The Squaresville algorithm: optimising Melbourne's buses

Diagram of the proposed layout (via Steve Pemberton)

Last week, RMIT Masters student Steve Pemberton released a paper on optimising Melbourne's bus network by changing from its current "tailor-made" network, where many buses meander through backstreets to ensure most of Melbourne has a one-seat journey to key destinations, to more of an "everywhere to everywhere" grid-based system, which prioritises speed and short waiting times over coverage, and was tested by a computer model. This is a great example of the tension that exists between patronage and coverage in public transport planning, and presents a good starting point for something that could actually be implemented in Melbourne, so let's take a look at it. 

The background to the paper is that Melbourne's bus network is something that evolved messily over many decades, and many routes are not very fast, direct, or frequent. Coverage is quite good - most of Melbourne is within walking distance of a bus stop - but many of those buses are slow, meandering, and only run every half hour or worse. While train and tram passenger numbers have been rising year on year, buses have languished - because while most people are near a bus, only a few are near a good bus. Few people would choose to use a bus like this if they had a better option. 

The Squaresville bus network, with north-south and east-west routes making any trip easy (via PTUA)

Pemberton's paper is based on Paul Mees' theory of "Squaresville", in which buses run to a simple grid, with evenly-spaced north-south and east-west buses running everywhere. This model means most people will need to change buses at some point in their journey, but the two keys to its success are: 1. you only ever need to change once, and 2. this is so efficient that you can run buses very frequently, meaning when you do need to change, you won't have a long wait. Pemberton takes this abstract theory and uses GIS data to model what it could actually look like in Melbourne, using the same amount of resources the government already allocates to buses. 

He ends up with is a grid that is, on average, 1.6km (one mile) square, with buses that run every 12 minutes in the peak and every 15 minutes off-peak - from everywhere, to everywhere. Routes only deviate from this pattern where the street structure doesn't allow for it, and in some cases to enter a train station precinct. Despite increasing the need to change between services, and increasing walking distance, the efficiency gains are so significant that it would reduce average journey times by about five minutes. And because it's truly "everywhere to everywhere" and allows for cross-suburb trips to the shops or to meet friends, not just the trip to work, it's the kind of network that not only allows people to leave their car in the driveway, but sell the car entirely. 

The frequent "mile grid" network

But it's not without its costs. It increases the distance people have to walk to their nearest route - rather than most people having a bus within 400m, they'd need to walk up to 800m. People who have a choice are usually willing to walk further for a better service (which is why people will generally walk further to a train station than to a bus stop) so purely from a patronage perspective that's not such a bad thing. But we need to remember that PT must also serve the coverage goal, so for a vulnerable person who would struggle to walk such a long way, this hypothetical network would effectively cut that person off. 

The other significant downside to this hypothetical is that it the improvements to many poor services are in large part achieved by downgrading a small number of good services. This is great for the person who currently only has an hourly bus, but not so great for the uni students trying to get to Monash. There are currently three routes that travel from Huntingdale Station to the Clayton campus, which add up to 26 buses in the morning peak hour - and they STILL have issues with overcrowding. This hypothetical network would see this reduced to a single route that ran every 12 minutes in the peak - just 4 or 5 buses in the peak hour - clearly not enough. This is without even considering that that new east-west route would not just need to handle passengers changing from the trains at Huntingdale, but also passengers changing off the hypothetical north-south route along Clayton Rd, since they'd no longer have the 703 which currently turns the corner and drops them right in the campus. 

There's a high-frequency route looping through farmland which could be eliminated

On the other end of the spectrum, the model also shows high-frequency routes in Werribee South, running through literal farmland with hardly any people. There are several examples like this around Melbourne, so a strict application of this flat-frequency grid system would clearly not be appropriate. 

The final issue is the lack of consideration given to major trip generators. The network sticks fairly rigidly to the arterial grid - railway stations were the only deviation programmed in. While the entire point of this network is to not meander through the backstreets or otherwise deviate too much from the grid, there should have been a few more exceptions to this rule than just the train stations. For example, the grid seems to have one bus running down Dandenong Rd past the front of Chadstone, but the route going down Warrigal Rd doesn't take the small deviation into the Chadstone precinct, and nor does the one going along Waverley Rd (which could provide a very useful connection between Chaddy and East Malvern Station, and tram route 3). Such a big centre for both shopping and jobs would clearly justify these minor deviations from the grid, just as much as railway stations do. 

Existing trams and trains, and proposed bus routes, around Chadstone

I should point out, I don't think these are all necessarily flaws with Pemberton's paper - he seems to have been trying to prove a point about what you could hypothetically do with existing resources, not suggesting that deploying exactly this as-is would meet best practice. It's just a starting point, and he touches on several issues that would need to be addressed in a real-world implementation. 

But it's clear that the purist version of this - with an indiscriminate grid and indiscriminate frequencies - clearly has some flaws. What would it take to get it to a saleable package? 

First things first, find the key trip generators that the old model didn't account for, like shopping centres and universities, and allow buses to deviate to them using similar parameters to the station deviations. Then, take whatever the computer spits out about this, and have a human being go over it with a fine-toothed comb to see if any further changes would be required. These changes would be unlikely to add significant route-kilometres to the overall network, nor add too much to travel times, but would make a huge difference in ensuring those destinations are properly served. 

Some routes should be rerouted to go to shopping centres (instead of just near them)

As soon as he notes the 12 peak/15 off-peak frequencies, Pemberton is quick to point out that this is just what's possible with the existing resources, and that he'd prefer to run them at a 10-minute frequency because this improves the connection times, so it'd be a priority for future funding increases. Which is fair enough, given the point of the exercise. In Peter Parker's review of the paper he points out that a 15-minute frequency does not harmonise well with the 10- and 20-minute frequencies of Melbourne's trains, and says "a 10 minute train intersecting  with a 15 minute bus provides good connections that recur only every 30 minutes." I don't know if I'd really agree with this, though. If you set the first train-bus pair up for a 5-minute connection time (to allow people to get from the platform to the bus stop) then the second one will have a 10-minute connection, and the third one won't have a connection at all. I'd argue that a 10-minute connection is completely fine, which means you effectively have four out of six trains per hour with a decent connection. 

The other aspect of this is that harmonisation is just one aspect of providing good connections. If your buses are running to one station, and their frequencies are harmonised, it's easy to set up a timetable with good connections. But with a grid network like this that will intersect with multiple train lines, if you set the bus up to connect with one line, it's very unlikely it will connect with the others. Plus, the whole Squaresville concept relies on short waits between every service, so if half your routes run every 20 minutes to harmonise with trains, it completely ruins the Squaresville network effect for the buses - waits of up to 19 minutes if you just miss a connection would kill any chance of competing with cars. 

So a mixture of 10- and 20-minute services that harmonise with trains would not, in my view, be a better outcome than consistent 15-minute frequencies. And while I'd certainly want to upgrade all the routes from 15 minutes to 10 if I had unlimited budget, this wouldn't be my highest priority either. 

This quite rural part of the Dandenongs probably doesn't need urban frequencies

My first priority would be to find examples like Werribee South, where the algorithm assumed high-frequency routes would run through farmland, and reduce these to more appropriate frequencies; the routes along Monbulk Rd and Mount Dandenong Tourist Rd are other examples that jump out. Not only are these routes isolated from houses and destinations, they're isolated from the rest of the network, so altering their frequencies doesn't have big flow-on effects. These savings should be reinvested in boosting frequencies on high-demand corridors like the Monash example above, going well beyond every 10 minutes if required. Would the savings you found in semi-rural areas be enough to cover the costs of the high demand corridors? I suspect not, in which case this would require some additional funding to be implemented. 

The other aspect that I think is essential, and which would definitely require additional funding, would be to fill in the grid a bit more, to provide more coverage. As much as I want to drive increased patronage and as much as I'm happy to accept some drop in coverage in order to achieve this, the model in the study just goes too far; the paper indicates that the existing network brings 88.82% of people within 800m of a train station or 400m of a tram or bus stop, while the algorithmically generated network would reduce this to 74.63%. We don't have a properly interactive zoomable map of the new network to play with, but from the high-level comparison tool on RMIT's website, it looks to me as though a lot of this coverage has been lost in fairly central, older, rectilinear suburbs - not just newer subdivisions full of squiggly roads and cul-de-sacs. 

The paper's route structure, with a new route added in red

For example, in Bentleigh East it seems to have north-south buses along Jasper Rd, East Boundary Rd, and Warrigal Rd, leaving gaps of 1.6km and 2km respectively - but there are some reasonably direct north-south paths in between these roads that could be utilised. You could for example fill the 2km gap with a route that ran up and down Poath, Mackie and Bignell Rds, and which could ultimately connect with Southland, Hughesdale Station, and Chaddy. These filler routes wouldn't necessarily need to be high-frequency (although given the shopping centre anchors, this one probably would). The main grid would provide your high-quality services to really drive patronage, and many people would be willing to walk a little further to access it - but this secondary grid would just back that up to provide a lifeline to those who cannot walk that far. Something as frequent as the average bus today would be sufficient to provide that lifeline. 

How many of these filler routes would you need, and how much budget would they use? Hard to be sure. But lots of sections like the inner north already have routes close enough together, so you wouldn't need these filler routes everywhere - you could probably bump the coverage up by 10% with only a handful of new routes. And with direct, efficient paths and modest frequencies, these wouldn't break the bank. 

Public transport has high mode share for trips from suburban Melbourne into the CBD, because that's what the trains and trams do well. Cross-suburb trips are mostly done by car, because our current bus network isn't fit for purpose; clearly we need dramatic change if we want our public transport network to offer the same freedom as the private car. The newly-minted Public Transport Minister, Ben Carroll, has indicated he's keen on substantial bus reform; papers like Pemberton's show that by reallocating the existing resources and by topping that up with a bit of extra funding, he could do truly revolutionary things. Let's hope he does.