Evaluating soil salinity dynamics under drip irrigation in the Manas River Basin, Xinjiang: a long-term analysis (1996–2019)

Study area

The Manas River Basin is located in the hinterland of the Eurasian continent, on the edge of the Gurbantungut Desert, the largest fixed and semi-fixed desert in China (84°43′–86°35′E, 43°21′–45°20′N) (Yang et al., 2020). The Xiayedi irrigation area, Anjihai irrigation area, Jingouhe irrigation area, Shihezi irrigation area, Xinhuchang irrigation area, Mosuowan irrigation area, and Manas River irrigation area are the seven irrigation regions in the basin. From south to north, the basin’s primary geomorphologic features—typical mountainous oasis basin ecosystems—include mountains, piedmont folds, alluvial fans, plains, dry deltas, and deserts. The average temperature in the basin was 6.5 °C. The highest temperature in the year was in July, and the lowest was in January. The frost-free period was 155 d. The sunshine duration was 2745 h. The annual precipitation was 125.0–207.7 mm, and the annual evaporation was 1,000–1,500 mm (2000–2017).

Soil sample collection and treatment

Field sampling and investigation were conducted at the Manas River Basin Oasis in October 2019. The sampling points were chosen based on their homogeneity, representativeness, salinization state, and surface characteristics in the study area. According to the five-point sampling approach, soil samples were collected in the 30 m by 30 m quadrat at 0 to 20 cm depth. Three hundred fifty effective sampling points were produced (Fig. 2) after precisely determined each sampling point’s longitude and latitude. Three hundred fifty soil samples were air-dried, crushed, and sieved through a 2 mm-aperture soil before being uniformly combined. A total of 200 g soil samples were collected using the quartering method. The electrical conductivity was measured, and soil salinity (SSC, g/kg) was determined using an empirical formula (Pouladi et al., 2019). As shown in Eq. (1):

(1) $$\mathrm{S}\mathrm{S}\mathrm{C}=\left(2.16\mathrm{E}{\mathrm{C}}_{5:1}+2.45\right).$$

In Eq. (1), EC5:1 is the soil conductivity measured by soil water ratio 1:5.

The soil salinization in the study area is divided into five grades. The grading standard of soil salinization used in this study is shown in Table 1.

Remote sensing data acquisition

This study used six distinct years’ worth of NASA’s Landsat series remote sensing photos, considering sampling time and remote sensing image quality.

Geometric correction

Remote sensing image deformation is generally caused by systematic and non-systematic two reasons (Roth et al., 2018). The influence of latitude, height, and external environment of different sensor platforms is called systematic influence; remote sensing images with uneven rows, different pixel sizes, and irregular shapes are called non-systematic effects .Geometric correction is usually used for non-systematic. The calibrated remote sensing image is chosen for geometric correction in this work. The ENVI 5.1 toolbox is used to choose the control points. The notion of prominent and simple-to-identify sites, such as typical buildings, river bends, road intersections, and so forth, is followed in selecting control points. The polynomial algorithm geometrically correct the remote sensing images after uniformly distributing as many control points as possible.

Image clipping

Cutting remote sensing images according to the boundary of the study area can effectively reduce the amount of data and improve the operation speed. In this study, ENVI 5.1 is used to cut the boundary of the study area, and the remote sensing image of the study area in 2019 is obtained.

Radiometric calibration

The foundation for achieving remote sensing information quantification is radiometric calibration, which involves converting distinct remote sensing images into radiance values (DN) using an atmospheric radiative transfer model (Xu et al., 2019). Based on ENVI 5.1, the visible-near-infrared data is chosen for this work, and the calibration type is radiation brightness value. The output type is float, while the storage form is BIL. Radiation brightness has a unit conversion coefficient of 0.1. The radiometric calibration of the photos from remote sensing is finally finished.

Flash atmospheric correction

The true reflectance of ground objects is determined by sensors using atmospheric correction which simultaneously supports a range of sensor types and has high accuracy and robust autonomy. Based on ENVI 5.1, this study performs atmospheric correction using radiometric calibration data, determines the actual reflectance of ground objects, and completes the Flash atmospheric correction of remote sensing photos.

Supervision classification

Remote sensing technology plays a vital role in land-use and land-cover classification, enabling spatial information extraction for environmental monitoring and resource management. Supervised classification is widely used among various classification techniques because it can categorize unknown pixels based on known training samples (Luo, 2019). This study employed supervised classification to analyze remote sensing images and generate land-use information for the Manas River Basin. Using ENVI 5.1 software, training areas representing distinct land-use types were selected to classify satellite imagery, ensuring accurate classification results. The generated thematic map provides valuable insights into the spatial distribution of oasis land-use types, contributing to a better understanding of land-use dynamics in the study area.

Calculation of remote sensing index

Nine multispectral remote sensing indicators closely connected to soil salinity were chosen for this study. Salinity index (SI, SI₁, SI₂, SI₃), ratio vegetation index (RVI), soil adjusted vegetation index (SAVI), normalized difference water index (NDWI), normalized difference vegetation index (NDVI), difference vegetation index (DVI), normalized difference vegetation index (NDVI), (Meivel & Maheswari, 2022). Calculations are made using ENVI 5.1’s Band math toolkit. In Table 2, the computation formula is displayed.

Data analysis and validation

In this study, 350 soil samples were randomly divided into two groups using the sampling data from 2019. One group of 280 samples was used for modeling, and the other 70 samples were used for verification.

The weighted average approach is utilized to give a bigger weight to the model with high verification accuracy and a lower weight to the model with low verification accuracy. The variable weight combination prediction method is chosen. Calculation approach as demonstrated by Eqs. (2) and (3):

(2) $$W_i={\displaystyle \frac{\sigma -{\sigma }_i}{\sigma }\times {\displaystyle \frac{1}{n-1}}}$$

(3) $$\sigma =\sum _{i=1}^n{\sigma }_{i}i=1,2,\dots n.$$

In the formula, W_i is the weight of model i, σ_i is the standard deviation of the prediction error of model i, and n is the number of single models.

The determination coefficient (R²) and root mean square error (RMSE) are the evaluation indices of the model’s stability and prediction accuracy. Strong prediction ability and great stability are characteristics of the model with a large determination coefficient and a modest RMSE. As shown by each test phrase in Eqs. (4) and (5),

(4) $${\mathrm{R}}^2={\displaystyle \frac{{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{X}}_{\mathrm{i}}-\overline{\mathrm{X}}\times ({\mathrm{Y}}_{\mathrm{i}}-\overline{\mathrm{Y}})}{\sqrt{{\sum }_{\mathrm{i}=1}^{\mathrm{n}}({\mathrm{X}}_{\mathrm{i}}-\overline{\mathrm{X}}{)}^2\times {\sum }_{\mathrm{i}=1}^{\mathrm{n}}({\mathrm{Y}}_{\mathrm{i}}-\overline{\mathrm{Y}}{)}^2}}}$$

(5) $$\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}=\sqrt{{\displaystyle \frac{{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{Y}}_{\mathrm{i}}^0-{{\mathrm{Y}}_{\mathrm{i}}^{\mathrm{s}}}^2}{\mathrm{n}}}}$$where: n is the number of sampling points; X_i is the measured value of soil sample I; $\overline{\mathrm{X}}$ is the average of measured values of soil samples; Y_i is the predicted value of soil sample i; $\overline{\mathrm{Y}}$ is the average value of soil sample prediction; ${\mathrm{Y}}_{\mathrm{i}}^0$ and ${\mathrm{Y}}_{\mathrm{i}}^{\mathrm{s}}$ are the predicted and real values of soil samples (Chicco, Warrens & Jurman, 2021).

Soil salinity inversion in the Manas River Basin Oasis, 1996–2019

The spatial distribution of soil surface salinization in six periods from 1996 to 2019 (Fig.3).The majority of the surface soil in the entire basin had significant salinization and saline soil in 1996, and the level of salinization was severe, as can be seen from the map. Following the initial application of large-area drip irrigation in 2002, soil salinization in the study region was much lower than it was in 1996. As of 2008, to accomplish complete drip irrigation, soil salinization had been further facilitated, and soil salinity had considerably lowered after 12 years of film drip irrigation technology combined with acceptable fertilization and other enhanced saline-alkali soil technology. In 2012, there was a clear upward trend in the salt content of surface soil, and from that year until 2019, there was a clear decrease trend.

Descriptive statistical analysis of soil salt content in different years

It is clear from Table 2 that the soil salt content in 1996 and 2002 was much higher than that in previous years. This is based on the average and standard deviation of soil salt content across various years. In 2016 and 2019, the soil salinity showed significant variance, with coefficients of variation of 128.64% and 127.92%, respectively. In 1996, 2002, 2008, and 2012, the coefficients of variation of soil salinity were, reasonably modest and corresponded to the range of medium fluctuation.

Figure 4 shows how drastically varied the distribution of soil salt is between various years. In 2008, 2012, 2016, and 2019, there was a noticeable skewed distribution of soil salinity to the left. In 2008 and 2019, the surface soil salinity had a higher skewness coefficient, and the kurtosis values were 3.36 and 2.86, respectively, deviating from the normal distribution. The variation in soil salinity in 1996, 2002, 2012, and 2016 was non-significant, and the kurtosis values, which complied with the normal distribution, were 0.29, 0.46, 0.76, and 0.39, respectively. The soil salt content from 1996 to 2019 was tested using the K–S normal test, and the results (p ≤ 0.05) showed that the soil salt level in 1996 and 2002 followed the normal distribution, while the soil salt content in 2008, 2012, 2016, and 2019 followed the logarithmic normal distribution.

Spatial-temporal evolution of soil salinity in Manas River Basin

According to the statistical findings, between 1996 and 2019, when drip irrigation under mulch was first introduced, the area of non-salinized soil increased by 1,403.46 km², the area of mildly salinized soil increased by 3,702.28 km², and the area of saline soil decreased by 7,685.6 km² in the study area . The transition of saline soil into severe and moderate salinization, and moderate and severe salinization into mild and non-saline soil, intensified salinization. When drip irrigation under mulch first gained popularity, the study area’s saline soil gradually changed to moderately and severely salinized soil. As a result, the salinization phenomenon was significantly reduced, but the moderately and severely salinized area increased from 1996 to 2019. After mulched drip irrigation became widely used, mild soil salinization was achieved using of flawless management technology. However, a significant amount of soil in 2019 is still moderately or severely salinized, and possible dangers could limit agricultural development and sustainable land use. Soil salinization management needs to be strengthened to secure future economic growth and create a healthy ecological environment.

Trend analysis of soil salinization

This article uses ArcGIS 10.2 software to calculate the difference in salt content between 1996 and 2019 to further explore the spatial evolution of the oasis in the Manas River Basin over the past 23 years. The evolution results are divided into five categories: noticeable improvement, stability, deterioration, and apparent deterioration. The specifics of the research area’s results for the evolution of soil salinization are displayed in the statistics table. Overall, there was a noticeable improvement in soil salinization from 1996 to 2019; the soil area with a substantial improvement in salinization reached 7,282.28 km², or 78.02% of the research region’s total area. The area where soil salinization improved was 238.07 km², making up 3.62% of the study area’s total area; the area where soil salinization remained stable was 620 km², making up 6.64% of the study area’s total area; and the area where soil salinization declined was 151 km², making up 1.62% of the study area’s total area. The area of the study region’s soil that had deteriorated due to salinization was 942.10 km², or 10.09% of the total area.

The distribution map of soil salinity evolution in the study area from 1996 to 2019 is shown in Fig. 5. The Shihezi irrigation area, Manas irrigation area, and Mosuowan irrigation area all exhibit more pronounced deterioration tendencies than the study area. Overall, the study area’s soil salinization has been significantly reduced; some saline soil has transformed into severe and moderate salinization soil, and some moderate and severe salinization soil has transformed into mild and non-salinized soil, indicating that, over the past 23 years, soil salinization has only intensified in a small portion of the study area and is still improving overall. In addition to improving soil quality and speeding up the process of self-healing, perennial drip irrigation conditions have helped to minimize the creation of soil salinization to some level.

Correlation analysis of spectral indices with soil salt concentration

The correlation analysis confirms the relationship between spectral indices and soil salt concentration (SSC), with SI, SI1, and SI3 positively correlated with salinity, while NDVI and SAVI show strong negative correlations (Table 3). These findings validate the selected indices’ effectiveness in detecting soil salinity patterns. The correlation results are presented in the Results section to align with their analytical (Ning et al., 2021; Qi et al., 2022).