Data from the ESA-LCCS product reveal that the 2015 global cropland area was 22.27 million km². Cropland area in Asia includes most of this total, about 41.79%, while the second largest contributor is Africa, encompassing 19.06%. The top five ranked countries are China, India, Brazil, Russia, and the USA, collectively encompassing 42.05% of global cropland in 2015 (Fig. 1).

The average net global cropland increase between 1992 and 2015 was 37,000 km²yr^-1, corresponding to an average annual growth rate of 0.17%. A transition point is evident during this period (Fig. 2). In 2004, the pace of cropland area increase began to slow down, and in 2012, the area of cropland started to decline slightly. The average annual cropland area growth rate was 0.32% before 2004, while this rate was about 0.03% between 2004 and 2012. The global area of cropland decreased slightly subsequent to 2012 at a rate of 0.08%, a potential predictor of the decreasing trend a few years later. The process of change was not consistent with the previously published FAO data, which show that the pace of cropland area increase began to speed up after 2010 (FAO, 2017).

Data at the regional level show that different areas are characterized by different trends in cropland change (Fig. 3). Notably, an increasing trend was only maintained in Africathroughout the whole study period, corroborating previously published FAO data (Cao et al., 2016). The data in this study show that average cropland area growth rates in Africa were different before and after the 2004 transition point; these values were 0.49% and 0.14%, respectively. In contrast, the area of cropland in Europe decreased between 2000 and 2004 at a rate of 0.26%, while areas in both South America and Oceania increased and then reached a consistent level. Within these data, the growth rate in South America (1.03%) was much higher than that of Oceania (0.36%) between 1992 and 2004. Asia and North America also showed a similar trend in cropland area change, an initial dramatic rise, followed by a gradual increase and then a final smooth decrease. This period of ‘gradual increase’ was much longer in Asia than North America; since 2004, the area of cropland in North America has decreased.

Fig. 3 also compares changes in cropland area between continents. Between 1992 and 2015, Europe first reached its peak area followed by North and South America, Asia and Africa. This may be indicative of an underlying rule: stepwise reduction in cropland area from developed continents (e.g. Europe) to developing continents (e.g. Africa). The stepwise reduction may be related to the increase in food trade (Dissanayake et al., 2017). Weinzettel et al. (2013) think that rich countries often displace land use to other countries through increasing food import, which may lead to cropland reduction in developed countries.

At the country level, Fig. 4 shows that many European countries first reached their maximum cropland areas before 1998, followed by some large countries, including India, Russia, China, and the USA between 1999 and 2003. The northern part of South America mostly reached their turning points in 2004 or 2005 while those in the southern part were delayed a few years. After 2005, many African countries and Central Asian countries reached their maximum cropland areas. This order implies that the reduction in cropland area is more likely to emerge earlier in high-income countries. Among the 74 countries reaching their maximum cropland areas before 2004, the average per capita GDP in the study period is about \$21,000 while that of the 98 countries after 2004 is only about \$7,000. Low-income countries offer low-price labor force for the global market, attracting high-income countries to displace their cropland abroad (Weinzettel et al., 2013). In these low-income countries, the rapid expansion of cropland is likely to occur in areas where large-scale farming is available, such as the northern parts of Kazakhstan (Swinnen et al., 2017), northeastern China (Xin et al., 2009), and sub-Saharan Africa (Schoneveld, 2014). In these regions, agricultural production heavily depends on blue water, which increases pressure on water resources and the risk for abrupt and irreversible environment change (Alexander et al., 2015; Johanssona et al., 2016).

The data assembled in this study show that the average increase in global cropland area (new cropland) between 1992 and 2004 was 99,000 km²yr^-1. This increase in area was mainly derived from forests (60.2%), shrub (16.7%), grassland (10.1%), and sparse vegetation (9.8%). In contrast, between 2004 and 2015, the average area increase was only a third of that seen during the previous time period, about 30,000 km²yr^-1. During this period, the proportion of forest decreased to 52.3% while the proportion of grass and bare land increased to 17.7% (Fig. 5).

The data clearly show that the largest proportions of cropland reclaimed globally between 1992 and 2004 were distributed on the southeastern edge of the Amazon forest, edge of the Eurasian Steppe, and southern margins of the Sahara Desert (Fig. 6). The area of new cropland reclaimed in Asia comprised 32.3% of the global total and was mostly converted from forest (45.2%), sparse vegetation (26.3%), and grassland (15.2%), mainly including southeastern India, Indonesia, and in other southeast Asian countries. In northwestern China cropland was mainly converted from bare areas, and in northeastern Iran and at the junction between Russia and Kazakhstan it was mainly converted from grass and sparse vegetation.

South America was second in this comparison, accounting for 31.7% of the total cropland increase, while Brazil, the largest country in this region, was also the nation that boasted the largest area of new cropland over the time period. At the same time, both Argentina and Bolivia also reclaimed large areas of cropland. Data show that the majority of reclaimed cropland in South America was converted from forest, as high as 86.6%, while a small amount was converted from shrub, just 12.5%.

Africa was third in this comparison, accounting for 21.9% of the total new cropland area, most of which was distributed in Nigeria, Sudan, and Mali to the south of the Sahara. New cropland in this region was largely converted from grassland and shrub, while in some Southern African countries (e.g., Tanzania, Mozambique, and Malawi) cropland was mostly converted from forests.

Previous studies also reported cropland expansion, which creates massive deforestation, grassland loss (Dissanayake et al., 2017). Chen et al. (2013) believe that cropland was substantially converted from of natural vegetation in Central Asia in the 1990s.

Compared to the first time period evaluated in this study, new cropland was distributed in Africa and Asia between 2004 and 2015 (Fig. 6). The proportions of this land use type in Asia and Africa increased by 11.9% and 8.3%, respectively, while the area in South America decreased sharply by 17.6%. Among the ten countries boasting the largest areas of new cropland, Asia includes the first six, Indonesia, China, Russia, Kazakhstan, Iran, and Malaysia. The rapid cropland expansion observed in Indonesia and Malaysia over the whole period may be related to the production of biofuels, such as oil palm (Tscharntke et al., 2012). However, data show that the loss of cropland was also significant in the two countries, which can be explained by the reduction in permanent crops (excluding oil palm) since 1990s (Wicke et al., 2011). As for cropland conversion in Asia, the proportion of forest as a source of cropland increased by 7.7% while the proportion of sparse vegetation as a source decreased by 11.5%, which indicates that cropland growth along the edge of the Eurasian Steppe slowed between 2004 and 2015. In Africa, new cropland areas were largely distributed in Nigeria, Zaire, Sudan, Algeria, Chad, and Tanzania. The proportion of grass and shrub converted to cropland remained stable during this period, while that of forest fell by 9.4% and the proportion of converted sparse vegetation increased 7.0%. Thus, in combination with the spatial distribution of new cropland, it is clear that the trend in Southern African cropland growth slowed while expansion along the northern edge of the Sahara Desert increased (Fig. 7). The area of new cropland also decreased sharply in comparison with the previous period in South America; data reveal a significant decrease in the proportion of forest converted to cropland, up to 17.8%, while the proportion of converted shrub increased by 15.2%. This implies that the reclamation of cropland at the expense of Amazonian edge forest was alleviated somewhat over this period. Previous studies have shown that tropical forests were the primary sources of new cropland in the 1980s and 1990s, and that these new areas were mainly distributed in South America, Sub-Saharan Africa, and southeast Asia (Gibbs et al., 2010). As this rapid deforestation of tropical area aroused a great degree of global concern, governments adopted a series of policies to protect their forests, including improved monitoring and the creation of a soybean moratorium in the Amazon. These measures may have significantly suppressed cropland expansion and mitigated the large-scale deforestation of tropical areas (Austin et al., 2017).

The data collated in this study show that the average global cropland area loss between 1992 and 2004 was about 30,000 km²yr^-1. These areas were mainly converted to forest (49.4%), settlements (34.3%), and grasslands (5.8%). In contrast, between 2004 and 2015, the average area lost remained approximately the same, about 28,000 km²yr^-1; over this period, the proportion of forest decreased by 4.6% while the proportion of settlements increased by 7.7%.

Globally, about 42.7% of cropland losses between 1992 and 2004 occurred in Asia, while about 21.0% were in Europe, and 17.8% were in South America. Losses of this land use type in Eastern Europe, and the Mediterranean are also obvious, most notably in Ukraine, Germany, France, Poland, Romania, and Britain, which corroborate the results of earlier work that addressed the period between 1990 and 2006 (Terres et al., 2015; Kuemmerle et al., 2016). In contrast, Asian cropland losses were concentrated in east and southeast Asia and Northern India, while in South America, cropland losses were mainly distributed in southern Brazil. Lost cropland was mostly converted to settlements in Europe and North America, accounting for 70.5% and 52.6%, respectively, while it was mostly converted to forests in other continents. In Asia, cropland loss may be related to forest transition. For instance, in Vietnam, the national-scale reforestation began in 1992 and forest cover has grown steadily from about 24.7% in 1992 to about 38.2% in 2005 (Meyfroidt et al., 2009).

Between 2004 and 2015, global cropland losses were highest in Asia, where the proportion was 53.8%, an increase of 11.1% compared to the earlier time period. This cropland shrinkage phenomenon was most notable within this region on the Huang-Huai-Hai Plain, China, Japan, and in southeast Asian countries, such as the Philippines. In contrast, the trend toward cropland losses slowed down in Europe, where the proportion was only 7.5%. Over this time period, the proportion of lost cropland converted to settlement decreased dramatically by 19.2%, in Europe with a corresponding increase in the proportion of forests (16.2%). Interestingly, the opposite situation occurred in Asia; in this region, the proportion of cropland converted to settlements increased substantially (24.5%) while that converted to forests decreased (19.3%). Similarly, a projection thinks that about 60% of global cropland loss caused by settlement expansion will take place in Asia by 2030 (Bren d’Amour et al., 2017).

Facing the challenges of cropland loss and increasing population, some researchers have recently focused on the issue of re-cultivating abandoned cropland in the countries of the former Soviet Union, notably Russia, Ukraine, and Kazakhstan (Meyfroidt et al., 2016). Indeed, one recent calculation (Schierhorn et al., 2014) suggests that the wheat production in European Russia could be increased by 6.0 million tons if the 4.4 million hectares of cropland abandoned within the former Soviet Union since 2000 was re-cultivated. It is also possible that this region will become a hot spot area for cropland increases rather than losses in the future.

There are various cropland datasets available at the country, regional, or continent levels. Here, this paper compared cropland and its changes based on different datasets. The FAO provides long-term land use data at the country, continent, and global levels. At the global level, cropland area in this study was higher than that extracted from FAO dataset for several reasons. First, in most cases, cropland area extracted from remote sensing data is often higher than that from statistical data because the former may cover small parcels of other land use types in cropland, such as small ponds, forests, and paths in farmland. Second, some countries do not update timely cropland data, which also affects the accuracy of cropland area based on the FAO dataset (FAO, 2017). For instance, during the period 2010- 2014, the area of cropland and permanent crops remained constant at 1.22×10⁵ km² in China. However, the change trends of cropland are similar at the global level. And, during the study period 1992-2015, global cropland areas extracted from the two datasets both showed an increasing trend at rates of about 4%.

In addition, this paper compared cropland in China and the United States at the county level, extracted from ESA-LCCS data and land use data with a spatial resolution of 30 m interpreted from Landsat Imagery (Fig. 8). The results showed that cropland from ESA-LCCS was highly consistent with cropland extracted from Landsat Imagery. In China and the United States, the R² values were about 0.78 and 0.94, respectively. The lower value in China may be caused by the more complicated landforms; China has a large proportion of croplands in mountain areas, which increases the difficulty for mapping croplands since cropland parcels are small. The results also show that the cropland areas from the ESA-LCCS dataset for the two countries were both higher than those from land use datasets with higher resolutions. Similarly, Cao et al. (2016) mapped global cropland in 2000 and 2010 based on Landsat and MODIS data, which was slightly lower than that based on the ESA-LCCS dataset.

Finally, the ESA-LCCS dataset has some limitations. First, the quality of the map varies according to the region of interest. Areas including the western part of the Amazonbasin, Chili, southern Argentina, western Congo basin, Gulf of Guinea, eastern Russia, eastern coast of China, and Indonesia are affected by lower MERIS FR coverage. Second, data between 1992 and 1999 were less reliable due to relatively lower quality AVHRR data.