Important progress has been achieved in global land cover mapping in the past decade. For example, spatial resolution has increased from 300 m to 30 m. The level of classification detail has also improved from a plane level to a two-layered hierarchical classification scheme with 29 classes. However, major challenges emerge in mapping at a fine spatial scale with primarily optical data. This paper introduces the major challenges in mapping croplands, human settlements, water, and wetlands. The challenges in the use of multi-temporal and multi-sensor data, which may be useful in the future applications of remotely sensed data, are also discussed. Some of the on-going efforts to improve the quality of global land cover maps are then summarized. We argue that although harmonizing and integrating various global land cover products may be worthwhile for land cover data developed in the past, existing technologies provide sufficient data for improved map making if extra efforts are exerted. Developing and selecting effective algorithms, as well as several input variables（new types of data or features） for classification, and utilizing representative training samples are among the effective conventional measures for improving mapping accuracies at local scales. Data are more important than algorithms with regard to improving mapping accuracies. Finally, a new paradigm for global land cover mapping is proposed. This new paradigm includes a view of vegetation classes based on their types and form, canopy cover, and height. The appropriate determination of a vegetation class requires complementary information on canopy cover and height that cannot be extracted with classification algorithms. The new paradigm also suggests that a universally applicable training sample set is effective in improving land cover classification at the continental scale of Africa. To ensure an easy transition from traditional land cover mapping to the new paradigm of global land cover mapping, we recommend the creation of an all-in-one data management and analysis system. This system can be used as a foundation for a global land cover mapping portal that links freely accessible cyberspace resources and bridges data users and producers, specialists, and laymen toward a gradually evolving online global land cover mapping system.

remote sensing;image classification;class definition;universal sampling;all-in-one system;on-line mapping