OPUS Upgrades

This page discusses updates to Urbansim that are discussed within the Sustaincity project. Most of these changes are to be on the python side. It still needs to be discussed in which priority these updates will be treated.

To be done by UCB

Here is a list of desirable improvements for OPUS. It may be completed by all partners, and more importantly a priority list should be established. Indeed, it is little likely that UCB will be able to proceed to all these upgrades (unless receiving substantial help from skillful programmers).

Improved demographic model [status: "option 1" see below; "option 2": seems that there is now (feb/2012) is now an OPUS implementation funded by a different project]

There are two options which should be discussed:

1. Creating an OPUS module that reads the results of an external, pre-computed demographic simulation.
   • Option 1A: This module would reproduce in OPUS the results of the simulation (which would likely take the form of a listing of events occuring at the household level), and also use them to drive the ensuing residential relocation decisions.
   • Option 1B: Create aggregate distributions from these external models and use these to form the control totals for UrbanSim?.
2. Use or refine the demographic models already implemented as rate-based models in OPUS (see DemographicProcessModels), with transition rates to be provided by the INED and the Bocconi University. Simple rate-based models for most demographic events have now been implemented in OPUS, and a matching model remains to be added. If these are insufficient, logit-based models could be easily created instead. If yet more sophisticated models are desired, they could be done as option 1.

While option 1 is currently the chosen one, option 2 holds various advantages:

• Initial versions of many of these models have already been implemented in OPUS and will be accessible from the GUI
• all demographic submodels will be able to use OPUS data structures,
• avoids computational burden of passing the data between UrbanSim? and an external model
• it will be easy to have spatial differentiation for the same submodels
• improvements will be reusable in the future
• coherence

Option 1 on the other hand offers INED better control over the demographic model and its outputs, and is a reasonable goal considering resources allowed to INED (according to INED). Important: Person id's in UrbanSim? must not be re-used (e.g. by re-using id's from emigrated persons for new born or immigrated persons). Use each person id only once.

Status (apr/2012, as I (kn) understood during skype conference):

• original data comes as csv. Converted to hdf.
• pytables is used to read the data efficiently since otherwise it would be too slow (several GB of input data; normal i/o would take more than 20min)
• pandas (?) is used to merge the external data with the pre-existing data inside urbansim.

From the conceptual side:

• looks like the external data only contains "working" or "not working", nothing about, say, part-time
• there may be a problem if there is an excess of "working" persons in the demographic data over the employment totals. Urbansim will set these people to "working without location". In matsim, they should be ignored (like "on vacation").

High resolution point-to-point generalized travel cost [status: it is planned that matsim will deliver link-based travel times]

A data structure inside urbansim that takes link travel times rather than zone2zone impedances as input, and from that generates high resolution point2point travel costs "on the fly".

• Some initial requested access measures (from Kay Axhausen):
• home to work
• home to education
• home to the closest store (beyond a certain size)
• home to previous home
• home to nearest motorway exit
• home to closest transit stop
• home to closest long-distance bus station / train station / airport

This would, as a possible extension(!), also allow the computation of configurable accessibility numbers.

Firmographics [status: ???]

Firmographics involve 4 submodels

1. birth and death of firms/establishments
2. change in the size of each establishment
3. relocation decision for existing firms
4. location choice of new and relocating firms

Submodels 3 and 4 already exist but have not been tested extensively. Such is not the case for submodels 1 and 2, but these could be easily derived from existing models provided that they remain simple (econometrically speaking).

Nested models [status: ???]

A 2-tier nested model has been partly implemented (some issue regarding the sampling of alternatives at the lower level still needs to be addressed). It could be completed in a reasonable amount of time. Having more tiers would involve far more work though (according to the person that developed this module).

Suggestion of Nathalie: the multi-level nested logit model could be implemented in a conditional rather than simultaneous form, which would only need programming the inclusive value computation.

New: Collective decision-making in households [status: ???]

1-person-hh: Provide OPUS template that uses something like "traveltime to job" as decision variable

2-person-hh: Provide OPUS template that uses something like "traveltime to job for person1" and "traveltime to job for person2" as decision variable

Make clear in template how to distinguish husband and wife.

To be done by others

Metropolis as a travel model [status: ???]

Work done by ENS Cachan (initially NicolasCoulombel and NavidKhamedi?, now MohammadSaifuzzaman? under the supervision of AndreDepalma?). They write:

"We are using Metropolis as a travel model for Paris scenario. We have calibrated Metropolis for Ile-de-France. The results are satisfactory. We will soon publish the report of Calibration and other works related to it. It will be a part of the deliverable 7.3: Calibration."

MATSim as a travel model [status: first release fall 2011]

Work done by TU Berlin (ThomasNicolai under the supervision of KaiNagel); see MATSim4Urbansim .

Accessibility indicators from MATSim [status: ongoing research]

Some standard accessibility measures, such as the standard logsum term, will be computed by MATSim and passed to UrbanSim. Check the publications here for results.

Possible to do's

Accessibilities by purpose [status: ongoing research]

Prerequisite: Pass land use info from UrbanSim to MATSim.

Probably not to be done in this project

Spatial econometric models

Preliminary investigation leads PW to believe that allowing for spatial autocorrelation of residuals is feasible (but this point would need further checking if decided to be one of the prioritary items).

Nathalie's pov: the spatial econometric model is not a priority since we can use lagged observed variables as explanatory variables, which would probably do a better job than spatial correlation of the residuals and would need no extra programming.

Simultaneous estimations

Supposedly not an easy thing to do in OPUS (to be confirmed).