Cropland data fusion and correction using spatial analysis techniques and the Google Earth Engine

Figures & data

Figure 1. Location and distribution of 65 countries along the B&R region (The base image in Figure 1 comes from World Topographic Map, Esri. The blue part represents the ocean, and the rest represents the land)

Table 1. Overview of the four land cover products

Product	Sensor	Creator	Time	Classification method	Spatial resolution	Classification scheme	Overall Accuracy(%)
CCI-LC	MERIS FR/RR, AVHRR, SPOT-VGT, PROBA-V	ESA	1992–2015	Unsupervised Classification and Machine learning	300 m	FAO LCCS 22 classes	74.1 (Santoro et al. 2017)
GFSAD 30	Landsat ETM+/OLI	USGS	2015	Machine learning	30 m	3 classes	94.0 (Teluguntla et al. 2018)
MCD12Q1	MODIS	Boston University	2001–2019	Supervised classification and Decision tree	500 m	IGBP 17 classes	71.6 (Friedl et al. 2010)
FROM-GLC	Landsat TM/ETM+	CESSC	2010, 2015, 2017	Supervised classification	30 m	11 classes	63.7 (Gong et al., 2013)

Notes: CESSC, Center for Earth System Science China; ESA, European Space Agency; FAO, Food and Agriculture Organization; IGBP, International Geosphere-Biosphere Programme; USGS, United States Geological Survey; MERIS, Medium Resolution Imaging Spectrometer; FR/RR, Full and Reduced Resolution; AVHRR, Advanced Very High Resolution Radiometer; SPOT, Satellite Pour l’Observation de la Terre; SPOT-VGT, SPOT- Vegetation; PROBA-V, Project for On-Board Autonomy, with the V standing for Vegetation; MODIS, Moderate Resolution Imaging Spectroradiometer; TM, Thematic Mapper; ETM+, Enhanced Thematic Mapper Plus; OLI, Operational Land Imager.

Figure 2. Flow chart of data fusion and correction of the four cropland products (CCI-LC, GFSAD30, MCD12Q1, and FROM-GLC)

Table 2. Comparison of cropland definition according to the classification system of four land-cover products

Product	Classes Description	Rules
CCI-LC	Cropland, rainfed	Cropland
	Cropland, irrigated or post-flooding
	Mosaic cropland (>50%)/natural vegetation (tree, shrub, herbaceous cover) (<50%)
	Mosaic natural vegetation (tree, shrub, herbaceous cover) (>50%) /cropland (<50%)	Non-cropland
GFSAD30	Cropland (Cropland and cropland fallows)	Cropland
MCD12Q1	Croplands (At least 60% of area is cultivated cropland)	Cropland
	Cropland/Natural Vegetation Mosaics (Mosaics of small-scale cultivation 40–60% with natural tree, shrub, or herbaceous vegetation.)	Cropland
FROM-GLC	Cropland (Rice, Greenhouse, Other)	Cropland

Figure 3. Consistency degree of the four cropland products CCI-LC, GFSAD30, MCD12Q1, and FROM-GLC in the B&R region. The values of 1–4, respectively, represent one to four cropland products classified as cropland, and the value of 0 represents four cropland products classified as non-cropland

Figure 4. Map of corrected cropland products in the B&R region

Figure 5. Degree of consistency of CCI-LC, GFSAD30, MCD12Q1, FROM-GLC and corrected cropland product in the B&R region. The values of 1–5 represent one to five cropland products classified as cropland

Figure 6. Dispersion degree and correlation analysis of the cropland proportion of cropland products and statistics. The dotted line is the 1:1 line of x:y, and the solid line is the linear regression line of the cropland products and statistics in 65 countries of the B&R region (RMSE: root mean square error, R²: coefficient of determination)

Figure 7. Spatial distribution of four cropland products and our corrected product. Based on the visual interpretation results of testing samples, CCI-LC and GFSAD 30 significantly overestimated the area of cropland in the plain of Eastern Europe (A) and Yunnan-Guizhou Plateaus (D), while MCD12Q1 and FROM-GLC clearly underestimated in the plain of Western Siberia (B) and Southeastern Asia (C)

Table 3. Accuracy error matrices of four cropland products and our corrected product in inconsistent areas according to visually interpreted testing samples

Cropland products		Producer’s accuracy	User’s accuracy	Overall accuracy
CCI-LC	Cropland	79.86%	41.68%	50.48%
	Non-cropland	32.96%	73.17%
GFSAD 30	Cropland	80.21%	54.26%	67.03%
	Non-cropland	59.43%	83.35%
MCD12Q1	Cropland	50.30%	54.66%	65.63%
	Non-cropland	74.96%	71.54%
FROM-GLC	Cropland	41.65%	52.20%	63.82%
	Non-cropland	77.10%	68.78%
Corrected cropland product	Cropland	86.72%	65.16%	77.54%
	Non-cropland	72.19%	90.06%

Figure 8. Enlarged window comparisons among four cropland products and corrected products, where (a) GFSAD 30 underestimated cropland on the Loess Plateau, particularly in Gansu and Henan provinces, China; (b) CCI-LC overestimated cropland in Southwest China, such as the Tibetan Plateau and Yunnan-Guizhou Plateau; (c) MCD12Q1 underestimated cropland in Southeast Asia; and (d) FROM-GLC had the lowest overall accuracy among all products. Artificial boundaries, caused by discontinuities where original image edges meet, can be observed in many areas such as China

Santoro, M., G. Kirches, J. Wevers, M. Boettcher, C. Brockmann, C. Lamarche, and P. Defourny. 2017. “Land Cover CCI: Product User Guide Version 2.0.” Accessed December 13. http://maps.elie.ucl.ac.be/CCI/viewer/download/ESACCI-LC/

 Google Scholar

Teluguntla, P., P. S. Thenkabail, A. Oliphant, J. Xiong, M. K. Gumma, R. G. Congalton, K. Yadav, and A. Huete. 2018. “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 Journal of Photogrammetry and Remote Sensing 144: 325–340. doi:10.1016/j.isprsjprs.2018.07.017.

 Web of Science ®Google Scholar

Friedl, M. A., D. Sulla-Menashe, B. Tan, A. Schneider, N. Ramankutty, A. Sibley, and X. Huang. 2010. “MODIS Collection 5 Global Land Cover: Algorithm Refinements and Characterization of New Datasets.” Remote Sensing of Environment 114 (1): 168–182. doi:10.1016/j.rse.2009.08.016.

 Web of Science ®Google Scholar

Gong, P., J. Wang, L. Yu, Y. Zhao, Y. Zhao, L. Liang, Z. Niu, et al. 2013. “Finer Resolution Observation and Monitoring of Global Land Cover: First Mapping Results with Landsat TM and ETM+ Data.” International Journal of Remote Sensing 34 (7): 2607–2654. doi:10.1080/01431161.2012.748992.

 Web of Science ®Google Scholar

Data availability statement

Google Earth Engine (https://code.earthengine.google.com/) is a free and open platform. All data, models, or code generated or used during this study are available from the corresponding author upon request.