Information Technology Reference
In-Depth Information
world (Worboys and Duckham
2004
). Uncertainty may arise because of
uncertain
specifications
. It may, for instance, not be entirely clear what we refer to when we say
a “daily trajectory”. When exactly does this start and end? Do we exclude stops? If so,
how long should stops be such that we exclude them? Second, measuring the accurate
and precise location of a moving object is difficult, hence resulting in
uncertain
measurements
. Uncertainty arises from the necessity that moving objects must be
sampled at discrete times—what happens in between remains uncertain. Third, in
most cases we will want to derive information from our raw location measurements
resulting in
uncertain transformations
.
Laube and Purves (
P13
.
2011
) showed that such uncertainty of GPS data should
not be neglected, especially when investigating movement at fine spatio-temporal
granularities. During that study initially aiming to discover multi-scale effects when
computing movement descriptors, it became obvious that what was assumed to
be “raw” GPS data in the first place was indeed smoothed by algorithmic post-
processing, adding positional uncertainty to the fixes. The study consequently
extended its focus and developed a methodological framework for giving an indica-
tion of those temporal scales for which the influence of uncertainty was less important
than the actual signal, the characteristics of the trajectory. To that end Monte Carlo
Simulation was used to model the uncertainty of the fixes. Each fix was assigned
an uncertainty sampled from a bivariate standard deviation before the movement
parameters were recalculated. T-tests then indicated at what scales and with what
uncertainties the found distributions for original and MC-simulated descriptors were
significantly different. This methodology—another GIScience classic—revealed for
the example studied in the paper, that for an uncertainty of 1 m speed is reliably
computed for temporal scales of 60 s and greater.
Imfeld et al. (
P1
.
2006
) illustrate the importance of quality control of raw data in a
field experiment. The study goes at great length to get an idea of the positional accu-
racy of movement and movement context data. The field experiment also showed
that specifically point in polygon tests for linking fixes to the environment are sen-
sitive to both inaccuracies in the location data but also the context information. This
article indicated that the implications of inaccurate data may depend on the task at
hand. If the goal is aggregation (for example the production of a density map) then
positional error is not such an issue. However, when point in polygon links are used,
the influence of the error may be larger, depending on the analysis scale.
2.4 Related Work
This section summarizes selected related work complementing and completing the
discussion of the topics covered in this chapter. The chapter then concludes with
insights and lessons learned from both the research covered in this chapter and from
the related work.
Conceptual models for movement and movement spaces.
Nathan et al. (
2008
)
present an often cited conceptual framework aiming at a unifying generic theory in
