Global maps of land cover derived from satellite-based Earth observations have existed for almost two decades and represent one of the most important sources of baseline terrestrial information to answer these questions. However there are often discrepancies in the information that they provide because of factors such as the use of different satellite sensors, various classification methodologies and the lack of sufficient in situ data to train, calibrate and validate land-cover maps.
In the past decade a number of global datasets have been developed to classify land cover. These are based primarily on two remote-sensing products – GlobCover and MODIS v5. While these have provided improved resolution and independently appear to have a moderate degree of accuracy – 78.4% for MODIS (+/–1.3%) and 67.1% for GlobCover – the remote-sensing community believes that this can be improved to at least 85% accuracy.
A study published in Environmental Research Letters demonstrates critical differences between GlobCover, MODIS v5 and other widely used land-cover products, and suggests options for reducing their spatial disagreement (that is, improving accuracy). This difference is particularly significant for cropland where, for example, 360 Mha are identified as cropland in GlobCover but as non-cropland in MODIS, a discrepancy that equates to approximately 20% of the global cropland area.
The researchers argue that caution must be applied in the choice of mapping product used, specifically the sensitivity of the product within a specific application and region. Comparison with high-resolution ground data or aerial photographs can improve confidence in a particular product. Another option is to use geo.wiki.org, a global land-cover validation tool that can visualize land-cover products and disagreements in the data through a Google Earth interface. Such a tool can help the user gain insight into which product is better in a given region or application.
The study was undertaken by scientists at the International Institute for Applied Systems Analysis in Vienna, in collaboration with the Department of Remote Sensing and Landscape Information Systems at the University of Freiburg, Germany, and the Institute for Environment and Sustainability, JRC of the European Commission, Italy. The team consisted of Steffen Fritz, Linda See, Ian McCallum, Christian Schill, Michael Obersteiner, Marijn van der Velde, Hannes Boettcher, Petr Havlík and Frédéric Achard.