Next Article in Journal
Varietal Differences in the Root Systems of Rice (Oryza sativa L.) under Drip Irrigation with Plastic Film Mulch
Previous Article in Journal
Water and Nitrogen Balance under Various Water and Fertilizer Regulation Modes
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Agronomy
Volume 13
Issue 12
10.3390/agronomy13122871
agronomy-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Research on the Inversion Model of Cultivated Land Quality Using High-Resolution Remote Sensing Data

by
Mengmeng Tang
1,
Qiang Wang
1,
Shuai Mei
1,
Chunyang Ying
1,
Zhengbao Gao
2,
Youhua Ma
1 and
Hongxiang Hu
1,*
1
College of Resources and Environment, Anhui Agricultural University, Hefei 230036, China
2
Department of Soil and Fertilizer, Chuzhou Agricultural and Rural Technology Promotion Center, Chuzhou 239000, China
*
Author to whom correspondence should be addressed.
Agronomy 2023, 13(12), 2871; https://doi.org/10.3390/agronomy13122871
Submission received: 10 October 2023 / Revised: 30 October 2023 / Accepted: 21 November 2023 / Published: 22 November 2023
(This article belongs to the Section Precision and Digital Agriculture)
Browse Figures
Versions Notes

Abstract

:
Cultivated land quality is an essential measure of cultivated land production capability. Establishing a cultivated land quality inversion model based on high-resolution remote sensing data provides a scientific basis for regional cultivated land resource management and sustainable utilization. Utilizing field survey data, cultivated land quality evaluation data, and high-resolution remote sensing data, a spectral index-cultivated land quality model was constructed and optimized with the machine learning method, and cultivated land quality inversion and verification in Chuzhou City in 2021 were carried out. The results showed that the distribution of cultivated land quality in the study area depicted with the remote sensing inversion model based on random forest was consistent with the actual cultivated land quality. Although the accuracy of the SVT-CLQ inversion model established using four spectral indices is slightly lower than that of the MSVT-CLQ group established using 15 indices, it can still accurately reflect the distribution of cultivated land quality in the study area. Compared with the two models of the MSVT-CLQ and SVT-CLQ groups, the field survey data of sampling points is reduced, the time and energy of field sampling and analysis are correspondingly saved, the efficiency of cultivated land quality evaluation is improved, and the dynamic monitoring and rapid evaluation of cultivated land quality are realized.

1. Introduction

Cultivated land is an important means of production for agricultural development, and the quality of cultivated land directly impacts food production, quality, and the sustainable development of agriculture [1,2]. Cultivated land is affected by natural and human factors and has a high risk of degradation, which has always been the focus of scholars’ research [3,4]. At present, the relevant quality research mostly adopts the analytic hierarchy process and other evaluation methods [5,6]. Although this method has high evaluation accuracy and a wide application range, it needs to consume a lot of manpower, material, and financial resources, and it is difficult to realize dynamic quality monitoring and rapid evaluation on large regional scales. The application of remote sensing technology provides a new way for rapid quantitative evaluation of cultivated land quality. Using remote sensing technology to improve the efficiency of cultivated land quality evaluation [7], monitor the dynamic change of cultivated land quality, give full play to the productive potential of cultivated land, and promote the rational development and utilization of cultivated land is an important direction of cultivated land quality evaluation [8,9].
In recent years, remote sensing has been used to evaluate the quality of cultivated land, mainly by using remote sensing images as data sources to extract information about current land use status [10] and update the spatial distribution information of cultivated land [11,12]. Remote sensing images can be used to decode relevant soil indicators [13] to build cultivated land quality evaluation models [14]. At present, the focus of research is to establish inversion models using remote sensing images to evaluate soil quality; for example, Peng et al. [15] used the extreme gradient enhancement algorithm (XGBoost) in combination with the variance inflation factor (VIF) to select the optimal crop spectral variables and used the BP neural network (BPNN) algorithm to build a soil fertility evaluation model based on the Sentinel-2 remote sensing images to estimate soil fertility in paddy fields. Duan et al. [16] extracted four aspects and eight indexes of soil fertility, natural conditions, construction level, and cultivated land productivity from multi-source remote sensing data to establish a new cultivated land quality (CLQ) evaluation system. Liu et al. [17] extracted and screened the vegetation index from GF-1 as an index of soil fertility and conducted the cultivated land quality inversion, among which the GA-BPNN model had the best effect. The application of machine learning algorithms in remote sensing tends to increase in recent years [18], and the employment of machine learning methods in the research of cultivated land quality inversion via remote sensing better reflects the dynamics of cultivated land quality, stochasticity, and nonlinearity. In general, the current systematic research on remote sensing inversion of cultivated land quality is still shallow, the inversion method is limited, mostly using the classical statistical model, but the cultivated land quality and evaluation indexes are not a simple linear relationship; deeper relationships need to be explored, and it is necessary to explore the machine learning algorithm further.
Based on summarizing the previous research, this paper extracts cultivated land from the Gaofen-6 PMS images of Chuzhou City and extracts spectral information and topographic indexes of cultivated land using remote sensing images and a digital elevation model. Various analysis methods are used to construct and optimize the inversion model, and the model’s accuracy is analyzed and applied from different angles. This paper aims to explore the new model and method of cultivated land quality inversion, improve the remote sensing inversion system for cultivated land quality, enhance the practical value of the model, and better serve the cultivated land quality evaluation.

2. Materials and Methods

2.1. Study Area

Chuzhou City is located in the easternmost part of Anhui Province, China. Chuzhou City is situated in the lower sections of the Yangtze River plain as well as the mountainous areas between the Yangtze and Huaihe rivers. With four distinct seasons, a pleasant climate, simultaneous rainfall, and heat, Chuzhou City has a humid monsoon climate. The plain region in the lower Yangtze River basin and the hilly Jianghuai region, with the terrain being higher in the west and lower in the east, make up the bulk of Chuzhou City. There are three different types of landforms in the city, making up 8.2%, 40.4%, and 39.2% of the total land area, respectively: plain areas, hilly areas, and downland areas. Chuzhou’s land is split into two major basins: the Yangtze River and the Huai River, which both cross the Jianghuai watershed. Mingguang County, Fengyang County, Dingyuan County, and some portions of Tianchang County belong to the Huai River Basin, which makes up around 66.9% of the city’s total land. Chuzhou urban area, Lai’an County, Quanjiao County, and Tianchang County make up 33.1% of the city’s total area.
The hilly area usually has an elevation of more than 100 m and a relative height of more than 50 m. Most of the slope deposits and alluvium on the hilly platform have exposed bedrock or gravel. However, there are diluvium and coarse slope deposits beneath the slope, along with exposed bedrock or gravel. In the northwest of Dingyuan County, around Fengyang Mountain in the southwest of Fengyang County, in the northwest of Mingguang County, in the northeast and southwest of Nanqiao District, in the northwest of Lai’an County, in the southeast of Quanjiao County, and on the southwest border of Tianchang County, there are platforms and undulating terrain surrounding hills. The average elevation of these areas is between 50 and 100 m. The soil is mostly composed of Xiashu loess and has a deep soil layer. The Xiashu loess is an eolian loess from the Quaternary Late Pleistocene that is found in China’s middle and lower portions of the Yangtze River. The plain area is mostly located along the banks of the Chuhe and Huaihe rivers as well as in the lakefront areas of the Nvshan and Gaoyou lakes. It has a relative height of less than 10 m and an altitude of less than 50 m.
Chuzhou City has a large area of cultivated land, which is dominated by paddy fields, accounting for 68% of the city’s cultivated land. Dingyuan County, Fengyang County, Mingguang City, and other counties have a large area of cultivated land, accounting for 59.15% of the city’s cultivated land. Quanjiao County, Lai’an County, Dingyuan County, Tianchang City, and other counties have larger areas of paddy fields, accounting for 71.10% of the city’s paddy fields. The geographic location of the study area is shown in Figure 1.

2.2. Sample Point

With a total of 1233 sampling points, the study’s sampling points span 8 counties in Chuzhou City. The following information is included in the sampling point data: sampling time; latitude and longitude; administrative division name; topographic areas; cultivated layer texture; texture configuration; soil organic matter; soil available phosphorus; soil available potassium; pH; soil bulk density; barriers; effective soil layer thickness; heavy metal content; biodiversity; etc. Biodiversity in cultivated land quality assessment is an important indicator of cultivated land health. Human involvement has the potential to reduce the biodiversity of the soil on cultivated land. In this study, biodiversity is classified as rich, generally rich, and not rich after being thoroughly assessed through a field survey and expert experience. A field sample of cultivated land in the study area was conducted in 2021 using a stainless steel soil extraction auger for operation, the collection depth was 0~20 cm, a plum-shaped sampling route was adopted, the sampling volume of each sample point was kept uniform and consistent, the soil samples of each sample point were mixed thoroughly, 1 kg of soil samples were retained in the end using the tetrad method, and the accurate coordinates of the sample points were determined by using a GPS locator at the same time of sampling. The soil samples collected in the field were air-dried, and crop roots, stones, and other foreign materials were removed. The soil samples were crushed and passed through a 2 mm aperture sieve, and other soil data were obtained through experimental analysis. Soil acidity and alkalinity were measured using potentiometry according to Chinese agricultural industry standards (NY-T 1377-2007) [19]. Soil organic matter was measured using the potassium dichromate method according to Chinese agricultural industry standards (NY/T 1121.6-2006) [20]. Soil available phosphorus was measured using the Mo-Sb colorimetric method according to Chinese agricultural industry standards (NY/T 1121.7-2014) [21]. Soil available potassium was measured using the extraction with neutral ammonium acetate solution and determination with the flame photometer method according to Chinese agricultural industry standards (NY/T 889-2004) [22].

2.3. Image Data Acquisition and Preprocessing

According to the regional crop cultivation practices and remote sensing images without cloud cover in Chuzhou City, the remote sensing images of various periods were chosen for land use information extraction and cultivated land quality inversion. Gaofen-6 is a low-orbit optical remote sensing satellite and the first high-precision agricultural observation satellite in China, characterized by a combination of high resolution and wide coverage. Gaofen-6 is equipped with a 2 m full color/8 m multispectral high-resolution camera (PMS) and a 16 m multispectral medium-resolution wide-frame camera (WFV), with a PMS observation width of 90 km and a WFV observation width of 800 km. The revisit period of the Gaofen-6 satellite is four days, and after networking with Gaofen-1, the revisit period can reach two days. The specific information from the remote sensing images is shown in Table 1, where the Gaofen-6 PMS images were used to extract land use information and the soil spectral index. Without clouds, the images may cover the entire research region, and the image quality is excellent.
The image was preprocessed with ENVI 5.3 (Exelis Visual Information Solutions, Boulder, CO, USA). The image was first radiometrically calibrated, which converts the image’s DN value to surface spectral reflectance. The calculation formula is as follows:
L = G a i n × D N + B i a s
L is the radiance, and G a i n and B i a s are calibration coefficients. Radiometric calibration coefficients are derived from the 2021 Land Observing Satellite Parameters published by the China Center for Resources Satellite Data and Application, and the absence of a labeled Bias value represents a Bias value of 0. The band information and the value of the radiance calibration parameter of the Gaofen-6 image are shown in Table 2.
The atmospheric adjustment was then performed using ENVI’s FLAASH module to obtain the true reflectance value of surface features. Using on-site control points and one DEM picture with a 30 m resolution, the image of the research region in this study was geometrically rectified. A remote-sensing image covering the full research region was produced after image mosaicking, vector cropping, and mask extraction.

2.4. Spectral Indicators and Topographic Indicators

Using a summary of pertinent study findings, remote sensing spectrum indicators that could be connected to cultivated land quality were chosen. This study selected B1, B3, B4, and B6 from Gaofen-6 WFV images as single-band spectral indicators. These four bands represent the blue band, the red band, the near-infrared band, and the red edge band, which is sensitive to vegetation conditions. When it is difficult to obtain remote sensing images of bare soil due to the influence of weather and surface cover, the quality of cultivated land can be indirectly estimated through surface vegetation. The vegetation index combines the visible light and near-infrared reflectance spectral information sensitive to vegetation with sensors and is an indicator that can be directly obtained through remote sensing to reflect the growth status of vegetation [23,24]. The vegetation index, as an important indicator that can reflect soil quality, was selected as the model variable in this study. Five vegetation indices with a wide application range and strong universality were selected for analysis. The vegetation indices and their calculation formulas are shown in Table 3, which were calculated by using the Band math tool in the ENVI software. For the outliers and invalid values encountered in the calculation of the vegetation indices, the outliers were excluded to improve the accuracy of the data.
The terrain of the study area is divided into hilly areas, downland areas, and plain areas. The altitude and terrain of the research area are shown in Figure 2. The terrain indicators are added to the inversion model because of the great changes in terrain undulation and the great influence of terrain on the quality of cultivated land [28]. Slope is one of the topographic factors that can most directly reflect the intensity of terrain undulation and elevation change. In soil erosion and terrain water flow simulation analysis, the slope factor is also a key factor affecting soil erosion resistance and water flow paths. In the study of surface terrain moisture index, the slope is a characterization of the soil’s ability to produce water [29].
The stream power index (SPI) is an index used to measure the three-dimensional spatial strength of water flow [30], which can indicate the strength distribution and rate of water flow. The larger the value of SPI, the greater the runoff concentration, which may lead to soil erosion.
The topographic moisture index (TWI) is an important quantitative index of convective path length, flow-producing area, and soil-runoff-generating capacity based on the digital elevation model [31]. The calculation formulas are as follows:
S P I = l n ⁡ ( S C A × t a n β )
T W I = ln ⁡ S C A t a n β
S C A is the specific catchment area, and t a n β is the slope. Specific catchment area (SCA) refers to the upstream catchment area per unit contour length or the runoff area per unit contour, which describes the catchment capacity of surface soil, and is an important parameter for various geomorphic structures and hydrological models.

2.5. Land Use Classification

The current methods applied to remote sensing image classification mainly include supervised classification, unsupervised classification, CART decision tree classification, etc. [32,33]. In this study, random forest classification was selected to perform land use classification. Combining the field survey data and the feature characteristics of remote sensing images, cultivated land and other land classes were classified and selected on remote sensing images in the study area. The training samples were selected based on the principles of representativeness and globalization. There are 56 cultivated land and 169 other land classes in the training samples selected in this study, and the samples in the selected training area were tested for the separability between land classes with the sample separability degree, and when the separability degree is more than 1.9, it indicates that the samples are well separable from each other.

2.6. Cultivated Land Quality Level Evaluation

Cultivated land quality, as a comprehensive concept, cannot be simply represented by one or two indicators. Cultivated land quality involves many aspects, such as the health status of cultivated land, standing conditions, soil nutrients, soil physicochemical properties, and farmland management [34,35]. In this study, 15 evaluation indexes were selected from the above six aspects to establish the evaluation index system of cultivated land quality level when evaluating cultivated land quality level. The classification of cultivated land quality evaluation indicators is shown in Table 4.
These indicators include farmland forest reticulation and topography, which represent the conditions of farmland, and three profile characteristics, such as obstacle factors, effective soil layer thickness, and texture configuration. The soil bulk density, acidity and alkalinity, and soil texture in the till layer represent soil physicochemical properties, soil available phosphorus, soil available potassium, and soil organic matter content characterizing soil nutrients, cleanliness, and biodiversity, reflecting the health of farmland. It also includes two indexes, drainage capacity and irrigation capacity, which reflect farmland management status.
Farmland forest reticulation is the ratio of the protected area of forest belts around farmland to the total area of farmland, playing an important role in the microclimate, wind and sand prevention, and pollution alleviation of farmland and its surrounding areas. On-site investigation of the protected area of forest belts around farmland and the total area of farmland, calculation of farmland forest network rate, and comprehensive judgment of farmland forest network degree, divided into high, medium, and low.
Cleanliness is an important indicator and reflects the health status of arable soil, mainly referring to the degree to which pollutants in arable soil do not have adverse or harmful effects on the ecosystem and human health. This study calculates the Nemero index based on the heavy metal content of the sampling points to determine the cleanliness level of the study area.
Irrigation is a key factor in ensuring crop water consumption, which directly affects the farming system and farmland production capacity. An on-site investigation of the type, location, irrigation method, and irrigation volume of water sources can be conducted, and the degree to which irrigation water consumption can be met in years of irrigation, which can be divided into fully satisfied, satisfied, basically satisfied, and not satisfied, can be comprehensively judged.
To ensure the normal growth of crops, timely drainage of surface water in farmland, and the ability to effectively control and reduce groundwater levels, an on-site investigation on drainage methods and the current status of drainage facilities can be conducted, the ability of farmland can be comprehensively assessed to ensure normal crop growth, surface water accumulation can be timely removed, and groundwater levels can be effectively controlled and reduced, which can be divided into fully satisfied, satisfied, basically satisfied, and not satisfied.
The 15 evaluation indicators were divided into textual and numerical evaluation indicators, and each map unit was assigned to obtain the corresponding attribute data. Table 4 displays the types and weights of evaluation indicators. Using techniques such as surface substitution and connecting attribute mapping, the textual indicators (topographic portion, farmland forest network, texture configuration, obstacle factors, tillage texture, biodiversity, cleanliness, irrigation capacity, drainage capacity, etc.) were assigned to each evaluation unit; the numerical indicators (effective soil thickness, bulk density, pH, soil organic matter, soil available phosphorus, soil available potassium, etc.) were spatially interpolated and superimposed on the map of each evaluation unit using the regional statistical method. After processing the spatial interpolation of numerical indicators, values were assigned to the evaluation unit by superimposing the evaluation unit map and the regional statistical method. These indicators included effective soil layer thickness, pH, soil organic matter, soil available phosphorus, and soil available potassium.
The weights of the indicators used in the assessment of the quality level of cultivated land were established by combining the hierarchical analysis approach with the Delphi method. First, the indicators were assigned weights using the Delphi approach. These weights were then utilized to categorize the indicators using the hierarchical analysis method. The judgment matrix was utilized to establish the weights of each indication after comparing the significance of each indicator within each indicator category. The comprehensive cultivated land quality index of each evaluation unit was computed using the weighted sum method following the establishment of the cultivated land quality level evaluation index system and the determination of the weights of each index. The calculation formula is as follows:
I F I = ∑ F i × C i
I F I represents the comprehensive index of cultivated land quality; F i is the i-th factor evaluation (score); and C i is the combined weight of the i-th factor.
Combined with the actual situation of the study area, based on the calculated comprehensive cultivated land quality index of each evaluation unit, the natural breakpoint method was used for grading, which was categorized into five grades, namely, higher, high, medium, low, and lower. The comprehensive index range of cultivated land quality level is shown in Table 5.

2.7. Cultivated Land Quality Inversion Model

In order to facilitate the use of machine learning algorithms and ensure comparability between different modeling methods, this study uses the evaluation unit, where 1233 sampling points are located, as the sample dataset. The 1233 sampling points are distributed in each topographic sub-district of the study area, and the locations of the distribution of the sampling points are shown in Figure 1.
This study adopted the 10-fold cross-validation method, in which the dataset was divided into 10 copies, and 9 of them were used as the training set and 1 as the test set in turn, and the validation was carried out. Each validation yielded a corresponding accuracy, and the average of the correctness of the ten results was used as an estimate of the accuracy of the algorithm. The use of 10-fold cross-validation ensures that all sample datasets have been used as both the training and test datasets, reduces chance due to a single division of the training and test datasets, and makes full use of the existing dataset for multiple divisions. Cross-validation is used to reduce chance and improve model generalization.
The random forest machine learning algorithm was used to model with the measured indexes, spectral indexes, vegetation indexes, and topographic indexes as the input variables. The composite cultivated land quality indexes were used as the output variables, and the accuracy of the inversion model was evaluated with R2 and the root-mean-square error (RMSE).
Random forest is a method based on the decision tree combination induced by Breiman, which is an integrated learning algorithm in sample space and feature space simultaneously. Each decision tree in random forest depends on a random vector consisting of parameters determined via training, and each tree forms an independently distributed set of training samples with the Bagging algorithm, uses these sets of training samples for training, and at the same time selects some of the features in the feature set for the construction of the decision tree.
There were 15 variable indicators that were entered into the model, with the measured indicators including soil organic matter, soil bulk density, effective soil thickness, and pH. The overall research workflow is shown in Figure 3. The variable indicators were further divided into four categories: measured indicators, spectral indicators, vegetation indices, and topographic indicators. The four bands—B1, B3, B4, and B6—of the GF-6 WFV images served as the spectral indicators. The vegetation indices included NDVI, DVI, RVI, EVI, and SAVI. Terrain indices include slope, TWI, and SPI.

3. Results

In this study, the data collected in the field were utilized as training samples, and random forest classification was used to finish the land-use classification, with a classification accuracy of 99.67% and a Kappa coefficient of 0.99. This study also used visually interpreted plots to aid in the classification process. Figure 4 depicts how the research area’s cultivated land is distributed. A small amount of cultivated land is also concentrated in the southwestern and southern regions with flat terrain. The cultivated land in the study area is primarily concentrated in the northern, eastern, and western regions. In the central part of the study area, there are urban areas and mountains, and the distribution of cultivated land is relatively scattered.
In the evaluation results of cultivated land quality grade, the first- and second-class land are the best cultivated land in the study area, which is mainly distributed in the plain area. Both first-class and second-class cultivated land have relatively deep soil layers, and the regions’ evaluation indicators are generally good. The crop growing conditions are met with the irrigation and drainage systems, the soil is fertile, and the production performance is high. The study area’s third- and fourth-class land is primarily the middle grain production area. The distribution area’s topography is primarily plain and hilly hillock terrain, with some undulation, a deep soil layer, good irrigation and drainage conditions, and soil nutrients at the middle to upper level, suitable for crop growth. Fifth-class land is the poorer quality of cultivated land in the study area. The distribution area is mainly mountainous and hilly areas, with relatively large terrain undulations. The overall quality of cultivated land is slightly lower, and water conservation measures are essentially limited to providing for the drainage and irrigation of cultivated land. There is a light to medium degree of soil erosion in the drylands, the soil layer of the gully and valley alluvial fields is shallow, the barrier layer is high, the effective nutrient content is generally low, the suitability for planting is poor, and the level of yield is relatively low.
In the random forest model training, the relative influence of each auxiliary variable was calculated, and the spectral, vegetation index, and terrain indicators with high relative influence were screened for modeling indicator preferences. The model input variable group and model accuracy are shown in Table 6. The model input variables were grouped according to indicator type and relative impact; the MSVT-CLQ group included all 15 indicators; the SVT-CLQ group included B4, B6, EVI, and TWI; and the cultivated land quality inversion was performed separately to compare the model accuracy and further screen the cultivated land quality inversion model.
Comparing and analyzing the result map of cultivated land quality level evaluation (Figure 5) and the distribution map of remote sensing inversion of MSVT-CLQ (Figure 6), it can be seen that the areas with lower cultivated land quality grades in the study area are concentrated in the mountainous and granitic areas with higher elevation, while the cultivated land quality in the plain area is higher, and the overall cultivated land quality of the MSVT-CLQ result is on the higher side compared with that of the CLQ result.
The analysis of the area comparison of various cultivated land quality evaluation methods is presented in Table 7, which indicates that the overall distribution of cultivated land quality levels in the study area is normal and that the majority of the cultivated land quality levels are concentrated in the second, third, and fourth grades of cultivated land. The cultivated land quality levels are concentrated in the second and third grades of cultivated land, and the overall cultivated land quality level is greater when comparing the findings of MSVT-CLQ and CLQ. The first-level area makes up a larger percentage of the total area. The percentage of each level of the cultivated land quality of SVT-CLQ and CLQ is similar, the same as the trend, but there is more first-level cultivated land. The results of the two inversion models are basically consistent with the area share of the cultivated land quality evaluation results, which proves that the inversion models are suitable for the prediction of cultivated land quality in the study area.

4. Discussion

Cultivated land quality monitoring and evaluation is generally carried out through field investigation and analysis of laboratory data collection [36], on the basis of which cultivated land quality evaluation is carried out. This evaluation method needs a lot of time and money, the efficiency is relatively low, the data obtained cannot meet the current needs of cultivated land quality evaluation, and it is difficult to realize the rapid and accurate evaluation of cultivated land quality. In this study, we extracted cultivated land evaluation units based on remote sensing images, which can refine the monitoring of cultivated land quality to the image element level and realize the quantitative monitoring of cultivated land quality. Most of the existing cultivated land quality evaluations use hierarchical analysis for the construction of evaluation models, which is highly subjective. In this paper, the importance of model variables is determined with the help of principal component analysis, and machine learning algorithms are used for the remote sensing inversion of cultivated land quality. The selection of model input variables is a prerequisite for accurate cultivated land quality inversion, and most of the previous studies on remote sensing inversion of cultivated land quality selected the vegetation index as the input variable of the model and inverted the cultivated land quality with the vegetation covered on the surface and the vegetation growth [37], which makes it difficult to accurately reflect the cultivated land quality.
In this study, we added remote sensing spectral indicators, vegetation index, and topographic indicators to the field survey and sampling point data for the evaluation of cultivated land quality. We also used a random forest algorithm to extract cultivated land, performed remote sensing inversion of cultivated land quality, improved the cultivated land quality evaluation unit, and increased the efficiency and accuracy of cultivated land quality evaluation. Although this study utilized spectral indicators and vegetation index as input variables of the model, it was limited by the temporal and spatial resolution of remote sensing images [38] and did not consider the differences in spectral images of different fertility periods of crops [32,39], which affected the accuracy of cultivated land quality evaluation to a certain extent. The cultivated land types, such as dryland and paddy fields, can be further subdivided [40] in order to improve the accuracy of the cultivated land quality inversion model, and the corresponding inversion models can be established on different cultivated lands. In this study, the random forest algorithm was adopted to extract the cultivated land as the evaluation unit in the extraction of the research scope. Correspondingly, in order to verify the reliability of the evaluation indexes and inversion model proposed in the study, a larger range for the sample size should be used to collect cultivated land information based on the cultivated land quality grade from high to low.

5. Conclusions

This paper takes Chuzhou City, Anhui Province, China, as the study area. We extract cultivated land based on Gaofen-6 remote sensing images and establish a cultivated land quality inversion model based on remote sensing images using a digital elevation model, vegetation index, and spectral index. We compared the inversion results with the results of the cultivated land quality level evaluation and carried out the inversion model preference in terms of model accuracy, inversion effect, and application analysis to draw the following conclusions:
1. The input variables of the selected cultivated land quality model can reflect the cultivated land quality in the study area, and the input variables include field survey data of sampling points, spectral index, vegetation index, and topographic index. In this study, a cultivated land quality inversion model was developed to be applicable to the evaluation of cultivated land quality in low-elevation areas such as plains and hills. Random forests are applied to establish the remote sensing inversion model of cultivated land quality and predict the distribution of cultivated land quality in the study area. The general trend is basically consistent with the actual cultivated land quality, which proves that it is feasible to use a machine learning algorithm to simulate the complex multivariate nonlinear relationship between cultivated land quality and a remote sensing spectral image and accurately evaluate cultivated land quality;
2. In the SVT-CLQ group, which selected spectral index, vegetation index, and topographic index as model input variables, the model R2 was 0.91, the RMSE was 3.13, and the distribution of cultivated land in each grade was similar to the proportion of cultivated land distribution area in CLQ and MSVT-CLQ, and the trend of cultivated land distribution was the same. Compared with the MSVT-CLQ group, although the model input variables lacked the field survey data of the sampling points, the inversion results were consistent with reality, indicating that the use of remote sensing imagery can be used for the inversion of cultivated land quality, reducing the time and energy for field sampling and analysis and laboratory tests, and improving the efficiency of cultivated land quality evaluation.

Author Contributions

M.T. performed the experiments and wrote the paper. M.T., Y.M. and Q.W. conceived and designed the framework. Y.M. and H.H. conceptualized and formulated overarching research goals and aims. S.M. and Z.G. contributed to data preparation and analysis. S.M. and C.Y. performed the experiments. All authors discussed the results and contributed to the final manuscript. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the modern agricultural remote sensing monitoring system construction and industrial application of the Science and Technology Major Project in Anhui Province, China (No. 202003a06020002).

Data Availability Statement

The data are available on request from the authors. The data are not publicly available due to confidentiality agreements.

Conflicts of Interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as potential conflict of interest.

References

  1. Cristina Rulli, M.; D’Odorico, P. Food appropriation through large scale land acquisitions. Environ. Res. Lett. 2014, 9, 064030. [Google Scholar] [CrossRef]
  2. Foley, J.A.; Ramankutty, N.; Brauman, K.A.; Cassidy, E.S.; Gerber, J.S.; Johnston, M.; Mueller, N.D.; O’Connell, C.; Ray, D.K.; West, P.C.; et al. Solutions for a cultivated planet. Nature 2011, 478, 337–342. [Google Scholar] [CrossRef]
  3. Rinot, O.; Levy, G.J.; Steinberger, Y.; Svoray, T.; Eshel, G. Soil health assessment: A critical review of current methodologies and a proposed new approach. Sci. Total Environ. 2019, 648, 1484–1491. [Google Scholar] [CrossRef]
  4. Fathizad, H.; Ardakani, M.A.H.; Heung, B.; Sodaiezadeh, H.; Rahmani, A.; Fathabadi, A.; Scholten, T.; Taghizadeh-Mehrjardi, R. Spatio-temporal dynamic of soil quality in the central Iranian desert modeled with machine learning and digital soil assessment techniques. Ecol. Indic. 2020, 118, 106736. [Google Scholar] [CrossRef]
  5. Everest, T.; Sungur, A.; Özcan, H. Applying the Best–Worst Method for land evaluation: A case study for paddy cultivation in northwest Turkey. Int. J. Environ. Sci. Technol. 2021, 19, 3233–3246. [Google Scholar] [CrossRef]
  6. Tashayo, B.; Honarbakhsh, A.; Azma, A.; Akbari, M. Combined Fuzzy AHP–GIS for Agricultural Land Suitability Modeling for a Watershed in Southern Iran. Environ. Manag. 2020, 66, 364–376. [Google Scholar] [CrossRef] [PubMed]
  7. Qader, S.H.; Dash, J.; Atkinson, P.M. Forecasting wheat and barley crop production in arid and semi-arid regions using remotely sensed primary productivity and crop phenology: A case study in Iraq. Sci. Total Environ. 2018, 613–614, 250–262. [Google Scholar] [CrossRef]
  8. Liu, L.; Zhou, D.; Chang, X.; Lin, Z. A new grading system for evaluating China’s cultivated land quality. Land Degrad. Dev. 2020, 31, 1482–1501. [Google Scholar] [CrossRef]
  9. Chen, S.; Lin, B.; Li, Y.; Zhou, S. Spatial and temporal changes of soil properties and soil fertility evaluation in a large grain-production area of subtropical plain, China. Geoderma 2020, 357, 113937. [Google Scholar] [CrossRef]
  10. Hoyos, L.; Cabido, M.; Cingolani, A. A Multivariate Approach to Study Drivers of Land-Cover Changes through Remote Sensing in the Dry Chaco of Argentina. ISPRS Int. J. Geo-Inf. 2018, 7, 170. [Google Scholar] [CrossRef]
  11. Tong, X.-Y.; Xia, G.-S.; Lu, Q.; Shen, H.; Li, S.; You, S.; Zhang, L. Land-cover classification with high-resolution remote sensing images using transferable deep models. Remote Sens. Environ. 2020, 237, 111322. [Google Scholar] [CrossRef]
  12. Zhang, S.; Ban, X.; Xiao, T.; Huang, L.; Zhao, J.; Huang, W.; Liang, D. Identification of Soybean Planting Areas Combining Fused Gaofen-1 Image Data and U-Net Model. Agronomy 2023, 13, 863. [Google Scholar] [CrossRef]
  13. Khosravi, V.; Doulati Ardejani, F.; Yousefi, S.; Aryafar, A. Monitoring soil lead and zinc contents via combination of spectroscopy with extreme learning machine and other data mining methods. Geoderma 2018, 318, 29–41. [Google Scholar] [CrossRef]
  14. Binte Mostafiz, R.; Noguchi, R.; Ahamed, T. Agricultural Land Suitability Assessment Using Satellite Remote Sensing-Derived Soil-Vegetation Indices. Land 2021, 10, 223. [Google Scholar] [CrossRef]
  15. Peng, Y.; Liu, Z.; Lin, C.; Hu, Y.; Zhao, L.; Zou, R.; Wen, Y.; Mao, X. A New Method for Estimating Soil Fertility Using Extreme Gradient Boosting and a Backpropagation Neural Network. Remote Sens. 2022, 14, 3311. [Google Scholar] [CrossRef]
  16. Duan, D.; Sun, X.; Liang, S.; Sun, J.; Fan, L.; Chen, H.; Xia, L.; Zhao, F.; Yang, W.; Yang, P. Spatiotemporal Patterns of Cultivated Land Quality Integrated with Multi-Source Remote Sensing: A Case Study of Guangzhou, China. Remote Sens. 2022, 14, 1250. [Google Scholar] [CrossRef]
  17. Liu, S.; Peng, Y.; Xia, Z.; Hu, Y.; Wang, G.; Zhu, A.; Liu, Z. The GA-BPNN-Based Evaluation of Cultivated Land Quality in the PSR Framework Using Gaofen-1 Satellite Data. Sensors 2019, 19, 5127. [Google Scholar] [CrossRef]
  18. Belgiu, M.; Drăguţ, L. Random forest in remote sensing: A review of applications and future directions. ISPRS J. Photogramm. Remote Sens. 2016, 114, 24–31. [Google Scholar] [CrossRef]
  19. NY/T 1377-2007; Determination of pH in Soil. Ministry of Agriculture and Rural Affairs, PRC: Beijing, China, 2007.
  20. NY/T 1121.6-2006; Soil Testing Part 6: Method for Determination of Soil Organic Matter. Ministry of Agriculture and Rural Affairs, PRC: Beijing, China, 2006.
  21. NY/T 1121.7-2014; Soil testing—Part 7: Method for Determination of Available Phosphorous in Soil. Ministry of Agriculture and Rural Affairs, PRC: Beijing, China, 2014.
  22. NY/T 889-2004; Determination of Exchangeable Potassium and Non-Exchangeable Potassium Content in Soil. Ministry of Agriculture and Rural Affairs, PRC: Beijing, China, 2004.
  23. Marti, J.; Bort, J.; Slafer, G.A.; Araus, J.L. Can wheat yield be assessed by early measurements of Normalized Difference Vegetation Index? Ann. Appl. Biol. 2007, 150, 253–257. [Google Scholar] [CrossRef]
  24. Richardson, A.J.; Wiegand, C.L. Distinguishing vegetation from soil background information. Photogramm. Eng. Remote Sens. 1977, 43, 1541–1552. [Google Scholar]
  25. Pearson, R.L.; Miller, L.D. Remote Mapping of Standing Crop Biomass for Estimation of the Productivity of the Shortgrass Prairie. Remote Sens. Environ. 1972, VIII, 1355. [Google Scholar]
  26. Zhang, Y.J.; Alfredo, R.H.; Kamel, D.; Tomoaki, M. Development of a two-band enhanced vegetation index without a blue band. Remote Sens. Environ. 2008, 112, 3833–3845. [Google Scholar]
  27. Huete, A.R. A soil-adjusted vegetation index (SAVI). Remote Sens. Environ. 1988, 25, 295–309. [Google Scholar] [CrossRef]
  28. Zhu, P.; Zhang, G.; Yang, Y.; Wang, C.; Chen, S.; Wan, Y. Infiltration properties affected by slope position on cropped hillslopes. Geoderma 2023, 432, 116379. [Google Scholar] [CrossRef]
  29. Moore, I.D.; Grayson, R.B.; Ladson, A.R. Digital terrain modelling: A review of hydrological, geomorphological, and biological applications. Hydrol. Process. 1991, 5, 3–30. [Google Scholar] [CrossRef]
  30. Jaafarzadeh, M.S.; Tahmasebipour, N.; Haghizadeh, A.; Pourghasemi, H.R.; Rouhani, H. Groundwater recharge potential zonation using an ensemble of machine learning and bivariate statistical models. Sci. Rep. 2021, 11, 5587. [Google Scholar] [CrossRef]
  31. Arabameri, A.; Rezaei, K.; Cerda, A.; Lombardo, L.; Rodrigo-Comino, J. GIS-based groundwater potential mapping in Shahroud plain, Iran. A comparison among statistical (bivariate and multivariate), data mining and MCDM approaches. Sci. Total Environ. 2019, 658, 160–177. [Google Scholar] [CrossRef]
  32. Wang, L.; Wang, J.; Liu, Z.; Zhu, J.; Qin, F. Evaluation of a deep-learning model for multispectral remote sensing of land use and crop classification. Crop J. 2022, 10, 1435–1451. [Google Scholar] [CrossRef]
  33. Deng, Z.; Lu, Z.; Wang, G.; Wang, D.; Ding, Z.; Zhao, H.; Xu, H.; Shi, Y.; Cheng, Z.; Zhao, X. Extraction of fractional vegetation cover in arid desert area based on Chinese GF-6 satellite. Open Geosci. 2021, 13, 416–430. [Google Scholar] [CrossRef]
  34. Vogel, H.-J.; Bartke, S.; Daedlow, K.; Helming, K.; Kögel-Knabner, I.; Lang, B.; Rabot, E.; Russell, D.; Stößel, B.; Weller, U.; et al. A systemic approach for modeling soil functions. Soil 2018, 4, 83–92. [Google Scholar] [CrossRef]
  35. Raiesi, F.; Kabiri, V. Identification of soil quality indicators for assessing the effect of different tillage practices through a soil quality index in a semi-arid environment. Ecol. Indic. 2016, 71, 198–207. [Google Scholar] [CrossRef]
  36. Greiner, L.; Nussbaum, M.; Papritz, A.; Fraefel, M.; Zimmermann, S.; Schwab, P.; Grêt-Regamey, A.; Keller, A. Assessment of soil multi-functionality to support the sustainable use of soil resources on the Swiss Plateau. Geoderma Reg. 2018, 14, e00181. [Google Scholar] [CrossRef]
  37. Li, C.; Wang, J.; Ge, L.; Zhou, Y.; Zhou, S. Optimization of Sample Construction Based on NDVI for Cultivated Land Quality Prediction. Int. J. Environ. Res. Public Health 2022, 19, 7781. [Google Scholar] [CrossRef]
  38. Eilola, S.; Käyhkö, N.; Fagerholm, N. Lessons learned from participatory land use planning with high-resolution remote sensing images in Tanzania: Practitioners’ and participants’ perspectives. Land Use Policy 2021, 109, 105649. [Google Scholar] [CrossRef]
  39. Blickensdörfer, L.; Schwieder, M.; Pflugmacher, D.; Nendel, C.; Erasmi, S.; Hostert, P. Mapping of crop types and crop sequences with combined time series of Sentinel-1, Sentinel-2 and Landsat 8 data for Germany. Remote Sens. Environ. 2022, 269, 112831. [Google Scholar] [CrossRef]
  40. Griffiths, P.; Nendel, C.; Hostert, P. Intra-annual reflectance composites from Sentinel-2 and Landsat for national-scale crop and land cover mapping. Remote Sens. Environ. 2019, 220, 135–151. [Google Scholar] [CrossRef]
Agronomy 13 02871 g001
Figure 1. Study area’s geographic location, the sampling sites’ dispersion, and the Gaofen-6 image map.
Figure 1. Study area’s geographic location, the sampling sites’ dispersion, and the Gaofen-6 image map.
Agronomy 13 02871 g001
Agronomy 13 02871 g002
Figure 2. The topographic map of the study area.
Figure 2. The topographic map of the study area.
Agronomy 13 02871 g002
Agronomy 13 02871 g003
Figure 3. Overall research workflow.
Figure 3. Overall research workflow.
Agronomy 13 02871 g003
Agronomy 13 02871 g004
Figure 4. Distribution of cultivated land in the study area.
Figure 4. Distribution of cultivated land in the study area.
Agronomy 13 02871 g004
Agronomy 13 02871 g005
Figure 5. Distribution map of cultivated land quality levels in the research area.
Figure 5. Distribution map of cultivated land quality levels in the research area.
Agronomy 13 02871 g005
Agronomy 13 02871 g006
Figure 6. MSVT-CLQ remote sensing inversion distribution map.
Figure 6. MSVT-CLQ remote sensing inversion distribution map.
Agronomy 13 02871 g006
Table 1. Specific information of Gaofen-6 remote sensing images.
Table 1. Specific information of Gaofen-6 remote sensing images.
Serial NumberImage Acquisition TimeSensor TypeSpatial Resolution (Meter)
119 February 2021PMS2
224 March 2021PMS2
324 March 2021PMS2
422 June 2021PMS2
525 November 2021PMS2
625 November 2021PMS2
725 November 2021WFV16
825 November 2021WFV16
Note: PMS is the abbreviation for a 2 m full-color/8 m multispectral high-resolution camera. WFV is the abbreviation for a 16 m multispectral medium-resolution wide-frame camera.
Table 2. Gaofen-6 image band information.
Table 2. Gaofen-6 image band information.
Sensor TypeBandWavelength (μm)Spatial Resolution (Meter)Gain
PMSPan0.45–0.9020.0577
Blue0.45–0.5280.0821
Green0.52–0.6080.0671
Red0.63–0.6980.0518
Near Infrared0.76–0.9080.031
WFVB10.45–0.52160.0633
B20.52–0.69160.0532
B30.63–0.69160.0508
B40.77–0.89160.0325
B50.69–0.73160.0523
B60.73–0.77160.0463
B70.40–0.45160.067
B80.59–0.63160.0591
Table 3. Vegetation index and calculation formula.
Table 3. Vegetation index and calculation formula.
Vegetation IndexAbbreviationCalculation FormulaReference
Normalized vegetation indexNDVINDVI = (B4 − B3)/(B4 + B3)[23]
Difference vegetation indexDVIDVI = B4 − B3[24]
Ratio vegetation indexRVIRVI = B4/B3[25]
Enhanced Vegetation IndexEVIEVI = 2.5(B4 − B3)/(B4 + 6.0B3 − 7.5B1 + 1)[26]
Soil-corrected vegetation indexSAVISAVI = (B4 − B3)/(B4 + B3 + 0.5) × (1 + 0.5)[27]
Note: B1, B3, and B4 represent reflectance in the blue, red, and near-infrared bands, respectively.
Table 4. Cultivated land quality evaluation indexes.
Table 4. Cultivated land quality evaluation indexes.
Guideline LayerIndicator LayerIndex TypeIndex Weight
Site conditionsParts of the terraintextual0.0988
Farmland forestry reticulationtextual0.0408
Profile traitsEffective soil layer thicknessnumerical0.0413
Texture configurationtextual0.0518
Obstacle factorstextual0.0536
Physical and chemical propertiesPlough layer texturetextual0.0797
Soil bulk densitynumerical0.0558
Soil acidity and alkalinitynumerical0.0491
Soil nutrientsSoil organic matternumerical0.1221
Soil available phosphorusnumerical0.0565
Soil available potassiumnumerical0.0594
Soil health statusBiodiversitytextual0.0345
Cleanlinesstextual0.0335
Farmland
management		Irrigation capacitytextual0.1089
Drainage capacitytextual0.1141
Table 5. The comprehensive index range of cultivated land quality level.
Table 5. The comprehensive index range of cultivated land quality level.
Serial NumberCultivated Land Quality LevelComprehensive Index Range
1Higher≥0.8924
2High0.8431–0.8924
3Medium0.7939–0.8431
4Low0.7446–0.7939
5Lower<0.7446
Table 6. Model input variable grouping and model accuracy.
Table 6. Model input variable grouping and model accuracy.
GroupsModel Input VariablesPerformance Indicator
R2RMSE
MSVT-CLQsoil organic matter, soil bulk density, effective soil thickness, pH, B1, B3, B4, B6, NDVI, DVI, RVI, EVI, SAVI, slope, TWI, SPI0.931.42
SVT-CLQB4, B6, EVI, TWI0.913.13
Table 7. Analysis of the proportion of cultivated land quality inversion area.
Table 7. Analysis of the proportion of cultivated land quality inversion area.
Quality Level of Cultivated LandProportion of Area at Different Levels
CLQMSVT-CLQSVT-CLQ
17.00%17.00%12.00%
226.00%32.00%25.00%
327.00%25.00%30.00%
425.00%14.00%23.00%
515.00%12.00%10.00%
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Tang, M.; Wang, Q.; Mei, S.; Ying, C.; Gao, Z.; Ma, Y.; Hu, H. Research on the Inversion Model of Cultivated Land Quality Using High-Resolution Remote Sensing Data. Agronomy 2023, 13, 2871. https://doi.org/10.3390/agronomy13122871

AMA Style

Tang M, Wang Q, Mei S, Ying C, Gao Z, Ma Y, Hu H. Research on the Inversion Model of Cultivated Land Quality Using High-Resolution Remote Sensing Data. Agronomy. 2023; 13(12):2871. https://doi.org/10.3390/agronomy13122871

Chicago/Turabian Style

Tang, Mengmeng, Qiang Wang, Shuai Mei, Chunyang Ying, Zhengbao Gao, Youhua Ma, and Hongxiang Hu. 2023. "Research on the Inversion Model of Cultivated Land Quality Using High-Resolution Remote Sensing Data" Agronomy 13, no. 12: 2871. https://doi.org/10.3390/agronomy13122871

APA Style

Tang, M., Wang, Q., Mei, S., Ying, C., Gao, Z., Ma, Y., & Hu, H. (2023). Research on the Inversion Model of Cultivated Land Quality Using High-Resolution Remote Sensing Data. Agronomy, 13(12), 2871. https://doi.org/10.3390/agronomy13122871

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop