Geographical Distribution of Crops - Part-1 Introduction, Crop Production | Crop Production Notes- Agricultural Engineering PDF Download

[1] Humans have transformed the surface of the planet through agricultural activities, and today, ∼12% of the land surface is used for cultivation and another 22% is used for pastures and range lands. In this paper, we have synthesized satellite-derived land cover data and agricultural census data to produce global data sets of the distribution of 18 major crops across the world. The resulting data are representative of the early 1990s, have a spatial resolution of 5 min. (∼10 km), and describe the fraction of a grid cell occupied by each of the 18 crops. The global crop data are consistent with our knowledge of agricultural geography, and compares favorably to another existing data set that partially overlaps with our product. We have also analyzed how different crops are grown in combination to form major crop belts throughout the world. Further, we analyzed the patterns of crop diversification across the world. While these data are not sufficiently accurate at local scales, they can be used to analyze crop geography in a regional-to-global context. They can also be used to understand the global patterns of farming systems, in analyses of food security, and within global ecosystem and climate models to understand the environmental consequences of cultivation. INDEX TERMS: 1699 Global Change: General or miscellaneous; 1640 Global Change: Remote sensing; 1630 Global Change: Impact phenomena; 1694 Global Change: Instruments and techniques; KEYWORDS: land use, land cover, croplands, crop distribution, agricultural census, crop harvested area

1. Introduction [2] Humans dominate the landscape in nearly every corner of the planet. Today, croplands occupy nearly 18 million km2 (an area roughly the size of South America), pastures take up another 34 million km2 (an area roughly the size of Africa), and urban areas use roughly 2.5 million km2 (an area roughly the size of a third of Europe) [Klein Goldewijk, 2001; Ramankutty and Foley, 1998; Turner et al., 1993]. Altogether, these three anthropogenic ecosystems currently occupy over a third of the global land surface. [3] Human land use practices have enormous consequences for the environment. While croplands and pastures provide food for the world’s population, agricultural practices have, to a large degree, led to the clearing of many forests [Goudie, 2000]. Human land use practices have also degraded many soils [Meyer and Turner, 1994], influenced global and regional climates [Betts, 1999; Bounoua et al., 2002; Brovkin et al., 1999; Pielke et al., 2002; Zhao and Pitman, 2002], changed the global cycles of carbon, nitrogen, and water [Houghton et al., 1999; Vitousek et al., 1997; Postel

et al., 1996], induced the loss of biodiversity [Dale et al., 2000; Pearce, 2001; McNeely, 1992], affected geomorphic processes, and changed the quality of many natural waterways [Goudie, 2000]. [4] Characterizing the geographic extent and nature of human-dominated ecosystems is vital to understanding the environmental impacts of land use and land cover change.

With the advent of remote sensing, there have been several efforts aimed at depicting regional and global patterns of land cover [Vogelmann et al., 2001; Cihlar, 2000; DeFries and Belward, 2000; Belward et al., 1999; Lobo et al., 1997; Brown et al., 1993; Loveland et al., 1991]. However, these studies have mainly focused on natural ecosystems, with only one or two land cover classes devoted to humandominated ecosystems. On the other hand, ground-based census data (routinely collected across counties, states, and nations) often contain detailed land use information, but lack the spatial detail and resolution of satellite-based data. [5] In order to characterize global agricultural land cover, Ramankutty and Foley [1998] developed a statistical ‘‘data fusion’’ technique to calibrate satellite-derived land cover data [Loveland and Belward, 1997] against agricultural census statistics. Their data represent the cropland land cover of the world on a continuous scale, depicting the percentage of each grid cell in cultivation (Figure 1). However, this data set is still limited in not delineating the distribution of

specific crop types and cropping systems. Detailed information on agricultural land use practices (such as cropping systems) is necessary to understand the environmental consequences of cultivated ecosystems. For example, the distinction between C3- and C4-based physiology crops is important in studies of the global carbon cycle [Lloyd and Farquhar, 1994; Still et al., 2003]. In addition, Donner and Kucharik [2003] discuss the importance of differentiating between maize and soybeans on nitrate export through the Mississippi River, and the spatial pattern of rice paddy is important to estimate the distribution of methane emissions [Cao et al., 1996; Matthews and Fung, 1987]. Moreover, model-based estimates of historical and current crop production [Kucharik, 2003] and future crop production [Leemans and Solomon, 1994; Fischer et al., 2000] have made distinctions between different crops. [ 6 ] Here we use an approach, similar to that of Ramankutty and Foley [1998], wherein we combine national and sub-national agricultural census data, along with land

cover data, to derive the spatial distribution of crop types across the world. Our focus is also on cultivation within permanent croplands, which follows the Food and Agriculture Organization (FAO) definition of arable lands and lands under permanent crops (Table 1). Other land use systems, including pastures and regions of shifting cultivation, are not considered in this study.

2. Methods [7] We derived the global distribution of 18 major crop types by synthesizing agricultural census data on harvested area, and the global cropland data set of Ramankutty and Foley [1998]. The crop categories were selected using distinct biogeochemical, phonological, and food resource characteristics. We arrived at 17 major crop categories (barley, cassava, cotton, groundnuts or peanuts, maize, millet, oil palm fruit, potatoes, rapeseed or canola, rice, rye, sorghum, soybeans, sugar cane, sugar beets, sunflower,

and wheat), one major crop group category (pulses, which includes many beans, peas, lentils, etc.), and 10 other ‘‘minor ’’ crop categories (fibers, fruit, nuts, oil-bearing crops, other cereals, other roots and tubers, others, spices, and vegetables) (Table 2). Agricultural census data were collected and regrouped into those categories and data sets were generated for all eighteen major crops, while all other crops and crop categories were grouped into a single ‘‘other crop’’ category (Table 2). [8] We collected agricultural census data from various census organizations (Table 3). At the national level, agricultural census data are available for every country in the world from the FAO in the FAOSTAT database [FAO, 2002]. This database contains harvested area data for over 100 individual crops, collected and reported annually for the time period 1961 – 2001. In addition, to achieve some level

of uniformity in the size of administrative units, we collected sub-national (one administrative level below the nation: state or province equivalent) census data for a number of countries, based on either their land area or their share of global cropland area. These countries were Argentina, Australia, Brazil, Canada, C hi na, India, Kazakhstan, Mexico, the Russian Federation, Turkey, and the United States of America (Figure 2). [9] We collected and averaged the census data for 6 years, 1990 – 1995. This period centers roughly around 1992, the year for which Ramankutty and Foley [1998] developed their cropland distribution data set. Often the national totals obtained from the sub-national censuses did not match the FAO national data. After checking for consistency in definitions and for errors, we scaled all sub-national data to the FAO national totals. Similarly, scaling was done in cases

where sub-national data were not available for all six years (see Appendix A). [10] The census data for each crop were converted to a relative crop fraction (the proportion of a particular crop’s

harvested area to the total harvested area) (see Table 1 for definition of terms used in this paper). This fraction was assigned to the administrative unit for which it was obtained (Figure 3). It should be noted that the harvested area data for

Figure 3. Flowchart showing the methodology used to construct the major crop data sets. Boxes with rounded corners indicate geospatial data sets, while boxes with sharp corners indicate tabular data at the level of political units. We first collected agricultural census data on harvested area for crops at the national or sub-national level. From these data, for each administrative unit, we then estimated the proportion of each of the 18 major crops to the total harvested area. We then masked non-cropland areas and applied a smoothing algorithm (see text for details). Finally, we multiplied the resulting data of individual crop proportions by the cropland data set (Figure 1) to obtain the per-pixel proportion of each of the 18 major crops.

all crops do not directly sum up to reported cropland area.

This is because of multiple cropping (the practice where the same land is cultivated 2 or even 3 times throughout the same year). Here we assumed that all crops within a region have similar multiple cropping patterns. Thus, by expressing the harvested area for each crop as a proportion of the total, we obtained for every pixel the probability that a particular crop is cultivated. [11] Next we used the data set of the extent of the world’s croplands [Ramankutty and Foley, 1998] to mask out noncropland areas. Then we smoothed the data to eliminate discontinuities at the borders between political units.

Although real crop discontinuities exist between some nations, all of the boundaries we have introduced are artificial and arise solely because of our use of administrative level data. Hence we smoothed these by applying a 41 ∼ 41 pixel Gaussian filter. Since the smoothing may have introduced biases in the data, we rescaled the data equally for all crops to ensure that all fractions indeed added to 1. This smoothing and normalization significantly altered the census data in administrative units where both of the following conditions are met: the unit has a width in any direction smaller than the width of the filter, and the unit has a crop proportion significantly different from that of the surrounding administrative units. [12] Finally, to obtain the spatial distribution of each crop (fraction of grid cell occupied by the crop, what we define as ‘‘crop fraction’’), we multiplied the spatially distributed relative crop fraction from the previous step by the cropland data set of Ramankutty and Foley [1998]. The 18 crop data sets thus produced are gridded, global, and at a 5-min latitude/longitude resolution (∼10 km) (Figures 4a – 4f). [13] These crop data have been created using a very simple procedure (Figure 3), with multiple assumptions, and thus

have several limitations. First, the use of national and state level data, without the ability to define cultivation intensities within administrative units, makes the data appropriate only for regional- to global-scale studies. Secondly, in regions where some crops have multiple growing seasons and others have single growing seasons, the assumption of the same number of growing seasons will lead to overestimation of the extent of the former (e.g., rice and pulses), and underestimation of the extent of the latter (e.g., millet, wheat, and barley). This is the case especially in the tropics where certain crops are planted and harvested multiple times, while others are not. Third, reporting errors in the census data may considerably alter the final results. This is especially true for countries where there may be political or economic incentives to over- or under-report cropland area (e.g., China [see Seto et al., 2000]). Fourth, the effects of the smoothing filter are difficult to quantify. Most frequently the affected areas are along the border between administrative units, across which cultivation intensities vary greatly because of cultural or environmental differences: small states and countries surrounded by larger administrative units, and long and narrow administrative units. Finally, errors in the Ramankutty and Foley [1998] croplands data set would carry over to these new data sets. [14] Despite the simplicity of the approach used to generate these data, they are the first of their kind. By merging the high spatial resolution of satellite-derived data and the high level of attribute differentiation in census data, these data represent a useful product to be used within numerical models (of the Earth’s climate, terrestrial ecosystems, and hydrological systems) and by land managers, planners, and policy makers. [15] One caveat to add is that the data, in their current form, do not identify some important phenological charac-

teristics of crops such as spring crops versus winter crops.

Such distinctions are very important when assessing the environmental consequences of crops. We recommend that climate and ecosystem modelers use climate data in conjunction with our crop data sets to distinguish between the different crop phenologies [e.g., Leemans and Solomon, 1993]. In future updates of this product, we plan to incorporate these distinctions. 

4. Global Cropping Patterns [20] Here we discuss the global distribution of major crop groups (Figures 4a – 4f, 5, and 6). 4.1. Cereals [21] The most prevalent group of crops across the world is cereals (Figures 4a – 4c). Taken together, cereals are the only group of crops with cultivation that exceeds 20% of global land area (Figures 5a – 5c), or 61% of the total cultivated land. In particular, wheat, maize, barley, rice, and millet are dominant over more than two thirds of the cropland of the world (Table 5). The only regions where a cereal is not the dominant crop are the Caribbean and central Africa. In the Caribbean, the dominant major crop is sugar cane; while in central Africa, the dominant major crop is cassava.

However, in both regions, a cereal (maize) is the second most prevalent crop (Tables 6 and 7). [22] A breakdown by latitude also reveals the dominance of cereals (Figures 5a – 5c). Wheat is the most abundant crop, occupying 22% of the total cultivated area in the world. The most intensive wheat cultivation occurs in the temperate latitudes of both hemispheres. Wheat is most prevalent in the Great Plains of the United States, the Canadian Prairie Provinces, the Indus and the upper Ganges Valleys, along the Kazakhstan and Russian border, and in southern Australia (Figure 4c). Wheat is also found throughout Europe, in southern South America, in parts of eastern Africa, and in eastern China.

[23] Maize is the most geographically ubiquitous crop and the crop with the third largest extent in the world. It is cultivated over 13% of the world’s croplands, with the most extensive cultivation occurring from approximately 50∼Nto45∼S (Figure 5c). Maize attains its highest cultivation intensity in the U.S. maize belt, but it is also a major commodity in northeastern China (Manchuria), along the Rift Valley in Africa, and in eastern Europe (Figure 4a). Less intensive cultivation of maize can be found in South America, western Europe, India, and southeastern China. [24] Barley and rye are preferentially grown in colder latitudes, with the majority being cultivated around 55∼N (Figure 5a) in Canada, the northern United States, and European Russia (Figures 4a and 4b). Barley is the crop with the fourth largest area, with 9% of the world’s croplands, while rye is cultivated over 2% of the world’s croplands. [25] Rice, sorghum, and millet dominate the tropical and sub-tropical belts, especially in the Northern Hemisphere (Figure 5b); these three crops occupy 11%, 3%, and 2%, respectively, of the global cultivated area. Rice is the second most extensive crop in the world, and is a major crop of south and southeast Asia. It is also cultivated in the Amazon Basin, the southern United States, and southern Australia (Figure 4b). Sorghum, the only cereal that does not emerge as a dominant crop in any region, is common throughout the Rift Valley and the Sahel region in Africa, the southern half of the Mississippi Valley, and India (Figure 4b). Millet appears in the

same regions of Africa as sorghum and to a lesser degree throughout parts of Asia, and is most abundant in western India (Figure 4a). 4.2. Roots and Tubers [26] Although roots and tubers cover a much smaller area than cereals, they are another important human staple. The two major crops from this category, potatoes and cassava, are both tubers. Together they cover 4% of the world’s total harvested area. Geographically, they are cultivated in contrasting climates (Figure 5d). Potatoes are extensively grown in the colder temperate latitudes between 40∼N and 75∼N, with the highest potato cultivation intensity occurring at 55∼N in the European part of the former Soviet Union and in northeastern Europe (Figure 4c). On the other hand, cassava is grown in the equatorial and tropical regions (Figure 4c). This crop is cultivated from ∼20∼Nto ∼30
S, but its cultivation intensity, even at its highest (5∼S), is only half that of potatoes. Cassava also appears in northern Brazil, south central Africa, Thailand, Micronesia, and Polynesia; cassava is also the major crop cultivated in central Africa. 4.3. Sugar Crops [27] Although sugar beets and sugar cane both produce sugar, the similarities between the two crops end there. They prefer different climates, with sugar cane favoring yearround warmth, while sugar beets favor much cooler conditions (Figure 5e). The two crops also have different plant

Figure 6. Regional distribution of major crops of the world. The world is divided into 24 agriculturally and culturally distinct regions. For each region, in a pie chart, we present the proportions of the top five most common crops (here we include minor crops, ones for which census data were collected, but for which no spatial data sets were generated). For completeness, we include the Islands as a region (defined as any island in the world with an area less than 5000 km2) although they are not part of our spatial maps or the discussion. The information for the Islands arises purely from the census data. See color version of this figure at back of this issue.