Sensitivity analysis and uncertainty analysis for vector geographical applications

Olivier Bonin 1
1 IGN / COGIT lab
2-4 avenue Pasteur, F-94165 Saint-Mandé CEDEX
Tel.: + 331 43 98 84 09
olivier.bonin@ign.fr

Abstract
The problem of the quality assessment of results from geographical  applications can be tackled with the help of sensitivity analysis and uncertainty  analysis techniques. Sensitivity analysis studies the relationships between the output and the inputs of an application. Uncertainty analysis aims at quantifying the overall uncertainty associated with the response of an application. Both techniques rely on the description of a geographical application by a numerical model. Uncertainty and sensitivity analyses are very well established techniques, with applications in many fields (nuclear, environmental, etc.). They rely on an array of techniques including Monte-Carlo simulations and ANOVA. However, they can only be performed when the output and input variables are scalar variables. This restriction can be handled when the geographical data is in raster format, but has many consequences for vector data. We discuss this point, and propose methods  to overcome these limitations in case of vector and raster data. Our methods rely on a statistical modeling of uncertainty in the input variables that takes into account correlation, on  the definition of summaries of the results in non-Euclidean spaces, and on simple meta-models linking the summaries of the output to the input variables. We also investigate the use of functional analysis, and high-dimensional model representations.

Keywords: uncertainty analysis, sensitivity analysis, statistical modeling, error propagation

In: Caetano, M. and Painho, M. (eds). Proceedings of the 7th International Symposium on Spatial Accuracy Assessment in Natural Resources and Environmental Sciences, 5 – 7 July 2006, Lisboa, Instituto Geográfico Português

AttachmentSize
Bonin2006accuracy.pdf667.72 KB