Baseline-Based Soil Salinity Index (BSSI): A Novel Remote Sensing Monitoring Method of Soil Salinization

Soil salinization leads to dehydration of plants, which seriously threatens ecologically sustainable development and food security guarantee. In the complex and diverse coastal wetland environment, the impervious surface and bare soil have similar spectral features with salinized soil, which make it difficult for traditional satellite data and algorithms to accurately and timely monitor the small surface features of salinization. This article presents a baseline-based soil salinity index (BSSI) for soil salinization monitoring using medium-resolution data. In BSSI, we construct a virtual salinization baseline by connecting the near-infrared (NIR) band and the short-wave infrared-2 (SWIR2) band to enhance the spectral feature of salinized soils which border on the impervious surface. In addition, we calculate the distance between the short-wave infrared-1 (SWIR1) band and the virtual salinization baseline as the BSSI, which can effectively improve the stability of salinity inversion for different soils. Through data comparison and model simulations, BSSI has shown advantages over a series of the traditional salinization spectral indices (SSIs). The results show that the saline soil extraction accuracy of BSSI exceeds 85% and the correlation coefficient of the BSSI and the degree of soil salinization exceeds 0.90. Since the related spectral bands, such as NIR, SWIR1, and SWIR2, are available on many existing satellite sensors such as Landsat TM/ETM+, OLI, and sentinel 2, the BSSI concept can be extended to establish long-term records for soil salinization monitoring.

, [5], [6]. In addition, saline soil also has great destructive power in economic growth, ecological protection, and water and soil conservation [7]. Therefore, saline soil monitoring is of great significance to the rational use of land resources and the sustainable development of society [8], [9].
Using traditional field surveys such as ground surveys, laboratory testing to study soil salinity is very time-consuming and limited to a small area, which cannot meet the practical information needs of saline-alkali land management and regional sustainable development planning [2], [4]. The advantages of remote sensing technology are that it does not require a lot of human and material costs, and it can obtain data from areas that humans cannot reach, as well as long-term Earth observation data [10], [11], [12], [13]. Therefore, it provides an efficient detection method for dynamic monitoring of saline soil [14], [15], [16], [17], [18], [19]. Since spectral reflectance measurements can successfully simulate various soil properties, optical remote sensing data are widely used in soil salinization monitoring [20]. The features of optical remote sensing data acquisition mainly include spectral bands and various salinization spectral indices (SSIs) [14], [21]. SSIs contain more information compared to spectral bands. In addition, compared with methods such as machine learning [10], [22], [23] and deep learning [24], [25], the index method is widely recognized due to its simplicity and obvious effect [20]. SSIs also provides them with effective substratum features. Therefore, further research is done in this article in terms of SSIs.
However, salinization problems are often noticed in areas related to human activities, such as coastal parks [26], urban wetlands [25], farmland [27], etc. It is very difficult to accurately identify and extract saline soils using index methods in such areas. The normalized difference salinity index (NDSI) combined with remote sensing images was proposed to identify salinized soils in farmland areas, and realized large-scale monitoring of saline soils [28], [29]. However, because the impervious surfaces such as the bright roof of the surrounding residents have the same spectral characteristics as the saline soil, they are often mistaken for the saline soil. At the same time, the brightness index (BI) added a new spectral band on the basis of NDSI, and achieved a certain effect in the detection of salinized soil in coastal wetlands [30]. Due  impervious surfaces such as bright roof and salt ponds brings great challenges to the extraction of salinized soils.
Furthermore, the soil salinity inversion is more complicated. Soil salinity inversion is the process of obtaining the soil salinity content (SSC) using model calculations on the indices basis [31], [32], [33], [34]. In the process of inversion of salinization soil in coastal wetlands, the vegetation soil salinity index (VSSI) found that the soil spectral characteristics of different humidity types were different [35]. The same index is difficult to take into account the inversion of different types of SSC [36]. In addition, when the intensity indices (Int) were used to study the salinity of dry desert soil [36], it was found that the dry bare soils which were non-saline soils could easily be incorrectly inverted to saline soils [37]. To sum up, the current index is susceptible to impervious surfaces, bare soil and salt ponds resulting in low accuracy of saline soil extraction and inversion.
The main objective of the work was to study the common problems of salinization, and on this basis, propose a new salinity index to monitor the surface salinization status. The baselinebased soil salinity index (BSSI), was developed and detected soil salinization using the medium-resolution data. In order to effectively suppress the impact of impervious surface and enhance the salinity characteristics of salt pools, a virtual salinization baseline was constructed in BSSI. The baseline can effectively integrate the spectral information of the near-infrared (NIR), the short-wave infrared-1 (SWIR-1), and the short-wave infrared-2 (SWIR-2), enhance the characteristics of surface salinity, and solve the problem of salt soil extraction. Moreover, in order to effectively identify the salinity of different types of soils, this article defines BSSI by calculating the differential value (D-value) between the reflectivity of SWIR-1 of the ground object and the virtual salinization baseline. The D-value effectively overcomes the spectral difference seen in salinized soils and has strong stability. Thus, it can be seen that the BSSI provides an excellent underlying feature for future salinity inversion models based on deep learning and machine learning. In addition, the huge computational volume and complex scenarios will no longer be an obstacle for BSSI in large area and long time series salinity monitoring tasks, and the simplified inversion model based on BSSI will break through this challenge.
The other parts of this article are respectively indicated as follows. The second part focuses on the study area and data, the methodology constructed in this article and the validation method. The experimental results are analyzed and discussed in Sections III and IV. Finally, the conclusion of this article is highlighted in Section V.

A. Study Area and Data
1) Study Area: Soil salinization is mainly distributed in arid areas [38] and coastal plain areas [33]. Therefore, this article selects the yellow river delta (YRD) which belongs to the coastal plain area and the southern margin of the Tarim basin (STB) which belongs to the arid area to test the effectiveness of BSSI, as shown in Fig. 1.
The climate of the region is temperate subhumid continental monsoon [39]. The YRD is located in the Tertiary period of the long-term sedimentation area of the crust, and it is mainly composed of silt, dark gray clay and silty clay. The northern and eastern tidal flat soils of this area were highly or moderately salinized, and the river highlands and parts of the plain soils were low or non-salinized [40]. The areas with severe salinization are mostly wasteland, and the areas with good desalting have become agricultural land.
The STB is a typical area in Xinjiang, which belongs to dry saline soil [see Fig. 1(b)]. The geographic coordinates are 36°50 30.34 ∼36°58 4.45 N, 81°32 5.717 ∼81°40 29.86 E and the land area is about 182 km 2 . The climate is a warm temperate continental arid climate. The STB is mainly Precambrian sedimentary strata, which mainly contain a large amount of metamorphic greywacke, metamorphic calcareous sandstone, and fine-grained schists. With the continuous increase of the surface soil salinity, the surface reflectance increases and the vegetation coverage decreases. In severe cases, salt crust will form, and the vegetation coverage will be zero, forming a severe saline-alkali area.
2) Landsat Data: This article uses Landsat ETM+ data provided by the United States Geological Survey (http://glovis.usgs. gov). For the YRD, we choose the data on May 3, 2006, and for the STB, we choose the data on September 2, 2001. To ensure the quality of image data, the selected data is visually clear and the cloud amount is less than 5%. The software ENVI 5.3 [41] was used to perform radiometric calibration, atmospheric correction and clipping operations on the image [2], [13].
3) Soil Sample Data: To accurately apply soil samples to establish the relationship between SSC and spectral reflectance, the field collection time of soil samples is required to be consistent with the imaging time of remote sensing satellites. The influence of seasonal changes and vegetation growth on the research is avoided. Therefore, on May 3, 2006, this article selected 70 samples from the soil samples in the YRD research area (http://www.geodata.cn/).
In addition, to further verify the capacity of extrapolation of the method, we also obtained field-measured data on the STB in Xinjiang [42], [43]. After excluding outliers, this article selects 31 samples from the study area on September 1, 2001. The depth of each sampling point was 0-30 cm and a total of five repetitions were set for each sampling point [1], [10], [13], [44].
The grading standard of electrical conductivity (EC) provided by the U.S. Salinity Laboratory [45] and the grading standard of SSC are given in Table I [46], [47]. The soil in the study area is divided into five grades: extremely saline soils, strongly saline soils, moderately saline soils, slightly saline soils, and non-saline soils [10].

B. Construction of BSSI 1) Extraction Analysis of Saline Soil:
The salinized area has a complex topography and a wide variety of ground features, which interferes with the extraction of saline soil. In order to extract saline soil, we need to select bands sensitive to saline soil information from the multispectral image. Therefore, we used a large number of pure samples of land cover types. We selected five types of land that are easily confused by salinized soil, including water, tidal flat, vegetation, impervious surfaces with high albedo (IS_High), and impervious surfaces with low albedo (IS_Low). Their average spectral profile is illustrated in Fig. 2.
From the arrows at the SWIR1 band (see Fig. 2), it can be seen that the virtual salinization baseline between NIR and SWIR2 band has obvious differences among the spectral values of the six types of ground objects. Compared with water and tidal flat, salinized soil has an obvious peak value in the SWIR1 band and other types have a valley value in the SWIR1 band. The spectral features of vegetation, IS-High and IS-Low do not vary significantly. In summary, the influence of several other types of features can be effectively suppressed by constructing virtual salinization baselines to achieve accurate extraction of salinized soils.
2) Characterization Analysis of the Saline Soil Degree: In order to accurately distinguish the degree of soil salinization, we need to look for spectral differences between different salinization levels. To achieve this goal, we selected extremely saline soils, strongly saline soils, moderately saline soils, slightly saline soils, and nonsaline soils from soil samples and obtained their average spectral profiles, as shown in Fig. 3.
As can be seen from arrows 1-5 in Fig. 3, the distance between SWIR1 and the baseline gradually increased with the increase in the degree of salinization. This phenomenon can provide a new monitoring idea for the monitoring of salinization, that is, the degree of salinization can be obtained by calculating the distance between SWIR1 and the baseline. In summary, this distance cannot only effectively distinguish salinized soil  from other ground objects, but also has unique advantages in monitoring the degree of salinization.
3) Formulation of BSSI: The difference distance from the baseline to the SWIR1 band provides effective help for the extraction of salinized soil and becomes a powerful weapon for monitoring the degree of salinization. We named the baseline as virtual salinization baseline, and this distance as BSSI. Combining the above ideas, we obtained the final BSSI formula. Fig. 4 shows the construction principle of BSSI, which is divided into two steps: construct the virtual salinization baseline; and calculate the distance between the SWIR1 and virtual salinization baseline.
The baseline BSSI is defined as where λ SWIR1 and λ SWIR2 are the spectral bands with the wavelength of 0.835, 1.648, and 2.206 um separately.

C. BSSI-Based Salinization Degree Assessment
We obtained the soil salinity thematic map by establishing a linear regression model between SSIs and soil salinity for inversion by using the SPSS 22.0 package [48], [49]. The operation process is shown in Fig. 5. First of all, we perform preprocessing operations, such as banding processing and radiometric calibration atmospheric correction on the Landsat 7 optical data to obtain basic data. BI, NDSI, SI and other indices were obtained from the basic data according to the calculation formula of the salinity index. Second, we established the linear relationship between the soil sample salinity data and the index to obtain a linear regression model. Third, by calculating the coefficient of determination of these linear models, the regression model with the best performance is selected. Finally, we divided the soil salinization degree of the study area into five grades according to the grading standard, and finally obtained the salinization grade map of this area.

D. Verification Method
In order to demonstrate the effectiveness of our approach, we conducted two sets of experiments with two sets of data separately. The first one is devised to verify the validity of the index extracted in the coastal plain area. The second one is to study the effectiveness of the index in arid and semi-arid regions. We selected a total of nine well-known SSIs, including SI, SI1, SI2, SI3, BI, VSSI, NDSI, Int1, and Int2, for comparison with our method. The details of these SSIs are apparent in Table II. We consider different types of soil objects to ensure sample validity.
where TP is an amount of true saline soil classed as saline soil, FP is an amount of true saline soil classed as nonsaline soil, TN is an amount of nonsaline soil classed as nonsaline soil and FN is an amount of nonsaline soil classed as saline soil, and TS is the whole amount of true saline soil in the reference maps.

2) Verification Method for Saline Soil Classification:
To better understand the effectiveness of different SSIs in reflecting true salinity values, Pearson correlation coefficients [49] (r) were used to analyze the correlation between these parameters and measured salinity. The formula is as follows: where x i is the SSI value, y i is the measured value of SSC,X is the average of the SSI, andȳ is the average value of measured SSC.
3) Verification Method of Saline Soil Mapping: To obtain the soil salinization status of the study area, we established a linear relationship between the index value and SSC. The fit of the linear model is judged by calculating its coefficient of determination (R2) [49]. The three concepts involved here are the regression sum of squares (SSR), the total sum of squares (SST), and the error sum of squares (SSE). The formulas are as follows: where y i is the measured value of SSC,ŷ i is the estimated value calculated by the model, andȳ is the average value of measured SSC.

1) Accuracy Assessment of Saline Soil Extraction Results:
We implement two experiments on the YDR dataset and the STB dataset, respectively, to highlight the effectiveness of our method. In order to obtain the optimal extraction result of salinized soil, we obtain the threshold of saline soil through trial and error. The optimum threshold (see Fig. 6) and the correctness of the salinity soil extraction for each SSI are given in Table III. As shown in Fig. 6 and Table III, we chose the segmentation threshold with the highest OA for each index to ensure the fairness of the SSIs comparison. For the YRD dataset (see Table III), the BI, Int1, Int2, SI, SI1, SI2, and SI3 had the bottommost OA and Kappa coefficients with the threshold around 0.4. Their E C or E O were higher. An omitted error indicates that some of the saline soil had been missed. Commission error denotes that non-saline soil surfaces, such as impervious surfaces with high albedo (IS_High), and impervious surfaces with low albedo (IS_Low), are wrongly labeled as saline soil. NDSI and VSSI had higher precision. Their E O was higher. Obviously, the reason for the lower extraction accuracy of saline soils is their weak resistance to external factors. On the contrary, BSSI had the highest extraction accuracy for saline soil, and BSSI had the lowest E O . Therefore, it showed its advantages in saline soil extraction.
For the STB dataset in Table III, the Int1, SI1, SI3, and BSSI had higher accuracy with the threshold at 0.14. But only BSSI was equally accurate in the YRD. Therefore, we can conclude that BSSI can accurately extract saline soil. In addition, BSSI is widely used in different regions, not only for saline soils in coastal plains but also for arid and semiarid saline soils.
2) Comparison of Saline Soil Extraction Results: In order to intuitively obtain the advantages of BSSI, we calculated the saline soil results of SSIs separately based on the YRD dataset. The details of the YRD dataset are described in Fig. 7(a). Fig. 7(b)-(k) shows the extraction results of saline soil with different indices. For the convenience of observation, we use "1-NDSI" to display the value of NDSI in reverse.
As shown in Fig. 7(a), we select: i) salt ponds with high salinity content and (ii) river banks where saline soils are easily distinguishable from nonsaline soils (ii). In (c), (e), (f) and (h) of Fig. 7(i), Int1, SI, SI1 and SI3 did not accurately extract the      Fig. 7 (ii), in (i), (j) and (k), the NDSI VSSI and BSSI accurately classify the rivers as nonsaline soils, and the other indices erroneously extract the rivers as salty soils. In addition, among the three indices, only the BSSI accurately extracts the non-saline soil along the river.
Similarly, we analyzed the saline soil extraction results of each SSI in the images of the STB dataset [see Fig. 8(a)]. Among them, we select the area of the transition area from the saline land to the river (in the dotted frame, the area is saline land, farmland, and river in order of arrows). Fig. 8(b)-(k) shows the saline soil extraction results for the ten SSIs.
As illustrated in Fig. 8(g), (i), and (j) in the square dashed box, SI2, NDSI and VSSI misclassify the river as saline soil. In Fig. 8(b), (c), (e), (f), and (h), BI, Int1, SI, SI1 and SI3 misclassified some rivers as saline soil. From Fig. 8(d) and (k), it can be found that Int2 and BSSI can better extract the river into nonsaline soil. Among them, BSSI has the best effect, which is in line with the actual situation of changing from saline-alkali land to river in this area.
In summary, BSSI performs very well in the extraction of saline soil.

B. Validity of Saline Soil Classification 1) Accuracy Assessment of Saline Soil Classification:
The Pearson correlation coefficient between the significant spectral data and soil parameters of the YRD dataset and the STB dataset are exhibited in Tables IV and V. For the YRD in Table IV,   inversion, but there are errors. Among them, BSSI has the highest correlation with salinity and can be better used for soil salinization inversion.
For the STB in Table V, the SSIs of BI(r−0.691), Int2 (r = 0.377), and SI2 (r = −0.513) have low correlations with SSC, indicating that the indices cannot be used for soil salinization inversion. BSSI(r = 0.963), Int1(r = 0.921), NDSI(r = 0.963), SI(r = 0.920), SI1(r = 0.921), SI2(r = 0.921), and SI3 (r = −0.932) have a high correlation with salinity, indicating that these indices can be used for soil salinization and salinity inversion, as shown in Table VI. In summary, BSSI has the highest correlation in the two research fields, so it is verified that it can best distinguish the degree of salinization.
2) Comparison Classification Results of Saline Soil: We further verified the effectiveness of BSSI in distinguishing soil salinization degree by analyzing the display effect of each SSI on the YRD dataset. The details of the YRD dataset are shown in Fig. 9(a). The inversion status of soil salinization degree by different indicators is shown in Fig. 9(b)-(k). In order to facilitate observation, we use the "1-NDSI" inversion to display the NDSI values.
As shown in Fig. 9(a), we select: (i) salt ponds and (ii) chemical factory with high salt content for analysis. Salt ponds are mainly engaged in the salt making industry, and a large amount of sea salt collects on the surface, hence the high salt content. The chemical factory is mainly engaged in industrial production, lack ecological protection, and the gradual intrusion of seawater into the interior, eventually leading to an increase in soil salinity.
First, BI, Int1, Int2, SI, SI1, SI2 and SI3 show different accurate soil salinity in Fig. 9(b)-(h). Areas (i) and (ii) have severe soil salinization, while the figure shows mild salinization. As shown in Fig. 9(i) and (j), NDSI and VSSI can represent the  degree of soil salinization in this area to a certain extent, but there are also errors that the degree of salinization along the river banks is too serious. Second, the degree of salinization of the ocean is obviously different from that of the land, so the boundary between the ocean and the land can be well obtained. However, only NDSI, VSSI, and BSSI have distinct boundaries in Fig. 9. Thus, the degree of soil salinization characterized by BSSI is the most accurate in Fig. 9(k).
Similarly, we also tested the dataset on the STB [see Fig. 10(a)]. Fig. 10(b)-(k) shows the inversion status of soil salinization degree by ten SSIs.
The dotted boxes in the figure are saline-alkali land, riverside cultivated land, and rivers in the direction indicated by the arrows, and the corresponding soil salinization degrees are severe salinization, mild salinization, and nonsalinization. The soil salinization degree of BI, Int2, and SI2 in the dotted box is relatively light, as shown in Fig. 10(b), (d), and (g). From Fig. 10(c), (e), (f) and (h), it can be seen that Int2, SI, SI1 and SI3 are in the dashed box according to the direction of the arrow in the direction of the corresponding soil salinization degree are slightly salinized salinization, nonsalinization, mild salinization. In the dashed box, the soil salinization degree corresponding to the direction indicated by the arrows in the NDSI is in the order of severe salinization, non-salinization, and severe salinization [see Fig. 10(i)]. In the dashed box, the VSSI is in the order of non-salinization, mild salinization, and nonsalinization in the direction indicated by the arrow [see Fig. 10(j)]. Only the change of soil salinization degree within the dashed box of BSSI is consistent with the actual situation [see Fig. 10(k)].

1) Assessing Soil Salinity With Salinity Index:
The coefficients of determination (R2) between models and ground-truth salinity experimental result on the YRD dataset and the STB dataset are given in Table IV. For the YRD in Table IV, the linear regression models of NDSI and VSSI both have a poor fitting degree and cannot characterize soil salinization degree well. The linear regression models of BI, Int1, Int2, SI, SI1, SI2, SI3, and BSSI all have a high degree of fitting and can better characterize the degree of soil salinization. Among them, the linear regression model of BSSI has the highest fitting degree and can best characterize the degree of soil salinization.
For the STB in Table IV, the linear regression models of BI, Int2, and SI2 all have a poor fitting degree and cannot characterize soil salinization degree well. The linear regression models of Int1, NDSI, SI, SI1, SI3, VSSI, and BSSI all have a high degree of fitting and can better characterize the degree of soil salinization. Among them, the linear regression model of BSSI has the highest fitting degree and can best characterize the degree of soil salinization. In summary, BSSI has the highest salinity inversion effect of soil salinization in both research fields, so it is verified that it can best distinguish the degree of salinization.
2) Spatial Distribution Analysis of Regional Soil Salinity: We performed soil salinity inversion in the Yangtze river delta region, based on the BSSI soil salinity linear model. The SSC in this area was between 0 and 13.3%, and the area with SSC between 0 and 2% accounted for the majority, which was consistent with the descriptive statistics of the research samples. According to the classification standard of total soil surface salinity in China, we obtained the thematic map of soil salinity in the YRD in Fig. 11(c).
Similarly, based on the BSSI linear model of soil salinity, we performed field inversion of soil salinity in the STB in Xinjiang. The SSC in this area was between 0 and 3.1%, and the area with SSC between 0 and 2% accounted for the majority, which was consistent with the descriptive statistics of the research samples. According to the classification standard of topsoil salt content in China, we used Landsat 7 ETM+ to obtain the spatial distribution map of soil salt content in the STB, as described in Fig. 11(d).

A. Applicability Analysis of Baseline Method in Soil Salinity
The monitoring of soil salinization can timely understand soil changes, and it is convenient to formulate reasonable measures, including applying chemical amendments and plant salt tolerant plants, to slow down the process of soil salinization [2]. However, the monitoring of saline soil has the difficulty in extracting saline soil from complex surface and distinguishing the grade of soil salinity.
Existing SSIs structures use sensitive bands for ratio calculations, which are sensitive to changes in environmental and observational conditions. Baseline method can enhance spectral difference and reduce environmental interference. In the past studies, Hu effectively inhibited the influence of thin clouds on the extraction of floating algae by constructing a baseline [51]. On the basis of Hu, Xing et al. used HJ-1 and Landsat data to map blooms in the Yellow Sea and East China Sea [52]. In conclusion, the baseline could be used to suppress objects whose spectral changes are gentle. In this article, it was found through research that the spectral curves of vegetation and impervious surfaces, which are likely to interfere with the extraction of saline soil, in Fig. 2 change gently from NIR to SWIR2. This provided solid data assurance for the application of the baseline method. The experimental results in Table III further prove that the baseline method play an effective role in the study of soil salinity, and the extraction accuracy of saline soils is about 10% higher than the comparison method.

B. Capacity of Extrapolation Analysis of the BSSI
Soil salinization is widely distributed, mainly in arid areas and coastal plain areas. Obtaining a universal spectral index showing satisfactory results under all environmental conditions is important for the monitoring of soil salinization [2]. Although both vegetation spectral indices and salinity spectral indices have shown satisfactory results in salinity monitoring worldwide, they are both targeted at specific environments [53].
In this article, the lithology of YRD is mainly sandstone, with more sandy surface and large voids. In addition, the salt seepage out of the surface crystallizes severely after the water evaporates due to the severe seawater erosion. The lithology of STB is mainly composed of metamorphic greywacke, metamorphic calcareous sandstone and fine schist. The weathering of parent material and the hydrogeological setting are produced a lot of saline. A large amount of salt is retained on the surface due to unreasonable irrigation, which eventually leads to soil salinization.
Although they have different genesis and salts, the formed salts precipitate out of the surface and are detected by satellites. For the commonality of their spectral features, we propose a virtual salinization baseline based on the spectral characteristics of saline soils and surface objects that are easily confused with them, in order to achieve the applicability of BSSI in different environments. This provides a basis for salinity detection on a national and global scale.
In addition, BSSI consists of NIR, SWIR1, and SWIR2 bands. Landsat TM/ETM+, OLI, and sentinel 2 all contain these three bands and their wavelength bands have similar wavelength ranges [50]. Therefore, BSSI can be extended to these satellite sensors to enable long time series monitoring of soil salinity.

V. CONCLUSION
In this article, we proposed the BSSI to extract salinized soil and characterize the degree of salinization. The index constructs a virtual salinization baseline that can enhance the spectral characteristics of salinized soil and suppress the influence of impervious surface and bare soil. The main conclusion described as follows.
1) The BSSI can accuracy extract saline soils. By comparing BSSI with others SSIs, NDSI, BI, SI, etc., we found that BSSI has the highest SSC extraction accuracy, exceed-ing85% in both coastal plains (YDR) and arid regions (STB, Xinjiang). 2) The BSSI can significantly characterize the grade of soil salinity and is suitable for different environments. The D-value of BSSI can effectively improve the stability of the soil salinity inversion. The correlation coefficient between BSSI and soil salinization degree exceeds 0.90 in both the YDR and the STB.
3) The BSSI method can be easily used for soil salinization inversion. The inversion model of BSSI has the highest fitting degree and can best characterize the degree of soil salinization. The coefficients of determination is exceeds 80% in both the YDR and the STB. Note that the feasibility of BSSI for deep soil salinization is limited due to optical remote sensing cannot penetrate the ground and only acquire information of surface features. In the future, BSSI will try to study deep soil salinization in combination with temporal changes in vegetation.