Peter Davidson & Anabelle Spinoulas

1. MANAGING MODELS IN THE
AGE OF OPEN DATA

2. Key topics to cover
 The use of spatial databases
 Underlying principles
 Model data – infrastructure networks +network
options
 The 4S model
 Conclusions

3. Why does this matter?
 We are no longer the custodians of data – we are more like
curators (collate and contextualise)
 Need to be able to incorporate updates
 Hierarchy of models – share across levels
 Community of modellers – share between modellers and
platforms
 Importance of auditing, licensing and change management
 Less tedious more fun!

4. GENERAL DATA MANAGEMENT

5. How Data is Stored – Traditional
 Developed without reference to
recent developments in data
management
 Stored in proprietary formats
 File based data/consolidated data
bank
 Data stored outside of the model
(e.g. GIS) – often deeply nested
folders of files on a network share

6. How Data is Stored - Improved
 Well established approach – Relational Database
Management System (RDBMS)
 Commercial and open source packages
 Large data sets, spatial
 Standardised access and analysis – SQL
 Integrate with other systems (GIS, stats, custom)
 Shared access, security and access control

7. Guiding principles
 Robustness principle – Fuzzy not brittle
 Be conservative in what you do, be liberal in what you accept from
others
 Separate data and processing
 BAD: Excel, GOOD: Database Queries
 Data normalisation – never repeat data
 [Every] non-key [attribute] must provide a fact about the key, the whole
key, and nothing but the key
 Unified data – Everything goes in the database
 Metadata – Data about data
 Source, context, limitations

8. NETWORKS

9. Data Sources
 Govt street centreline data
 Freely available but limited
 Commercial products
 Full routing information
 License issues with derivative works
 Crowd sourced (OpenStreetMap)
 Road networks, points of interest, commercial centres,
schools, airports, parking and many other elements
 Good quality – but some missing/inconsistent
 Can fix errors/omissions

10. Network Geometry and Connectivity
Traditional approach:
 Series of links and nodes
 Anode, Bnode and fixed number of attributes
 Semi-automatic/semi-manual process that creates a new
stand-alone artifact
Weaknesses:
 Cannot distinguish defaults from overrides
 Breaks links to original data sources
 Hard to bring in update to external sources
 Difficult to unify changes (node number conflicts)

11. The Goal
 No manual processing in network creation
 Repeatable, automatic process
 Share process not all data
 Fast enough to run every time
 “Fuzzy" enough that it can still work even if there are
changes to the underlying spatial data
 No node numbers!

12. Creating a network from GIS layers
 Two ways of viewing a network
 Geographically (polylines in GIS)
 Topologically (links + nodes in transport model)
 Conversion between these views
 Network connectivity from spatial join
 Cannot use exact coincident points
 sensitive to minor changes
 Not too fuzzy or else incorrect topology

13. Network Connection Points

14. Adding more detailed information
 Need to add detail to source data (lanes,
capacities)
 Common approaches – both break connection
 Edit source data
 Make model network and then edit

15. Our Approach – “Link Transitions”
A point with a
bearing (unit vector)
Specify start or end
of an attribute
change

16. Directional Points - Link Transitions
 Bearing allows direction
 Better identification when position data is ambiguous -
location + bearing eliminates most ambiguities
 Remaining problems can be identified and solved through
more careful coding
 Works with named roads – consistent with the way that we
think about roads
 Robust when network changes – coordinates, added or
removed links

17. Link Transitions

18. Option Coding
o Option Links
o Option
Connection
Points
o Option Link
Transitions
o Option Nodes
o All have
OptionCode
o Scenarios have
hierarchical sets
of OptionCodes

19. THE 4S MODEL

20. 4S
Structure
Stochastic:
● Monte Carlo methods to draw
values from probability
distributions
● Random variable parameters
● Number of slices can be
varied
SIMULTANEOUS
Segmented:
● Comprehensive
breakdown of travel
markets (20 private + 40
CV segments)
● Behavioural parameters
vary by market segment
EXPLICIT RANDOM UTILITY
Slice:
● Takes slices of the travel
market
○ across model area
○ through probability
distributions
● Very efficient – detailed
networks, large models
Simulation:
● Uses state-machine with
very flexible transition rules
● Simulates all aspects of
travel choice
● Complex public transport
● Multimodal freight
● Easily extended

21. Key features of 4S model
 No matrices, no skims, no zones, no centroid connectors
 All travel is from node to node
 Models constructed with MUCH less manual effort
 Include all roads, all paths, timetabled transit
 Population and employment from multiple sources
 Multimodal with all modes assigned
 Continuous time and simultaneous choice
 Easily include any demand based effects and capacity constraints (not
just roads and transit)
 Much more detailed outputs (volumes by purpose)

22. Australia wide model
All roads except local streets
Some timetabled PT
Walk/cycle
Commercial vehicles
Runs in under 2 hrs (500k links, 400k nodes)

23. Detailed Australia model
All roads
Some timetabled PT
Walk/cycle
Commercial vehicles
Runs in under 8 hrs (2m links (2way), 1.5m nodes)

24. NSW

25. Central Sydney

26. ACT

27. Hobart

28. Orange, NSW

29. Great Britain
Excluding
residential
streets
864k Links
293,000 km
3:19 hrs

30. California
All Roads and
paths
1.9m Links
509,000 km
316,000 mi
8:44 hrs