Global irrigated area map (GIAM), derived from remote sensing, for the end of the last millennium

A Global Irrigated Area Map (GIAM) has been produced for the end of the last millennium using multiple satellite sensor, secondary, Google Earth and groundtruth data. The data included: (a) Advanced Very High Resolution Radiometer (AVHRR) 3‐band and Normalized Difference Vegetation Index (NDVI) 10 km monthly time‐series for 1997–1999, (b) Système pour l'Observation de la Terre Vegetation (SPOT VGT) NDVI 1 km monthly time series for 1999, (c) East Anglia University Climate Research Unit (CRU) rainfall 50 km monthly time series for 1961–2000, (d) Global 30 Arc‐Second Elevation Data Set (GTOPO30) 1 km digital elevation data of the World, (e) Japanese Earth Resources Satellite‐1 Synthetic Aperture Radar (JERS‐1 SAR) data for the rain forests during two seasons in 1996 and (f) University of Maryland Global Tree Cover 1 km data for 1992–1993. A single mega‐file data‐cube (MFDC) of the World with 159 layers, akin to hyperspectral data, was composed by re‐sampling different data types into a common 1 km resolution. The MFDC was segmented based on elevation, temperature and precipitation zones. Classification was performed on the segments. Quantitative spectral matching techniques (SMTs) used in hyperspectral data analysis were adopted to group class spectra derived from unsupervised classification and match them with ideal or target spectra. A rigorous class identification and labelling process involved the use of: (a) space–time spiral curve (ST‐SC) plots, (b) brightness–greenness–wetness (BGW) plots, (c) time series NDVI plots, (d) Google Earth very‐high‐resolution imagery (VHRI) ‘zoom‐in views’ in over 11 000 locations, (e) groundtruth data broadly sourced from the degree confluence project (3 864 sample locations) and from the GIAM project (1 790 sample locations), (f) high‐resolution Landsat‐ETM+ Geocover 150 m mosaic of the World and (g) secondary data (e.g. national and global land use and land cover data). Mixed classes were resolved based on decision tree algorithms and spatial modelling, and when that did not work, the problem class was used to mask and re‐classify the MDFC, and the class identification and labelling protocol repeated. The sub‐pixel area (SPA) calculations were performed by multiplying full‐pixel areas (FPAs) with irrigated area fractions (IAFs) for every class. A 28 class GIAM was produced and the area statistics reported as: (a) annualized irrigated areas (AIAs), which consider intensity of irrigation (i.e. sum of irrigated areas from different seasons in a year plus continuous year‐round irrigation or gross irrigated areas), and (b) total area available for irrigation (TAAI), which does not consider intensity of irrigation (i.e. irrigated areas at any given point of time plus the areas left fallow but ‘equipped for irrigation’ at the same point of time or net irrigated areas). The AIA of the World at the end of the last millennium was 467 million hectares (Mha), which is sum of the non‐overlapping areas of: (a) 252 Mha from season one, (b) 174 Mha from season two and (c) 41 Mha from continuous year‐round crops. The TAAI at the end of the last millennium was 399 Mha. The distribution of irrigated areas is highly skewed amongst continents and countries. Asia accounts for 79% (370 Mha) of all AIAs, followed by Europe (7%) and North America (7%). Three continents, South America (4%), Africa (2%) and Australia (1%), have a very low proportion of the global irrigation. The GIAM had an accuracy of 79–91%, with errors of omission not exceeding 21%, and the errors of commission not exceeding 23%. The GIAM statistics were also compared with: (a) the United Nations Food and Agricultural Organization (FAO) and University of Frankfurt (UF) derived irrigated areas and (b) national census data for India. The relationships and causes of differences are discussed in detail. The GIAM products are made available through a web portal (http://www.iwmigiam.org).