UCGIS Research Areas
in Error & Uncertainty in Spatial Data

Prepared by
A-Xing Zhu (axing@geography.wisc.edu)
and
Brian Yandell (yandell@stat.wisc.edu)

Modified according to the discussion in the Uncertainty group (uncertainty@solim.geography.wisc.edu)
Dear Colleagues,

After browsing through the UCGIS research priority nominations, we noticed that there is a common concern among many of the member universities about the need for more research in the area of errors and uncertainty in spatial databases. We (at U.W.-Madison) felt that there are three basic sub-research- areas on errors and uncertainty in spatial databases and attempted to group the nominations on errors and uncertainty based on the three subareas. The purpose of this effort is to prompt discussion among the nominators and to facilitate any possible future collaborations.

We are not sure that we have put your nominations in the proper groups. We would appreciate your comments on the designation of the groups and the assignments of your nominations. Any other comments are also welcome.

We have also created an email list for the purpose of discussion. Please send your comments to uncertainty@solim.geography.wisc.edu.

  1. Conceptual Models for Representing Spatial Data
  2. Integrating Spatial Data Across Scales
  3. Error Propagation Through Models

Conceptual Models for Representing Spatial Data

In addition to the instrument (measurement) error, further error can be introduced during the data collection process. This is due in part to the inadequacy of the conceptual model for representing spatial data. Efforts involve developing new models and/or extensions to existing models for adequately capturing and representing spatial data to minimize the discretization. The new models and/or extionsions should include adding variance estimates as layers, to attributes in tables, to entire object classes, to regions, and to entire data sets; adding variograms, correlograms, and other descriptors of joint probabilities; adding misclassification matrices.

Oak Ridge Nat Lab | OH St U | U CA Berkeley | U WI Madison | U N C

Integrating Spatial Data Across Scales

Data in spatial data bases come from a variety of sources (remote sensing, terrain analysis, digitization of existing paper maps, etc.) at multiple scales. These data have undergone different degrees of generalization at both spatial and attributes levels. They may or may not represent the scale at which natural processes occur. Thus integration of these diverse data sets in the environmental analysis process can introduce errors due to their scale incompatibility. A clear conceptual framework defining the detailed relationships between generalization and data quality is needed to guide the development of methodologies (de)generalizing datasets to compatible scales for the purpose of analysis.

U WI Madison | MI St U | U CA Santa Barbara | OH St U

Error Propagation Through Models

Data from different sources contain inherent errors which are propagated when these layers are combined in a modeling process. Error propagation should be an integral part of any environmental modeling activity. This can ensure that results from environmental analysis can be interpreted with a proper understanding of uncertainty. Impact of uncertainty on decision making needs to be assessed as well.

San Diego St U | U MN | Clark U