Overview and State-of-the-Art of Uncertainty Visualization - Scientific Visualization: Uncertainty, Multifield, Biomedical, and Scalable Visualization - page 5

Fig. 1.1 Sources of uncertainty. Both sampling and modeling uncertainties affect each other and
add to visualization uncertainties
be trusted. Data sets where missing or incomplete instances provide too little infor-
mation present challenges to many visualization methods. Filtering out data with
missing elements can ignore valuable information and produce awkward holes. Fill-
ing in missing values or instances by interpolation, imputation, or other estimation
techniques from known values can introduce error. In such cases, data quality metrics
might indicate the confidence in estimated quantities. For instance, estimating a sin-
gle missing data value from a dense set of similar instances would be expected to
produce a smaller error than an estimation from a sparser or more disparate set.
Data sets where multiple, contradictory measurements seem to provide too much
data also offer challenges for visualization. Such situations can be caused by noisy
data, noisy instruments, human error in the data gathering process, or sampling at
a scale different than that natural to the phenomenon. One special case of error in
data measurements is that of spatial data where the error might be in the position of
a sampled location, rather than in its measured values, resulting in uncertainty about
where values should be displayed. Similarly, data with contradictory values might
be characterized by data quality metrics based on sample value range, variance, or
another measure of variability. Finally, metadata about a data source may cast doubt
on its certainty. For instance, data that is old, from an untrusted source, or gathered
through a nonstandard process might be regarded with some skepticism (Fig. 1.1 ).
1.1.1.2 Models Containing Uncertainty
Sophisticated computational models may contain elements designed to estimate the
uncertainty or variability in the model predictions. The sources of this type of uncer-
tainty include residual variability from simplifying abstractions, variability in the
mechanism or magnitude of causality and relationships, potential error in model
inputs, incorrect model parameters, and imprecision in tacit knowledge incorporated
in the model. The range of predictions made by model ensembles, where differ-