A new combined uncertainty and sensitivity assessment of spatial models with object-based features
Rios, J. Fernando
MetadataShow full item record
Uncertainty and sensitivity quantification is not often performed in spatial analysis and modeling applications. One reason is the lack of tools which can handle various kinds of uncertainty associated with spatial data. In particular, positional uncertainty in objects is often ignored. There has been much research in regards to the modeling of different kinds of positional uncertainty (e.g., measurement error, indeterminacy). However, a conceptualization which encompasses them in a single framework has not been developed. This work addresses the issue with a new conceptualization of positional uncertainty, termed F-Objects, along with a Python-based implementation, called Wiggly, which handles both positional and non-positional variables. The novel Wiggly framework brings together probabilistic and fuzzy-set representations of uncertainty, positional and otherwise, under the F-Objects conceptual model using a simulation-based approach. Modularity allows for the future addition of functionality. The framework improves on previous efforts by: proposing a new flexible categorization of uncertainty in objects and an implementation, presenting a combined fuzzy-probabilistic uncertainty assessment method which also includes uncertainty in the position of objects by using fuzzy Monte Carlo simulation, and providing the ability for quantitative sensitivity analysis using a novel fuzzy-probabilistic approach. Using the framework, an analysis of three common GIS operations (linear referencing, point-to-polygon distance, overlay) shows a strong dependence on the geometry of the features. In the case of linear referencing, this is significant since there is no such dependence under certainty. A case study using a GIS-based groundwater contamination model shows how the framework can be applied to determine the uncertainty associated with the contaminant load to a river. This realistic example shows how the tool is able to answer the question ''what is the effect of positional uncertainty in my model''? In this case, is found that non-positional uncertainty dominates, though if that uncertainty can be reduced (e.g., by collecting additional data), it is shown that the river's boundary is slightly more influential than the contaminant source locations.