Edzer Pebesma presenting in Geography Seminar

Prof. Edzer Pebesma from University of M√ľnster in Germany is currently spending part of his sabbatical at CARSI. He’ll present some of his research in the Geography Seminar on Monday, October 5 at 5:35PM in the Hunter College Geography meeting room:

On generating spatio-temporal data

In order to gain knowledge, we (scientists) often generate new
data products from primary observation data, and disseminate what
we did by (i) text, e.g. in the form of a scientific publication,
and (ii) publishing the input and output data and computational
procedures used, e.g. an R script along with software version
information. Although this may be sufficient to understand, reproduce
and verify the work, it is still too difficult to assess whether
two independently generated products can be compared, or whether
method Y can be meaningfully applied to data set X. And how can we
discover all datasets generated by procedure Z, or advertise the
one we just generated?

We [1] introduce a generative algebra for spatio-temporal information.
Using functions on the basic types Space, Time, Quality, and
Entity, we construct data generation procedures (fields, lattices,
point patterns, objects, events, trajectories) that can be executed
to generate actual data. Data derivation operations (the algebra)
are used for generating new data types, e.g. for generating objects
from fields, fields from objects, or lattices from trajectories.
As opposed to data which is always discrete, data generation
procedures can be continuous. They also make the data support
explicit, i.e. whether values refer to points/time instances or to
areas/time intervals. In contrast to the when/where/what questions
usually addressed by semantics, we believe that the algebra can be
used to describe the why and how questions, and as such describe
data provenance.

[1] joint work with Simon Scheider, Benedikt Graeler and Christoph
Stasch; see http://www.meaningfulspatialstatistics.org/