The analysis of geospatial information necessarily employs a wide range of approaches, including data mining, simulation and modeling, verification and communication of results. These activities take place along a continuum (figure 1), often beginning with abductive tasks such as hypothesis formation and knowledge construction, through inductive tasks such as classification and learning from examples and then moving to deductive systems that build deterministic models (that are common in geographic infor-mation systems). Ideally, feedback between these stages informs analysts of potential problems or the need for knowledge refinement. However, there is no single package or system that currently facilitates these different types of inference as an integrated whole; users must instead resort to a set of disparate (and often clumsy) programs that are difficult to connect together operationally and that do not engage the domain knowledge of experts in an efficient manner, nor do they support joint work on difficult tasks. This is a serious problem; creativity is stifled by temporally separating observation from hypothesis, by separating data interpretation from data manipulation, and by separating subjective visual forms of information representation from quantitative analysis. Similarly, despite enormous efforts in quantification, many aspects of geographical analysis remain non-axiomatic; it is not possible to deduce all outcomes from known laws. Therefore it is vital to approach geospatial sciences in a manner that encourages the creation or discovery of new knowledge (Baker, 1999). To achieve this we must provide analysts with an environment that encourages the development and testing of new hypotheses concerning the structure of complex systems and that enables joint work on these tasks.

Here, we propose (1) to develop, implement, and assess the impact of an integrated environment for scientific geospatial knowledge construction and analysis that provides tightly integrated visual and computational methods and tools in support of the scientific process. Within any knowledge construction activity using geospatial data, data quality and uncertainty must be dealt with at many levels (e.g., in the selection of data to use and discard, the formation of categories, the assignment of data entities to categories, and the communication of the end result). To meet this need we will (2) develop visualization methods that help assure quality, allow analysts to reduce uncertainty where it exists, and support effective communication of the remaining uncertainty to the user community. In addition, we will (3) develop and implement collaborative visualization methods to enable coordinated work by teams of analysts who bring different expertise and perspectives to dealing with geospatial knowledge construction and application. Support will be provided for same and different place as well as same and different time work.