The Value of New York-Boston

This post mainly responds to an argument made by Jim in comments that the core portion of the Northeast Corridor is New York-Washington, and New York-Boston is more expensive and less useful.

The Northeast Corridor has two halves: New York-Philadelphia-Washington and New York-Boston. The southern half is the more important one: according to PDF-page 41 of the Master Plan it carries 70% of the traffic, and the top city pairs are New York-Washington and New York-Philadelphia, both substantially larger than New York-Boston. Although the corridor is always treated as a single line, it is worth checking the value of upgrades separately, especially because the southern half is straighter and would take less investment to bring up to full high-speed rail standards.

However, even with such a disaggregation, New York-Boston is a valuable route, with more potential benefits than any other in the US except New York-Washington. For a first filter, we can look at city populations. New York’s size is such that even much smaller cities can be fruitfully connected to it by greenfield HSR. Seoul, of comparable size to New York, is connected to Busan, which is slightly smaller than Boston and slightly farther apart from Seoul than Boston is from New York; the intermediate cities, Daejeon and Daegu, are somewhat larger than New Haven and Providence. Total ridership on the Seoul-Busan segment of the KTX is about 30 million per year, and Korail appears to be making profits on the line, despite very high construction costs coming from heavy tunneling (about 40% of the route).

The first filter alone warrants further investigation of the route, even independently of New York-Washington. The costs of New York-Boston, while higher than those of New York-Washington, are not very high. We can put much of the New Haven Line in the “too hard” basket initially, but instead focus on high speeds between New York and Stamford and between New Haven and Boston and somewhat higher speeds between Stamford and New Haven than today. The average speed that can be squeezed with such investment is comparable to that of the KTX today, before the high-speed segments through two intermediate cities have been completed; those segments can then be compared to tackling Stamford-New Haven.

Of course, while adequate, is still nowhere nearly as good as what can be done on New York-Washington. It’s not just that this segment connects New York to two different cities the size of Greater Boston. Although today New York-Philadelphia has higher ridership than New York-Boston, this may well be reversed in the presence of full-fat HSR, since cutting a 1:05 trip to 0:38 does not provide the same competitive boost as cutting a 3:30 trip to 2:00, and Amtrak’s primary competition is highways rather than planes.

Rather, the benefits of New York-Washington are more in the distribution of the secondary centers. The biggest satellite metro area of Washington is Baltimore, on the line. Of Boston’s satellite metro areas – Providence, Worcester, and Nashua-Manchester – at most one can be served by any semi-reasonable HSR alignment. Philadelphia-Washington provides an additional market to be served. Washington’s population growth is much faster than that of the rest of the Northeast, and is comparable to that of the Sunbelt.

At the same time, there are benefits to building both lines; since under a phased program that built one first and then the other New York-Washington would be first, those benefits should be counted as benefits of New York-Boston. The biggest one is service through New York. Boston-Washington is a major air market: the O&D passenger volume between the Boston and Washington areas is 3.1 million a year; this is more than Boston-New York and New York-Washington combined, and Boston-Washington is both Boston and Washington’s busiest air city pair. In general, Boston-Washington and Boston-Philadelphia are located at more favorable ranges for HSR’s competitiveness than New York-Boston and New York-Washington, on which the trip time advantage versus cars and buses is smaller.

If we take Boston-Philadelphia and Boston-Washington and slightly more than double the size of today’s air market – the same ratio of present-day HSR traffic to pre-HSR air traffic between London and Paris – we get an additional 9 million passengers a year, small compared to the possibilities of the other city pairs but non-negligible.

The other issue is rolling stock and operating plan. Because the Northeast Corridor runs as one route, and there is no point in separating its two halves operationally, the same rolling stock that runs south of New York also has to run north of New York. In particular, we obtain the following situation:

– Traffic purely south of New York justifies rolling stock that could in principle achieve much higher speeds north of New York.

– Because of certain capacity problems, both real and imagined, a few strategic bypasses and (near Boston) commuter rail modernization can have outsized intercity trip time benefit relative to the cost. A high-speed train that gets stuck behind commuter trains pulled by diesel locomotives does Boston-Providence in about 35 minutes. One that shares tracks with punctual modern EMUs can do the same in 20.

– At low levels of infrastructure investment, the cost of new rolling stock can be sizable compared to the cost of infrastructure. Half-hourly service with 16-car trains, one possible initial service plan, costs $250 million times the one-way Boston-Washington trip time in hours including turnaround time, plus spare ratio. 15-minute service requires double that cost, naturally. None of the New York-Boston projects, not even “buy the MBTA better trains,” flips to negative cost this way, but some have lower cost as a result.

– Shared-track HSR requires very good punctuality. Coming south from Boston, trains have no reason to be off-schedule in New York – Metro-North is reasonably punctual (though not enough for HSR) and both between the Rhode Island/Massachusetts state line and New Haven and between New Rochelle and New York Amtrak owns the tracks and runs the most trains – but the bypass and junction separation will help further. Coming north from Virginia there are much bigger problems, but trains can dwell arbitrarily long in Washington for schedule buffer, which they can’t in New York.

What this means is that it’s best to phase Northeast HSR investment throughout the entire corridor. New York-Washington should have the priority because it’s cheaper and has more traffic potential, but unless for some reason there is no money (not even about $4 billion for immediate improvements plus a little more for a bypass), parts of New York-Boston should be in the first phase.

Arguably, even true branches, such as to the Keystone Corridor and Springfield, can get through-service early. Those do not have the ridership potential of the core route, but electrifying New Haven-Springfield to run trains through, and programming extra trains for Keystone, can be done within a few years. The limit perhaps is only rolling stock, or more precisely what to do with the electric locomotives that would become obsolete (they already are, but are a sunk cost, and buying new HSR trainsets now becomes an additional one-time cost).

Northeast Corridor: Dealing With Capacity

To build high-speed rail on the Northeast Corridor cheaply, intercity trains will have to share tracks with regional trains at several locations, which between them comprise a majority of the corridor. At most of these, commuter traffic is heavy enough that it must be accommodated in some way; only in a minority is it so insignificant that Amtrak can feasibly kick trains out if need be. So far I’ve only explained how track-sharing can be done between Boston and Providence and between New York and New Rochelle, both far from the busiest segments of the corridor.

I’d like to start tackling the more difficult segments, New York-Trenton and New Rochelle-Stamford. Thankfully, they are almost fully four-tracked; the one exception is the North River Tunnels and the immediate approaches, where there is little speed difference between intercity and regional trains. Unfortunately, even four tracks are not enough to provide full separation between services that do not run at the same speed, because those corridors are busy enough to warrant both local and express commuter service. This requires some scheduling creativity. In both cases, what is required is having express commuter trains weave between the local tracks and the intercity tracks.

As before, my explicit assumptions are that the rolling stock is optimized, and that speed limits except those coming from right-of-way geometry have already been eliminated. However, since unlike the MBTA, Metro-North runs good rolling stock, and New Jersey Transit runs passable rolling stock, we can’t realistically expect either to buy the most powerful regional trains on the market; that said, New Jersey Transit is looking into new trains, and we will assume those trains will be in line with the high-performance but heavy Silverliner Vs. What we can expect is better on-time performance and less schedule padding. The amount of commuter traffic is assumed to be similar to or slightly higher than today; the outer ends of both lines can be expected to lose traffic from commuter to intercity trains, but the rest will not.

Complicating all this is the requirement of making all the trains cohere into one line. In other words, unlike the situation for Boston-Providence, Newark-Trenton can’t stand on its own; the trains need to depart Newark with suitable gaps to allow trains to come in from the Kearny Connection. Likewise, New Rochelle-New Haven needs to feed into New Rochelle-New York in such a way that trains can share tracks on that segment. The most difficult portion is then combining the two commuter halves of the Northeast Corridor together to allow through-running, without holding commuter trains for too long at Penn Station. One possibility is to expand the entire route between the tunnels and Newark to a long overtake segment, and have all commuter trains stop at Secaucus to further slow them down and permit intercity trains that arrived at New York second to arrive at Newark first.

Newark-Trenton

Current peak traffic on New Jersey Transit’s Northeast Corridor and North Jersey Coast lines is 13 trains per hour. Of those, only 2 make local stops from Rahway north, and both are North Jersey Coast trains. An additional 2 trains are express North Jersey Coast trains, leaving us with 9 trains at Metropark (3 stopping from New Brunswick, 3 more stopping from Jersey Avenue, 3 super-express from Trenton).

The introduction of high-speed trains would change the distribution of demand dramatically. From Trenton, HSR would be far faster. Even from Princeton Junction, it would be substantially faster to take a commuter train south to Trenton and connect to HSR to New York. For passengers desiring a one-seat ride, trains could continue to run, but make more stops along the way. We may suppose that no commuter train will skip any stop from Metropark south, and that an additional 2 trains that currently run express to New Brunswick or Jersey Avenue will run local, providing the Rahway-Newark segment with a peak local traffic of 4 tph.

The desired ideal is that all commuter trains will stay away from the inner two tracks, with brief forays when absolutely necessary. The above rule regarding local runs ensures no overtakes among the commuter trains take place south of Rahway. We are then left with the task of ensuring all overtakes north of Rahway make use of the two existing six-track segments, around Newark Airport, and from just south of the Elizabeth curve to Union Interlocking between the mainline and the North Jersey Coast Line, the latter segment including Linden and Rahway.

Because the southern six-track segment persists through the interlocking, it provides a fully separated route between local coast trains and express mainline trains, and also between local mainline trains and express coast trains southbound. Slightly modifying the interlocking to allow a separated northbound path between local mainline and express coast trains that does not use the inner two tracks may be required.

Now, a Silverliner V running at 160 km/h appears to lose about 90 seconds to a high-platform stop. For the record, a FLIRT would lose 75 seconds. Since there are five local stops north of the interlocking, we have to deal with 7.5 minutes. With 2-minute headways, this means local trains can depart Newark 9.5 technical minutes ahead of express trains on the same branch (mainline or coast), and 6.5 ahead of express trains on the opposite branch. Since 11/9.5 = 1.16 and a 16% pad is excessive, 11 scheduled minutes of separation are enough, and we obtain the following option for departure times out of Newark:

Express :00
Express :02
Express :04
Local :06
Express :15
Express :17
…

Each local must serve a different branch from the express immediately following, since 9 < 9.5. The express afterward – for example the :17 express after the :06 local – is separated by 11 minutes, and so can run on any branch. This allows 16 tph, of which 4 are local, and at least 4 have to run on the New Jersey Coast Line. Of course not every slot has to have a train scheduled in it.

Adding local frequency at the expense of express frequency is possible, but requires tightening the gap between a local train and the express that follows it, unless we allow inconveniences such as serving the local stations at highly irregular intervals. For example, 10-minute local headways allow trains to depart Newark at,

Express :00
Local :02
Express :10
Local :12
…

The 8-minute difference means each express must serve a different destination from the preceding local, and this underserves the mainline at only 6 tph. We can add stops to the express trains (or saddle them with inferior, locomotive-hauled rolling stock), but two stops are required, unless New Jersey Transit makes sure to get cutting-edge trains, which is unlikely; with FLIRTs, the time difference shrinks from 7.5 minutes to 6.25 and only one stop is required. In either case, we might as well squeeze an :x8 express, also serving a different destination from the local ahead of it.

Another option is to use the Linden-Rahway segment for overtakes. Trains lose 3 minutes there. If we add an infill stop, they lose 4.5, which is very close to the 4 minutes required to switch the order of two trains. This means express trains need to approach Linden 2-2.5 minutes behind the locals, and thus leave Newark 7 minutes behind. We obtain,

Local :00
Express :07
Express :09 (different destination from the :00 local)
…
(12-minute clockface pattern)

Since 12-minute schedules are generally awkward and my New York-New Rochelle proposal uses 10- (below) or 15-minute schedules (in the original link), we should add a Newark Airport or Elizabeth stop to the express trains, and then they leave Newark about 5-5.5 minutes behind the locals, and we can have a 7.5-minute pattern, which divides 15 evenly. Alternatively, we can add the stop and then have a 10-minute pattern again. We either get 6 local and 12 express slots with each express serving a different destination from the local behind it, or 8 local and 8 express slots, and there is no restriction on destination:

Local :00
Express :05
Express :07
Local :10
Express :15
Express :17
…

or

Local :00
Express :05
Local :07
Express :13
Local :15
Express :20
…

Note that nowhere here does HSR share tracks with anything except maybe in the Newark Penn Station throat, under any of the options. Thus, any discussion of HSR speed zones is irrelevant, except perhaps at the final stage when some tweaks to the basic schedule are under consideration.

New Rochelle-Stamford

Like Newark-Trenton, this is a four-track segment. However, commuter traffic here is heavier, and there are no six-track segments. Instead, overtakes between express commuter trains and intercity trains must be done on bypass segments. The one that we will consider is, as I outlined before, a route from just south of Rye to between Greenwich and Cos Cob, following the I-95 right-of-way.

No HSR on the New Haven Line should be considered with New Rochelle as it is. The flat junction and S-curve together severely constrain train speed and capacity. Since the junction has to be grade-separated, and some takings are required, we might as well assume the separation allows trains to proceed without crossing opposing traffic no matter where they go. Furthermore, the station should be six-tracked if necessary.

We will also assume the curve has been partially eased, to a radius of 700 meters with appropriate superelevation spirals, permitting our example 375-mm-equivalent-cant trains 150 km/h. We will also assume that Harrison has been partially eased to 1,500 meters, permitting 220; that the curve toward the bypass around Rye is 1,500 meters, which may be slightly too optimistic but not by more than a few seconds of travel time; and that the curves farther north until the Stamford approach are 2,000, permitting 250. The approach to Stamford consists of two curves forming a wide S, the western one at 1,000 (180 km/h) and the eastern one at 800 (160). Note that these upgrades allow express commuter trains to travel at 160 on the shared segments – indeed, they require it to avoid or at least limit cant excess. Local trains have more limited cant because of the needs of freight.

We will assume the number of trains is about the same, with capacity boosted with longer trains; where New Jersey Transit runs 12-car trains because of limited capacity across the Hudson, Metro-North tends to run 8-car trains. Unlike in New Jersey’s case, there’s little point then in programming more slots for trains.

Let us now consider Easy Mode, with all Metro-North trains using the existing route to Grand Central. We have two track-sharing segments, one between New Rochelle and Rye, and one between Greenwich and Stamford. The first segment is 12 km long and has 3 stations with 2 more at the ends; the second is 8 km long and has 3 stations as well, with just 1 more at one end.

On the first segment, there are 18 Metro-North tph peak today: 12 not stopping at all, 2 stopping only at Harrison, and 4 local. Now if the express trains share tracks with HSR rather than the locals, we will want to schedule trains HSR-express-express repeating every 10 minutes, or HSR-express-express-express repeating every 15; the former allows more versatile HSR slots (local and express), and the 15-minute assumption of New York-New Rochelle has no relevance in Easy Mode.

Current scheduled time between Rye and New Rochelle is 17 minutes for local trains; judging by both rolling stock capability and the local-express schedule difference, nonstop trains take about 10 minutes, and trains stopping at Rye but not New Rochelle take 12. Sped up to 160 km/h, with 7% schedule padding, nonstop trains would take about 6 minutes. HSR would take 3:35, again padded 7%. This means that, with 2-minute headways in both directions (fast-ahead-of-slow, and slow-ahead-of-fast), and 6-minute express commuter trains, we can have (northbound times from New Rochelle, HSR passing and commuter trains passing or stopping):

HSR :00
Express :02
Express :04
HSR :10
…

This is because by Rye, the :10 HSR is still 3.5 minutes behind the :04 express Metro-North train.

Alternatively, it’s possible but very tight to have an express-express-local schedule, assuming commuter trains are not sped up but the stop penalty is reduced to 80 seconds (so, 4 minutes for 3 stops), which is feasible at the speeds of the line:

Express :00
Express :02
Local :04
Express :10
…

This requires express trains to weave effortlessly to tracks 2 and 5 of a 6-track New Rochelle station, or to stop at New Rochelle.

A mixed schedule, with half the express trains sharing track with HSR and half with the local trains, is also feasible, but is essentially like drinking half coffee and half tea: while the 8-minute local-express gap on the local tracks is fine as it is, the gap on the HSR tracks requires speeding up the express trains anyway.

Note that if the all express trains share tracks with HSR, it is trivial to add local service, or to replace express trains with locals.

The other segment is easier, because it is shorter and lower-traffic, with only 13 tph (3 local, the rest express). HSR would take 2:24 with pad; express commuter trains take 7 today and could take 3:24 with a speed-up. The speed-up would be very significant here as this is a slow segment today, with the movable Cos Cob Bridge restricting speeds. The present speed difference is already almost small enough to allow HSR-express-express, as above; we need to cut another 36 seconds from the express travel time, which we can do with improved reliability reducing padding (the pad I’ve observed between Stamford and Grand Central is 10 minutes). Local-express-express would be 7 vs. 12 minutes, and if locals could consistently take just 11 minutes, as some already do, or if expresses could easily weave to the express tracks just north of Greenwich after HSR trains diverge to the bypass, then it would be feasible.

To finish Easy Mode, let us reconcile the two segments, which after all are populated by the same trains. If we have HSR-express-express on both segments and use the bypass as an overtake, we need to ensure that a commuter train that was 2 minutes ahead of HSR before is now 4 minutes behind after: 4, and not 2, because each HSR train overtakes 2 express trains. Losing 6 minutes is difficult, as the current local travel time between Greenwich and Rye is just 7 minutes. But with approaches this is a bit more than 8 minutes, and HSR would do the segment in 2:08, padded. This has the only drawback of awkwardly making express trains make stops at Greenwich (understandable given traffic), Port Chester, and Rye.

As an alternative, we can also do local-express-express on one segment and HSR-express-express on the other. Since the HSR-express-express schedule is tighter on the southern segment, the southern segment should have local-express-express. This only requires us to avoid having trains run a mixture of local and express too much: as two of the Greenwich-Stamford locals run express south of Greenwich and three of the Harrison-New Rochelle locals only run as far north as Harrison, we can simply combine those to create more locals going all the way from Stamford to Grand Central.

Now, let us move on to Hard Mode, which includes New Rochelle-New York. A consistent 15-minute schedule does not look possible to me on New Rochelle-Stamford without reducing peak commuter traffic to 16 tph, for example by lengthening trains and platforms. If that were done, 8 tph local on the local tracks with 4 turning at or south of Greenwich, and 8 tph express in HSR-express-express pattern on the southern shared segment, would be feasible.

So let us consider 10-minute schedules. HSR and express trains run at almost the same speed, since there are few areas south of New Rochelle on which even 200 km/h is at all feasible. The difference between HSR and express 145 km/h M8s (stopping at New Rochelle and Sunnyside), as already investigated in my original post on the subject, is 3:15 without pad, and 3:29 with. This means the express needs to leave New Rochelle up to 4.5 minutes after the HSR train, so that it will arrive at Sunnyside 8 minutes after, and 2 ahead of the next HSR. With a 9-minute time difference between HSR and the express trains from Greenwich to New Rochelle, this requires the express to be at least 4.5 minutes ahead of HSR at Greenwich, which with very minor speed-up is possible. What this means is that with mixed HSR-express-express and local-express-express as in Easy Mode, the first express after each HSR will go to Penn Station and the second one will go to Grand Central.

The question then is what to do with local trains. If they only go as far north as Co-op City, then it’s easy; with the exception of Hell Gate Bridge, the tracks in the Bronx would have to be four-tracked anyway to allow some overtake, and since there’s room, there’s not much traffic now, and this represents an expansion of Metro-North service, we can safely assume four-tracking. In that case local trains, making no stops in Queens except Sunnyside, would run at the same speed as the express trains on the two-track segment south of Hunts Point, and could be scheduled anywhere. An Astoria stop would require them to be scheduled immediately after the HSR and express trains southbound, which is feasible as near Sunnyside those would be close together already, with the express just ahead of HSR.

Of course, as this is Hard Mode, we cannot assume the local trains turn at Co-op City. Instead, we will make them turn at New Rochelle, or ideally run through farther north as locals. Now, southbound HSR trains, we have established, pass New Rochelle at :00, express trains leave at :04:30 and :06:30, and local trains leave at :02:30. This means we should extend four-tracking at Co-op City such that the local goes at the exact same speed as the express, which does stop at New Rochelle, until after it diverges to the local tracks at Co-op City; alternatively, if there is room in the schedule, we can have the local trains leave at :02 and then there is enough room.

Rearrangement of trains heading toward Hell Gate should be considered a trivial problem: if locals are too far ahead, or too far behind, the number of stops could be adjusted, or they could be held at the Bronx stations longer. Because the time difference between a local and an HSR in the Bronx and Queens combined is just under 10 minutes, the first option would require an extra stop or two or longer dwell times, making the local lose a full 10-minute headway and thus come immediately after the next HSR. By then the express has shifted back to about :07-:08 and so it’s not a threat even if the local does make an Astoria stop.

Conclusion

None of this is elegant. The schedules don’t necessarily match. Local HSR trains would add extra complications, though at least they’d reduce the speed difference with commuter trains in Metro-North territory. All agencies involved need to be on the same page. Through-running would involve a multi-overtake schedule, in which the most local trains get overtaken several times, by different classes of trains (HSR and express). Punctuality doesn’t have to be Japanese, but it needs to be Swiss, or else the entire edifice collapses.

And it’s still far cheaper than trying to overbuild everything to prevent this mess. The only commuter trains sharing track with HSR in this region are Metro-North, and those are fairly punctual, though this involves heavy padding. The rolling stock assumed is already in operation or in the procurement stage. The track repairs required are straightforward, and the curve modifications required, while annoying, are not the end of the world; the one greenfield bypass follows an existing Interstate, and the takings required, while nonzero, are low.

The travel time implied for this is a little more than 17 minutes from Newark to Trenton, for an intercity train stopping only at Newark (though this requires a top speed of 360 km/h causing severe noise impact in New Brunswick and Trenton), a little more than 8 between New Rochelle and Stamford, and just less than 10 between Sunnyside and New Rochelle. Depending on how much speed can be squeezed out of narrow tunnels and a new Portal Bridge, about 11 minutes Sunnyside-Newark, including New York and Newark dwell times, could be done; this is about 46 minutes Stamford-Trenton, a segment that Amtrak currently does in 1:45 excluding the long New York dwell time. And the amount of concrete pouring required is quite small for an hour’s worth of travel time reduction. Even a top speed that’s less noisy and ambitious, and lower speeds through the existing tunnels, do not raise this travel time far above 50 minutes.

Great things are possible if we first look at what is feasible, and then demand that agencies cooperate to achieve it, instead of program everything around public transportation agencies that act like rival gangs. If everything is optimized right, travel times not much higher than those Amtrak is targeting become possible, for a small fraction of the price, and capacity constraints can be kicked down the road to when passenger rail makes enough money to pay to relieve them. Organization, electronics, and small, strategic concrete pouring can go a long way. The choice is not between HSR for a twelve-figure sum and small improvements for an eleven-figure sum; it’s between low-cost HSR and agency turf battles.

Surreptitious Cost Escalations and Spurious Cost Savings

In response to my previous post regarding the extreme cost of Amtrak’s new Northeast Corridor Vision plan, people both on forums and on blogs have said that it’s actually a cost saving coming from bundling the Vision with the earlier Master Plan. Although the original cost was $117 billion and the current one is $151 billion, the current one is still lower than the sum of the original cost plus the cost of the Master Plan, by $15 billion. This looks like a cost saving, but it’s actually not.

The explanation is that the Master Plan still contains elements that are unnecessary if large portions of the line, including nearly the entire New York-Boston segment, are bypassed. The list of projects on PDF-page 21 of the plan contains additional tracks in eastern Connecticut and a replacement of the bridge over the Connecticut, boosting capacity. However, if the intercity trains are removed from the line, there is no need to boost capacity. Low-performing branch lines – and this is what Shore Line East is without intercity trains – can be and are spun off to regional agencies: JR East abandoned the northern reaches of the Tohoku Main Line as it extended the Tohoku Shinkansen, spinning them off to the prefectures to run as it is not interested in running regional rail at the low densities of northern Japan and the intercity functions were all rolled into the Shinkansen.

So in that sense, any cost saving was spurious: Amtrak simply removed some, but not all, Master Plan projects that are obviated by the plan for a bypass. It’s no different from the fact that the Tokaido Main Line and the PLM Line are still double-tracked, as in both cases the national railroad chose to build high-speed rail parallel to them instead of to quadruple-track them to boost capacity.

But on top of that, there is at least some cost overrun implied in the plan. The cost breakdown is not detailed enough to make this clear, but the cost of the Gateway Tunnel is up to $14.7 billion, from $10-13.5 billion last year. It’s buried deep enough that it’s hard to see, or discern what the total overrun is, but it’s there. So Amtrak has a surreptitious cost escalation for the Gateway project at the same time as a spurious cost saving from partially merging the Vision and the Master Plan.

The CAHSR-SNCF Bombshell

The most important HSR news right now is the recent revelation on the LA Times, strategically made immediately after the state legislature had voted to appropriate the required money to begin construction, that the California HSR Authority had brushed off an offer from SNCF, which came with funding attached, to take over and build the project. SNCF’s offer would run trains through I-5 all the way instead of the chosen route vaguely along State Route 99, bypassing Bakersfield and Fresno.

Stephen Smith, who’s talked to the same sources who spoke with the LA Times, says that SNCF was interested in either I-5 or a greenfield alignment just west of SR 99 that would serve Bakersfield and Fresno with edge-of-urban area stations, though I-5 was “the only alignment… that private backers felt was financially viable.”

Although in 2009 SNCF submitted a document proposing to build the project along the chosen alignment, serving Bakersfield and Fresno at city-center stations, the document is stamped “Do not circulate outside government,” and the source says explicitly that the HSR Authority had pressured SNCF not to say anything about alignments, and more recently rejected its I-5 (or west-of-99) proposal out of hand. The HSR Authority responded, brushing off some of the article’s concerns and raising what is essentially FUD: HSR Authority Chair Dan Richard made sure to mention the manufactured controversy over the fact that SNCF had been forced by the Nazis to help ship Jews to extermination camps.

I do not have any access to sources, confidential or otherwise, but at least some analysis of this can be made from public information. The key cost numbers the LA Times provided are,

The I-5 route would have been the shortest, fastest and lowest-cost alignment, with a price tag of about $38 billion — sharply less than the rail authority’s current route, which has been estimated at various times to cost $34 billion, $43 billion, $98 billion and now $68 billion.

The problem: the cost of the Central Valley segment is a sufficiently small portion of the cost that it can’t possibly make the entire or even most of difference between $38 billion and the current price tag. It’s unclear to me what $38 billion should be compared to – 2010 dollars or year-of-expenditure dollars, and the Blended Plan ($68 billion YOE) or the full Phase 1 ($98 billion YOE) – but the lowest number, the Blended Plan in 2010 dollars, is $53 billion, $15 billion more than SNCF’s proposal. I have asked what exactly the comparable Authority number is and will update when I get an answer.

In contrast, the Initial Construction Segment, which includes a large majority of the Phase 1 Central Valley segment (though not the most difficult part, through Bakersfield) is $5.2 billion in 2010 dollars (see PDF-page 15 of the 2010 business plan); the actual money appropriated is just over $6 billion, but if we’re doing YOE numbers then we must compare $38 to $68 and then the difference doubles. Since the cost of construction along I-5, although lower than along the chosen route with its viaducts and grade separations, is nonzero, we get that a relatively small fraction of the cost difference, perhaps a quarter or a third, is attributable to this design choice.

So if it’s not just I-5, what is it, and what can we learn from this? I believe the results should if anything make the HSR Authority look even worse than it already does in light of this story and its lackluster response. This is because it means the entire amount of money required to build to SNCF’s specs but serve Bakersfield and Fresno, at edge-of-urban-area stations if the cities object to the noise of trains through downtown (which at least Fresno does not), is a small number of billions of dollars. This means that if service to those two cities was the true dealbreaker, the Authority could have asked SNCF to change the alignment back to the chosen route or a greenfield route just west of it, and then demanded that Fresno and Bakersfield pay for the difference.

Fresno had been hoping to use statewide HSR money to bundle its own project of grade-separating the freight tracks through the city along the Union Pacific right-of-way. The poor relationship between the HSR Authority and Union Pacific dashed the plans to use its right-of-way where it is superior to the BNSF alignment. That said, the threat of being left out of the network entirely could have induced it to come up with money for this on its own; the segment of the project through the Fresno area is $1-1.5 billion. A downtown station in Bakersfield is more difficult, especially if one gets from the Central Valley to the LA Basin via the Grapevine rather than via Palmdale, but in Bakersfield there are some complaints about the impacts that a downtown alignment would cause, and at any rate even I-5 would come close to serving the urban area.

In addition, portions of the cost savings that do not come from alignment choice have to be attributed to superior cost control. Part of the difference between American and rest-of-world construction costs has to come from more mundane issues such as proper supervision of contractors, since the difference is large and persistent and remains in place even after one controls for such issues as the percentage of the route that is in tunnel. (For example, recall that the Tohoku Shinkansen extension cost $4.6 billion for 82 km, of which a third is just one long tunnel and another sixth additional shorter tunnels).

The other lesson we can learn from this episode is political, regarding cost escalations and strategic misrepresentation. Too many political transit supporters downplay the issue. LightRailNow claims that a cost escalation that occurs before construction starts is not a cost escalation, but just a more accurate cost estimate; Robert Cruickshank did not quite say the same when the 2010 business plan for CAHSR revealed costs had doubled, but came close to it by describing the plan as more careful and thorough. In reality, large bombshell reports shortly after money has been obligated are a hallmark of secretive, untrustworthy planning, precisely the kind likeliest to lie about costs.

The main problem with megaprojects is not the dollar cost. In the grand scheme of things, a lot of them can generate enough social rate of return, and sometimes even a purely financial rate of return; at any rate, even when they are cost-ineffective, they are a small proportion of total GDP. The problem is getting politicians to vote for them. This means that issues such as institutional inertia are in play. It’s harder to get people to rescind money than to get them to vote against spending money.

If the primary cause of cost escalations is unforeseeable challenges, then we will see them come in timed with engineering developments, contract awards, and actual construction. If instead it is strategic misrepresentation, then they will be timed to come just after major political hurdles regarding funding: the passage of a referendum, legislative funding, an electoral victory by a supportive politician. The California HSR bombshells aren’t quite this clean, but they come a lot closer to the outright lying hypothesis.

Northeast Corridor HSR, 90% Cheaper

Amtrak’s latest Next-Generation High-Speed Rail plan is now up to $151 billion, from a prior cost of $117 billion. This is partially a small cost escalation, but mostly including Master Plan upgrades to the legacy line. Per kilometer of route length, this means the project has now crossed the $200 million/km mark, a higher cost than 60%-underground Chuo Shinkansen maglev. The primary cause of the high cost of Amtrak’s project is the heavy amount of deep-cavern urban tunneling: nearly a tenth of the cost is the Gateway Tunnel, a rebranded bundling of ARC into the project, and a similar amount is a similar project in Philadelphia. At least this time they’re serving Rhode Island with a stop in or near Providence rather than Woonsocket.

In contrast with this extravaganza, it is possible to achieve comparable travel times for about one tenth the cost. The important thing is to build the projects with the most benefit measured in travel time reduced or reliability gained per unit of cost, and also share tracks heavily with commuter rail, using timed overtakes to reduce the required amount of multi-tracking.

I propose the following general principles, guiding any future development in the corridor:

1. Rolling stock is cheaper than infrastructure. This is not true everywhere, but the Northeastern US and Japan both have high infrastructure-to-equipment cost ratios. A Shinkansen train today costs about $4 million per car judging by how much Taiwan pays. A 16-car train every 15 minutes from Washington to Boston, with a one-way travel time including turnaround of about 3:30, would require 30 sets, or 480 cars, or $2 billion. Therefore, it makes financial sense to demand more of the rolling stock: some tilting as present on the Talgo, Pendolino, N700, or E5; high initial acceleration as present on the N700-I; and high power-to-weight ratio as on the Talgo and Shinkansen models, or even possibly an all-cars-powered Pendolino.

The difference between an average and a top-rate train could easily amount to 20 minutes between Washington and Boston. Making up those 20 minutes with infrastructure, once the easiest projects have been completed, would cost far more than $2 billion.

2. Speed up commuter trains instead of bypassing them. The place where this is most obvious and useful is the Boston-Providence segment. I have nothing to add that I didn’t already say in my pair of posts on the subject last year. Something similar is true between Baltimore and Washington. It is more difficult between New York and New Haven, but at least there are curves that have to be bypassed anyway, and so the track sharing can be reduced to a manageable degree given the line’s heavy commuter traffic.

This requires fixing agency turf battles, which costs a lot of political capital but is almost free to the taxpayer. In contrast, long multi-track segments, often with new viaducts, easily run into the billions. Amtrak’s single biggest question mark east of New York is the string of tunnels from Penn Station to New Rochelle to Danbury, all so that it doesn’t have to share tracks with Metro-North. It could buy the commuter operations and subsidize them forever and still come out ahead of all those tunnels.

3. The regulations should be based on service needs. As a corollary of #1 and 2 and the every minute counts philosophy they espouse, the regulations should allow trains that can operate safely. Here safety is determined by actual practice and track record, rather than what the FRA thinks safety is, which has an incidental relationship with reality. That Shinkansen trains do not meet UIC standards should not be even a minor issue; trains in Japan are safer than in the UIC’s prime-mover European countries.

4. On shared segments that aren’t bypassed, build infrastructure that allows higher speeds. This is a corollary of #2: if legacy routes are to be upgraded rather than bypassed, there’s no point in assuming present-day speed limits, such as Metro-North’s 75 mph/120 km/h limit in Connecticut, will remain in place, and therefore projects should be built with high radius of curvature. Assume that large portions of the New Haven Line will host trains going at 240 km/h.

5. Make sure station throats allow full speed. Every non-geometric restriction on speed – tunnel diameter, track condition, switches – should be eliminated. Higher-speed switches are cheaper than new concrete pouring; more precise track maintenance is cheaper than most people realize, standing at about $100,000 per double-track-km on average; Shinkansen trains’ noses are designed (and European trains’ noses can be modified) to allow full speed through narrow tunnels, as Shinkansen construction standards minimize tunnel diameter to reduce costs.

The time cost of even a short segment inhibiting full-throttle acceleration in station throats is higher than most people realize. A kilometer a train has to wend at 50 km/h when it could go 200, such as the Penn Station throat, is worth 54 seconds. At stations closer to full-speed zone, this speed-restricted kilometer slows the train’s acceleration to full speed further down the line, and thus it comes at the expense of a kilometer at 300-360 km/h, raising the time cost even further.

6. Fix curves in higher speed zones. This applies mainly to the S-curve flanking I-287 in Metuchen: its curvature is not terrible, but because to its south there are no geometric speed restrictions for tens of kilometers and to its north the curves are also reasonably gentle, its bang for the buck can be surprisingly high.

7. Worry about track capacity when all other capacity factors have been optimized. An intercity railroad that runs 8-car trains is definitionally not at capacity. Running 16-car trains requires lengthening a small number of platforms, most at easy locations. Doubling train capacity across the Hudson chokepoint requires building a new tunnel under the river. Amtrak currently runs 4 trains per hour into Penn Station at the peak; if after everything else has been built it has exhausted the capacity of 4 trains per hour each with 16 cars and a thousand seats, its operating profits will let it pay for any further expansion.

With the above seven principles, one could come up with a reasonable set of projects of immediate significance. With a total cost in the single-digit billions, they’d eliminate most of the barriers to full-speed travel between New York and Washington, and leave New York-Boston with just one major problem section between Stamford and Milford. Best-practice trains, even ones optimized for a straighter route – for example, Shinkansen or the Talgo, but not the Pendolino, which is both heavier and less powerful but has a much larger degree of tilting – could go from Boston to Washington in about 4 hours, or not much more.

Getting this further down to 3 hours would require further investment according to the same principles, but even 4 hours, by virtue of the markets to and from New York, would generate the profits required to pay for them. Moreover, the contrast between fast travel on bypass segments in eastern Connecticut or straight legacy segments in Rhode Island and New Jersey and the remaining slower problem segments would create political will to complete the system. The areas with the most NIMBY resistance should be left for last, because today’s train riders as well as Amtrak itself are not nearly as powerful as they will be if the mostly NIMBY-free projects cut train travel time from 7 hours to 4.

Cost Bundling

It’s common to bundle multiple construction projects into one, either to save money or to take advantage of a charismatic piece of infrastructure that can fund the rest. For example, on-street light rail is frequently bundled with street reconstruction or drainage work, and rail lines can also be bundled with freeway construction in the same corridor (as in Denver) or widening the road they run under (as in New York). Combining different constructions into one project can be a powerful cost saver, as seen in the Denver example and also in Houston.

The problem is when it leads to scope creep. In case there is one charismatic project that carries the rest, it’s always tempting to add more features to the project to get more funding. If the funding comes from a pot specific to one use – in the examples in this post transit, but it could be anything – then it will also lead to a misleading reporting of the total cost, making it look higher than it is. Part of the surreptitious underfunding of transit in the US comes from such bundling, for example parking garages for commuter rail. More commonly the projects in question will be transit, just not necessarily cost-effective on their own.

Because one agency tends to have the lead on such projects, there is no incentive for cost control. The worst case I know of is high-speed rail construction on the Caltrain corridor; the segment from San Francisco to San Jose incurred the highest cost overrun in the system, its cost rising by a factor of nearly 3 versus a systemwide average of 2, and most of the overrun came from tunnels and viaducts reinforcing various agency turf boundaries.

The flip side is bundling projects not so that a charismatic major project can support others, but rather so that a major project can get the support of others by throwing them bones. This is essentially Amtrak’s Vision plan for the Northeast: Gateway is meant to get support from New York and New Jersey now that ARC is canceled, Market East is meant to get support from Philadelphia on the dubious idea that the city wants a Center City stop, and so on. In this case, there is a symbiotic relationship: the charismatic project, in this case HSR, gets to brand all these separate projects as necessary for a grand goal, while the presence of the smaller project ensures that local politicians, whose priorities rarely include providing intercity transportation maximally efficiently, support the project.

The Recession Won’t Last Forever

The article about New York State’s decision to discontinue studying high-speed rail between New York and Buffalo is by itself not terribly surprising. Although Andrew Cuomo likes flashy public works projects, of which HSR is one, he is consistently pro-road and anti-rail.

The study released by the state sandbagged actual HSR on cost grounds – it did not provide any further analysis, and in two ways (lower average speed than most HSR lines, and a requirement for tilting) stacked the deck against it – but instead looked into medium-speed rail, with top speed of 110-125 mph, which is frequently misnamed HSR in the US. This, too, is not surprising. State DOTs have no idea how HSR works, and tend to make mistakes, not know how to do cost control, and so on.

What’s most surprising is the explanation for why not to do anything substantial: as one of the HSR proponents quoted in the article complains, “The State of New York is worried about making ends meet; the economy is not doing so great. That’s the reason in the short term.” Taking his argument at face value, the state is refusing to advance study of an HSR line because economic conditions are bad now, a decade or more before such line could even open.

The recession won’t last forever; if it does, there are bigger things to worry about than transportation. Other than immediate reconstruction projects, for which the environmental reviews are fast-tracked, major projects take years to do all the design and environmental studies. California has been planning HSR since the late 1990s. It intended to go to ballot in 2004, and after delays did so in 2008. HSR is scheduled to break ground later this year, assuming the state does not cancel the project. An HSR project for which planning starts now will start construction after the economy recovers not from this recession, but from the next one.

The recurrent theme in the article is the state’s preference for mundane over flashy projects, but rejecting HSR shows the exact opposite.Starting planning now costs very little. In fact, the best thing any state agency can do is keep planning multiple big-ticket project contingencies pending an infusion of money; this way, it can dust off plans and execute them faster if there’s a stimulus bill in the next recession. That’s long-term planning. Refusing to advance construction because it won’t start until long after Cuomo’s Presidential run in 2006 2016 isn’t.

Of course, the same goes in the other direction. Too many people, building on Keynesian stimulus ideas, want massive infrastructure spending now as a public works program. For example, Robert Cruickshank (and in comments, Bruce McF) argues for long-term benefits coming from the stimulus effect. Although construction in 2012-3 would have an impact, a multi-decade project spanning periods of both growth and recession should not rely on estimates of job creation solely from periods of recession. On the contrary, economic costs and benefits should be based on a long-term multi-business cycle trend.

I propose the following principles for interaction between business cycles and very long-term investment:

1. Assume your project will be undertaken in a period of close to (but not quite) full employment, in terms of both funding sources and economic effects, unless you specifically intend to advance construction in a recession.

2. If you want to use a recession to lock in lower interest rates, higher job impacts, or lower construction costs, make sure you have a shovel-ready project, or else try to advocate for better staffing at the requisite regulatory agencies well ahead of time so that they can fast-track it.

3. Treat fiscal surpluses coming from an economy at full employment as one-time shots rather than an ongoing situation that can be used for regular spending or tax cuts. Growth doesn’t last forever, either.

Plan B for HSR

Now that the California state legislature’s dragging its feet on releasing the state’s money for high-speed rail, there’s talk about a Plan B. The official Plan B, supported by the chair of the State Senate’s transportation committee, is to redistribute most of the money from the Central Valley toward the train stations of San Francisco and Los Angeles. Since the federal money was conditioned on sending everything to the Central Valley, and a last-ditch Plan B is unlikely to get USDOT to change the rules, most likely the actual Plan B is to kill California HSR and redistribute the $3.3 billion in federal elsewhere within the US. Illinois and North Carolina both want money for their medium-speed projects, and Amtrak wants money for Northeast Corridor improvements.

Because the Northeast Corridor improvements Amtrak wants are not necessarily the most cost-effective, I think it’s most paramount to look for projects that are in the intersection: part of the Master Plan, ideally as ready as possible (e.g. ones that are considered state of good repair), but also compatible with future upgrades to full HSR standards. In particular, this means no investment in parts of the mainline that should be bypassed in the future, but high investment in parts that shouldn’t.

Although the cost projected by Amtrak for these upgrades is in all cases higher than it should be, the high value of investment in the Northeast Corridor is such that they are still cost-effective. This is similar to Second Avenue Subway, which despite the immense cost has such a high projected ridership that its cost per rider is fine, if higher than it should be.

Projects that are to my knowledge still in progress, such as Portal Bridge, are excluded. The same is true of projects that are too big or too cost-ineffective at present construction costs.

Constant Tension Catenary

Cost: $1 billion for “high-speed territory,” which appears to be a small subset of the New York-Washington mainline; including related upgrades, just the 40 km between New Brunswick and Morrisville that are already funded are $450 million. For the full line, figure $2-3 billion. The non-US cost should be about $1 billion, but because of benefits, paying the premium is worth it.

Benefit: higher reliability in summer. No limit to top speed except for the curves; although present-day rolling stock can only do 150 or 160 mph (240-255 km/h), up from 135 mph (215 km/h) allowed by the existing catenary, the time savings for future rolling stock capable of higher speed are substantial. The more curves are fixed along the line, the greater the benefit.

New Haven Line Bridge Replacement

Two bridges (Devon, over the Housatonic River, and Cos Cob, over Mianus) require replacement; two more (Saga, in Westport, and Walk, in Norwalk) require rehabilitation. Except for the Walk Bridge, which can be bypassed on I-95, those bridges should carry high-speed traffic in the future.

Cost: unclear – the plan says $4.4 billion for many projects on the New Haven Line, and a separate breakdown only says that replacing both the Saga and Walk Bridges costs $600 million. For what it’s worth, replacing the (two-track) movable Connecticut River Bridge with a high-level fixed bridge is pegged at $500 million, over a wider river.

Benefit: higher reliability and capacity. No speed limit on unpowered bridges, versus the 40 mph (65 km/h) limit today. More subtly, on both sides of the Cos Cob Bridge there are short, sharp curves; rebuilding the bridge as a high-level bridge with a single very gentle curve imposing no speed limit could be done more or less within right-of-way, though the Cos Cob station platforms might have to be moved slightly. Even more subtly, more reliability means less padding on both Metro-North and Amtrak’s part, and with federal funding obtained by Amtrak this can potentially allow intercity trains to go at a higher speed elsewhere on the New Haven Line than Metro-North currently permits.

The segment between the NY/CT state line and Stamford is in my experience the slowest on the Northeast Corridor outside immediate major-station areas, and when I timed the trains on it, the northbound trains did it in about 11 minutes, for an average speed of 60 km/h. The curves immediately west of Stamford are actually fairly gentle, and letting the trains run on this segment at speed could nearly halve this travel time. While this would require higher cant and cant deficiency than the low values currently used on the New Haven Line, there’s little point in raising them while speeds are so limited on the Cos Cob Bridge.

B&P Tunnel Replacement

The tunnels immediately west of Baltimore were poorly engineered and impose a tight speed limit, slowing down trains by about 1.5-2 minutes even though they are adjacent to a station. While this is a relatively straightforward project, it may not be sufficiently advanced in the design and environmental clearance phase, making it a candidate for future funding but not for stimulus funding.

Cost: Amtrak’s Master Plan says $1 billion. The FRA’s study on the matter says $770 million. Both figures are within the normal non-US range for urban tunneling of this length, though the Amtrak figure is toward the upper end of it.

Benefit: reliability, and on the margins some extra space for intercity trains to pass commuter trains (on the margins, because for the next two stations south of the tunnels there are four tracks already). Some trip time improvement, and even more trip time improvement if there is new high-acceleration rolling stock, for which speed limits in station throats add more to trip times. Reduction in maintenance costs – curves as sharp as those in the existing tunnels (about 250-meter radius) begin to wear the wheels of trains, and the best available future rolling stock, Shinkansen trains, has the highest minimum curve radius, though it is well below 250 meters (I believe it is 190).

Pelham Bay Bridge Replacement

Cost: $500 million together with curve modifications between New York and New Rochelle. Just repairing the bridge more, which is not the same as replacement, is $100 million.

Benefit: like Cos Cob, Pelham Bay is flanked by two sharp, short curves. Replacing it even without doing anything else would eliminate a speed restriction in a zone that for a few km could support 200 km/h.

Medium-Term Future

There are additional projects that can be undertaken, in relatively small chunks. Some have been hinted at; some haven’t been studied at all that I know of, but have tantalizing benefits for future high-speed service. Because there’s no design yet, except possibly for Elizabeth, it’s unlikely anything can be done by any deadline, but design should begin promptly to make the next round of funding. At any rate, the above shorter-term projects are more than enough to soak up all funding that could become available if California fails to appropriate money for its own HSR project and returns the federal funds.

New Rolling Stock

The Acelas are heavy, low-capacity, low-performance, and high-maintenance. New trains can’t be FRA-compliant, and in practice some time (measured in years, not decades) can pass before the best rolling stock is legal on US track. But Amtrak and all involved in HSR on existing track should be at the forefront of asking for an overhaul. High-acceleration trains, capable of about the same cant deficiency as the Acela (for example the E5 Series Shinkansen and the high-speed Talgos), can achieve much faster trip times than possible today, with trivial changes to right-of-way geometry. Of course the tracks would have to be maintained to higher standards, but that’s much cheaper than moving a viaduct or carving a new right-of-way through a residential suburb.

Elizabeth S-Curve Modification

Cost: ??? The project would entail stretching the present reverse-curve, and probably demolishing all or parts of Union County College’s Elizabeth Kellogg building, a new medium-sized building that cost $48 million to build, as well as a parking garage between the college and the train station. The chief difficulty is easing a curve that’s on a viaduct.

Benefit: current speed limit on the curve is 55 mph (90 km/h), and because the limiting factor is not radius but how fast one can reverse a curve, there’s not much that can be done by raising superelevation. If only the above two buildings are removed, and some parking lots are taken, the curve appears to be modifiable to a radius of about 1,500 meters, which with cutting-edge superelevation (200 mm) and the E5 or Talgo 350’s cant deficiency (about 175 mm) corresponds to 220 km/h. This effectively extends the high-speed zone in New Jersey farther north, closer to Newark.

An express New Jersey Transit train taking a curve with radius 1,500 meters and superelevation 200 mm at its top speed of 160 km/h would have perfectly balanced cant to within a millimeter, and so there is no need to reduce cant to accommodate it.

Metuchen S-Curve Modification

Metuchen is Elizabeth’s shy, ignored sister. Amtrak’s Vision travel time simulation does not fix the curve at all. Update: as Jim notes in comments, the Master Plan does talk about some fix, calling it the Lincoln Interlocking. The total cost of this, Elizabeth, constant tension catenary, additional curve realignments in Pennsylvania and Connecticut, and other projects Amtrak identifies as immediate trip time improvements is $4 billion, of which a portion has already been allocated.

Cost: ??? The project entails straightening two reverse curves, an easier one between Metropark and Metuchen and a harder one on both sides of I-287. Some residential takings may be required, especially for the former; the latter may require partial takings at a strip mall and an industrial building. Since the railroad is not on viaduct here, structure costs should be far lower than in Elizabeth.

Benefit: current speed limit is 100 mph (160 km/h). The S-curve is not as tight as at Elizabeth and this means there’s more potential for an increase in speed, but not too much. With minor takings, the curves in the area can all be straightened to 2,500 meters (280 km/h) except the I-287 curve, whose maximum feasible radius depends on how many takings are allowed; with very superficial takings, 1,800 meters (240 km/h) is possible, and with completely taking the strip mall and industrial site there’s practically no limit. Although the existing speed is much higher than at Elizabeth, this is smack in the middle of an otherwise full-speed zone, and so the benefits of speed boost are higher.

Second update: I forgot to say – with the same assumptions as for Elizabeth, a 160 km/h NJT express would have 17 mm of cant excess on an 1,800-meter curve and 80 mm on a 2,500-meter curve, both lower than the cant excess of stopping trains on some of the curvy stations in southeastern Connecticut.

Port Chester-Greenwich Bypass

Most of the slowness of the segment between the NY/CT state line and Stamford comes from Cos Cob, but part of it comes from a sharp curve in Port Chester that can’t be modified without too many takings. As an alternative, trains should leave the existing line just south of Rye, travel along I-95 and its gentler curves, bypass Port Chester and Greenwich, and rejoin in the vicinity of the newly-raised Cos Cob Bridge. Curve radius without significant residential takings would be about 1,300 meters through the I-95 S-curve in Rye and Port Chester, and 2,000 meters elsewhere.

Cost: ??? This is about 7 km of new line, with significant portions on viaduct. Parts of the Greenwich station house may need to be knocked down or moved.

Benefit: the direct benefit is bypassing two curves in the middle of what would be, in the presence of a fixed Cos Cob Bridge, a relatively high-speed segment. The indirect benefit is that it gives intercity trains several fast kilometers to overtake express commuter trains. Not only does this boost reliability, but also, like the Cos Cob fix, it makes it possible for intercity trains to travel faster elsewhere on the line without mucking up commuter trains’ schedule. Currently permitted top speed in Metro-North territory is 90 mph (145 km/h) in New York State and 75 (120) in Connecticut, but those curve fixes would allow much higher speed on a long continuous segment. With higher superelevation, current curvature would permit a continuous 200 km/h south to Harrison and north to Stamford, 170 km/h through Harrison south to New Rochelle, and 160 km/h through Stamford.

New Rochelle Interlocking Grade-Separation

Cost: ??? Harold Interlocking, a more complex project, is about $300 million. But this project conversely would require minor curve modification between New Rochelle and Pelham Bay for full benefit, and also some takings through New Rochelle to straighten the existing S-curve. Ultimate cost depends on how much straightening is involved.

Benefit: current speed through the interlocking is 30 mph (50 km/h). The flat junction also leads to capacity constraints at rush hour, limiting intercity train movements and forcing them into slots that may be suboptimal in other parts of the line. Depending on how many takings one is willing to engage in, an S-curve with enough space to fully reverse the curve could have a radius anywhere from 700 meters (150 km/h) and up. 700 meters represents minimal takings; the point of diminishing returns is about 1,800 meters (about as much as other curves farther north can be eased to, permitting 240 km/h), which would require taking a row or two of buildings east of the tracks.

Eastern CT I-95 Bypass

Not a small project at all, but it can be broken into segments, some of which allow postponing or canceling projects on the existing Shore Line. In addition, Connecticut wants to widen I-95 in this area from 4 to 6 lanes, and since the capacity of HSR is much higher, the money can be reprogrammed without net loss of auto capacity.

This project would start right at New Haven Union Station, cross the Quinnipiac River at a new bridge near US 1 and the new I-95 bridge, follow I-95 to the state line, and then cut across barely-populated territory to the Shore Line at Kingston, where it straightens.

Cost: this is 121 km of tunnel-free route, and based on similar costs in Europe, it should be $2.5 billion. Carefully tracing through the unit costs implied by the Penn Design group, following California HSR costs, produces a figure of $2 billion. But this assumes much lower costs for the bridges over the rivers than Amtrak has produced so far; Amtrak costs are likely much higher, though not by orders of magnitude.

Benefit: New Haven-Providence in about 40 minutes, New Haven-Boston in about an hour. Current travel time can be improved using better rolling stock, but there’s a point of diminishing returns, and reliability with present-day movable bridges, especially over the Connecticut River, is low, requiring extra schedule padding.

The three basic segments of this are New Haven-East Haven (i.e. the Quinnipiac bridge), East Haven-Old Saybrook, and Old Saybrook-Kingston; the Old Saybrook point comes from the fact that I-95 and the Shore Line are close there and there’s room for a track connection. The eastern segment bypasses the curviest segment with the worst bridges, but requires difficult bridges of its own; that said, the Penn Design methodology, under which a single bridge over a river is not as expensive as multiple grade separations, makes this segment look cheaper than it probably is. The western segment offers new capacity for commuter rail in the New Haven area, because it completely removes Boston-bound trains from State Street and points north.

Commuter Rail-HSR Compatibility

Cost: ??? This involves strategic four-track overtake segments; see example for the MBTA here and here, and for Metro-North to Penn Station here. For comparison, 17 km of four-tracking the three-track gap between the Devon Bridge and New Haven is $15 million, and 8 km of three-tracking the two-track Readville-Canton segment is $80 million. The much higher cost of the latter project presumably comes from the fact that this is new track rather than what appears to be restoring a fourth track that used to exist. But since those four-tracking segments are quite short, not much longer than a station and approaches, the cost of each should be in the low tens of millions.

In addition, MARC and especially the MBTA would need to obtain more modern rolling stock, to minimize infrastructure costs. An 8-car EMU is $20 million at Metro-North/LIRR/SEPTA costs (as well as the costs of European countries; American EMU orders are hardly more expensive than European ones, in contrast with the situation for infrastructure). That said, operating costs would be reduced due to lower energy consumption and a lower breakdown rate.

Benefit: de jure only capacity, but de facto those are busy commuter lines and intercity traffic should not take absolute precedence. As a result, those overtakes are crucial for letting HSR run at the full speed allowed by right-of-way geometry, rather than at reduced speed to avoid interfering with regional traffic. The new rolling stock and more rigorous operating schedule would also speed up regional trains significantly; MBTA trains could run from Boston to Providence in 51 minutes, down from 1:10 today, even while being overtaken by HSR twice during each run and making 3 additional stops.

The Urban Geography of Park-and-Rides

The urban geography of transit cities and of car cities is relatively well-understood. In a transit city, there will be a strong CBD surrounded by residences with spiky secondary centers, all quite small geographically but dense, centered around train stations and junctions; because density is high throughout, minor trips are done on foot. In a car city, all trips are done by car, the core is weak, and most employment is in suburban edge cities and edgeless cities.

What I haven’t seen is an explanation of how urban geography works in mixed metro areas: there are those in which short trips are done on foot and long ones in cars, such as new urbanist developments, and those in which short trips are done by car and long ones on transit, such as park-and-ride-oriented commuter suburbs. It is the latter that I want to address in this post.

The first feature of park-and-rides is that of all combinations of modes of transportation, they are the fastest and enable suburbs to sprout the farthest from the center. This is because the segment of the trip done in a car is uncongested and so driving is faster than transit, while the segment done on a train parallels a congested road, and conversely makes few stops so that average speeds are high.

On top of this, because intra-suburban trips are done by car, the density in the suburbs is very low, comparable to proper car cities (see the lower end of the density profiles of the New York, Chicago, and Boston metro areas), and this forces sprawl to go outward. New York is the world’s most sprawling city measured in total built-up area; the only other city of comparable size that’s not a transit city or a bus/jitney city is Los Angeles, which is forced to have denser suburbs because of the mountains. Of course Houston and Dallas sprawl even more relative to size, but because they lack New York’s transit-oriented core, there’s an inherent limit to their size.

The other feature is that there’s a definite socioeconomic history to the development of the auto-oriented commuter suburbs of transit cities. First, people move to the suburbs and commute into the city, almost always by train due to road congestion (or, as in the earliest New York suburbs, because mass motorization hasn’t arrived yet). The mass exodus into these suburbs comes from cars rather than commuter rail, and so the local services for people living in those suburbs are built at automobile scale, rather than at the walkable town center scale of 1910.

In North America there’s also a definite class element here – the early movers are the rich rather than the poor. Historically this was partly because poor people couldn’t afford regular train fare, and partly because the impetus for suburbanization was idyllic country homes with access to urban jobs rather than cheap housing for the poor. If I’m not mistaken, this wasn’t the case in Australian cities’ suburbanization, leading to a more urban transit-style mode of running mainline rail. The result of this class distinction is that North American commuter rail styles itself as for the rich: agencies make an effort to ensure everyone has a seat and downplay comfortable standing space, and the expectation is that transit is a last-ditch mode of transportation for when cars just don’t have the capacity to get people downtown, and so nobody needs to take the trains in the off-peak or take a bus to the train.

The result is that the park-and-ride city will still have a strong core with high-capacity transportation, and the primary CBD will maintain its supremacy for high-income jobs. Establishing edge cities in the direction of the favored quarter can happen, but there’s still a congested city nearby, and so from many directions it’s impossible to drive, and taking transit is impossible. Thus jobs in White Plains and Stamford are not nearly as high-paying as jobs in Manhattan.

There can even be secondary CBDs, if the inner part of the metro area, where people take transit more regularly than the suburban commuters do, is large enough. But those secondary CBDs are frequently quite auto-oriented. Brooklyn’s mode share for jobs is only 42-39 in favor of transit (for residents, it’s 60-25), and all other counties in the New York region except Manhattan have more workers driving than taking transit, a situation that is not true if one looks at residents. Those secondary CBDs then have mixed characteristics: they are dense and fairly walkable, as can be expected based on their history and location, but also have plentiful parking and a large share of drivers demanding even more. They can accommodate multiple modes of transportation, but driving is more convenient, and from the suburbs the commuter rail system isn’t always geared to serve them.

Core Connectors and In-Between Neighborhoods

In some American cities, new or proposed transit lines are either core connectors, i.e. city-center circulator streetcars built for development purposes, or far-flung commuter rail extensions with few urban stops. Both are present in Providence, with the South County extension of the MBTA and the Core Connector, but worse circulators than in Providence are proposed elsewhere (for example, in New Haven), and exurb-focused commuter rail with parking lot stations is the standard in most Sunbelt cities and also in Massachusetts. At first I thought my opposition to both was just a matter of wonky support of a specific stop distance and service pattern that falls in between those two extremes, but recently, after attending Providence urbanist blogger meetings and also rereading old threads about New Haven, I realized there’s a political and social dimension to all this.

Recall that old American cities have a donut-shaped income distribution: gentrified in the center, poor in most other urban neighborhoods and inner suburbs, and middle-class to rich in most suburbs. Those two forms of bad transit are specifically built to cater to the rich parts of the metro area, and ignore the poor parts. The problem, of course, is that the poor parts are precisely where transit ridership is concentrated. People in the gentrified cores of smaller cities can walk; people in the suburbs own cars, and those cities have too many roads and too much parking for buses to be an even semi-reasonable alternative.

In Providence, as I recently brought up, the busiest buses follow Broad and North Main, and serve working-class and poor populations. The same is true in New Haven: the busiest line by far runs on Dixwell, connecting the Yale student ghetto, the in-between poor neighborhoods, and the strip malls in middle-class Hamden. So what service addition does a study by the South Central Regional Council of Governments (SCORAG) propose? Naturally, a circulator connecting Union Station with the New Haven Green. You could chalk this up to a belief in systemwide upgrades over building a few high-performance lines, but many outlying bus stops have no shelter, and the study says nothing about that.

When Peter Brassard first pitched the idea of a local rail shuttle service in Providence and its inner suburbs to us privately, the observation one of us made (I think it was Jef Nickerson, but I’m not sure) is that it would invert the usual relationship between infrastructure investment and income. This is mostly accidental – the mainline serves Olneyville and Pawtucket but not the East Side. But something like this is more likely than not when the focus is on serving reasonably dense neighborhoods and perhaps inner-suburban malls outside walking range.

The same is true of what I believe to be the most promising rail shuttle service in New Haven – namely, a service using the Farmington Canal Trail, which runs about 200 meters east of Dixwell, and could be reused by light rail reaching downtown New Haven on city streets or rapid transit connecting to the mainline with a very short tunnel or trench. With a stop spacing of a little less than a kilometer, modern rolling stock could average 35-40 km/h in service, double the speed of the current bus.

I suspect part of the bias against such service comes from the belief that building ten kilometers of light rail is expensive. Because there’s an implicit hierarchy in planners’ mind between services, they think a downgrade is an automatic cost saver, even when it’s not – for instance, when a bus on an abandoned railroad costs far more than most rail reactivation projects do. One of these mantras is that commuter rail infill is less expensive (and then they build infill stations at $100 million apiece, strategically located away from the intersection with the main bus corridor). As a rule of thumb, each of these downgrades just raises unit costs because of various overbuilding schemes until total cost is the same as if they’d built regular urban rail, but the benefits are much lower.

But it’s more than a technical bias; it’s also political bias. The Core Connector is explicitly a development project. It may even be a successful one, if it convinces local power broker Colin Kane to drop plans for building 7,000 parking spaces in the Jewelry District, as described in a recent paywalled article in Next American City. Development projects like this never go to extant low-income neighborhoods, unless there’s an explicit effort at gentrification, and usually locals protest against the displacement; neglect is much easier and less controversial than redevelopment.

The technical and political biases merge in one of the less challenged cost-effectiveness metrics, the cost per new rider. Although it’s presented in neutral terms – the cost is compared to the predicted total transit ridership if the project is built minus the predicted total if it is not – the results privilege adding choice riders (that is, those who already own a car and drive to work) over retaining existing riders. Although transit revivals happen, most of the world’s transit cities built out their systems before most people got cars, and people simply kept using transit instead of buying cars even as they moved into the middle class. Portland may have about the same metro area transit mode share as before it built light rail, but other cities of similar age lost ground and have even lower transit use.

It’s tricks like ignoring retention that lead Boston to downrate replacing the southern half of the Silver Line with light rail on its list of possible projects even though it would be very cheap by US standards per rider, and rate new commuter rail branches well beyond the continuous built-up area as more cost-effective. The rail bias factor implied by the computation for new riders is less than 0.5%: 130 new riders against 34,000 existing ones. A Transportation Research Board analysis finds the rail bias is in the 34-43% range. I suspect that if the Silver Line served richer areas than Roxbury, Boston would use a more reasonable rail bias than 130/34,000, bringing down costs per new rider by two orders of magnitude. New York went ahead with Second Avenue Subway; it is undoubtedly the most important subway project in the region, but the next best corridors, e.g. Utica, serving less chic neighborhoods than the Upper East Side, are ignored.

The technical reason to build urban rail a certain way – own-right-of-way, stops roughly every kilometer within the city, etc. – is of course separate. Technical characteristics do not tell you which neighborhoods to serve, not without first looking into existing demand patterns. It is just fortunate that New Haven has a right-of-way closely paralleling Dixwell, and unfortunate that Providence has none paralleling Broad. But the income donuts, and more generally the connection between density and old industrial development that is usually working-class (since gentrification in such cities is within walking distance of the core rather than within transit distance), have certain social implications. The most annoying to the planner and the government official is that they must invest in poor neighborhoods as they are, and do not have a special reason to try to foist change upon them.

Or they can just build core connectors for the cities and park-and-ride extensions for the suburbs. The FTA will fund these no matter what; its cost-effectiveness metrics are biased that way to avoid having to send every penny it has available to a few expensive but high-ridership lines such as Second Avenue Subway. The developers will like them, because of real or imagined property value benefits. The state will like them – state governments are dominated by suburbanites and urban developers and view transit as pork rather than as useful spending based on ridership metrics; Rhode Island is much likelier to find support for development in the Jewelry District than for boring rail lines in already-developed Providence neighborhoods. It’s a win-win for everyone except the riders, and they don’t count.