. The S1DI1 model is most suitable for retrieving the degree of soil salinization in the cotton field in the Alar reclamation area, and the S1DI2, S1DI3, and S1DI4 models are also very useful for monitoring soil salinization in the Alar reclamation area. Using the S1DI1 model to invert the soil salinization level of the cotton fields in the Alar reclamation area, it was found that the cotton field in the reclamation area is dominated by non-saline soil, and the light saline soil and moderate saline soil are mainly distributed in the 9th and 12th clusters of the reclamation area. As the S1DI1 model possesses the highest accuracy in extracting the soil salinization information of the cotton fields in the Alar reclamation area, it can be used as a remote sensing salinity 1 monitoring index model for the inversion of the soil salinization of the cotton fields in the reclamation area, which is expected to provide an effective reference value for soil salinization monitoring.

1. Introduction

Soil salinization is a land degradation phenomenon under the joint action of nature and human activities [1,2,3]. As a global ecological environment problem, it seriously restricts the sustainable development of the agricultural economy and the stability of the ecosystem [2,4]. The Alar Reclamation Area is located in the arid area of southern Xinjiang, where the problem of soil salinization is serious [5]. Although the use of water-saving irrigation technology has effectively reduced farmland water use, the problem of the secondary salinization of the soil has increased in recent years. How to quickly and effectively extract the spatial information of soil salinization is the key to sustainable land development and agricultural scientific management in the Alar Reclamation Area [6].

The traditional field sampling survey method is characterized by consuming substantial human sources, and it is mostly limited to small-scale research obtaining point-like information [1]. On the contrary, remote sensing (RS) technology can quickly obtain large-scale ground object information in a repeat manner, which is helpful for long-term soil salinization research. Benefiting from the advantages of macroscopic coverage, comprehensive analysis, dynamical processing, and the low cost of RS technology, it has been widely used in the (near) real-time monitoring of soil salinization, the quantitative estimation of soil salinity, etc. [7,8,9,10].

According to the RS-based soil salinization analysis model based on the NDVI-SI feature space, the salinity index 1 (salinity index 1, SI1) and four types of vegetation indices that are sensitive to crop growth monitoring were selected, including the normalized difference vegetation index in a green–red band (GRNDVI), normalized vegetation index of greenness (GNDVI) [8], normalized difference vegetation index (normalized difference vegetation index, NDVI), and difference environmental vegetation index (difference environmental vegetation index, DVI). Then, based on the Sentinel-2 remote sensing image data and the field measured data in the Alar reclamation area, combined with the above five spectral indices, four types of remote sensing salinity monitoring index models are constructed: S1DI1 (SI1-GRNDVI), S1DI2 (SI1-GNDVI), S1DI3 (SI1-NDVI), and SDI4 (SI1-DVI), and the inversion of cotton field soil salinity in the Alar reclamation area is explored in this study. The optimal remote sensing salinity monitoring index model for waterlogging in this study aims to provide a scientific basis for the precise irrigation of cotton fields in the reclamation area and the good growth and development of cotton.

2. Materials and Methods

2.1. Study Area

The Alar Reclamation Area is located in the transition zone between the Taklamakan Desert and the Tianshan Mountains, as shown in Figure 1. It is a county-level city directly under the jurisdiction of the Xinjiang Uygur Autonomous Region (80°30′~81°58′ E, 40°22′~40°57′ N). It is warm, temperate, and continental and has an arid desert climate and a very fragile ecological environment. It is adjacent to the Tarim River, with a total area of 4197.58 km². There are nine regiments in the reclamation area (Numbers 7 to 16 are regiments, while Number 15 is withdrawn) and three major reservoirs, including the Shengli reservoir. The average annual evaporation, average annual precipitation, and average annual temperature are approximately 1876.6~2558.9 mm, 40.1~82.5 mm, and 10.7 °C, respectively [11]. Its evaporation is far greater than the precipitation. The wild woody plants in the Alar Reclamation Area are dominated by Populus euphratica and Tamarix, and the main economic crops include cotton, wheat, apples, fragrant pears, walnuts, etc. Among them, cotton is the largest planting area and the most widely distributed area.

2.2. Data Source and Processing

The RS data used in this study were collected from the Google Earth Engine (GEE) cloud-computing platform (https://code.earthengine.google.com, accessed on 11 January 2022) [12,13], and the Sentinel-2 surface reflectance data from 17 July to 22 July 2022 were selected. Only images with no more than 15% cloud coverage were used for the composite. The Sentinel-2 images have a spatial resolution of 10 m. After preprocessing, such as radiometric calibration and atmospheric correction, the image map of the research area was cropped to avoid the extraction of non-cotton areas such as water bodies, deserts, buildings, and fruit trees. The accuracy caused interference, so the Google Earth Engine cloud platform was used to classify the images to extract the cotton field planting area and to remove the masks of water bodies (Figure 2b), deserts(Figure 2d), buildings (Figure 2a), fruit trees(Figure 2c), etc. in the research area, as shown in Figure 2.

Soil samples were collected in the field from 15 July to 24 July 2022, and the sample points were evenly distributed in each company in the reclamation regiment. According to the triangular sampling method [12], three soil samples of 0–15 cm beneath the surface were collected for each point, and the GPS coordinates were measured by Huace S8 (Figure 1). In total, there were 115 field sample points and 345 soil samples. The soil sampling data were placed in the laboratory to dry naturally for one month, and then the three samples of the same sample were ground and evenly mixed into one part. The plant roots and stones in the soil were removed and then sieved with 0.15 mm and 100 mL of distilled water. The solution was prepared according to a water-to-soil ratio of 5:1, it was stirred fully and evenly, it was placed statically for 8~10 h and filtered, and the soil conductivity was measured with the Shanghai Keyou Conductivity Meter DDS-11A.

2.3. Models

GRNDVI, GNDVI, NDVI, and DVI are vegetation indices which can better eliminate the images of soil and the cotton canopy background in cotton density areas, thus showing a better ability to extract cotton information [14]. Cotton growth is affected by soil salinization in cotton fields. With a more serious degree of soil salinization in cotton fields, the growth of cotton will be inhibited. The vegetation index can be used as a basis to indirectly reflect the degree of soil salinization in cotton fields [15]. The study found that the green band, red band, and short-infrared waves have better responses to soil salinization information. The formula of the spectral index and the formula of the remote sensing salinity monitoring index model are shown in

Table 1. The expression of spectral indices.

Spectral Index	Expression
SI1	SI1 = √(G × R)
GRNDVI	GRNDVI = (N − R − G)/(N + R + G)
GNDVI	GNDVI = (N − G)/(N + G)
NDVI	NDVI = (N − R)/(N + R)
DVI	DVI = N − R
S1DI1 Model (SI1-GRNDVI)	S1DI1 = √(GRNDVI − 1)2 + SI12
S1DI2 Model (SI1-GNDVI)	S1DI2 = √(GNDVI − 1)2 + SI12
S1DI3 Model (SI1-NDVI)	S1DI3 = √(NDVI − 1)2 + SI12
S1DI4 Model (SI1-DVI)	S1DI4 = √(DVI − 1)2 + SI12

.

3. Results and Analysis

3.1. Scatterplot of Spectral Indices

In order to analyze the correlations between SI1 and GRNDVI, GNDVI, NDVI, and DVI, this study takes SI1 as the horizontal axis of coordinates and GRNDVI, GNDVI, NDVI, and DVI as the vertical axes of coordinates, respectively, to establish an SI1-GRNDVI, SI1-GNDVI, SI1- NDVI, and SI1-DVI two-dimensional scatter plot, as shown in Figure 3. The fitness among different indices is also provided in

Table 2. The fitness of different indices.

2D Scatter Chart	Spectral Index Fitting Formula	Degree of Fit
SI1-GRNDVI	Linear: Y = −3.8334X + 0.9859	R2 = 0.9200
	Quadratic: Y = −34.4211X2 + 10.013X − 0.4013	R2 = 0.9332
	Exponential: Y = 10.349e−19.41X	R2 = 0.8776
	Power: Y = 0.0004X−3.854	R2 = 0.8627
	Logarithm: Y = −0.765ln(X) − 1.0135	R2 = 0.9117
SI1-GNDVI	Linear: Y = −2.682X + 1.0323	R2 = 0.9123
	Quadratic: Y = −29.197X2 + 9.0629X − 0.1445	R2 = 0.9317
	Exponential: Y = 1.4957e−5.532X	R2 = 0.9014
	Power: Y = 0.0838X − 1.102	R2 = 0.8899
	Logarithm: Y = −0.5351n(X) − 0.3659	R2 = 0.9026
SI1-NDVI	Linear: Y = −3.2938X + 1.1922	R2 = 0.9060
	Quadratic: Y = −31.881X2 + 9.5304X − 0.0927	R2 = 0.9211
	Exponential: Y = 1.8827e−6.321X	R2 = 0.8945
	Power: Y = 0.0699X−1.259	R2 = 0.8834
	Logarithm: Y = −0.657ln(X) − 0.5254	R2 = 0.8972
SI1-DVI	Linear: Y = −3.053X + 1.0464	R2 = 0.7921
	Quadratic: Y = −45.511X2 + 15.254X − 0.7879	R2 = 0.8236
	Exponential: Y = 1.18526e−7.26X	R2 = 0.7751
	Power: Y = 0.0424X−1.443	R2 = 0.7615
	Logarithm: Y = −0.6081n(X) − 0.5436	R2 = 0.7807

.

In Figure 3 and Table 2, the salinity index 1 (SI1), green–red band normalized difference vegetation index (GRNDVI), green normalized vegetation index (GNDVI), and normalized difference vegetation index (NDVI) can be seen. The SI1-GRNDVI, SI1-GNDVI, SI1-NDVI, and SI1-DVI two-dimensional scatter plots constructed by the difference environmental vegetation index (DVI) have the best fitting degrees: R² = 0.9332, R² = 0.9317, R² = 0.9211, and R² = 0.8236, respectively. The above analysis shows that the salinity index 1 and the four vegetation indexes have a very large correlation. Consequently, the RS-based salinity 1 monitoring index model can be built. Based on the four types of remote sensing salinity monitoring 1 index models for the cotton field soil salinity in the Alar reclamation area, for determining whether the accuracy of chemical inversion is good or bad, further verification is needed.

3.2. Salinity Monitoring

The RS salinity monitoring index models S1DI1, S1DI2, S1DI3, and S1DI4 were calculated using the formulas in Table 1, and the Jenk natural discontinuity point classification method was used to divide the value range of the above model into three saline soil grades [16] (

Table 3. Soil Salinization Classification of Cotton Fields in the Alar Reclamation Area.

Model	Non-Saline Soil	Mildly Saline Soil	Moderately Saline Soil
S1DI1	<0.8770	0.8770~1.0486	1.0486~1.13918
S1DI2	<0.5935	0.5935~0.7171	0.7171~0.9643
S1DI3	<0.5692	0.5692~0.7176	0.7176~1.0144
S1DI4	<0.6586	0.6586~0.8036	0.8036~1.0936

). The soil salinization distributions of the four monitoring models were obtained, as illustrated in Figure 4.

It can be seen in Figure 4 that the soil salinity in the cotton fields in the Alar reclamation area retrieved by the S1DI1, S1DI2, S1DI3, and S1DI4 models is mainly non-saline soil, while the distribution of light saline soil and moderately saline soil is slightly different. For the S1DI1-derived results, the light saline soil is mainly distributed in the southeast area of the 9th regiment and the area along the Tarim River in the 12th regiment of the reclamation area. The moderate saline soil is scattered in the Alar reclamation area based on the S1DI1 model. Among them, the central region of the 9th Regiment and the northwest region of the 10th Regiment are the most numerous and widely distributed. For the S1DI2-derived result, the light saline soil is mainly distributed in the middle and northwest of the 9th regiment and the area along the Tarim River in the 12th regiment of the reclamation area; the moderate saline soil is scattered in the Alar reclamation area. Moreover, based on the S1DI3-derived result, it can be found that the light saline soil is mainly distributed in the central area of the 9th regiment and the northeast area of the 12th regiment in the reclamation area; the moderate saline soil is scattered in the Alar reclamation area. The central part of the 9th regiment and the 12th regiment are the most widely distributed and the largest in the Taklamakan Desert. For the S1DI4-derived results, the light saline soil is mainly distributed in the central area of the 9th regiment and the area along the Tarim River in the 12th regiment of the reclamation area; the moderate saline soil is scattered in the Alar reclamation area. Among them, the central region of the 9th and 10th regiments and the 12th regiments are the most widely distributed in the Taklamakan Desert. The inversion results of the four models for the non-salinized cotton soil in the Alar reclamation area are roughly the same, but there are some differences in the inversion results for the mildly saline soil and the moderately saline soil. As a result, it is necessary to select a suitable and proper salinity monitoring index model with a higher inversion accuracy and remote sensing images.

3.3. Accuracy Assessment

Refer to the relevant RS saline soil-related research [17], the categories of different salinization levels are classified according to the soil electrical conductivity value: the salt content of non-saline soil is less than 2 mS·cm⁻¹; The salt content of light saline soil is 2~4 mS·cm⁻¹; the salt content of moderately saline soil is 4~8 mS·cm⁻¹; the salt content of severely saline soil is 8~16 mS·cm⁻¹; the salt content of extremely severe saline soil is >16 mS·cm⁻¹. As a result, this study measures the electrical conductivity of the soil sampling data based on this standard, classifies the salinization, and determines the accuracy of the RS-based salinity monitoring index models S1DI1, S1DI2, S1DI3, and S1DI4. The results are shown in

Table 4. Model accuracy.

Model	Correct Sample Classification	Sample Classification Error			Overall Accuracy %
		Non-Salt	Light	Total
S1DI1	111	3	1	4	96.52
S1DI2	110	4	1	5	95.65
S1DI3	109	4	2	6	95.78
S1DI4	109	5	1	6	95.78

and Figure 5.

It can be seen in Table 4 that, using the remote sensing salinity monitoring index model to invert the soil salinity of cotton fields in the Alar reclamation area, the S1DI1 model has the largest number of correctly classified sample points, while the S1D3 and S1DI4 models have relatively few correctly classified sample points. The overall accuracy of the S1DI1 model, S1DI2 model, S1DI3 model, and S1DI4 model is 96.52%, 95.65%, 95.78%, and 95.784%, respectively. Among the sample classification errors, the S1DI1 model, S1DI2 model, S1DI3 model, and S1DI4 model sample point classification errors were mainly caused by non-saline soil.

In Figure 5, it can be seen that the fitting degrees of the S1D1 model, S1DI2 model, S1DI3 model, and S1DI4 model are R² = 0.8335, R² = 0.8310, R² = 0.8296, and R² = 0.8025, respectively. Moreover, it can be observed that the S1DI1 model, S1DI2 model, S1DI3 model, and S1DI4 model are positively correlated with the measured conductivity.

In summary, the S1DI1 model constructed by SI1 and GRNDVI has an overall classification accuracy of 96.52% and a fitting degree of R² = 0.8335 between the model and the measured conductivity. According to the above analysis, the S1DI1 model is more suitable for extracting information on soil salinization in cotton fields in the Alar reclamation area.

3.4. Spatio-Temporal Distribution of Soil Salinization

The S1DI1 model was used to classify the soil salinization level of the reclamation area in 2019 and 2022, as shown in Figure 6. Additionally, the area of different degrees of soil salinization in the Alar reclamation area after removing the water body and sandy area was counted, as listed in

Table 5. The estimated areas of the soil salinization level in 2019 and 2022.

Particular Year	Non-Saline Soil	Mildly Saline Soil	Moderately Saline Soil
2019 year	1048.91	40.65	14.65
2022 year	986.14	113.64	4.43

.

From 2019 to 2022, the area of soil salinization in cotton fields in the reclamation area showed a decrease in the area of non-saline soil and moderately saline soil, while it showed an increase in the area of lightly saline soil. Among them, the area of non-saline soil decreased by 62.77 km², and the reduced areas were mainly distributed in the northern, southern, and eastern regions of the reclamation area, mainly shifting from non-saline soil to lightly saline soil. The area of lightly saline soil increased by 72.99 km². It can be seen that the increased area is scattered in the reclamation area. Additionally, the saline soil area of moderate saline soil decreased by 10.22 km², and the decreased area was mainly in the southern and eastern areas of the reclamation area. In the past four years, the cotton fields in the Alar reclamation area are mainly non-saline soil, but some non-saline soil and moderate saline soil have been shifted to light saline soil, and in general, soil salinization has slightly increased. Based on the analysis, we need to know that, for the area where the soil changes from non-salinization to slight salinization, it is necessary to plant suitable crops reasonably and use flood irrigation to intervene in the soil salinization of cotton fields so as to transform the soil from slight salinization to non-salinization. Areas with severe salinization turning into mild salinization should be further rationally planned to prevent the transformation from salinization to moderate salinization or even more severe salinization.

4. Discussion

(1) Model selection. Conventional soil salinity monitoring is mostly obtained from field data collection combined with laboratory measurements. Multispectral remote sensing data acquisition is simple, low-cost, and easy to operate. The remote sensing salinity monitoring index model can be constructed directly through the spectral index, which can quickly measure the soil salinity in the study area. When selecting a model, considering the differences in terrain, climate, and vegetation coverage in different geographical locations of the study area, the precision results of the remote sensing salinity monitoring index models constructed by different spectral indices are inconsistent. As a result, it is suggested that any single model should not be directly selected to invert the salinity of the reclamation area in advance, as in this case, optimal accuracy cannot be guaranteed.

(2) Model accuracy. The overall accuracy of the four types of remote sensing salinity monitoring index models was verified, and the remote sensing salinity monitoring index model constructed by the salinity index 1 and the normalized difference vegetation index of the green–red band was better for the inversion of soil salinization in the Alar reclamation area. From the comparison of the overall accuracy, it can be seen that the remote sensing salinity monitoring index models S1DI1, S1DI2, S1DI3, and S1DI4 constructed by the vegetation index are all very high. The reason may be that the remote sensing image is selected as the period of vigorous vegetation coverage, and the vegetation index can reflect the vegetation growth condition at this time. Vegetation growth is affected by different salinity contents, so it can more indirectly reflect the soil salinization status in the reclamation area. Among them, the S1DI1 model, S1DI2 model, S1DI3 model, and S1DI4 model showed a decreasing trend in the inversion effect of the cotton field saline soil in the Alar reclamation area. This is likely to be because the GRNDVI vegetation index, GNDVI vegetation index, NDVI vegetation index, and DVI vegetation index had an impact on cotton. The sensitivity of growth also shows a decreasing trend. In this paper, the remote sensing salinity monitoring index model is constructed according to different vegetation indexes and salinity index 1 (SI1). As a result, for different research areas, the vegetation index should be reasonably selected according to its own characteristics to construct the remote sensing salinity monitoring index model so as to accurately retrieve the soil salinity information.

(3) Distribution of soil salinity in the reclamation area. The soil salinization in the Alar reclamation area was retrieved through the remote sensing salinity monitoring index parameter model S1DI1, and the distribution of farmland in the inversion results was basically consistent with the previous research results [18]. The Alar Reclamation Area is a fine-soil plain alluvial by the Tarim River, resulting in relatively high salinity in the cotton field soil. The distribution of soil salinization in cotton fields in the reclamation area is affected by both human factors and natural factors. Soil moisture in cotton fields has a greater impact on soil salinity. The moisture in cotton field soil will migrate salt to the surface soil as cotton plants absorb and evaporate water. The Alar Reclamation Area is located in the arid area of southern Xinjiang, with strong sunlight and large evaporation. Most cotton fields use flood irrigation in spring or winter to reduce the soil salt return in cotton fields and drip irrigation under mulch to control the water volume and soil salinity, which greatly reduces the surface soil in cotton fields. Therefore, the cotton field in the reclamation area is mainly composed of non-saline soil [19].

The inversion accuracy of soil salinization in cotton fields is affected to a certain extent by the resolution of remote sensing image data, whether the field sampling date is consistent with the satellite transit date, and whether the weather is suitable for field sampling. In this study, the 10 m Sentine-2 RS data were selected, so the classification accuracy can only be approximate. If the inversion accuracy of cotton field soil salinization is further improved, high-resolution remote sensing image data can be used, such as Planet series, domestic high-resolution series, Quickbird satellite, etc. On the other hand, the Sentinl-2 images in this study were composited from 17 July to 22 July 2022, but the soil samples were collected in the field from 15 July to 24 July 2022. There is a certain difference between the date of remote sensing images and the date of sampling. To further improve the precision of the inversion, it is necessary to calculate whether the satellite transit date and the weather on the transit date are suitable for field sampling and arrange the sampling time reasonably so that the sampling date is within the composite date of the remote sensing image.

5. Conclusions

In this study, based on the 10 m Sentinel-2 images on the Google Earth Engine platforms, and through exploring SI1, the green–red band normalized difference vegetation index (GRNDVI), the green normalized difference vegetation index (GNDVI), the normalized difference vegetation index (NDVI), and the difference environmental vegetation index (DVI), four remote sensing salinity monitoring index models including S1DI1, S1DI2, S1DI3, and S1DI4 were built to extract information on soil salinization in cotton fields in the Alar reclamation area, combined with the measured electrical conductivity. With the conducted accuracy evaluation, we can draw the following conclusions:

(1)

The fitting degrees of the remote sensing salinity monitoring index models S1DI1, S1DI2, S1DI3, and S1DI4 and the measured conductivity are R² = 0.8335, R² = 0.8310, R² = 0.8296, and R² = 0.8025, and the overall classification accuracy of the samples is 96.52%, compared to 95.65%, 95.78%, and 95.78%. As a result, the model accuracy of cotton field soil salinization information extraction in the Alar reclamation area is S1DI1 > S1DI2 > S1DI3 > S1DI4.

(2)

Using the S1DI1 model to extract the soil salinization information of the cotton fields in the Alar reclamation area, it can be found that the cotton fields in the Alar reclamation area are dominated by non-saline soil, which is distributed in the 7th to 16th clusters of the reclamation area. The content of light saline soil occupies second place, mainly distributed in the southeast area of the 9th regiment and the area along the Tarim River of the 12th regiment in the reclamation area, and the other small parts are scattered in the cotton fields of each regiment.

(3)

This study utilizes the Sentinel-2 images on the Google Earth Engine cloud platform combined with the optimal salinity monitoring index model S1DI1 to conduct research on the monitoring and identification of salinity in cotton fields, which has the advantages of free remote sensing image data, easy access, and good performance. With the further development of optical satellite technology, using satellite remote sensing images to obtain high-precision, real-time information on the spatial distribution of salinization in cotton fields is an important trend in future agricultural development, for which some guidance can be found in this study.