Remote Sensing Evaluation of Cultivated Land Soil Quality in Soda–Saline Soil Areas

Abstract

Rapid evaluations of farmland soil quality can provide data support for farmland protection and utilization. This study focuses on the soda–saline soil region of Da’an City, Jilin Province, covering an area of 4879 km²; it proposes a framework for evaluating farmland soil quality based on multi-source remote sensing data (Sentinel-2 MSI, GF-5 AHSI hyperspectral and field hyperspectral data). Soil organic matter content, salt content, and pH were selected as indicators of cultivated land soil quality in soda–saline soil areas. A threshold of 20% crop residue cover was set to mask high-cover areas, extracting bare soil information. The spectral indices SI1 and SI2 were utilized to predict the comprehensive grade of soil organic matter + salinity based on the cloud model ($M{E}_c$ = 0.74 and $M{E}_v$ = 0.68). The pH grade was predicted using the red-edge ratio vegetation index (RVIre) ($M{E}_c$ = 0.95 and $M{E}_v$ = 0.98). The short-board method was used to construct a soil quality evaluation system. The results indicate that 13.73% of the cultivated land in Da’an City is of high quality (grade 1), 80.63% is of medium quality (grades 2–3), and 5.65% is of poor quality (grade 4). This study provides a rapid assessment tool for the sustainable management of cultivated land in saline–alkali areas at the county level.

1. Introduction

Cultivated land is a natural economic complex composed of natural elements, representing the integration of production, ecology, and life, as well as the synthesis of multifunctional needs [1,2,3,4]. The concept of cultivated land quality is highly complex and constantly evolving. Many scholars consider various aspects such as site conditions, soil characteristics, utilization status, ecological environment, and biodiversity [5,6,7]. Other scholars mainly approach the study of cultivated land quality from the perspective of soil quality or soil health [8]. Soil quality is widely considered to refer to the ability of soil to maintain biological productivity, preserve environmental quality, and promote the health of plants, animals, and humans within the boundaries of a specific land use or ecosystem [9,10,11,12].

Soil is the foundation of cultivated land, and the evaluation of cultivated land quality and soil quality intersect with each other, each having its own emphasis. Cornell University in the United States has established a soil health evaluation system, which considers soil to be an important ecosystem supporting the survival of plants, animals, and humans. Soil health refers to its ability to continuously perform this function. The evaluation system involves soil physics, biology, chemistry, and other aspects, including soil bulk density, organic matter, and the residue degradation rate [13]. Zhao et al. evaluated the health of cultivated land soil by integrating inherent and dynamic attribute indicators such as soil physics, soil chemistry, soil biology, topography, climatic conditions, and water conditions [14]. In 2018, Bünemann et al. analyzed 65 studies related to soil quality and summarized evaluation indicators with a usage frequency of more than 10%, which also included soil physics, chemistry, and biology [15].

Remote sensing technology can quickly, widely, and periodically acquire surface feature information, facilitating the exploration of soil properties and their variation patterns across spatial scales. Near-ground remote sensing, aerial remote sensing, and spaceborne remote sensing in the optical field represent an economical and rapid extraction method for soil attributes. Optical remote sensing observations have been proven to be powerful techniques for quantitatively measuring and simulating soil properties such as soil organic carbon content and salt content, with high fitting degrees of prediction functions. Remote sensing-based soil property mapping mainly includes the following methods: First, there is soil property remote sensing mapping based on soil–landscape models. Using the sample data from China’s second national soil survey as soil input data, combined with soil–landscape models, a three-dimensional mapping of soil organic carbon was achieved in China. Although the overall accuracy was not high, it revealed the spatial variation trend in soil organic carbon content in China [16,17,18]. Remote sensing data improves the accuracy of soil property mapping in large-scale or complex terrain areas, revealing spatial variation trends for soil properties, but they are not precise enough in expressing the spatial differences in soil properties at small regional scales. Second, there is soil property remote sensing that makes quantitative predictions based on remote sensing information. The Earth’s surface is intricate, and remote sensing data offer a comprehensive reflection of surface cover information. The surface of cultivated land is often covered by vegetation or crop straw. Direct use for soil property prediction generally requires selecting the bare soil period during the agricultural off-season [19,20]. Aerial or drone remote sensing provides flexibility in terms of time and can effectively supplement data for soil property predictions [21]. The Salinity Index (SI) and Normalized Difference Salinity Index (NDSI) were constructed to effectively monitor soil salinity using Vis-NIR spectroscopy and remote sensing data [22,23]. Third, we must consider the indirect prediction of soil properties based on vegetation remote sensing information. When it is difficult to directly obtain image data for the bare soil period due to weather, satellite transit cycles, and surface cover, soil properties can be indirectly estimated by monitoring vegetation information [24]. The soil pH value has been accurately estimated using the Ratio Vegetation Index (RVI), Normalized Difference Vegetation Index (NDVI), and Normalized Difference Water Index (NDWI) at the regional scale land use [25]. In addition, crop growth indicators such as NDVI, Net Primary Productivity (NPP), Gross Primary Production (GPP), and the Leaf Area Index (LAI) were obtained based on remote sensing for evaluating soil quality [26].

The quality of the soil of cultivated land is commonly evaluated at the county scale, emphasizing the improvement of regional soil productivity [27]. Da’an City is a county-level city in Jilin Province, China. It is a typical area of soda–saline soil and has been selected as the research area in this study. There are differences in the spatial spectral response characteristics of soil, crops, and crop straw on the surface of cultivated land. This study comprehensively utilizes the change information of these three elements, combined with remote sensing data, to evaluate soil quality in cultivated land. Existing research has found that soil attribute extraction based on remote sensing is mainly carried out on bare soil on cultivated land, with the following findings: (1) the properties of the topsoil (0–30 cm) that is regularly plowed are relatively homogenous; (2) during the intervals between crop sowings, the proportion of bare soil is large and no crust formation occurs; (3) the surface soil moisture content is low [28]. In most areas of Northeast China, conservation tillage is vigorously promoted. Due to the one-year cropping cycle, the intervals between crop sowings are long, but the surface is often covered with crop straw. It remains to be determined how remote sensing data can be better utilized to predict soil attributes to achieve remote sensing assessments of key soil indicators for cultivated land quality. On this basis, this study has the following aims: (1) develop a framework for evaluating county-level cultivated land soil quality based on multi-source remote sensing data in Soda–Saline Soil Areas; and (2) utilize extensive remote sensing data to swiftly acquire the soil attributes necessary for cultivated land soil evaluation, thereby achieving the rapid assessment of cultivated land soil quality in Soda–Saline Soil Areas at the county-level.

2. Materials and Methods

2.1. Overview of the Study Area

Da’an City (44°57′00″–45°45′51″ N, 123°08′45″–124°21′56″ E), located in the northwest of Jilin Province, China, is situated in the hinterland of the Songnen Plain, covering an area of 4879 km² (Figure 1). The city has a mid-temperate monsoon climate with four distinct seasons. The annual average sunshine duration is 3012.8 h, the annual average temperature is 4.3 °C, the annual average accumulated temperature is 2921.3 °C, and the annual average rainfall is 413.7 mm. The water areas of the Nenjiang River, Taoer River, and Huolin River in Da’an City cover an area of 726 km², accounting for one-seventh of the total area; the annual surface water runoff is 22.3 billion m³, ranking second in Jilin Province; the natural groundwater resources are 560 million m³, and the exploitable resources are 420 million m³, accounting for 75% of the natural resources. The cultivated land area of Da’an City is 2267.07 km², including 492.53 km² of paddy fields and 1774.54 km² of dry land. Among them, there are 20.6 km² of saline–alkali cultivated land. The main factors affecting the quality of cultivated land in Da’an City include soil organic matter content, degree of salinization, and soil pH value. Problems related to salinization are prominent in Da’an City, and the area of saline–alkali land is continuously increasing. The increasing use of agricultural inputs such as pesticides and fertilizers has led to the worsening of agricultural non-point source pollution.

2.2. Data Acquisition and Processing

2.2.1. Soil Collection and Processing

Based on the data on cultivated land extent, crop type, soil properties, DEM, and NDVI, sample points were set up on the cultivated land in Da’an. Field surveys and sampling were conducted multiple times from 11 to 18 May 2018, 19 to 26 August 2018, 25 to 28 August 2019, and 16 to 25 May 2021, resulting in a total of 169 sampling points (Figure 2). The number of soil samples was 138 and 31 from dry land and paddy fields, respectively. Maize and rice were planted on dry land and paddy fields, respectively. At each sampling point, soil was collected using a soil auger using the five-point method at a distance of 10 m, and then they were thoroughly mixed. The soil was collected from a depth of 0 to 20 cm. The coordinates were recorded using a handheld GPS. The statistical results of soil properties for the soil sampling points are shown in

Table 1. Statistics of soil properties for soil sampling points.

	Minimum	Mean	Maximum	Standard Deviation
Soil organic matter content/g/kg	2.01	16.03	36.46	8.98
pH	6.99	8.49	10.02	0.70
Salinity content/g/kg	0.00	5.30	55.00	8.60

.

2.2.2. Field Hyperspectral Data Measurement

The measurements were conducted under calm and cloudless conditions using the SVC HR1024i spectrometer (Spectral Vista Corporation, Poughkeepsie, NY, USA) with a spectral range of 340 to 2510 nm. Field measurements of ground reflectance data were taken between 10:00 and 14:00 after calibrating the spectrometer with a standard whiteboard. Spectra were measured near the sample points, with each spectrum being measured twice using the spectrometer. Simultaneously, a handheld GPS was used to record the longitude and latitude of each sample with a positioning accuracy of 5 m.

2.2.3. Acquisition and Preprocessing of Sentinel-2 MSI

The Sentinel-2 MSI images were downloaded from the website https://dataspace.copernicus.eu/ (accessed on 20 August 2021). The Level 1C data yielded the top-of-atmosphere reflectance. The SNAP software 8.0.0 provided by the European Space Agency was used for atmospheric correction to obtain the bottom-of-atmosphere reflectance. The time of the Sentinel-2 MSI (2A and 2B) data used is shown in the

Table 2. Utilization time and purposes of different data.

Image	Spatial Resolution	Time	Purpose
GF-5 AHSI	30 m	21 April 2019, 4 November 2019, 11 November 2019	Extracting bare soil
Sentinel-2 MSI	10/20 m	27 August 2019	Monitoring crop growth
UAV	2.5 cm	4–5 November 2019	Extracting bare soil
Field survey	--	11–18 May 2018, 19–26 August 2018, 25–28 August 2019, 16–25 May 2021	Constructing the soil attribute prediction model

. The four bands (blue, green, red, and near-infrared) with a spatial resolution of 10 m were resampled to 20 m.

2.2.4. Acquisition and Preprocessing of GF-5 Hyperspectral Data

The GF-5 Advanced Hyper-Spectral Imager (AHSI) operates within a spectral range of 400 to 2500 nm, achieving a spectral resolution of 5 nm in the visible–near-infrared (VNIR) range and 10 nm in the short-wave infrared (SWIR) range, encompassing a total of 330 channels. AHSI hyperspectral imagery with cloud cover less than 10% from April to November 2019 was acquired, and bands overlapping between VNIR and SWIR (spectral channels 151–153) were removed. Subsequently, radiometric calibration and atmospheric correction were performed, and 29 bands significantly affected by water vapor in the atmosphere were eliminated. These bands span the spectral ranges of 390 to 400 nm, 1360 to 1420 nm, 1800 to 1950 nm, and 2505 to 2513 nm, leaving a final set of 301 bands. The time of the GF-5 AHSI data used is shown in the Table 2.

2.3. The Evaluation Framework for Cultivated Land Soil Quality Based on Remote Sensing in Soda–Saline Soil Areas

Drawing on previous research findings and guided by the principles of dominance, regional variability, and remote sensing accessibility, this study constructs a theoretical framework for remote the sensing-based evaluation of cultivated land soil quality, encompassing “demand–indicator–remote sensing response” [6,7]. The study area, Da’an City in Jilin Province, is in a typical soda–saline soil region [29,30]. Soil organic matter is the main source of crop nutrition, which can improve the physical properties of soil, enhance soil fertility, retention, and buffering capacity, and determine the conditions of soil water, fertilizer, air, and heat. In Da’an City, precipitation is low, and evaporation is high, causing soluble salt in soil and groundwater to move upward with water. In soda–saline soil areas, salinity and alkalinity occur together, mainly due to the arid climate and abundant Na⁺. Leaching is weak, and salts such as calcium carbonate and sodium carbonate accumulate in significant amounts in the soil. Hydrolysis can produce OH⁻, resulting in alkalinity. Meanwhile, due to the dry climate and strong evaporation, the soil surface is salinized. Therefore, to evaluate the crop production capacity and production stability grade of cultivated soil, soil organic matter content, salt content, and pH value are selected as key indicators for evaluation.

2.4. Classification of Soil Attribute Grades

Based on the actual situation in the study area, relevant soil remote sensing fundamentals, the “Regulation for gradation on agriculture land quality” (GB/T 28407-2012) [31], the “Cultivated land quality grades” (GB/T 33469-2016) [32], soil organic matter (SOM), salt content (SAC), and pH value are graded. The specific situation is shown in the

Table 3. Grading standards for soil indicators in cultivated land quality evaluation.

Index	Grade	Score Value	Describe
Organic matter content	Grade 1	100	≥40 g/kg
	Grade 2	90	30 g/kg~40 g/kg
	Grade 3	80	20 g/kg~30 g/kg
	Grade 4	70	10 g/kg~20 g/kg
	Grade 5	60	6 g/kg~10 g/kg
	Grade 6	50	<6 g/kg
pH value	Grade 1	100	6.0~7.9
	Grade 2	90	7.9~8.5
	Grade 3	80	8.5~9.0
	Grade 4	30	9.0~9.5
	Grade 5	10	≥9.5
Salt content	Grade 1	100	Desalination, soil desalination, crops are free from seedling loss and ridge breakage caused by salinization, and the salt content in the topsoil is less than 0.1% (soluble salts are mainly sodium carbonate)
	Grade 2	90	Mild salinization, resulting in a 20% to 30% crop seedling loss due to salinization, with a surface soil salt content of 0.1% to 0.3% (soluble salts are mainly sodium carbonate)
	Grade 3	70	Moderate salinization, resulting in a 30% to 50% crop seedling loss due to salinization, with a surface soil salt content of 0.3% to 0.5% (mainly soluble salts, primarily sodium carbonate)
	Grade 4	40	Severe salinization, resulting in crop seedling loss of ≥50% due to salinization, with surface soil salt content ≥ 0.5% (mainly composed of soluble salts such as sodium carbonate)

.

2.5. Soil Attribute Grade Prediction

First, soil organic matter (SOM) and soil salt content (SAC) directly affect soil spectral information. The spatial distribution of soil organic matter and salt content can be obtained by utilizing bare soil spectral signals. In the wild environment, soil reflectance is influenced by SOM and SAC. Attempts have been made to use spectral indices to predict the comprehensive grade of two attributes: SOM and SAC. Second, the pH value primarily depends on the concentration of H⁺ ions in the soil solution, and hydrogen ions lack spectral characteristics. The pH value has no direct impact on soil spectra. The spatial heterogeneity of soil pH is reflected by crop growth information [25].

The relationship between soil properties and remote sensing data exhibits ambiguity and randomness, making it difficult to establish a completely qualitative or quantitative mapping relationship. In fuzzy mathematics, complex phenomena with ambiguity and randomness, as well as intermediate transition phenomena, are described using “membership functions”, breaking the absolute relationship of either/or. An attempt is made to introduce the cloud model from artificial intelligence to characterize the relationship between soil properties and remote sensing data. The cloud model can use the expected value ($Ex$), entropy ($En$), and hyperentropy ($He$) cloud numerical characteristics to comprehensively represent qualitative concepts [33]. $Ex$ is the center of cloud droplet distribution in the domain space, representing the numerical value that best characterizes the qualitative concept; $En$ reflects the degree of dispersion of cloud droplets and is used to measure the fuzziness and randomness of qualitative concepts; $He$ entropy is the entropy of entropy, reflecting the degree of condensation of cloud droplets, and the uncertainty measure of entropy. The greater the hyperentropy, the lower the degree of condensation of the cloud, the thicker the cloud becomes, and the greater the randomness of membership degree [34].

The comprehensive grade of SOM + SAC is reflected using the SI1 and SI2 spectral indices. The red-edge ratio vegetation index (RVIre) was constructed to monitor crop growth changes, indirectly reflecting the differences in soil pH [35]. Spectral indices are described as follows:

$$\mathrm{SI}1={\displaystyle \frac{\sqrt{{\mathsf{\rho }}_{NIR}\times {\mathsf{\rho }}_{NIR}+{\mathsf{\rho }}_{SWIR1}\times {\mathsf{\rho }}_{SWIR1}}}{2}}$$

(1)

$$\mathrm{SI}2={\displaystyle \frac{\sqrt{{\mathsf{\rho }}_{SWIR1}\times {\mathsf{\rho }}_{SWIR1}+{\mathsf{\rho }}_{SWIR2}\times {\mathsf{\rho }}_{SWIR2}}}{2}}$$

(2)

$$\mathrm{RVIre}={\mathsf{\rho }}_{NIR}/{\mathsf{\rho }}_{RE}$$

(3)

In the formula, ${\mathsf{\rho }}_{RE}$, ${\mathsf{\rho }}_{NIR}$, and ${\mathsf{\rho }}_{SWIR}$ represent the reflectance of red-edge, near-infrared, and shortwave infrared bands, respectively.

Based on the comprehensive spectral index and classification criteria, the SOM + SAC comprehensive grade and pH value grade are divided into four grades using the natural breaks classification method. The $E_x$ values corresponding to different grades vary, and as the numerical value that best represents qualitative concepts, $E_x$ can effectively predict the grades of soil properties.

The cloud model is utilized to calculate the membership degrees of soil properties in different grades of spectral indices. The entropy weight method is employed to compute the weights of the membership degrees of soil properties in different grades of spectral indices. The calculation steps are as follows:

Assuming there are $m$ evaluation samples and $n$ spectral indices, a normalized judgment matrix is constructed:

$$X={\left(x_{ij}\right)}_{m\times n}$$

(4)

The entropy of each spectral index is

$$H_j=-{\displaystyle \frac{1}{\ln m}}{\displaystyle \sum }_{i=1}^m{f}_{ij}\ln f_{ij}$$

(5)

$$f_{ij}={\displaystyle \frac{1+x_{ij}}{{{\displaystyle \sum }}_{i=1}^m\left(1+x_{ij}\right)}}$$

(6)

The entropy weight of each spectral index is calculated:

$$W_j={\displaystyle \frac{1-H_j}{{{\displaystyle \sum }}_j^n\left(1-H_j\right)}}$$

(7)

According to the above calculation steps, the weights of the membership degrees of the two spectral indices SI1 and SI2 are 0.56 and 0.44, respectively.

2.6. Comprehensive Evaluation Model of Cultivated Land Soil Quality Based on Remote Sensing

Soil spectroscopy is a comprehensive manifestation of soil properties, and crop growth is also influenced by soil properties. Coupling relationships between the SI1 and SI2 spectral indices and the comprehensive grade of SOM + SAC, as well as the coupling relationship between the RVIre spectral index and pH value, are established. Based on the SI1 and SI2 spectral indices $F_{Soil quality 1}$, the comprehensive grade of soil organic matter + salt content is characterized, and the grade of the soil pH value is represented as $F_{Soil quality 2}$. The cultivated land soil quality grade is obtained using the short-board method $F_{Soil quality}$, as detailed in formulas 15 to 17. The natural discontinuity point classification method is used to cluster the cultivated land soil quality values into four grades, namely, cultivated land soil quality evaluation grade 1 ($F_{Soil quality}$ ≤ 0.5), grade 2 (0.5 < $F_{Soil quality}$ ≤ 1), grade 3 (1 < $F_{Soil quality}$ ≤ 2.5), and grade 4 (2.5 < $F_{Soil quality}$ ≤ 4). Grade 1 is the highest, indicating the best cultivated land soil quality.

$$F_{Soil quality 1}=\alpha \times G_{SI1}+\beta \times G_{SI2}$$

(8)

$$F_{Soil quality 2}=G_{RVI}$$

(9)

$$F_{Soil quality}=MIN\left(F_{Soil quality 1}, F_{Soil quality 2}\right)$$

(10)

In the formula, $F_{Soil quality}$ represents the grade of cultivated soil quality, $G_{SI1}$ and $G_{SI2}$ are used to characterize the comprehensive grade of soil organic matter + salt content, $G_{RVI}$ represents the grade of the soil pH value, $\alpha$ and $\beta$ are the weights of SI1 and SI2 spectral indices, $F_{Soil quality 1}$ and $F_{Soil quality 2}$, respectively, with values ranging from 1 to 4.

2.7. Prediction Model Accuracy Verification

Two-thirds of the samples are randomly selected to establish a prediction model, and one-third of them are used for model validation. The average error ($ME$) and relative error ($RE$) are selected to test the model. The smaller the sum of $ME$ and $RE$ values for the model, the higher the model accuracy. The formulas for $ME$ and $RE$ are as follows:

$$ME={{\displaystyle \sum }}_{i=1}^n{\displaystyle \frac{\left|y_i-f_i\right|}{n}}$$

(11)

$$RE={{\displaystyle \sum }}_{i=1}^n{\displaystyle \frac{\left|y_i-f_i\right|}{n{y}_i}}\times 100\%$$

(12)

3. Results

3.1. Preparation of Remote Sensing Dataset of Bare Soil in Cultivated Land

Bare soil areas in cultivated land were extracted based on GF-5 AHSI hyperspectral data. The single scene image of GF-5 AHSI could not fully cover Da’an City, so a total of six scenes were used: two scenes each on 21 April 2019, 4 November 2019, and 11 November 2019 (Table 2). For detailed research on the quantitative prediction of crop straw coverage using remote sensing, see Gao et al. [36]. Based on GF-5 AHSI hyperspectral data, the average value of the cellulose absorption index (CAI) was constructed across different ranges of crop straw coverage. It can be observed that, when the crop residue coverage (CRC) is less than 20%, the CAI is negative, indicating that the surface coverage is mainly soil, with an average CAI of −0.21. When the CRC is greater than 20%, the CAI is positive, with an average value of 0.54. Due to the influence of adjacent effects on pixel spectral reflectance based on GF-5 AHSI hyperspectral data, the boundary of CRC at 20% is evident in Figure 3. Therefore, a threshold of CRC at 20% was selected to extract bare soil, masking pixels with a high CRC. The extraction of bare soil pixels in cultivated land was based on GF-5 AHSI hyperspectral data.

3.2. Prediction of Soil Attribute Grades Based on Cloud Model

The SI1 and SI2 spectral indices are used to reflect the comprehensive grade of SOM + SAC, while the RVIre index reflects the grade of the pH value. The comprehensive grade classification criteria for SOM + SAC are as shown in

Table 4. Comprehensive classification standards for soil organic matter and salinity.

Grade 1	Grade 2	Grade 3	Grade 4
SOM1, SAC1	SOM2, SAC2	SOM1, SAC4	SOM5, SAC4
SOM1, SAC2	SOM3, SAC1	SOM2, SAC4	SOM6, SAC3
SOM2, SAC1	SOM3, SAC2	SOM3, SAC4	SOM6, SAC4
	SOM3, SAC3	SOM4, SAC3
	SOM4, SAC1	SOM4, SAC4
	SOM4, SAC2	SOM5, SAC2
	SOM5, SAC1	SOM5, SAC3

In the table, the numerical values of SOM and SAC represent the classification of soil organic matter and salt content in Table 3, respectively.

. The grade of pH value is based on the pH value grading standards given in Table 3 for cultivated land quality evaluation. The cloud model is used to analyze the $Ex$, $En$, and $He$ values for different grades of RVIre and pH values. For grades 4 and 5 of pH value, that is, when pH ≥ 9.0, the differences in $Ex$ are small in Da’an City. Therefore, the RVIre grading standards are as follows: the RVIre grades 1, 2, and 3 correspond to grades 1, 2, and 3 of pH value, while RVIre grade 4 corresponds to grades 4 and 5 of pH value in Table 3. Based on the remote sensing dataset of bare soil in cultivated land, the SI1 and SI2 spectral indices are calculated. The cloud model is used to perform fuzzy prediction of the comprehensive grade of SOM and SAC based on the SI1 and SI2 spectral indices. The $Ex$, $En$, and $He$ values for each grade are calculated (

Table 5. Digital characteristics of cloud models for soil organic matter, salinity, and pH values at different grades of cultivated land soil quality evaluation.

Spectral Index	Cloud Digital Features	Grade 1	Grade 2	Grade 3	Grade 4
SI1	Ex	0.17	0.21	0.25	0.27
	En	0.06	0.05	0.05	0.10
	He	0.02	0.01	0.01	0.10
SI2	Ex	0.21	0.24	0.28	0.29
	En	0.04	0.05	0.06	0.11
	He	0.01	0.02	0.02	0.10
RVIre	Ex	6.82	5.78	4.35	2.21
	En	4.97	4.25	3.42	0.96
	He	2.10	1.25	0.88	0.84

), forming the evaluation cloud standard. The $M{E}_c$ and $M{E}_v$ of the SOM + SAC prediction model were 0.74 and 0.68, respectively, and the $M{E}_c$ and $M{E}_v$ of pH values for the prediction model were 0.95 and 0.98, respectively (

Table 6. Prediction and verification of soil organic matter, salinity, and pH grades based on the cloud model.

Spectral Index	$\mathit{M}{\mathit{E}}_{\mathit{c}}$	$\mathit{R}{\mathit{E}}_{\mathit{c}}$/%	$\mathit{M}{\mathit{E}}_{\mathit{v}}$	$\mathit{R}{\mathit{E}}_{\mathit{v}}$/%
Soil organic matter + salinity	0.74	27.69	0.68	22.87
pH value	0.95	33.36	0.98	36.03

In the table, $M{E}_c$ represents the average error of modeling, $R{E}_c$ represents the relative error of modeling, $M{E}_v$ represents the average error of validation, and $R{E}_v$ represents the relative error of validation.

). Based on the cloud model and SI1, SI2, and RVIre spectral indices, the comprehensive grade and pH value grade of SOM + SAC at the county-level can be well predicted. The spatial distribution of the comprehensive grade and pH value grade of soil organic matter + salinity in Da’an City is shown in Figure 4 and Figure 5, respectively. The areas with higher comprehensive grades of soil organic matter + salinity are mainly distributed in the northeast and northwest of Da’an City, while the areas with poorer pH value grade are mainly distributed in southern areas such as Haituo Township and Longzhao Town in Da’an City.

3.3. Comprehensive Evaluation of Cultivated Land Soil Quality Based on Remote Sensing

The spatial distribution of different grades of cultivated soil quality in Da’an City is shown in Figure 6. The proportion of the area of cultivated soil at quality evaluation grade 1, indicating high quality, is 13.73%, and it is mainly distributed in the northeast and northwest areas of Da’an City; the main grades of cultivated soil quality evaluation are grade 2 and grade 3, with medium quality accounting for 80.63%; the proportion of the cultivated soil quality area at evaluation grade 4, indicating poor quality, is relatively low, only 5.65%, and this land is mainly distributed in the southern areas of Dagangzi Town, Haituo Township, and Longzhao Town in Da’an City. The poor quality of cultivated soil in Dagangzi Town is mainly related to the comprehensive grade of soil organic matter + salt content, while the poor quality in Haituo Township is mainly affected by soil pH grade, and the poor quality in Longzhao Town is related to both the comprehensive grade of soil organic matter + salt content and soil pH grade. From Figure 7, it can be seen that the comprehensive grade of soil organic matter + salt content in cultivated soil in Da’an City is mainly grade 2 and grade 3, accounting for 81.35%; the soil pH grade of cultivated soil is mainly grade 1, accounting for 62.35%; the proportion of different grades of cultivated soil quality is more similar to the proportion of comprehensive grade of soil organic matter + salt content.

4. Discussion

The county-level scale is one of the commonly used scales in farmland soil quality evaluation. The evaluation results can provide a more precise decision-making basis for improving crop productivity and grain yield. Soil organic matter and pH are the most frequently proposed soil quality indicators [15]. Soil organic matter is a primary limiting factor for crop yields and serves as one of the most crucial indicators for monitoring soil fertility. The rapid and accurate monitoring of soil organic matter content enables timely adjustments to soil nutrient status, ultimately leading to increased production. In soda–saline soil areas, both salt and alkali components are present simultaneously. Soil alkalinity also plays an important role in the decomposition of soil organic matter. Microbial activity is sensitive to soil pH, and pH mainly affects the transformation of soil organic matter. To evaluate the productivity of cultivated land, soil organic matter content, salt content, and pH value were selected as indicators of cultivated land soil quality in soda–saline soil areas.

Rapid and large-scale mapping of farmland soil quality evaluation indicators is a difficult problem in evaluations. The key issue addressed in this study is how to estimate soil organic matter content, salt content, and pH values based on remote sensing data. Based on the changes in terms of the types, quantities, and spatial structures of apparent characteristics of soil–crop–crop straw systems, remote sensing evaluations of cultivated land soil quality were conducted. First, soil organic matter content and salt content interact with electromagnetic waves, forming spectral absorption and reflection characteristics in specific soil spectral bands. Under certain conditions, specific soil properties can have a major impact on soil spectra, masking the spectral characteristics of other soil components, providing data support for predicting this single soil attribute. Soil organic matter and salt content directly affect soil spectral information. Generally, there is a negative correlation between soil spectra and soil organic matter content [37]. Baumgardner et al. [38,39] contend that, when the soil organic matter content is greater than 20 g/kg, it has a major impact on soil spectral characteristics; when the soil organic matter content is less than 20 g/kg, its ability to mask the spectral characteristics of other soil components is weakened. The range of soil organic matter content is from 2.01 to 36.46 g/kg in Da’an City, increasing the difficulty of remote sensing prediction of soil organic matter. A predictive model based on the relationship between bare soil spectral signals and the comprehensive grade of soil organic matter + salt content has been established. In addition, crop straw hinders the accuracy of the remote sensing extraction of soil attributes. The impact of crop straw on spectral information was analyzed. A threshold of 20% crop residue coverage was set to minimize the interference factors of crop straw and effectively extract bare soil areas. This provides a foundation for the establishment of a prediction model for soil properties.

Secondly, pH value has no direct impact on soil spectra. Crop growth is used to reflect the spatial variability in soil alkalinity. Healthy crops exhibit significant spectral differences from soil and crop straw due to pigments and other substances in the leaf and stem. The “red edge” is one of the most distinct features of green vegetation, referring to the transition platform formed by the strong absorption of red light by chlorophyll and multiple scattering into the near-infrared range. It is usually located between 680 and 750 nm. When crops are thriving, chlorophyll content is high, and the red edge shifts towards longer wavelengths, known as a “red shift”; conversely, if the red edge shifts towards shorter wavelengths, it is referred to as a “blue shift”. The displacement of the crop’s red edge can indicate the health status of the crop and indirectly reflects soil quality. By comprehensively considering the spectral characteristics of green vegetation and the bands of Sentinel-2 MSI imagery, the red-edge ratio vegetation index (RVIre) was constructed to monitor crop growth changes, indirectly reflecting the differences in soil properties. Therefore, a coupling relationship between pH value and RVIre was established, using crop growth to reflect soil alkalinity.

From the perspective of land use status, dry land accounts for 78.27% of cultivated land, with an increase of 36.84% in newly added dry land compared to the 1980s. Dry land with soil quality grades of 3 and 4 account for 36.34% and 5.25% of the total dry land area, respectively, of which 50.92% and 46.43% are newly added cultivated land. Among the newly added dry land with grades 3 and 4, 20.86% and 20.35% were soda–saline soil in the 1980s. Paddy fields account for 21.73% of cultivated land, with an increase of 76.95% in newly added paddy field area compared to the 1980s. Paddy fields with soil quality grades of 3 and 4 account for 60.17% and 7.06% of the total paddy field area, respectively, of which 85.78% and 46.09% are newly added cultivated land. Among the newly added paddy fields with grades 3 and 4, 42.95% and 65.60% were soda–saline soil in the 1980s. The newly added cultivated land with soil quality grades 3 and 4 is mostly reclaimed from soda–saline soil. The low soil quality grade of these cultivated lands is mainly influenced by the low natural endowment of soda–saline soil. This article has assessed the current situation of cultivated land soil quality in Da’an City. In subsequent research, we will add more comparative analyses between the current situation and the past situation. These analyses will provide guidance for future research.

5. Conclusions

In this study, the multi-source remote sensing data of field hyperspectral data, Sentinel-2 MSI, and GF-5 AHSI hyperspectral data were acquired. A prediction model of spectral indices and evaluation indicators were constructed based on the cloud model, ultimately enabling the assessment of farmland soil quality. A theoretical framework was constructed for evaluating farmland soil quality, encompassing “demand–indicator–remote sensing response”. To evaluate the productivity of cultivated land, soil organic matter content, salt content, and pH value were selected as key indicators for the assessment of cultivated land soil quality in Soda–Saline Soil Areas. Moreover, a threshold of 20% crop residue cover was set to mask crop residue cover areas and obtain a bare soil dataset. The spectral indices SI1 and SI2 were utilized to predict the comprehensive grade of soil organic matter + salinity based on the cloud model. The $M{E}_c$ and $M{E}_v$ of the SOM + SAC prediction model were 0.74 and 0.68, respectively. A predictive model based on the relationship between pH and red-edge ratio vegetation index (RVIre) grades was also established. The $M{E}_c$ and $M{E}_v$ of the pH prediction model were 0.95 and 0.98, respectively. Thus, remote sensing can predict the grades of SOM + SAC and pH. The proportion of high-quality (grade 1) areas was 13.73%, mainly distributed in the northeast and northwest of Da’an City; the main grades were medium quality (grades 2 and 3), accounting for 80.63%; and the proportion of poor-quality (grade 4) land was relatively low, only 5.65%, in cultivated soil quality evaluation. The results of this study can provide a data foundation and technical support for remote sensing-based evaluations of cultivated land soil quality. They offer scientific guidance for assessing soil quality in cultivated land in saline–alkali areas at the county level.