1. Looking Forward Mike Goodchild

2. Where is ESRI going? 9.0 –massively expanded toolbox –script management and metadata –Python, JScript, Perl –visual modeling interface 9.1 –transportation and routing –many improvements to modeling

3. Towards an infrastructure for sharing models Infrastructure for sharing –search –discovery –evaluation of fitness for use –acquisition –execution

4. Falling through the cracks Text-sharing infrastructure –libraries, bookstores, books, journals, WWW, search engines Data-sharing infrastructure –metadata schema, archives, clearinghouses, data centers Model-sharing infrastructure –models are the highest form of sharable knowledge of the Earth system

5. Current status Some archives –some pre-WWW No standards No clearinghouses www.ncgia.ucsb.edu/~scott

7. The locations of computing User location u –the user interface Processing location p –||u-p|| 1960s < 10m dedicated lines ca 1970 <10km now no limit Data storage location d –independent of u, p Subject location s –independent of u, p, d

8. Options for p Where to process? –server or client, which server? –published services directories www.geographynetwork.com –description standards UDDI: Universal Description, Discovery and Integration WSDL: Web Service Definition Language

9. p and u ||p-u|| = 0 –computing in the client using local data, ||u-d|| = 0 using remote data ||p-u||>0 –send data to the service from the client –link a remote service to a remote data source, p  u, d  u

10. Costs and benefits More cycles available remotely –integrating and exploiting waste cycles –the Grid –SETI Intellectual property issues –intellectual value of service –risk of dissemination –commercial value Update, versioning issues –distributed service has versioning problems Process coupled to data, well defined

12. High-priority geoservices Geocoding –tied to data, update issue Gazetteer –conversion between general or domain-specific placename and coordinates –geoparsing identification and decoding of placename references in text mapping and associating news stories –queries based on placenames how far is the capital of Belgium from the capital of France? What else, is there a general model?

13. Evaluation of models What determines the value of a model? Excess of benefits over costs Cost of execution –depends on data volume, model complexity Cost of data –depends on spatial resolution

14. Determining benefits Value of improved decision making Model accuracy –an inaccurate model has no value Numbers beat no numbers every time –and a picture is worth a thousand words –and a GIS has both numbers and pictures –and results come out of a computer

15. Sources of error and uncertainty Inadequate spatial resolution –necessary resolution is defined by the process being modeled –how to combine models of different processes with different resolutions? Inadequate temporal resolution Measurement error in the data Error in the parameters

16. Error propagation Determining the effects of errors in input data on the output of modeling –confidence limits on every result The butterfly effect –nonlinear response –the effects of spatial autocorrelation relative accuracy versus absolute accuracy Modeling error in data –with Monte Carlo simulation –a very simple illustrationa very simple illustration

19. Sensitivity analysis Repeat the modeling with various values of parameters –original value + 10% –original value – 10% Observe effects on results –identifying parameters whose values are most critical An example –J.C.J.H. Aerts, M.F. Goodchild, and G.B. M. Heuvelink (2003) Accounting for spatial uncertainty in optimization with spatial decision support systems. Transactions in GIS 7(2): 211–230.

20. Other strategies Hind-casting etc. –run the model backwards in time, and compare to the historical record –start the model at some previous time and replicate the historical record –used to calibrate the rules of urban growth models But no-one can predict the future…

21. Yet more strategies The model is only as good as its conceptual inputs –the rules and data If the model doesn't predict correctly it could be because: –the rules are wrong or incomplete –the data are wrong or have inadequate resolution –the time steps are too long –and there is no way to tell which of these is true –likely they are all true

22. In summary: A model is not a way to find out how the world works –but a way to implement what we know in a convenient, integrated package –a tool for spatial decision support –a link between basic science and decision making