The Metro-North Accident and Train Control

Early in the morning on Sunday, a Metro-North train derailed on the Hudson Line, immediately south of the junction with Amtrak’s Empire Connection: maps of the derailment area can be found on the BBC, while The LIRR Today has a map and a diagram with speed limits. Four cars overturned, and four people died while more than 70 others were injured. The train was going at 82 mph (132 km/h) through a tight curve at Spuyten Duyvil with a 30 mph limit; the speed limit on the straight segment before the curve is 75 mph according to Rich E. Green’s map, which may be a few years out of date, and 70 mph according to the first New York Times article about the derailment. The curve radius appears to be 230 meters on Google Earth, putting the lateral acceleration rate at 5.8 m/s^2, minus a small amount of superelevation (at most 0.8 m/s^2, or 125 mm, to perfectly match the centrifugal force at the curve’s speed limit, and likely lower); the cutting edge of tilting trains allows about 2 m/s^2 lateral acceleration (see PDF-p. 2 of this article about the Pendolino), or 300 mm cant deficiency.

Initial reports of a mechanical brake failure seem unfounded: a National Transportation Safety Board briefing mentions that the brakes had functioned properly on brake tests and at previous stops on the journey (starting at 00:40 in the video). The focus is now on human error: the NTSB refused to say this outright, but beginning at 03:00 in its briefing video it trumpets positive train control as something that “could have” prevented the accident. Rick Gallant, who led California’s rail regulatory agency at the time of the 2005 Glendale crash, is also quoted as saying positive train control “probably could have” prevented the accident on NBC. Moreover, the train driver is quoted as having told investigators “he had become dazed before the accident, suffering what his lawyer referred to as ‘highway hypnosis.'” Metro-North’s spokeswoman made the strongest statement: “if the accident was caused by speeding, positive train control would have stopped it.”

It is extremely likely that a robust train control system would have prevented the accident, as it is capable of slowing the train sufficiently before it reaches a speed restriction. The bulk of this post will be dedicated to talking about what train control systems can do. There’s a large array of acronyms, some of which mean different things in different countries, and one of which has two different meanings.

Broadly speaking, train control can prevent two types of dangerous driving: crashing into another train on the same track, and excessive speeding. If the system detects dangerous behavior, it will automatically stop or slow down the train. Driverless trains are based on robust enough systems that are so automated they no longer need the driver. The hard part is having an on-board system figure out whether the train is traveling too close to another train or too fast, which requires communication with the signaling system; automatically slowing the train down is comparatively easy. In nearly all cases, the signals are static and embedded in the track systems, but in a few, usually high-frequency subways rather than mainline rail, the system directly communicates with the train ahead on the same track (this is moving block signaling, or communication-based train control).

It is century-old technology to stop a train that is about to enter a segment of track too close to another train (“signal passed at danger,” or SPAD). A train’s steel wheels close an electric circuit that detects whether there is a train on a block of track, and this communicates to the signals entering this block of track to prohibit trains from proceeding; see diagrams in the moving-block signaling link, which also show how it works in the more common fixed-block setup. A situation that electrically insulates the train from the track is therefore extremely dangerous and may lead to line shutdowns for safety. Any system with the capability to stop a train in such a situation is called automatic train stop, or ATS. The 79 mph speed limit on nearly all passenger train lines in the US comes from a 1947 regulation by the Interstate Commerce Commission (which has since morphed into the FRA) requiring ATS or in-cab signaling at higher speed; the intention was to force the railroads to install ATS by threatening a crippling speed limit, not to actually reduce train speed.

It is much harder to enforce speed limits. ATS systems do not have to enforce speed limits: at Amagasaki, there was an ATS system that would have stopped a train running a stop signal (as it had earlier on the trip), but no protection from excessive speeding, which is what led to the crash. The signaling system needs to be able to communicate both permanent and temporary speed restrictions. It is nontrivial to maintain an up-to-date database of all speed restrictions on an on-board computer, or alternatively communicate many different speeds from wayside track signals to the train’s computer.

In 2008, the FRA mandated positive train control (PTC) as a result of the Chatsworth crash; PTC is a term that doesn’t exist outside North America, and refers to an automatic train control system capable of not just ATS but also enforcement of all speed restrictions. In Europe it is called automatic train protection, or ATP, and in Japan it is called automatic train control, or ATC. It is common in the US to do trackwork on one track of a multiple-track railroad and slap a temporary speed restriction on adjacent track, and enforcing such limits to protect wayside workers is specifically part of PTC.

Because the ATC system requires trainside equipment, a train that travels between different systems will need more equipment, raising its cost. In Europe, with its hodgepodge of national standards, some international trains require 7 different systems, raising locomotive costs by up to 60%. This led to the development of a unified Europe-wide standard, European Train Control System (ETCS), which combined with GSM radio for communication between lineside signals and the train is called European Rail Traffic Management System (ERTMS). The obligatory cost and schedule overruns of any IT project have plagued this system, and led to delays in installing train protection on some lines, which led to a fatal accident in Belgium. However, the agony of the ERTMS project has for the most part already passed, and now there is a wide variety of vendors manufacturing equipment to the specified standards, leading to widespread installations on new and upgraded lines outside Europe. As of September of 2013, ETCS is installed on 68,000 track-km and 9,000 vehicles worldwide.

Although ETCS is an emerging global standard (outside Japan, which has a vast system of domestic ATC with multiple domestic vendors), American agencies forced to install PTC have not used it. California HSR is planning to use ETCS, and Amtrak’s signaling system on much of the Northeast Corridor, Advanced Civil Speed Enforcement System (ACSES), with full implementation on the Northeast Corridor expected by this year, is similar to ETCS but not the same. Elsewhere in the US, systems have been bespoke (e.g. on Caltrain), or based on the lower-capacity systems used by the freight operators.

Metro-North does not have PTC. It has an ATS system that protects against SPAD, but can only enforce one speed limit, the maximum speed on the line (MAS). As the maximum speed on the outer Hudson Line is 90 mph, the system cannot enforce any lower speed, and so the train could travel at 82 mph even in 70 or 75 mph territory, let alone 30 mph territory. More modern systems can enforce several speed limits (e.g. the TGV’s TVM), and the most modern can enforce any speed limit, in 1 km/h or 1 mph increments.

Metro-North and the LIRR have been trying to wrangle their way out of the PTC mandate, saying it offers “marginal benefits”; a year and a half ago, the New York Post used the word “outrageous” to describe the PTC mandate, saying it would cost over a billion dollars and that the money could go to capacity improvements instead, such as station parking. Lobbying on behalf of Metro-North and the LIRR, Senator Charles Schumer made sure to amend a proposed Senate transportation bill to give the railroads waivers until 2018, so that they could devote resources to more rush hour capacity from the outer suburbs (such as Ronkonkoma) to Manhattan and fewer to safety. According to Siemens, the work will actually take until 2019, and Siemens says it “has developed PTC specifically for the North American market,” in other words built a bespoke system instead of ETCS. (ACSES was developed by Alstom.)

Because the systems developed for the US are based on the needs of American freight railroads and perhaps Amtrak, which do not need as much capacity in terms of trains per hour as the busiest commuter lines, they are much lower-capacity than those used in Europe. The LIRR and Metro-North have far busier mainline tracks than any other US commuter rail system with the exception of the inner part of New Jersey Transit, which is equipped with ACSES as part of the Northeast Corridor; to modify the system to their needs raises costs, as per the New York Post article. The MTA released the following statement (see also mirrors on Fox and CBS):

The MTA began work to install Positive Train Control on the Long Island Rail Road and Metro-North Railroad in 2009. To date, the MTA has budgeted nearly $600 million for elements of PTC installation, including a $428 million procurement last month for a system integrator. Full implementation is estimated to cost $900 million, and the MTA will make sure the appropriate funding is made to implement PTC on the most aggressive schedule possible. However, implementing PTC by the 2015 deadline will be very difficult for the MTA as well as for other commuter railroads, as the Federal Railroad Administration (FRA) and the Government Accountability Office (GAO) have both concluded.  Much of the technology is still under development and is untested and unproven for commuter railroads the size and complexity of Metro-North and LIRR, and all of the radio spectrum necessary to operate PTC has not been made available. The MTA will continue its efforts to install PTC as quickly as possible, and will continue to make all prudent and necessary investments to keep its network safe.

Of course, the technology is no longer under development or untested. Just ask the Belgians, the Swiss, the Chinese, the Saudi, or the Taiwanese. Older technologies meeting the definition of PTC exist practically everywhere on mainline trains in the European and Asian first world. Urban commuter lines in Tokyo such as the Tokaido Main Line and the Yamanote Line, each with more ridership than all North American commuter lines combined, are equipped with ATC. The RER A, with slightly less ridership than all North American commuter lines combined, has a train control system providing moving-block signaling capability on the central trunk. A Swiss mainline with 242 passenger and freight trains per day and minimum train spacing of 110 seconds at 200 km/h has ERTMS as its only ATP system, and Switzerland expects to fully equip its network with ERTMS by 2017.

Although the US mainline rail system is freight-primary, with different needs from those of Europe south of Scandinavia (e.g. critical trunk lines are thousands of kilometers long and lie in sparsely-populated territory), the same can’t be said of the Northeastern commuter rail lines, most of which only see a few daily freight trains and are dominated by tidal flows of commuter trains with high traffic density at rush hour. Rush hour traffic levels approaching 20 tph per track are routine, with 24-26 on the Northeast Corridor entering Penn Station from New Jersey. It is incompetent to try to adapt a system developed for long-distance low-cost freight railroads and ignore one developed for busy commuter lines just because it has an E for European in its name.

While most European countries have long implementation timelines coming from a large installed base of good but not top-line legacy signaling, countries with inferior systems sometimes choose to replace their entire signaling systems, as the passenger-primary parts of the US should. Denmark, whose intercity rail far lags that of most peer European countries, decided to replace its signaling system entirely with ERTMS. The projected cost is €3.2 billion, of which €2 billion is for ERTMS on the network, €400 million is for equipping the Copenhagen S-Bahn with CBTC, and €800 million is contingency; the total length of the system is 2,132 route-km and 3,240 track-km.

At a million euros per route-km, exclusive of contingency, Metro-North could install the system on all east-of-Hudson lines, except the New Haven Line, where Amtrak plans to install ACSES, for about $450 million, and the LIRR could install the system on its entire system (including parts currently without any signaling) for about $650 million. Denmark has about 700 trainsets and locomotives to install the system on, in addition to tracks; on the LIRR and Metro-North, those figures are about 150 each, although this assumes that trainsets would be permanently coupled, whereas today they run in married pairs, so that in an eight-car unit there are four cabs where only two are needed. If the LIRR and Metro-North agreed to treat trains as permanently-coupled sets, then the scope of the order would be about 40% of the size of the Danish fleet, consistent with a total cost of about a billion dollars.

This would also allow higher capacity than the current systems, which could squeeze more trains onto busy lines, so it wouldn’t be at the expense of capacity improvements. In particular, the LIRR could keep postponing the $1.5 billion Main Line third track to Hicksville project, and instead run trains on the currently double-track bidirectionally (today they run one-way at rush hour, to accommodate local and express service) using the very high frequency that ETCS permits. Another project, which Sen. Schumer thinks is more important than PTC, a $400 million plan to double-tracking the outer part of the Main Line from Farmingdale to Ronkonkoma, could also be postponed while still providing the necessary capacity.

Although both of the LIRR multi-tracking projects’ cost figures are enormous – the third track is about $100 million per kilometer, almost what a subway in suburbia should cost, and the outer second track is $15 million per km, more reasonable but still very high – adding tracks is in general more expensive than adding signals. IT procurement is expensive and prone to cost overruns, but once the initial system has been developed, the marginal cost of implementing it in new but similar environments is relatively low; ETCS would cost about the same on the LIRR and Metro-North as the MTA plans to spend on signaling, but provides better functionality as it’s compatible with their high traffic density. Organisation vor Elektronik vor Beton.

Of course the first step in the organization before electronics before concrete slogan is improving the state of the organization. In terms of safety, there may be scope for better training, but the train driver according to the NTSB has 10 years’ experience (start at 02:20 in the video) and based on his work schedule would have had enough time to get a full night’s sleep before his shift started (start at 07:25). Since there is no obvious organizational way to further improve safety, electronics is the next step, and this means installing a good PTC system in a timely manner.

However, in terms of cost, there is something to be done. While the MTA claims PTC is too expensive and provides little benefit, Metro-North spent $80 million a year on conductors’ salaries in 2010 (although it’s been going down, to about $65 million by 2012) and the LIRR spent another $95 million (in either 2010 or 2012), both numbers coming from the Empire Center’s SeeThroughNY. About six years’ worth of conductor salaries would pay for full PTC; future savings are free. The NTSB briefing said there were 4 conductors on the train (start at 09:15). The main duty of conductors is to sell, check, and punch tickets, an old-time rail practice that has been abolished in modern commuter railroads throughout the first world.

A commuter train needs between 0 and 1 conductor. Stephen Smith quotes Vukan Vuchic, a professor of transportation engineering at Penn who was involved in the implementation of SEPTA’s through-running in the 1980s, as saying that ticket-punching is “extremely obsolete” and “very 19th century.” A tour of any of the major urban commuter rail systems of Europe will reveal that a few, such as the Paris RER and the London systems, use turnstile, while most use proof-of-payment, in which roving teams of ticket inspectors only check a small proportion of the trains, slapping fines on people caught without a valid ticket. On American light rail lines, which are often similar in role to German commuter rail lines (especially tram-trains) except that they run on dedicated greenfield tracks, this is routine; this can and should extend to commuter mainlines. While the electronics is needed to handle safety, this organizational improvement would pay for the electronics.

Although the investigation seems to be going in a competent manner, the MTA’s position on the relevant issues in general does not come from a position of competence. It is not competent to have this many redundant employees but then cry poverty when it comes to avoiding crashes and derailments. And it is not competent to pretend that there is nothing in Europe or Japan worth using for American signaling systems. The US did not invent PTC – at most, it invented the term for what’s called ATP or ATC elsewhere. It shouldn’t act like it’s the only place in the world that uses it.

Sometimes, Half a Line is as Good as No Line

The perfect is not the enemy of the good when it comes to rail projects. The half-done job is. In a trivial sense it’s obvious that half a tunnel across a mountain is useless. But even partial lines that have some uses are sometimes so much less useful than the full line, that the economic benefits of completing the half line to the full system are actually greater than those of building the first half. In many cases, even partial lines that are very good on their own have relatively easy extensions with very good economics.

This is primarily true for intercity rail, since costs are roughly proportional to route-km whereas benefits (e.g. high-speed rail operating profits) are proportional to passenger-km: once a first-phase rail line is in place, any future phase such that passengers will use the first phase for much of their travel will generate a large amount of passenger traffic relative to infrastructure construction. Probably the simplest example of this is extending California HSR to Sacramento: once a Los Angeles-San Francisco system is in place, especially if the route goes over Altamont Pass, extending to Sacramento requires only about 100 km of additional construction (180 if the LA-SF route is via the currently planned Pacheco Pass route), in flat land, but people would be taking the train from Sacramento to Los Angeles, a distance of about 600 km. Thus, despite generating much lower ridership than San Francisco, Sacramento is a highly beneficial extension of California HSR, once the LA-SF first phase is in place.

There are several more places in North America that are like this. When I tried applying a very primitive ridership model to American city pairs, what I found is that next to the Northeast Corridor, the highest-performing lines are extensions of the Northeast Corridor to the south. This is for the same reason as with Sacramento: once Boston-New York-Washington is in place, an extension to Richmond would generate 540 passenger-km of New York-Richmond travel on just 180 route-km of Washington-Richmond HSR, and thence extensions to Raleigh and Norfolk would be similarly high-performing, and so on. Some of those extensions would add about 40 million passenger-km per route-km of new construction, compared with about 28 million on the Northeast Corridor alone; in other words, assuming constant per-km cost, the rate of return on some of the extensions is higher than on the Northeast Corridor trunk. Similarly, although international HSR links are overrated, once New York-Buffalo is in place, an extension into Toronto becomes high-performing (with about 30 million passenger-km per new route-km after a fudge factor accounting for the underperformance of international city pairs), which is especially useful given that New York-Buffalo’s projected traffic based on said primitive model is marginal.

In those cases, the picture is bright, in that the first phase is strong on its own, and then future phases become natural extensions, which can be funded on the heels of the first phase’s success.Unfortunately, in many cases the situation is different, and the first phase is really a half-built line that isn’t much better than nothing, at least on the proposed merits. For example, High Speed 2’s rising costs are causing the cost-benefit analysis to head well into marginal territory: as per PDF-page 15 of a Parliamentary primer, the benefit-cost ratio of the first phase, London-Birmingham, is now down to 1.4, while this of the full system as proposed by the Cameron administration, going to Manchester and Leeds, is 1.8. Although 1.4 > 1, common practice in Europe is to build only projects with benefit-cost ratios higher than 1.2 or 1.3, because of the risk of further cost escalations (although the stated cost includes a generous contingency factor). The environmental benefits are likewise lopsided in favor of full construction, according to pro-HSR group Greengauge 21: three quarters of the benefits come from the second phase. This is because few people fly from London to Birmingham or Manchester already, since the existing medium-speed trains are fast enough at these distances to outcompete low-cost flights; however, there’s a large volume of people flying from London to Glasgow, and it is expected to take the full opening of HS2 to get enough of those fliers to switch to make a significant difference.

In this case, HS2’s first phase is better than nothing, and the problem stems from extremely high costs: without contingency, London-Birmingham, a distance of about 180 km, is projected to be about $23 billion after PPP conversion, which at nearly $130 million per km is worse than California HSR, which has to tunnel under tall mountain ranges. With contingency, it is $175 million per km, not much less than the projected cost of the majority-underground Chuo Shinkansen maglev. If the costs were brought down to reasonable levels, the first phase alone would be highly beneficial, as can only be expected given the size of London and the secondary cities of the West Coast Main Line.

In some other proposed cases, even the benefits are marginal. Worse, sometimes attempts to cut costs lead to steeper cuts in benefits. The example that motivated this post is a recent story of a proposal for HSR in Colorado, which is not planned to serve the built-up area of Denver at all, but instead stop at the airport. An airport stop without a downtown stop is unacceptable anywhere, especially given Denver’s airport’s large distance from downtown (30 km, vs. 15 km in Shanghai, where most HSR trains stop at the domestic airport and only a few stop downtown at Shanghai Station). It is especially unacceptable given that Denver is to be connected to cities that are within easy driving or medium-speed rail distance: Fort Collins is 100 km north of Denver, Colorado Springs is 110 km south, and Pueblo, which is only proposed as part of a larger second phase together with a Rocky Mountain crossing, is 180 km south. At the distances of Fort Collins and Colorado Springs, the egress time would eat all time advantage of HSR over driving; at the distance of Pueblo, it would eat most of the time advantage. Saving money is nice, but not when it makes the entire project useless except to the occasional Fort Collins- or Colorado Springs-based flier.

One can go further and ask why even build HSR at such short distances. On the Northeast Corridor, full-service HSR is a great investment, because of the combination of extremely thick city pairs at the 360 km mark (New York-Washington and New York-Boston) and one reasonably thick pair at the 720 km mark (Boston-Washington), which is too far for medium-speed rail to compete with air. Philadelphia’s presence boosts the case for HSR – it conveniently provides a source of reverse-peak traffic away from New York and Washington, adds long-distance travelers to Boston, and adds short-distance high-speed travelers to New York – but by itself it’s not worth it to build HSR at the distance of New York-Philadelphia. If Boston and Washington weren’t there, then incremental upgrades with a top speed of 200 km/h or maybe 250 km/h would be best, and higher speeds would just waste money on more expensive trains and create noise pollution and higher energy consumption.

The same analysis is true of faster-than-HSR travel modes. The other motivation for this post, in addition to Colorado’s proposal, is Japan’s attempt to export maglev to the US, proposing the Northeast Corridor as the route to run maglev on, with Baltimore-Washington as the first segment, which Japan proposed to build for free, as a loss leader. Nobody needs maglev from Baltimore to Washington: the egress time is going to ensure the benefits of maglev speeds over HSR speeds are small, and even the benefits of HSR speeds over fast commuter rail speeds are limited. The Chuo Shinkansen is only planned to be about 440 km long, but it’s a capacity boost on a line that already has HSR with extremely high ridership, and not just a speed upgrade. Elsewhere, Japan builds conventional HSR rather than maglev, even for inter-island travel, where people fly today since the Shinkansen takes 5+ hours and flying takes an hour.

Part of my distaste for Hyperloop essentially comes from the same problem: it tries to compete with HSR at a distance where HSR is appropriate and faster trains are not. All of the technical problems of Hyperloop – thermal expansion, claustrophobic vehicles, extreme levels of lateral acceleration – are solvable, at the cost of more money. The technology is feasible; it’s Musk’s order-of-magnitude-too-low cost estimate that I object to. The problem is that at LA-SF distance, access and egress time and security will eat the entire time advantage over conventional HSR, in similar vein to the problem with siting Denver’s HSR station at the airport. Conventional HSR still involves regular trains that can run on electrified legacy lines, so it’s cheap to go the first and last miles within the Bay Area and the LA Basin; maglev doesn’t have this ability and neither do vactrains. Thus there will always be the problem with the first and last mile, which can be solved only by spending even more money – even in the case of the Chuo Shinkansen, JR Central decided that Shinagawa, just outside Central Tokyo, is good enough, and there’s no need to spend further money to get trains into Tokyo Station. But the access, egress, and security time penalties are constant, whereas the time advantage over slower modes of transportation grows with distance.

So by all means, let’s think about maglev from New York to Chicago and Miami and from Los Angeles to Seattle, where HSR is too slow to compete with air travel; let’s think about a vactrain at transcontinental scales, were open-air maglev is too slow. There’s a reason this year’s April Fool’s post emphasized that the vactrain system should be intercontinental and globally connected. I don’t think maglev in the US or a vactrain anywhere pans out in the next few decades, but at least at this greater scale they wouldn’t be crowding out a technology that can succeed, i.e. conventional HSR at the scale of the Northeast Corridor or California.

Sometimes, starting small means failing. A strong first phase with stronger second phases, such as LA-SF or Boston-NY-DC, is likely to become a success and motivate the political system to spend additional money, partly from first-phase profits, on extensions. A weak first phase that needs additional phases to pan out won’t lead to the same extensions. When a white elephant project opens, nobody listens to critics who say it should’ve been built bigger, even in the uncommon cases when those critics are right. Colorado HSR as proposed is going to get faltering ridership, not enough to justify the cost, and cause widespread disaffection even with potentially strong rail projects in Colorado. The same is true of any faster-than-HSR project that tries to replace HSR instead of capitalize on its strength in serving much longer-distance city pairs. If Musk succeeds in causing the median Californian to turn away from HSR and build Hyperloop instead, then first Hyperloop will turn out to cost ten or more times as much as Musk predicted (for which people won’t blame Musk but the government – Musk’s sycophants will tsk-tsk from the sideline and say that if only he had been in charge), and second the ridership won’t cover the costs, leading people to decide that any linear transportation corridor is bad and the government should stick to highways and airports.

The Difference Between Bus and Subway Alignments

Reading design guidelines for bus routes reminds me of how different surface transit is from rapid transit. Buses need to follow straight, wide, two-way roads. Subway trains do not: those roads make construction easier, but it’s normal for train lines to detour and turn, even in rigidly gridded cities like New York. The upshot is that sometimes the optimal route for a bus is different from that of a subway, and this limits the usefulness of preexisting bus routes for subway planning.

For a relatively simple example of this, consider the plans for a subway under Wilshire Boulevard in Los Angeles. The buses follow Wilshire all the way from Downtown to Santa Monica. The trains were never intended to: there’s a short stretch where Wilshire isn’t as important while somewhat off the street lies Century City, and all alignments studied for the Wilshire subway have involved some deviation. The chosen alignment is the one that deviates more than the other, to serve Century City more centrally.

This is relevant specifically to the example of Tel Aviv. When I criticized the Tel Aviv subway route choice for being politically motivated to avoid certain neighborhoods, Alan of Tel Aviv Bus Mappa said,

To minimise cost, the planners looked at what works today (the existing high-demand bus routes) and decided that connecting Petah Tikva and Bat Yam to Tel Aviv was the highest demand corridor. They also looked at what was wide (boulevards and arterials), as their aim was to maximise segregated on-street running. This is also the reason that the plan makes use of the ‘Turkish line’ alignment connecting Jaffa and Allenby rather than the more direct, but narrow, Derech Yafo and Eilat Street.

Central Bus Station would have been a huge diversion for the route and is not a particularly in-demand destination. However, the planned Green line will serve that location.

The problem with this line of thought is that subways are not buses. Subways can use the more direct but narrower alignments if they need to: it may be somewhat more expensive to construct, but there’s no disutility to passengers. A bus running on a narrow street is slowed down, especially if the street is twisty. A subway that can go under private property is not.

Even in New York there are some twists – for examples, the route of the L through Brooklyn, and the route of the 2/3 from the Upper West Side to Harlem. But those twists are not critical, and the city doesn’t really need them. The Wilshire deviation in Los Angeles is also in this category.

It’s ungridded cities where the ability of trains to cut under the street network becomes critical to providing service to major destinations, which may not be anywhere near the wide streets. A look at the inner network of the London Underground will confirm that the lines bear little relationship to the street network, which was built incrementally over the centuries and would not be good at serving the major destinations in the desired directions. In Paris the older lines were built subsurface and do follow streets (which at any case are more rationalized than in London due to heavyhanded central planning), but the newer ones were built deeper and do not.

In Tel Aviv, the problem is that many of the neighborhoods that need public transportation service the most do not have wide streets for buses, or have wide streets configured in the wrong directions. The oldest parts of the city, the Old City of Jaffa and Ajami, have very narrow streets since they predate modern boulevard design by a few centuries. The next oldest – the Jewish neighborhoods of Jaffa, South Tel Aviv, the western parts of Central Tel Aviv, and the Old North – do have wide streets, but often pointing in the wrong direction, for examples nothing serves the Port or the Basel Heights compound, and the east-west streets going through the Old North are very narrow. They have no reason to form a coherent rapid transit network, since they were built as interurban streets or as neighborhood main streets, not as subway alignments. They barely even form a coherent bus network, but the hacks made over the decades to create bus trunk lines are different from the optimal route a subway would follow.

In fact, the recently-reelected Huldai administration has plans to upzone around the central parts of the route to build a new CBD. The area in question, around Begin Road, is unwalkable and almost unfixable to be made pedestrian-friendly, the road is so wide and fast. This is not service to an existing destination that follows a linear corridor as in New York and other strongly gridded cities.

In a city like Tel Aviv – or any other city without a strong grid that influences development – subway planning should start from a list of major destinations and dense residential neighborhoods and their locations on a map. The subway routes should form somewhat straight lines connecting them, with the first line chosen in a way that connects to the most and the most important ones. It’s fine to have somewhat kinked routes – nobody likes riding a C-shaped route, but it’s okay to have small deviations such as the ones proposed for Wilshire or even larger ones such as the one Shanghai’s Line 1 takes to reach People’s Square. The junctions should be the most important destinations, or the ones with the most potential for CBD formation; in Tel Aviv those are generally to the west of the planned CBD, because of the potential for waterfront upzoning and the preexisting density in the neighborhoods south of the Yarkon and west of the Ayalon Freeway.

Buses are of course not planned like this. A city that wants a vigorous bus network needs to do what Los Angeles and Vancouver have done: put the buses on a grid as much as possible, and have them go straight along major roads, with as few deviations as possible. Vancouver’s north-south buses deviate a little bit to serve Downtown, and even those deviations are sometimes questionable since people transfer from the buses to SkyTrain before the buses reach Downtown. The grid allows for an efficient network of transfers, with the transfer penalty reduced by high frequency on the trunk lines. It’s nothing like subway lines, which form a tight bus-like mesh in about one city in the world (Mexico City) and everywhere else have a mesh-like core surrounded by what is an essentially hub-and-spoke system.

Even when the busiest bus routes do indicate something about subway demand, there are exceptions. In New York, the busiest bus lines today are the M15, on First and Second Avenues, and the B46, on Utica Avenue: they are almost even in ridership and have traded places for first place citywide recently. But nobody expects a Utica subway to get the ridership of Second Avenue Subway, even people like myself who believe such a subway is underrated and should be considered in medium-term transit planning. The third busiest route in New York is the Bx12, on Fordham, and I do not know a single transit activist who believes it should be railstituted, even ones who believe other routes with somewhat lower ridership should be railstituted (such as Nostrand, whose bus, the B44, ranks fifth). The issue here is that First and Second are in Manhattan, where bus speeds are so low that ridership is suppressed, as people walk longer distances to the parallel subway or don’t take the trip; if both Second and Utica get subways, the lower amount of congestion in outer Brooklyn is irrelevant and the trains will travel at the same speed, whereas today there are factors working against Second that make the rail bias there higher than on Utica.

Something similar is the case in Tel Aviv. The widest boulevards have the largest concentrations of bus trunk lines, but that’s because they are the only streets on which buses are even remotely feasible as modes of transportation. In Jaffa, Jerusalem Boulevard is wide enough for fast surface transit but Yefet Street is not. Based on Jerusalem’s width, the planners chose to keep trains at-grade on Jerusalem, which they could not do on Yefet. But if trains were underground, Jerusalem’s current advantage would evaporate, leaving Yefet with the advantages of proximity to the Old City and the Flea Market and of higher density.

It is wrong to plan buses as if they were subways or early-20th century streetcars, where frequent twists were not a problem since there were few cars on the road, and where the dominance of the traditional downtown favored a hub-and-spoke network. Recent bus successes in North America have involved discarding those ideas and planning buses based on modern travel needs and modern traffic levels. By the same token, it is wrong to plan subways as if they were buses when they are capable of following alignments that buses cannot.

Who Regional Rail is For

A few rail proposals have happened in the last few months that begin with the concept of improving transit access in the suburbs, and end in a bad direction. These center on airport-oriented rail extension, which in the case of New York means building transit to Newark, JFK, and LaGuardia, as a high priority; consider Chris Christie’s proposal for a PATH link to Newark Airport, and proposals on PDF-pp. 17-18 of Next New York for airport service. Instead of this, let me expound a bit on what the most promising travel markets for regional rail are:

1. The through-running aspect is useful for people whose commute requires them to cross the CBD or go around it. In New York, this means people who live in New Jersey and work in Brooklyn, Queens, the Bronx, or Long Island, or vice versa; and people who live in Westchester and points north, including Connecticut, and work in Brooklyn, Staten Island, possibly Queens or Long Island, or Newark and points south, and again vice versa. None of these travel markets is by itself very large, but some, especially those involving people working in Brooklyn and Queens, are of moderate size and together they’re about 150,000 commuters, about as many as use each of New York’s three commuter rail system at two trips per person. (All numbers are as of 2000 and come from the census.)

2. Additional lines allow travel even on markets that are not really through-running. A Staten Island-Manhattan tunnel is likely to be used primarily by people from Staten Island working in Manhattan or Downtown Brooklyn rather than by suburb-to-suburb commuters. Staten Island itself produces about 80,000 commuters bound toward Manhattan and Brooklyn, and electrification of the Erie Lines and a connection to Lower Manhattan opens up rail service to about 70,000 Manhattan-bound commuters from Bergen and Passaic Counties.

3. As a continuation of point 2, lines laid out in a way that serves secondary CBDs on the way from the suburbs to the primary CBD can produce additional ridership. For example, the LIRR already has some Brooklyn-bound commuters, and New Jersey Transit some Newark-bound ones; the Erie Lines could produce Jersey City-bound commuters, and one of the reasons to build the Lower Manhattan tunnel via Pavonia or Exchange Place rather than Hoboken is to serve the larger secondary CBDs there. Hudson County has about 30,000 workers commuting in from Bergen and Passaic Counties and 50,000 from Essex County and points west and south.

4. High all-day frequency of local trains together with fare integration with local transit allows people living and working within each inner-suburban region to use regional rail to get to work. The urban analog is that Brooklynites who work in Brooklyn often use the subway, and drive mainly if their commute is orthogonal to the Manhattan-bound orientation of the subway lines. Residents of Newark, Yonkers, Elizabeth, Paterson, Mount Vernon, New Rochelle, and Hempstead drive at higher rates than residents of the Outer Boroughs even when the poverty rates are comparable: a transit trip from Elizabeth to Newark today is either a bus that gets stuck in traffic or an expensive train that comes twice every hour off-peak and only stops at Downtown Elizabeth, the airport, and Downtown Newark. In 2000, only 26% of people working in Downtown Newark got there on public transit (see PDF-p. 13 of this report).

Airports are not very significant traffic generators. The AirTrain JFK has 5.5 million annual riders; the average ratio of annual to weekday ridership on the subway is 300 (on commuter rail, which has a more pronounced peak, it’s about 270), so that’s equivalent to about 18,000 weekday riders. The Newark version has 2 million annual riders. Regional rail is a way to build low-cost rapid transit in areas where there already are mainline railroads that can be used for local and regional service. Deviations need very high ridership to be justified. The tunnels through the CBD, such as the central RER and S-Bahn tunnels or the tunnels under Manhattan that I propose, bring in commuters from many suburbs into the primary CBD and also connect multiple secondary CBDs. Greenfield lines used for some airport extensions, such as in Zurich, are justified by their short length, connections to trains from all over Switzerland, and very high traffic (with nearly 50% mode share) coming from the use of the airport’s landside concessions as a shopping destination.

In contrast, an examination of the four above main travel markets suggests specific ways regional rail must be built and operated to maximize its usefulness. Brooklyn is the largest destination in the region outside Manhattan, and this means that tunnels serving it from more directions than just that of Long Island should be a higher priority. Queens is the second largest destination, and this means that commuter trains using the Northeast Corridor should stop there, with easy transfers to Jamaica, Flushing, and Long Island City for trains not serving those destinations; Sunnyside Junction would especially useful for this.

Moreover, travel market #4 is the most underrated. The potential traffic volume dwarfs all others. Newark has about 4,000 workers who live in areas who would be served by through-running, such as Brooklyn and the Bronx. It has 36,000 workers who live in the city itself, 30,000 who live in the rest of Essex County, 17,000 who live in Union County, and another 17,000 who live in points farther south. The Northeast Corridor, North Jersey Coast, and Morris and Essex Lines already exist, but provide expensive, infrequent service, with stations spaced too far apart for walking to the station. Christie’s PATH extension tellingly does not include a stop at South Street, but instead goes nonstop from Newark Penn Station to the Newark Airport train station. It’s of paramount importance to raise the transit mode share on these internal inner-suburban travel markets.

Tokyo’s CBD has about 2 million workers, the same as Downtown and Midtown Manhattan. The reason Tokyo has so much more rail ridership than New York is not a bigger downtown, or better airport service, but better rail service to secondary job centers, which themselves grow around train stations more closely than in New York. But Downtown Brooklyn, parts of Queens, and Downtown Newark at least already have the transit access, both by subway/PATH and by commuter rail. Present-day commuter rail just doesn’t provide good enough service to compete with parking rates and traffic jams outside Manhattan.

More on Vancouver’s Obsession With Filling Buses (Hoisted from Comments)

Via Human Transit, I learn that Translink has a bus service performance summary with an infographic on PDF-page 16 contrasting high- and low-performing routes. As usual, Translink claims that the high-performing routes have strong anchors at their ends as one of the reasons for their success. Unfortunately, the claim is not completely correct, and on top of that the definition of “high-performing” is stretched to make anchored routes look better. In particular, this implicates Vancouver’s strategy of upzoning the most intensely at its southern rim while ignoring its center.

To paraphrase my second comment on Human Transit, the summary rates routes on three metrics: boardings per hour, capacity utilization, and cost per boarding; a high-performing route is one that is in the metro area’s top 25 on all three, and a low-performing one is one that is in the bottom 25 on all three. However, only the first and the third are actually useful for the passenger. The second is a measure of pain – it’s the product of turnover with crowding, and although it can be raised by raising turnover, it can also be raised by making the bus more crowded.

An updated list of Vancouver buses and their productivity measures is available here. Measured by cost per rider and boardings per hour, the unanchored 8 is more productive than the 49, which has anchors but nothing in between. But the 8 ranks 29th in capacity utilization, so it’s penalized. The 5 and the 6, which are very short routes serving the West End, are also penalized solely because of their low crowding levels and their short length, which makes turnover more difficult. The 8 has high turnover (like the 3 and 20, which did make the infographic), so it achieves more passenger boardings per hour but fewer passenger-km despite its weak outer-end anchor, and the 5 and 6 are so short that even passengers riding all the way through provide many boardings per hour relative to capacity utilization.

Translink unfortunately does not break down capacity utilization into its two components, and only cites the crowding level at the most crowded point of the average trip. But we can still construct a table of some routes with their performance on the three metrics as well as their crowding level:

Route Boardings/hr Capacity use Peak Crowding Cost/boarding
3 113 110% 49% $0.89
5 105 67% 53% $0.95
6 98 68% 48% $1.02
8 102 98% 41% $0.98
9 103 160% 64% $0.97
20 113 124% 50% $0.89
25 81 187% 74% $1.23
41 105 180% 78% $0.95
49 96 169% 82% $1.04
99 176 167% 86% $0.57

The 3, 9, 20, 41, 49, and 99 are in the infographic on the list of most productive routes; the 25 narrowly misses on cost per passenger and boardings per hour but is second systemwide in capacity utilization, and the 5, 6, and 8 miss only on capacity utilization. The 25 and 49 have strong anchors at their outer ends, a single strong central anchor at the Canada Line, and nothing else; On the metrics relevant to the passenger who’s expected to ride the bus and fund it by paying a fare, they are somewhat lower-performing than the short 5 and 6 and the short-trip-encouraging 3, 8, and 20, but have far more crowding. The 9 and 41, both in the infographic, are about on a par with the 3, 5, 6, 8, and 20, and have more turnover due to additional destinations on Broadway and 9th that don’t exist on King Edward and 49th, but are still much more crowded than the unanchored routes. The 99 beats all others in performance, but the cost in terms of crowding is even higher.

The purpose of anchoring is explicitly to keep buses full all the way; the 25 and 49 are great at that, since people ride them longer distances, not having much to go to between their major destinations. However, it’s not a measure of passenger satisfaction or of transit agency finances, but of passenger-km. The surreptitious focus on passenger-km is dubious as a performance metric for urban transit, since transit-using city dwellers usually prefer shorter commutes and do most non-work trips on foot.

And if it’s dubious as a transit proposition, then as an urbanist proposition it’s destructive. As discussed in my previous post on the subject, Vancouver is upzoning Marine Drive (slightly) more intensely than the area south of Broadway and near the stations on the Canada Line between Broadway and Marine Drive – see PDF-pp. 26-27 of the draft plan. Despite the hysteria about urban planners using social engineering to make people live close to city center and take transit instead of driving, here we have a city with an otherwise well-deserved reputation for greenness using social engineering to make people live farther out.

This focus on anchors is making Vancouver build itself to be on a regional scale like how the 25 and 49 look on the local scale. The famous high-rise Vancouverism is really about looking like the 5 and 6 – i.e. upzoning near Downtown so that people will walk or take short trips – but future development is not intended to occur near Downtown but rather in strategically chosen secondary CBDs farther away. And what is really needed is continuous corridor development, as is practiced on the corridors hosting the 3, the 8, and the 20.

Monorail Construction Costs

Supporters of monorail and other sleek structures argue that because the structures are thinner than conventional rail viaducts, they’re cheaper and more aesthetic. They even argue that viaducts, which are more expensive than at-grade construction, are actually better. Transrapid does that, and Hyperloop does that as well. Hyperloop proponents specifically mention the the structure’s lighter weight as an explanation for the lower proposed cost – see my post update and the comments for extensive discussion and explanation for why the proposed Hyperloop costs are still an order of magnitude too low to be realistic.

Not having reliable construction costs for the intercity modes, I went and looked for construction costs of urban monorails, which are usually put above-ground, where their sleekness is a major advantage over conventional rail since they do not darken the street as much.

The resource with the most information is a JRTR article about Japanese intermediate-capacity rail, including both monorail and significantly less sleek automated guideway transit (AGT). It includes a diagram of monorail structures, which can be seen to be quite light and thin. The width of the structure from guideway to guideway is 4.5 meters including both guideway widths, and including the outside appears to raise it to 5.5. Two-track elevated conventional rail structures typically range from 7 to 10.5 meters wide.

The most recent Japanese monorail on the list, the Tama Monorail in suburban Tokyo, opened 1998, was $2.422 billion for 16 km: $151 million per km.

The one Japanese project more recent than the JRTR article, the Okinawa Monorail, built from 1996 to 2003, was $1.1 billion for 13 km: $85 million per km. The cost cited on the Monorail Society’s webpage is less than a third that amount. An extension to begin construction soon is projected at $350 million for 4 km, about the same cost per km.

Other Tokyo projects are not cheaper than the Tama Monorail. The AGT Yurikamome, opened 1995, cost $140 million per km as per JRTR; the Tohoku Jukan Line, a conventional elevated structure on top of an older elevated structure located in Central Tokyo, is $400 million for 1.3 km of new el and 2.5 additional km of new track on existing structure, which is $300 million per km if one considers the cost of everything except the new el to be zero, and about $150 million per km if the 2.5 km of existing structure is deemed to cost as much as at-grade rail, which is about half as much as an el typically. The ratio of elevated to underground cost is 2-3 and this is also in line with $150 million per km of baseline cost.

Outside Japan, we have the following projects, with their costs:

Chongqing Rail Transit Line 3: the first phase, built from 2007 to 2011, is ¥13.8 billion for 39 km, or in PPP dollars $88 million per km. About a third of the line is underground. An extension opened in 2013 cost ¥5.7 billion for 16.5 km (Chin.), or $85 million per km, all elevated. This is in a country where fully underground subways average about $150 million per km; but I cannot find cost figures for other lines in Chongqing itself, and any help would be appreciated.

Moscow Monorail: according to Wikipedia, 6.33 billion rubles in 2001-4, or $514 million in 2010 PPP dollars, for 4.7 km. This is $109 million per km, all elevated.

AirTrain Newark: the airport-internal people mover opened in 1996 and cost $354 million for 3 km, while the extension to the mainline train station was built from 1997 to 2000, added another 1.8 km to the project, and cost $415 million; both numbers are taken from the New York Times. Deflating both numbers to 1998, this is $1.03 billion for 4.8 km, or $215 million per km. In contrast, AirTrain JFK, a SkyTrain-like system, was $1.9 billion for 13 km and was built from 1998 to 2002, which in 2010 dollars is $185 million per km, actually lower than the cost of the monorail. Note that the AirTrain construction cost was not that high by normal-world standards: the same technology in Vancouver, in all-elevated configuration, is projected at C$116-150 million per km into Surrey when all elevated (see RRT alternatives 1 and 3 with distances of 15.5 km for 1 and 6 km for 3 as measured on a map), which is about US$95-120 million after PPP conversion. This is a 50-100% premium for New York over Vancouver prices, compared with a 400-600% premium for subway construction.

Palm Jumeirah Monorail, Dubai: Dh4.1 billion for 5.45 km, built 2006-8, about $1 billion in 2010 PPP dollars. This is $183 million per km, all elevated. Compare with the 17% underground Dubai Metro, mentioned in my previous post, which costs half as much per km.

Mumbai Monorail: the master plan is to spend 20,000 crore on 135 km, which after PPP conversion is $66 million per km. The under-construction first line is 3,000 crore for 19.54 km, or $69 million per km, all elevated. This compares with a parallel Mumbai Metro plan to build 146 km for 36,000 crore, or $111 million per km, of which according to Wikipedia 32 km, or 22%, is to be underground. This by itself suggests no monorail cost saving. But the first Mumbai Metro line is already over budget, at 3,800 crore, which again using Wikipedia for length (11.07 km) gives $154 million per km, all elevated. This suggests that in Mumbai there is a cost saving from using monorail, assuming all numbers are correct and that the Monorail Line 1 cost is not just the first phase, which is only 8 km.

Is Low-Cost Intercity Rail Possible?

Update: see corrected Shinkansen staffing numbers below

The last few decades have seen the growth of airlines and bus operators that reduce operating costs using a variety of lean-production ideas, chiefly using the equipment for more hours per day to earn more revenue with the same fixed costs. This hasn’t generally happened for rail, even in the presence of competition between operators. There is one low-cost option, on the TGV network, which like Ryanair and easyJey cuts costs not only by leaner production but also by reducing passenger comfort and convenience. I contend that an intermediate solution should be investigated: lean like Southwest and JetBlue, but without the extra fees, which are lower on those two airlines than on legacy US airlines.

First, the preexisting fares. In Japan, JR Central charges an average of $0.228 per passenger-km on the Shinkansen, JR East charges $0.245, JR West charges $0.208. In Japan nearly all intercity service is Shinkansen; averaging all JR East rail other than Tokyo-area commuter rail, even commuter rail around Sendai and Niigata, drops the average marginally, to $0.217. European intercity rail fares per passenger-km are lower: €0.104 on RENFE (PDF-p. 27), €0.108 on DB, and €0.112 on SNCF. All of those companies are profitable and do not receive subsidies for intercity rail, with the exception of RENFE, which loses small amounts of money (-0.8% profit margin). This is far lower than Northeast Corridor fare, which, as of the most recent monthly report, averages $0.534 per passenger-km on the Acela and $0.292 on the Regional.

Now, we can try penciling what operating costs should be. The most marginal costs, which grow linearly with the addition of new service, look a lot like those of low-cost private bus operators: crew, cleaners, energy, rolling stock acquisition, rolling stock maintenance. I am specifically handwaving the peak factor – frequency is assumed to be constant, to establish the operating cost of the base rather than that of the peak. I am going to assume 1,120 seats per train, all coach, about the same as a 16-car Shinkansen with 2+2 standard-class seating, or 70 per car. First class should be thought of as an equivalent of buying extra seats – fares should scale with the amount of space per passenger, and at any rate most cars are coach. Occupancy rate will be taken to be 57%, for a round 40 passengers per car; this is well within the range of HSR occupancy.

The cost turns out to be quite low – this is similar to the analysis in Reason & Rail from 2 years ago, except for now I’m leaving out infrastructure costs, which in that analysis are the dominant term, and so excluding them leads to very low costs. It is about three cents per passenger-km in operating and maintenance costs. This is of course not what HSR currently costs, but should be thought of as a lower limit or as the marginal cost of increasing base service.

A crew on a high-speed train is a train driver and a conductor. A 16-car Shinkansen train appears to have one conductor judging by the single conductor’s compartment has three conductors (see Andrew in Ezo’s comment below); the TGV has much more staffing, with the low-cost TGV having four. US salaries are high because the railroads have good unions: according to the Manhattan Institute’s applet for public employees’ salaries, on the LIRR, the average train driver makes $103,000 a year (search for “engineer”) and on Metro-North $115,000 (search for “locomotive engineer”). This is higher than on the Shinkansen. A conductor makes $98,000 on the LIRR and $105,000 on Metro-North. Figure $240,000 per year for a two-person crew $440,000 per year for a four-person crew.

We need to convert this to operating hours. On the LIRR and Metro-North, there are about 4,500 revenue car-hours per driver-year, which translates to about 600 revenue train-hours. At an average speed of 200 km/h, HSR would cost $2 $3.67 per train-km, or $0.003125 $0.0057 per passenger-km. But Metro-North and the LIRR are inefficient due to a prominent peak making smooth scheduling difficult; HSR can schedule a simple shift with a roundtrip of about 6-7 hours plus rest time, and if each employee does this 5 days a week minus holidays this is 1,200 revenue hours. This halves the cost. Conversely, going to 4 conductors, with a five-person crew paid a total of $540,000 per year, raises the cost to $0.007 per passenger-km, still low.

Electricity consumption can be calculated from first principles based on acceleration characteristics, or based on real-life HSR consumption levels. For the latter, a UIC paper claims 73 Wh/passenger-km on PDF-p. 17; this appears to be based on an assumption (see PDF-p. 33) of 70% occupancy but a train that is smaller (397 seats for 8 cars) and heavier (425 t vs. 365 t for an 8-car Shinkansen). Correcting for these gives 54 Wh/p-km. When I try to derive this from first principles assuming Northeast Corridor characteristics but with substantial segments upgraded to 360 km/h, I get about 50 Wh/p-km; this doesn’t include losses between catenary and wheel or regenerative braking, which mostly cancel each other out with losses being a little bigger. Rounding up to 56 Wh/p-km and using a transportation-sector electricity cost of $0.125 per kWh, we get $0.007 in electricity cost per passenger-km.

Cleaning should be done as fast as possible, with large crews working to turn trains around in the minimum amount of time based on safety margins and schedule recovery. JR East cleans Shinkansen trains in 12 minutes of Tokyo turnaround time minus 5 minutes for letting passengers disembark; the team size is 1 cleaner per standard-class car and 2-3 per green car, for a total of 22. This does not mean we can pencil in just 7 minutes of cleaning, since this doesn’t take into account the cleaning crew’s time waiting for a train to arrive, or downtime in case trains don’t arrive exactly one turnaround time apart. For a 4 tph operation, 15 minutes are fine, but for a 6 tph one, 10 may not be enough, requiring going up to 20. This is once per train run, so once per 720 km. With a team size of 24, that’s 24 person-hours per 720 train-km, or 32 in the 6 tph version.

Again using Manhattan Institute data, cleaners make $50,000 a year; it’s possible wages will have to go up to attract people who can consistently clean a car on the tight schedules posited, but there’s no base of comparison of companies having both Japanese standards for scheduling and American union scales. Say $30 per hour on the job (including downtime and waiting for a train, but not scheduled breaks). In the 6 tph version, this costs $0.002 per passenger-km.

RENFE’s above-linked executive summary includes a breakdown of employees by category (regular, support, and managerial) and gender on PDF-p. 46, whence we can obtain that for each operations employee there are 0.2 managers and 0.07 support employees. For capital projects, the California HSR estimates add 20% for overhead, management, and design, not including contingency, and the Penn Design estimate adds 18% (PDF-p. 247). This should be taken as the marginal cost of extra managers to oversee extra employees hired to provide additional service. In total, this is roughly $0.019 per passenger-km assuming higher crew staffing, and $0.013 $0.0175 assuming lower staffing.

Rolling stock is more expensive, and should spend as much time earning revenue as feasible based on established maintenance protocols. A large share of the operating costs of high-speed rail comes from the rolling stock: 20% on Madrid-Barcelona according to a RENFE presentation to California HSR whose official source is now a dead link, and, from eyeballing, perhaps 25% according to PDF-p. 8 of a UIC presentation about track access charges. The low-cost TGV doubles train utilization to about a million kilometers a year. This should be routine on Northeast Corridor operations: two round-trips per train, about 14-15 hours per day including turnaround time, 1 million train-km a year. Procurement of new N700s costs about $3 million per car, and Japanese depreciation schedules are over 20 years. Other trains capable of more than 250 km/h cost $4 million per car in China; with mid-life refurbishment of non-trivial cost, they can last up to 40. With 4% interest cost, depreciation and interest are about $280,000 per car-year either way, and if a car travels a million km with 40 people on average, that’s another $0.007 per passenger-km, a substantial sum so far.

Rolling stock maintenance is also relatively expensive. California HSR’s 2012 business plan has a list of costs around the world on PDF-p. 136. JR Central’s rolling stock maintenance is $7.20 per trainset-mile, which with our assumptions translates to $0.007 per passenger-km. European rolling stock maintenance costs are $4.16 per trainset-mile, which appears to be for an 8-car train, so scaling up by a factor of two gives $0.008 per passenger-km. Note that the maintenance of the rolling stock costs as much as the depreciation and interest on its acquisition.

In reality, maintenance depends on both time and distance, so increasing rolling stock utilization leads to lower costs per train-km. Since with those assumptions, the rolling stock costs about as much as the actual operations, this is a major cost cutter, though not a game changer given other costs. Note that the RENFE presentation slide also includes a large array of fixed costs and infrastructure (maintenance, which is very cheap at about $100,000 per route-km per year, and depreciation and interest on construction, which aren’t so cheap) as well as managerial overheads, hence the 20%; the UIC presentation includes some overheads as well. However, those fixed costs are more affordable if they’re spread across more service. A line built to have a 6 tph capacity has the same infrastructure cost at any frequency up to 6 tph.

So far, adding up all the operating and rolling stock costs totals to about $0.03 $0.033 per passenger-km. This means $11 $12 direct operating costs between New York and Washington or New York and Boston. It’s also a quarter what the Europeans charge for HSR tickets, and an eighth of what the Japanese charge. Despite this, the California HSR numbers are similar, so this analysis passes a sanity check. Again referring to the business plan’s PDF-p. 136, the table claims operating costs per trainset-mile that, after scaling from 8- to 16-car trains, are $0.04 per passenger-km. They exclude rolling stock acquisition, but include maintenance; but the assumptions in the Operations and Maintenance Peer Review are worse than in this post, with worse train utilization (turnaround times are assumed to be 40 minutes on PDF-p. 21) and more staff on board each train (an engineer, a conductor, an assistant conductor, a ticket collector, and a special services employee per 8-car unit, for a total of ten employees for 16 cars).

Still, I have no expectation that anyone can charge $11 $12 profitably for HSR service between New York and Washington. However, I strongly believe costs could be brought substantially below current rates. I believe the reason SNCF has only begun to do that and other operators not at all comes from two places.

First, infrastructure charges, a third of the cost of both the TGV and the Madrid-Barcelona AVE, are not just about paying off infrastructure costs (both Spain and France are low-construction cost countries for HSR). They transfer profits from the HSR operator to the monopoly infrastructure owner: track access charges were specifically increased in France ahead of the opening of the European rail market to competition, ensuring HSR surplus would go to state-owned infrastructure owner RFF rather than to foreign companies or the customers.

And second, unlike in the US, in Europe low-cost airlines are associated with terrible service: low seat pitch, hidden fees, rigid policies toward carry-on baggage, rigid policies toward missed flights, worse customer satisfaction, secondary airports located far from the cities they purportedly serve. The US has some of this in Spirit Airlines and Allegiant Airways, but it also has Southwest, JetBlue, and Virgin Atlantic, which have high customer satisfaction, flexible tickets, secondary airports located close to city centers (such as Dallas Love Field), and seat pitch equal to or better than that of the legacy airlines, which have degraded service. Europeans hate low-cost flying; Americans hate flying. The result is that Ryanair tars any attempt to lower costs in Europe by associating lean production and high equipment utilization with no-frills third-class service. This might make managers more wary of adopting some of the more positive aspects of low-cost carriers. Japan has no major low-cost carriers, so although it does not have the stigma, it doesn’t have the domestic experience, either.

I do not believe it’s possible for a train to charge $11 $12 one-way between New York and Washington and stay in business. There needs to be some profit margin, plus paying back infrastructure construction costs. However, I do believe it’s possible to charge closer to that than to present European HSR fares for the same distance (about $45), let alone present Amtrak fares. California HSR is actually pointing the way, but has such high construction costs that paying off even part of construction represents a major rise in ticket fares. The Northeast can and should do better.

Loopy Ideas Are Fine, If You’re an Entrepreneur

There is a belief within American media that a successful person can succeed at anything. He (and it’s invariably he) is omnicompetent, and people who question him and laugh at his outlandish ideas will invariably fail and end up working for him. If he cares about something, it’s important; if he says something can be done, it can. The people who are already doing the same thing are peons and their opinions are to be discounted, since they are biased and he never is. He doesn’t need to provide references or evidence – even supposedly scientific science fiction falls into this trope, in which the hero gets ideas from his gut, is always right, and never needs to do experiments.

Thus we get Hyperloop, a loopy intercity rail transit idea proposed by Tesla Motors’ Elon Musk, an entrepreneur who hopes to make a living some day building cars. And thus a fair amount of the media coverage is analysis-free summary of what Tesla already said: see stenography by ABC, Forbes, the Washington Post’s Wonkblog, and even BusinessWeek (which added that critics deal with “limited information”). Some media channels are more nuanced, sometimes even critical; the Wall Street Journal deserves especial credit, but Wonkblog also has a second, mildly critical post. But none has pressed Musk or Tesla about the inconsistencies in his proposal, which far exceed the obvious questions about the proposed $6 billion price tag (compare $53 billion in today’s money for California HSR). For better prior criticism, see James Sinclair’s post and Clem Tillier’s comment on California HSR Blog.

My specific problems are that Hyperloop a) made up the cost projections, b) has awful passenger comfort, c) has very little capacity, and d) lies about energy consumption of conventional HSR. All of these come from Musk’s complex in which he must reinvent everything and ignore prior work done in the field; these also raise doubts about the systems safety that he claims is impeccable.

In principle, Hyperloop is supposed to get people from Los Angeles to San Francisco in half an hour, running in a tube with near-vacuum at speeds topping at 1,220 km/h. In practice, both the costs and the running times are full of magic asterisks. The LA end is really Sylmar, at the edge of the LA Basin; with additional access time and security checks, this is no faster than conventional HSR doing the trip in 2:40. There is a crossing of the San Francisco Bay, but there’s no mention of the high cost of bridging over or tunneling under the Bay – we’re supposed to take it on faith the unit cost is the same as along the I-5 corridor in the Central Valley.

There is no systematic attempt at figuring out standard practices for cost, or earthquake safety (about which the report is full of FUD about the risks of a “ground-based system”). There are no references for anything; they’re beneath the entrepreneur’s dignity. It’s fine if Musk thinks he can build certain structures for lower cost than is normal, or achieve better safety, but he should at least mention how. Instead, we get “it is expected” and “targeted” language. On Wikipedia, it would get hammered with “citation needed” and “avoid weasel words.”

The worst is the cost of the civil infrastructure, the dominant term in any major transportation project’s cost. Hundreds of years of incrementally-built expertise in bridge building is brushed aside with the following passage:

The pods and linear motors are relatively minor expenses compared to the tube itself – several hundred million dollars at most, compared with several billion dollars for the tube. Even several billion is a low number when compared with several tens of billion proposed for the track of the California rail project.

The key advantages of a tube vs. a railway track are that it can be built above the ground on pylons and it can be built in prefabricated sections that are dropped in place and joined with an orbital seam welder. By building it on pylons, you can almost entirely avoid the need to buy land by following alongside the mostly very straight California Interstate 5 highway, with only minor deviations when the highway makes a sharp turn.

In reality, an all-elevated system is a bug rather than a feature. Central Valley land is cheap; pylons are expensive, as can be readily seen by the costs of elevated highways and trains all over the world. The unit costs for viaducts on California HSR, without overhead and management fees, are already several times as high as Musk’s cost: as per PDF-page 15 of the cost overrun breakdown, unit costs for viaducts range from $50 million to $80 million per mile. Overheads and contingencies convert per-mile cost almost perfectly to per-km costs. And yet Musk thinks he can build more than 500 km of viaduct for $2.5 billion, as per PDF-page 28 of his proposal: a tenth the unit cost. The unrealistically low tunnel unit cost is at least excused on PDF-page 31 on the grounds that the tunnel diameter is low (this can also be done with trains if they’re as narrow as Hyperloop, whose capsule seating is 2-abreast rather than 4- or 5-abreast as on HSR; see below on capacity). The low viaduct unit cost is not.

This alone suggests that the real cost of constructing civil infrastructure for Hyperloop is ten times as high as advertised, to say nothing of the Bay crossing. So it’s the same cost as standard HSR. It’s supposedly faster, but since it doesn’t go all the way to Downtown Los Angeles it doesn’t actually provide faster door-to-door trip times.

Nor is the system more comfortable for the passenger. Levitating systems can get away with higher cant than conventional rail because they sway less: Transrapid’s lateral acceleration in the horizontal plane is about 3.6 m/s^2 in Shanghai, and the company claims 4.37 m/s^2 is possible. On standard-gauge rail, the conversion rate is approximately 150 mm of total equivalent cant per 1 m/s^2. HSR cant tops at 180-200 mm, and cant deficiency tops at 180 mm for Talgos and 270-300 mm for medium-speed Pendolinos, so about 2.5 m/s^2 at high speed; this was shown safe by simulation in Martin Lindahl’s thesis, which is also a good source for track construction standards.

But Hyperloop goes one step further and proposes a lateral acceleration of 4.9 m/s^2: 0.5 g. This is after canting, according to the standards proposed:

The Hyperloop will be capable of traveling between Los Angeles and San Francisco in approximately 35 minutes. This requirement tends to size other portions of the system. Given the performance specification of the Hyperloop, a route has been devised to satisfy this design requirement. The Hyperloop route should be based on several considerations, including:

  1. Maintaining the tube as closely as possible to existing rights of way (e.g., following the I-5).
  2. Limiting the maximum capsule speed to 760 mph (1,220 kph) for aerodynamic considerations.
  3. Limiting accelerations on the passengers to 0.5g.
  4. Optimizing locations of the linear motor tube sections driving the capsules.
  5. Local geographical constraints, including location of urban areas, mountain ranges, reservoirs, national parks, roads, railroads, airports, etc. The route must respect existing structures.

For aerodynamic efficiency, the velocity of a capsule in the Hyperloop is
typically:

  • 300 mph (480 kph) where local geography necessitates a tube bend radii < 1.0 mile (1.6 km)
  • 760 mph (1,220 kph) where local geography allows a tube bend > 3.0 miles (4.8 km) or where local geography permits a straight tube.

These bend radii have been calculated so that the passenger does not experience inertial accelerations that exceed 0.5 g. This is deemed the maximum inertial acceleration that can be comfortably sustained by humans for short periods. To further reduce the inertial acceleration experienced by passengers, the capsule and/or tube will incorporate a mechanism that will allow a degree of ‘banking’.

0.5 g, or 4.9 m/s^2, is extreme. Non-tilting trains do not accelerate laterally at more than 1.2 m/s^2 in the plane of the track (i.e. after accounting for cant), and at high speed they have lower lateral acceleration, about 0.67 m/s^2 with limiting cases of about 0.8 for some tilting trains relative to the plane of the train floor. For example, the Tokaido Shinkansen has 200 mm of cant and maximum speed of 255 km/h on non-tilting trains on 2,500-meter curves, for 100 mm of cant deficiency, or 0.67 m/s^2.

The proposed relationship between curve radius and speed in the Hyperloop standards is for a lateral acceleration much greater than 4.9 m/s^2 in the horizontal plane: 480 km/h at 1,600 meters is 11.1 m/s^2. This only drops to 5 m/s^2 after perfectly canting the track, converting the downward 9.8 m/s^2 gravity and the sideways acceleration into a single 14.8 m/s^2 acceleration vector downward in the plane of the capsule floor, or 5 m/s^2 more than passengers are used to. This is worse than sideways acceleration: track standards for vertical acceleration are tighter than for horizontal acceleration, about 0.5-0.67 m/s^2, one tenth to one seventh what Musk wants to subject his passengers to. It’s not transportation; it’s a barf ride.

Even 4.9 m/s^2 in the horizontal plane is too much. With perfect canting, it combines with gravity to accelerate passengers downward by 11 m/s^2, 1.2 m/s^2 more than the usual, twice as high as the usual standards. Motion sickness is still to be fully expected in such a case. Transrapid’s 4.37 m/s^2, which adds 0.93 m/s^2 in the vertical component with perfect canting, is the limit of what’s possible.

Speaking of vertical acceleration, this gets no comment at all in the Hyperloop proposal. At 1,220 km/h, it is very hard to climb grades, which would require very tall viaducts and deep tunnels under mountains. Climbing grades is easy, but vertical acceleration is such that the vertical curve radius has to be very large. A lateral acceleration of 0.67 m/s^2 would impose a minimum vertical curve radius of 170 km, versus 15 km at 360 km/h HSR speed. Changing the grade from flat to 2% would take 3.4 km, and changing back would take the same, so for climbing small hills, the effective average grade is very low (it takes 6.8 km to climb 68 meters).

Nor does jerk get any treatment. Reversing a curve takes several seconds at the cant and cant deficiency of conventional HSR (about 3 seconds by Swedish standards, more by German ones); reversing a curve with the extreme canting levels of Hyperloop would take much longer. Maintaining comfort at high total equivalent cant requires tight control of the third derivative as well as the second one; see a tilting train thesis for references.

The barf ride that is as expensive as California HSR and takes as long door-to-door is also very low-capacity. The capsules are inexplicably very short, with 28 passengers per capsule. The proposed headway is 30 seconds, for 3,360 passengers per direction per hour. A freeway lane can do better: about 2,000 vehicles, with an average intercity car occupancy of 2. HSR can do 12,000 passengers per direction per hour: 12 trains per hour is possible, and each train can easily fit 1,000 people (the Tokaido Shinkansen tops at 14 tph and 1,323 passengers per train).

But even 30 seconds appears well beyond the limit of emergency braking. It’s common in gadgetbahn to propose extremely tight headways, presuming computerized control allowing vehicles to behave as if they’re connected by a rod. Personal rapid transit proponents argue the same. In reality, such systems have been a subject of research for train control for quite a while now, with no positive results so far. Safety today still means safe stopping distances. If vehicles brake at a constant rate, the safe headway is half the total deceleration time; if a vehicle brakes from 1,220 km/h to zero in 60 seconds, the average acceleration is more than 5 m/s^2, twice the current regulatory safety limit for passengers with seat belts.

Most of this could be chalked to the feeling of some entrepreneurs that they must reinvent everything. The indifference to civil engineering costs, passenger comfort issues, and signal safety could all be chalked to this. So could the FUD about earthquake safety of HSR on PDF-page 5.

However, one thing could not: the chart on PDF-page 9 showing that only the Hyperloop is energy-efficient. The chart has a train consuming nearly 900 megajoules per person for an LA-San Francisco trip, about as much as a car or a plane; this is about 1,300 kJ per passenger-km. This may be true of Amtrak’s diesel locomotives; but energy consumption for HSR in Spain is on average 73 Watt-hour (263 kJ) per passenger-km (see PDF-page 17 on a UIC paper on the subject of HSR carbon emissions), one fifth as much as Tesla claims. Tesla either engages in fraud or is channeling dodgy research about the electricity consumption of high-speed trains.

Indeed, a train with a thousand seats, 20 MW of power drawn, 60% seat occupancy, and a speed of 360 km/h can only ever expend 333 kJ per passenger-km while accelerating, and much less while cruising (acceleration at lower speed requires more energy per unit of distance, but cruising at lower speed expends only a fraction of the energy of full-power acceleration). Tesla’s train energy consumption numbers do not pass a sanity check, which suggests either reckless disregard for the research or fraud. I wouldn’t put either past Musk: the lack of references is consistent with the former, and the fact that Musk’s current primary endeavor is a car company is consistent with the latter.

There is no redeeming feature of Hyperloop. Small things can possibly be fixed; the cost problems, the locations of the stations, and the passenger comfort issues given cost constraints can’t. Industry insiders with ties to other speculative proposals meant to replace conventional rail, such as maglev, are in fact skeptical of Hyperloop’s promises of perfect safety.

It’s possible to discover something new, but people who do almost always realize the context of the discovery. If Musk really found a way to build viaducts for $5 million per kilometer, this is a huge thing for civil engineering in general and he should announce this in the most general context of urban transportation, rather than the niche of intercity transportation. If Musk has experiments showing that it’s possible to have sharper turns or faster deceleration than claimed by Transrapid, then he’s made a major discovery in aviation and should announce it as such. That he thinks it just applies to his project suggests he doesn’t really have any real improvement.

In math, one common sanity check on a result is, “does it prove too much?” If my ten-page paper proves a result that implies a famous open problem, then either my paper is wrong or I’ve proved the famous open problem, and it’s up to me to take extra care to make sure I did not miss anything. Most people in this situation do this extra step and then realize that they were subtly wrong. If a famous question could be solved in ten pages, it probably wouldn’t still be open. The same is even true in undergrad-level proof classes: if your homework answer proves things that are too strong, you’ve almost certainly made a mistake.

Musk’s real sin is not the elementary mistakes; it’s this lack of context. The lack of references comes from the same place, and so does the utter indifference to the unrealistically low costs. This turns it from a wrong idea that still has interesting contributions to make to a hackneyed proposal that should be dismissed and forgotten as soon as possible.

I write this not to help bury Musk; I’m not nearly famous enough to even hit a nail in his coffin. I write this to point out that, in the US, people will treat any crank seriously if he has enough money or enough prowess in another field. A sufficiently rich person is surrounded by sycophants and stenographers who won’t check his numbers against anything.

There are two stories here. In the less interesting one, Musk is a modern-day streetcar conspiracy mogul: he has a car company, he hopes to make money off of it in the future and uses non-generally accepted accounting to claim he already does, and he constantly trash-talks high-speed rail, which competes with his product. Since he’s not proposing to build Hyperloop soon, it could be viewed as clever distraction or FUD.

The more interesting possibility, which I am inclined toward, is that this is not fraud, or not primarily fraud. Musk is the sort of person who thinks he can wend his way from starting online companies to building cars and selling them without dealerships. I have not seen a single defense of the technical details of the proposal except for one Facebook comment that claims, doubly erroneously, that the high lateral acceleration is no problem because the tubes can be canted. Everyone, including the Facebook comment, instead gushes about Musk personally. The thinking is that he’s rich, so he must always have something interesting to say; he can’t be a huckster when venturing outside his field. It would be unthinkable to treat people as professionals in their own fields, who take years to make a successful sideways move and who need to be extremely careful not to make elementary mistakes. The superheros of American media coverage would instantly collapse, relegated to a specialized role while mere mortals take over most functions.

This culture of superstars is a major obstacle frustrating any attempt to improve existing technology. It more or less works for commercial websites, where the startup capital requirements are low, profits per employee are vast, and employee turnover is such that corporate culture is impossible. People get extremely rich for doing something first, even if in their absence their competitors would’ve done the same six months later. Valve, a video game company that recognizes this, oriented its entire structure around having no formal management at all, but for the most part what this leads to is extremely rich people like Bill Gates and Mark Zuckerberg who get treated like superstars and think they can do anything.

In infrastructure, this is not workable. Trains are 19th-century technology, as are cars and buses. Planes are from the 20th century. Companies can get extremely successful improving the technology somehow, but this works differently from the kind of entrepreneurship that’s successful in the software and internet sectors. The most important airline invention since the jet engine is either the widebody (i.e. more capacity) or the suite of features that make for low-cost flights, such as quick turnarounds. What Southwest and its ultra low-cost successors have done is precious: they’ve figured how to trim every airline expense, from better crew utilization to incentives for lower-transaction cost booking methods. This requires perfect knowledge of preexisting practices and still takes decades to do. The growth rate of Microsoft, Google, and Facebook is not possible in such an environment, and so the individual superstar matters far less than a positive corporate culture that can transmit itself over multiple generations of managers.

There is plenty of room for improvement in HSR technology, then, but it’s of a different kind. It involves adapting techniques used by low-cost airlines to reduce costs, as SNCF is doing right now with its new low-cost TGV product. It perhaps involves controlling construction costs more tightly, though $5 million per km for viaducts seems like an impossible fantasy. But it has to come from within the business, or from someone who intimately understands the business.

And with the kind of success that US media harps on, this is almost impossible to do domestically. Someone as smart as Musk, or any of many other Silicon Valley entrepreneurs, could find a detailed breakdown of the operating and construction costs of civil infrastructure, and figure out ways of reducing them, Megabus- or Southwest-style. That’s what I would do if I had the unlimited resources Musk has: I’d obtain unit costs at far greater detail than “X meters of tunnel cost $Y” and compare what New York is doing wrong that Madrid is doing right. But I don’t have the resources – in money, in ability to manage people, in time. And the people who do are constantly told that they don’t need to do that, that they’re smart enough they can reinvent everything and that the world will bow to their greatness.

Update: people all over the Internet, including in comments below, defend the low cost projections on the grounds that the system is lighter and thinner than your average train. The proposal itself also defends the low tunneling costs on those same grounds. To see to what extent Musk takes his own idea seriously, compare the two proposals: the first for a passenger-only tube, and the second for a larger tube capable of carrying both passengers and vehicles. On PDF-pp. 25-26, the proposal states that the passenger-only tube would have an internal diameter of 2.23 meters and the passenger-plus-vehicle tube would have an internal diameter of 3.3 meters, 47% more. Despite that, the tunneling costs on PDF-p. 28 are $600 and $700 million, a difference of just 17%.

The same is true of the “but the Hyperloop capsule is lighter than a train” argument for lower pylon construction costs. Together with the differences in tube thickness posited on PDF-p. 27, 20-23 mm versus 23-25, there is 60% more tube lining in the passenger-plus-vehicle version, but the tube and pylons are projected to cost just 24% more. In this larger version, the twin tube has 0.025*3.3*pi*2 = 0.5 cubic meters of steel per meter of length, weighing about 4 tons. This ranges from a bit less than twice to a bit more than twice the weight of a train. To say nothing of the pylons’ need to support their own considerable weight, which is larger than for HSR due to the need for taller viaducts coming from the constrained ability to change grade. They are far more obtrusive than trees and telephone poles, contra the claims of minimal obtrusiveness and disruption.

Update update (12/24): Hyperloop is in the news again; I’ve been getting a lot of pingbacks copying this article. You can read the plan here; the construction costs are now up from a laughable sub-$10 million per kilometer to $10-30 million, which is perfectly feasible if you’re building in flat terrain and if what you’re building is conventional rail and not a vactrain. There’s virtually no discussion of why the costs are so much lower, just an assurance that the team ran the numbers and that they’re looking into minimizing the costs of the construction material (costs that, for conventional HSR, are a small proportion of the total construction costs – concrete is cheap, it’s pouring it that’s expensive). On PDF-p. 19 of the new plan, the accelerations are explicitly stated to be 0.5 g in normal service, which the person heading the team trying to build it claims is not a barf ride in the article, but which is in reality is again worse than the acceleration felt by passengers on an airplane taking off. There already exists a mode of transportation that involves security theater, travel at 1,000 km/h, poor comfort, and motion sickness.

Tel Aviv Needs a Subway, Done Right

After decades of false starts, Tel Aviv is finally building a subway-surface line. The political opinions of activists and urban planners in Israel are divided between supporters, who believe the line is long overdue, and opponents, who instead believe buses remain the solution and also oppose the Jerusalem light rail. I on the contrary think that on the one hand Tel Aviv needs a subway, but on the other hand the current plan has deep flaws, both political and technical, and is learning the wrong lessons from recent first-world greenfield subways.

In some ways, the Tel Aviv subway resembles New York’s Second Avenue Subway. It passes through neighborhoods that are very dense – the line under construction connects some of the densest cities in Israel, albeit poorly. Nobody believes it will be built because of all the false starts. Real incompetence in construction leading to cost overruns has led to speculation about much greater cost overruns.

For nearly a hundred years, the conurbation around Tel Aviv and Jaffa has been the largest metro area in what is now Israel; it is also the largest first-world metro area outside the US that has no urban rail. There were preliminary plans for a Tel Aviv subway in the 1930s, followed by repeated plans since independence, all of which were shelved. A proposal from just after independence for developing coastal Israel around rail and rapid transit trunks was rejected by Prime Minister David Ben Gurion because it conflicted with the political goal of Jewish population dispersal; to further its political goals, the state concentrated on building roads instead. In the late 1950s there was a new integrated national rail plan that was not implemented. Haifa got a six-station, one-line funicular, but Tel Aviv and Jerusalem remained bus-only. In the 1960s a skyscraper in Central Tel Aviv was built with a subway station, but there were no tunnels built; a subsequent 1971 plan was abandoned in 1973 due to the Yom Kippur War. The current subway plan dates to the 1990s, and has suffered from repeated delays, and construction only began recently, with opening expected for 2016.

Unlike in the North American debate, in Israel the left is pro-BRT and anti-rail, due to a long tradition of mistrust in mainstream (center-right to right-wing) politics. The same is true of urban planners who follow the Jacobsian tradition, such as Yoav Lerner Lerman (Heb.). The article I translated two years ago about Jerusalem’s light rail is in that tradition: it attacks genuine problems with cost overruns and a politicized route choice process, but then concludes that BRT is the solution because it’s been implemented in Curitiba and Bogota successfully. The result is that people whose ideas about trade, energy, health care, education, and housing are well to the left of what is considered acceptable in the US end up channeling the Reason Foundation on bus versus rail issues.

In reality, Tel Aviv’s urban form is quite dense. The city itself has 8,000 people per square kilometer, much lower than Paris and Barcelona, but higher than most other European central cities (say, every single German city). Like Los Angeles, its municipal borders do not conform to the informal borders of the inner-urban area, since it contains lower-density modernist neighborhoods north of the Yarkon, while dense Ramat Gan, Giv’atayim, Bnei Brak, and Bat Yam are separate municipalities. The inner ring of suburbs, including the above-named four, has 7,400 people per square kilometer; excluding the more affluent but emptier northern suburbs, this approaches 10,000/km^2.

However, the urban form is quite old, in the sense that the density is fairly constant, without the concentrations of density near nodes that typify modern transit cities. Tel Aviv’s residential high-rise construction is not very dense because it still follows the modernist paradigm of a tower in a park, leading to low lot coverage and a density that’s not much higher than that of the old four-story apartment blocks. The Old North achieves about 15,000 people per square kilometer with a floor area ratio of 2: the setbacks are such that only about half of each lot is buildable, and there are four floors per building. The Akirov Towers complex averages about 2.5.

Although this density pattern favors surface transit rather than rapid transit, Tel Aviv doesn’t have the street network for efficient surface transit. Paris, a poster child for efficient recent construction of light rail (see costs and ridership estimates on The Transport Politic), is a city of wide boulevards. Central Tel Aviv has about two such streets – Ibn Gabirol and Rothschild – and one auto-oriented arterial, Namir Road, which the subway line under construction will go under. The street network is too haphazard to leverage those two for surface BRT or light rail, and the major destinations of the central areas are often on narrower streets, for example Dizengoff. On top of that, light rail speeds in Paris are lower than 20 km/h, whereas newly built subways are much faster, approaching 40 km/h in Vancouver and Copenhagen. Outside Central Tel Aviv, the roads become wider, but not nearly as wide as those used for BRT in Bogota, and there is nothing for surface transit on those streets to connect to on the surface. A surface implementation of Route 66, following Jabotinsky Street (the eastern leg of the subway line under construction) in Ramat Gan, Bnei Brak, and Petah Tikva, wouldn’t be very fast on the surface to begin with, but would come to a crawl once crossing the freeway into Tel Aviv.

Tel Aviv also has two more important reasons to imitate Vancouver and Copenhagen, besides speed: religious politics, and economic and demographic comparability. Public transportation in Israel operates six days a week, with few exceptions, to avoid running on the Sabbath. A driverless train, built to be quiet even on elevated sections, with no turnstiles and free fares on the Sabbath, could circumvent religious opposition to seven-days-a-week operation.

Even without the religious question, Copenhagen and especially Vancouver are good models for Tel Aviv to follow, more so than middle-income Curitiba or Bogota. Israel is a high-construction cost country, but Canada is not very cheap, and Vancouver has cut construction costs by making elevated trains more palatable and reducing station lengths. Greater Tel Aviv has 2.5-3.5 million people depending on who you ask, not much higher than the range for Copenhagen and Vancouver. Tel Aviv is about as dense as Copenhagen and Vancouver, though Vancouver’s density is spikier. Tel Aviv expects fast population growth, like Vancouver, though in Tel Aviv’s case it’s a matter of high birth rates whereas in Vancouver it’s only immigration.

One way in which Vancouver is not a good model is the role of regional rail. Israel has no equivalent of Transport Canada or FRA regulations. It even connected Tel Aviv’s northern and southern rail networks and through-routes nearly all commuter and intercity trains. However, the network has real limitations, coming from its poor urban station locations, often in highway medians; the through-running project was completed simultaneously with the construction of the freeway. For example, the Tel Aviv University station is located far downhill from the actual university. As a result, even when there is development near the train stations, it is usually not walkable. This compels new rail service with stations in more central locations as well as east-west service, complementing the north-south mainline.

However, for service to the less dense suburbs, the construction of new lines, and electrification of the entire national network (so far only the Haifa commuter network is scheduled for electrification), should provide the backbone. There is no integrated planning between regional rail and shorter-distance urban rail, the first failing of the current plan.

More broadly, the plan fails not just because of the wrong mode choice – subway-surface rather than driverless metro with a regional rail complement – but also because of how it treats urban geography. The proposed network – on which the red line is under construction and the green line is intended to be the second built – is too sparse in the center, and ignores the older urban centers. The phasing ignores preexisting transportation centers, and often the choice of who to serve and how to serve them is political.

The worst political decision concerns Jaffa, the old core of the metro area. (Tel Aviv was founded as a nominally independent city, but really as a Jewish suburb of Jaffa.) The most activity is in the Old City and the Flea Market, going down along Yefet Street to Ajami, since 1948 the only majority-Arabic speaking neighborhood in the municipality, and the only neighborhood that is completely unplanned. The streets are narrow, favoring a subway, and the residents are poor and have low car ownership rates. Instead, the route through Jaffa is on the surface and follows Jerusalem Boulevard, a less busy road built by the city’s then-mayor out of envy of then-separate Tel Aviv’s Rothschild Boulevard. This serves the more gentrified Jewish parts. Ajami is gentrifying – it’s close to Central Tel Aviv, is right next to the coast, and has stunning architecture – but is still majority-Arab.

The other neighborhood that due to ethnic differences is viewed separately from Tel Aviv, Hatikva, is also underserved. In this case, the residents are Jewish, but are predominantly Mizrahi and traditional-to-religious, with high poverty levels. The plan does serve Hatikva, but much later than it should given the neighborhood’s density, intensity of low-end commercial activity, and proximity to Central Tel Aviv. A northwest-southeast line, following Dizengoff and then serving Central Bus Station (a larger transportation center still than any mainline rail station) and Hatikva before continuing east into the inner suburbs, should be a high priority, but isn’t. The Central Bus Station area is also a concentration of refugees, another low-income, low-car ownership population, though since this concentration is more recent than the plans for the subway, the lack of priority service to the bus station is not a result of racism.

It’s not only about class reasons, or racial ones: Tel Aviv had to fight to get the Ministry of Transportation to agree to build the second line underground under Ibn Gabirol, and that’s to an upper middle-class Ashkenazi neighborhoods. The common thread within the city proper is a preference for new modernist luxury towers over serving existing walkable density, even when that density is hardly lower than what the towers are providing. (The towers can be built more densely, with less open space; by the same token, the low-rise buildings could be upzoned from one half the lot and four story to three-quarters and six stories.)

Another example of bad politics is the way military bases are served. The very center of Tel Aviv is home to the Ministry of Defense and the main military headquarters, the Kirya. The inner urban area is ringed with much larger military bases, including Tsrifin to the south, Glilot to the north, and the Bakum to the east. But the officer corps is concentrated in the Kirya, while Tsrifin is a more general base, Bakum is dedicated to new draftees so that they can be told what unit they’re to be sent to, and Glilot is somewhat higher-end than Tsrifin due to its role in military intelligence but still lacks the Kirya’s concentration of high-ranking officers. Since draftees almost never own cars and often ride buses for hours, the three outlying bases are all natural outer anchors for lines, and Glilot and Tsrifin both lie on easy spurs from the mainline rail network. Despite this, there are no plans for regular service, while the Kirya is part of the subway line under construction and is the intersection point with the second line to be built.

Even on pure geography, the plan makes critical mistakes. The eastern leg of the line under construction is much better than its southern leg: it goes straight from the train station through Ramat Gan and Bnei Brak to a secondary anchor in Petah Tikva. And yet, the station spacing in Bnei Brak, the densest city in Israel, is the widest, even though higher density allows shorter station spacing. In contrast, the surface segment in less dense Petah Tikva is intended to have denser stop spacing. Moreover, despite the advantages subway-surface operation has in terms of branching, the branching is meant to be really a short-turn, with half of all trains going straight to the depot still in the underground section and half continuing to Petah Tikva. Central Petah Tikva is well to the south of the line, which is intended to terminate at Petah Tikva’s peripherally located central bus station, but there is no branch serving that center, despite high intended frequencies (3 minutes on the surface, 1.5 minutes underground).

I believe that in addition to an electrified mainline rail trunk, Tel Aviv needs a driverless subway network that looks roughly like an E: one or two north-south lines (west of the freeway if one, one on each side if two), three east-west lines intersecting the mainline rail at the three main Tel Aviv stations. The east-west lines should be anchored at the eastern ends at Petah Tikva, Bar Ilan University, and the Bakum or Kiryat Ono; the north-south lines should go about as far north and south as required to serve the center, letting mainline rail take care of destinations roughly from Glilot or Herzliya north and from Tsrifin south. Such a network would not serve political goals of making Tel Aviv a luxury city; it would just serve the transportation goals of the urban area’s residents.

Quick Note: Why Quinn is Unfit to be Mayor

The Triboro RX plan calls for using preexisting freight rail rights-of-way with minimal freight traffic to build a circumferential subway line through the Bronx, Queens, and Brooklyn. It was mentioned as a possible project by then-MTA head Lee Sander and more recently by Scott Stringer and on The Atlantic Cities by Eric Jaffe. Despite not having nearly as much ridership potential as Second Avenue Subway or a future Utica subway, the presumed low cost of reactivating the right-of-way makes it a promising project.

According to Capital New York, leading mayoral contender Christine Quinn has just made up a price tag of $25 billion for Triboro, while claiming that paving portions of the right-of-way for buses will cost only $25 million. This is on the heels of city council member Brad Lander’s proposal for more investment in bus service. The difference is that Lander proposed using buses for what buses do well, that is service along city streets, and his plan includes bus lanes on major street and what appears to be systemwide off-board fare collection. In contrast, Quinn is just channeling the “buses are always cheaper than rail” mantra and proposing to expand bus service at the expense of a future subway line.

There is no support offered for either of the two cost figures Quinn is using, and plenty of contradictory evidence. Paving over rail lines for bus service is expensive; a recent example from Hartford and a proposal from Staten Island both point to about $40 million per km in the US. The map in the Capital New York article suggests significant detours away from the right-of-way, including on-street turns making the bus as slow as the existing circumferential B35 route, but also several kilometers on the railroad in Queens. Conversely, reusing rail lines for rail service is not nearly as expensive as building a subway. The MTA’s own biased study says a combined on-street and existing-right-of-way North Shore service would cost 65% more if it were light rail than if it were a busway; since the Triboro right-of-way is intact, the cost of service is in the light rail range, rather than the $25 billion for 35 km Quinn says.

But the reason Quinn is unfit for office rather than just wrong is the trust factor coming from this. She isn’t just sandbagging a project she thinks is too hard; the MTA is doing that on its own already. She appears to be brazenly making up outlandish numbers in support of a mantra about bus and rail construction costs. Nor has anyone else proposed a Triboro busway – she made the logical leap herself, despite not having any background in transit advocacy. Politicians who want to succeed need to know which advocates’ ideas to channel, and Quinn is failing at that on the transit front. If I can’t trust anything she says about transit, how can I trust anything she says about the effectiveness of stop-and-frisk, or about housing affordability, or about the consequences of labor regulations?

Update: Stephen Smith asked Quinn’s spokesperson, who cited a $21 billion figure for a far larger RPA plan including Second Avenue Subway and commuter rail through-running with new lines through Manhattan. I am not holding my breath for a retraction of the bus paving plan from the Quinn campaign.

Update 2: Quinn admitted the mistake on the rail plan, and revised the estimate of the cost down to $1 billion, but sticks to the bus plan and its $25 million estimate.