Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Design and characterization of spatial units for monitoring global impacts of environmental factors on major crops and food security

Design and characterization of spatial units for monitoring global impacts of environmental... Crop distribution, crop monitoring, This paper describes the design and characterization of a limited number of environmental variability, global agriculture, spatial units. areas to be used as spatial monitoring and reporting units (MRU) by Crop- Watch, the global crop monitoring system (http://www.cropwatch.com.cn/) Correspondence hosted by the Chinese Academy of Sciences. The MRUs are modified and adapted Bingfang Wu, CAS Key Laboratory of Digital from an existing standard vegetation zoning map. They are designed to be areas Earth Science, Institute of Remote Sensing of uniform vulnerability as assessed by three sets of variables: environmental and Digital Earth (RADI), Chinese Academy of (such as climate and NDVI, a satellite-based Normalized Difference Vegetation Sciences, Beijing 100101, China. Tel: Index), seasonality (such as intra- annual climate variations and interannual NDVI +86 10 64855689; Fax: +86 10 64858721; E-mail: wubf@radi.ac.cn variability) and agronomy (such as presence/absence of major crops and their yield). The paper pays due attention to within- zone spatial variability for each Funding Information of the three groups of variables: in general, the variability measured as the aver- In addition to regular program funding, the age of the spatial coefficient of variation inside MRUs, is much larger for ag - Digital Agriculture Division, Remote Sensing ronomic variables (330%) than for environmental variables (38%) and seasonality and Digital Earth Institute (RADI), Chinese (55%). The MRUs provide a rather coherent picture of the variations of global Academy of Sciences received support from agriculture, for instance, the links between the distribution of crops, agricultural the special Fund for Grains-scientific Research in the Public Interest (Grants No. 201313009- production potential, and environmental variability (over space and time). They 2 and No. 201413003-7), the National High closely delineate the distribution of the major nonsugar food crops (barley, Technology Research and Development cassava, maize, potatoes, rice, soybean, and wheat). The discussion focuses on Program of China (863 program), Grant No. the relations between the size of the MRUs and the within- MRU variability 2012AA12A307. The first author is supported (spatial heterogeneity) of cropping, and environmental conditions including by Grant No. 2013T1Z0016 for Visiting seasonality. The conclusion stresses that spatial variability in agriculture is bound Professorships for Senior International to be larger than the variability in the environmental variables used to define Scientists. the units, regardless of the units’ size. The size of the spatial units (and therefore Received: 29 April 2015; Revised: 23 July their number) is not a very critical constraint for operational impact reporting, 2015; Accepted: 3 September 2015 especially if impact indicators focus on agricultural areas inside MRUs. Food and Energy Security 2016; 5(1): 40–55 doi: 10.1002/fes3.73 information collected for impact assessments includes both Introduction impacting factors and impacted system. For global crop monitoring and reporting, some generalization of methods Most definitions of “monitoring” and “crop monitoring” is required because of the gap between the scale of the refer to the regular and standardized collection and analysis observations (whether points or large pixels) and the scale of information over a specific area, often by an external at which impacts are assessed and decisions are taken, and neutral observer, with a view to understanding the usually administrative or other large units (Dalgaard et al. causes of, preventing or limiting damage (Babu and Quinn 2003). 1994; Schmid 1998; OECD 2007; Snodderly 2011). The © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. R. Gommes et al. Spatial Units for Global Environmental Monitoring There exists a number of different global monitoring characterizing a limited number of zones to be used by units, such as in the developed and developing countries, CropWatch for global monitoring and reporting (Wu et al. Southern and Northern Hemisphere, continents, countries 2014), based on an existing standard agricultural zoning map. at risk, etc. The most popular monitoring or reporting The zones are characterized by three sets of variables: ag- unit is the country since it is consistent with statistical ronomic (such as presence/absence of major crops and their data, economic, and policy conditions, even if inhomo- yield), environmental (such as climate) and seasonality (such geneous for impacting factors. Size also varies a lot from as intra- annual climate or NDVI variability). The paper also small countries such as Singapore to the largest one (Russia, pays attention to within-zone spatial variability in order to more than 9000 times larger than Singapore). identify those where impacts may be inhomogeneous. When designing a reporting unit, adequate indicators and monitoring units are interlinked (Wu et al. 2015). Data Ideal indicators are both systemic and normative (Binder et al. 2010). Systemic: they organically relate to the im- Spatial data pacted systems (reporting unit); normative: they can be used to intercompare different spatial units or the same The CropWatch Mapping and Reporting Units (MRUs) are spatial unit at different times. While impacting factors based on several existing global maps, starting with the may vary a lot spatially, the impacted systems or report- FAO Global Ecological Zones map (GEZ), a map designed ing unit definition can be designed in such a way that for reporting forest and forest change statistics in the ambit they are homogeneous, that is, the same impacting factors of the Forest Resources Assessment (FRA; FAO 1999, 2012, will yield a similar impact over most of the impacted downloadable from GeoNetwork, FAO 2014a). The GEZ systems. In other terms, the vulnerability patterns are map is basically a map of natural vegetation types. expected to be uniform over the monitoring units. Next to a recent Köppen climate map (1976–2000 Impacts, which are the object of monitoring, are brought data; Grieser et al. 2006b), the MRUs were also assessed about by a variety of factors, individually or collectively against global land use and ecosystem maps and especially (Zhang et al. 2013). They include economic conditions (cost the suitability for agriculture maps from the ongoing of oil and fertilizer: Chen et al. 2010; Kelly 2010; Akpan Atlas of the Biosphere project, established in 2002 (SAGE, et al. 2012), policies (Lu 2002; Wiggins and Brooks 2010) 2002). It is also in order to mention several global prod- but mostly, directly or indirectly, weather (Gornall et al. ucts primarily developed for climate change impact studies 2010; Reynolds 2010; Iizumi et al. 2014; Leblois et al. 2014) (Ramankutty et al. 2002; Monfreda et al. 2008, 2009), and agronomy, especially mechanization, inputs, improved as well as most of the global zoning schemes described cultivars and management (Loyce et al. 2012; Yu et al. 2012; by van Wart et al. (2013), and others (including an Zhang et al. 2013; George 2014b; Rozbicki et al. 2015). The array of current and potential land use maps available economic and policy factors follow country units in a sta- in digital form as grids or polygons from FAO, 2014a). tistical sense; weather and agronomic factors don’t: they For China, the standard ecological zones of Sun (1994; mostly follow agroecological zones over long time periods. of which an English language version is available in Hu There exists a number of different global classifications of and Zhang 2006) were integrated into the MRU map. agriculture, agricultural land use and agroecological zones. Some of them focus on the environment of specific crops, Impact factors such as the wheat mega- environments developed by the CGIAR institutes (Hodson and White 2007). Others put more weight Impact factors are classified into three categories for char- on functional, social or environmental aspects (management, actering MRUs: agronomic (such as presence/absence of biodiversity), exemplified by the vast literature on agroeco- major crops and their yield), environmental (such as systems (Doré et al. 2011). All agroecological zones work is climate) and seasonality (such as intra- annual climate or based on the implicit or explicit assumption that environmental NDVI variability). Impacting factor values and statistics variables and the human activities (such as vegetation and are referred to whole MRUs (e.g., wheat yield in the agriculture) that depend on them are correlated and behave Pampas), even if they “exist” only in cropped areas. All coherently. The link between them is strong and they con- variables as well as their sources and spatial resolution stitute a “complex”: natural vegetation directly links to dominant (“pixel size”) are succinctly described in Table 1. crops, which in turn often constitute, of have constituted the basis of agro- economic zones (Sombroek and Gommes 1996; Agronomic variables Haberle and Chepstow-Lusty 2000; Gommes et al. 2004). This paper examines the validity of the concept and the Agronomic variables are prefixed with A, for example, feasibility of global monitoring units for identifying and A1: Arland%, average of the percentage of each pixel that © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. 41 Spatial Units for Global Environmental Monitoring R. Gommes et al. Table 1. List of variables used to characterize MRUs. Type: A, agronomic; E: Environmental; S, seasonality. As a rule, the values in the table are spatial averages over all grid points inside the MRUs. Period: the period to which the data refer (“Current” means that the data are continually updated). “Pixel” is the grid size (in km) of the original rasters provided by the respective Sources. Type Name Unit Definition Pixel Period Source A1 Arland% 0–100 Percent of arable land per pixel 23 Current IIASA A2 Irr% 0–100 Percent of pixel area equipped for irrigation 8 1990–2013 GMIA A3 Bar% 0–100 Percent of pixels where barley is cultivated 7 Current JRC A4 Baryld t/ha Average barley yield 57 2008–2012 CW- 1 A5 Casyld t/ha Average cassava yield 57 2008–2012 CW- 1 A6 Mz% 0–100 Percent of pixels where maize is cultivated 3 Current JRC A7 Mzyd t/ha Average maize yield 57 2008–2012 CW- 1 A8 Pot% 0–100 Percent of pixels where potato is cultivated 7 Current JRC A9 Potyld t/ha Average potato yield 57 2008–2012 CW- 1 A10 Rc% 0–100 Percent of pixels where rice is cultivated 7 Current JRC A11 Padyld t/ha Average paddy yield 57 2008–2012 CW- 1 A12 Soy% 0–100 Percent of pixels where soybean is cultivated Current JRC A13 Soyyld t/ha Average soybean yield 57 2008–2012 CW- 1 A14 Wh% 0–100 Percent of pixels where wheat is cultivated 7 Current JRC A15 Whyld t/ha Average wheat yield 57 2008–2012 CW- 1 E1 Area 1000 km² Area in thousands of Km² n.a. n.a. CW- 2 E2 Z m a.s.l. Altitude 16 n.a. Worldclim E3 Rain mm Annual rainfall total in mm 16 1950–2000 Worldclim E4 Tavg °C Average annual temperature 16 1961–1990 Climond E5 Rady W/m Annual mean radiation total 16 1961–1990 Climond E6 NDVIavg NDVI Average annual NDVI (2 + 5 + 8 + 11)/4 21 1999–2012 VITO E7 Avm mmH O/m Easily available soil moisture 24 n.a. Geo- 1 E8 NPP gDM/(m year) Net Primary production potential 24 1976–2000 Grieser S1 Rn0510 0–1 Fraction of annual precipitation that falls from May to October 16 1950–2000 CW- 3 S2 T02- 08 °C Difference between average February and average August 23 1950–2000 CW- 3 temperature S3 TCV CV Temperature seasonality, coefficient of variation in monthly 16 1961–1990 Climond temperature S4 Tamp °C Temperature of warmest week minus temperature of coldest week 16 1961–1990 Climond S5 NDVI02- 08 NDVI (no unit) Average difference between February and August NDVI 21 1999–2012 CW- 4 S6 NDVIStDev NDVI (no unit) Variability of NDVI over time (standard deviation) 8 1981–2003 Geo- 2 Data sources. Climond: Kriticos et al. (2012), and Xu and Hutchinson (2011, 2013); CW-1: see text under 3.1; CW- 2: computed based on the MRU polygon using QGIS 2.2.0, http://www.qgis.org/en/docs/; CW- 3: computed based on WorldClim grids; CW-4: computed based on VITO (2014); GMIA: Global Map of Irrigation Agriculture, FAO (2013) or Siebert et al. (2013); IIASA: Geo- Wiki (2014); Geo- 1: http://data.fao.org/ map?entryId=61946540-bdbf-11db-a0f6-000d939bc5d8; Geo-2: http://www.fao.org/geonetwork/srv/en/metadata.show?id=37059; Grieser: Grieser et al. (2006a); JRC: data made available by the EC/JRC MARS unit http://mars.jrc.ec.europa.eu, based on the methodology described by Vancutsem et al. (2013); VITO: VITO (2014); Worldclim: Hijmans et al. (2005). MRU, monitoring and reporting units; CV, coefficient of variation. is cultivated; Irr%, average of the percentage of each pixel country centroids and interpolating them using inverse that is equipped for irrigation; Bar%, percentage of pixels distance weighting (exponent 2) to a coarse 57 km grid, in the MRU where barley is cultivated; Baryld, average as shown in Table 1. This is, admittedly, a crude method yield achieved for barley in the pixels where the crop is when applied to large areas with a low density of sample cultivated; Casyld, average yield achieved for cassava in points, but deemed acceptable to estimate 5- year reference the pixels where the crop is cultivated. Similar variables “zonal yields.” The gridded yields were subsequently used are defined for maize (Mz% and Mzyd), white potatoes to derive the average and the standard deviation by crop (Pot%, Potyld), soybean (Soy%, Soyyld), and wheat (Wh%, over each MRU. Values were retained only for the MRUs Whyld). For rice, the percentage of pixels in the MRU where the crop is actually grown, according to the JRC the crop is cultivated (Rc%) is complemented by yield crop masks (Vancutsem et al. 2013). (Padyld) expressed as paddy (husked grain). In order to compare agronomic performance, a simple Average MRU yields were computed by assigning aver- “yield index” (YldInd) was derived as follows: for each age 2008–2012 yield from FAOSTAT (FAO 2014b) to of the seven crops, yields were ranked from 1 to 52 42 © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. R. Gommes et al. Spatial Units for Global Environmental Monitoring (potatoes, which occur in 52 MRUs out of 65), 1 to 21 1 (highest variability), providing a variability/seasonality (both cassava and barley, which occur in 21 Crop MRUs) indicator (VSI). etc.; the highest rank was assigned the highest value. Ranks were then divided by the number of occurrences, resulting Method in the highest value becoming 1 and the lowest 1 divided by the number of occurrences. The final step was the Generation of MRUs calculation of the average across the seven crops for each MRU, ignoring crops that do not occur. In the case of The spatial units used for global crop monitoring and New Zealand (only potatoes and barley), only two values reporting by CropWatch are referred to as Crop were averaged. Monitoring and reporting units (MRUs). They are essentially a simplification of the FAO GEZ map with additional adjustments made based on the recent Environmental variables 1976–2000 Köppen climate map by Grieser et al. (2006b). Table 1 lists eight environmental variables (E as prefix), The following four “rules” were broadly followed. First, of which the first (E1, area of the MRU) is conven- a rather low level of detail (equivalent to merging GEZs) tionally assigned to this group. The seven variables E2 was adopted for minor- or nonagricultural areas, such to E8 thus include altitude (Z), total annual rainfall as the Central Northern Andes (MRU 21 in the (Rain), average annual temperature (Tavg), total annual map of Fig. 1) which actually cover the whole spectrum radiation (Rady), average annual NDVI, easily available from hyperarid climate (BW in the Atacama desert) soil moisture, and net primary production potential to polar (E) at the highest elevation. Second, inclusions (NPP). The pixel count, spatial average values, and which are “small” relative to the size of the MRU spatial standard deviation were directly extracted from were ignored, for instance the Alps in “non- the original grids, using a Mollweide projection for the Mediterranean western Europe” (MRU 60) or the semi- area. arid area (BS) on the border between Karnataka and Andhra Pradesh states in “Southern Asia” (MRU 45). Third, the borders of the GEZs were modified where Seasonality variables the map by Grieser and colleagues showed similar cli- There are six seasonality variables (S-variables): Rn0510, mate conditions (due to recent shifts in climate) across percentage of rainfall falling during the May to October GEZ borders. This usually involves climate warming semester (in combination with latitude, the variable indicates in high latitude or high altitude areas, typically the whether the MRUs enjoys winter or summer rainfall); T02- north- east “ appendix” of eastern- central Asia (MRU 08, the difference between average February and August 52) or the Andean part of the semiarid southern cone average monthly temperature; TCV, the coefficient of vari- (MRU 28). Fourth, some MRUs are “residual” areas, ation of 12 monthly average temperature values; Tamp, the that is, very inhomogeneous hilly terrain where the difference of temperature between the warmest week and “inhomogeneity” is actual a defining feature, typically coldest week (annual thermal amplitude); NDVI02-08, the the area between the Black Sea and the Caspian difference between February and August the average NDVI (Caucasus, MRU 29) and the Pamir area (MRU 30). and finally NDVIStDev, and the standard deviation of aver- In China, the authors adopted the ecological zones of age annual NDVI over time. NDVIStDev is a particularly Sun (1994) which are closer to the new Köppen map relevant variable as it is the only one that assesses variability than the GEZ map. The GEZ was found to be very over time, that is, a major component of farming risk. largely compatible with the cereal suitability map (FAO In order to derive a generic measure of environmental 2007) and the map of major crop types which is part variability, some of the S-variables were modified in such of FAO (2010). a way that high values express high variability, regardless Agricultural and nonagricultural areas were included of the hemisphere where the MRU occurs. For instance, in order to provide global coverage and to include some the percentage of rainfall during May to October was agriculturally marginal areas, such as low-rainfall rangeland taken as the absolute departure from 50%, as 50% cor- at the edge of deserts. responds to well distributed rainfall over the year. Variables derived as the difference between February and August Characterization of MRUs values were taken as the absolute value to make northern and southern hemisphere comparable. The values were The 65 MRUs are identified by a code from M01 to then ranked and the average rank was rescaled in order M65. The naming is conventional and was chosen only to provide a variable between 0 (lowest variability) and for easy reference (Fig. 1). ESRI shapefiles as well as the © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. 43 Spatial Units for Global Environmental Monitoring R. Gommes et al. Figure 1. Spatial extent of 65 Crop MRUs designed for global crop monitoring, together with their numbers and conventional “long” names. The matrix of the coefficients of variation values of the variables for individual MRUs are available as supporting information (Tables S1 and S2) as well as In addition to average MRU values mentioned above, the from GeoNetwork (FAO 2014a). They can also be re- spatial standard deviation for each variable was computed quested from the corresponding author. over the MRU polygons, and the spatial coefficient of variation (CV) was computed. As the CV is a standard- ized statistic, it can be used for intervariable and inter- Principal components MRU comparisons. The matrices of the CV of MRUs × variables (15 A- The environmental factors used to characterize MRUs variables, 6 S- variables and 8 E-variables of which E1 is (Table 1) are redundant because they are correlated. For area and E2 is elevation) were analyzed. As the matrix instance, temperature is linked to elevation, and radiation of A- variables is incomplete (not all crops are cultivated is linked to rainfall through cloudiness. Principal com- everywhere), a new matrix was prepared including the ponents are therefore used below in several figures (starting coefficients of spatial variation of A1, E2 to E8, S1 to S6 with Fig. 3) to characterize the MRUs based on a limited as well as the averages of A- variables (hereafter referred number of synthetic factors. 44 © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. R. Gommes et al. Spatial Units for Global Environmental Monitoring Table 2. “Top 20” of MRUs for several agronomic variables. Number of crops: number of crops between barley, cassava, maize, potatoes, rice, soybean, wheat that are cultivated in the MRU; YldInd, Yield index (unit less, defined under 2.2.1), percentage of arable land and percentage of irrigation. No. of crops cultivated Yield Arable land Irrigation No. MRU YldInd YldInd MRU Arland% MRU YldInd Irr% MRU YldInd 1 7 M26 0.75 1.00 M56 100 M34 0.58 35 M34 0.58 2 6 M04 0.20 0.96 M60 100 M45 0.45 25 M48 0.40 3 6 M23 0.50 0.94 M46 99 M12 0.91 18 M44 0.24 4 6 M45 0.45 0.91 M12 99 M37 0.54 17 M37 0.54 5 5 M01 0.11 0.84 M15 99 M40 0.47 17 M45 0.45 6 5 M03 0.15 0.81 M28 99 M59 0.78 13 M42 0.58 7 5 M05 0.09 0.81 M61 98 M33 0.53 9 M16 0.77 8 5 M07 0.68 0.80 M18 98 M41 0.57 9 M36 0.65 9 5 M09 0.21 0.78 M59 98 M60 0.96 9 M46 0.94 10 5 M12 0.91 0.77 M16 97 M17 0.65 9 M50 0.51 11 5 M14 0.72 0.75 M26 97 M29 0.47 8 M30 0.58 12 5 M16 0.77 0.74 M58 96 M14 0.72 8 M40 0.47 13 5 M17 0.65 0.72 M14 96 M36 0.65 7 M29 0.47 14 5 M21 0.46 0.71 M27 96 M48 0.40 7 M33 0.53 15 5 M25 0.61 0.70 M13 90 M26 0.75 7 M59 0.78 16 5 M29 0.47 0.68 M54 89 M03 0.15 6 M20 0.06 17 5 M34 0.58 0.68 M07 88 M50 0.51 6 M38 0.49 18 5 M37 0.54 0.65 M36 87 M20 0.06 6 M41 0.57 19 5 M38 0.49 0.65 M17 87 M44 0.24 5 M07 0.68 20 5 M40 0.47 0.64 M57 86 M46 0.94 5 M14 0.72 MRU, monitoring and reporting units; YldInd, yield index; varA, A- variables; varE, E- variables; varS, S- variables; NPP, net primary production. to as varA), E- variables (varE) and S- variables (varS), The largest MRU is about 300 times larger than the small- resulting in a 65 × 18 matrix. est ones. M53 (“North Australia”) also includes the southern fringe of maritime South-east Asia; M59 (“Mediterranean Results Europe and Turkey”) includes all of the Iberian Peninsula (including the Atlantic coast) and “Southern Africa” (M09) Map of MRUs reaches as far as the coastal areas of southern Kenya. Figure 1 shows the extent and distribution of the 65 MRUs. The largest MRU correspond to mostly nonagri- Characteristics of crops in MRUs cultural areas such as the deserts (M63, Australian desert; M64, old world deserts from the Sahara to the Afghan Table 2 summarizes some of the main features about the desert), boreal forest (tundra, taiga and permafrost areas: crops grown in the various MRUs. It focuses on the seven M51, Eastern Siberia; M57, boreal Eurasia and M65, sub- major crops in terms of worldwide production and trade arctic America - which includes Greenland and Iceland). if sugar and oil crops are excluded: barley, cassava, maize, There are some other minor instances of transcontinental potatoes, rice, soybean and wheat. It is also stressed that zones, although they are not easily visible in the map the conclusions are only partly true in areas where other above: M53, North Australia also includes the southern crops play an important part (e.g., coarse grains such as fringe of the Western province in Papua New Guinea oats, or sunflower and oil palm, etc.). Additionally, the and areas in Indonesia and Timor Leste at the same lati- spatial variability inside MRUs could also depend on the tude, as already mentioned above. farm size, field size (e.g., huge fields in USA, Kazakhstan The smallest zones correspond to the southernmost or Russia), the applied agro- techniques (mixed dry farm- islands of China (Hainan, M33; China Taiwan, M42). ing and irrigation) etc. The next smallest area corresponds to Western Cape in Only one zone (M26, the humid Pampas) grows all South Africa (M10, a Mediterranean climate “enclave”); seven crops, while three grow six: M04, the Horn of variables S1 and S6 show that seasons are inverted com- Africa; M23, Central-eastern Brazil and M45, southern pared with the surrounding C9, Southern Africa MRU. Asia, including mainly India. The areas are identified by © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. 45 Spatial Units for Global Environmental Monitoring R. Gommes et al. Figure 2. Number of crops (among barley, cassava, maize, potatoes, rice, soybean, wheat) that are gown in various areas. Seven crops occur in the Pampas (M26) but the number is not shown in the legend because the locations are not visible at the scale of the map. the presence of ecosystems at relatively high elevations northern Great Plains (M12). Yields are mostly modest which can accommodate both tropical and temperate crops, (in the YldInd range from 0.5 to 0.6) with the exception sometimes with relatively low yields by world standards of the northern Mediterranean and Turkey (YldInd = 0.78) (e.g., the Horn of Africa). and Western Europe (0.96). A number of regions grow five crops, covering the The observation about modest yields also applies to yield spectrum from low (M5, north and central the main irrigated areas, with values of 0.58 in Huang Madagascar; M1, equatorial central Africa and M3, the Huaihai (M34, 35% of the area equipped for irrigation) gulf of Guinea) to high (M12, American northern Great and 0.40 in Punjab to Gujarat (M48, 25% irrigated). Plains). Figure 2 shows the distribution of the number of crops with additional detail. It is striking how the Environmental features of MRUs zones mentioned are all characterized by one or two dominant “background” crops with patches of ecologically Based on the correlations between them, the seven envi- diverse areas where additional crops occur. ronmental variables from E2 to E8 (refer to Table 1) can The highest yields occur in New Zealand (M56), non- very clearly be subdivided into three groups of factors, Mediterranean Western Europe (M60), Southern Japan corresponding to the three- first principal components. They and Korea (M46) and in the northern Great Plains in account for a total of 92% of the variance of the variables, the USA. It is to be stressed that the number of crops distributed among the components as follows: 49%, 28% in M56 is just two crops (barley and potatoes) which and 14%. Figure 3 shows a plot of the production zones benefit from very long days, resulting in long periods of against the two- first principal components. photosynthesis. The remaining areas cultivate between The first component correlates very strongly with av- three and five crops. Lowest yields occur in Equatorial erage NDVI (R = −0.91), NPP (R = −0.95) and total central Africa (M01), the east African highlands (M02), rainfall (R = −0.90). It can thus be interpreted as “un- Madagascar (M05) and the Caribbean (M20). It is stressed productivity” (MLA, 2015) a biological productivity vari- that the listed crops are marginal in some of the areas, able, increasing from right to left. The second component where yams, millets, teff, pigeon peas, plantain, and others PC2 (or “coldness”) is clearly and negatively associated play the greatest role as staple foods. with radiation (Rady; R = −0.93) and with annual aver- Table 2 also indicates how widespread arable land is age temperature (R = −0.71). More surprisingly, PC2 is in the various production zones. In 15 MRUs the arable positively correlated with the soil moisture storage ca- land fraction exceeds 90%. Most of them occur in China pacity (E7). The relation between E7, temperature and and India (including M34, Huang Huaihai, southwest radiation is not direct, but it can be hypothesized that China M41 and M45, southern China), but also in western low temperature is associated slow soil organic matter Europe and the northern Mediterranean (M59 and M60), degradation, as organic matter and other colloids play Sierra Madre in Mexico and the US (M17) and the an important part in the soil moisture storage capacity. 46 © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. R. Gommes et al. Spatial Units for Global Environmental Monitoring Figure 4. Seasonality in the 65 MRUs represented on the plane of the Figure 3. Plots of MRUs (numbers defined in Fig. 1) against the 1st and two- first principal components, where the first captures temperature second principal components extracted from the six environmental and NDVI variability and the second represents rainfall seasonality. The variables E2–E7. The first component (unproductivity) corresponds to colors indicate the percentage of arable land in five categories. the production potential (decreasing from left to right) and the second Additional symbols (arrows) refer to the variability in NDVI over time: up (coldness) captures radiation and temperature (high at the bottom to for the five most variable MRUs, down for the least variable ones and a low at the top). The colors indicate the percentage of arable land in five horizontal bar for average values. categories. The correlation between E7 and the factors which posi- to a variety of factors, including low population densities tively affect soil organic matter production is positive and topography. (NDVI: R = 0.42; NPP: R = 0.24 and Rain: R = 0.28) which is an additional argument in favor of the climatic Seasonality and variability over time in determinants of E7. MRUs An interesting observation refers to elevation, which is the variable most closely correlated with the third principal Based on the six S-variables in Table 1 modified as component (R = 0.73). This is not so surprising since, described under 2.2.3, the 6 × 65 data matrix was given the large geographic coverage of the production subjected to a principal components analysis and the zones; rainfall and temperature are not correlated with MRUs were plotted against the two-first components altitude, as would happen for a spatially more detailed (Fig. 4). The 67.8% of the variance is absorbed by the study. Therefore, the altitude variable is best regarded as first component which correlates with the temperature- a proxy for “type of landscape” as high elevations tend related variables and intra- annual NDVI amplitude (R to be characterized by rugged terrain. between 0.89 and 0.96 for N = 65 MRUs); it can con- Unfavorable temperature and biomass conditions occur veniently be referred to as “temperature variability”. in the ecologically similar M51 (Eastern Siberia) and M65 The second (16.8% of variance) is virtually equivalent (subarctic America), both characterized by little or no to rainfall variability (the absolute value of February agriculture. Very favorable conditions of productivity oc- rain minus August rain, R = 0.99) while the third cur in M24 (Amazon) and in C9 (Southern Africa) in (13.1% of variance) is best correlated with interannual spite of the presence of the semiarid Kalahari “desert”. variability of NDVI, R = 0.62). Together, the three M50 (mainland South-East Asia) and M19 (Central and components account for 97.6% of the variance of the Northern South America) are among the areas where a original variability/seasonality matrix. The above- high fraction of arable land coincides with a high pro- mentioned VSI correlates well with temperature vari- duction potential. It is worth noting that some areas such ability (R = 0.96). It is interesting to observe that, at as M42 (China Taiwan), M05 (northern and central the scale of the MRUs, rainfall seasonality, intra-annual Madagascar) M01 (Equatorial central Africa) do not, at temperature variability and interannual NDVI are largely present, fully exploit their potential, which may be due uncorrelated variables. © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. 47 Spatial Units for Global Environmental Monitoring R. Gommes et al. (A) (B) (C) (D) According to Fig. 4, the MRUs with low intra-annual rainfall variability occur in two types of situations: Figure 5. Within- MRU spatial variability (coefficients of variation) of agronomic (A), environmental (B), seasonality (C) variables, and elevation (D). 48 © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. R. Gommes et al. Spatial Units for Global Environmental Monitoring Figure 6. Plots of the average coefficients of spatial variation against each other. The indicated outliers and extremes are M28 (semiarid southern cone in Latin America), M39 (Qinghai- Tibet in China), M63 (Australian desert) and M65 (subarctic America). rainfall is very low (M64, Sahara; M28, semiarid southern continental and the islands, M49 and M50). Most variable cone) and therefore cannot vary a lot, or rainfall is temperatures are found in the most continental areas of actually high but well distributed over the year (M56, north- east China (M38), central Asia (M52), and eastern New Zealand; M60, non- Mediterranean Western Europe). Siberia (M51). High rainfall seasonality characterizes the West African Sahel (M08), parts of China (North- East China, Spatial variability inside MRUs M38 and Inner Mongolia, M35), South-west Madagascar (M06) etc. While the A- , E- , and S- variables are relevant in describ- As expected, low temperature intra- annual variability ing the potential and actual farming in the CropWatch affects equatorial areas (M19, Central America and northern monitoring zones, spatial variability in the same variables South America; M20, Caribbean and south- east Asia, both needs to be looked at to understand how homogeneous © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. 49 Spatial Units for Global Environmental Monitoring R. Gommes et al. the zones are, which is an important requisite for global general, the spatial variability in yield inside the MRUs monitoring. is low, mostly between 10% and 15%, while the variability A principal components analysis of the matrix of the in pixels cultivated under various crops or irrigated pixels CV shows that no <11 factors (components) are required is very high, usually between 500% and 1000%. to account for 95% of the variance of the matrix. If the It is also stressed that there is virtually no link between original A, E and S variables are added (65 × 38), 18 the spatial variability in the three variable types (Fig. 6, components are needed to reach the same percentage. but also Fig. 5). In fact, the principal components of the VarA is different from varE and varS in the sense that average spatial variability carry 50%, 25%, and 25% of it is based largely on the “presence” (pixel value of 1) the variance. Figure 6 also shows the significant positive and absence (pixel value of 0) of crops. For a given crop, skew affecting all three variability variables. the average thus indicates the fraction of the area of the Among the varA variables, the lowest spatial variability MRU where the crop is cultivated. As a result, the coef- occurs, on average in M33 (11% in Hainan, China) while ficient of variation is directly depends on the mentioned the most heterogeneous area, from an agronomic point fraction. For values between 0.3 and 0.9, CV depends of view, is M65 (2587%, subarctic America). The Sahara directly on the fraction and varies almost linearly from for instance, which belongs to M64 together with the 150% (fraction = 0.3) to 17% (0.9). Arabian Peninsula and hyperarid regions in Iran and The spatial variability in elevation (E2) can probably Afghanistan displays a spatial variability of 668% as it is be regarded as a good yardstick when comparing spatial made up by low- variability desert with some irrigated variability. The lowest values correspond to M39 (Qinghai- pixels, mostly in the Nile valley. Tibet, 19%) and the east African Highlands (M02, 27%) E- variables, as mentioned, display limited variability, while the five highest values from 132% to 146% cor- which results from the definition of the MRUs as envi- respond to M19 (central and northern South America), ronmentally homogeneous areas: the variation covers the M20 (Caribbean), M44 (Southern Himalayas), M49 (mari- range from 19% (M22, Brazilian Nordeste) to 96% (M39, time south- east Asia) and M62 (Ural to Altai Mountains). Qinghai- Tibet in China), for an average of 38%. The As expected, plateaus show low elevation variability while spatial variability in the seasonality variables is larger: 72% MRUs including lowlands and highlands display high on average, from a low value around 10% in M38 (north- values. east China) and M34 (Huang Huaihai) to values between Figure 5 shows how varA, varE and varS vary among 200% and 400% (207% in M49, maritime south-east Asia; the MRUs, in comparison with the spatial variability in 230% in M4, the Horn of Africa and 385% in M28, the elevation, which is the first factor that disturbs zonality. semiarid southern cone in Latin America). The second is a weak E- W gradient that is visible in Fig. 5B (varE) at high northern latitudes. To interpret Discussion the spatial variability in elevation, it is necessary to keep in mind the definition of the coefficient of variation, which Although primarily based on the natural vegetation GEZ explains why highlands such as the East African Highlands map (FAO 2012), the MRUs shows a good agreement with (M2) and Qinghai- Tibet (M39) display relatively low val- the spatial distribution of major crops zones (Fig. 2). Both ues: the average elevation that divides standard deviation natural ecosystems and crops depend on natural resources, is high. On the other hand, the North China Plain (M34, but crops benefit in addition from genetic improvements Huang Huaihai) is at a low elevation, but includes some and management practices that take them to the “limit” higher terrain in the west. of their ecophysiology into environments that would be Western Asia (M31) and the central- northern Andes (M21) marginal without human intervention. This is particularly are among the most spatially inhomogeneous areas, as the visible in the areas where cropping is most intensive in heterogeneity affects all variables. The areas that display low terms of percentage of arable land and yields achieved. variability for several groups of variables can fall in that It is also noted, as expected (Fig. 4), that high potential category for several reasons. Hainan (M33), for instance, areas are also areas of low variability, which may result is a tropical island, but other areas (M34, Huang Huaihai) in high cropping intensities, as in continental and maritime belong there because they are very extensively cropped, SE Asia (M49 and M50) and in Central and northern resulting in exceptionally low values of spatial variability South America (M19). Altogether, the MRUs thus retain (11%) for a variable that is very high in most other the logic that underlies the GEZ, which is not surprising areas. because agriculture remains a major component of the Large differences in coefficient of variation are observed environment (Monfreda et al. 2008) and because arable between the categories of variables, 330%, 38%, and 55% land occupies a large fraction of land in most MRUs on average for varA, varE, and varS, respectively. In (Table 2). In fact, according to recent estimates, as much 50 © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. R. Gommes et al. Spatial Units for Global Environmental Monitoring Figure 7. Plot of average coefficient of variation for agricultural (A for varA), environmental (E for varE) and seasonality (S for varS) variables against MRU area, after removal of mostly nonagricultural near-polar and/or arid areas (M27, M28, M51, M57, M61, M63, M64 and M65). The numbered points correspond to equatorial Africa (1, M01), the Amazon (24, M24), western Asia (29, M31), Gansu-Xinjiang (30, M32), Qinghai- Tibet (37, M39) and maritime south- east Asia (47, M49). Table 3. Comparison of some MRU descriptors as a function of the fraction of arable land. <75%, less than 75% arable (36 MRUs); >75%, more than 75% arable (29 MRUs). R = coefficient of correlation of agricultural variability versus Area, S is the corresponding slope. Definition of variables is as in Table 1. Avg. is average and CV% is the coefficient of variation in % between the respective variables over MRUs belonging to each of the two groups. Reg. stands for the regression of varA against MRU area. Reg. varA varE varS Area Arland% Irr% Z Rain Avm NPP <75% R = 0.30 Avg. 446 43 63 2610 39 2 894 693 94 80 97 94 83 17 64 S = 0.05 CV% 105 35 114 67 150 >75% R = 0.19 Avg. 185 32 45 1380 91 7 556 1123 100 135 S = 0.06 CV% 91 25 93 71 9 114 86 50 17 33 MRU, monitoring and reporting units; CV, coefficient of variation; NPP, net primary production. as 23.8% of the potential global NPP is directly or in- designed to assess water-related issues. They result from directly managed by man (Haberl et al. 2007). the intersection of 115 geopolitical regions and 126 hy- In addition, natural vegetation and crops are subjected drographic basins and are generally found acceptable for to the same global environmental patterns of not only water- related studies (Kummu et al. 2011). On the other resources but variability as well, where latitude and to- hand, for statistical purposes, the United Nations geo- pography play a major role. In general, a rather large scheme (UN, 2010) adopts just 21 global units, which are number of agroecological classes are required to capture also used by OECD. spatial variations of agriculture with sufficient detail. What In reality, the issue of what amount of variability within does “sufficient” stand for? van Wart et al. (2013) found zones is acceptable is bound to remain subjective as it that, in general, all global zoning approaches that include depends on the intended application. In this study, we less than 100 units retain significant within- unit variability. found that large differences (as measured by the average Indeed, all national agroecological zone studies go into coefficient of variation varA, varE, and varS) are observed significant detail, as amply documented by the national between the categories of variables: 330% for varA, which agro-ecological maps available in GeoNetwork (FAO, refers mostly to crop distribution within MRUs. The value 2014a). To take just one example, for India, the agricultural of varA would drop to significantly lower values if the area of which is smaller than China’s, Pal et al. (2009) calculations were restricted to arable land (0% would be deem that about 130 zones are required for national work, possible only in the case of one single crop). Spatial vari- while they encourage states to add additional level for ability in yields, on the other hand, is in the range of more detailed local studies. In China, on the other hand, 10%–15% (no doubt also partly due to the method used the large zones proposed by Sun are just nine, which are to estimate crop yields by MRU). about the same size as the MRUs adopted for India. The Environmental variables show a much lower range of 281 Food Production Units (FPU) proposed by Nelson spatial variability (average varE is 38%) and so do sea- et al. (2010) and Rosengrant et al. (2012) were originally sonality variables (average varS is 55%). It is stressed that © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. 51 Spatial Units for Global Environmental Monitoring R. Gommes et al. seasonality is expected to be more variable than the en- low; it does not seriously affect the validity of large vironmental variables as it also includes NDVI variability agroecological zones as monitoring as well as reporting over time, as a measure of cropping risk. units. Regardless of the size of the MRUs, it is the pre- Figure 7 shows the plot of the three average spatial vailing environmental conditions that condition vulnerabil- variability indicators against the MRUs’ area, after exclu- ity patterns and the type of farming that can be practiced, sion of major cold and/or dry noncropped areas. an extreme case being irrigated crops in desert areas (e.g., Correlations are all positive, as expected, but significant in the Nile valley, or the semiarid areas in central Asia only for VarA (agricultural variability, R = 0.45, significant or the hills immediately bordering the Amazon). Even if below P = 0.01). As far as environmental and seasonality the number of spatial units were increased, because of conditions are concerned, the MRUs may be considered the fractal nature of landscapes, there would always remain homogeneous, which directly results from the derivation some MRUs that include portions that are transitional of MRUs from the FAO GEZs. to another MRU, and that attract or exclude agriculture In Fig. 7A, the slope of varA against the area of the more than neighboring homogeneous areas. Therefore, for MRUs is 0.08 (the figure includes all but seven nonag- the purpose of operational global monitoring and report- ricultural MRUs). Table 3 shows, among other descriptors ing, a limited but manageable number of spatial units from Table 1, that varA is highest in areas with low defined based on general environmental variables is ac- arable land, contrary to varE and varS, and some other ceptable: the proposed CropWatch MRUs are meaningful variables. In Table 3 the arable land limit of 75% was reporting units at the global scale. chosen because it is close to the median (74%), so that the number of MRUs inside each category is approximately Acknowledgments the same. Area with low varA, coincide with high potential areas (high rainfall, high NPP, widespread irrigation) while The authors specifically acknowledge the financial support other variables, including elevation and soil moisture do through the Special Fund for Grains-scientific Research in not play a part that is commensurate with the role of the Public Interest (Grant No. 201313009-2 and 201413003-7) the factors that directly affect NPP (e.g., rainfall). the National High Technology Research and Development It is also noted that the correlation between varA and Program of China (863 program), Grant No. 2012AA12A307; MRU area is no longer significant (significance threshold and the Visiting Professorships for Senior International for 35 observations is R = 0.32 at P ≤ 0.05, and larger Scientists, Grant No. 2013T1Z0016. The authors thank the for 29 observations as well as for P ≤ 0.01). In general, reviewers for providing thorough comments that have im- areas with high arable land fractions also correspond to proved the quality and readability of the paper. higher average annual temperature and higher average NDVI, although this is not shown in Table 3 (17°C vs. Conflict of Interest 11°C and 0.54 vs. 0.39). On the other hand, the average standard deviation of NDVI over time (S6) and average None declared. annual NDVI amplitude (S5) are virtually identical between References the groups (0.21–0.22). The fact that the CV between MRUs in the groups in Table 3 is comparable (in par- Akpan, S. B., E. J. Udoh, and V. S. Nkanta. 2012. Factors ticular varA: 105% and 91%) indicates that the “ecological” influencing fertilizer use intensity among small holder logic behind the selection of the MRUs remains meaningful crop farmers in Abak Agricultural Zone in Akwa Ibom for agriculture, across different scales. State, Nigeria. J. Biol. Agric. Healthc. 2:54–65. Babu, S. C., and V. J. Quinn. 1994. Food security and nutrition monitoring in Africa: introduction and Conclusions historical background. Food Policy 19:211–217. The current MRU subdivision by CropWatch adopts 65 Binder, C. R., G. Feola, and J. K. Steinberger. 2010. units; it is derived from a GEZ map and used for op- Considering the normative, systemic and procedural erational monitoring, including areas with very limited dimensions in indicator- based sustainability assessments or no crop or livestock production. in agriculture. Environ. Impact Assess. Rev. 30:71–81. Spatial variability in environmental and seasonality vari- Chen, S. T., H. I. Kuo, and C. C. Chen. 2010. Modeling ables inside MRUs only marginally depends on their size, the relationship between the oil price and global food which stresses their intrinsic homogeneity. They are spatially prices. Appl. Energy 87:2517–2525. less homogeneous for agricultural variables. Dalgaard, T., N. J. Hutchings, and J. R. Porter. 2003. This is a direct consequence of the low density of ag- Agroecology, scaling and interdisciplinarity. Agric. Ecosyst. riculture in the areas where the production potential is Environ. 100:39–51. 52 © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. R. Gommes et al. Spatial Units for Global Environmental Monitoring Doré, T., D. Makowski, E. Malézieux, N. Munier-Jolain, M. agricultural productivity in the early twenty- first Tchamitchian, and P. Tittonell. 2011. Facing up to the century. Philos. Trans. R. Soc. Lond. B Biol. Sci. paradigm of ecological intensification in agronomy: 365:2973–2989. revisiting methods, concepts and knowledge. Eur. J. Grieser, J., R. Gommes, S. Cofield, and M. Bernardi, Agron. 34:197–210. 2006a. World maps of climatological net primary FAO. 1999. FRA 2000, A concept and strategy for ecological production of biomass, NPP. FAO, 2005. 5 pp. http:// zoning for the global forest resources assessment 2000. www.juergen-grieser.de/downloads/NetPrimaryProduction/ INTERIM REPORT. FAO, Forestry Dep. Working Paper npp.pdf (accessed September 2015). 20. Rome. 28 pp. Grieser, J., R. Gommes, S. Cofield, and M. Bernardi, 2006b. FAO. 2007. Most suitable cereal. Digital map downloadable New gridded maps of Koeppen’s climate classification. from FAO GeoNetwork web site http://www.fao.org/ Data, methodology and gridded data Available at: http:// geonetwork/srv/en/main.home. Plate 48b in the FAO/ www.fao.org/nr/climpag/globgrids/KC_classification_en.asp. IIASA GAEZ study http://www.fao.org/nr/gaez/en/. Methodology also downloadable from http://www. Methodology in 2002 document jointly published by juergen-grieser.de/downloads/Koeppen-Climatology/ FAO and IIASA and available from http://webarchive. Koeppen_Climatology.pdf (accessed September 2015). iiasa.ac.at/Research/LUC/SAEZ/pdf/gaez2002.pdf (accessed Haberl, H., K. H. Erb, F. Krausmann, V. Gaube, A. September 2015). Bondeau, C. Plutzar, et al. 2007. Quantifying and FAO. 2010. Land Use Systems of the World. Digital map mapping the human appropriation of net primary downloadable from FAO GeoNetwork web site http:// production in earth’s terrestrial ecosystems. Proc. Natl www.fao.org/geonetwork/srv/en/main.home (accessed Acad. Sci. USA 104:12942–12947. September 2015). Haberle, S. G., and A. Chepstow-Lusty. 2000. Can climate FAO. 2012. Global ecological zones for FAO forest influence cultural development? A view through time reporting: 2010 Update. Forest Resources Assessment Environ. Hist., 34:9–369. Working Paper N. 179. 42 pp. FAO, Rome. Hijmans, R. J., S. E. Cameron, J. L. Parra, P. G. Jones, and FAO. 2013. Global map of irrigation areas - version 5. A. Jarvis. 2005. Very high resolution interpolated climate Digital map downloadable from FAO GeoNetwork web surfaces for global land areas. Int. J. Climatol. site http://www.fao.org/geonetwork/srv/en/main.home. 25:1965–1978. The version used was downloaded on 20130901. Hodson, D. P., and J. W. White. 2007. Use of spatial Version 5 was published in October 2013 (accessed analyses for global characterization of wheat- based September 2015). production systems. J. Agric. Sci. 145:115–125. FAO. 2014a. GeoNetwork, a repository of spatial information Hu, Z., and D. Zhang. 2006. China country pasture/forage about world agriculture. http://www.fao.org/geonetwork/ resource profiles. Pp. 63. FAO, Rome. srv/en/main.home (accessed September 2015). Iizumi, T., J. J. Luo, A. J. Challinor, G. Sakurai, M. FAO. 2014b. FOSTAT, the global statistical database of Yokozawa, H. Sakuma, et al. 2014. Impacts of El Niño FAO. Available at: http://faostat3.fao.org/faostat-gateway/ Southern Oscillation on the global yields of major crops. go/to/home/E (accessed September 2015). Nat. Commun. 5:3712. George, T. 2014b. Why crop yields in developing countries Kelly, V. A., 2010. Factors affecting demand for fertilizer in have not kept pace with advances in agronomy. Glob. sub-Saharan Africa. Agriculture and Rural Development Food Sec. 3:49–58. Discussion Paper N. 23. The World Bank, Washington, Geo-Wiki. 2014. Data downloaded from the Geo-wiki USA. 89 pp. project. Available at http://agriculture.geo-wiki.org/login. Kriticos, D. J., B. L. Webber, A. Leriche, N. Ota, I. Macadam, php?menu=results. See also http://www.iiasa.ac.at/web/ J. Bathols, et al. 2012. CliMond: global high- resolution home/research/researchPrograms/ historical and future scenario climate surfaces for EcosystemsServicesandManagement/Geo-Wiki.en.html. bioclimatic modelling. Method Ecol. Evol. 3:53–64. Dataset downloaded from Beta-hybrid.geo-wiki.org on Kummu, M., H. de Moel, P. J. Ward, and O. Varis. 2011. 20130901 (accessed September 2015). How close do we live to water? A global analysis of Gommes, R., J. du Guerny, M. H. Glantz, and L.-N. Hsu, population distance to freshwater bodies. PLoS One 2004. Climate and HIV/AIDS: a hotspots analysis for 6:e20578. early warning rapid response systems. UNDP/FAO/NCAR, Leblois, A., P. Quirion, and B. Sultan. 2014. Price vs. UNDP SE Asia and Development Programme, Bangkok, weather shock hedging for cash crops: ex ante evaluation 20 pp. Available at: http://www.fao.org/forestry/15532-0-0. for cotton producers in Cameroon. Ecol. Econ. pdf (accessed September 2015). 101:67–80. Gornall, J., R. Betts, E. Burke, R. Clark, J. Camp, K. Loyce, C., J.M. Meynard, C. Bouchard, B. Rolland, P. Lonnet, Willett, et al. 2010. Implications of climate change for P. Bataillon, et al. 2012. Growing winter wheat cultivars © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. 53 Spatial Units for Global Environmental Monitoring R. Gommes et al. under different management intensities in France: a Schmid, A. P., 1998. Thesaurus and glossary of early multicriteria assessment based on economic, energetic warning and conflict prevention terms (abridged version). and environmental indicators. Field Crops Res. Synthesis Foundation. Erasmus University, Amsterdam. 125:167–178. Edited for Forum on Early Warning and Early Response Lu, W. C. 2002. Effects of agricultural market policy on (FEWER) by S. B. Anderlini. FEWER Secretariat, London. crop production in China. Food Policy 27:561–573. 30 pp. MLA. 2015. Dictionary.com Unabridged. Random House, Siebert, S., V. Henrich, K. Frenken, and J. Burke. Inc. 22 Jul. 2015. Available at: http://dictionary.reference. 2013. Update of the digital Global Map of com/browse/unproductivity. Irrigation Areas (GMIA) to version 5. 171 pp. Monfreda, C., N. Ramankutty, and J. A. Foley. 2008. Farming Institute of Crop Science and Resource Conservation, the planet: 2. Geographic distribution of crop areas, yields, Rheinische Friedrich-Wilhelms-Universität Bonn, physiological types, and net primary production in the year Bonn, Germany. 2000. Global Biogeochem. Cycles, 22:GB1022. Snodderly, D., 2011. Peace Terms: Glossary of Terms for Monfreda, C., N. Ramankutty, and T. W. Hertel. 2009. Global Conflict Management and Peacebuilding. United States agricultural land use data for climate change analysis. Institute of Peace, Academy for International Conflict Chap. 2 in Economic analysis of land use in global climate Management and Peacebuilding, Washington USA, change policy (33–48) edited by Hertel, T.W., Rose, S., 60 pp. Tol, R. 2009. 348 pp. Routledge Press, London and New Sombroek, W. S., and R. Gommes, 1996. The climate York. ISBN 978-0415619813. change-agriculture conundrum, Pp. 1–14 in F. Bazzaz, W. Nelson, G. C., R. W. Rosegrant, A. Palazzo, I. Gray, C. Sombroek, eds. Global climate change and agricultural Ingersoll, R. Robertson, et al. , 2010. Food Security, production. 345 pp. FAO and John Wiley, & Sons, Farming, and Climate Change to 2050: Scenarios, Results, Chichester, UK. ISBN 0 471 96927 3. Policy Options. International Food Policy Research Sun, He., 1994. Agricultural natural resources and regional Institute IFPRI, Washington, D.C. 131 pp. Available at: development of China. Jiangsu Science and Technology http://www.ifpri.org/sites/default/files/publications/rr172.pdf Press, Nanjing. (in Chinese) (accessed September 2015). UN. 2010. United Nations Statistics Division. Composition OECD. 2007. Glossary of statistical term. 863 pp. OECD, of Macro-Geographical (Continental) Regions, Paris. Geographical Sub-Regions, and Selected Economic and Pal, D. K., D. K. Mandal, T. Bhattacharyya, C. Mandal, and Other Groupings. Available at: http://millenniumindicators. D. Sarkar. 2009. Revisiting the agro- ecological zones for un.org/unsd/methods/m49/m49regin.htm (accessed crop evaluation. Indian J. Genet. Plant Breed. September 2015). 69:315–318. Vancutsem, C., E. Marinho, F. Kayitakire, L. See, and Ramankutty, N., J. A. Foley, J. Norman, and K. McSweeney. S. Fritz. 2013. Harmonizing and combining existing 2002. The global distribution of cultivable lands: current land cover/land use datasets for cropland area patterns and sensitivity to possible climate change. Glob. monitoring at the African continental scale. Remote Ecol. Biogeogr. 11:377–392. Sens. 5:19–41. Reynolds, M. P., ed. 2010. Climate change and crop VITO. 2014. NDVI Based on average 1999-2012 production. CABI climate change series 1. 292 pp. CABI, monthly SPOT VEGETATION NDVI, downloadable Wallingford, UK. from https://earth.esa.int/web/guest/pi-community/ Rosengrant, M. W., and the Impact development team. apply-for-data and http://www.vito-eodata.be/PDF/portal/ 2012. International Model for Policy Analysis of Application.html#Home. (accessed in September 2015). Agricultural Commodities and Trade (IMPACT). Model van Wart, J., L. G. van Bussel, J. Wolf, R. Licker, P. Description. 50 pp. International Food Policy Research Grassini, A. Nelson, et al. 2013. Use of agro- climatic Institute IFPRI, Washington, DC. Available at: http:// zones to upscale simulated crop yield potential. Field ebrary.ifpri.org/cdm/ref/collection/p15738coll2/id/12735 Crops Res. 143:44–55. (accessed September 2015). Wiggins, S., and J. Brooks. 2010. The use of input subsidies Rozbicki, J., A. Cegli, D. Gozdowski, M. Jakubczak, G. in developing countries. 22 pp. OECD, Paris. Cacak-Pietrzak, W. Madry, et al. 2015. Influence of the Wu, B., J. Meng, Q. Li, N. Yan, X. Du, and M. Zhang. cultivar, environment and management on the grain yield 2014. Remote sensing- based global crop monitoring: and bread- making quality in winter wheat. J. Cereal Sci. experiences with China’s CropWatch system. Int. J. 61:126–132. Digital Earth 7:113–137. SAGE. 2002. Available at: and http://library.mcmaster.ca/ Wu, B., R. Gommes, M. Zhang, H. Zeng, N. Yan, maps/geospatial/atlas-biosphere (accessed September 2015). W. Zou, et al. 2015. Global crop monitoring: a 54 © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. R. Gommes et al. Spatial Units for Global Environmental Monitoring satellite- based hierarchical approach. Remote Sens. Supporting Information 7:3907–3933. Additional supporting information may be found in the Xu, T., and M. Hutchinson. 2011. ANUCLIM version online version of this article at the publisher’s web-site. 6.1 user guide. 85 pp. The Australian National Table S1. Spatially averaged values of agronomic vari- University, Fenner School of Environment and ables by MRU. A1 Arland%, A2 Irr%, A3 Bar%, A4 Baryld, Society, Canberra. A5 Casyld, A6 Mz%, A7 Mzyd, A8 Pot%, A9 Potyld, A10 Xu, T., and M. F. Hutchinson. 2013. New developments and Rc%, A11 Padyld, A12 Soy%, A13 Soyyld, A14 Wh%, A15 applications in the ANUCLIM spatial climatic and Whyld. Refer to table 1 for additional details of variable bioclimatic modelling package. Environ. Model. Softw. definitions. 40:267–279. Table S2. Spatially averaged values for environmental Yu, Y., Y. Huang, and W. Zhang. 2012. Changes in rice and seasonality/time variability for each PSZ. E1 Area, E2 yields in China since 1980 associated with cultivar elevation, E3 Rain, E4 Tavg, E5 Rady, E6 NDVIavg, E7 improvement, climate and crop management. Field Crops Avm, E8 Npp, S1 rn0510%, S2 T02-08, S3 bio04, S4 bio07, Res. 136:65–75. Zhang, X., S. Wang, H. Sun, S. Chen, L. Shao, and X. Liu. S5 dvi2-8, S6 NDVIStDev and VSI, the variability/season- 2013. Contribution of cultivar, fertilizer and weather ality indicators which varies from 0 to 1 (least variable to yield variation of winter wheat over three decades: a to most variable MRU). Refer to table 1 for additional case study in the North China Plain. Eur. J. Agron. details of variable definitions. 50:52–59. Data S1. ESRI shapefiles for MRU polygons. © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. 55 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Food and Energy Security Wiley

Design and characterization of spatial units for monitoring global impacts of environmental factors on major crops and food security

Download PDF
16 pages

Loading next page...
 
/lp/wiley/design-and-characterization-of-spatial-units-for-monitoring-global-4Oix88y5is

References (37)

Publisher
Wiley
Copyright
© 2016 John Wiley & Sons Ltd and the Association of Applied Biologists
ISSN
2048-3694
eISSN
20483694
DOI
10.1002/fes3.73
Publisher site
See Article on Publisher Site

Abstract

Crop distribution, crop monitoring, This paper describes the design and characterization of a limited number of environmental variability, global agriculture, spatial units. areas to be used as spatial monitoring and reporting units (MRU) by Crop- Watch, the global crop monitoring system (http://www.cropwatch.com.cn/) Correspondence hosted by the Chinese Academy of Sciences. The MRUs are modified and adapted Bingfang Wu, CAS Key Laboratory of Digital from an existing standard vegetation zoning map. They are designed to be areas Earth Science, Institute of Remote Sensing of uniform vulnerability as assessed by three sets of variables: environmental and Digital Earth (RADI), Chinese Academy of (such as climate and NDVI, a satellite-based Normalized Difference Vegetation Sciences, Beijing 100101, China. Tel: Index), seasonality (such as intra- annual climate variations and interannual NDVI +86 10 64855689; Fax: +86 10 64858721; E-mail: wubf@radi.ac.cn variability) and agronomy (such as presence/absence of major crops and their yield). The paper pays due attention to within- zone spatial variability for each Funding Information of the three groups of variables: in general, the variability measured as the aver- In addition to regular program funding, the age of the spatial coefficient of variation inside MRUs, is much larger for ag - Digital Agriculture Division, Remote Sensing ronomic variables (330%) than for environmental variables (38%) and seasonality and Digital Earth Institute (RADI), Chinese (55%). The MRUs provide a rather coherent picture of the variations of global Academy of Sciences received support from agriculture, for instance, the links between the distribution of crops, agricultural the special Fund for Grains-scientific Research in the Public Interest (Grants No. 201313009- production potential, and environmental variability (over space and time). They 2 and No. 201413003-7), the National High closely delineate the distribution of the major nonsugar food crops (barley, Technology Research and Development cassava, maize, potatoes, rice, soybean, and wheat). The discussion focuses on Program of China (863 program), Grant No. the relations between the size of the MRUs and the within- MRU variability 2012AA12A307. The first author is supported (spatial heterogeneity) of cropping, and environmental conditions including by Grant No. 2013T1Z0016 for Visiting seasonality. The conclusion stresses that spatial variability in agriculture is bound Professorships for Senior International to be larger than the variability in the environmental variables used to define Scientists. the units, regardless of the units’ size. The size of the spatial units (and therefore Received: 29 April 2015; Revised: 23 July their number) is not a very critical constraint for operational impact reporting, 2015; Accepted: 3 September 2015 especially if impact indicators focus on agricultural areas inside MRUs. Food and Energy Security 2016; 5(1): 40–55 doi: 10.1002/fes3.73 information collected for impact assessments includes both Introduction impacting factors and impacted system. For global crop monitoring and reporting, some generalization of methods Most definitions of “monitoring” and “crop monitoring” is required because of the gap between the scale of the refer to the regular and standardized collection and analysis observations (whether points or large pixels) and the scale of information over a specific area, often by an external at which impacts are assessed and decisions are taken, and neutral observer, with a view to understanding the usually administrative or other large units (Dalgaard et al. causes of, preventing or limiting damage (Babu and Quinn 2003). 1994; Schmid 1998; OECD 2007; Snodderly 2011). The © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. R. Gommes et al. Spatial Units for Global Environmental Monitoring There exists a number of different global monitoring characterizing a limited number of zones to be used by units, such as in the developed and developing countries, CropWatch for global monitoring and reporting (Wu et al. Southern and Northern Hemisphere, continents, countries 2014), based on an existing standard agricultural zoning map. at risk, etc. The most popular monitoring or reporting The zones are characterized by three sets of variables: ag- unit is the country since it is consistent with statistical ronomic (such as presence/absence of major crops and their data, economic, and policy conditions, even if inhomo- yield), environmental (such as climate) and seasonality (such geneous for impacting factors. Size also varies a lot from as intra- annual climate or NDVI variability). The paper also small countries such as Singapore to the largest one (Russia, pays attention to within-zone spatial variability in order to more than 9000 times larger than Singapore). identify those where impacts may be inhomogeneous. When designing a reporting unit, adequate indicators and monitoring units are interlinked (Wu et al. 2015). Data Ideal indicators are both systemic and normative (Binder et al. 2010). Systemic: they organically relate to the im- Spatial data pacted systems (reporting unit); normative: they can be used to intercompare different spatial units or the same The CropWatch Mapping and Reporting Units (MRUs) are spatial unit at different times. While impacting factors based on several existing global maps, starting with the may vary a lot spatially, the impacted systems or report- FAO Global Ecological Zones map (GEZ), a map designed ing unit definition can be designed in such a way that for reporting forest and forest change statistics in the ambit they are homogeneous, that is, the same impacting factors of the Forest Resources Assessment (FRA; FAO 1999, 2012, will yield a similar impact over most of the impacted downloadable from GeoNetwork, FAO 2014a). The GEZ systems. In other terms, the vulnerability patterns are map is basically a map of natural vegetation types. expected to be uniform over the monitoring units. Next to a recent Köppen climate map (1976–2000 Impacts, which are the object of monitoring, are brought data; Grieser et al. 2006b), the MRUs were also assessed about by a variety of factors, individually or collectively against global land use and ecosystem maps and especially (Zhang et al. 2013). They include economic conditions (cost the suitability for agriculture maps from the ongoing of oil and fertilizer: Chen et al. 2010; Kelly 2010; Akpan Atlas of the Biosphere project, established in 2002 (SAGE, et al. 2012), policies (Lu 2002; Wiggins and Brooks 2010) 2002). It is also in order to mention several global prod- but mostly, directly or indirectly, weather (Gornall et al. ucts primarily developed for climate change impact studies 2010; Reynolds 2010; Iizumi et al. 2014; Leblois et al. 2014) (Ramankutty et al. 2002; Monfreda et al. 2008, 2009), and agronomy, especially mechanization, inputs, improved as well as most of the global zoning schemes described cultivars and management (Loyce et al. 2012; Yu et al. 2012; by van Wart et al. (2013), and others (including an Zhang et al. 2013; George 2014b; Rozbicki et al. 2015). The array of current and potential land use maps available economic and policy factors follow country units in a sta- in digital form as grids or polygons from FAO, 2014a). tistical sense; weather and agronomic factors don’t: they For China, the standard ecological zones of Sun (1994; mostly follow agroecological zones over long time periods. of which an English language version is available in Hu There exists a number of different global classifications of and Zhang 2006) were integrated into the MRU map. agriculture, agricultural land use and agroecological zones. Some of them focus on the environment of specific crops, Impact factors such as the wheat mega- environments developed by the CGIAR institutes (Hodson and White 2007). Others put more weight Impact factors are classified into three categories for char- on functional, social or environmental aspects (management, actering MRUs: agronomic (such as presence/absence of biodiversity), exemplified by the vast literature on agroeco- major crops and their yield), environmental (such as systems (Doré et al. 2011). All agroecological zones work is climate) and seasonality (such as intra- annual climate or based on the implicit or explicit assumption that environmental NDVI variability). Impacting factor values and statistics variables and the human activities (such as vegetation and are referred to whole MRUs (e.g., wheat yield in the agriculture) that depend on them are correlated and behave Pampas), even if they “exist” only in cropped areas. All coherently. The link between them is strong and they con- variables as well as their sources and spatial resolution stitute a “complex”: natural vegetation directly links to dominant (“pixel size”) are succinctly described in Table 1. crops, which in turn often constitute, of have constituted the basis of agro- economic zones (Sombroek and Gommes 1996; Agronomic variables Haberle and Chepstow-Lusty 2000; Gommes et al. 2004). This paper examines the validity of the concept and the Agronomic variables are prefixed with A, for example, feasibility of global monitoring units for identifying and A1: Arland%, average of the percentage of each pixel that © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. 41 Spatial Units for Global Environmental Monitoring R. Gommes et al. Table 1. List of variables used to characterize MRUs. Type: A, agronomic; E: Environmental; S, seasonality. As a rule, the values in the table are spatial averages over all grid points inside the MRUs. Period: the period to which the data refer (“Current” means that the data are continually updated). “Pixel” is the grid size (in km) of the original rasters provided by the respective Sources. Type Name Unit Definition Pixel Period Source A1 Arland% 0–100 Percent of arable land per pixel 23 Current IIASA A2 Irr% 0–100 Percent of pixel area equipped for irrigation 8 1990–2013 GMIA A3 Bar% 0–100 Percent of pixels where barley is cultivated 7 Current JRC A4 Baryld t/ha Average barley yield 57 2008–2012 CW- 1 A5 Casyld t/ha Average cassava yield 57 2008–2012 CW- 1 A6 Mz% 0–100 Percent of pixels where maize is cultivated 3 Current JRC A7 Mzyd t/ha Average maize yield 57 2008–2012 CW- 1 A8 Pot% 0–100 Percent of pixels where potato is cultivated 7 Current JRC A9 Potyld t/ha Average potato yield 57 2008–2012 CW- 1 A10 Rc% 0–100 Percent of pixels where rice is cultivated 7 Current JRC A11 Padyld t/ha Average paddy yield 57 2008–2012 CW- 1 A12 Soy% 0–100 Percent of pixels where soybean is cultivated Current JRC A13 Soyyld t/ha Average soybean yield 57 2008–2012 CW- 1 A14 Wh% 0–100 Percent of pixels where wheat is cultivated 7 Current JRC A15 Whyld t/ha Average wheat yield 57 2008–2012 CW- 1 E1 Area 1000 km² Area in thousands of Km² n.a. n.a. CW- 2 E2 Z m a.s.l. Altitude 16 n.a. Worldclim E3 Rain mm Annual rainfall total in mm 16 1950–2000 Worldclim E4 Tavg °C Average annual temperature 16 1961–1990 Climond E5 Rady W/m Annual mean radiation total 16 1961–1990 Climond E6 NDVIavg NDVI Average annual NDVI (2 + 5 + 8 + 11)/4 21 1999–2012 VITO E7 Avm mmH O/m Easily available soil moisture 24 n.a. Geo- 1 E8 NPP gDM/(m year) Net Primary production potential 24 1976–2000 Grieser S1 Rn0510 0–1 Fraction of annual precipitation that falls from May to October 16 1950–2000 CW- 3 S2 T02- 08 °C Difference between average February and average August 23 1950–2000 CW- 3 temperature S3 TCV CV Temperature seasonality, coefficient of variation in monthly 16 1961–1990 Climond temperature S4 Tamp °C Temperature of warmest week minus temperature of coldest week 16 1961–1990 Climond S5 NDVI02- 08 NDVI (no unit) Average difference between February and August NDVI 21 1999–2012 CW- 4 S6 NDVIStDev NDVI (no unit) Variability of NDVI over time (standard deviation) 8 1981–2003 Geo- 2 Data sources. Climond: Kriticos et al. (2012), and Xu and Hutchinson (2011, 2013); CW-1: see text under 3.1; CW- 2: computed based on the MRU polygon using QGIS 2.2.0, http://www.qgis.org/en/docs/; CW- 3: computed based on WorldClim grids; CW-4: computed based on VITO (2014); GMIA: Global Map of Irrigation Agriculture, FAO (2013) or Siebert et al. (2013); IIASA: Geo- Wiki (2014); Geo- 1: http://data.fao.org/ map?entryId=61946540-bdbf-11db-a0f6-000d939bc5d8; Geo-2: http://www.fao.org/geonetwork/srv/en/metadata.show?id=37059; Grieser: Grieser et al. (2006a); JRC: data made available by the EC/JRC MARS unit http://mars.jrc.ec.europa.eu, based on the methodology described by Vancutsem et al. (2013); VITO: VITO (2014); Worldclim: Hijmans et al. (2005). MRU, monitoring and reporting units; CV, coefficient of variation. is cultivated; Irr%, average of the percentage of each pixel country centroids and interpolating them using inverse that is equipped for irrigation; Bar%, percentage of pixels distance weighting (exponent 2) to a coarse 57 km grid, in the MRU where barley is cultivated; Baryld, average as shown in Table 1. This is, admittedly, a crude method yield achieved for barley in the pixels where the crop is when applied to large areas with a low density of sample cultivated; Casyld, average yield achieved for cassava in points, but deemed acceptable to estimate 5- year reference the pixels where the crop is cultivated. Similar variables “zonal yields.” The gridded yields were subsequently used are defined for maize (Mz% and Mzyd), white potatoes to derive the average and the standard deviation by crop (Pot%, Potyld), soybean (Soy%, Soyyld), and wheat (Wh%, over each MRU. Values were retained only for the MRUs Whyld). For rice, the percentage of pixels in the MRU where the crop is actually grown, according to the JRC the crop is cultivated (Rc%) is complemented by yield crop masks (Vancutsem et al. 2013). (Padyld) expressed as paddy (husked grain). In order to compare agronomic performance, a simple Average MRU yields were computed by assigning aver- “yield index” (YldInd) was derived as follows: for each age 2008–2012 yield from FAOSTAT (FAO 2014b) to of the seven crops, yields were ranked from 1 to 52 42 © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. R. Gommes et al. Spatial Units for Global Environmental Monitoring (potatoes, which occur in 52 MRUs out of 65), 1 to 21 1 (highest variability), providing a variability/seasonality (both cassava and barley, which occur in 21 Crop MRUs) indicator (VSI). etc.; the highest rank was assigned the highest value. Ranks were then divided by the number of occurrences, resulting Method in the highest value becoming 1 and the lowest 1 divided by the number of occurrences. The final step was the Generation of MRUs calculation of the average across the seven crops for each MRU, ignoring crops that do not occur. In the case of The spatial units used for global crop monitoring and New Zealand (only potatoes and barley), only two values reporting by CropWatch are referred to as Crop were averaged. Monitoring and reporting units (MRUs). They are essentially a simplification of the FAO GEZ map with additional adjustments made based on the recent Environmental variables 1976–2000 Köppen climate map by Grieser et al. (2006b). Table 1 lists eight environmental variables (E as prefix), The following four “rules” were broadly followed. First, of which the first (E1, area of the MRU) is conven- a rather low level of detail (equivalent to merging GEZs) tionally assigned to this group. The seven variables E2 was adopted for minor- or nonagricultural areas, such to E8 thus include altitude (Z), total annual rainfall as the Central Northern Andes (MRU 21 in the (Rain), average annual temperature (Tavg), total annual map of Fig. 1) which actually cover the whole spectrum radiation (Rady), average annual NDVI, easily available from hyperarid climate (BW in the Atacama desert) soil moisture, and net primary production potential to polar (E) at the highest elevation. Second, inclusions (NPP). The pixel count, spatial average values, and which are “small” relative to the size of the MRU spatial standard deviation were directly extracted from were ignored, for instance the Alps in “non- the original grids, using a Mollweide projection for the Mediterranean western Europe” (MRU 60) or the semi- area. arid area (BS) on the border between Karnataka and Andhra Pradesh states in “Southern Asia” (MRU 45). Third, the borders of the GEZs were modified where Seasonality variables the map by Grieser and colleagues showed similar cli- There are six seasonality variables (S-variables): Rn0510, mate conditions (due to recent shifts in climate) across percentage of rainfall falling during the May to October GEZ borders. This usually involves climate warming semester (in combination with latitude, the variable indicates in high latitude or high altitude areas, typically the whether the MRUs enjoys winter or summer rainfall); T02- north- east “ appendix” of eastern- central Asia (MRU 08, the difference between average February and August 52) or the Andean part of the semiarid southern cone average monthly temperature; TCV, the coefficient of vari- (MRU 28). Fourth, some MRUs are “residual” areas, ation of 12 monthly average temperature values; Tamp, the that is, very inhomogeneous hilly terrain where the difference of temperature between the warmest week and “inhomogeneity” is actual a defining feature, typically coldest week (annual thermal amplitude); NDVI02-08, the the area between the Black Sea and the Caspian difference between February and August the average NDVI (Caucasus, MRU 29) and the Pamir area (MRU 30). and finally NDVIStDev, and the standard deviation of aver- In China, the authors adopted the ecological zones of age annual NDVI over time. NDVIStDev is a particularly Sun (1994) which are closer to the new Köppen map relevant variable as it is the only one that assesses variability than the GEZ map. The GEZ was found to be very over time, that is, a major component of farming risk. largely compatible with the cereal suitability map (FAO In order to derive a generic measure of environmental 2007) and the map of major crop types which is part variability, some of the S-variables were modified in such of FAO (2010). a way that high values express high variability, regardless Agricultural and nonagricultural areas were included of the hemisphere where the MRU occurs. For instance, in order to provide global coverage and to include some the percentage of rainfall during May to October was agriculturally marginal areas, such as low-rainfall rangeland taken as the absolute departure from 50%, as 50% cor- at the edge of deserts. responds to well distributed rainfall over the year. Variables derived as the difference between February and August Characterization of MRUs values were taken as the absolute value to make northern and southern hemisphere comparable. The values were The 65 MRUs are identified by a code from M01 to then ranked and the average rank was rescaled in order M65. The naming is conventional and was chosen only to provide a variable between 0 (lowest variability) and for easy reference (Fig. 1). ESRI shapefiles as well as the © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. 43 Spatial Units for Global Environmental Monitoring R. Gommes et al. Figure 1. Spatial extent of 65 Crop MRUs designed for global crop monitoring, together with their numbers and conventional “long” names. The matrix of the coefficients of variation values of the variables for individual MRUs are available as supporting information (Tables S1 and S2) as well as In addition to average MRU values mentioned above, the from GeoNetwork (FAO 2014a). They can also be re- spatial standard deviation for each variable was computed quested from the corresponding author. over the MRU polygons, and the spatial coefficient of variation (CV) was computed. As the CV is a standard- ized statistic, it can be used for intervariable and inter- Principal components MRU comparisons. The matrices of the CV of MRUs × variables (15 A- The environmental factors used to characterize MRUs variables, 6 S- variables and 8 E-variables of which E1 is (Table 1) are redundant because they are correlated. For area and E2 is elevation) were analyzed. As the matrix instance, temperature is linked to elevation, and radiation of A- variables is incomplete (not all crops are cultivated is linked to rainfall through cloudiness. Principal com- everywhere), a new matrix was prepared including the ponents are therefore used below in several figures (starting coefficients of spatial variation of A1, E2 to E8, S1 to S6 with Fig. 3) to characterize the MRUs based on a limited as well as the averages of A- variables (hereafter referred number of synthetic factors. 44 © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. R. Gommes et al. Spatial Units for Global Environmental Monitoring Table 2. “Top 20” of MRUs for several agronomic variables. Number of crops: number of crops between barley, cassava, maize, potatoes, rice, soybean, wheat that are cultivated in the MRU; YldInd, Yield index (unit less, defined under 2.2.1), percentage of arable land and percentage of irrigation. No. of crops cultivated Yield Arable land Irrigation No. MRU YldInd YldInd MRU Arland% MRU YldInd Irr% MRU YldInd 1 7 M26 0.75 1.00 M56 100 M34 0.58 35 M34 0.58 2 6 M04 0.20 0.96 M60 100 M45 0.45 25 M48 0.40 3 6 M23 0.50 0.94 M46 99 M12 0.91 18 M44 0.24 4 6 M45 0.45 0.91 M12 99 M37 0.54 17 M37 0.54 5 5 M01 0.11 0.84 M15 99 M40 0.47 17 M45 0.45 6 5 M03 0.15 0.81 M28 99 M59 0.78 13 M42 0.58 7 5 M05 0.09 0.81 M61 98 M33 0.53 9 M16 0.77 8 5 M07 0.68 0.80 M18 98 M41 0.57 9 M36 0.65 9 5 M09 0.21 0.78 M59 98 M60 0.96 9 M46 0.94 10 5 M12 0.91 0.77 M16 97 M17 0.65 9 M50 0.51 11 5 M14 0.72 0.75 M26 97 M29 0.47 8 M30 0.58 12 5 M16 0.77 0.74 M58 96 M14 0.72 8 M40 0.47 13 5 M17 0.65 0.72 M14 96 M36 0.65 7 M29 0.47 14 5 M21 0.46 0.71 M27 96 M48 0.40 7 M33 0.53 15 5 M25 0.61 0.70 M13 90 M26 0.75 7 M59 0.78 16 5 M29 0.47 0.68 M54 89 M03 0.15 6 M20 0.06 17 5 M34 0.58 0.68 M07 88 M50 0.51 6 M38 0.49 18 5 M37 0.54 0.65 M36 87 M20 0.06 6 M41 0.57 19 5 M38 0.49 0.65 M17 87 M44 0.24 5 M07 0.68 20 5 M40 0.47 0.64 M57 86 M46 0.94 5 M14 0.72 MRU, monitoring and reporting units; YldInd, yield index; varA, A- variables; varE, E- variables; varS, S- variables; NPP, net primary production. to as varA), E- variables (varE) and S- variables (varS), The largest MRU is about 300 times larger than the small- resulting in a 65 × 18 matrix. est ones. M53 (“North Australia”) also includes the southern fringe of maritime South-east Asia; M59 (“Mediterranean Results Europe and Turkey”) includes all of the Iberian Peninsula (including the Atlantic coast) and “Southern Africa” (M09) Map of MRUs reaches as far as the coastal areas of southern Kenya. Figure 1 shows the extent and distribution of the 65 MRUs. The largest MRU correspond to mostly nonagri- Characteristics of crops in MRUs cultural areas such as the deserts (M63, Australian desert; M64, old world deserts from the Sahara to the Afghan Table 2 summarizes some of the main features about the desert), boreal forest (tundra, taiga and permafrost areas: crops grown in the various MRUs. It focuses on the seven M51, Eastern Siberia; M57, boreal Eurasia and M65, sub- major crops in terms of worldwide production and trade arctic America - which includes Greenland and Iceland). if sugar and oil crops are excluded: barley, cassava, maize, There are some other minor instances of transcontinental potatoes, rice, soybean and wheat. It is also stressed that zones, although they are not easily visible in the map the conclusions are only partly true in areas where other above: M53, North Australia also includes the southern crops play an important part (e.g., coarse grains such as fringe of the Western province in Papua New Guinea oats, or sunflower and oil palm, etc.). Additionally, the and areas in Indonesia and Timor Leste at the same lati- spatial variability inside MRUs could also depend on the tude, as already mentioned above. farm size, field size (e.g., huge fields in USA, Kazakhstan The smallest zones correspond to the southernmost or Russia), the applied agro- techniques (mixed dry farm- islands of China (Hainan, M33; China Taiwan, M42). ing and irrigation) etc. The next smallest area corresponds to Western Cape in Only one zone (M26, the humid Pampas) grows all South Africa (M10, a Mediterranean climate “enclave”); seven crops, while three grow six: M04, the Horn of variables S1 and S6 show that seasons are inverted com- Africa; M23, Central-eastern Brazil and M45, southern pared with the surrounding C9, Southern Africa MRU. Asia, including mainly India. The areas are identified by © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. 45 Spatial Units for Global Environmental Monitoring R. Gommes et al. Figure 2. Number of crops (among barley, cassava, maize, potatoes, rice, soybean, wheat) that are gown in various areas. Seven crops occur in the Pampas (M26) but the number is not shown in the legend because the locations are not visible at the scale of the map. the presence of ecosystems at relatively high elevations northern Great Plains (M12). Yields are mostly modest which can accommodate both tropical and temperate crops, (in the YldInd range from 0.5 to 0.6) with the exception sometimes with relatively low yields by world standards of the northern Mediterranean and Turkey (YldInd = 0.78) (e.g., the Horn of Africa). and Western Europe (0.96). A number of regions grow five crops, covering the The observation about modest yields also applies to yield spectrum from low (M5, north and central the main irrigated areas, with values of 0.58 in Huang Madagascar; M1, equatorial central Africa and M3, the Huaihai (M34, 35% of the area equipped for irrigation) gulf of Guinea) to high (M12, American northern Great and 0.40 in Punjab to Gujarat (M48, 25% irrigated). Plains). Figure 2 shows the distribution of the number of crops with additional detail. It is striking how the Environmental features of MRUs zones mentioned are all characterized by one or two dominant “background” crops with patches of ecologically Based on the correlations between them, the seven envi- diverse areas where additional crops occur. ronmental variables from E2 to E8 (refer to Table 1) can The highest yields occur in New Zealand (M56), non- very clearly be subdivided into three groups of factors, Mediterranean Western Europe (M60), Southern Japan corresponding to the three- first principal components. They and Korea (M46) and in the northern Great Plains in account for a total of 92% of the variance of the variables, the USA. It is to be stressed that the number of crops distributed among the components as follows: 49%, 28% in M56 is just two crops (barley and potatoes) which and 14%. Figure 3 shows a plot of the production zones benefit from very long days, resulting in long periods of against the two- first principal components. photosynthesis. The remaining areas cultivate between The first component correlates very strongly with av- three and five crops. Lowest yields occur in Equatorial erage NDVI (R = −0.91), NPP (R = −0.95) and total central Africa (M01), the east African highlands (M02), rainfall (R = −0.90). It can thus be interpreted as “un- Madagascar (M05) and the Caribbean (M20). It is stressed productivity” (MLA, 2015) a biological productivity vari- that the listed crops are marginal in some of the areas, able, increasing from right to left. The second component where yams, millets, teff, pigeon peas, plantain, and others PC2 (or “coldness”) is clearly and negatively associated play the greatest role as staple foods. with radiation (Rady; R = −0.93) and with annual aver- Table 2 also indicates how widespread arable land is age temperature (R = −0.71). More surprisingly, PC2 is in the various production zones. In 15 MRUs the arable positively correlated with the soil moisture storage ca- land fraction exceeds 90%. Most of them occur in China pacity (E7). The relation between E7, temperature and and India (including M34, Huang Huaihai, southwest radiation is not direct, but it can be hypothesized that China M41 and M45, southern China), but also in western low temperature is associated slow soil organic matter Europe and the northern Mediterranean (M59 and M60), degradation, as organic matter and other colloids play Sierra Madre in Mexico and the US (M17) and the an important part in the soil moisture storage capacity. 46 © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. R. Gommes et al. Spatial Units for Global Environmental Monitoring Figure 4. Seasonality in the 65 MRUs represented on the plane of the Figure 3. Plots of MRUs (numbers defined in Fig. 1) against the 1st and two- first principal components, where the first captures temperature second principal components extracted from the six environmental and NDVI variability and the second represents rainfall seasonality. The variables E2–E7. The first component (unproductivity) corresponds to colors indicate the percentage of arable land in five categories. the production potential (decreasing from left to right) and the second Additional symbols (arrows) refer to the variability in NDVI over time: up (coldness) captures radiation and temperature (high at the bottom to for the five most variable MRUs, down for the least variable ones and a low at the top). The colors indicate the percentage of arable land in five horizontal bar for average values. categories. The correlation between E7 and the factors which posi- to a variety of factors, including low population densities tively affect soil organic matter production is positive and topography. (NDVI: R = 0.42; NPP: R = 0.24 and Rain: R = 0.28) which is an additional argument in favor of the climatic Seasonality and variability over time in determinants of E7. MRUs An interesting observation refers to elevation, which is the variable most closely correlated with the third principal Based on the six S-variables in Table 1 modified as component (R = 0.73). This is not so surprising since, described under 2.2.3, the 6 × 65 data matrix was given the large geographic coverage of the production subjected to a principal components analysis and the zones; rainfall and temperature are not correlated with MRUs were plotted against the two-first components altitude, as would happen for a spatially more detailed (Fig. 4). The 67.8% of the variance is absorbed by the study. Therefore, the altitude variable is best regarded as first component which correlates with the temperature- a proxy for “type of landscape” as high elevations tend related variables and intra- annual NDVI amplitude (R to be characterized by rugged terrain. between 0.89 and 0.96 for N = 65 MRUs); it can con- Unfavorable temperature and biomass conditions occur veniently be referred to as “temperature variability”. in the ecologically similar M51 (Eastern Siberia) and M65 The second (16.8% of variance) is virtually equivalent (subarctic America), both characterized by little or no to rainfall variability (the absolute value of February agriculture. Very favorable conditions of productivity oc- rain minus August rain, R = 0.99) while the third cur in M24 (Amazon) and in C9 (Southern Africa) in (13.1% of variance) is best correlated with interannual spite of the presence of the semiarid Kalahari “desert”. variability of NDVI, R = 0.62). Together, the three M50 (mainland South-East Asia) and M19 (Central and components account for 97.6% of the variance of the Northern South America) are among the areas where a original variability/seasonality matrix. The above- high fraction of arable land coincides with a high pro- mentioned VSI correlates well with temperature vari- duction potential. It is worth noting that some areas such ability (R = 0.96). It is interesting to observe that, at as M42 (China Taiwan), M05 (northern and central the scale of the MRUs, rainfall seasonality, intra-annual Madagascar) M01 (Equatorial central Africa) do not, at temperature variability and interannual NDVI are largely present, fully exploit their potential, which may be due uncorrelated variables. © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. 47 Spatial Units for Global Environmental Monitoring R. Gommes et al. (A) (B) (C) (D) According to Fig. 4, the MRUs with low intra-annual rainfall variability occur in two types of situations: Figure 5. Within- MRU spatial variability (coefficients of variation) of agronomic (A), environmental (B), seasonality (C) variables, and elevation (D). 48 © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. R. Gommes et al. Spatial Units for Global Environmental Monitoring Figure 6. Plots of the average coefficients of spatial variation against each other. The indicated outliers and extremes are M28 (semiarid southern cone in Latin America), M39 (Qinghai- Tibet in China), M63 (Australian desert) and M65 (subarctic America). rainfall is very low (M64, Sahara; M28, semiarid southern continental and the islands, M49 and M50). Most variable cone) and therefore cannot vary a lot, or rainfall is temperatures are found in the most continental areas of actually high but well distributed over the year (M56, north- east China (M38), central Asia (M52), and eastern New Zealand; M60, non- Mediterranean Western Europe). Siberia (M51). High rainfall seasonality characterizes the West African Sahel (M08), parts of China (North- East China, Spatial variability inside MRUs M38 and Inner Mongolia, M35), South-west Madagascar (M06) etc. While the A- , E- , and S- variables are relevant in describ- As expected, low temperature intra- annual variability ing the potential and actual farming in the CropWatch affects equatorial areas (M19, Central America and northern monitoring zones, spatial variability in the same variables South America; M20, Caribbean and south- east Asia, both needs to be looked at to understand how homogeneous © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. 49 Spatial Units for Global Environmental Monitoring R. Gommes et al. the zones are, which is an important requisite for global general, the spatial variability in yield inside the MRUs monitoring. is low, mostly between 10% and 15%, while the variability A principal components analysis of the matrix of the in pixels cultivated under various crops or irrigated pixels CV shows that no <11 factors (components) are required is very high, usually between 500% and 1000%. to account for 95% of the variance of the matrix. If the It is also stressed that there is virtually no link between original A, E and S variables are added (65 × 38), 18 the spatial variability in the three variable types (Fig. 6, components are needed to reach the same percentage. but also Fig. 5). In fact, the principal components of the VarA is different from varE and varS in the sense that average spatial variability carry 50%, 25%, and 25% of it is based largely on the “presence” (pixel value of 1) the variance. Figure 6 also shows the significant positive and absence (pixel value of 0) of crops. For a given crop, skew affecting all three variability variables. the average thus indicates the fraction of the area of the Among the varA variables, the lowest spatial variability MRU where the crop is cultivated. As a result, the coef- occurs, on average in M33 (11% in Hainan, China) while ficient of variation is directly depends on the mentioned the most heterogeneous area, from an agronomic point fraction. For values between 0.3 and 0.9, CV depends of view, is M65 (2587%, subarctic America). The Sahara directly on the fraction and varies almost linearly from for instance, which belongs to M64 together with the 150% (fraction = 0.3) to 17% (0.9). Arabian Peninsula and hyperarid regions in Iran and The spatial variability in elevation (E2) can probably Afghanistan displays a spatial variability of 668% as it is be regarded as a good yardstick when comparing spatial made up by low- variability desert with some irrigated variability. The lowest values correspond to M39 (Qinghai- pixels, mostly in the Nile valley. Tibet, 19%) and the east African Highlands (M02, 27%) E- variables, as mentioned, display limited variability, while the five highest values from 132% to 146% cor- which results from the definition of the MRUs as envi- respond to M19 (central and northern South America), ronmentally homogeneous areas: the variation covers the M20 (Caribbean), M44 (Southern Himalayas), M49 (mari- range from 19% (M22, Brazilian Nordeste) to 96% (M39, time south- east Asia) and M62 (Ural to Altai Mountains). Qinghai- Tibet in China), for an average of 38%. The As expected, plateaus show low elevation variability while spatial variability in the seasonality variables is larger: 72% MRUs including lowlands and highlands display high on average, from a low value around 10% in M38 (north- values. east China) and M34 (Huang Huaihai) to values between Figure 5 shows how varA, varE and varS vary among 200% and 400% (207% in M49, maritime south-east Asia; the MRUs, in comparison with the spatial variability in 230% in M4, the Horn of Africa and 385% in M28, the elevation, which is the first factor that disturbs zonality. semiarid southern cone in Latin America). The second is a weak E- W gradient that is visible in Fig. 5B (varE) at high northern latitudes. To interpret Discussion the spatial variability in elevation, it is necessary to keep in mind the definition of the coefficient of variation, which Although primarily based on the natural vegetation GEZ explains why highlands such as the East African Highlands map (FAO 2012), the MRUs shows a good agreement with (M2) and Qinghai- Tibet (M39) display relatively low val- the spatial distribution of major crops zones (Fig. 2). Both ues: the average elevation that divides standard deviation natural ecosystems and crops depend on natural resources, is high. On the other hand, the North China Plain (M34, but crops benefit in addition from genetic improvements Huang Huaihai) is at a low elevation, but includes some and management practices that take them to the “limit” higher terrain in the west. of their ecophysiology into environments that would be Western Asia (M31) and the central- northern Andes (M21) marginal without human intervention. This is particularly are among the most spatially inhomogeneous areas, as the visible in the areas where cropping is most intensive in heterogeneity affects all variables. The areas that display low terms of percentage of arable land and yields achieved. variability for several groups of variables can fall in that It is also noted, as expected (Fig. 4), that high potential category for several reasons. Hainan (M33), for instance, areas are also areas of low variability, which may result is a tropical island, but other areas (M34, Huang Huaihai) in high cropping intensities, as in continental and maritime belong there because they are very extensively cropped, SE Asia (M49 and M50) and in Central and northern resulting in exceptionally low values of spatial variability South America (M19). Altogether, the MRUs thus retain (11%) for a variable that is very high in most other the logic that underlies the GEZ, which is not surprising areas. because agriculture remains a major component of the Large differences in coefficient of variation are observed environment (Monfreda et al. 2008) and because arable between the categories of variables, 330%, 38%, and 55% land occupies a large fraction of land in most MRUs on average for varA, varE, and varS, respectively. In (Table 2). In fact, according to recent estimates, as much 50 © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. R. Gommes et al. Spatial Units for Global Environmental Monitoring Figure 7. Plot of average coefficient of variation for agricultural (A for varA), environmental (E for varE) and seasonality (S for varS) variables against MRU area, after removal of mostly nonagricultural near-polar and/or arid areas (M27, M28, M51, M57, M61, M63, M64 and M65). The numbered points correspond to equatorial Africa (1, M01), the Amazon (24, M24), western Asia (29, M31), Gansu-Xinjiang (30, M32), Qinghai- Tibet (37, M39) and maritime south- east Asia (47, M49). Table 3. Comparison of some MRU descriptors as a function of the fraction of arable land. <75%, less than 75% arable (36 MRUs); >75%, more than 75% arable (29 MRUs). R = coefficient of correlation of agricultural variability versus Area, S is the corresponding slope. Definition of variables is as in Table 1. Avg. is average and CV% is the coefficient of variation in % between the respective variables over MRUs belonging to each of the two groups. Reg. stands for the regression of varA against MRU area. Reg. varA varE varS Area Arland% Irr% Z Rain Avm NPP <75% R = 0.30 Avg. 446 43 63 2610 39 2 894 693 94 80 97 94 83 17 64 S = 0.05 CV% 105 35 114 67 150 >75% R = 0.19 Avg. 185 32 45 1380 91 7 556 1123 100 135 S = 0.06 CV% 91 25 93 71 9 114 86 50 17 33 MRU, monitoring and reporting units; CV, coefficient of variation; NPP, net primary production. as 23.8% of the potential global NPP is directly or in- designed to assess water-related issues. They result from directly managed by man (Haberl et al. 2007). the intersection of 115 geopolitical regions and 126 hy- In addition, natural vegetation and crops are subjected drographic basins and are generally found acceptable for to the same global environmental patterns of not only water- related studies (Kummu et al. 2011). On the other resources but variability as well, where latitude and to- hand, for statistical purposes, the United Nations geo- pography play a major role. In general, a rather large scheme (UN, 2010) adopts just 21 global units, which are number of agroecological classes are required to capture also used by OECD. spatial variations of agriculture with sufficient detail. What In reality, the issue of what amount of variability within does “sufficient” stand for? van Wart et al. (2013) found zones is acceptable is bound to remain subjective as it that, in general, all global zoning approaches that include depends on the intended application. In this study, we less than 100 units retain significant within- unit variability. found that large differences (as measured by the average Indeed, all national agroecological zone studies go into coefficient of variation varA, varE, and varS) are observed significant detail, as amply documented by the national between the categories of variables: 330% for varA, which agro-ecological maps available in GeoNetwork (FAO, refers mostly to crop distribution within MRUs. The value 2014a). To take just one example, for India, the agricultural of varA would drop to significantly lower values if the area of which is smaller than China’s, Pal et al. (2009) calculations were restricted to arable land (0% would be deem that about 130 zones are required for national work, possible only in the case of one single crop). Spatial vari- while they encourage states to add additional level for ability in yields, on the other hand, is in the range of more detailed local studies. In China, on the other hand, 10%–15% (no doubt also partly due to the method used the large zones proposed by Sun are just nine, which are to estimate crop yields by MRU). about the same size as the MRUs adopted for India. The Environmental variables show a much lower range of 281 Food Production Units (FPU) proposed by Nelson spatial variability (average varE is 38%) and so do sea- et al. (2010) and Rosengrant et al. (2012) were originally sonality variables (average varS is 55%). It is stressed that © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. 51 Spatial Units for Global Environmental Monitoring R. Gommes et al. seasonality is expected to be more variable than the en- low; it does not seriously affect the validity of large vironmental variables as it also includes NDVI variability agroecological zones as monitoring as well as reporting over time, as a measure of cropping risk. units. Regardless of the size of the MRUs, it is the pre- Figure 7 shows the plot of the three average spatial vailing environmental conditions that condition vulnerabil- variability indicators against the MRUs’ area, after exclu- ity patterns and the type of farming that can be practiced, sion of major cold and/or dry noncropped areas. an extreme case being irrigated crops in desert areas (e.g., Correlations are all positive, as expected, but significant in the Nile valley, or the semiarid areas in central Asia only for VarA (agricultural variability, R = 0.45, significant or the hills immediately bordering the Amazon). Even if below P = 0.01). As far as environmental and seasonality the number of spatial units were increased, because of conditions are concerned, the MRUs may be considered the fractal nature of landscapes, there would always remain homogeneous, which directly results from the derivation some MRUs that include portions that are transitional of MRUs from the FAO GEZs. to another MRU, and that attract or exclude agriculture In Fig. 7A, the slope of varA against the area of the more than neighboring homogeneous areas. Therefore, for MRUs is 0.08 (the figure includes all but seven nonag- the purpose of operational global monitoring and report- ricultural MRUs). Table 3 shows, among other descriptors ing, a limited but manageable number of spatial units from Table 1, that varA is highest in areas with low defined based on general environmental variables is ac- arable land, contrary to varE and varS, and some other ceptable: the proposed CropWatch MRUs are meaningful variables. In Table 3 the arable land limit of 75% was reporting units at the global scale. chosen because it is close to the median (74%), so that the number of MRUs inside each category is approximately Acknowledgments the same. Area with low varA, coincide with high potential areas (high rainfall, high NPP, widespread irrigation) while The authors specifically acknowledge the financial support other variables, including elevation and soil moisture do through the Special Fund for Grains-scientific Research in not play a part that is commensurate with the role of the Public Interest (Grant No. 201313009-2 and 201413003-7) the factors that directly affect NPP (e.g., rainfall). the National High Technology Research and Development It is also noted that the correlation between varA and Program of China (863 program), Grant No. 2012AA12A307; MRU area is no longer significant (significance threshold and the Visiting Professorships for Senior International for 35 observations is R = 0.32 at P ≤ 0.05, and larger Scientists, Grant No. 2013T1Z0016. The authors thank the for 29 observations as well as for P ≤ 0.01). In general, reviewers for providing thorough comments that have im- areas with high arable land fractions also correspond to proved the quality and readability of the paper. higher average annual temperature and higher average NDVI, although this is not shown in Table 3 (17°C vs. Conflict of Interest 11°C and 0.54 vs. 0.39). On the other hand, the average standard deviation of NDVI over time (S6) and average None declared. annual NDVI amplitude (S5) are virtually identical between References the groups (0.21–0.22). The fact that the CV between MRUs in the groups in Table 3 is comparable (in par- Akpan, S. B., E. J. Udoh, and V. S. Nkanta. 2012. Factors ticular varA: 105% and 91%) indicates that the “ecological” influencing fertilizer use intensity among small holder logic behind the selection of the MRUs remains meaningful crop farmers in Abak Agricultural Zone in Akwa Ibom for agriculture, across different scales. State, Nigeria. J. Biol. Agric. Healthc. 2:54–65. Babu, S. C., and V. J. Quinn. 1994. Food security and nutrition monitoring in Africa: introduction and Conclusions historical background. Food Policy 19:211–217. The current MRU subdivision by CropWatch adopts 65 Binder, C. R., G. Feola, and J. K. Steinberger. 2010. units; it is derived from a GEZ map and used for op- Considering the normative, systemic and procedural erational monitoring, including areas with very limited dimensions in indicator- based sustainability assessments or no crop or livestock production. in agriculture. Environ. Impact Assess. Rev. 30:71–81. Spatial variability in environmental and seasonality vari- Chen, S. T., H. I. Kuo, and C. C. Chen. 2010. Modeling ables inside MRUs only marginally depends on their size, the relationship between the oil price and global food which stresses their intrinsic homogeneity. They are spatially prices. Appl. Energy 87:2517–2525. less homogeneous for agricultural variables. Dalgaard, T., N. J. Hutchings, and J. R. Porter. 2003. This is a direct consequence of the low density of ag- Agroecology, scaling and interdisciplinarity. Agric. Ecosyst. riculture in the areas where the production potential is Environ. 100:39–51. 52 © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. R. Gommes et al. Spatial Units for Global Environmental Monitoring Doré, T., D. Makowski, E. Malézieux, N. Munier-Jolain, M. agricultural productivity in the early twenty- first Tchamitchian, and P. Tittonell. 2011. Facing up to the century. Philos. Trans. R. Soc. Lond. B Biol. Sci. paradigm of ecological intensification in agronomy: 365:2973–2989. revisiting methods, concepts and knowledge. Eur. J. Grieser, J., R. Gommes, S. Cofield, and M. Bernardi, Agron. 34:197–210. 2006a. World maps of climatological net primary FAO. 1999. FRA 2000, A concept and strategy for ecological production of biomass, NPP. FAO, 2005. 5 pp. http:// zoning for the global forest resources assessment 2000. www.juergen-grieser.de/downloads/NetPrimaryProduction/ INTERIM REPORT. FAO, Forestry Dep. Working Paper npp.pdf (accessed September 2015). 20. Rome. 28 pp. Grieser, J., R. Gommes, S. Cofield, and M. Bernardi, 2006b. FAO. 2007. Most suitable cereal. Digital map downloadable New gridded maps of Koeppen’s climate classification. from FAO GeoNetwork web site http://www.fao.org/ Data, methodology and gridded data Available at: http:// geonetwork/srv/en/main.home. Plate 48b in the FAO/ www.fao.org/nr/climpag/globgrids/KC_classification_en.asp. IIASA GAEZ study http://www.fao.org/nr/gaez/en/. Methodology also downloadable from http://www. Methodology in 2002 document jointly published by juergen-grieser.de/downloads/Koeppen-Climatology/ FAO and IIASA and available from http://webarchive. Koeppen_Climatology.pdf (accessed September 2015). iiasa.ac.at/Research/LUC/SAEZ/pdf/gaez2002.pdf (accessed Haberl, H., K. H. Erb, F. Krausmann, V. Gaube, A. September 2015). Bondeau, C. Plutzar, et al. 2007. Quantifying and FAO. 2010. Land Use Systems of the World. Digital map mapping the human appropriation of net primary downloadable from FAO GeoNetwork web site http:// production in earth’s terrestrial ecosystems. Proc. Natl www.fao.org/geonetwork/srv/en/main.home (accessed Acad. Sci. USA 104:12942–12947. September 2015). Haberle, S. G., and A. Chepstow-Lusty. 2000. Can climate FAO. 2012. Global ecological zones for FAO forest influence cultural development? A view through time reporting: 2010 Update. Forest Resources Assessment Environ. Hist., 34:9–369. Working Paper N. 179. 42 pp. FAO, Rome. Hijmans, R. J., S. E. Cameron, J. L. Parra, P. G. Jones, and FAO. 2013. Global map of irrigation areas - version 5. A. Jarvis. 2005. Very high resolution interpolated climate Digital map downloadable from FAO GeoNetwork web surfaces for global land areas. Int. J. Climatol. site http://www.fao.org/geonetwork/srv/en/main.home. 25:1965–1978. The version used was downloaded on 20130901. Hodson, D. P., and J. W. White. 2007. Use of spatial Version 5 was published in October 2013 (accessed analyses for global characterization of wheat- based September 2015). production systems. J. Agric. Sci. 145:115–125. FAO. 2014a. GeoNetwork, a repository of spatial information Hu, Z., and D. Zhang. 2006. China country pasture/forage about world agriculture. http://www.fao.org/geonetwork/ resource profiles. Pp. 63. FAO, Rome. srv/en/main.home (accessed September 2015). Iizumi, T., J. J. Luo, A. J. Challinor, G. Sakurai, M. FAO. 2014b. FOSTAT, the global statistical database of Yokozawa, H. Sakuma, et al. 2014. Impacts of El Niño FAO. Available at: http://faostat3.fao.org/faostat-gateway/ Southern Oscillation on the global yields of major crops. go/to/home/E (accessed September 2015). Nat. Commun. 5:3712. George, T. 2014b. Why crop yields in developing countries Kelly, V. A., 2010. Factors affecting demand for fertilizer in have not kept pace with advances in agronomy. Glob. sub-Saharan Africa. Agriculture and Rural Development Food Sec. 3:49–58. Discussion Paper N. 23. The World Bank, Washington, Geo-Wiki. 2014. Data downloaded from the Geo-wiki USA. 89 pp. project. Available at http://agriculture.geo-wiki.org/login. Kriticos, D. J., B. L. Webber, A. Leriche, N. Ota, I. Macadam, php?menu=results. See also http://www.iiasa.ac.at/web/ J. Bathols, et al. 2012. CliMond: global high- resolution home/research/researchPrograms/ historical and future scenario climate surfaces for EcosystemsServicesandManagement/Geo-Wiki.en.html. bioclimatic modelling. Method Ecol. Evol. 3:53–64. Dataset downloaded from Beta-hybrid.geo-wiki.org on Kummu, M., H. de Moel, P. J. Ward, and O. Varis. 2011. 20130901 (accessed September 2015). How close do we live to water? A global analysis of Gommes, R., J. du Guerny, M. H. Glantz, and L.-N. Hsu, population distance to freshwater bodies. PLoS One 2004. Climate and HIV/AIDS: a hotspots analysis for 6:e20578. early warning rapid response systems. UNDP/FAO/NCAR, Leblois, A., P. Quirion, and B. Sultan. 2014. Price vs. UNDP SE Asia and Development Programme, Bangkok, weather shock hedging for cash crops: ex ante evaluation 20 pp. Available at: http://www.fao.org/forestry/15532-0-0. for cotton producers in Cameroon. Ecol. Econ. pdf (accessed September 2015). 101:67–80. Gornall, J., R. Betts, E. Burke, R. Clark, J. Camp, K. Loyce, C., J.M. Meynard, C. Bouchard, B. Rolland, P. Lonnet, Willett, et al. 2010. Implications of climate change for P. Bataillon, et al. 2012. Growing winter wheat cultivars © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. 53 Spatial Units for Global Environmental Monitoring R. Gommes et al. under different management intensities in France: a Schmid, A. P., 1998. Thesaurus and glossary of early multicriteria assessment based on economic, energetic warning and conflict prevention terms (abridged version). and environmental indicators. Field Crops Res. Synthesis Foundation. Erasmus University, Amsterdam. 125:167–178. Edited for Forum on Early Warning and Early Response Lu, W. C. 2002. Effects of agricultural market policy on (FEWER) by S. B. Anderlini. FEWER Secretariat, London. crop production in China. Food Policy 27:561–573. 30 pp. MLA. 2015. Dictionary.com Unabridged. Random House, Siebert, S., V. Henrich, K. Frenken, and J. Burke. Inc. 22 Jul. 2015. Available at: http://dictionary.reference. 2013. Update of the digital Global Map of com/browse/unproductivity. Irrigation Areas (GMIA) to version 5. 171 pp. Monfreda, C., N. Ramankutty, and J. A. Foley. 2008. Farming Institute of Crop Science and Resource Conservation, the planet: 2. Geographic distribution of crop areas, yields, Rheinische Friedrich-Wilhelms-Universität Bonn, physiological types, and net primary production in the year Bonn, Germany. 2000. Global Biogeochem. Cycles, 22:GB1022. Snodderly, D., 2011. Peace Terms: Glossary of Terms for Monfreda, C., N. Ramankutty, and T. W. Hertel. 2009. Global Conflict Management and Peacebuilding. United States agricultural land use data for climate change analysis. Institute of Peace, Academy for International Conflict Chap. 2 in Economic analysis of land use in global climate Management and Peacebuilding, Washington USA, change policy (33–48) edited by Hertel, T.W., Rose, S., 60 pp. Tol, R. 2009. 348 pp. Routledge Press, London and New Sombroek, W. S., and R. Gommes, 1996. The climate York. ISBN 978-0415619813. change-agriculture conundrum, Pp. 1–14 in F. Bazzaz, W. Nelson, G. C., R. W. Rosegrant, A. Palazzo, I. Gray, C. Sombroek, eds. Global climate change and agricultural Ingersoll, R. Robertson, et al. , 2010. Food Security, production. 345 pp. FAO and John Wiley, & Sons, Farming, and Climate Change to 2050: Scenarios, Results, Chichester, UK. ISBN 0 471 96927 3. Policy Options. International Food Policy Research Sun, He., 1994. Agricultural natural resources and regional Institute IFPRI, Washington, D.C. 131 pp. Available at: development of China. Jiangsu Science and Technology http://www.ifpri.org/sites/default/files/publications/rr172.pdf Press, Nanjing. (in Chinese) (accessed September 2015). UN. 2010. United Nations Statistics Division. Composition OECD. 2007. Glossary of statistical term. 863 pp. OECD, of Macro-Geographical (Continental) Regions, Paris. Geographical Sub-Regions, and Selected Economic and Pal, D. K., D. K. Mandal, T. Bhattacharyya, C. Mandal, and Other Groupings. Available at: http://millenniumindicators. D. Sarkar. 2009. Revisiting the agro- ecological zones for un.org/unsd/methods/m49/m49regin.htm (accessed crop evaluation. Indian J. Genet. Plant Breed. September 2015). 69:315–318. Vancutsem, C., E. Marinho, F. Kayitakire, L. See, and Ramankutty, N., J. A. Foley, J. Norman, and K. McSweeney. S. Fritz. 2013. Harmonizing and combining existing 2002. The global distribution of cultivable lands: current land cover/land use datasets for cropland area patterns and sensitivity to possible climate change. Glob. monitoring at the African continental scale. Remote Ecol. Biogeogr. 11:377–392. Sens. 5:19–41. Reynolds, M. P., ed. 2010. Climate change and crop VITO. 2014. NDVI Based on average 1999-2012 production. CABI climate change series 1. 292 pp. CABI, monthly SPOT VEGETATION NDVI, downloadable Wallingford, UK. from https://earth.esa.int/web/guest/pi-community/ Rosengrant, M. W., and the Impact development team. apply-for-data and http://www.vito-eodata.be/PDF/portal/ 2012. International Model for Policy Analysis of Application.html#Home. (accessed in September 2015). Agricultural Commodities and Trade (IMPACT). Model van Wart, J., L. G. van Bussel, J. Wolf, R. Licker, P. Description. 50 pp. International Food Policy Research Grassini, A. Nelson, et al. 2013. Use of agro- climatic Institute IFPRI, Washington, DC. Available at: http:// zones to upscale simulated crop yield potential. Field ebrary.ifpri.org/cdm/ref/collection/p15738coll2/id/12735 Crops Res. 143:44–55. (accessed September 2015). Wiggins, S., and J. Brooks. 2010. The use of input subsidies Rozbicki, J., A. Cegli, D. Gozdowski, M. Jakubczak, G. in developing countries. 22 pp. OECD, Paris. Cacak-Pietrzak, W. Madry, et al. 2015. Influence of the Wu, B., J. Meng, Q. Li, N. Yan, X. Du, and M. Zhang. cultivar, environment and management on the grain yield 2014. Remote sensing- based global crop monitoring: and bread- making quality in winter wheat. J. Cereal Sci. experiences with China’s CropWatch system. Int. J. 61:126–132. Digital Earth 7:113–137. SAGE. 2002. Available at: and http://library.mcmaster.ca/ Wu, B., R. Gommes, M. Zhang, H. Zeng, N. Yan, maps/geospatial/atlas-biosphere (accessed September 2015). W. Zou, et al. 2015. Global crop monitoring: a 54 © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. R. Gommes et al. Spatial Units for Global Environmental Monitoring satellite- based hierarchical approach. Remote Sens. Supporting Information 7:3907–3933. Additional supporting information may be found in the Xu, T., and M. Hutchinson. 2011. ANUCLIM version online version of this article at the publisher’s web-site. 6.1 user guide. 85 pp. The Australian National Table S1. Spatially averaged values of agronomic vari- University, Fenner School of Environment and ables by MRU. A1 Arland%, A2 Irr%, A3 Bar%, A4 Baryld, Society, Canberra. A5 Casyld, A6 Mz%, A7 Mzyd, A8 Pot%, A9 Potyld, A10 Xu, T., and M. F. Hutchinson. 2013. New developments and Rc%, A11 Padyld, A12 Soy%, A13 Soyyld, A14 Wh%, A15 applications in the ANUCLIM spatial climatic and Whyld. Refer to table 1 for additional details of variable bioclimatic modelling package. Environ. Model. Softw. definitions. 40:267–279. Table S2. Spatially averaged values for environmental Yu, Y., Y. Huang, and W. Zhang. 2012. Changes in rice and seasonality/time variability for each PSZ. E1 Area, E2 yields in China since 1980 associated with cultivar elevation, E3 Rain, E4 Tavg, E5 Rady, E6 NDVIavg, E7 improvement, climate and crop management. Field Crops Avm, E8 Npp, S1 rn0510%, S2 T02-08, S3 bio04, S4 bio07, Res. 136:65–75. Zhang, X., S. Wang, H. Sun, S. Chen, L. Shao, and X. Liu. S5 dvi2-8, S6 NDVIStDev and VSI, the variability/season- 2013. Contribution of cultivar, fertilizer and weather ality indicators which varies from 0 to 1 (least variable to yield variation of winter wheat over three decades: a to most variable MRU). Refer to table 1 for additional case study in the North China Plain. Eur. J. Agron. details of variable definitions. 50:52–59. Data S1. ESRI shapefiles for MRU polygons. © 2015 The Authors. Food and Energy Security published by John Wiley & Sons Ltd. and the Association of Applied Biologists. 55

Journal

Food and Energy Security – Wiley

Published: Feb 1, 2016

Keywords: ; ; ; ;

There are no references for this article.