GlobeLand30: Operational global land cover mapping and big-data analysis

Information regarding Land cover and change (LCC) over time is essential for a variety of Societal Benefits Areas (SBA), such as environmental change analysis, geographical condition monitoring, urban and rural management, earth surface process modeling, and sustainable development. Since the middle of 1990s, the international scientific communities have devoted tremendous efforts to Global Land Cover (GLC) mapping, and developed a number of coarser resolution (ranging from 300-m to 1 km) data products. As these products could not provide sufficient spatial details and are far from satisfactory for many applications, the Group on Earth Observations (GEO) and some other international organizations called for actions to move towards finer resolution GLC mapping and monitoring in 2010. In order to meet increasing user needs, China launched an operational GLC mapping project and produced a 30-m GLC data product, GlobeLand30, with 10 classes for years 2000 and 2010 (Chen et al., 2015). In September 2014, GlobeLand30 was donated by China to the United Nations for open access and international sharing. It was reported by Nature as “China: Open access to Earth land-cover map” (Chen et al., 2014) and recognized by international experts as “a milestone achievement in the Earth Observation and open geo-information access” (Ban et al., 2015). In order to further report the innovative developments and applications of GlobeLand30, Science China Earth Sciences has published a special issue, entitled “GlobeLand30 remote sensing mapping innovation and big data analysis”, in the end of 2016. An operational finer-resolution GLC mapping aims to deliver high quality data products and is therefore facing a number of significant scientific and technical challenges, such as characterization of complex landscapes with remote sensing and assurance of data product quality. Due to the high spectral heterogeneity within a single land cover class and significant spectral confusion among different classes, it is extremely difficult to achieve satisfactory thematic accuracy with single per-pixel spectral classifier for the entire globe. A Pixel-Object-Knowledge-based (POK-based) approach was developed to produce GlobeLand30 by integrating pixel-based classification, object-based processing and knowledge-based interactive verification (Chen et al., 2015). With the POK approach, the omission and commission errors caused by spectral confusion within and among land cover types have been significantly reduced, and an overall classification accuracy of 83% was achieved for GlobeLand30. It has been recognized as “feasible and reliable for global land cover mapping” (Ban et al., 2015). In the special issue of Science China Earth Sciences, the detailed methodology and operational utilization of the POK approach have been presented with two land cover classes, i.e., global cropland (Cao et al., 2016) and artificial surfaces (Chen X H et al., 2016). In addition, some other innovative methods developed for GlobeLand30 were also reported, such as geospatial-knowledge-based verification and improvement (Zhang et al., 2016) and spatial heterogeneitybased adaptive sampling (Chen F et al., 2016).