Remote Sensing Technology Based on an Algorithm for Cotton Spatial Distribution in Aksu-Alaer Region (2020)

Zhang, P.^1,2  Fan, J. L.^1*  Ding, G.³  Li, S. Y.¹

1. National Engineering Technology Research Center for Desert-Oasis Ecological Construction, Xinjiang Institute of Ecology and Geography, Chinese Academy of Sciences, Urumqi 830054, China;

2. University of Chinese Academy of Sciences, Beijing 100049, China;

3. Division of Risk Monitoring and Comprehensive Disaster Reduction, Department of Emergency Management, Xinjiang Uygur Autonomous Region, Urumqi 830011, China

Abstract: Aksu and Alaer region means the Aksu prefecture and Alaer city in the central region of Xinjiang, China. Based on images from Landsat 8, Sentinel-2, and MOD13Q1 acquired in 2020 and the Google Earth Engine (GEE) platform, the authors extracted the cotton planting area (cotton fields) using the random forest method and classification post-processing in the Aksu Prefecture and Alaer city (excluding Wushi and Baicheng due to extremely low cotton planting). The overall classification accuracy of the images in each county was above 0.9, with Kappa coefficients all exceeding 0.8. The dataset includes: (1) distribution of cotton fields with spatial resolution of 250 m; (2) sample point data. This dataset is archived in .tif and .shp formats, and consists of 17 data files with data size of 385 KB (compressed into 1 file with 134 KB).

Keywords: Aksu prefecture; Alaer city; cotton; random forest

DOI: https://doi.org/10.3974/geodp.2023.04.05

CSTR: https://cstr.escience.org.cn/CSTR:20146.14.2023.04.05

Dataset Availability Statement:

The dataset supporting this paper was published and is accessible through the Digital Journal of Global Change Data Repository at: https://doi.org/10.3974/geodb.2024.02.10.V1 or https://cstr.escience.org.cn/CSTR:20146.11.2024.02.10.V1.

1 Introduction

Cotton, the second largest crop after grain in China’s agricultural landscape, is becoming increasingly prominent. In China, there are three major cotton-growing regions: the Yangtze River basin, the Yellow River basin, and the northwest inland region, with Xinjiang as the main hub^[1]. Xinjiang cotton, a crucial link in the global cotton industry chain, represents a pillar industry in the domestic economy and international market. Among the key cotton-producing areas in Xinjiang, the Aksu-Alaer region is of paramount importance; according to statistics from 2020, the cotton cultivation area in the Aksu-Alaer region reached 664 thousand hectares, accounting for 26.52% of the total cotton cultivation area in Xinjiang^[2,3]. Therefore, gaining a comprehensive understanding of the spatial distribution pattern of cotton in the Aksu-Alaer region is crucial for the effective planning of planting spaces for cotton in this area. However, extracting spatial distribution patterns of cotton fields using remote sensing imagery often requires large amounts of ground survey data as training samples and field surveys consume considerable manpower and resources; thus, obtaining sufficient training samples over large areas remains challenging^[4]. However, the Google Earth Engine (GEE) has emerged as an effective approach to address this issue. The GEE is a cloud-based platform for planetary-scale geographic spatial analysis that leverages Google’s immense computational power to address various high-impact societal issues, including deforestation, droughts, disasters, diseases, food security, water management, climate monitoring, and environmental protection. Many machine learning algorithms can be employed for remote sensing image classification, such as artificial neural networks^[5], decision trees^[6,7], and support vector machines^[8]. Of these, the Random Forest (RF) approach, which has the advantages of high classification accuracy, ability to handle large numbers of input variables, and capability to balance errors, has been widely used for land-cover classification^[9,10]. Remote-sensing cloud-computing technology and RF methods have been extensively applied to extract cotton information in Xinjiang. For instance, Zhou^[11] utilised the PIE Engine Studio and GEE platforms to extract the spatial distribution of cotton in Shihezi, Xinjiang, using the RF method based on NDVI and EVI data as feature indices. Similarly, Lv^[12] utilised the PIE platform and Sentinel-2 data from the GEE platform to extract the spatial distribution of cotton in Alaer city in 2020. Finally, Wang^[13], using Sentinel-2 data from the GEE platform, applied random forest, support vector machine, and decision tree methods to extract cotton information in the Mosuowan reclamation area. However, further research is needed on the spatial distribution of cotton in the Aksu-Alaer region.

Therefore, this study, using the GEE platform and data from Landsat 8, Sentinel-2, and MOD13Q1, used the random forest classification method to construct a dataset of cotton spatial distribution in the Aksu-Alaer region of Xinjiang in 2020, with the aim of providing a reference for planning the spatial distribution pattern of cotton planting in the Aksu-Alaer region.

2 Metadata of the Dataset

The metadata of the Cotton field dataset based on multi-satellite images in the Aksu and Alaer region (2020)^[14] are summarised in Table 1.

3 Methods

3.1 Data Sources

This study uses the Google Earth Engine (GEE) platform and high-resolution remote sensing imagery, including 30-m spatial resolution Landsat8 data and 10-m spatial resolution Sentinel-2 data, for visual interpretation of the cotton area in the Aksu-Alaer region. Enhanced Vegetation Index (EVI) data were sourced from the MOD13Q1 dataset available on the GEE platform. The basic parameters of remote sensing data are shown in Table 2. Land use data were obtained from the Chinese 30-m annual land cover dataset (CLCD)^[16], and spatial distribution data of the digital elevation models were sourced from Shuttle Radar Topography Mission (SRTM) data collected by the United States Space Shuttle Endeavour.

Table 1  Metadata summary of the Cotton field dataset based on multi-satellite images in the Aksu and Alaer region (2020)

Table 2  Data sources and basic parameters

3.2 Research Methodology

Data used in this study were obtained from the GEE platform. First, vector maps of the cultivated land boundaries for each county in the Aksu-Alaer region were utilised with the geographic coordinate system set to GCS_WGS_1984. Then, high spatial resolution remote sensing imagery from Sentinel-2 and Landsat 8 during the 2020 cotton-growing season was visually interpreted to obtain sample points of cotton and non-cotton areas in each region. The acquired sample point data for each county were archived in .shp files. This study employed RF as the classifier owing to its efficiency in handling large training samples and high-dimensional data, along with its strong fault-tolerance capability^[17]. The RF model comprised multiple classification trees^[18]. During the training of the RF model, two-thirds of the total training samples were used to construct each decision tree, and the remaining samples were used to validate the classification results of each tree. During classification, each decision tree in the RF produced a classification result, which was then combined using a majority voting method to obtain the final RF classification result. Sentinel-2 NDVI data and MOD13Q1 EVI data were used as feature values for RF classification. An RF classifier was built to classify each county and to determine the cotton planting distribution for each county. Additionally, the connectedPixelCount method was applied to the GEE platform to eliminate the influence of small patches. Finally, the spatial distribution of cotton in the Aksu-Alaer region was obtained. A flowchart of the study process is shown in Figure 1.

Figure 1  Aksu-Alaer region cotton distribution dataset technical workflow diagram

4 Data Results and Validation

4.1 Dataset Characteristics

The dataset comprises two data files: (1) the spatial distribution of cotton in the Aksu-Alaer region (.tif), with a spatial resolution of 250 m and a timeframe of 2020, and (2) sample point data for cotton and non-cotton areas in the Aksu-Alaer region (.shp).

4.2 Data Analysis

Aksu prefecture, located in the Xinjiang Uygur autonomous region of China, consists of two county-level cities and seven counties: Aksu, Kuqa, Wensu, Xayar, Xinhe, Awat, Kalpin, Baicheng, and Wushi. Considering the research scope, Alaer city was also included in the study. However, Wushi and Baicheng counties, where cotton distribution is minimal, were not considered for cotton extraction in this study. This research was based on the GEE platform, utilising visual interpretation to select cotton sample points, as shown in Figure 2. A total of 1,706 cotton sample points and 1,277 non-cotton sample points were selected, amounting to 2,893 total sample points. The Aksu-Alaer region was classified by county using supervised classification, and the cotton distribution was obtained using the random forest method. From the spatial distribution pattern of cotton in the Aksu-Alaer region (Figure 3), cotton fields are concentrated in the central area (mainly in Alaer city), the northern parts of Awat county, the southern parts of Wensu county, the northern parts of Xayar county, and the southern parts of Kuqa city. Furthermore, there were significant differences in the proportion of cotton-grown area to arable land among the counties in the Aksu-Alaer region, as shown in Table 3. Among them, Alaer city had the highest proportion, at 67.33%, followed by Xayar county, with a cotton-to-arable-land ratio of 64.63%. Kuqa city and Awat county also had high proportions of 58.82% and 52.41%, respectively.

Figure 2  The distribution map of sample points of cotton and non-cotton in the Aksu-Alaer region

Figure 3  The spatial distribution map of cotton in the Aksu-Alaer region of Xinjiang

Table 3  Cotton area percentages in the Aksu-Alaer region

4.3 Data Validation

The study also assessed the accuracy of the extraction results for each county by validating the dataset using the overall classification accuracy, Kappa coefficient, producer's accuracy, and user’s accuracy, as detailed in Table 4. The validation results for each county indicated that the overall classification accuracy was above 0.9 and the Kappa coefficient was above 0.8 for all counties. Among them, Wensu county has the highest accuracy, with an overall classification accuracy of 0.99 and a Kappa accuracy of 0.97, while Kalpin county has the lowest accuracy, with an overall classification accuracy of 0.94 and a Kappa accuracy of 0.83.

Table 4  Accuracy validation results by county in the Aksu-Alaer region

5 Discussion and Conclusions

This study constructed a dataset based on in-depth analysis of cotton distribution in the Aksu-Alaer region using the GEE platform. Specifically, by visually interpreting high-resolution remote sensing images from Sentinel-2 and Landsat 8 data, a total of 2,893 cotton and non-cotton sample points were obtained. Supervised classification using the RF method was then used to generate cotton distribution maps for each county in the region. Finally, a spatial distribution dataset of cotton in the Aksu-Alaer region in 2020 was constructed. The results revealed that cotton fields in the Aksu-Alaer region are centrally-concentrated, primarily distributed in the northern parts of Alaer city and Awat county, the southern parts of Wensu county, the northern part of Xayar county, and the southern part of Kuqa city. In recent years, research on cotton in the Aksu-Alaer region has mostly treated Alaer city and the Aksu area as a whole^[19], focusing solely on Alaer city^[12] for cotton extraction, or has carried out detailed cotton extraction for only specific regions^[20]. Therefore, this study extracted cotton by county in the Aksu-Alaer region, and through validation, demonstrated accurate classification for each county, reflected by the overall classification accuracy and Kappa coefficient both exceeding 0.8 in all counties. The research results provide important spatial information for agricultural planning and resource management in the Aksu-Alaer region. In future cotton extraction efforts, improving spatial resolution can achieve greater precision and accuracy.

Author Contributions

Fan, J. L., Ding, G., and Li, S. Y. designed algorithms for the dataset. Zhang, P. contributed to data processing and wrote the paper.

Conflicts of Interest

The authors declare no conflicts of interest.

References

[1]       Yu, S. X., Zhang, L., Feng, W. J. Study on strategy of large scale, mechanization, informationization, intelligence and social services for cotton production [J]. Engineering Science 2016, 18: 137–148.

[2]       National Bureau of Statistics. China Statistical Yearbook [M]. Beijing: China Statistics Press, 2021: 4-5.

[3]       Statistics Bureau of XPCC and NBS Survey Office in XPCC. Xinjiang Production & Construction Corps Statistical Yearbook [M]. Beijing: China Statistics Press, 2021: 5.

[4]       Hao, P. Y., Wang, L., Zhan, Y. L., et al. Using moderate-resolution temporal NDVI profiles for high-resolution crop mapping in years of absent ground reference data: a case study of bole and Manas counties in Xinjiang, China [J]. ISPRS International Journal of Geo-Information, 2016, 5: 23. DOI: 10.3390/ijgi5050067.

[5]       Hassan-Esfahani, L., Torres-Rua, A., Jensen, A., et al. Assessment of surface soil moisture using high-resolution multi-spectral imagery and artificial neural networks [J]. Remote Sensing, 2015, 7: 2627–2646. DOI: 10.3390/rs70302627.

[6]       Berhane, T. M., Lane, C. R., Wu, Q. S., et al. Decision-tree, rule-based, and random forest classification of high-resolution multispectral imagery for wetland mapping and inventory [J]. Remote Sensing, 2018, 10: 26. DOI: 10.3390/rs10040580.

[7]       Hubert-Moy, L., Thibault, J., Fabre, E., et al. Mapping grassland frequency using decadal MODIS 250 m time-series: towards a national inventory of semi-natural grasslands [J]. Remote Sensing, 2019, 11: 21. DOI: 10.3390/rs11243041.

[8]       Xiong, J., Thenkabail, P. S., Tilton, J. C., et al. Nominal 30-m cropland extent map of continental Africa by Integrating pixel-based and object-based algorithms using Sentinel-2 and Landsat-8 data on Google Earth Engine [J]. Remote Sensing. 2017, 9: 27. DOI: 10.3390/rs9101065.

[9]       Rodriguez-Galiano, V. F., Ghimire, B., Rogan, J., et al. An assessment of the effectiveness of a random forest classifier for land-cover classification [J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2012, 67: 93–104. DOI: 10.1016/j.isprsjprs.2011.11.002.

[10]    Rodriguez-Galiano, V. F., Chica-Olmo, M., Abarca-Hernandez, F., et al. Random forest classification of Mediterranean land cover using multi-seasonal imagery and multi-seasonal texture [J]. Remote Sensing of Environment, 2012, 121: 93–107. DOI: 10.1016/j.rse.2011.12.003.

[11]    Zhou, L., Lin, Z. S., Wang, L. H., et al. Dynamic monitoring of cotton planting area under PIE platform [J]. Spacecraft Recovery & Remote Sensing, 2023, 44(3): 108–118.

[12]    Lv, S. L., Zhao, Y., Chen, W. J., et al. Extraction of cotton planting area in Alaer based on remote sensing cloud computing [J]. Cotton Sciences 2022, 44: 19–25.

[13]    Wang, H. H., Zhang, Z., Kang, X. Y., et al. Cotton planting area extraction and yield prediction based on Sentinel-2A [J]. Transactions of the Chinese Society of Agricultural Engineering, 2022, 38(9): 205–214.

[14]    Zhang, P., Fan, J. L., Li, S. Y. Cotton field dataset based on multi-satellite images in Aksu and Alaer region (2020) [J/DB/OL]. Digital Journal of Global Change Data Repository, 2024. https://doi.org/10.3974/geodb.2024.02.10.V1. https://cstr.escience.org.cn/CSTR:20146.11.2024.02.10.V1.

[15]    GCdataPR Editorial Office. GCdataPR data sharing policy [OL]. https://doi.org/10.3974/dp.policy.2014.05 (Updated 2017).

[16]    Yang, J., Huang, X. The 30 m annual land cover dataset and its dynamics in China from 1990 to 2019 [J]. Earth System Science Data 2021, 13: 3907–3925. DOI: 10.5194/essd-13-3907-2021.

[17]    Immitzer, M., Vuolo, F., Atzberger, C. First experience with Sentinel-2 data for crop and tree species classifications in Central Europe [J]. Remote Sensing, 2016, 8: 27. DOI: 10.3390/rs8030166.

[18]    Breiman, L. Random forests [J]. Machine Learning, 2001, 4: 5–32. DOI: 10.1023/a:1010933404324.

[19]    Liu, C. J., Jin, X. B., Xu, W. Y., et al. Analysis of the spatial distribution and variation characteristics of cotton planting in southern Xinjiang from 2000 to 2020 [J]. Transactions of the Chinese Society of Agricultural Engineering, 2021, 37(16): 223–232.

[20]             Zhang, N. N., Zhang, X., Bai, T. C., et al. Field scale cotton land feature recognition based on UAV visible light images in Xinjiang [J]. Transactions of the Chinese Society for Agricultural Machinery, 2023, 54: 199–205.