Next Article in Journal
Evaluation of Tree-Growth Rate in the Laurentides Wildlife Reserve Using GEDI and Airborne-LiDAR Data
Previous Article in Journal
Unimodular Waveform Design for the DFRC System with Constrained Communication QoS
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Communication

Soil Texture Mapping in Songnen Plain of China Using Sentinel-2 Imagery

1
College of Geographic Science and Tourism, Jilin Normal University, Siping 136000, China
2
State Key Laboratory of Black Soils Conservation and Utilization, Northeast Institute of Geography and Agroecology, Chinese Academy of Sciences, Changchun 130102, China
*
Author to whom correspondence should be addressed.
Remote Sens. 2023, 15(22), 5351; https://doi.org/10.3390/rs15225351
Submission received: 14 September 2023 / Revised: 26 October 2023 / Accepted: 10 November 2023 / Published: 14 November 2023

Abstract

:
Soil texture is a key physical property that affects the soil’s ability to retain moisture and nutrients. As a result, it is of extreme importance to conduct remote sensing monitoring of soil texture. Songnen Plain is located in the black soil belt of Northeast China. The development of satellite imagery in remote sensing technology enables the rapid monitoring of large areas. This study aimed to map the surface soil texture of cultivated land in Songnen Plain using Sentinel-2 images and Random Forest (RF) algorithm. We conducted this study by collecting 354 topsoil (0–20 cm) samples in Songnen Plain and evaluating the effectiveness of the bands and spectral indices of Sentinel-2 images and RF algorithm in predicting soil texture (sand, silt, and clay fractions). The results demonstrated that the 16 covariates were moderately and highly correlated with soil texture. And, Band11 of Sentinel-2 images could be used as the corresponding band of soil texture. For sand fraction, the Sentinel-2 images and RF algorithm’s Coefficient of Determination (R2) and Root Mean Square Error (RMSE) were 0.77 and 10.48%, respectively, and for silt fraction, they were 0.75 and 9.38%. Sand fraction decreased from southwest to northeast in Songnen Plain, while silt and clay fractions increased. We found that the Songnen Plain was affected by water erosion and wind erosion, in the northeast and southwest, respectively, providing reference for the implementation of Conservation Tillage policies. The outcome of the study can provide reference for future soil texture mapping with a high resolution.

Graphical Abstract

1. Introduction

The properties and processes of the soil’s physical, chemical, and biological components are significantly influenced by the texture of the soil [1,2]. According to the USDA (United States Department of Agriculture) soil texture classification, soil texture is expressed by the soil particle-size fractions [3], which is divided into sand (2–0.05 mm), silt (0.05–0.002 mm), and clay (<0.002 mm), with a sum of 1 (or 100%) [4]. To comprehend soil information and make agricultural decisions, it is essential to map soil particle-size fractions. Digital Soil Mapping (DSM) has been used environmental covariates to predict the spatial distribution of soil types or properties [5,6]. The environmental covariates of DSM are mainly derived from Digital Elevation Models (DEMs), satellite images, and previously generated maps [7]. As remote sensing technology has grown and developed, many satellite images have been used as covariates in the prediction of soil attributes such as soil organic matter [8], soil total nitrogen [9], soil texture [10], and soil moisture [11].
Soil texture influences the spectral intensity and absorption band depth of visible, Near-InfraRed and Short-Wave InfraRed (VNIR-SWIR) spectral domain [12,13,14]. The European Space Agency (ESA) successfully launched the super spectral Sentinel-2A on 23 June 2015, and Sentinel-2B on 7 March 2017. Sentinel-2 has a spatial resolution ranging from 10 to 60 m and a minimum five-day revisit cycle [10]. Sentinel-2 has 13 bands from the VNIR-SWIR spectral domain, and its images are frequently used to forecast the soil properties [10,14,15]. Gholizadeh et al. [10] used the ten bands of Sentinel-2 images, Band2, Band3, Band4, Band5, Band6, Band7, Band8, Band8a, Band11, and Band12, as well as the 18 spectral indices constructed by Sentinel-2 images to predict clay fraction in four agricultural sites in the Czech Republic. Among the bands of Sentinel-2 images, Band7 showed the highest correlation coefficient with clay fraction at the Jičín site [10]. The findings demonstrated that clay fraction could be predicted from Sentinel-2 images for mapping larger geographic areas [10]. The study by Swain et al. [16] was based on 295 soil samples, with an ensemble modeling approach of the surface soil particle-size fractions in the western catchment of Chilika lagoon using the ten bands of Sentinel-2 images, Band2, Band3, Band4, Band5, Band6, Band7, Band8, Band8a, Band11, and Band12, and the results were comparable to hyperspectral data observations measured in the laboratory. Gomez et al. [14] used 130 soil surface samples from a cultivated region in India to map soil texture classification through the ten bands of Sentinel-2 images, Band2, Band3, Band4, Band5, Band6, Band7, Band8, Band8a, Band11, and Band12, and demonstrated that Sentinel-2 images provided the potential for inputs of environmental process and crop management modeling. Zhang et al. [15] used Sentinel-2A and MODIS images to predict soil organic matter through three machine learning algorithms, based on 281 soil samples in the north of Songnen Plain. The outcomes demonstrated that the three machine learning algorithms using Sentinel-2A images performed better in predicting soil organic matter than MODIS images. In earlier studies, soil texture prediction and classification used the bands of Sentinel-2 images and their constructed vegetation indices, moisture indices, and bright-related indices. The Sentinel-2 images have a high resolution to improve mapping accuracy. Therefore, soil texture based on the Sentinel-2 images helped researchers to obtain more detailed information on the soil physical properties.
The measurement methods of soil particle-size fractions were mainly divided into sedimentation method and laser diffraction method. The sedimentation includes pipette, hydrometer and sieving [17]. Most of the data in the world soil database were obtained based on sedimentation methods [17]. The sieve-pipette method is to determine the sand fraction by sieving, and the silt and clay fractions are measured by pipette [18]. The speed, reliability, and automation of laser diffraction method [17] made the application of laser diffraction method increase gradually. Most authors believed that the laser diffraction method underestimated the clay fraction and overestimated the silt fraction compared to the sieve-pipette method, while the difference in the sand fraction was small [17,19,20]. However, Bittell et al. [17] believed that the laser diffraction method had better accuracy compared to the sieve-pipette method, and suggested using laser diffraction as the standard method for soil particle-size analysis. Most of the previous studies were to map the spatial distribution of soil particle-size fractions of laser diffraction method [21] or the spatial distribution of soil particle-size fractions of sedimentation method [4], and less to compare the differences in the spatial distribution of the two methods.Songnen Plain is located in the black soil area of Northeast China, which is one of the four major black soil areas in the world. Although the spatial distribution maps of soil organic matter with high resolution were available for this region [15], the spatial distribution map of soil particle-size fractions with high resolution had not been adequately addressed. Algorithms for DSM have significant implications for model building and spatial mapping, and an increase in computational power contributes to model development. Random Forest (RF) is an integrated learning algorithm based on decision trees that can reduce overfitting and improve robustness. In this study, Sentinel-2 images and RF algorithm are used for spatial mapping of surface (0–20 cm) soil particle-size fractions of cultivated land in Songnen Plain. Our study objectives are (1) to evaluate the correlation of the bands and spectral indices of Sentinel-2 images and soil particle-size fractions followed by (2) evaluating the RF model to predict soil particle-size fractions in Songnen Plain, (3) to map the spatial distribution of surface soil particle-size fractions of cultivated land in Songnen Plain based on laser diffraction method and transformation sieve-pipette method, and finally (4) to analyze the causes for the change in surface soil particle-size fractions of cultivated land in Songnen Plain.

2. Materials and Methods

2.1. Study Area and Soil Data

Our study area was located in Songnen Plain, China (Figure 1). The study area is connected to the Xiao Hinggan Mountains in the north, the Greater Hinggan Mountains in the west, and the Changbai Mountain in the east, covering the agricultural land of 142,337.70 Km2. This region is a significant base for China’s commodity grain production. It has a temperate continental semi-arid and semi-humid monsoon climate [15] with a mean annual air temperature (MAAT) of 3.3 °C and a mean annual precipitation (MAP) of 504 mm [22]. The elevation of the Songnen Plain increases from southwest to northeast. The mean elevation of the study area is 208.53 m, and it is higher in Heihe City, Suihua City, and Harbin City, while the rest of the study area is relatively flat. According to the World Reference Base (WRB), the main soil types in the study area are Phaeozems, Chernozems, Cambisols, and Arenosols [23]. Soil type data were derived from the second national land survey. This area is mainly planted with corn, soybean, and rice.
Three hundred and fifty-four samples of topsoil (0–20 cm) from the study area’s cultivated land were taken in 2021. In order to ensure that the collected soil samples had a variety of soil types and were situated within the cultivated land range, we combined the data on soil types and cultivated land range to aid in the selection of soil samples. Soil samples were collected by using the five-point sampling method. Each soil sample was collected in the cultivated land at least 200 m off the road, and the distance of each soil sample was 5–10 km [24]. In the interim, we used a GPS receiver to locate and record each soil sample. Soil samples were transported to the laboratory and then air-dried. Soil particle-size fractions were measured using Malvern MS-2000 laser particle-size analyzer.

2.2. Sentinel-2 Satellite Images

In the study area, the bare soil period ran from April to May, when there was less crop stubble and more exposed soil [15]. The images of the bare soil period helped to obtain ground soil information and reduce the interference of other factors. We chose to download the Sentinel-2 images from April to May 2021 from the Google Earth Engine (GEE) platform (https://code.earthengine.google.com/) [25,26]. The soil properties of the surface layer change over time, but the soil particle-size fractions do not change in the short term [14] and the bare soil period is short. Additionally, Sentinel-2 images were affected by the presence of cloud, and those from April to May 2021 did not completely cover the study area. Therefore, the missing Sentinel-2 images in 2021 were supplemented by 2019 and 2020. The Sentinel-2 images downloaded in this study were Level 2A processed, and they were atmosphere corrected. We resampled their resolution to 10 m by GEE. We chose to download the Blue (Band2), Green (Band3), Red (Band4), NIR (Band8), SWIR1 (Band11), and SWIR2 (Band12) of the Sentinel-2 images.

2.3. Covariates Selection

It has been demonstrated that the bands and spectral indices of Sentinel-2 images have been used to predict the soil particle-size fractions [10]. In this study, we calculated 13 spectral indices that were created using the six bands of the downloaded Sentinel-2 images as covariates. Major climate, ecosystem, terrain, and soil physical properties have been linked to spatial variations in vegetation indices [27]. We calculated the following vegetation indices [10]: Normalized Differences Vegetation Index (NDVI) [28], Transformed Vegetation Index (TVI) [29], Enhanced Vegetation Index (EVI) [30], Soil Adjusted Vegetation Index (SAVI) [31], Soil Adjusted Total Vegetation Index (SATVI) [32], Green-Red Vegetation Index (GRVI) [33], the Second Modified Soil Adjusted Vegetation Index (MSAVI2) [34], and Vegetation (V) [35]. Soil moisture was measured by water indices, and soil texture was detected by bright-related indices [10,13]. Then, we chose Moisture Stress Index (MSI) [36], Redness Index (RI) [37], Brightness Index (BI) [34], the Second Brightness Index (BI2) [38], and Color Index (CI) [37]. Thereafter, we performed feature selection of covariates by Pearson’s correlation analysis on covariates and soil particle-size fractions, selecting covariates with better correlation as the final predicted covariates. And, the low correlation covariates would not be used as the final predicted covariates. The rasters of all spectral indices were calculated by MATLAB 2021a. The equations of spectral indices are listed in Table 1.

2.4. Mean Annual Precipitation (MAP) Data

Mean Annual Precipitation (MAP) data at a spatial resolution of 1000 m were obtained from the National Meteorological Information Centre (http://data.cma.cn/), and they were resampled to 10 m resolution using the NEAREST in ArcGIS. The MAP data had a low resolution. Additionally, the MAP data were not used as covariates in the RF models. We only used them as analytical data to analyze the soil particle-size fractions in the study area.

2.5. Random Forest (RF) Model

Random Forest (RF) [39] has been applied in soil particle-size fractions modeling studies, which can deal with complex soil–environmental relationships [4]. RF is a decision tree collection that performs better than individual decision trees. From a randomly chosen subset of the available features, it can choose the best feature to be divided at each node [40]. This random selection can produce the diversity of trees, reduce overfitting, and improve prediction accuracy. The final prediction is the average of the predicted values of all trees [41]. There are two parameters in RF to be determined, the number of decision trees (ntree) and the number of random variables of the split nodes (mtry). In this study, the bands and final predicted spectral indices of Sentinel-2 images were used to model the surface soil particle-size fractions of cultivated land in the study area. We set the ntree as 500 and the mtry as 1/3 of the number of inputs when both the results and the error within the model were relatively stable. The RF model was implemented through the Random Forest package (https://cran.r-project.org/web/packages/randomForest/index.html) in R software [42].

2.6. Model Training and Validation

Based on the 354 samples, 2/3 were randomly used as training data and 1/3 as validation data. The Coefficient of Determination (R2), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Ratio of Performance to Interquartile Distance (RPIQ) were used to assess the performance of the RF model. The stability of the model was tested by the Coefficient of Determination (R2). The larger the R2, the more stable the model. The lower the RMSE and MAE values and higher RPIQ value [43], the higher the accuracy of the model.
R 2 = 1 i = 1 n X i Y i 2 i = 1 n ( Y i Y ¯ ) 2
M A E = 1 n i = 1 n | X i Y i |
R M S E = 1 n i = 1 n ( X i Y i ) 2
R P I Q = I Q R M S E
Note: Xi is the predicted value of the soil particle-size fractions, Yi is the observed value of the soil particle-size fractions, n is the number of soil samples, Y ¯ is the average of the observed values, and IQ is the interquartile range.

2.7. The Transformation of Laser Diffraction Method and Sieve-Pipette Method

This study was based on the laser diffraction method for measuring the soil particle-size fractions in Songnen Plain. Most of the existing soil texture products (such as HWSD v2.0 and Soil Grids 250 m v2.0) were measured based on sedimentation methods. In order to remain consistent with the range of their values and explore the differences between the values of the two methods, we transformed the spatial distribution maps of the soil particle-size fractions based on the laser diffraction method. Yang et al. [19] believed that the laser diffraction and sieve-pipette methods had a good correlation and transformation relationship, and the transformation equations could be used. Yang et al. [19] constructed the transformation equations for laser diffraction and sieve-pipette methods based on 265 soil samples in China. Li et al. [20] constructed the transformation equations for laser diffraction and sieve-pipette methods based on five soil samples in Heilongjiang Province, Jilin Province, and Inner Mongolia Autonomous Region of China. Therefore, we transformed the distribution map of the soil particle-size fractions predicted in laser diffraction method according to the transformation equations of Yang et al. [19] and Li et al. [20] for the laser diffraction and sieve-pipette methods. The mean of these two transformation results was then selected as the distribution map result of soil particle-size fractions predicted by the transformation sieve-pipette method. The transformation equations for sand fraction were (5) [19] and (6) [20], respectively. The transformation equations for silt fraction were (7) [19] and (8) [20], respectively. The transformation equations for clay fraction were (9) [19] and (10) [20], respectively.
y 1 = 1.0298 x 1 0.3695
y 2 = 0.77 x 1 + 19.18
y 3 = 0.9122 x 2 7.3149
y 4 = 0.48 x 2 + 3.89
y 5 = 1.4334 x 3 + 3.4414
y 6 = 2.66 x 3 + 1.10
Note: y 1   [19] and y 2 [20] denote the sand fraction measured by the sieve-pipette method, x 1 is the sand fraction measured by the laser diffraction method, y 3   [19] and y 4 [20] denote the silt fraction measured by the sieve-pipette method, x 2 is the silt fraction measured by the laser diffraction method, y 5   [19] and y 6 [20] denote the clay fraction measured by the sieve-pipette method, and x 3 is the clay fraction measured by the laser diffraction method.

3. Results

3.1. Descriptive Statistics

Table 2 displays descriptive statistics for samples of soil particle-size fractions taken from the study area, including minimum, maximum, mean, median, Standard Deviation (SD), and Coefficient of Variation (CV). The fractions of sand, silt, and clay range from 3.00% to 90.04%, 7.60% to 84.79%, and 2.36% to 22.18%, respectively. The sand and silt fractions in the study area change greatly, but only the clay fraction changes slightly. The mean of silt fraction (60.34%) is the highest, greater than the mean of sand fraction (28.02%) and clay fraction (11.64%). The Coefficient of Variation (CV) of sand fraction (77.65) is higher than that of clay (31.82) fraction and silt (31.15) fraction, indicating that the sand fraction in the study area has high variability.

3.2. Analysis of Soil Reflectance Spectral Characteristics of Sentinel-2

Figure 2 shows the reflectance spectral characteristics of soil with different sand, silt, and clay fractions. Reflectance rises gradually from Band2 to Band11, rises slowly, invisibly, and quickly in Band8 to Band11, but falls in Band12. With the increase in sand fraction, the reflectance of each band of the spectral curve increases. The reflectance of each band of the spectral curve decreases as the silt and clay fractions rise. The reflectance of sand fraction changes the most with an increase in sand, silt, and clay fractions, while the reflectance of clay fraction changes the least, particularly in the visible area. Meanwhile, the standard deviations in Figure 2a–c from Band2 to Band8 are 635, 706, and 551, respectively. The standard deviations in Figure 2a–c from Band8 to Band11 are 1014, 875, and 712, respectively, and the standard deviations in Figure 2a–c from Band11 to Band12 are 1011, 769, and 606, respectively. The standard deviations from Band8 to Band11 and Band11 to Band12 are greater than those from Band2 to Band8, which also shows that the changes from Band8 to Band11 and Band11 to Band12 are faster. Soil sand, silt, and clay fractions all changed more significantly in SWIR1 (Band11), followed by SWIR2 (Band12).

3.3. Correlation of Covariates

Covariates are significantly correlated with soil particle-size fractions at a significance level of p < 0.01 as shown in Table 3. This helped to understand which bands and spectral indices of Sentinel-2 images were the most important factors affecting soil particle-size fractions.
Sand fraction and the bands of Sentinel-2 images are positively correlated, while both silt and clay fractions are negatively correlated with the bands of Sentinel-2 images. Sand fraction and Band11 (r = 0.789) are strongly correlated, MSI (r = −0.242), GRVI (r = 0.187), and CI (r = −0.187) are weakly related, and other covariates are moderately correlated (r > 0.26). Silt fraction and Band11 (r = −0.761) are strongly correlated, MSI (r = 0.240), GRVI (r = −0.174), and CI (r = 0.174) are weakly related, and other covariates are moderately correlated (r > 0.26). Clay fraction and Band11 (r = −0.775) are strongly correlated, MSI (r = 0.204), GRVI (r = −0.214), and CI (r = 0.214) are weakly related, and other covariates are moderately correlated (r > 0.26). Sand, silt, and clay fractions all showed a high correlation with Band11, followed by Band12. In the spectral indices constructed by Sentinel-2 images, SATVI (composed of Short-Wave InfraRed and Near-InfraRed) revealed the highest correlation with sand, silt, and clay fractions. Sand, silt, and clay fractions in the study area were correlated to the amount of visible, Near-InfraRed and Short-Wave InfraRed absorbed and/or reflected, according to the high correlations of Band11, Band12, and SATVI [16].
Feature selection of selected covariates occurred before the RF model construction to ensure the selection of significant and high correlation covariates and the exclusion of low correlation covariates. Therefore, using all covariates and response covariates, we calculated Pearson’s correlation coefficients. Due to their weak correlation with soil particle-size fractions, MSI, GRVI, and CI were not used to forecast these fractions. Finally, we chose Band2, Band3, Band4, Band8, Band11, Band12, NDVI, TVI, EVI, SAVI, SATVI, MSAVI2, BI, BI2, RI, and V to build RF models in this study.

3.4. Random Forest Model Accuracy Assessment

In spatial modeling of soil texture, the sum of sand, silt, and clay fractions at each location is 100% [1]. In this study, the prediction accuracy of sand and silt fractions was higher than that of clay fraction; therefore, we used 100% to subtract sand and silt fractions to obtain the values of clay fraction [44]. The training and validation accuracy of the sand and silt fractions using RF models in the study area is displayed in scatterplots in Figure 3. The predicted sand fraction training values for R2, RMSE, MAE, and RPIQ are 0.95, 5.06%, 3.65%, and 5.42, whereas validation values for R2, RMSE, MAE, and RPIQ are 0.77, 10.48%, 8.31%, and 2.6. The predicted silt fraction training values for R2, RMSE, MAE, and RPIQ are 0.95, 4.62%, 3.43%, and 5.58, whereas validation values for R2, RMSE, MAE, and RPIQ are 0.75, 9.38%, 7.37%, and 2.67.

3.5. Spatial Distribution of Soil Particle-Size Fractions

The images were segmented in accordance with the cultivated range boundaries because our samples were found within the cultivated land. Then, using RF models and the final 16 covariates selected from Sentinel-2 images, the distribution of surface soil particle-size fractions of cultivated land in Songnen Plain was mapped. Figure 4a,c,e show the predicted maps of sand, silt, and clay fractions in Songnen Plain, with ribbons ranging from green to red, where green is lower and red is higher. In Songnen Plain, the fractions of silt and clay tend to be higher in the northeast and lower in the southwest, while the fractions of sand tend to be lower in the northeast and higher in the southwest. The average values of sand, silt, and clay fractions were 32.33%, 57.2%, and 10.56%, respectively. Figure 4b,d,f show the spatial distribution of sand, silt, and clay fractions by the transformation sieve-pipette method, respectively. The sand fraction value of transformation sieve-pipette and laser diffraction methods is relatively consistent, the silt fraction value of laser diffraction method is higher than that of transformation sieve-pipette method, and the clay fraction value of laser diffraction method is less than that of transformation sieve-pipette method. This is consistent with previous studies [17,19,20]. Based on the transformation sieve-pipette method, compared to the laser diffraction method, the average values of sand fraction increased by 6.16%, the mean of silt fraction decreased by 19.09%, and the mean of clay fraction increased by 13.32%.

4. Discussion

Our study revealed that Band11 had the best correlation coefficients and more significant spectral characteristics with soil texture. Demattê et al. [45] believed that quartz reflected in SWIR, and Band5 of Landsat-5 TM image could significantly distinguish between different soil textures. Therefore, Band11 of Sentinel-2 images could act as the response band of soil particle-size fractions. The high correlation of the bands and spectral indices of Sentinel-2 images contributed to improving the accuracy of the model. To date, there have been few studies using only single remote sensing covariates to predict soil texture [10,44]. Our study demonstrated the high accuracy of the model created using just remote sensing covariates. Also, using Sentinel-2 images increased the precision of predicting soil particle-size fractions and Sentinel-2 had high resolution. As a result, Sentinel-2 images were able to create maps of soil particle-size fractions with greater precision and resolution.
Compared with the results from previous studies [46], our model constructed using Sentinel-2 images and RF algorithm had higher R2, but the RMSE had a larger value. The wide range of our study area and the wide value variation of soil particle-size fractions, such as sand fraction (3.00–90.04%) and silt fraction (7.60–84.79%), which might have an impact on the RMSE value of the model, were potential causes of this. The RF algorithm was widely recognized for region-scale SOM prediction and achieved good model performance [15]. In this study, the RF algorithm achieved good accuracy in the prediction of soil particle-size fractions, which provided a reference for predicting soil particle-size fractions in Songnen Plain or other areas.
We mapped the soil particle-size fractions in Songnen Plain using the Sentinel-2 images and RF algorithm. Sandy desertification exists in the west of Songnen Plain, which had an important influence on the soil pattern [47]. This was also confirmed by our study, where the west of Songnen Plain had a high sand fraction. The sand fraction in the study area tended to increase from the northeast to the southwest. The Horqin Sandy Land in the southwest of the Songnen Plain provided the sand source [48], where there was a high (>60) sand fraction in the southwest of the study area. The silt fraction in the study area tended to decrease from the northeast to the southwest, with the opposite trend of sand fraction. Liu et al. [4] stated that the sand fraction in the northeast of Songnen Plain was low, while the silt fraction in the east of Songnen Plain was high, and the sand fraction showed the opposite trend of silt fraction, which was consistent with our study. The annual precipitation in the northeast of the study area was higher than that in the southwest, resulting in strong water erosion, transportation, and sorting of soil particles [4]. Therefore, a higher silt fraction was formed in the northeast of the study area. The clay fraction in the study area was very low, and it was in a downward trend from northeast to southwest. The clay fraction in the southwest of the study area was lower than that in the northeast, which might be due to the low precipitation and high wind speed in the southwest of the study area [4].
The eastern water erosion and the western wind erosion in the northeast black soil region [49] shaped the distribution of soil texture in Songnen Plain, and these two phenomena were clear in Songnen Plain. In Songnen Plain, water erosion and wind erosion were discovered in the northeast and southwest of cultivated land, respectively. Soil wind erosion reduces the soil organic matter and nutrient content, threatens agricultural production, and causes great harm to the ecological environment [50]. Soil water erosion causes the loss of topsoil with surface runoff, resulting in the barren of soil nutrients [51]. The land of Songnen Plain is fertile, suitable for the growth of crops, and has a long history of reclamation. In areas that have become farmland, soil texture is not only affected by natural factors such as terrain, temperature, and precipitation, but also by artificial measures such as tillage and fertilization. With the increase in reclamation scope, forest and grassland became cultivated land, losing the role of vegetation water conservation [49] and resisting wind and sand, resulting in the serious water erosion and wind erosion. With the growth of reclamation time, the unreasonable use of land, and the abuse of chemical fertilizer, the soil fertility of cultivated land decreases, and the water and soil loss are serious. Soil erosion leading to soil degradation and decreased food production poses a large threat to food security; therefore, the concept of Conservation Tillage has been proposed. Conservation Tillage reduces the number of soil tillage and increases the surface straw stubble, which can increase soil organic matter, improve soil structure, and reduce water erosion and wind erosion [52]. We can identify the study areas where it is necessary to implement Conservation Tillage by using our study to reflect the trends and extent of water and wind erosion in Songnen Plain. Conservation Tillage significantly increased the silt and clay fractions compared with traditional tillage [53]. The change in soil particle-size fractions, in the areas where Conservation Tillage has been implemented, can be understood through the past soil particle-size fractions and our study. It also provides a reference for soil structure in the areas where Conservation Tillage has been implemented for a long time in the study area.
The images from 2019 and 2020 were used to supplement the Sentinel-2 images from 2021. Despite research suggesting that soil texture would not change in the near future [14], future studies should look for other satellite images with shorter revisit dates to make sure that all the images in the study area are acquired in the same year. Our study illustrated that the use of RF algorithm to map the soil particle-size fractions in Songnen Plain obtained good accuracy. It is anticipated that new machine learning algorithms will be developed in subsequent studies to increase prediction accuracy. For example, convolutional neural networks and other ML algorithms that can predict multiple outputs would be useful for getting various soil particle-size fractions. This study mapped only the recent soil texture, but soil texture might change in the long term. Future agricultural policy and fertilization conditions depend on our ability to comprehend the change over time of soil texture and draw a change trend map of soil texture. More studies are needed to compare and analyze the accuracy of laser diffraction and sieve-pipette methods in measuring soil particle-size fractions, as well as to determine the measurement method of soil particle size fractions that most conforms with the actual situation. Bittelli et al. [17] believed that the purpose of particle size analysis was not for soil texture classification, that this conversion was not required, and that the original data from the laser diffraction method could be used. Based on the convenience and speed of measuring soil particle-size fractions by the laser diffraction method, future studies are needed to develop standards for soil texture classification based on the laser diffraction method.

5. Conclusions

The soil texture map of the top layer of cultivated land in Songnen Plain was created for this study using the final predicted 16 covariates from Sentinel-2 images and the RF algorithm. Band11 of Sentinel-2 images could serve as the response band of soil texture. The spatial distribution map of the 10 m soil particle-size fractions provided more details in Songnen Plain. The west of Songnen Plain had a high sand fraction and appropriate measures are needed to resist sand in this region. This study served as a crucial resource for comprehending the physical properties of soil and developing agricultural policies.

Author Contributions

Conceptualization, M.Z. and K.S.; methodology, M.Z. and X.W.; software, M.Z. and X.W.; validation, K.S., X.W., S.L., B.Z. and J.H.; formal analysis, M.Z.; investigation, M.Z.; resources, K.S.; data curation, M.Z.; writing—original draft preparation, M.Z.; writing—review and editing, M.Z. and X.W.; visualization, X.W., S.L. and B.Z.; supervision, S.L., J.H. and B.Z.; project administration, K.S.; funding acquisition, K.S. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the National Key Research and Development Program of China (2021YFD1500101).

Data Availability Statement

No new data were created or analyzed in this study. Data sharing is not applicable to this article.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Amirian-Chakan, A.; Minasny, B.; Taghizadeh-Mehrjardi, R.; Akbarifazli, R.; Darvishpasand, Z.; Khordehbin, S. Some practical aspects of predicting texture data in digital soil mapping. Soil Tillage Res. 2019, 194, 104289. [Google Scholar] [CrossRef]
  2. Dunkl, I.; Ließ, M. On the benefits of clustering approaches in digital soil mapping: An application example concerning soil texture regionalization. SOIL 2022, 8, 541–558. [Google Scholar] [CrossRef]
  3. He, N.P.; Wu, L.; Wang, Y.S.; Han, X.G. Changes in carbon and nitrogen in soil particle-size fractions along a grassland restoration chronosequence in northern China. Geoderma 2009, 150, 302–308. [Google Scholar] [CrossRef]
  4. Liu, F.; Zhang, G.L.; Song, X.D.; Li, D.C.; Zhao, Y.G.; Yang, J.L.; Wu, H.Y.; Yang, F. High-resolution and three-dimensional mapping of soil texture of China. Geoderma 2020, 361, 114061. [Google Scholar] [CrossRef]
  5. Zeraatpisheh, M.; Ayoubi, S.; Jafari, A.; Finke, P. Comparing the efficiency of digital and conventional soil mapping to predict soil types in a semi-arid region in Iran. Geomorphology 2017, 285, 186–204. [Google Scholar] [CrossRef]
  6. Zhang, M.; Shi, W.J.; Xu, Z.W. Systematic comparison of five machine-learning models in classification and interpolation of soil particle size fractions using different transformed data. Hydrol. Earth Syst. Sci. 2020, 24, 2505–2526. [Google Scholar] [CrossRef]
  7. Shahriari, M.; Delbari, M.; Afrasiab, P.; Pahlavan-Rad, M.R. Predicting regional spatial distribution of soil texture in floodplains using remote sensing data: A case of southeastern Iran. CATENA 2019, 182, 104149. [Google Scholar] [CrossRef]
  8. Zhang, M.; Liu, H.; Zhang, M.; Yang, H.; Jin, Y.; Han, Y.; Tang, H.; Zhang, X.; Zhang, X. Mapping Soil Organic Matter and Analyzing the Prediction Accuracy of Typical Cropland Soil Types on the Northern Songnen Plain. Remote Sens. 2021, 13, 5162. [Google Scholar] [CrossRef]
  9. Zhang, Y.; Sui, B.; Shen, H.O.; Ouyang, L. Mapping stocks of soil total nitrogen using remote sensing data: A comparison of random forest models with different predictors. Comput. Electron. Agric. 2019, 160, 23–30. [Google Scholar] [CrossRef]
  10. Gholizadeh, A.; Žižala, D.; Saberioon, M.; Borůvka, L. Soil organic carbon and texture retrieving and mapping using proximal, airborne and Sentinel-2 spectral imaging. Remote Sens. Environ. 2018, 218, 89–103. [Google Scholar] [CrossRef]
  11. Fathololoumi, S.; Vaezi, A.R.; Alavipanah, S.K.; Ghorbani, A.; Saurette, D.; Biswas, A. Effect of multi-temporal satellite images on soil moisture prediction using a digital soil mapping approach. Geoderma 2021, 385, 114901. [Google Scholar] [CrossRef]
  12. Baumgardner, M.F.; Silva, L.F.; Biehl, L.L.; Stoner, E.R. Reflectance Properties of Soils. Adv. Agron. 1986, 38, 1–44. [Google Scholar]
  13. Ben-Dor, E.; Banin, A. Near infrared analysis (NIRA) as a simultaneously method to evaluate spectral featureless constituents in soils. Soil Sci. 1995, 159, 259–270. [Google Scholar] [CrossRef]
  14. Gomez, C.; Dharumarajan, S.; Féret, J.-B.; Lagacherie, P.; Ruiz, L.; Sekhar, M. Use of Sentinel-2 Time-Series Images for Classification and Uncertainty Analysis of Inherent Biophysical Property: Case of Soil Texture Mapping. Remote Sens. 2019, 11, 565. [Google Scholar] [CrossRef]
  15. Zhang, M.; Zhang, M.; Yang, H.; Jin, Y.; Zhang, X.; Liu, H. Mapping Regional Soil Organic Matter Based on Sentinel-2A and MODIS Imagery Using Machine Learning Algorithms and Google Earth Engine. Remote Sens. 2021, 13, 2934. [Google Scholar] [CrossRef]
  16. Swain, S.R.; Chakraborty, P.; Panigrahi, N.; Vasava, H.B.; Reddy, N.N.; Roy, S.; Majeed, I.; Das, B.S. Estimation of soil texture using Sentinel-2 multispectral imaging data: An ensemble modeling approach. Soil Tillage Res. 2021, 213, 105134. [Google Scholar] [CrossRef]
  17. Bittelli, M.; Pellegrini, S.; Olmi, R.; Andrenelli, M.C.; Simonetti, G.; Borrelli, E.; Morari, F. Experimental evidence of laser diffraction accuracy for particle size analysis. Geoderma 2022, 409, 115627. [Google Scholar] [CrossRef]
  18. Peng, Y.H.; Keating, K.; Myers, D.B. NMR relaxation times for soil texture estimation in the laboratory: A comparison to the laser diffraction and sieve–pipette methods. Eur. J. Soil Sci. 2021, 72, 918–933. [Google Scholar] [CrossRef]
  19. Yang, J.L.; Zhang, G.L.; Li, D.C.; Pan, J.H. Relationships of Soil Particle Size Distribution between Sieve-Pipette and Laser Diffraction Methods. Acta Pedol. Sin. 2009, 46, 772–780. [Google Scholar]
  20. Li, H.R.; Liu, B.; Wang, R.X.; Liu, W.; Fang, Y.; Yang, D.L.; Zou, X.Y. Particle-size Distribution Affected by Testing Method. J. Desert Res. 2018, 38, 619–627. [Google Scholar]
  21. Li, J.; Wan, H.Y.; Shang, S.H. Comparison of interpolation methods for mapping layered soil particle-size fractions and texture in an arid oasis. Catena 2020, 190, 104514. [Google Scholar] [CrossRef]
  22. Song, X.D.; Yang, F.; Ju, B.; Li, D.C.; Zhao, Y.G.; Yang, J.L.; Zhang, G.L. The influence of the conversion of grassland to cropland on changes in soil organic carbon and total nitrogen stocks in the Songnen Plain of Northeast China. Catena 2018, 171, 588–601. [Google Scholar] [CrossRef]
  23. Zhang, W.L.; Xu, A.G.; Zhang, R.L.; Ji, H.J. Review of Soil Classification and Revision of China Soil Classification System. Sci. Agric. Sin. 2014, 47, 3214–3230. [Google Scholar]
  24. Wang, X.; Li, S.; Wang, L.; Zheng, M.; Wang, Z.; Song, K. Effects of cropland reclamation on soil organic carbon in China’s black soil region over the past 35 years. Glob. Chang. Biol. 2023, 29, 5460–5477. [Google Scholar] [CrossRef]
  25. Main-Knorn, M.; Pflug, B.; Louis, J.; Debaecker, V.; Müller-Wilm, U.; Gascon, F. Sen2Cor for Sentinel-2. Proc. SPIE 2017, 10427, 1042704. [Google Scholar]
  26. Li, S.J.; Song, K.S.; Wang, S.; Liu, G.; Wen, Z.D.; Shang, Y.X.; Lyu, L.L.; Chen, F.F.; Xu, S.Q.; Tao, H.; et al. Quantification of chlorophyll-a in typical lakes across China using Sentinel-2 MSI imagery with machine learning algorithm. Sci. Total Environ. 2021, 778, 146271. [Google Scholar] [CrossRef]
  27. Singh, D.; Herlin, I.; Berroir, J.P.; Silva, E.F.; Meirelles, M.S. An approach to correlate NDVI with soil colour for erosion process using NOAA/AVHRR data. Adv. Space Res. 2004, 33, 328–332. [Google Scholar] [CrossRef]
  28. Rouse, J.W.; Haas, R.H.; Schell, J.A.; Deering, D.W. Monitoring vegetation systems in the Great Plains with ERTS. In Proceedings of the 3rd ERTS Symposium, Washington, DC, USA, 1 January 1974. [Google Scholar]
  29. Nellis, M.D.; Briggs, J.M. Transformed vegetation index for measuring spatial variation in drought impactedbiomass on Konza Prairie, Kansas. Trans. Kans. Acad. Sci. 1992, 95, 93–99. [Google Scholar] [CrossRef]
  30. Huete, A.; Didan, K.; Miura, T.; Rodriguez, E.P.; Gao, X.; Ferreira, L.G. Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sens. Environ. 2002, 83, 195–213. [Google Scholar] [CrossRef]
  31. Huete, A.R. A soil-adjusted vegetation index (SAVI). Remote Sens. Environ. 1988, 25, 295–309. [Google Scholar] [CrossRef]
  32. Marsett, R.C.; Qi, J.G.; Heilman, P.; Biedenbender, S.H.; Watson, M.C.; Amer, S.; Weltz, M.; Goodrich, D.; Marsett, R. Remote Sensing for Grassland Management in the Arid Southwest. Rangel. Ecol. Manag. 1988, 59, 530–540. [Google Scholar] [CrossRef]
  33. Tucker, C.J. Red and photographic infrared linear combinations for monitoring vegetation. Remote Sens. Environ. 1979, 8, 127–150. [Google Scholar] [CrossRef]
  34. Qi, J.; Kerr, Y.; Chehbouni, A. External factor consideration in vegetation index development. In Proceedings of the International Symposium on Physical Measurements and Signatures in Remote Sensing, Val d’Isere, France, 1 January 1994. [Google Scholar]
  35. Jordan, C.F. Derivation of leaf area index from quality of light on the forest floor. Ecology 1969, 50, 663–666. [Google Scholar] [CrossRef]
  36. Rock, B.N.; Williams, D.L.; Vogelmann, J.E. Field and airborne spectral characterization of suspected damage in red spruce (picea rubens) from Vermont. In Proceedings of the Symposium on Machine Processing of Remotely Sensed Data, West Lafayette, IN, USA, 1 January 1985. [Google Scholar]
  37. Pouget, M.; Madeira, J.; Le, F.E.; Kamel, S. Caractéristiques spectrales des surfaces sableuses de la région côtière Nord-Ouest de l’Egypte: Application aux données satellitaires SPOT. In Proceedings of the Caractérisation et Suivi des Milieux Terrestres en Régions Arides et Tropicales, Paris, France, 4–6 December 1990. [Google Scholar]
  38. Escadafal, R. Remote sensing of arid soil surface color with Landsat thematic mapper. Adv. Space Res. 1989, 9, 159–163. [Google Scholar] [CrossRef]
  39. Breiman, L. Random Forests. Mach. Learn. 2001, 45, 5–32. [Google Scholar] [CrossRef]
  40. Denisko, D.; Hoffman, M.M. Classification and interaction in random forests. Proc. Natl. Acad. Sci. USA 2018, 115, 1690–1692. [Google Scholar] [CrossRef]
  41. Ji, W.J.; Adamchuk, V.I.; Chen, S.C.; Mat Su, A.S.; Ismail, A.; Gan, Q.J.; Shi, Z.; Biswas, A. Simultaneous measurement of multiple soil properties through proximal sensor data fusion: A case study. Geoderma 2019, 341, 111–128. [Google Scholar] [CrossRef]
  42. Wang, X.; Wang, L.P.; Li, S.J.; Wang, Z.M.; Zheng, M.; Song, K.S. Remote estimates of soil organic carbon using multi-temporal synthetic images and the probability hybrid model. Geoderma 2022, 425, 116066. [Google Scholar] [CrossRef]
  43. Meng, X.T.; Bao, Y.L.; Wang, Y.; Zhang, X.L.; Liu, H.J. An advanced soil organic carbon content prediction model via fused temporal-spatial-spectral (TSS) information based on machine learning and deep learning algorithms. Remote Sens. Environ. 2022, 280, 113166. [Google Scholar] [CrossRef]
  44. Adhikari, K.; Kheir, R.B.; Greve, M.B.; Bøcher, P.K.; Malone, B.P.; Minasny, B.; McBratney, A.B.; Greve, M.H. High-Resolution 3-D Mapping of Soil Texture in Denmark. Soil Sci. Soc. Am. J. 2013, 77, 860–876. [Google Scholar] [CrossRef]
  45. Demattê, J.A.M.; Ramirez-Lopez, L.; Rizzo, R.; Nanni, M.R.; Fiorio, P.R.; Fongaro, C.T.; Medeiros Neto, L.G.; Safanelli, J.L.; Da, S.; Barros, P.P. Remote Sensing from Ground to Space Platforms Associated with Terrain Attributes as a Hybrid Strategy on the Development of a Pedological Map. Remote Sens. 2016, 8, 826. [Google Scholar] [CrossRef]
  46. Chagas, C.D.S.; Junior, W.D.C.; Bhering, S.B.; Filho, B.C. Spatial prediction of soil surface texture in a semiarid region using random forest and multiple linear regressions. CATENA 2016, 139, 232–240. [Google Scholar] [CrossRef]
  47. Gao, C.Y.; Wei, C.F.; Zhang, L.N.; Han, D.X.; Liu, H.X.; Yu, X.F.; Wang, G.P. Historical (1880s–2000s) impact of wind erosion on wetland patches in semi-arid regions: A case study in the western Songnen Plain (China). Aeolian Res. 2019, 38, 13–23. [Google Scholar] [CrossRef]
  48. Zhao, H.Q.; Zhang, Z.H.; Wang, C.Q. Actuality, dynamic change and the prevention countermeasure of desertification in the Songnan Plain. J. Arid Land Resour. Environ. 2009, 23, 107–113. [Google Scholar]
  49. Li, F.P.; Li, J.Y.; Xu, Z.X. The Status Quo of Black Soil Degradation and Water and Soil Loss in Northeast China. Res. Soil Water Conserv. 2006, 3, 50–54. [Google Scholar]
  50. Li, S.L.; Li, H.P.; Lin, Y.; Xiao, B.; Wang, G.P. Effects of Tillage Methods on Wind Erosion in Farmland of Northeastern China. J. Soil Water Conserv. 2019, 33, 110–118+220. [Google Scholar]
  51. Lyu, Y.H.; Liu, G.H.; Feng, X.M. Environmental Impacts of Soil Water Erosion: A Review of Influence Factors, Hot Research Topics and Evaluation Indices. J. Ecol. Rural Environ. 2011, 27, 93–99. [Google Scholar]
  52. Zhang, H.L.; Gao, W.S.; Chen, F.; Zhu, W.S. Prospects and present situation of conservation tillage. J. China Agric. Univ. 2005, 1, 16–20. [Google Scholar]
  53. He, W.Y. Diversity of Farmland Black SoilStoichiometric Characteristics and Microbial Effects of Conservation Tillage on Ecological. Master’s Thesis, Northeast Agricultural University, Harbin, China, 2023. [Google Scholar]
Figure 1. The location of the study area and sampling locations.
Figure 1. The location of the study area and sampling locations.
Remotesensing 15 05351 g001
Figure 2. Soil reflectance spectral characteristics with different sand fractions (a), silt fractions (b), and clay fractions (c), the two red arrows represent the rising and decreasing trends, respectively.
Figure 2. Soil reflectance spectral characteristics with different sand fractions (a), silt fractions (b), and clay fractions (c), the two red arrows represent the rising and decreasing trends, respectively.
Remotesensing 15 05351 g002
Figure 3. (a,b) represent the results of the sand fraction prediction models using training and validation samples, respectively; (c,d) represent the results of the silt fraction prediction models using training and validation samples, respectively.
Figure 3. (a,b) represent the results of the sand fraction prediction models using training and validation samples, respectively; (c,d) represent the results of the silt fraction prediction models using training and validation samples, respectively.
Remotesensing 15 05351 g003
Figure 4. Spatial distribution maps for (a) sand, (c) silt, and (e) clay fractions by laser diffraction method, spatial distribution maps for (b) sand, (d) silt, and (f) clay fractions by the transformation sieve-pipette method.
Figure 4. Spatial distribution maps for (a) sand, (c) silt, and (e) clay fractions by laser diffraction method, spatial distribution maps for (b) sand, (d) silt, and (f) clay fractions by the transformation sieve-pipette method.
Remotesensing 15 05351 g004
Table 1. The equations of spectral indices.
Table 1. The equations of spectral indices.
IndexDefinitionReference
NDVI ρ N I R ρ R e d ρ N I R + ρ R e d [28]
TVI ρ N I R ρ R e d ρ N I R + ρ R e d + 0.5 1 / 2 × 100 [29]
EVI 2.5 × ρ N I R ρ R e d ρ N I R + 6 × ρ R e d 7.5 × ρ B l u e + 1 [30]
SAVI ( ρ N I R ρ R e d ) × 1.5 ρ N I R ρ R e d + 0.5 [31]
SATVI ρ S W I R 1 ρ R e d ρ S W I R 1 + ρ R e d + 1 × 2 ρ S W I R 2 2 [32]
GRVI ρ G r e e n ρ R e d ρ G r e e n + ρ R e d [33]
MSAVI2 2 × ρ N I R + 1 ( 2 × ρ N I R + 1 ) 2 8 × ( ρ N I R ρ R e d ) 2 [34]
V ρ N I R ρ R e d [35]
MSI ρ S W I R 1 ρ N I R [36]
RI ρ R e d × ρ R e d ρ G r e e n × ρ G r e e n × ρ G r e e n [37]
BI ρ R e d × ρ R e d + ( ρ G r e e n × ρ G r e e n ) 2 [38]
BI2 ρ R e d × ρ R e d + ρ G r e e n × ρ G r e e n + ( ρ N I R × ρ N I R ) 3 [38]
CI ρ R e d ρ G r e e n ρ R e d + ρ G r e e n [37]
Table 2. Descriptive statistics of the soil particle-size fractions.
Table 2. Descriptive statistics of the soil particle-size fractions.
TextureMin (%)Max (%)Mean (%)Median (%)SD (%)CV
Sand3.0090.0428.0218.4721.7677.65
Silt7.6084.7960.3468.5018.8031.15
Clay2.3622.1811.6412.243.7031.82
Table 3. Pearson’s correlation coefficients between soil particle-size fractions and covariates.
Table 3. Pearson’s correlation coefficients between soil particle-size fractions and covariates.
CovariatesSoil Particle-Size Fractions (%)
SandSiltClay
Band20.665 **−0.639 **−0.664 **
Band30.711 **−0.686 **−0.693 **
Band40.734 **−0.709 **−0.714 **
Band80.735 **−0.711 **−0.710 **
Band110.789 **−0.761 **−0.775 **
Band120.778 **−0.749 **−0.770 **
NDVI−0.403 **0.384 **0.417 **
TVI−0.411 **0.392 **0.425 **
EVI−0.325 **0.324 **0.265 **
SAVI0.489 **−0.475 **−0.460 **
SATVI−0.778 **0.749 **0.770 **
GRVI0.187 **−0.174 **−0.214 **
MSAVI2−0.420 **0.401 **0.434 **
V−0.359 **0.342 **0.372 **
MSI−0.242 **0.240 **0.204 **
RI−0.370 **0.354 **0.373 **
BI0.728 **−0.703 **−0.709 **
BI20.737 **−0.712 **−0.714 **
CI−0.187 **0.174 **0.214 **
Note: ** Significant at 0.01 probability.
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

Zheng, M.; Wang, X.; Li, S.; Zhu, B.; Hou, J.; Song, K. Soil Texture Mapping in Songnen Plain of China Using Sentinel-2 Imagery. Remote Sens. 2023, 15, 5351. https://doi.org/10.3390/rs15225351

AMA Style

Zheng M, Wang X, Li S, Zhu B, Hou J, Song K. Soil Texture Mapping in Songnen Plain of China Using Sentinel-2 Imagery. Remote Sensing. 2023; 15(22):5351. https://doi.org/10.3390/rs15225351

Chicago/Turabian Style

Zheng, Miao, Xiang Wang, Sijia Li, Bingxue Zhu, Junbin Hou, and Kaishan Song. 2023. "Soil Texture Mapping in Songnen Plain of China Using Sentinel-2 Imagery" Remote Sensing 15, no. 22: 5351. https://doi.org/10.3390/rs15225351

APA Style

Zheng, M., Wang, X., Li, S., Zhu, B., Hou, J., & Song, K. (2023). Soil Texture Mapping in Songnen Plain of China Using Sentinel-2 Imagery. Remote Sensing, 15(22), 5351. https://doi.org/10.3390/rs15225351

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