Next Article in Journal
Exploring the 4IR Drivers for Sustainable Residential Building Delivery from Social Work Residential Perspective—A Structural Equation Modelling Approach
Previous Article in Journal
Does Green Credit Policy Promote or Inhibit Firms’ Green Innovation in China? Moderating Effect of Environmental Information Disclosure
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Sustainability
Volume 15
Issue 1
10.3390/su15010469
sustainability-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 thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Remote Sensing Prediction Model of Cultivated Land Soil Organic Matter Considering the Best Time Window

by
Yiang Wang
1,2,
Chong Luo
2,
Wenqi Zhang
1,*,
Xiangtian Meng
2,
Qiong Liu
3,
Xinle Zhang
4 and
Huanjun Liu
2,3
1
School of Economics and Management, Jilin Agricultural University, Changchun 130118, China
2
Northeast Institute of Geography and Agroecology, Chinese Academy of Sciences, Changchun 130102, China
3
School of Pubilc Adminstration and Law, Northeast Agricultural University, Harbin 150030, China
4
College of Information Technology, Jilin Agricultural University, Changchun 130118, China
*
Author to whom correspondence should be addressed.
Sustainability 2023, 15(1), 469; https://doi.org/10.3390/su15010469
Submission received: 26 November 2022 / Revised: 25 December 2022 / Accepted: 26 December 2022 / Published: 27 December 2022
Browse Figures
Versions Notes

Abstract

:
Soil organic matter (SOM) is very important to the quality evaluation of cultivated land, especially in fertile black soil areas. Many studies use remote sensing images combined with different machine learning algorithms to predict the regional SOM content. However, the information provided by remote sensing images in different time windows is very different. Taking Youyi Farm, a typical black soil area in Northeast China, as the research area, this study obtains all available Sentinel-2 images covering the research area from 2019 to 2021, calculates the spectral index of single-phase and multi-temporal synthesis images, takes the spectral index and band of each image as the input, and employs the random forest regression algorithm to evaluate the performance of SOM prediction using remote sensing images with different time windows. The results show that: (1) the accuracy of SOM prediction using image band and spectral index is generally improved compared to using only the band; (2) when using single-phase images, the R2 range of SOM prediction using image band and spectral index is from 0.16 to 0.59 and the RMSE ranges from 0.82% to 1.23%; When using multi-temporal synthesis images, the R2 range of SOM prediction using image band and spectral index is from 0.18 to 0.56 and the RMSE ranges from 0.85% to 1.19%; (3) the highest accuracy of SOM prediction using synthetic images is lower than that of single-phase images; (4) the best time window of the bare soil period in the study area is May. This study emphasizes the importance of the time window to SOM prediction. In subsequent SOM prediction research, remote sensing images with appropriate time windows should be selected first, and then the model should be optimized.

1. Introduction

Soil organic matter (SOM) is usually produced by vegetation activity in soil and is an important index of soil fertility [1,2]. Under the influence of human farming and crop growth, agroecosystems are an active and renewable soil carbon pool [3,4]. On one hand, SOM can support the growth and development of crops [5,6]. On the other hand, SOM stores a large amount of carbon [7,8]. SOM contributes to the maintenance and turnover of nutrients, the maintenance and utilization of water, and to agricultural production [9]. Monitoring the spatial distribution of SOM is helpful in analyzing the trend of carbon sequestration to make reasonable decisions on future environmental changes; it can provide valuable information for government measures and farmland activities [10,11,12].
The traditional SOM monitoring method is mainly through field sampling combined with laboratory analysis [13]. This method has high accuracy but can only obtain point data. Furthermore, it is difficult to obtain the overall regional spatial distribution data. At present, there are two mainstream methods to obtain SOM spatial distribution data. The first is the spatial interpolation method, and the second is the remote sensing prediction method [14]. Compared with the spatial interpolation method, the remote sensing prediction method can better avoid SOM prediction error caused by soil spatial heterogeneity [15,16]. In recent years, many studies have relied on the relationship between SOM and the soil reflection spectrum and have used bare soil images to map the spatial distribution of SOM [17,18,19]. However, most studies usually improve the SOM prediction accuracy from model improvement and input optimization and rarely consider the impact of the bare soil time window on the accuracy of SOM prediction. Because bare soil pixels are easily disturbed by environmental factors, the impact of a bare soil window on the accuracy of SOM prediction may be higher than that of other factors.
Many studies have attempted to improve the accuracy of digital soil mapping (DSM) through various methods [20,21]. There are two main ways to select the input variables. First, the common methods include extracting a large number of secondary variables from traditional environmental factors (topography, temperature, and precipitation) and obtaining valuable information from these variables [22,23,24]. Second, the spectral bands and derived spectral indexes based on remote sensing images are also increasingly used in DSM. The use of remote sensing data can directly observe bare soil surface, which has been proved to have a good effect in DSM [25,26]. As the bare soil surface is very vulnerable to the impact of the surrounding environment, selecting a suitable remote sensing image is crucial to providing the accuracy of DSM [27,28].
At present, commonly used prediction methods mainly include linear models, machine learning models, and deep learning models [29,30]. The random forest (RF) model in machine learning models is widely used in remote sensing prediction because of its good performance and model interpretability [31,32,33]. Many studies have used RF regression in grain yield estimation, surface temperature prediction, water quality parameter prediction, and soil physical and chemical property prediction and have achieved good results [34,35,36,37,38]. The prediction of soil organic carbon and/or organic matter using the RF algorithm has been proven to be simple and effective [39].
Sentinel-2 is a high-resolution multispectral imaging satellite equipped with a multispectral imager (MSI) and composed of “twin” Sentinel-2A and Sentinel-2B satellites. It has good spatial, temporal, and spectral resolutions and has great potential in soil mapping [40,41]. There have been many studies on soil mapping using Sentinel-2 images, including soil salinity mapping, soil organic matter mapping, and soil texture mapping, which have achieved good results [42,43]. Recently, research has been conducted to mask unsuitable pixels based on NDVI and NBR2 thresholds, compared with the regions studied (tropical Brazil and Belgium, where conservation tillage is very common). There is a long period of bare soil in Northeast China. When the research area has a precise range of cultivated land, simple median synthesis can also be used to obtain suitable pixels for SOM mapping [25,44].
Existing research on SOM prediction generally uses remote sensing images and different environmental covariates as inputs to explore their impact on the accuracy of SOM prediction. However, even the remote sensing images obtained in the same study area will produce different prediction accuracies. This is mainly because the image may be affected by precipitation, straw mulching, surface morphology, and other factors [45]. Some studies have proved that the accuracy difference caused by SOM mapping using bare soil images in different periods is much higher than that caused by the model method [17,18,19]. It has also been proved that the accuracy of SOM mapping using multi-temporal remote sensing images is higher than that using single-temporal remote sensing images. In a word, screening suitable remote sensing images is the first step of SOM remote sensing mapping [27].
Therefore, it is necessary to determine a specific time window in which relatively good spectral data can be obtained, and the performance of the spectral model is also very high for predicting SOM. The Sentinel-2 satellite remote sensing image with high spatial resolution was used as the data source, and the band and the constructed spectral index were used as the input variables based on the RF algorithm. The specific aims of this study are as follows: (i) to determine the best time window for SOM prediction according to the prediction accuracy; (ii) to compare the accuracy difference of SOM prediction using single-phase image and multi-phase image, and finally; (iii) to map the spatial distribution of SOM content.

2. Materials and Methods

2.1. Study Area

Youyi Farm is located in the hinterland of Sanjiang Plain, Heilongjiang Province, Northeast China [46]. The coordinate range is 131°27′~132°15′ E, 46°28′~46°59′ N (Figure 1). The farm area is 56 km long from east to west and 44 km wide from north to south. The total control area in the area is 1888 km2. It is adjacent to Baoqing County in the east, Jixian County in the west, Shuangyashan city in the south, and Fujin city in the north. Youyi Farm, with its flat terrain and contiguous land, is the largest mechanized state-owned farm in China, with a comprehensive degree of agricultural mechanization of more than 90%. The main soil types include Phaeozems, GleysolS, and Cambisols. The crop planting mode is one season a year, and there is no crop cover from March to May and October to November every year, which is the window period of bare soil. The depth of regular tillage is 20 cm. Due to the early development and high intensity of cultivated land use and serious soil degradation, it is urgent to map the spatial distribution of SOM and other indicators to clarify the key areas and directions of cultivated land protection.

2.2. Data Acquisition and Processing

2.2.1. Soil Data Acquisition

A total of 188 topsoil soil samples (0–20 cm) were collected in the study area from 1 April to 7 April 2021. The SOM concentration will remain stable for several years, so we used the images of nearly three years for SOM prediction [27]. Based on the accessibility of soil sampling, we used the layered sampling method to design the soil sample points, which covered the main soil types, landform types, and cultivated land use types in the study area; using these sampling points ensured the high universality of the SOM prediction model established. During sampling, we first ensured that the sample points were more than 50 m away from roads, forest belts, ditches, and other ground objects to avoid the influence of mixed pixels, and then we collected 5–6 (0–20 cm) soil samples within a 30 m × 30 m square for mixing to avoid the influence of random factors. A portable global positioning system (GPS, g350, Unistrong, Beijing, China) was used to record the longitude and latitude information and acquisition time of the sample points. Finally, the soil samples were ground, air dried, and screened through a 2 mm sieve. The SOC content of the samples was measured by the potassium dichromate heating method, and then the SOC content was multiplied by 1.724 according to the “van Bemmelen factor” to convert it into the SOM content [47].

2.2.2. Sentinel-2 Remote Sensing Image Acquisition and Preprocessing

(1) Acquisition and preprocessing of single-phase Sentinel-2 image
In this study, the L2a products of Sentinel-2 (6 in 2019, 7 in 2020, and 10 in 2021, a total of 23) used in the study area from 2019 to 2021 (cloud cover less than 1% and no snow cover) were screened and downloaded in the Google Earth engine for SOM mapping (Table 1). The L2a product of Sentinel-2 was the surface reflectance data after geometric correction and atmospheric treatment [48]. Among them, band 1 was the aerosol band, band 9 was the water vapor band, and band 10 was the reaction atmosphere band. Therefore, these three bands were removed in the study, and 10 bands remained. Among them, band 2, band 3, band 4, and band 8 had a 10 m spatial resolution, and band 5, band 6, band 7, band 8a, band 11, and band 12 had a 20 m spatial resolution. Then, the remote sensing images were cut and mosaicked according to the boundary of the study area; all bands were spatially resampled to a 10 m resolution.
(2) Acquisition and preprocessing of multi-temporal Sentinel-2 synthesis images
The L2a products of Sentinel-2 used in the study area from 2019 to 2021 (March to November) were screened and downloaded in the Google Earth engine for SOM mapping. The official cloud removal algorithm provided in the GEE platform for cloud mask processing was used on all images. Finally, monthly median synthesis was carried out, and some studies have proved that the median is robust to outliers. The strategy of using median synthesis to select the best pixel has been proved to have good results and high computational efficiency [49]. However, the results obtained by the average value synthesis method may not be the actual physical observation results, and are more vulnerable to the impact of extreme outliers [50].
According to the agricultural planting habits in the study area, the bare soil period was in March, April, May, and November, and crop growth period was in June, July, August, September, and October.

2.3. Construction of the Spectral Index

Because there are few bands in multispectral images, the construction of a spectral index can make full use of the useful information in multispectral images. The construction of the difference index (DI), ratio index (RI), normalized difference index (NDI), and other spectral indices can reduce the effects of moisture, roughness, and atmosphere. Therefore, the spectral index set was established by performing band operation on the band reflectance of Sentinel-2 image data. The formula is as follows:
D I i j = ρ i − ρ j
R I i j = ρ i / ρ j
N D I i j = ρ i − ρ j / ρ i + ρ j
where ρ i and ρ j are the reflectivity of the i and j bands, respectively, where i > j. A total of 135 spectral indices (45 each for difference indices, ratio indices, normalized difference indices) were generated in this study.

2.4. RF Prediction Model

A RF model is a nonlinear machine learning model based on decision trees. In this study, RF modeling was implemented through the RF package in the R language. There were three key parameters involved in the RF model: the number of decision trees (ntree), the number of random variables that split nodes (mtry), and the importance of the independent variable on the SOM (importance). The RF package ranked the importance of the independent variables that affect the SOM prediction during the operation. The larger the value, the greater the influence of the independent variable on the SOM prediction, and the stronger the correlation. In addition, the RF model can effectively avoid the multicollinearity problem between features. In the actual RF modeling operation in this study, ntree was set to 500, and ntry was set to 1/3 of the number of inputs. At this time, not only could a relatively stable out-of-bag error rate be generated, but the value was also small, and the model was relatively stable.

2.5. Model Validation and Establishment

In this study, the RF algorithm model was used. Ten bands and 135 spectral indices of Sentinel-2 image data were used as inputs to carry out SOM inversion modeling. Among the 188 samples, the modeling set and the validation set were divided according to the ratio of 3:1 by Kennard stone algorithm [51], including 141 training samples and 47 validation samples. The coefficient of determination (R2) was used to check the model stability. The root mean square error (RMSE) was used to measure the accuracy of the model.

2.6. Technology Roadmap

The technical route of this study is shown in Figure 2.

3. Results

3.1. Descriptive Statistics of SOM Content

The maximum value of the entire dataset was 9.91%, the minimum value was 0.48%, and the SOM content range was very wide, with a mean value of 3.92%, a standard deviation of 1.40%, and a coefficient of variation of 35.84%, showing strong variation (Table 2). The training and validation sets were similar to the descriptive statistics of the entire dataset.

3.2. Analysis of Spectral Characteristics of Different SOM Contents

The spectral reflectance of soils with different SOM contents is shown in Figure 3. The spectral curves of soils with different organic matter contents were generally similar. Similar to previous studies, the higher the SOM content, the lower the soil spectral reflectance. For the Sentinel-2 image, the soil spectral reflectance gradually increased and showed a trend of first increasing and then decreasing. Soil samples with an SOM content lower than 3% had one or two peaks in the B8 band and B11 band, and soil samples with an SOM content higher than 3% had one or two peaks. The soil sample point had only one peak of B11.

3.3. SOM Prediction Using Single-Phase Images

We used all available imagery in the study area from 2019 to 2021 for SOM prediction (Table 3). In general, the SOM prediction using the image band and the spectral index generally improved the accuracy compared to using only the image band. The accuracy of SOM prediction using remote sensing images from different periods varied greatly. The R2 range of SOM prediction using only bands ranged from 0.05 to 0.53, and the RMSE range was from 0.89% to 1.30%. The R2 range of SOM prediction using bands and the spectral index ranged from 0.16 to 0.59, and the RMSE ranged from 0.82% to 1.23%.
The prediction accuracy of SOM using remote sensing images in the growing period could not reach the highest value but was still relatively stable. The accuracy of SOM prediction using remote sensing images in the bare soil period varied greatly. The period with the highest SOM prediction accuracy was the beginning of May in the bare soil period, which is the best time window for SOM prediction in the study area.

3.4. SOM Prediction Using Synthetic Images

The precision of SOM prediction using multi-year synthetic images in different months is shown in Table 4. The results show that when only the band was used, the accuracy of SOM prediction using May synthesis image was the highest, R2 was the highest 0.55, RMSE was 0.97%. When both the band and the spectral index was used, the accuracy of SOM prediction using May synthesis image was the highest, R2 was the highest 0.56, and RMSE was 0.85%. The highest accuracy of SOM prediction using synthetic images was lower than that of single-phase images. The results using synthetic images were similar to those of single-phase images. May was the window period for SOM prediction in the study area.

3.5. Spatial Distribution of SOM Content

The image and spectral index with the highest prediction accuracy of SOM in the bare soil period and growing period (17 May 2021 and 1 July 2011, respectively) were used as inputs for the RF algorithm. SOM mapping was carried out on the cultivated land of the Youyi Farm in the study area (Figure 4). The results showed that the SOM content was lower in the cultivated land near the urban area and the western mountainous area of the Youyi Farm, and that the SOM content was higher in the cultivated land near the wetland reserve in the east. The spatial difference in SOM prediction using the optimal bare soil image was more obvious than that using the optimal growing season image.

4. Discussion

4.1. Selection of the Best Window

Compared with the model improvement and input optimization that concerns traditional research, this study analyzed the time window of the key factor that is easy to ignore in SOM prediction using bare soil images. General soil studies usually randomly select the images of the bare soil window period for soil mapping because they believe that the images of the bare soil window period have little difference [10,16,52]. Our research compared the accuracy of SOM prediction of different bare soil window images. The results showed that the accuracy of SOM prediction varied greatly in different time periods, which showed that although there are bare soil images without crop cover, the spectral reflectance difference of different bare soil images greatly affected the accuracy of SOM prediction (Figure 5). The analysis of this study showed that the best time window for SOM prediction in the study area was early May, which is the most stable period in a year after cultivated land plowing and before crop emergence [17]. The SOM mapping accuracy during the peak crop growth period (July to September) was relatively stable, not very high or very low, which may be because the crop growth can reflect the SOM content to a certain extent [53]. October was the time for large-scale harvesting in the region, which was the reason for the poor prediction accuracy [33,54].

4.2. Factors Affecting the Prediction Accuracy of SOM Content

The research showed that the highest accuracy of SOM prediction in different years always appeared in the traditional bare soil period (from the end of March to the end of May and after October). The accuracy of SOM prediction using growth period images was relatively stable (R2 was approximately 0.3), which indicated that crop growth would also reflect the spatial distribution of SOM content to a certain extent. The accuracy of SOM prediction using remote sensing images in the bare soil period varied greatly. We analyzed the soil water content and straw residue of all training samples in images in different periods (Figure 6). We used the normalized tillage index (DNTI) to characterize straw coverage and the normalized differential water index (NDMI) to represent the soil water content [55,56,57]. The results showed that soil water content was negatively correlated with the prediction accuracy of SOM (Table 5); that is, the lower the soil water content, the higher the prediction accuracy of SOM. The correlation between straw coverage and SOM was not obvious, which may be related to the local habit of plowing in autumn. After plowing in autumn, almost all images in the bare soil period would be unaffected by straw images.

4.3. Differences between Single-Temporal and Multi-Temporal Synthetic Images

We compared the accuracy difference of SOM prediction using single-phase and multi-phase synthetic images in the study area. The results showed that the highest accuracy of single-phase synthetic images was higher than that of multi-phase synthetic images. Some studies have compared the accuracy of SOM prediction between single-phase and multi-phase synthetic images in tropical Brazil, and the results show that the effect of SOM prediction using multi-phase images is better [25]. The main reason why the results of this study differ from those of that study is that the characteristics of the study area are different. In the tropical region of Brazil, it is almost impossible to obtain an image with a single period and full bare soil pixels, so a single time phase image often has other impurity pixels, which will in turn affect the prediction of SOM [44]. However, in our study area, the farming mode of one season a year plus no promotion of conservation tillage (even promotion of “black wintering”– a farming measure of plowing after crop harvest in October), pure soil pixels were very common (late March to early June) [58,59]. The single-period image is more consistent with the actual situation than the multi-year synthesis image, so SOM prediction of some single-phase images can obtain higher accuracy than multi-phase synthesis images.
Comparing the importance ranking of SOM prediction using single-phase and synthetic images (Figure 7), it can be found that the spectral index of the bands with wavelengths 490–860 nm (B2 to B8A) and the combination of these bands are of high importance, which is similar to previous studies [16]. Spectral index is a powerful supplement to SOM prediction: five of the top ten inputs in SOM prediction of single-phase images are spectral indexes. When using synthetic images for SOM prediction, the contribution of the spectral index is not so strong, and only three of the top ten inputs are spectral indexes.

4.4. Comparison of Different SOM Content Mapping Results

We compared the spatial distribution map of SOM content in this study with the mapping results of SOM content in other studies (Figure 8). We found that the SOM content predicted by Soil Subcenter (http://soil.geodata.cn, accessed on 15 June 2022) and Poggio, et al. [60] was higher than that in our study, which may be because the results of the Soil Subcenter were produced in 1990, when the soil was not yet greatly degraded [61]. These results were obtained from national and global perspectives. In the mapping results, our mapping results had large spatial differences, which could better reflect the actual situation of the soil surface. In the mapping results of Soil Subcenter and Poggio, et al., the spatial variability of SOM content in local areas was small. It should be emphasized that the mapping results of the Soil Subcenter and previous study were of great significance to the spatial distribution of soil organic carbon content in China and even the world.

4.5. Limitations and Prospects

In this study, the influence of images from different shooting dates on SOM prediction was considered. The relationships amongst soil moisture, straw content, and SOM content were analyzed, and the single-phase and multi-phase synthetic images were compared, which could provide a reference for selecting the best bare soil period image for SOM content prediction at the same latitude. It has been proved that although feature selection can improve the accuracy of SOM prediction, the improvement was very small [18]. Combined with the characteristics of random forest regression, without feature selection the accuracy of SOM prediction would not be affected too much, and the use efficiency of the model would be improved. Since this study only evaluated the highest accuracy that could be obtained from experimental remote sensing data, many studies have proven that adding environmental elements (meteorological data, terrain data) can further improve the accuracy of SOM prediction, which is needed for the next step of research. There are many paddy fields in this study area, which impacted the prediction accuracy of SOM [54,62]. Therefore, in future research, paddy fields and dry land will be modeled separately, and more input variables considering soil forming factors will be added to improve the prediction accuracy of SOM.

5. Conclusions

This study used 2019–2021 to cover all available Sentinel-2 remote sensing images of Youyi Farm in the study area. Combined with a RF algorithm, this study evaluated the performance difference of SOM prediction using remote sensing images in different periods to select the time window of SOM prediction in the study area. Compared with SOM prediction using remote sensing in the growth period, bare soil images can describe the soil surface more directly. However, they are also more vulnerable to changes in the soil environment, and the SOM prediction result was more unstable. The results showed that the image time difference had a great impact on the SOM prediction image. The best time window for SOM prediction in the study area was May; this was mainly because the soil moisture content in May was more appropriate than that in other bare soil periods. The difference in SOM prediction accuracy in the bare soil period was affected by factors such as soil water content and straw coverage. This study emphasized that selecting the correct remote sensing image is the first step for soil mapping; otherwise, the subsequent model optimization may be useless.

Author Contributions

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

Funding

This work was supported by the National Key R&D Program of China (Grant No. 2021YFD1500105).

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Informed consent was obtained from all subjects involved in the study.

Data Availability Statement

Data are available upon request by email to the author.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Kallenbach, C.M.; Frey, S.D.; Grandy, A.S. Direct evidence for microbial-derived soil organic matter formation and its ecophysiological controls. Nat. Commun. 2016, 7, 1–10. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  2. Tadini, A.M.; Xavier, A.A.; Milori, D.M.; Oliveira, P.P.; Pezzopane, J.R.; Bernardi, A.C.; Martin-Neto, L. Evaluation of soil organic matter from integrated production systems using laser-induced fluorescence spectroscopy. Soil Tillage Res. 2021, 211, 105001. [Google Scholar] [CrossRef]
  3. Lal, R. Soil carbon management and climate change. Carbon Manag. 2013, 4, 439–462. [Google Scholar] [CrossRef]
  4. Lal, R. Soil carbon dynamics in cropland and rangeland. Environ. Pollut. 2002, 116, 353–362. [Google Scholar] [CrossRef]
  5. Soon, Y.; Arshad, M.; Haq, A.; Lupwayi, N. The influence of 12 years of tillage and crop rotation on total and labile organic carbon in a sandy loam soil. Soil Tillage Res. 2007, 95, 38–46. [Google Scholar] [CrossRef]
  6. Dube, E.; Chiduza, C.; Muchaonyerwa, P. Conservation agriculture effects on soil organic matter on a Haplic Cambisol after four years of maize–oat and maize–grazing vetch rotations in South Africa. Soil Tillage Res. 2012, 123, 21–28. [Google Scholar] [CrossRef]
  7. Lal, R. Soil carbon sequestration impacts on global climate change and food security. Science 2004, 304, 1623–1627. [Google Scholar] [CrossRef] [Green Version]
  8. Van der Werf, G.R.; Randerson, J.T.; Collatz, G.J.; Giglio, L. Carbon emissions from fires in tropical and subtropical ecosystems. Glob. Change Biol. 2003, 9, 547–562. [Google Scholar] [CrossRef] [Green Version]
  9. Gong, W.; Yan, X.; Wang, J.; Hu, T.; Gong, Y. Long-term manure and fertilizer effects on soil organic matter fractions and microbes under a wheat–maize cropping system in northern China. Geoderma 2009, 149, 318–324. [Google Scholar] [CrossRef]
  10. Fleming, K.; Heermann, D.; Westfall, D. Evaluating soil color with farmer input and apparent soil electrical conductivity for management zone delineation. Agron. J. 2004, 96, 1581–1587. [Google Scholar] [CrossRef]
  11. Seely, B.; Welham, C.; Blanco, J.A. Towards the application of soil organic matter as an indicator of forest ecosystem productivity: Deriving thresholds, developing monitoring systems, and evaluating practices. Ecol. Indic. 2010, 10, 999–1008. [Google Scholar] [CrossRef]
  12. Six, J.; Paustian, K. Aggregate-associated soil organic matter as an ecosystem property and a measurement tool. Soil Biol. Biochem. 2014, 68, A4–A9. [Google Scholar] [CrossRef]
  13. Bao, Y.; Meng, X.; Ustin, S.; Wang, X.; Zhang, X.; Liu, H.; Tang, H. Vis-SWIR spectral prediction model for soil organic matter with different grouping strategies. Catena 2020, 195, 104703. [Google Scholar] [CrossRef]
  14. Fan, M.; Lal, R.; Zhang, H.; Margenot, A.J.; Wu, J.; Wu, P.; Zhang, L.; Yao, J.; Chen, F.; Gao, C. Variability and determinants of soil organic matter under different land uses and soil types in eastern China. Soil Tillage Res. 2020, 198, 104544. [Google Scholar] [CrossRef]
  15. Meng, X.; Bao, Y.; Ye, Q.; Liu, H.; Zhang, X.; Tang, H.; Zhang, X. Soil Organic Matter Prediction Model with Satellite Hyperspectral Image Based on Optimized Denoising Method. Remote Sens. 2021, 13, 2273. [Google Scholar] [CrossRef]
  16. Meng, X.; Bao, Y.; Liu, J.; Liu, H.; Zhang, X.; Zhang, Y.; Wang, P.; Tang, H.; Kong, F. Regional soil organic carbon prediction model based on a discrete wavelet analysis of hyperspectral satellite data. Int. J. Appl. Earth Obs. Geoinf. 2020, 89, 102111. [Google Scholar] [CrossRef]
  17. Luo, C.; Zhang, X.; Meng, X.; Zhu, H.; Ni, C.; Chen, M.; Liu, H. Regional mapping of soil organic matter content using multitemporal synthetic Landsat 8 images in Google Earth Engine. Catena 2022, 209, 105842. [Google Scholar] [CrossRef]
  18. Luo, C.; Zhang, X.; Wang, Y.; Men, Z.; Liu, H. Regional soil organic matter mapping models based on the optimal time window, feature selection algorithm and Google Earth Engine. Soil Tillage Res. 2022, 219, 105325. [Google Scholar] [CrossRef]
  19. Luo, C.; Wang, Y.; Zhang, X.; Zhang, W.; Liu, H. Spatial prediction of soil organic matter content using multiyear synthetic images and partitioning algorithms. Catena 2022, 211, 106023. [Google Scholar] [CrossRef]
  20. Ge, X.; Ding, J.; Jin, X.; Wang, J.; Chen, X.; Li, X.; Liu, J.; Xie, B. Estimating Agricultural Soil Moisture Content through UAV-Based Hyperspectral Images in the Arid Region. Remote Sens. 2021, 13, 1562. [Google Scholar] [CrossRef]
  21. Ge, X.; Ding, J.; Teng, D.; Xie, B.; Zhang, X.; Wang, J.; Han, L.; Bao, Q.; Wang, J. Exploring the capability of Gaofen-5 hyperspectral data for assessing soil salinity risks. Int. J. Appl. Earth Obs. Geoinf. 2022, 112, 102969. [Google Scholar] [CrossRef]
  22. 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]
  23. Liu, T.; Liu, H.; Qi, Y. Construction land expansion and cultivated land protection in urbanizing China: Insights from national land surveys, 1996–2006. Habitat Int. 2015, 46, 13–22. [Google Scholar] [CrossRef]
  24. Pahlavan-Rad, M.R.; Dahmardeh, K.; Brungard, C. Predicting soil organic carbon concentrations in a low relief landscape, eastern Iran. Geoderma Reg. 2018, 15, e00195. [Google Scholar] [CrossRef]
  25. Silvero, N.E.Q.; Demattê, J.A.M.; Amorim, M.T.A.; dos Santos, N.V.; Rizzo, R.; Safanelli, J.L.; Poppiel, R.R.; de Sousa Mendes, W.; Bonfatti, B.R. Soil variability and quantification based on Sentinel-2 and Landsat-8 bare soil images: A comparison. Remote Sens. Environ. 2021, 252, 112117. [Google Scholar] [CrossRef]
  26. Shafizadeh-Moghadam, H.; Minaei, F.; Talebi-khiyavi, H.; Xu, T.; Homaee, M. Synergetic use of multi-temporal Sentinel-1, Sentinel-2, NDVI, and topographic factors for estimating soil organic carbon. Catena 2022, 212, 106077. [Google Scholar] [CrossRef]
  27. Dou, X.; Wang, X.; Liu, H.; Zhang, X.; Meng, L.; Pan, Y.; Yu, Z.; Cui, Y. Prediction of soil organic matter using multi-temporal satellite images in the Songnen Plain, China. Geoderma 2019, 356, 113896. [Google Scholar] [CrossRef]
  28. Meng, X.; Bao, Y.; Wang, Y.; Zhang, X.; Liu, H. 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]
  29. Bouktif, S.; Fiaz, A.; Ouni, A.; Serhani, M.A. Optimal deep learning lstm model for electric load forecasting using feature selection and genetic algorithm: Comparison with machine learning approaches. Energies 2018, 11, 1636. [Google Scholar] [CrossRef] [Green Version]
  30. Wang, H.; Lei, Z.; Zhang, X.; Zhou, B.; Peng, J. A review of deep learning for renewable energy forecasting. Energy Convers. Manag. 2019, 198, 111799. [Google Scholar] [CrossRef]
  31. Zhang, F.; Yang, X. Improving land cover classification in an urbanized coastal area by random forests: The role of variable selection. Remote Sens. Environ. 2020, 251, 112105. [Google Scholar] [CrossRef]
  32. Taghizadeh-Mehrjardi, R.; Schmidt, K.; Amirian-Chakan, A.; Rentschler, T.; Zeraatpisheh, M.; Sarmadian, F.; Valavi, R.; Davatgar, N.; Behrens, T.; Scholten, T. Improving the spatial prediction of soil organic carbon content in two contrasting climatic regions by stacking machine learning models and rescanning covariate space. Remote Sens. 2020, 12, 1095. [Google Scholar] [CrossRef] [Green Version]
  33. Luo, C.; Qi, B.; Liu, H.; Guo, D.; Lu, L.; Fu, Q.; Shao, Y. Using Time Series Sentinel-1 Images for Object-Oriented Crop Classification in Google Earth Engine. Remote Sens. 2021, 13, 561. [Google Scholar] [CrossRef]
  34. Chen, D.; Zhang, F.; Zhang, M.; Meng, Q.; Jim, C.Y.; Shi, J.; Tan, M.L.; Ma, X. Landscape and vegetation traits of urban green space can predict local surface temperature. Sci. Total Environ. 2022, 825, 154006. [Google Scholar] [CrossRef] [PubMed]
  35. Chen, M.; Qiu, X.; Zeng, W.; Peng, D. Combining Sample Plot Stratification and Machine Learning Algorithms to Improve Forest Aboveground Carbon Density Estimation in Northeast China Using Airborne LiDAR Data. Remote Sens. 2022, 14, 1477. [Google Scholar] [CrossRef]
  36. Hou, Y.; Zhang, A.; Lv, R.; Zhao, S.; Ma, J.; Zhang, H.; Li, Z. A study on water quality parameters estimation for urban rivers based on ground hyperspectral remote sensing technology. Environ. Sci. Pollut. Res. 2022, 29, 63640–63654. [Google Scholar] [CrossRef]
  37. Ji, Y.; Chen, Z.; Cheng, Q.; Liu, R.; Li, M.; Yan, X.; Li, G.; Wang, D.; Fu, L.; Ma, Y.; et al. Estimation of plant height and yield based on UAV imagery in faba bean (Vicia faba L.). Plant Methods 2022, 18, 1–13. [Google Scholar] [CrossRef]
  38. Marshall, M.; Belgiu, M.; Boschetti, M.; Pepe, M.; Stein, A.; Nelson, A. Field-level crop yield estimation with PRISMA and Sentinel-2. Isprs J. Photogramm. Remote Sens. 2022, 187, 191–210. [Google Scholar] [CrossRef]
  39. Duarte, E.; Zagal, E.; Barrera, J.A.; Dube, F.; Casco, F.; Hernandez, A.J. Digital mapping of soil organic carbon stocks in the forest lands of Dominican Republic. Eur. J. Remote Sens. 2022, 55, 213–231. [Google Scholar] [CrossRef]
  40. Pahlevan, N.; Sarkar, S.; Franz, B.; Balasubramanian, S.; He, J. Sentinel-2 MultiSpectral Instrument (MSI) data processing for aquatic science applications: Demonstrations and validations. Remote Sens. Environ. 2017, 201, 47–56. [Google Scholar] [CrossRef]
  41. Segarra, J.; Buchaillot, M.L.; Araus, J.L.; Kefauver, S.C. Remote sensing for precision agriculture: Sentinel-2 improved features and applications. Agronomy 2020, 10, 641. [Google Scholar] [CrossRef]
  42. Ge, X.; Ding, J.; Teng, D.; Wang, J.; Huo, T.; Jin, X.; Wang, J.; He, B.; Han, L. Updated soil salinity with fine spatial resolution and high accuracy: The synergy of Sentinel-2 MSI, environmental covariates and hybrid machine learning approaches. Catena 2022, 212, 106054. [Google Scholar] [CrossRef]
  43. Wang, X.; Wang, L.; Li, S.; Wang, Z.; Zheng, M.; Song, K. Remote estimates of soil organic carbon using multi-temporal synthetic images and the probability hybrid model. Geoderma 2022, 425, 116066. [Google Scholar] [CrossRef]
  44. Demattê, J.A.M.; Fongaro, C.T.; Rizzo, R.; Safanelli, J.L. Geospatial Soil Sensing System (GEOS3): A powerful data mining procedure to retrieve soil spectral reflectance from satellite images. Remote Sens. Environ. 2018, 212, 161–175. [Google Scholar] [CrossRef]
  45. Vaudour, E.; Baghdadi, N.; Gilliot, J.-M. Mapping tillage operations over a peri-urban region using combined SPOT4 and ASAR/ENVISAT images. Int. J. Appl. Earth Obs. Geoinf. 2014, 28, 43–59. [Google Scholar] [CrossRef]
  46. Song, B.; Liu, L.; Du, S.; Zhang, X.; Chen, X.; Zhang, H. ValLAI_Crop, a validation dataset for coarse-resolution satellite LAI products over Chinese cropland. Sci. Data 2021, 8, 1–16. [Google Scholar] [CrossRef]
  47. Pribyl, D.W. A critical review of the conventional SOC to SOM conversion factor. Geoderma 2010, 156, 75–83. [Google Scholar] [CrossRef]
  48. Claverie, M.; Ju, J.; Masek, J.G.; Dungan, J.L.; Vermote, E.F.; Roger, J.-C.; Skakun, S.V.; Justice, C. The Harmonized Landsat and Sentinel-2 surface reflectance data set. Remote Sens. Environ. 2018, 219, 145–161. [Google Scholar] [CrossRef]
  49. Teluguntla, P.; Thenkabail, P.S.; Oliphant, A.; Xiong, J.; Gumma, M.K.; Congalton, R.G.; Yadav, K.; Huete, A. A 30-m landsat-derived cropland extent product of Australia and China using random forest machine learning algorithm on Google Earth Engine cloud computing platform. Isprs J. Photogramm. Remote Sens. 2018, 144, 325–340. [Google Scholar] [CrossRef]
  50. 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]
  51. Saptoro, A.; Tadé, M.O.; Vuthaluru, H. A modified Kennard-Stone algorithm for optimal division of data for developing artificial neural network models. Chem. Prod. Process Model. 2012, 7, 16. [Google Scholar] [CrossRef]
  52. Liu, F.; Geng, X.; Zhu, A.-X.; Fraser, W.; Waddell, A. Soil texture mapping over low relief areas using land surface feedback dynamic patterns extracted from MODIS. Geoderma 2012, 171, 44–52. [Google Scholar] [CrossRef]
  53. Guo, L.; Sun, X.; Fu, P.; Shi, T.; Dang, L.; Chen, Y.; Linderman, M.; Zhang, G.; Zhang, Y.; Jiang, Q. Mapping soil organic carbon stock by hyperspectral and time-series multispectral remote sensing images in low-relief agricultural areas. Geoderma 2021, 398, 115118. [Google Scholar] [CrossRef]
  54. Chong, L.U.O.; Liu, H.J.; Lu, L.P.; Liu, Z.R.; Kong, F.C.; Zhang, X.L. Monthly composites from Sentinel-1 and Sentinel-2 images for regional major crop mapping with Google Earth Engine. J. Integr. Agric. 2021, 20, 1944–1957. [Google Scholar] [CrossRef]
  55. Gao, B.-C. NDWI—A normalized difference water index for remote sensing of vegetation liquid water from space. Remote Sens. Environ. 1996, 58, 257–266. [Google Scholar] [CrossRef]
  56. Jin, X.; Du, J.; Liu, H.; Wang, Z.; Song, K. Remote estimation of soil organic matter content in the Sanjiang Plain, Northest China: The optimal band algorithm versus the GRA-ANN model. Agric. For. Meteorol. 2016, 218, 250–260. [Google Scholar] [CrossRef]
  57. Zheng, B.; Campbell, J.B.; de Beurs, K.M. Remote sensing of crop residue cover using multi-temporal Landsat imagery. Remote Sens. Environ. 2012, 117, 177–183. [Google Scholar] [CrossRef]
  58. Xiang, X.; Du, J.; Jacinthe, P.-A.; Zhao, B.; Zhou, H.; Liu, H.; Song, K. Integration of tillage indices and textural features of Sentinel-2A multispectral images for maize residue cover estimation. Soil Tillage Res. 2022, 221, 105405. [Google Scholar] [CrossRef]
  59. Pan, Y.; Zhang, X.; Liu, H.; Wu, D.; Dou, X.; Xu, M.; Jiang, Y. Remote sensing inversion of soil organic matter by using the subregion method at the field scale. Precis. Agric. 2022, 23, 1813–1835. [Google Scholar] [CrossRef]
  60. Poggio, L.; de Sousa, L.M.; Batjes, N.H.; Heuvelink, G.; Kempen, B.; Ribeiro, E.; Rossiter, D. SoilGrids 2.0: Producing soil information for the globe with quantified spatial uncertainty. Soil 2021, 7, 217–240. [Google Scholar] [CrossRef]
  61. Zhao, M.-S.; Rossiter, D.G.; Li, D.-C.; Zhao, Y.-G.; Liu, F.; Zhang, G.-L. Mapping soil organic matter in low-relief areas based on land surface diurnal temperature difference and a vegetation index. Ecol. Indic. 2014, 39, 120–133. [Google Scholar] [CrossRef]
  62. Chong, L.U.O.; Liu, H.J.; Qiang, F.U.; Guan, H.X.; Qiang, Y.E.; Zhang, X.L.; Kong, F.C. Mapping the fallowed area of paddy fields on Sanjiang Plain of Northeast China to assist water security assessments. J. Integr. Agric. 2020, 19, 1885–1896. [Google Scholar] [CrossRef]
Sustainability 15 00469 g001 550
Figure 1. Overview of the study area. (a) Distribution map of soil sampling points; (b–d) Photos of bare soil period in the study area. All photos were taken in early April 2021.
Figure 1. Overview of the study area. (a) Distribution map of soil sampling points; (b–d) Photos of bare soil period in the study area. All photos were taken in early April 2021.
Sustainability 15 00469 g001
Sustainability 15 00469 g002 550
Figure 2. Technology roadmap.
Figure 2. Technology roadmap.
Sustainability 15 00469 g002
Sustainability 15 00469 g003 550
Figure 3. Sentinel-2 (date: 2021-05-17) spectral curve of different organic matter contents.
Figure 3. Sentinel-2 (date: 2021-05-17) spectral curve of different organic matter contents.
Sustainability 15 00469 g003
Sustainability 15 00469 g004 550
Figure 4. Spatial distribution of SOM content in Youyi Farm. (a) The bare soil image with the highest precision of SOM inversion on 17 May 2021; (b) the growing season image with the highest precision of SOM inversion on 18 July 2021.
Figure 4. Spatial distribution of SOM content in Youyi Farm. (a) The bare soil image with the highest precision of SOM inversion on 17 May 2021; (b) the growing season image with the highest precision of SOM inversion on 18 July 2021.
Sustainability 15 00469 g004
Sustainability 15 00469 g005 550
Figure 5. Field photos of the study area in different periods. (a) Shot on 2 April 2021; (b) Shot on 28 May 2021.
Figure 5. Field photos of the study area in different periods. (a) Shot on 2 April 2021; (b) Shot on 28 May 2021.
Sustainability 15 00469 g005
Sustainability 15 00469 g006 550
Figure 6. NDTI and NDMI values of image sampling points at different times.
Figure 6. NDTI and NDMI values of image sampling points at different times.
Sustainability 15 00469 g006
Sustainability 15 00469 g007 550
Figure 7. Top 20 input for importance. (a) 17 May 2021; (b) Synthetic image in May.
Figure 7. Top 20 input for importance. (a) 17 May 2021; (b) Synthetic image in May.
Sustainability 15 00469 g007
Sustainability 15 00469 g008 550
Figure 8. Comparison of mapping results of different SOM contents. (a–c) The mapping results of the study, Soil Subcenter, and Poggio, et al.
Figure 8. Comparison of mapping results of different SOM contents. (a–c) The mapping results of the study, Soil Subcenter, and Poggio, et al.
Sustainability 15 00469 g008
Table
Table 1. Properties of images selected for this study.
Table 1. Properties of images selected for this study.
YearDate (Mon/D)YearDate (Mon/D)YearDate (Mon/D)
201903/29202004/12202104/04
05/0304/2904/19
06/2405/0705/17
08/3105/2906/08
09/2707/2306/13
11/0409/2306/23
10/2607/18
08/17
09/06
10/29
Table
Table 2. Descriptive statistical analysis of SOM content.
Table 2. Descriptive statistical analysis of SOM content.
SetNMax (%)Min (%)Mean (%)SD (%)CV (%)
Whole dataset1889.910.483.921.4035.84
Training set1419.010.483.891.3735.26
validation set479.911.754.021.5037.29
Note: SD represents standard deviation; CV stands for coefficient of variation.
Table
Table 3. SOM prediction accuracy of single-phase images at different times.
Table 3. SOM prediction accuracy of single-phase images at different times.
TimeYearImage TimeOnly BandsBands + Spectral Indices
R2RMSE (%)R2RMSE (%)
growing season2019201906240.450.950.470.94
201908310.301.080.411.00
201909270.241.140.261.14
2020202007230.271.120.251.13
202009210.421.100.281.01
2021202106080.381.030.470.95
202106130.361.040.391.01
202106230.371.020.361.03
202107180.500.920.510.92
202108170.231.140.261.15
202109060.351.040.271.10
bare soil period2019201903290.231.140.281.11
201905030.480.940.520.90
201911040.251.130.281.11
2020202004120.051.300.171.18
202004290.301.100.411.00
202005070.301.100.460.95
202005290.421.000.430.99
202010260.261.120.410.10
2021202104040.181.170.191.17
202104190.530.890.590.83
202105170.420.980.590.82
202110290.101.270.161.23
Table
Table 4. SOM prediction accuracy of multi-year synthetic images at different times.
Table 4. SOM prediction accuracy of multi-year synthetic images at different times.
YearImage TimeOnly BandsBands + Spectral Indices
R2RMSE (%)R2RMSE (%)
2019–2021Apr.0.321.060.351.04
May0.550.870.560.85
Jun.0.061.270.221.14
Jul.0.321.070.391.01
Aug.0.371.040.401.10
Sep.0.371.020.401.00
Oct.0.111.240.181.19
Table
Table 5. Correlation analysis between RMSE of SOM prediction using different time images and NDTI and NDMI values of sampling points.
Table 5. Correlation analysis between RMSE of SOM prediction using different time images and NDTI and NDMI values of sampling points.
CorrelationNDTINDMIRMSE
NDTI1 **
NDMI−0.22 **1 **
RMSE0.18 **−0.60 **1 **
Note: ** Represents significance at p < 0.01 level.
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

Wang, Y.; Luo, C.; Zhang, W.; Meng, X.; Liu, Q.; Zhang, X.; Liu, H. Remote Sensing Prediction Model of Cultivated Land Soil Organic Matter Considering the Best Time Window. Sustainability 2023, 15, 469. https://doi.org/10.3390/su15010469

AMA Style

Wang Y, Luo C, Zhang W, Meng X, Liu Q, Zhang X, Liu H. Remote Sensing Prediction Model of Cultivated Land Soil Organic Matter Considering the Best Time Window. Sustainability. 2023; 15(1):469. https://doi.org/10.3390/su15010469

Chicago/Turabian Style

Wang, Yiang, Chong Luo, Wenqi Zhang, Xiangtian Meng, Qiong Liu, Xinle Zhang, and Huanjun Liu. 2023. "Remote Sensing Prediction Model of Cultivated Land Soil Organic Matter Considering the Best Time Window" Sustainability 15, no. 1: 469. https://doi.org/10.3390/su15010469

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