Movement Spaces and Movement Traces - Computational Movement Analysis - page 22

to compute the same basic movement parameters (trip distance and duration, mean
speed; Shamoun-Baranes et al. 2012 ). So, whereas there is still surprisingly little work
on the implications of choices of conceptual models and data structures, issues related
to data quality and granularity are often investigated, especially in application-driven
movement analysis research fields such as ecology or transportation research, where
careful conduct with data emerges from a long tradition of empirical research.
Scale . Whereas data-driven and theoretical research areas tend to simply accept a
given spatial or temporal granularity, in problem-driven research areas, most promi-
nently in movement ecology, it is widely acknowledged that the observed movement
signal varies with different observation and analysis scales (Nams 2005 ; Fryxell et al.
2008 ). Since often the data representing the different scales derived from rather dif-
ferent data capture procedures (as for example Fryxell et al. 2008 , in VHS, GPS and
tracks in the fresh snow), it remains difficult to separate multi-scale effects of used
methods from differences of the data capturing techniques. Some exceptions explic-
itly performed multi-scale analysis like that of Laube and Purves ( P13 . 2011 ) aiming
at isolating methodological effects. For example, Nams ( 2005 ) derives fractal dimen-
sion D (as a measure for sinuosity, or tortuosity, as it is often termed in behavioral
ecology) for the same trajectories at various spatial scales. The hypothesis under-
lying this research states that animals express different movement behaviors (e.g.,
tortuous foraging at fine spatial scales but directed advances at coarse spatial scales)
at different spatial scale sections (so called “domains”), which are identified through
cross-scale analysis. Similar work studied the influence of the sampling regime on
the computation of the home ranges (Borger et al. 2006 ) and the computation of a
straightness index (Postlethwaite et al. 2013 ). Such work is important but remains
difficult since obtaining the required fine spatio-temporal granularities still is difficult
and costly.
Uncertainty . It is widely accepted that uncertainty is an inherent property of
movement data (Andrienko et al. 2008 ). Giannotti and Pedreschi ( 2008 ) list mea-
surement error and unavoidable discrete sampling regimes as two major sources of
imperfection for movement data. It is however interesting to observe that only some
areas concerned with movement analysis address this uncertainty while others prefer
to largely neglect it. Most theoretical work studying trajectories mainly as geomet-
ric features defines in their preliminaries that fix locations are perfectly known and
accurate (for example, Gudmundsson et al. 2007 ; Benkert et al. 2008 ). However, it is
known that especially in urban areas GPS can be inaccurate, having an effect on the
actual analysis task. For example, imperfection of tracking data had implications for
the segmentation and travel mode allocation in NYC (Gong et al. 2012 ) and for the
size of movement pattern clusters in pedestrian movement in a recreational appli-
cation (Moreira et al. 2010 ). Whereas such methodological research remains rare
in the GIScience and computational geometry fields, application areas with urgent
applied research questions such as for example, again, movement ecology, have a
much stronger interest in the implications of imperfect data on the actual outcomes
of the analytical process. For instance, Hurford ( 2009 ) shows in an experimental
piece the emergence of systematic bias when computing turning angle due to GPS
measurement error. Similarly, Jerde and Visscher ( 2005 ) use Monte Carlo simulation