Geoscience Reference
In-Depth Information
18
Limits to GeoComputation
Linda See
CONTENTS
Abstract .......................................................................................................................................... 417
18.1 Introduction .......................................................................................................................... 417
18.2 Limits of Computational Power ............................................................................................ 418
18.3 Data Limitations ................................................................................................................... 421
18.4 Limits to Predictability and Understanding ......................................................................... 422
18.5 Computation-Led Limits ...................................................................................................... 424
18.6 Uncertainty ........................................................................................................................... 424
18.7 Conclusions ........................................................................................................................... 425
References ...................................................................................................................................... 425
ABSTRACT
When the first GeoComputation (GC) conference took place in 1996, the computational environ-
ment was much more limited. With increased computational power and data storage, we can now
model millions of agents and begin to tackle the challenges associated with big spatial data streams.
In light of the many changes that have taken place since the
GeoComputation
book first appeared
(Openshaw and Abrahart 2000), this chapter considers what the current limitations to GC are right
now. In the first edition, Kirkby (2000) identified a series of limitations that included inherent
predictability and the need to use additional computing power to build more complex models,
estimate uncertainty and improve the process of model calibration and validation, framed within the
context of the earth and environmental sciences. This chapter takes a broader view that encompasses
both physical and human geographical domains. Five main limitations are discussed: computational
power, data-led limitations, limitations in predictability and understanding, computation or artifi-
cial intelligence (AI)-led limits and limitations as a result of uncertainty. Although some of these
limitations represent real barriers, others can simply be viewed as exciting challenges that require
further research within and beyond the field of GC.
18.1 INTRODUCTION
Models are a fundamental component of scientific research with a long history that has gone hand in
hand with the development of the scientific method. The latter approach involves the investigation of
phenomena through observation and measurement, experimentation and the testing of hypotheses
or research questions in order to expand our knowledge about many different phenomena of inter-
est. As part of this process, we develop analytical or numerical models that attempt to represent the
behaviour of these phenomena, where models are essentially simplified constructs of a reality that is
generally characterised by complexity, non-linearity and at times chaotic behaviour. Geographical
models in particular have the added element of space and/or time depending on whether they are
capturing dynamic or static behaviours. Classic examples of early geographical models are gravity
and spatial interaction models, which have been used in the past to capture flows such as those of
commodities, people and information (Wilson 1971; Fotheringham and O'Kelly 1989; Openshaw
1998). These types of models have helped us to better understand reality, for example, what factors
417
