Keynote Dr. Dan Cornford - Abstract

Uncertainty: the key to interoperable geospatial processing?

Modelling is a ubiquitous activity, however by construction all models are wrong. The degree to which they are wrong is a crucial aspect of a models uncertainty characterisation, both static and dynamic. When used in isolation, with humans making decisions and relating models in an interactive manner this uncertainty is often implicitly taken into account, but in the automated service chaining envisioned in many geospatial applications, built using Web Processing Services, explicitly taking account of uncertainty becomes more critical. Using the example of an interoperable, automated interpolation service, being developed as part of the INTAMAP project, I will outline some of the key aspects of uncertainty that need to be represented. I will argue that while a semantic solution seems attractive, a more realistic solution is through standards. I will introduce UncertML, an XML schema being developed to exchange uncertainty, which currently focuses on a range of probabilistic representations. I will show how this has been used in a sensor web context to realise a probabilistic web processing service that undertakes interpolation with sensors with a range of observation and error characteristics. I will argue that any future developments in geospatial modelling, both static and process based, must be accompanied by quantified uncertainty estimates, and present a framework which permits this to be realised.