Next Article in Journal
Investigating Stability Parameters for Agronomic and Quality Traits of Durum Wheat Grown under Mediterranean Conditions
Previous Article in Journal
How Do Long Term Crop Rotations Influence Weed Populations: Exploring the Impacts of More than 50 Years of Crop Management in Serbia
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Agronomy
Volume 12
Issue 8
10.3390/agronomy12081773
agronomy-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles 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

Recognition and Changes Analysis of Complex Planting Patterns Based Time Series Landsat and Sentinel-2 Images in Jianghan Plain, China

by
Zijing Zhang
1,
Li Hua
1,*,
Qi Wei
1,
Jialin Li
1 and
Jianxun Wang
2
1
College of Resources and Environment, Huazhong Agricultural University, Wuhan 430070, China
2
State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan 430070, China
*
Author to whom correspondence should be addressed.
Agronomy 2022, 12(8), 1773; https://doi.org/10.3390/agronomy12081773
Submission received: 2 July 2022 / Revised: 24 July 2022 / Accepted: 26 July 2022 / Published: 28 July 2022
Browse Figures
Versions Notes

Abstract

:
Accurate and timely information on crop planting patterns is crucial for research on sustainable agriculture, regional resources, and food security. However, existing spatial datasets have few high-precision and wide-range planting pattern maps. The production may be limited by the unbalanced spatiotemporal resolution, insufficient massive field sample data, low local computer processing speed, and other factors. To overcome these limitations, we proposed semi-automatic expansion and spatiotemporal migration strategies for sample points and performed a pixel-and-phenology-based random forest algorithm on the Google Earth Engine platform to generate crop planting pattern maps at high spatiotemporal resolution by integrating Landsat-8 and Sentinel-2 time series image data. In this study, we report planting pattern maps for 2017–2021 at a 10-m spatial resolution of the Jianghan Plain, including six crops and nine planting patterns, with an overall accuracy of 84–94% and a kappa coefficient of 0.80–0.93. The spatiotemporal distribution is driven by multiple factors, such as subjectivity and social economy. This research indicates that the proposed approach is effective for mapping large-scale planting patterns and can be readily applied to other regions.

1. Introduction

Scientific planting patterns are not only crucial for crop production and food security promotion [1,2,3] but also related to environmentally friendly agriculture and sustainability [4,5]. Recently, the agro-ecological environment has gradually deteriorated and the sustainability of cultivated land has become a challenge, which is attributed to the world population and food demand growths [6,7]. These problems can be ameliorated by implementing scientific planting patterns and monitoring them in a timely and accurate manner [8]. Additionally, high-precision spatial planting pattern datasets are crucial for agricultural ecology and crop models [9,10]. The model outputs aid in improving food production evaluation and prediction accuracies and serve as an essential basis for optimizing regional planting structures [10,11]. Crop planting pattern maps can be produced using fieldwork or remote sensing images. Remote sensing techniques are more prevalent than ground surveying methods owing to their low cost, broad coverage, and regular acquisition [12].
Several remote sensing approaches have generated crop type and planting pattern maps at a moderate spatial resolution. Many studies have identified planting patterns by analyzing crop phenological cycles within vegetation index (VI) data, including normalized difference vegetation index (NDVI) [13], enhanced vegetation index (EVI) [14], and ratio vegetation index(RVI) [15], which are typically calculated from MODIS images at 500-m and 250-m spatial resolutions. For example, Chen et al. [16] introduced MODIS data and a decision tree classifier to map six planting patterns of five crops in Brazil to find mapping accuracies for farmland, planting pattern, and crop type as 90%, 73%, and 86%, respectively. Huang et al. [17] obtained spatial distribution information of the dominant planting patterns in northeast China by establishing a MODIS-NDVI model, with an overall extraction accuracy of over 87%; however, affected by the spatial resolution of MODIS, its pixels usually contain a mixture of different crops, which might lead to sub-pixel heterogeneity of crop types and planting patterns [18]. Therefore, it is difficult to accurately monitor precise planting pattern information in large-scale spaces.
To overcome the sub-pixel heterogeneity of MODIS, higher spatial resolution images should be used, to map planting patterns [19]. For example, using Landsat images with 30-m resolutions, the cropland data layer (CDL) with more than 100 crop types classified has been produced annually by the U.S. Department of Agriculture (USDA) National Agricultural Statistics Service (NASS) [20] and planting/tillage classifications [21,22] and global food security support analysis data (GFSAD30) [23] have been produced in the National Land Cover Dataset (NLCD) by the U.S. Geological Survey (USGS). However, mapping accuracy relies heavily on the availability of Landsat time series imagery [19]. In some regions, no high-quality Landsat data can be obtained during the planting cycle because of the heavy clouds or shadows [18]. Although Landsat provides higher spatial resolutions than MODIS, it may not contribute the temporal resolutions required to identify complex planting patterns.
Sentinel series satellites can return large-scale, long-term, and high-precision data [24]. Sentinel-2 satellites have a 10-m multispectral band resolution and a revisit frequency of 5 d, which can provide similar spectral information and spatial resolutions as Landsat. The integration of Landsat and Sentinel-2 data will effectively improve the spatiotemporal resolution of the data [25], provide more precise extraction of planting patterns than MODIS and Landsat. Some studies have combined Landsat and Sentinel-2 for extraction research on a regional scale or single crop [26,27,28,29,30], but there is a need to evaluate the potential of combining both data in high-precision extraction of complex planting patterns in large areas.
At present, planting pattern classification based on remote sensing data mainly adopts supervised classification [16,31,32] and is restricted by samples. Large-area field sampling is limited by time, manpower, and geography, so massive sample data covering the entire study area cannot be easily acquired for model training [33]. Seasonal samples are often only used for current-season planting pattern classification and are rarely used for the next few seasons, resulting in a low utilization rate of historical data [34]. Additionally, supervised classification relies heavily on the quantity and quality of training samples, and sufficient and representative samples can allow for higher classification result accuracies. Previous studies showed that this accuracy is affected by “high quality samples” [35,36]; therefore, it is important for large-area and refined planting pattern mapping to overcome sample quantity and quality constraints.
As an important grain-producing area in southern China, the vast Jianghan Plain has complex crop types and planting patterns, making it is difficult to map its precise planting patterns. In this study, we proposed semi-automatic expansion and spatiotemporal migration strategies for sample points and performed a pixel-and-phenology-based random forest (RF) algorithm on the Google Earth Engine (GEE) platform to generate 10-m resolution planting pattern maps over Jianghan Plain by integrating Landsat-8 and Sentinel-2 time series data. The specific objectives of this study are to explore the potential of Sentinel-2/Landsat-8 time series integration data and semi-automatic spatiotemporal migration strategies of sample points in complex planting pattern recognition for four years; to analyze the spatiotemporal changes in planting patterns and the driving factors in the Jianghan Plain; and to provide available high-precision spatial datasets for related research for a scientific basis of the formulation of related policies.

2. Materials and Methods

2.1. Study Area

Located in the central and southern parts of Hubei Province, the Jianghan Plain is named after the impact of the Yangtze and Hanjiang Rivers (Figure 1a). In our study, 17 county/city-level administrative divisions were selected as demarcation lines, covering an area of 28,000 km2.
Local terrain, climate, and hydrological conditions are suitable for farming, and the study area is a plain, except for some eastern hills (Figure 1c), which are dominated by cropland (Figure 1b). Controlled by the subtropical monsoon climate zone, the Jianghan Plain is mild throughout the year, with sufficient sunshine and an annual average temperature of 16.3 °C. Additionally, there are dense rivers and lakes, and the average annual precipitation is between 1100–1300 mm. Superior natural and geographical conditions are conducive for the development of local agriculture and aquaculture.
The Jianghan Plain is an important commodity grain base in China with complex crop types and planting patterns. It primarily produces grain crops, such as wheat and rice, and economic crops, such as rapeseed and cotton [37]. Benefitting from good natural conditions, most planting patterns are double cropping a year, forming a complex agricultural landscape with multiple planting patterns, such as “rapeseed–rice,” “wheat–rice,” “wheat–soybean,” “rapeseed–corn,” and “crayfish farming–rice”.

2.2. Landsat-8 and Sentinel-2 Images and Pre-Processing

2.2.1. Image Selection

To map planting patterns from 2017 to 2021, Landsat-8 and Sentinel-2 data between April 2017 and April 2021 were used because the study area was dominated by a planting pattern of double cropping a year; the first crop was sown from around April, and the second crop was harvested in April of the following year. GEE platform hosts both Sentinel-2 MultiSpectral Instrument (MSI) images and Landsat-8 Operational Land Imager (OLI) images. All Sentinel-2 and Landsat-8 images we used were top-of-atmosphere (TOA) data (Figure 2).

2.2.2. Imagery Harmonization and Image Fusion

Remote sensing data must be unified because of the slight differences in the wavelengths of the bands between sensors. Therefore, we harmonized the OLI data with the MSI data. For the near-infrared, shortwave-infrared, red, and blue bands ordinary least squares regression coefficients were used for this transformation, using regression coefficients based on Zhang et al. [38]. Due to the differences in spatial resolutions between OLI and MSI data, bicubic resampling was introduced to resample the OLI data and shortwave-infrared MSI data to 10 m × 10 m.
Sentinel-2 images with cloud cover of less than 10% were screened. Due to rainy weather, hollows and gaps were found in Sentinel-2 time series images after cloud removal and screening. Re-sampled Landsat-8 images of the same period, which had varying coefficients, were cut and spliced for fusion to fill Sentinel-2 data vacancies, and they formed a complete time series fusion image of the study area with a spatial resolution of 10 m.

2.2.3. Preparation of a Monthly Composite Images

Images within close temporal proximity in different periods were combined to display a complete image of the study area because of poor image quality; however, the acquired images will cause some differences in the observed VI values. Therefore, to generate a time series of the same lengths and intervals, we obtained monthly NDVI, EVI, and land surface water index (LSWI) composite datasets by calculating the average VI values of all possible observations in each the month, as follows:
NDVI = ( NIR − RED ) / ( NIR + RED )
EVI = 2.5 × NIR − RED NIR + 6.0 × RED − 7.5 × BLUE + 1
LSWI = ( SWIR − NIR ) / ( SWIR + NIR )
where RED, NIR, BLUE, and SWIR correspond to bands B4, B8, B2, and B11 in Sentinel-2 and B4, B5, B2, and B6 in Landsat-8, respectively.

2.3. Field-Based Samples and Survey Data

As the study area is dominated by annual double cropping, instantaneous field sampling data cannot be used to reflect the planting pattern there. Field surveys were conducted in June 2019 and March 2020 with the aid of an unmanned aerial vehicle (UAV) to shoot and collect as many ground data samples as possible for training and validation. Questionnaires were distributed to local farmers, including the questions on crop type, planting pattern, and phenological date. The field surveys collected 258 UAV images, 300 photographs, and 165 questionnaires across the study area (Figure 3).
The main planting patterns in the study area were determined, based on photographs, UAV images, and questionnaires, as wheat–cotton, wheat–rice, wheat–soybean, wheat–corn, rapeseed–cotton, rapeseed–rice, rapeseed–soybean, and rapeseed–corn rotations and single rice. Planting pattern classifications of the 508 samples were determined through vectorization and were used to determine the standard VI profiles. The samples were divided into training samples and verification points at a 7:3 ratio for model training and accuracy verification.

2.4. Pixel-Based Planting Pattern Extraction Method

We proposed semi-automatic expansion and spatiotemporal migration strategies for sample points and performed a RF algorithm on the GEE platform to generate crop planting pattern maps, as shown in the flow chart (Figure 4).

2.4.1. Analysis of Phenological Characteristics

In order to observe the phenological characteristics of different crop planting patterns, NDVI, LSWI, and EVI time series were created using field sampling data and questionnaires and were then smoothed using an adaptive Savitzky–Golay filter [39], a common technique used in crop identification. Crop phenological characteristics in the VIs time series can help to identify planting patterns and crop types (Figure 5).
In a one-year phenological cycle (i.e., April of the current year to April of the following year), compared with other double-cropping crops, the NDVI value profile of single rice had only one peak (sowing period in May and harvest period in September), which can be distinguished from other bimodal crop patterns. For double-cropping planting patterns, winter crops included rapeseed and wheat (sowing period in October and harvest period in May). March is the most important period for identifying rapeseed and wheat, as the NDVI of rapeseed undergoes an obvious trough in full bloom, while wheat has a high NDVI value at the jointing and booting stages. Summer crops included cotton, soybean (summer sowing soybean), corn (summer sowing corn), and rice. The NDVI profile shows that cotton is sown the earliest; therefore, its NDVI value rises first and has the shortest growth cycle (sowing period in April and matured and harvested several times from August to November), which can be used to distinguish cotton from other crops. After the LSWI profile was added, the LSWI of rice in the transplanting period (April–May) was higher than that of the other crops because of the higher soil moisture content. The EVI profile can be used to distinguish soybean and corn, and it explains why corn has higher NDVI and EVI values than soybean during the growth period [40].
Based on the phenological characteristics of crops and planting patterns, NDVI, EVI, and LSWI were used to calculate monthly time series images (Section 2.2.3) for subsequent classification.

2.4.2. Expansion and Spatiotemporal Migration of Sample Points

Considering the time-, manpower-, and region-based limitations of field sampling, we expanded the number of samples in the spatial domain in the same year. Sample set was enlarged based on preliminary classification result by obtaining a preliminary map of planting patterns from 2019 to 2020 based on field sample points and RF model, as reference; then, based on this preliminarily map, we adopted the “Create Random Points” and “Extract Multi Values To Points” tools in the ArcGIS platform to automatically obtain 4000 sample points with different VI values; finally, we compared VI temporal profiles of these samples with the standard and deleted the nonstandard to obtain the final sample set of 2019–2020.
For the spatiotemporal migration of the sample set in other years, we superimposed this expanded 2019–2020 sample data with spatial layers of other years, extracted VI values of corresponding years and compared them with the 2019–2020 sample data. Samples with planting pattern changes were deleted and new samples were added manually to ensure sample quantity and diversity. Through annual analysis using this method, a year-by-year sample set from 2017 to 2021 was obtained (Figure 6). Although some manual operations were required, this method considerably reduced labor and time costs compared to full manual field sampling.

2.4.3. Random Forest Model

RF is an ensemble learning classifier [41] and is a common model used for classification problems. In this study, RF was performed using the RF (v. 4.6–12) R package-based script, published by Millard and Richardson [42]. There were two parameters that needed to be adjusted, that is, the number of trees created by randomly selecting samples from the training samples (ntree parameter) and the number of variables used for tree node splitting (mtry parameter). In the RF classifier, the NDVI, EVI, and LSWI values calculated from multitemporal images were used as the input variables.

2.4.4. Sample Training and Accuracy Verification

In this study, we used RF as the target classifier and input the field samples and their post-expansion and post-migration data into the classifier year-by-year for model training. Each year corresponded to a classifier to realize the planting pattern classification in 2017–2021.
The sample set obtained through Section 2.4.2 contained 4000 samples of different planting patterns every year, which were divided into training and verification points according to the 7:3 ratio for model training and accuracy verification. The accuracy was evaluated using a confusion matrix, with indices including user accuracy (UA, %) [43], producer accuracy (PA, %) [43], overall accuracy (OA, %) [43], and Kappa coefficient [44].

2.5. Comparison with the National Statistical Data

The sown area obtained from the classification results was compared with the national county-level statistical data for verification. The statistics came from the 2017–2021 statistical yearbooks of cities in Hubei province. All 17 cities/counties in the study area were selected for verification, covering all regions of the study area. The sown area of each city/county was estimated based on crop pattern data and compared with the statistical yearbook data of corresponding years.

3. Results

3.1. Accuracy Assessment of Multi-Year Planting Pattern Maps

Verification points were used to calculate confusion matrix and evaluate the accuracy of planting pattern map (Table 1). The OA was 90% and the Kappa coefficient was 0.86 during the study period, while UA and PA were 90% and 88%, respectively, indicating that the classification results were accurate overall.
The accuracy of the planting pattern maps differed among years, with an OA of 84–94% and a Kappa coefficient of 0.80–0.93. All planting pattern maps were accurate, indicating that the spatiotemporal migration strategy of the samples was reliable. The accuracy was highest in 2019–2020 (OA: 94%; Kappa: 0.93), followed by 2020–2021 (OA: 91%; Kappa: 0.84), while a lower accuracy was found in 2017–2018 (OA: 84%; Kappa: 0.80) and 2018–2019 (OA: 88%; Kappa: 0.87). Generally, the accuracy of the 2019–2020 planting pattern map with field samples was considerably higher than that of the other years, and the classification accuracy decreased further from the field sampling year. The main reason for this difference is that the planting pattern was most affected by temperature, climate, and human factors during 2019–2020. There are subtle differences in the VIs between different years [18], and the real ground sample points obtained by field sampling were the most accurate data for classification; however, the sample expansion and spatiotemporal migration strategy proposed in this study met the classification requirements with an accuracy above 88% in four years, which can effectively ensure precise long-term and large-scope planting pattern classification.
Combined with the confusion matrix thermal diagrams of the four years (Figure 7), the misclassification results were similar throughout the four years. With only one NDVI growth cycle, single rice had the best classification results; meanwhile, wheat/rapeseed–cotton and wheat/rapeseed–corn had more misclassifications, mainly because they had similar phenological curves. Although planting times differed, they might be affected by the subjective will of farmers, resulting in the curve shifts over time.

3.2. Comparison of Planting Pattern Maps with National Statistical Data

We compared planting pattern results with city/county statistics for 2017–2021. There was a significant linear relationship between the planting pattern results and the sown area of the national statistical data of cities/counties, and the R2 was 0.85, 0.82, 0.91, and 0.86, respectively. (Figure 8). The mean absolute error (MAE) and root mean square error (RMSE) can be used to compare datasets (Table 2). The experimental results for the planting area were smaller than the statistics, primarily because some minor crops in the statistical yearbook were not included in the classification results, such as the soybean area, including spring and summer soybean sowing in the statistical yearbook; however, in this study, the classification only considered summer sowing soybean. The results showed that the correlation was strongest in 2019–2020 (RESE: 2.875; MAE: 2.068) and weak in 2017–2018. Overall, the statistical results for the four years had a strong correlation and conformed to the statistical data results.

3.3. Maps of Planting Patterns from 2017 to 2021

We mapped planting patterns from 2017 to 2021 at a 10 m spatial resolution. Figure 9 shows an enlarged view of the study area and region. Spatiotemporal characteristics of the planting patterns varied annually and were affected by climate, policy revisions, and human factors in different years.
Planting patterns over the four years showed similar spatial regularity in the regional distribution structure. Single rice was the dominant crop in the Jianghan Plain, which was widely distributed in various areas of the Plain. Wheat is intensively planted in the western and northern parts. The northern wheat area was dominated by the planting pattern of wheat–soybean, wheat–rice, and wheat–cotton, whereas wheat–corn was extensively distributed in the western hilly area. Rapeseed was primarily distributed in the middle of the Jianghan Plain, mostly in rapeseed–cotton and rapeseed–rice rotations, and cultivation was much denser in areas close to the Yangtze and Hanjiang Rivers. For corn, most wheat–corn patterns were found in the hilly western plains, while rapeseed–corn was not intensively planted on a large scale. Lastly, cotton was primarily planted near the Yangtze River Basin in the central and western parts of the plain, dominated by a rapeseed–cotton planting pattern.
Several planting patterns changed temporally during 2017–2021. The rapeseed–cotton area was dramatically reduced, especially in the Yangtze River Basin in the center of the Plain (area (a) in Figure 8), and most was replaced with rapeseed–rice and single-cropping rice. One major reason for this was that in recent years, affected by planting efficiency and agricultural structure adjustments, farms have planted less cotton, and national cotton planting has been concentrated in the Xinjiang cotton area, which is an advantageous place. Simultaneously, field investigations prove that many laborers have shifted to nonfarm employment [45,46], leaving fewer young, rural laborers. The remaining farmers prefer to plant crops requiring less labor, such as single rice; during the four years, the wheat and soybean areas have grown steadily. In the wheat part of area B, a stable planting pattern characterized by wheat–soybean and wheat–cotton was formed. In area C, the rapeseed–rice area decreased each year, but the wheat–corn area prevailed. This increase in wheat and soybean planting areas has benefited from the release of relevant policies, such as the national encouragement to tap the potential of farmland in Jianghan Plain, moderate expansion of dry-land wheat planting area, and the 2019 soybean revitalization plan. Area D is a typical representative of the southeastern plain, and it has been cultivated by single rice for many years, and small portions of rapeseed–cotton and rapeseed–rice areas were gradually converted to single rice, because the region has been vigorously developing crayfish and single rice specialty industries in recent years. In the same field, single rice is planted in summer and crayfish are raised in winter, which is popular and allows farmers to increase production and income, becoming a new benchmark for modern agriculture in the Jianghan Plain. Notably, the rapeseed planting area in the study area experienced a significant decline in 2018–2019 but increased steadily afterward. Relevant data indicate that the rapeseed planting area fell dramatically in the study area as farmland soil was too wet to be plowed due to continuous rain and low temperatures at the end of 2018. Subsequently, the 2019 Central Document No. 1 was released during this year, which supported rapeseed production in the Yangtze River Basin and boosted mechanization of the farming process, which allowed for steady recovery.

3.4. Change Modes of Planting Patterns

An overview of planting pattern maps in Section 3.3 shows the modern change frequency and planting patterns in the Jianghan Plain. A change frequency of 0 means planting patterns have not changed in four years, while a change frequency of 1 represents only one change, and the following change frequencies show a similar pattern. As shown in Figure 10, planting patterns changed frequently in cultivated lands near the water area in the central-western and central-northern areas of the study area. However, no change was found in the center, northeast, or west with massive, cultivated land remaining unchanged throughout the years. Table 3 lists the change modes of planting patterns that can be used to guide the adjustment of planting policies and to maintain green agricultural ecology.

4. Discussion

4.1. Composition of Landsat and Sentinel-2 Imagery

Among the existing studies, planting patterns tend to be identified by algorithms based on low- and medium-resolution images, such as MODIS. However, arable land covers a small area in most of China [47]. In particular, the planting system in southern China is characterized by smallholder farms, and most of the farmland is less than 0.04 hectares [48]; therefore, MODIS data accuracy is limited, and it cannot identify planting patterns precisely [18].
The combination of Landsat and Sentinel-2 data overcomes “slits” in time and “holes” in space in precise agricultural monitoring, providing an opportunity to map high-resolution crop planting patterns based on phenological analysis [49,50]. In our study, Landsat-8 data were used to fill vacancies in the Sentinel-2 data, which greatly support long-term, large-scale, and high-precision research on planting patterns. The classification accuracy for four years was above 88%, showing that the fusion of these two types of data has the potential to be applied in regions with complex agricultural landscape heterogeneity.
To further prove the applicability of data fusion in extracting high-precision and complex planting patterns, we attempted to extract planting patterns in the study area using MODIS, Landsat-8, and Sentinel-2 images. For MODIS, the overall classification accuracy for the four years was only 60%, and planting patterns were seriously distributed, making it impossible to achieve accurate planting pattern extractions. After cloud removal in Landsat-8, limited images were available in the study area. During important phenological growth periods, such as June and August, Landsat-8 cannot create a high-quality cloud-free image. Although many images of Sentinel-2 were available, holes were found in them after cloud removal. Meanwhile, Ghaderpour and Vujadinovic [51] have proposed a method named jumps upon spectrum and trend (JUST) that can create spatiotemporal maps when there are missing values without any need for gap-filling, but the application of this method in the extraction of crop types and planting patterns remains to be explored. In summary, the planting pattern map created by integrating Landsat-8 and Sentinel-2 images effectively improved the usability of images owing to the availability of high-quality observations and provided more crop information for planting pattern mapping.

4.2. Spatiotemporal Migration Strategy of Samples

Various methods have been proposed in previous studies to generate training samples to address the lack of such training samples in agricultural mapping, such as transfer learning methods [52,53], crowdsourcing initiatives [54], use of target class spectral and temporal features [55], automated clustering method based on an iterative K-means refinement procedure [56], or dependence on existing inventory to guide the labeling of new training samples [57]. It is important to automatically generate training samples by examining spectral and temporal features of the class of interest.
More studies recommend training supervised classifiers using existing samples [57,58]. Furthermore, different agricultural practices and weather conditions lead to considerable differences in crop phenology that hinder the reusability of labels from one geographic region to another and from one year to another [59].
In Section 2.4.3, we developed a strategy that semi-automatically generates numerous samples and spatiotemporally transfers them in accordance with field samples and a preliminary classification map. This strategy helped to precisely extract planting patterns in the study area over the past four years. Figure 7 shows that the training samples obtained by this method were widely distributed and increased considerably. In practice, samples were quickly and accurately generated. Nevertheless, this method has limitations; for example, it is difficult to apply it to areas without field sampling data or study areas with few field samples but that are too large. In summary, more work should be conducted to further study crop mapping methods based on supervised classification and training samples.

5. Conclusions

In this study, we developed and implemented semi-automatic expansion and spatiotemporal migration strategies of sample points, defined VIs temporal profiles of planting patterns within a specified time period, and identified four years of planting patterns using a RF model from integrated Landsat-8 and Sentinel-2 time series data. The VIs was obtained from a combination of Landsat-8 and Sentinel-2 imagery and was then smoothed with an adaptive Savitzky–Golay filter. We plotted planting pattern maps of the study period with a 10-m spatial resolution of the Jianghan Plain, including six crops and nine planting patterns, the resulting maps had OA of 84–94%, with Kappa coefficients of 0.80–0.93. The average OA of the maps was 90%, and the Kappa coefficient was 0.86 over all four years. Compared to planting pattern maps with medium and low spatial resolution, such as those from MODIS data, the spatial distribution of 10-m planting pattern was more detailed, and the accuracy was greatly improved. The distribution of spatiotemporal changes was driven by multiple factors, such as subjectivity and social economics. This study indicates that the proposed approach is promising for mapping large-scale and complex planting pattern maps and can be readily applied to other regions. The resulting spatial datasets can be used for related research and can serve as a scientific basis for the formulation of related policies.

Author Contributions

L.H. conceived the review and designed the research; Z.Z. performed the data calculation and cartography; Z.Z. drafted the manuscript; L.H. and J.L. revised and edited the paper; L.H. and Q.W. participated in the discussion and provided useful comments; L.H., J.W. and J.L. participated in field sampling. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the National Natural Science Foundation of China (41601280).

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

Not applicable.

Conflicts of Interest

The authors declare no conflict of interest.

Abbreviations

The following abbreviations are used in this article:
CDLvegetation index
EVInormalized difference vegetation index
EVIenhanced vegetation index
GEEratio vegetation index
GFSAD30cropland data layer
LSWIU.S. Department of Agriculture
MAENational Agricultural Statistics Service
MSIglobal food security support analysis data
NASSNational Land Cover Dataset
NDVIU.S. Geological Survey
NDVIrandom forest
NLCDGoogle Earth Engine
OAMultiSpectral Instrument
OLIOperational Land Imager
PAtop-of-atmosphere
RFland surface water index
RMSEunmanned aerial vehicle
RVInormalized difference vegetation index
TOAenhanced vegetation index
UAuser accuracy
UAVproducer accuracy
USDAoverall accuracy
USGSmean absolute error
VIroot mean square error

References

  1. Iizumi, T.; Ramankutty, N. How Do Weather and Climate Influence Cropping Area and Intensity? Glob. Food Secur. 2015, 4, 46–50. [Google Scholar] [CrossRef] [Green Version]
  2. USA World Agricultural Outlook Board. Major World Crop Areas and Climatic Profiles. In Agriculture Handbook; World Agricultural Outlook Board: Washington, DC, USA, 1994; p. 299. [Google Scholar]
  3. Wu, W.B.; Yu, Q.Y.; You, L.Z.; Chen, K.; Tang, H.J.; Liu, J.G. Global Cropping Intensity Gaps: Increasing Food Production without Cropland Expansion. Land Use Policy 2018, 76, 515–525. [Google Scholar] [CrossRef]
  4. Wu, W.B.; Yu, Q.Y.; Peter, V.H.; You, L.Z.; Peng, Y.; Tang, H.J. How Could Agricultural Land Systems Contribute to Raise Food Production under Global Change? J. Integr. Agric. 2014, 13, 1432–1442. [Google Scholar] [CrossRef]
  5. Zhang, J.; Feng, L.; Yao, F. Improved Maize Cultivated Area Estimation over a Large Scale Combining Modis–Evi Time Series Data and Crop Phenological Information. ISPRS J. Photogramm. 2014, 94, 102–113. [Google Scholar] [CrossRef]
  6. Alvarado, R.; Toledo, E. Environmental Degradation and Economic Growth: Evidence for a Developing Country. Environ. Dev. Sustain. 2017, 19, 1205–1218. [Google Scholar] [CrossRef]
  7. Li, A.; Wu, J.; Zhang, X.; Xue, J.; Liu, Z.; Han, X.; Huang, J. China’s New Rural “Separating Three Property Rights” Land Reform Results in Grassland Degradation: Evidence from Inner Mongolia. Land Use Policy 2018, 71, 170–182. [Google Scholar] [CrossRef]
  8. Liu, L.; Xu, X.; Zhuang, D.; Chen, X.; Li, S. Changes in the Potential Multiple Cropping System in Response to Climate Change in China from 1960–2010. PLoS ONE 2013, 8, e80990. [Google Scholar]
  9. Belcher, K.; Boehm, M.; Fulton, M. Agroecosystem Sustainability: A System Simulation Model Approach. Agric. Syst. 2004, 79, 225–241. [Google Scholar] [CrossRef]
  10. Dietrich, J.P.; Schmitz, C.; Müller, C.; Fader, M.; Lotze-Campen, H.; Popp, A. Measuring Agricultural Land-Use Intensity—A Global Analysis Using a Model-Assisted Approach. Ecol. Model. 2012, 232, 109–118. [Google Scholar] [CrossRef]
  11. Liu, H. Extraction of Crop Planting Structure in Hetao Irrigation Area Based on Sentinel-2 Images. Resour. Environ. Arid Areas 2021, 35, 88–95. [Google Scholar]
  12. Ashourloo, D.; Shahrabi, H.S.; Azadbakht, M.; Rad, A.M.; Aghighi, H.; Radiom, S. A Novel Method for Automatic Potato Mapping Using Time Series of Sentinel-2 Images. Comput. Electron. Agric. 2020, 175, 1–11. [Google Scholar] [CrossRef]
  13. Rouse, J.W.; Hass, R.H., Jr.; Schell, J.A.; Deering, D.W. Monitoring vegetation systems in the great plains with ERTS. In Proceedings of the Third Earth Resources Technology Satellite-1 Symposium 1 (A), Washington, DC, USA, 10–14 December 1973; Texas A&M University: College Station, TX, USA, 1974; pp. 309–317. [Google Scholar]
  14. Liu, H.Q.; Huete, A. A feedback based modification of the NDVI to minimize canopy background and atmospheric noise. IEEE Trans. Geosci. Remote Sens. 1995, 33, 457–465. [Google Scholar] [CrossRef]
  15. Chen, X.Y.; Meng, J.H.; Du, X.; Zhang, F.F.; Zhang, M.; Wu, B.F. Research on the Remote Sensing Monitoring Model of Winter Wheat Leaf Area Index Based on Ccd Data from Environment Star. Remote Sens. Land Res. 2010, 55, 62. [Google Scholar]
  16. Chen, Y.; Lu, D.; Moran, E.; Batistella, M.; Dutra, L.V.; Sanches, I.D.A.; da Silva, R.F.B.; Huang, J.; Luiz, A.J.B.; de Oliveira, M.A.F. Mapping Croplands, Cropping Patterns, and Crop Types Using Modis Time-Series Data. Int. J. Appl. Earth Obs. Geoinf. 2018, 69, 133–147. [Google Scholar] [CrossRef]
  17. Huang, Q.; Tang, H.J.; Zhou, Q.B.; Wu, W.B.; Wang, L.M.; Zhang, L. Remote Sensing Extraction and Growth Monitoring of Main Crop Planting Structures in Northeast China. Chin. J. Agric. Eng. 2010, 218, 386. [Google Scholar]
  18. Liu, L.; Xiao, X.; Qin, Y.; Wang, J.; Xu, X.; Hu, Y.; Qiao, Z. Mapping Cropping Intensity in China Using Time Series Landsat and Sentinel-2 Images and Google Earth Engine. Remote Sens. Environ. 2020, 239, 111624. [Google Scholar] [CrossRef]
  19. Ding, M.; Guan, Q.; Li, L.; Zhang, H.; Liu, C.; Zhang, L. Phenology-Based Rice Paddy Mapping Using Multi-Source Satellite Imagery and a Fusion Algorithm Applied to the Poyang Lake Plain, Southern China. Remote Sens. 2020, 12, 1022. [Google Scholar] [CrossRef] [Green Version]
  20. Johnson, D.M.; Mueller, R. The 2009 Cropland Data Layer. Photogramm. Eng. Remote Sens. 2010, 76, 1201–1205. [Google Scholar]
  21. Homer, C.; Dewitz, J.; Fry, J.; Coan, M.; Hossain, N.; Larson, C.; Herold, N.; McKerrow, A.; VanDriel, J.N.; Wickham, J. Completion of the 2001 National Land Cover Database for the Counterminous United States. Photogramm. Eng. Remote Sens. 2007, 73, 337. [Google Scholar]
  22. Homer, C.; Dewitz, J.; Yang, L.; Jin, S.; Danielson, P.; Xian, G.; Coulston, J.; Herold, N.; Wickham, J.; Megown, K. Completion of the 2011 National Land Cover Database for the Conterminous United States–Representing a Decade of Land Cover Change Information. Photogramm. Eng. Remote Sens. 2015, 81, 345–354. [Google Scholar]
  23. 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. 2018, 144, 325–340. [Google Scholar] [CrossRef]
  24. 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]
  25. Li, J.; Roy, D.P. A Global Analysis of Sentinel-2a, Sentinel-2b and Landsat-8 Data Revisit Intervals and Implications for Terrestrial Monitoring. Remote Sens. 2017, 9, 902. [Google Scholar] [CrossRef] [Green Version]
  26. Nguyen, M.D.; Baez-Villanueva, O.M.; Bui, D.D.; Nguyen, P.T.; Ribbe, L. Harmonization of Landsat and Sentinel 2 for Crop Monitoring in Drought Prone Areas: Case Studies of Ninh Thuan (Vietnam) and Bekaa (Lebanon). Remote Sens. 2020, 12, 281. [Google Scholar] [CrossRef] [Green Version]
  27. Fieuzal, R.; Bustillo, V.; Collado, D.; Dedieu, G. Combined Use of Multi-Temporal Landsat-8 and Sentinel-2 Images for Wheat Yield Estimates at the Intra-Plot Spatial Scale. Agronomy 2020, 10, 327. [Google Scholar]
  28. Han, J.; Zhang, Z.; Cao, J. Developing a New Method to Identify Flowering Dynamics of Rapeseed Using Landsat 8 and Sentinel-1/2. Remote Sens. 2020, 13, 105. [Google Scholar] [CrossRef]
  29. He, Y.; Wang, C.; Chen, F.; Jia, H.; Liang, D.; Yang, A. Feature Comparison and Optimization for 30-M Winter Wheat Mapping Based on Landsat-8 and Sentinel-2 Data Using Random Forest Algorithm. Remote Sens. 2019, 11, 535. [Google Scholar] [CrossRef] [Green Version]
  30. Xu, F.; Li, Z.; Zhang, S.; Huang, N.; Quan, Z.; Zhang, W.; Liu, X.; Jiang, X.; Pan, J.; Prishchepov, A.V. Mapping Winter Wheat with Combinations of Temporally Aggregated Sentinel-2 and Landsat-8 Data in Shandong Province, China. Remote Sens. 2020, 12, 2065. [Google Scholar] [CrossRef]
  31. Wardlow, B.D.; Egbert, S.L.; Kastens, J.H. Analysis of Time-Series Modis 250 M Vegetation Index Data for Crop Classification in the Us Central Great Plains. Remote Sens. Environ. 2007, 108, 290–310. [Google Scholar] [CrossRef] [Green Version]
  32. Yang, N.; Liu, D.; Feng, Q.; Xiong, Q.; Zhang, L.; Ren, T.; Zhao, Y.; Zhu, D.; Huang, J. Large-Scale Crop Mapping Based on Machine Learning and Parallel Computation with Grids. Remote Sens. 2019, 11, 1500. [Google Scholar]
  33. Li, G.F. Research on Small-Sample Enhancement Methods for High-Resolution Remote Sensing Image Classification; Xi’an University of Technology: Xi’an, China, 2021. [Google Scholar]
  34. Belgiu, M.; Csillik, O. Sentinel-2 Cropland Mapping Using Pixel-Based and Object-Based Time-Weighted Dynamic Time Warping Analysis. Remote Sens. Environ. 2018, 204, 509–523. [Google Scholar] [CrossRef]
  35. Hixson, P.K. Language Stimulation Group: Habilitation Model. Ann. Otol. Rhinol. Laryngol. 1980, 89, 175–178. [Google Scholar] [CrossRef]
  36. Fan, D.D.; Li, Q.Z.; Wang, H.Y.; Zhang, Y.; Du, X.; Shen, Y. Improving the Accuracy of Remote Sensing Identification of Small Crops by Sampling Training Samples. Acta Remote Sens. 2019, 23, 730–742. [Google Scholar]
  37. Zeng, L.; Wardlow, B.D.; Xiang, D.; Hu, S.; Li, D. A Review of Vegetation Phenological Metrics Extraction Using Time-Series, Multispectral Satellite Data. Remote Sens. Environ. 2020, 237, 111511. [Google Scholar] [CrossRef]
  38. Zhang, H.K.; Roy, D.P.; Yan, L.; Li, Z.; Huang, H.; Vermote, E.; Skakun, S.; Roger, J.-C. Characterization of Sentinel-2a and Landsat-8 Top of Atmosphere, Surface, and Nadir Brdf Adjusted Reflectance and Ndvi Differences. Remote Sens. Environ. 2018, 215, 482–494. [Google Scholar] [CrossRef]
  39. Savitzky, A.; Golay, M.J. Smoothing and differentiation of data by simplified least squares procedures. Anal. Chem. 1964, 36, 1627–1639. [Google Scholar]
  40. Becker, W.R.; Johann, J.A.; Richetti, J.; Silva, L.C. Data Mining Techniques for Separation of Summer Crop Based on Satellite Images1. Eng. Agrícola 2017, 37, 750–759. [Google Scholar] [CrossRef] [Green Version]
  41. Breiman, L. Random Forests. Mach. Learn. 2001, 45, 5–32. [Google Scholar] [CrossRef] [Green Version]
  42. Millard, K.; Richardson, M. On the Importance of Training Data Sample Selection in Random Forest Image Classification: A Case Study in Peatland Ecosystem Mapping. Remote Sens. 2015, 7, 8489–8515. [Google Scholar] [CrossRef] [Green Version]
  43. Congalton, R.G.; Green, K. Assessing the Accuracy of Remotely Sensed Data: Principles and Practices; CRC Press: Boca Raton, FL, USA, 2019. [Google Scholar]
  44. Cohen, J. A Coefficient of Agreement for Nominal Scales. Educ. Psychol. Meas. 1960, 20, 37–46. [Google Scholar] [CrossRef]
  45. Liu, Y.; Li, J.; Yang, Y. Strategic Adjustment of Land Use Policy under the Economic Transformation. Land Use Policy 2018, 74, 5–14. [Google Scholar] [CrossRef]
  46. Cheng, M.; Liu, Y.; Zhou, Y. Measuring the Symbiotic Development of Rural Housing and Industry: A Case Study of Fuping County in the Taihang Mountains in China. Land Use Policy 2019, 82, 307–316. [Google Scholar]
  47. Fritz, S.; See, L.; McCallum, I.; You, L.; Bun, A.; Moltchanova, E.; Duerauer, M.; Albrecht, F.; Schill, C.; Perger, C.; et al. Mapping Global Cropland and Field Size. Glob. Chang. Biol. 2015, 21, 1980–1992. [Google Scholar] [CrossRef]
  48. Tan, M.; Robinson, G.M.; Li, X.; Xin, L. Spatial and Temporal Variability of Farm Size in China in Context of Rapid Urbanization. Chin. Geogr. Sci. 2013, 23, 607–619. [Google Scholar]
  49. 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]
  50. Wang, S.; Azzari, G.; Lobell, D.B. Crop Type Mapping without Field-Level Labels: Random Forest Transfer and Unsupervised Clustering Techniques. Remote Sens. Environ. 2019, 222, 303–317. [Google Scholar]
  51. Ghaderpour, E.; Vujadinovic, T. Change Detection within Remotely Sensed Satellite Image Time Series Via Spectral Analysis. Remote Sens. 2020, 12, 4001. [Google Scholar]
  52. Tuia, D.; Ratle, F.; Pacifici, F.; Kanevski, M.F.; Emery, W.J. Active Learning Methods for Remote Sensing Image Classification. IEEE Trans. Geosci. Remote Sens. 2009, 47, 2218–2232. [Google Scholar] [CrossRef]
  53. Tuia, D.; Volpi, M.; Copa, L.; Kanevski, M.; Munoz-Mari, J. A Survey of Active Learning Algorithms for Supervised Remote Sensing Image Classification. IEEE J. Sel. Top. Signal Process. 2011, 5, 606–617. [Google Scholar]
  54. Fritz, S.; McCallum, I.; Schill, C.; Perger, C.; Grillmayer, R.; Achard, F.; Kraxner, F.; Obersteiner, M. Geo-Wiki. Org: The Use of Crowdsourcing to Improve Global Land Cover. Remote Sens. 2009, 1, 345–354. [Google Scholar] [CrossRef] [Green Version]
  55. Malambo, L.; Heatwole, C.D. Automated Training Sample Definition for Seasonal Burned Area Mapping. ISPRS J. Photogramm. 2020, 160, 107–123. [Google Scholar] [CrossRef]
  56. Taheri Dehkordi, A.; Valadan Zoej, M.J.; Ghasemi, H.; Ghaderpour, E.; Hassan, Q.K. A New Clustering Method to Generate Training Samples for Supervised Monitoring of Long-Term Water Surface Dynamics Using Landsat Data through Google Earth Engine. Sustainability 2022, 14, 8046. [Google Scholar] [CrossRef]
  57. Huang, H.; Wang, J.; Liu, C.; Liang, L.; Li, C.; Gong, P. The Migration of Training Samples towards Dynamic Global Land Cover Mapping. ISPRS J. Photogramm. 2020, 161, 27–36. [Google Scholar] [CrossRef]
  58. Radoux, J.; Lamarche, C.; Van Bogaert, E.; Bontemps, S.; Brockmann, C.; Defourny, P. Automated Training Sample Extraction for Global Land Cover Mapping. Remote Sens. 2014, 6, 3965–3987. [Google Scholar] [CrossRef] [Green Version]
  59. Belgiu, M.; Bijker, W.; Csillik, O.; Stein, A. Phenology-Based Sample Generation for Supervised Crop Type Classification. Int. J. Appl. Earth Obs. Geoinf. 2021, 95, 102264. [Google Scholar] [CrossRef]
Agronomy 12 01773 g001 550
Figure 1. Overview of the study area showing (a) the geographical location of the study area in China, (b) the different types of land use in the study area, and (c) the elevation of the study area.
Figure 1. Overview of the study area showing (a) the geographical location of the study area in China, (b) the different types of land use in the study area, and (c) the elevation of the study area.
Agronomy 12 01773 g001
Agronomy 12 01773 g002 550
Figure 2. Temporal coverage of the Sentinel-2 and Landsat-8 time series data used in this study.
Figure 2. Temporal coverage of the Sentinel-2 and Landsat-8 time series data used in this study.
Agronomy 12 01773 g002
Agronomy 12 01773 g003 550
Figure 3. (a–f): Field images based on UAV observations and (g,h) photos of adjusting the UAV in the field and the field survey questionnaire.
Figure 3. (a–f): Field images based on UAV observations and (g,h) photos of adjusting the UAV in the field and the field survey questionnaire.
Agronomy 12 01773 g003
Agronomy 12 01773 g004 550
Figure 4. Flow chart of planting pattern mapping.
Figure 4. Flow chart of planting pattern mapping.
Agronomy 12 01773 g004
Agronomy 12 01773 g005 550
Figure 5. Temporal profile of VIs (NDVI, EVI, and LSWI) for planting pattern with (a) NDVI of all crop planting patterns, (b) EVI of summer crops in double cropping, and (c) LSWI of summer crops in double cropping.
Figure 5. Temporal profile of VIs (NDVI, EVI, and LSWI) for planting pattern with (a) NDVI of all crop planting patterns, (b) EVI of summer crops in double cropping, and (c) LSWI of summer crops in double cropping.
Agronomy 12 01773 g005
Agronomy 12 01773 g006 550
Figure 6. Multi-year sample points including the (a) field sampling points; (b) 2017–2018 sample set obtained through the sample migration strategy; (c) 2018–2019 sample set obtained through the sample migration strategy; (d) 2019–2020 sample set obtained through the sample expansion strategy; and (e) 2020–2021 sample set obtained through the sample migration strategy.
Figure 6. Multi-year sample points including the (a) field sampling points; (b) 2017–2018 sample set obtained through the sample migration strategy; (c) 2018–2019 sample set obtained through the sample migration strategy; (d) 2019–2020 sample set obtained through the sample expansion strategy; and (e) 2020–2021 sample set obtained through the sample migration strategy.
Agronomy 12 01773 g006
Agronomy 12 01773 g007 550
Figure 7. Confusion matrix thermal diagrams (SR: single rice; W–Cot: wheat–cotton; W–R: wheat–rice; R–Cot: rapeseed-cotton; R–R: rapeseed–rice; R–S: rapeseed–soybean; R–Cor: rapeseed–corn; W–S: wheat–soybean; W–Cor: wheat–corn).
Figure 7. Confusion matrix thermal diagrams (SR: single rice; W–Cot: wheat–cotton; W–R: wheat–rice; R–Cot: rapeseed-cotton; R–R: rapeseed–rice; R–S: rapeseed–soybean; R–Cor: rapeseed–corn; W–S: wheat–soybean; W–Cor: wheat–corn).
Agronomy 12 01773 g007
Agronomy 12 01773 g008 550
Figure 8. Comparison of sown area (thousand ha) obtained from planting pattern maps with area obtained from national statistical data ((a) is the year of 2017–2018, (b) is the year of 2018–2019, (c) is the year of 2019–2020, (d) is the year of 2020–2021; y is the area from cropping map, x is the area from the national statistics, R2 is the determination coefficient, and n is the is the number of cities/counties).
Figure 8. Comparison of sown area (thousand ha) obtained from planting pattern maps with area obtained from national statistical data ((a) is the year of 2017–2018, (b) is the year of 2018–2019, (c) is the year of 2019–2020, (d) is the year of 2020–2021; y is the area from cropping map, x is the area from the national statistics, R2 is the determination coefficient, and n is the is the number of cities/counties).
Agronomy 12 01773 g008
Agronomy 12 01773 g009 550
Figure 9. Spatial distribution of planting pattern in each year. (a–d) are the detail areas showed in the figure of the study area).
Figure 9. Spatial distribution of planting pattern in each year. (a–d) are the detail areas showed in the figure of the study area).
Agronomy 12 01773 g009
Agronomy 12 01773 g010 550
Figure 10. Frequency of change of planting pattern.
Figure 10. Frequency of change of planting pattern.
Agronomy 12 01773 g010
Table
Table 1. Overall accuracy of classification in each year 2017–2021.
Table 1. Overall accuracy of classification in each year 2017–2021.
YearOAKappaUAPA
April 2017–April 201884%0.8087%89%
April 2018–April 201988%0.8789%90%
April 2019–April 202094%0.9395%89%
April 2020–April 202191%0.8490%83%
Total90%0.8690%88%
Table
Table 2. Comparison of sown area (thousand ha) obtained from planting pattern maps and from national statistical data.
Table 2. Comparison of sown area (thousand ha) obtained from planting pattern maps and from national statistical data.
2017–20182018–20192019–20202020–2021Total
MAE2.6192.4172.0682.8052.477
RMSE3.4533.5612.8753.8663.438
Table
Table 3. Frequency and change modes of planting patterns.
Table 3. Frequency and change modes of planting patterns.
FrequencyChange Mode of Planting Patterns
1Rapeseed/Cotton→Rapeseed/Rice
Rapeseed/Cotton→Single rice
Rapeseed/Cotton→Rapeseed/Soybean
Rapeseed/Rice→Single rice
Wheat/Rice→Singlerice
Wheat/Cotton→Rapeseed/Soybean
Wheat/Corn→Wheat/Cotton
Rapeseed/Corn→Single rice
Single rice→Rapeseed/Rice
Rapeseed/Rice→Rapeseed/Soybean
2Wheat/Soybean→Rapeseed/Rice→Rapeseed/Soybean
Rapeseed/Cotton→Rapeseed/Rice→Single rice
Single rice→Rapeseed/Corn→Wheat/Corn
Wheat/Soybean→Rapeseed/Rice→Single rice
Rapeseed/Rice→Wheat/Corn→Single rice
Wheat/Corn→Wheat/Cotton→Rapeseed/Cotton
3Wheat/Cotton→Rapeseed/Rice→Wheat/Corn→Rapeseed/Cotton
Single rice→Wheat/Soybean→Rapeseed/Soybean→Wheat/Soybean
Single rice→Rapeseed/Rice→Wheat/Corn→Single rice
Rapeseed/Cotton→Rapeseed/Rice→Wheat/Cotton→Single rice
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Share and Cite

MDPI and ACS Style

Zhang, Z.; Hua, L.; Wei, Q.; Li, J.; Wang, J. Recognition and Changes Analysis of Complex Planting Patterns Based Time Series Landsat and Sentinel-2 Images in Jianghan Plain, China. Agronomy 2022, 12, 1773. https://doi.org/10.3390/agronomy12081773

AMA Style

Zhang Z, Hua L, Wei Q, Li J, Wang J. Recognition and Changes Analysis of Complex Planting Patterns Based Time Series Landsat and Sentinel-2 Images in Jianghan Plain, China. Agronomy. 2022; 12(8):1773. https://doi.org/10.3390/agronomy12081773

Chicago/Turabian Style

Zhang, Zijing, Li Hua, Qi Wei, Jialin Li, and Jianxun Wang. 2022. "Recognition and Changes Analysis of Complex Planting Patterns Based Time Series Landsat and Sentinel-2 Images in Jianghan Plain, China" Agronomy 12, no. 8: 1773. https://doi.org/10.3390/agronomy12081773

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