The Future of GeoComputation Ian Turton Centre for Computational Geography University of Leeds
Summary People Data –Space –Time Computing Methods –Explorative –Explanative –Exploitative
The CCG Some of them anyway
Mountains of Data
Swamps of Data
We know what you spend...
…where you spend it...
…who you talk to...
…where you live... What your neighbours are like, what your house is
...Crime data and... crime type crime location insurance data
...Health data environmental data socio-economic data admissions data
The Cray T3D and T3E High Performance Computing Time machines Just big enough for modern geographical problems
The Internet GIS and the Web –Public participation in planning Distributed Computing –“many hands make light work”
What can we do with all this data and computer power? Explore it Explain it Exploit it
Exploration Given some (large amount of) data find anything that is “interesting” in that data
Pattern Analysis GAM GEM Automated analysis Easy to understand output No statistical assumptions crime, health, education...
Spatial Search Agents If we don’t know where to look Look every where? Or let something else do the looking?
Urban Social Structure Glasgow and London
Fourier-Mellin space Glasgow and London
Rezoning Census variables and areas Sales areas Voting districts
Explanation Having found something “interesting” in a data set Attempt to explain it or model it
Spatial Interaction Models Migration flows Commuting flows –GB Ward to Wards flows (10,000) Phone flows –(20+ Million) EU Flows
Cellular Automata Simple CA Life Complex multi-state CA forest fires Pedestrian or traffic movements
Neural Nets Black Box Non-linear parameter free estimations Used any where a “normal” model could be used.
Fuzzy Logic Allows the introduction of imprecision to model More computation gives better answers
Agents on a Ring Catherine Dibble Agents can move along the lines GROW MAKE SERV INFO Generate reasonable patterns
Exploitation Having found something of interest and explained it (in some way) make use of this knowledge
Spatial Location Optimisation Based on spatial interaction model Run the model 1000’s of times In this case 10,000 zones
Flood Forecasting How likely is it to flood in the next 6 hours? Neural nets Fuzzy Logic
Sensitivity Analysis on Models Run the model 1000’s of times with perturbations to inputs Get out real error estimates Population Models Flood Models Drainage Models
Conclusions More data –better data More computing –better computing More models –better models