Temporal and Spatial Characteristics of Soil Salinization and Its Impact on Cultivated Land Productivity in the BOHAI Rim Region

Abstract

Soil salinization can have an inhibitory effect on crop productivity and pose a serious threat to regional agricultural resource utilization and sustainable development. However, there are scarce studies on the quantitative impact of soil salinization on crop productivity. The purpose of this study is to identify the spatial and temporal characteristics of the cultivated land, the soil salinization degree, and the net primary productivity (NPP) of vegetation in the Shandong region around the Bohai Sea and to couple the quantitative relationship between soil salinization and the productivity of the cultivated land. We found that the amount of cultivated land in the study area gradually decreased from 2001 to 2019. The degree of soil salinization in the cultivated land in the north of the study area increased slightly from 2001 to 2005 and decreased continuously after 2011. The NPP value of the cultivated land increased, fluctuating from 2001 to 2019. The spatial distribution of the NPP value was generally lower in the north than in the south, with Dongying District as the boundary. In addition, in different soil salt content (g/kg) intervals, the change trends in cultivated land NPP varied. When the salt content of cultivated soil was in the range of 0–3 g/kg, 3–7 g/kg, or 7–15 g/kg, for every 1 g/kg increase in salt content, the NPP value decreased by 43.62 gC m⁻²a⁻¹, 30.36 gC m⁻²a⁻¹, and 44.19 gC m⁻²a⁻¹, respectively. These results are expected to provide decision-making support for strengthening the dynamic monitoring and regulation of cultivated land salinization and promoting sustainable utilization of salinized cultivated land in the coastal areas to alleviate the food crisis.

1. Introduction

Soil salinization is a form of land degradation caused by natural or human activities [1]. About 8.31 × 10⁸ hm² of soil in the world is threatened by salinization [2]. With the continuous impact of global climate change and human activities, the land area affected by salt is expanding [3]. At the same time, the world has been on the verge of a serious food crisis for the past 50 years [4]. In 2021, the number of hungry people resulting from the worldwide food shortage reached 828 million, 46 million more than that in 2020 [5]. Under the influence of the global COVID-19 pandemic and the locust plague in Africa, the number of hungry people in the world may further increase[6,7]. The total area of salinized soil with utilization potential in China is about 3.6 × 10⁷ hm², accounting for 4.88% of the nationally available land area [8]. This has seriously affected our ability to realize the goal of increasing grain production, farmers’ income, and the sustainable utilization of resources throughout the country. Therefore, the comprehensive utilization of salinized land is of great strategic significance to ensure national food security. The low plain area around the Bohai Sea has been the main battlefield for the development and utilization of salinized land in China in recent years, and it is also the focus of social and academic attention [9,10]. For two consecutive years, from 2015 to 2016, the “No. 1 Central Document” proposed the “implementation of scientific and technological projects for high grain yield and scientific and technological demonstration of saline alkali land transformation” and to “implement the scientific and technological demonstration project of the Bohai granary, increase scientific and technological support, and accelerate the transformation of saline alkali land” [11]. However, due to the shortage of freshwater resources, less precipitation with uneven temporal and spatial distribution, frequent and abnormal droughts and floods caused by climate anomalies, poor saline–alkali land, and urbanization occupation [11], the amount and quality of the cultivated land in the Bohai Rim region have changed greatly in temporal and spatial distribution. In addition, the stability and sustainability of cultivated land productivity have been seriously threatened. It is an urgent task to determine the spatial and temporal change characteristics of cultivated land in salinized areas, the changing trend in the productivity of salinized cultivated land, and the quantitative relationship between the degree of salinization and productivity of cultivated land. This is of great significance for the sustainable development of coastal agriculture, food security, and achieving the goal of zero growth in land degradation [12].

The traditional dynamic monitoring methods for soil salinization mostly use data on soil salt content obtained with field sampling and laboratory testing. Although this method has high accuracy in obtaining data, it is time-consuming, laborious, expensive, and limited in sample size, which is not conducive to large-scale and long-term monitoring [13]. The soil salinometer based on conductivity simplifies the sampling process and helps to quickly obtain soil salt content data at different depths. However, it is still not suitable for long-term monitoring of regional saline soil [14]. Optical remote sensing data have the advantages of wide coverage, low cost, less ground limitation, and rich historical data, and these data have been widely used for the dynamic monitoring and evaluation of soil salinization. This makes it possible to monitor salinized soil on a large scale rapidly and dynamically [15]. Masoud et al. [16] built an inversion model for soil salinization in the central plain of Morocco based on Landsat 8 spectral data, which was used to monitor the degree of salinization in the Dakhla Oasis. Wang et al. [17] constructed inversion models for salinization at different degrees in typical areas of the Yellow River Delta based on UAV and Sentinel-2A images. Tran et al. [18] constructed an enhanced salinity index (ESI) and analyzed the dynamic change characteristics of soil salinization in the Mekong Delta according to a long time series of Landsat images from 1989 to 2018. These are successful cases that used optical remote sensing data to monitor regional soil salinization.

The net primary productivity (NPP) of vegetation refers to the total amount of organic matter accumulated with vegetation photosynthesis minus its own respiration [19]. It reflects the productive capacity of vegetation under the comprehensive influence of climate, terrain, and human factors in the natural environment [20,21]. Research shows that NPP is more sensitive to climate change and human activities than the normalized difference vegetation index (NDVI) [22]. Therefore, it can accurately reflect the status of vegetation productivity and can be also used as an important factor to measure the productivity of cultivated land. Past research confirmed that NPP data in the MOD17A3 dataset could represent the productivity of cultivated land [23]. Soil salinization of cultivated land affects regional NPP values, which harms crop productivity by inhibiting crop growth. Coastal areas have a wide range of salinized farmland, which is a key area for the transformation of low- and medium-yield farmland and improvement in farmland production potential in the future. An in-depth analysis of the impact of soil salinization on cultivated land productivity is essential for achieving efficient utilization of cultivated land resources and ensuring national food security. However, the existing research focused on the correlation between the quality of non-salinized farmland and NPP [24,25], and research on the relationship between salinization and farmland productivity has not received due attention. Compared with previous studies, this paper attempts to quantitatively elucidate the effects of different degrees of salinized soils on cultivated land productivity by analyzing time series characteristics and their interrelationships between soil salinity and cultivated land productivity. Using these results, this paper then explains the comprehensive management effect of saline cultivated land in a certain period. This approach is more conducive to understanding the characteristics of and trends in the changes to salinized cultivated land, and it is of great significance for the improvement and regulation of local salinized cultivated land. Due to the fact that cultivated land productivity can reflect the degree of soil salinization, we assume that the productivity of cultivated land is highly correlated with soil salt content, and based on the aforementioned analysis, we conduct our research in this paper.

Our main research objectives are as follows: (1) analyze the temporal and spatial characteristics of soil salinization and NPP of cultivated land in the Shandong region around the Bohai Sea from 2001 to 2019 and (2) quantitatively express the influence of soil salinization on cultivated land productivity. Our study aims to elucidate the impact of salinity on cultivated land productivity by analyzing the spatial and temporal dynamic characteristics of salinity and NPP of the cultivated land and to provide support for using quantitative measures to regulate salinity, improve farmland productivity, and alleviate food security problems.

2. Materials and Methods

2.1. Study Area

The study area is located in the Shandong part of the Bohai Sea rim, mainly including the whole areas of Qingyun County, Wudi County, Zhanhua District, Hekou District, Lijin County, Kenli District, Dongying District, Guangrao County, Shouguang City, Hanting District, and Changyi City and partial areas of Pingdu City and Laizhou City. The geographical coordinates are 117°26′–120°19′ E, 36°26′38°19′ N (Figure 1). The area has a high elevation in the south and a low elevation in the north. The geomorphic types include river beach highlands (including riverbed, shoal, natural levee, and crevasse splay in each stage), flood plains, and depressions [26]. The study area is located in the warm temperate zone and is cold in the winter and hot in the summer. The annual average temperature is 9–20 °C, the annual average precipitation is 530–630 mm, and the annual potential evapotranspiration is 1500 mm. The distribution of precipitation is uneven in time and space. In the spring, there is little precipitation and strong evaporation, and salt easily accumulates on the land surface. In summer, there is more rainwater, which easily leads to waterlogging. The soil types are mainly fluvo-aquic soil and saline soil, and silty soil is most widely distributed in the cultivated layer. Groundwater is mainly brackish water, salt water, and brine. The special sedimentary environment, climate conditions, and soil parent materials lead to serious salinization of the cultivated land, and the area of primary salinization in the cultivated land is as high as 70%.

2.2. Data Sources

2.2.1. Remote Sensing Data

The land cover type data were obtained from the results of Landsat remote sensing images produced by Huang Xin’s team at Wuhan University [27]. There are five types of land: cultivated land, forest land, water area, construction land (impervious water surface), and tidal flats (unused land). The time series of the images was 2001–2019, and the spatial resolution was 30 m. Landsat remote sensing images from five phases in 2001, 2005, 2011, 2015, and 2019, used for soil salinization inversion, were downloaded from the USGS website (https://earthplorer.usgs.gov/ (accessed on 13 November 2022)). ENVI5.3 software was used for preprocessing including radiation calibration, atmospheric correction, and clipping.

The net primary productivity (NPP) data for the vegetation were obtained from the MOD17A3HGF V6 dataset provided by NASA. The annual NPP of the product was synthesized using MOD17A2H [28] to obtain a net photosynthesis product for all 8 days in a given year. The time series was 2001–2019, and the spatial resolution was 500 m. The above data were downloaded using code programming in PIE—Engine cloud platform (https://engine.piesat.cn/ (accessed on 13 November 2022)). To ensure consistency in the data spatial resolution, Arcgis10.2 software was used to resample NPP on a 30 m grid.

2.2.2. Soil Data

The grid method was used for the layout of the sample sites. Using the remote sensing image map, the study area was covered with 3 km × 3 km grids (avoiding aquaculture water surface, salt pans, and current construction land), and one sampling site was designed in the middle of each grid. During the field surveys, the specific sampling locations were determined according to the traffic accessibility and the field cultivated land conditions. A hand-held GPS was used to read the coordinates of the sample sites, and the land use status was also recorded. An EC110 portable conductivity meter was used to measure the soil conductivity at different depths for each layer at a distance of 2.5 cm, 7.5 cm, 15.0 cm, and 22.5 cm from the ground surface. At typical sample points, about 1 kg of soil samples from the plow layer was obtained with soil drills, bagged and sealed, and taken back to the laboratory to test the soil salt content (the soil salt was extracted with water and soil at a ratio of 5:1 and determined using the drying method). The field surveys were carried out from early March to early May 2019. A total of 720 sets of conductivity data were measured, and 537 soil samples were collected.

Using the soil salt content data from the 537 samples obtained from the cultivated layer (measured with the laboratory drying method) and the average soil conductivity data from each depth level in the field, a relationship model between soil salt content and conductivity was established, and the soil salt content of each depth layer at the 720 sample sites in the study area was calculated, respectively. In this way, the tedious laboratory analysis process of obtaining soil salinity data from all sample sites was reduced.

2.3. Methods

A general flowchart for this study is shown in Figure 2. Firstly, the characteristics of soil salinization at different depths in the cultivated land were analyzed using descriptive statistics. Secondly, the transfer matrix method was used to analyze the mutual transformation process between the areas of various land use types from 2001 to 2019, so as to express the spatio-temporal evolution characteristics of the quantity of cultivated land and other land types. Thirdly, the random forest model method was used to inverse the soil salt content for each year, and the spatial interpolation and mapping of inverse distance weights were combined to analyze the temporal and spatial distribution pattern changes in soil salinization of the cultivated land. Finally, the quantitative impact of cultivated land salinization on NPP was analyzed using the correlation regression method.

2.3.1. Descriptive Statistical Analysis Method

Statistical Product Service Solutions (SPSS, version: 23.0) software was used with descriptive statistical methods to analyze the maximum, minimum, variance, mean, median, standard error, kurtosis, skewness, coefficient of variation (CV), and other characteristics of salt content at different depths in the soil layers. The CV reflects the dispersion degree and relative variation in random variables, which can be divided into three levels of variation: weak [0, 0.10), medium [0.10, 1.00), and strong [1.00, ∞) [29].

2.3.2. Analysis of the Land Type Transfer Matrix

The transfer matrix method reflects the following contents: (a) cultivated land area in two periods and (b) the source of increases and the flow direction of decreases in cultivated land. The expression is as follows [30]:

$$S_{ij}=\left[\begin{array}{ccc}{S}_{11}& \cdots & S_{1n}\\ S_{21}& \cdots & S_{2n}\\ ⋮& ⋮& ⋮\\ S_{n1}& \cdots & S_{nn}\end{array}\right]$$

(1)

where S is the total land area; n is the number of land types; and i, j is the land type at the beginning and end of the study period, respectively.

2.3.3. Random Forest Inversion of Soil Salt Content

The random forest machine learning method was proposed by Breiman and Adele Cutler et al. [31] at the beginning of the 21st century. Its basic principle is an integrated learning algorithm based on decision trees as individual learners. It is widely used to solve the problem of overfitting that is present in traditional machine learning models. Its advantage lies in the fact that each decision tree unit is unrelated and randomly arranged. Compared with other machine learning models, it has the advantages of fast training speed, a good prediction effect, and strong robustness [32]. Based on the five Landsat remote sensing images preprocessed in Section 2.2.1 above, the vegetation index [33] and salinity index [34] were calculated using ENVI5.3 software (

Table 1. Correlation spectral index.

Spectral Index	Expression	Reference
Normalized difference vegetation index	(NIR − Red)/(NIR + Red)	[33]
Difference vegetation index	NIR − Red
Enhanced vegetation index	2.5 × (NIR − Red)/(NIR + 6×Red − 7.5 × Blue + 1)
Ratio vegetation index	NIR/Red
Soil adjusted vegetation index	(NIR − Red) × 1.5/(NIR + Red + 0.5)
Salinity index (SI1)	(Green × Red)0.5	[34]
Salinity index (SI2)	(Green2 + Red2 + NIR2)0.5
Salinity index (SI3)	(Green2 × Red2)0.5
Salinity index (SI4)	SWIR1/NIR
Salinity index (SI5)	(Red − SWIR1)/(Red + SWIR1)
Salinity index (SI6)	(Red × Blue)/Green
Salinity index (SI7)	(Red × NIR)/Green

). According to the correlation coefficient and variance expansion factor (VIF) for each spectral index and soil salt content, the final spectral indexes for inversion such as RVI, EVI, SI₃, and SI₅ were determined. An inversion model was established based on the measured soil salt content data from 2019 and the spectral index from the same period using the random forest method. Regarding the parameter setting for the random forest, the number of decision trees was 200 and the minimum number of leaves was 3. The above process was implemented in Matlab (version: 2021b) software. Equally spaced years were selected to invert the soil salinity in 2001, 2005, 2011, and 2015. (The 2011 dataset was selected to replace the 2010 dataset for inversion due to high cloud cover in the study area during the same period in 2010, resulting in poor inversion results.)

2.3.4. Inverse Distance Weight Space Interpolation and Mapping

The principle underlying spatial interpolation analysis with the inverse distance weighting method is based on the basic assumption of “the first law of geography”: “the similarity between two objects decreases with the increase of their distance” [35], and the corresponding weight is calculated according to the distance between sampling sites. The two-dimensional inverse distance weighted interpolation method (IDW) is a multivariate interpolation method used to evaluate the prediction value of S0 at any site based on the assumption of known datasets. It is often used to analyze the spatial distribution of soil salt content, organic matter content, etc. This process was mainly implemented in Arcgis10.7 [36].

2.3.5. Correlation and Regression Analysis

(1)

The Spearman correlation coefficient method

The Spearman correlation coefficient is used to determine the correlation between factor variables. The formula is as follows [37]:

$$P=1-\frac{6{{\displaystyle \sum }}_{i=1}^n{d}_i^2}{n\left(n^2-1\right)}$$

(2)

where P represents the Spearman correlation coefficient; n is the total number of samples; and d_i is the rank difference between the soil salt content and NPP of cultivated land.

(2)

Regression model construction

The functional relationship between the soil salt content (independent variable) and NPP value (dependent variable) of the cultivated land was established to quantify the impact of salinization on cultivated land crop productivity. It mainly included linear (unitary linear) and curve (unitary quadratic linear, power regression, exponential regression, reciprocal regression) function models to identify the possible relationship (linear or nonlinear) between the degree of soil salinization and NPP of the cultivated land. These analyses were implemented in Origin 2021. The fitting models are as follows:

Univariate linear regression: Y = β₁ X + β₀

(3)

$$\mathrm{Monadic} \mathrm{quadratic} \mathrm{linear} \mathrm{regression}: Y={\beta }_2{X}^2+{\beta }_{1}X+{\beta }_0$$

(4)

Power regression: Y = β₂X^β1 + β₀

(5)

$$\mathrm{Exponential} \mathrm{regression}: Y={\beta }_2{e}^{{\beta }_{1}X}+{\beta }_0$$

(6)

$$\mathrm{Piecewise} \mathrm{linear} \mathrm{regression}: Y=\{\begin{array}{c}{\beta }_0+{\beta }_{1}X,X\le \alpha \hfill \\ {\beta }_0+{\beta }_{1}X+{\beta }_0\left(X-\alpha \right),X>\alpha \end{array}$$

(7)

where $\alpha , {\beta }_0, {\beta }_1, \mathrm{and} {\beta }_2$ are constants.

3. Results

3.1. Descriptive Statistical Characteristics of Cultivated Land Salinization

Based on the conductivity measured using an EC110 conductivity meter, the relationship between soil conductivity EC₀ (us/m) and salt content S_t (g/kg) measured in the laboratory was established as follows:

S_t = 0.0021EC₀ + 0.7507, n = 376, R² = 0.873

(8)

The soil salt contents in the plow layer and each layer at 2.5 cm, 7.5 cm, 15.0 cm, and 22.5 cm at all sample sites were calculated using Equation (8), and then a descriptive statistical analysis of these data was conducted (

Table 2. Descriptive statistics for the salt content at different depths in the cultivated land in the Shandong area around the Bohai Sea.

Depth of Soil Layer (cm)	Sample Size	Min (g/kg)	Max (g/kg)	Average (g/kg)	Median (g/kg)	Variance	Standard Deviation	Kurtosis	Skewness	Coefficient of Variation
2.50	720	0.76	55.98	3.56	1.74	44.97	6.71	28.78	5.12	1.88
7.50	720	0.76	40.57	2.95	2.03	13.63	3.69	40.68	5.63	1.25
15.00	720	0.76	33.09	3.00	2.27	8.88	2.98	38.07	5.37	0.99
22.50	720	0.76	29.29	3.13	2.45	6.80	2.61	29.36	4.52	0.83
plow layer	720	0.76	37.53	3.16	2.16	13.59	3.69	30.56	4.94	1.17

). According to the grading standard of “China Salinized Soil”, the degrees of salinization were divided into five grades: non-salinization, mild salinization, moderate salinization, severe salinization, and saline soil. The salt content (g/kg) ranges are (0, 1), [1, 2], [2, 4], [4, 6], [6, ∞] [38].

It can be seen from Table 1 that the average salt content in the plow layer was 3.16 g/kg, showing moderate salinization. The minimum and maximum values were 0.76 g/kg and 37.53 g/kg, respectively, and the coefficient of variation was 1.17, showing strong variation. The average salt contents in each layer from top to bottom were 3.56 g/kg, 2.95 g/kg, 3.00 g/kg, and 3.13 g/kg, respectively, which were all moderate salinization, showing the characteristic of first decreasing and then increasing in the vertical direction. The kurtosis and skewness of salt content in each soil layer were large, and the data did not show a normal distribution. The medians were in the range of 1.74–2.45 g/kg, which were less than the average values, indicating that the data distributions were right-skewed (positively skewed). The coefficient of variation tended to increase from bottom to top. The coefficients of variation in the 22.50 cm and 15.00 cm layers were 0.83 and 0.99, respectively, showing moderate variation. The coefficients of variation for the salt content in the 2.5 cm and 7.5 cm layers were greater than 1.00, showing strong variation. This was mainly because the upper soil in the cultivated land was greatly affected by random factors (such as climate, human activities, etc.), which led to strong spatial variability in soil salt content and large difference in salt content in different regions.

3.2. Spectral Index Screening

The correlation analysis results for the soil salt content and spectral index in the study area are shown in Figure 3. There was a highly significant negative correlation between soil salinity and the vegetation index. There was a highly significant correlation between soil salt content and most salt indices (except for SI7). Using the correlation coefficient and variance expansion factor (VIF) for each spectral index and soil salt content, the final spectral indexes for inversion including RVI, EVI, SI3, and SI5 were determined.

3.3. Temporal and Spatial Pattern Change Characteristics in Cultivated Land Quantity

The land type area transfer matrix for the study area from 2001 to 2019 is listed in

Table 3. Transfer matrix for land use types in the Shandong area around the Bohai Sea from 2001 to 2019.

Year	Type	2019						2001–2019
		Grassland/km2	Cultivated Land/km2	Construction Land/km2	Woodland/km2	Water Area/km2	Mudflats/km2	Change/km2	Change (% of Total Area)
2001	Grassland	23.87	85.13	30.87	0.08	26.41	42.59	−87.82	−30.83
	Cultivated land	29.66	18006.31	1914.59	0.97	537.20	252.77	−2004.51	−9.67
	Construction land	0.29	323.88	3370.83	0.04	195.98	123.90	1840.71	+45.98
	Woodland	0.00	2.38	0.26	0.45	0.00	0.00	−0.92	−50.80
	Water area	0.97	147.26	113.18	0.02	941.18	238.67	1197.92	+82.43
	Mudflats	9.67	168.80	432.83	\	929.08	1772.68	−888.42	−26.67

, and the spatial distribution and proportion for each category are shown in Figure 4. It can be seen that from 2001 to 2019, the land use types were mainly cultivated land and construction land, and the sum of the two areas accounted for more than 80% of the total study area each year. However, the overall area and proportion of cultivated land showed a decreasing trend year by year, and the construction land increased year by year. The total area of cultivated land was 20,724.08 km² in 2001 and 18,719.57 km² in 2019, with a cumulative decrease of 2004.51 km² and a decreased proportion of 9.67%. The average annual decrease was 105.50 km², and the average annual decrease rate was 0.51%. The total area of construction land was 4002.87 km² in 2001 and 5843.58 km² in 2019, with a cumulative increase of 1840.71 km² and an increased proportion of 45.98%. The average annual increase was 96.88 km², and the average annual increase rate was 2.42%. The transfer matrix showed that over the 19 studied years, 1914.59 km² of cultivated land was converted into construction land, while only 323.88 km² of construction land was converted into cultivated land. The net transfer of cultivated land to construction land reached 1590.71 km².

As seen in Figure 4, from 2001 to 2019, the increase in construction land occupied the cultivated land around the original cities and towns and spread around the original cities and towns in 2001. These areas were mainly concentrated in the middle and south of the study area.

The net transfer of cultivated land to water area was 389.94 km², and to mudflats, the net transfer was 83.97 km². In addition, the net transfer of grassland to cultivated land was 55.47 km², and the spatial distribution was very scattered, which benefited from the policy and engineering measures for land consolidation to supplement cultivated land.

3.4. Temporal and Spatial Characteristics of Soil Salinization in Cultivated Land

The 2019 sample data were divided into a training set and a verification set in a ratio of 7:3, and the inversion model was established using the random forest method. The results are shown in Figure 5. The training set’s R² was 0.833 and RMSE was 1.982 g/kg. The validation set’s R² was 0.714, and the RMSE was 2.134 g/kg. The training set and validation set had good modeling effects, which were used for the inversion of soil salt content in the study area.

Using this model, the soil salt contents of the cultivated land in the Shandong region around the Bohai Sea in 2001, 2005, 2011, and 2015 were inversed, respectively. According to the type of salinization degree, a spatial distribution map was drawn, and then the area and proportions were counted. The results are shown in

Table 4. The areas and proportions of cultivated land with different salinization degrees.

Salinization Degree	Soil Salt Content (g/kg)	Area (km2)					Proportion (%)
		2001	2005	2011	2015	2019	2001	2005	2011	2015	2019
Non salinization	<0.1	337.64	169.56	494.48	232.60	27.86	1.63	0.83	2.51	1.23	0.15
Mild salinization	0.1–0.2	4567.23	4215.51	4237.21	5023.18	7149.28	22.04	20.63	21.48	26.47	38.19
Moderate salinization	0.2–0.4	4614.43	3552.67	4469.16	6069.45	7289.96	22.27	17.38	22.65	31.98	38.94
Severe salinization	0.4–0.6	4038.99	4019.65	5352.03	3676.43	1921.91	19.49	19.67	27.13	19.37	10.27
Saline soil	>0.6	7165.80	8480.89	5177.61	3975.10	2330.56	34.58	41.50	26.24	20.95	12.45

and Figure 6.

3.4.1. Spatial Distribution of Soil Salinization in Cultivated Land

The area and proportion of cultivated land with different salinization degrees in the study area in 2019 are shown in Table 4, and the spatial distribution is shown in Figure 6e. In terms of the proportion of quantity, the area with moderate salinization accounted for the largest proportion of the total area of cultivated land (38.94%), followed by mild salinization (38.19%), which added up to 77.13%. Severe salinization and saline soil accounted for 12.45% and 10.27%, respectively, while non-salinization only accounted for 0.15%. In summary, the cultivated land in the study area was mainly mild and moderate salinization, which was the main body determining the regional cultivated land productivity. However, the severe salinization and saline soil, accounting for 22.82%, had a significant impact on the quality and productivity of the cultivated land. From the perspective of spatial distribution, the degree of soil salinization in the cultivated land in the study area gradually increased from south to north and from east to west.

3.4.2. Spatial and Temporal Evolutionary Characteristics of Soil Salinization in the Cultivated Land

Figure 6a–e shows the overall characteristics of the spatial distribution of soil salinization in the cultivated land in the study area over time from 2001 to 2019. The overall spatial distribution pattern in soil salinization did not change significantly from 2001 to 2005. Saline soil was mainly distributed in the north and slightly increased, while mild salinization was mainly distributed in the south, and severe salinization and moderate salinization were between the two. From 2005 to 2011, the northern saline soil area decreased significantly, the severe and moderate salinization increased correspondingly, and the southern part was still dominated by mild salinization. From 2011 to 2019, most of the severe salinization and saline soil in the north and middle areas were reduced to moderate salinization.

Figure 6f shows the Sankey diagram for the transfer of cultivated land area, and Figure 6g shows the change in the proportion of cultivated land area with different salinization degrees in the Shandong area around the Bohai Sea from 2001 to 2019. According to the spatial change characteristics shown in Figure 6a–e and the quantitative change characteristics of the cultivated land area with different soil salinization degrees shown in Figure 6f,g, the temporal and spatial evolution characteristics of soil salinization in the cultivated land in the study area can be divided into three stages. (1) From 2001 to 2005, the degree of soil salinization increased slightly. The proportion of saline soil in the cultivated land increased from 34.58% to 41.50%. Most of the areas were transformed from severe salinization, with a transformation area of 1924.46 km². The second was moderate salinization, with a conversion area of 687.45 km². The area with severe salinization soil was roughly kept in balance, with an area of 1147.64 km² transferred into saline soil, and an area of 939.04 km² transferred in from moderate salinization. The moderate and mild salinization soil areas decreased slightly. (2) From 2005 to 2011, the degree of soil salinization decreased slightly. The proportion of saline soil area decreased from 41.50% to 26.24%, and most of the areas were transformed into severe salinization, with a transformation area of 3302.12 km². Severe salinization soil increased from 19.67% to 27.13%, and moderate salinization and mild salinization soils increased slightly from 22.65% to 31.98% and 21.48% to 26.47%, respectively. (3) From 2011 to 2019, the soil salinization degree significantly decreased further. The areas of saline soil and severe salinization decreased on a large scale, and the areas with the two types transformed into moderate salinization soil at 1453.77 km² and 23,365.91 km², respectively. The areas with mild salinization and moderate salinization soils were increasing. This showed that the salinization disaster was greatly improved in the past 10 years with great attention from the country and the joint efforts of the country, academia, and industry.

3.5. Temporal and Spatial Characteristics of the NPP of Cultivated Land

The temporal and spatial characteristics of the NPP of cultivated land in the Shandong area around the Bohai Sea from 2001 to 2019 are shown in Figure 7. It can be seen from Figure 7a that from 2001 to 2019, the NPP values for the cultivated land in the Shandong area around the Bohai Sea in each year were in the range of 296.04–364.54 gC m⁻²a⁻¹, and the average value across all years was 330.79 gC m⁻²a⁻¹. In the 19 studied years, the NPP value increased by 68.50 gC m⁻²a⁻¹, with an average annual growth of 3.61 gC m⁻²a⁻¹. The curve showed that the average value for the regional NPP showed an increasing trend with fluctuation, and the linear growth reached a significant level (p < 0.05). From 2001 to 2009, the value first increased and then decreased, and from 2009 to 2019, it continued to rise. This period basically coincided with the proposal and implementation of the national and Shandong Province saline–alkali land treatment project, the “Bohai granary” Science and Technology Demonstration Project, and other strategies [10,11]. The increase in the NPP of the cultivated land showed that the improvement in salinized cultivated land produced significant results during this period.

Figure 7b shows that the distribution of the average NPP of the cultivated land in the Shandong area around the Bohai Sea from 2001 to 2019 had obvious spatial heterogeneity. The NPP value in the north was significantly lower than that in the south, roughly bounded by Dongying District in the middle of the study area.

3.6. Analysis of the Influence of the Salinization Degree of Cultivated Land on NPP

3.6.1. Correlation between Soil Salinization and NPP in Cultivated Land

A 3 km × 3 km grid was established using ArcGIS software, and the NPP value for the cultivated land and soil salt content in different years was extracted for correlation analysis. Figure 8a shows the Spearman correlation coefficient analysis results for both. It can be seen that there was an extremely significant negative correlation between NPP and soil salt content in different years. The correlation coefficients were greater than 0.4, but the correlation coefficients were generally decreasing, with a total decrease of 0.266 from 2001 to 2019. We speculate that the causes of this phenomenon may be the following two points: ➀ Figure 8b–g shows the results for the correlation between the soil salt content and NPP of cultivated land in different years calculated between partitions. In the range of “salt content > 6 g/kg and NPP value between 0–200 gC m⁻²a⁻¹”, “salt content between 2–4 g/kg and NPP value between 200–400 gC m⁻²a⁻¹”, and “salt content between 1–2 g/kg and NPP value between 400–600 gC m⁻²a⁻¹”, there were high degrees of correlation, which was a “significant” or “extremely significant” negative correlation. ➁ In addition, the “soil salt content > 6 g/kg and NPP value between 400–600 gC m⁻²a⁻¹” in 2005 and 2011, the “soil salt content > 6 and NPP value between 600–800 gC m⁻²a⁻¹” in 2019, and the multi-year average “soil salt content between 4–6 g/kg and NPP value between 400–600 gC m⁻²a⁻¹” were positively correlated, which also verified the above speculation about the reasons for the downward trend in their correlation coefficients.

3.6.2. Fitted Relationship between Soil Salinization and NPP of Cultivated Land

The fitted results for the NPP values and soil salt content of cultivated land in different years are shown in Figure 9. During this study, the R² for the fitted nonlinear function results were higher than the fitted linear function results, indicating that the correlation between them was not purely linear. For 2001, 2005, 2011, 2015, 2019, and the multi-year average, the R² values of the curve for the fitted models were 0.06–0.12, 0.01–0.02, 0.01–0.07, 0.01–0.04, 0.01–0.05, and 0.03–0.09 higher than that of linear fitted models, respectively. For the curve fitting models, the power function regression model was the best. In summary, the precision of different fitting regression models had advantages and disadvantages, but their overall trends were roughly the same, showing a trend that the NPP value gradually decreased with the increase in the soil salt content of the cultivated land.

In order to perform a quantitative analysis of the soil salt content and NPP, three intervals of 0–3 g/kg, 3–7 g/kg, and 7–15 g/kg were finally selected to achieve piecewise fitting using the power function regression model, considering the difference in slope change and the fitting degree of the piecewise linear regression. The results are shown in Figure 10. In summary, when the salt content of the cultivated soil was in the range of 0–3 g/kg, the fitting accuracy R² was 0.54, and the fitting effect was the best. For every 1 g/kg increase in salt content, the NPP value was predicted to decrease by 43.62 gC m⁻²a⁻¹. When the soil salt content of the cultivated land was in the range of 3–7 g/kg, the fitting accuracy R² was 0.44. For every 1 g/kg increase in salt content, the NPP value was predicted to decrease by 30.36 gC m⁻²a⁻¹. When the soil salt content of the cultivated land was in the range of 7–15 g/kg, the fitting accuracy R² was 0.45. For every 1 g/kg increase in salt content, the NPP value was predicted to decrease by 44.19 gC m⁻²a⁻¹.

4. Discussion

4.1. Quantity Change in Salinized Cultivated Land

From 2001 to 2019, the amount of cultivated land in the Shandong area around the Bohai Sea showed a decreasing trend, and the proportion of cultivated land decreased from 69.71% in 2001 to 62.98% in 2019, with a net area decrease of 2004.51 km². This was mainly related to the rapid development of industrialization and urbanization in the region and the large use of cultivated land, which is consistent with the research results of Zhu et al. [39] on the changes in cultivated land in the Yellow River Delta in the recent 20 years. The development and utilization of unused land have always been one of the main sources of new cultivated land in China [40]. According to statistics, 12.59% of the newly cultivated land in Xinjiang from 2005 to 2008 came from saline–alkali land [41]. The “Bohai Sea Granary Science and Technology Demonstration Project” implemented in the Bohai Rim region from 2013 to 2018 involved 2.6667 × 10⁶ hm² of medium- and low-yield salinized cultivated land treatments. Overall, 3.647 × 10⁵ hm² of the salinized cotton fields was transformed into grain fields, 1.935 × 10⁵ hm² of the saline–alkali land was improved into cultivated land, and the cumulative increase in grain production in five years was 1.41 × 10⁷ t [2,42]. Only by protecting both the quality and quantity of cultivated land can we truly achieve the goal of steadily improving the quality of arable land and food production in the Bohai Sea Rim region.

4.2. Change in Soil Salinization Degree in Cultivated Land

Due to the special geographical environment in the coastal area, the cultivated land in this area is more seriously affected by salinization than that in the inland area [43]. Fully understanding the temporal and spatial characteristics of soil salinization in coastal areas is essential for mitigating salinization and ensuring regional agricultural development and food security. The salinization degree in the cultivated land in the Shandong area around the Bohai Sea presented a trend with “high salt content near the sea, low salt content far from the sea”, which is similar to the research results of Das et al. (2020) [44] from the coastal areas in central and southern Bangladesh. For areas closer to the sea, there is a higher frequency of seawater immersion of the soil mass, and the salinity of the groundwater is significantly higher than that in other regions. At the same time, due to the low terrain, the soil salt in the high terrain region converges here with the water flow. In this way, under the combined effect of multiple factors, the salt content of cultivated land in the offshore area is high, even several times higher than that in other regions.

The change in the soil salinization degree of cultivated land in the study area over the past 20 years is characterized by several stages. From 2011 to 2019, the improvement in salinized cultivated land achieved remarkable results, and the area of saline soil and severe salinized soil decreased significantly. This is closely related to the increasing attention paid to the salinization hazards in recent years [45,46,47], especially the extensive development of saline–alkali land treatment and salinized cultivated land improvement. In future improvement processes, proper consideration should be given to zoning for improvement. For saline soil and severe salinized areas, mechanical deep plowing should be carried out to cut off the soil capillaries, improve the soil water holding capacity, and inhibit salt return, and salt drainage using concealed pipes should be improved to reduce the groundwater level at the same time. For moderately and mildly salinized areas, the drainage and irrigation system should be further optimized, especially using water-saving irrigation technologies such as sprinkler irrigation and drip irrigation, so as to achieve accurate salt regulation.

4.3. Temporal and Spatial Changes in the NPP of Salinized Cultivated Land

At present, there are no research reports on the NPP of salinized cultivated land in the Shandong area around the Bohai Sea. Most of the existing studies focused on all land use types in the Yellow River Delta [48] and the NPP of forest vegetation in a large range in the Yellow River basin [49,50]. The results of this paper showed that, in terms of the time change, the overall NPP of salinized cultivated land in the Shandong area around the Bohai Sea showed a fluctuating upward trend in the past 20 years. With 2009 as the node, the fluctuating trend increased first and then decreased from 2001 to 2009, and then it continued to increase from 2009 to 2019. This may be related to the implementation of projects such as national and Shandong Province’s saline–alkali land treatment, saline–alkali medium- and low-yield cultivated land improvement, and other projects. The ecological water supplement measures in the Yellow River Delta may also play a role in diluting the soil salinity and groundwater mineralization in this region [51]. From the perspective of spatial distribution, the average NPP of the cultivated land over the years was high in the southeast and low in the coastal area. The southeast area is relatively high in elevation, with little impact of salinization, and the soil salt content is generally low (mostly mild salinization). As the NPP is affected by the expansion of urban construction land, the values for the cultivated land around some cities and towns were also low (0–200 gC m⁻²a⁻¹). This is also a phenomenon that should be given more attention in the future, and measures should be taken to curb urban expansion, maintain stability, and improve cultivated land productivity.

4.4. Quantitative Relationship between Annual Average NPP of the Cultivated Land and Soil Salt Content

The soil quality of the cultivated land in coastal areas depends to a large extent on its soil salt content (the higher the soil salt content, the lower the quality index) [52]. Some existing research focused on the relationship between cultivated land quality and NPP, and there are a lot of research results [53,54,55]. Zhuang et al. [12] analyzed the relationship between cultivated land quality (CLQ) and NPP in Sangong River Basin, Xinjiang, and found that the correlation coefficient R² was 0.706. For every point of increase in the CLQ value, the annual average NPP increased by 35.318 gC m⁻²a⁻¹. Wang et al. [56] inversed the temporal and spatial characteristics of cultivated land salinization in the Hetao Plain, Inner Mongolia. Using an overlay analysis of NPP and salt content, it was shown that the NPP in the salinized soil improvement area was in a sustainable growth trend, indicating that the saline–alkali land treatment had achieved certain good results. This is similar to the results of this study. However, the previous studies failed to clarify the quantitative relationship between the salinization degree of cultivated land and its productivity. Different from the above assumptions, this study preliminarily identified the quantitative relationship between the two and divided the influence of the soil salt content on the NPP value into three intervals(g/kg): [0, 3), [3, 7), [7, 15). It was found that the influence degrees of different intervals were different. There was an overall decreasing trend in the correlation coefficient between soil salinity and NPP in cultivated land. We speculate that this may be due to two factors: (1) the impact of the accuracy error between local NPP and the predicted soil salt content and (2) the impact of the regional improvement in saline soil led to the phenomenon of “high salt content, large NPP value”, which reduced the negative correlation between them. This result made up for the gaps in this field and can help to estimate the productive capacity of cultivated land, delimit key improvement areas, and formulate improvement plans in the future.

4.5. Research Limitations and Future Prospects

We need to acknowledge some small limitations and provide guidance for future research directions. First of all, when inversing the soil salinization of the cultivated land in different years, we only considered one machine learning method and remote sensing image. In the future, we should comprehensively explore multiple methods (linear, nonlinear, and machine learning), use higher resolution satellite images (such as Sentinel series), and fully combine remote sensing and machine learning algorithms to further improve the remote sensing monitoring accuracy of cultivated land salinization. Secondly, a typical coastal region, the Shandong region of the Bohai Rim, was chosen as the case study. We recommend caution in applying the results of this study to inland regions due to the large differences in the soil-forming parent material and the environment from inland arid and semi-arid regions.

5. Conclusions

Soil salinization of cultivated land in coastal area poses a great threat to the sustainable development of agricultural system. The contradiction between the inhibition of soil salinization on productivity and the growing demand for food is becoming increasingly fierce. The research results of this paper confirmed the following. First of all, the temporal change characteristics in the amount of salinized cultivated land in the Shandong area around the Bohai Sea from 2001 to 2019 were explored, showing an overall decreasing trend. We realized the inversion of cultivated land salinization in this region using a long time-series, which indicated that the change processes in cultivated land salinization in this region were mainly divided into three stages: slightly increased in 2001–2005, slightly reduced in 2005–2011, and significantly reduced in 2011–2019. Secondly, the NPP of the cultivated land in this region kept a fluctuating and rising trend. The NPP of the cultivated land in the south of the study area remained stable, and the NPP of the cultivated land in the north had a very significant increasing trend. Finally, the quantitative relationship between soil salt content and NPP in the cultivated land was analyzed. The soil salt content was divided into three intervals, and the change degrees between soil salt content and NPP in different intervals were different. Our research results provide decision-making support for strengthening the dynamic monitoring and regulation of cultivated land salinization and for promoting sustainable utilization of salinized cultivated land in the coastal areas to alleviate the food crisis.