There are recurrent discussions of how to best connect public transportation to airports; I, too, have made my comments both on how desirable such connections are and how to best build them. What I think is less discussed is how to build airports in a way that makes it easier to serve them by public transit. Airport authorities spend billions every few decades rebuilding terminals, sometimes even moving the entire airport to a new location, but they never consider how to do so in a way that makes transit access the easiest; this means airport access is done by car, or another high-cost scheme must be implemented to bring a rail line to the airport. Now that there is a plan to replace the Newark AirTrain for a billion dollars, just twenty years after it was first built, it’s worth discussing what capital projects on the airport side should facilitate transit access.

First, recall from previous discussions on this blog that the best way to serve a major international airport is by a mainline train, which is capable of both providing fast service to the CBD (where most inbound air travel is headed) and to many suburbs (which have outbound travelers).

However, we can say more: it is better, other things being equal, for the airport to be on the way, rather than at the end of a line. If the airport must be at the end of a line, it should not be far from where the line would’ve ended if the airport were not there. For example, LaGuardia is a few kilometers east of the end of the Astoria Line, which can be extended. Vancouver’s airport is on a short branch of the Canada Line, which would have been built to Richmond Centre even without the airport.

An even better example would have been Floyd Bennett Field, just past where the Utica subway should end; there were plans in the 1930s to build such a subway, but not only were they never realized, but also Mayor LaGuardia preferred to build the airport that currently bears his name, for easier auto access to Manhattan than Bennett Field had. Thus we can catalog the decision to open the new airport and close Bennett Field as bad for transit access, and oppose similar moves when cities today propose them. The best location for an airport, from the point of view of transit access, is near a subway or commuter rail station, ideally close enough that no further people mover is required.

Let us now discuss internal airport design. I claim that, to maximize transit accessibility, airports should have just one terminal (or several terminals that can be served from the same station), or, failing that, one dominant terminal, as at such fortress hubs as Detroit, Frankfurt, and Charles-de-Gaulle. The reason is that trains are slowed down by additional stations, whereas cars are not slowed down by additional bays and driveways. Mainline trains, in particular, rarely make more than one stop at an airport, and in the cases I know of where they do, the airport is at the end of a branch (such as the RER B and the lines serving Narita), rather than on the way.

This introduces some tension into airport design. Large airport terminals are dendritic, to maximize the perimeter available for gates and jetways; in some cases, they feature satellite terminals, connected to the main terminal by underground passageways, people movers, or even landside buses (as at Charles-de-Gaulle). I encourage people to look at satellite images of Frankfurt, O’Hare, Atlanta, Zurich, and Charles-de-Gaulle. Frankfurt’s Terminal 1 is a kilometer from entry to the farthest branches to the west. This creates some demand for quicker small terminals, which are harder to serve by rail. In addition, the most efficient dendritic design has branches coming out from the center in every direction, except perhaps one direction for an access road; this makes it harder to be on the way of a rail line.

I think it is telling that the single- or dominant-terminal design is less common at airports that are not a single airline’s fortress hub. Haneda and Narita have two major terminals each, one used by Japan Airlines and one used by ANA. Madrid has four terminals, one for Iberia and three connected ones, sharing a Metro station, for competitors, including several low-cost airlines. In all three cases, there are two train stations per line connecting to the airport (with the understanding that Narita has multiple lines, operating by competing railroads).

Usually, airports make an effort to group airlines by alliance. Thus Charles-de-Gaulle and Frankfurt put their respective dominant airlines and partners in their main terminals, and competing airlines in smaller terminals; and Narita makes sure to group Star Alliance airlines with ANA and Oneworld airlines with JAL. Among the largest airports of Europe, Heathrow is the big exception, since it organizes terminals by alliances but splits Oneworld between Terminal 5 for British Airways and Terminal 3 for the rest.

In the US, this is not common, with some exceptions such as Detroit and JFK’s Terminals 7 and 8. This is because the US does not permit connecting air passengers to transit its airports. All passengers arriving at a US airport from a foreign airport without preclearance, even ones in transit, have to go through immigration, collect their bags, go through customs, recheck their bags, and go through security again; between the inconvenience and the real risk of literally being disappeared, few people connect in the US between two foreign countries. Hub terminals elsewhere facilitate easy transfers by maintaining large international areas where passengers can walk between gates, and keeping the passport controls between the international and domestic terminals short. Regardless, even with the vagaries of American immigration policy, it is easier to connect without having to go between terminals; moreover, for passengers leaving the US rather than arriving, the situation is if anything easier than in Europe since there’s no passport control at exit.

Let us now apply these concepts to New York’s two main international airports. Newark may be a fortress hub, but it is not configured as one; United and its Star Alliance partners are sprawled across all three terminals. Moreover, the terminals are just far enough from the commuter rail station to require a people mover. Since it’s better for an airport to be on the way, and have just a single terminal, what this suggests is that Terminal C should be lengthened to approach the train station.

There is currently a plan to replace Terminal A, for $1-1.25 billion of construction budget and $2 billion total development budget. Under this single-terminal paradigm, the terminal should not be redeveloped. Instead, it should be demolished, and replaced by extensions of Terminal C to the west, with additional concourses and piers both to the north and to the south, replacing the current road loop serving the terminals. People would arrive by road via US 1 or by rail via the commuter rail station. Security checkpoints would be conducted at a building just west of Route 1, and the airside terminal’s western end would be an overpass over the road. Rail passengers would have enclosed overpasses to the checkpoints; there would not be any need for a people mover, only moving walkways given the distance between the station and the terminal’s current eastern end. There is enough space for the new concourses to also replace Terminal B, which is of similar vintage to Terminal A.

At JFK, the situation is different. First, it is not a fortress hub. Its top three carriers – JetBlue, Delta, and American – are all reasonably happy with their terminals (Delta’s terminal is 4, not 2, which it is abandoning). British Airways is considering abandoning Terminal 7 and joining American at Terminal 8. Consolidating the airlines that use Terminal 1 at Terminal 4 is impossible until the US resolves its endless immigration lines, which at Terminal 4 are often longer than an hour.

Second and more fundamentally, the transit access situation there is good enough. JFK is far from any subway or commuter rail line, so the only way to serve it by rail is by a dedicated people mover, of which the AirTrain is not bad. The connection to Jamaica approaches the “be on the way” maxim well, since Jamaica is central to the LIRR network and has fast service to Manhattan on the subway as well. Some transit advocates in the region periodically propose a direct subway or commuter rail line to replace the AirTrain connection, but such plans always run against network design issues, since the branching is set up in a way that reduces frequency to Jamaica, a more important station. Given that there must be some people mover connection, traveling in a circle among the terminals is not terrible; straightening the route has some benefits, but the cost of rebuilding the infrastructure is almost certainly too high to be justified.

Update: James Sinclair argues convincingly that the Newark AirTrain is not really at the end of its life, but Port Authority is saying that to justify spending billions of dollars on a better replacement, including either a PATH extension to the airport station (which is largely dead) or an extension of the AirTrain to Newark Penn Station, as a sweetener for United.

I’ve written a lot of articles over the years about what should be done with regional rail in the New York area, focusing either on the overall shape of the system (as on The Transport Politic) or on specific aspects of the central links (as in past posts here). I’d like to synthesize these ideas into one coherent proposal. Unlike my posts on The Transport Politic, I’m going to pay relatively little attention to how to match branches for ridership, but more attention to what to do in a central region consisting of the city proper, New Jersey as far as Newark, and New Rochelle. I will also indicate things that can be done to keep construction costs under control for a plan that includes 30 kilometers of urban and underwater tunnel, about six times as much as the planned tunnels across the Hudson.

The Principles, Restated

The most important principle for infrastructure planning in developed countries is organization before electronics before concrete. In New York, it’s possible to squeeze some extra capacity out of the first two: notably, the LIRR and Amtrak together only run about 40 trains per hour into Penn Station from the east on four tracks, whereas the maximum capacity is about 50, and this is before trains are diverted to the East Side Access tunnels to Grand Central. The LIRR’s bottleneck is not the East River Tunnels, but the platforms at Penn Station, and this means it’s possible to use improved operations, including through-running, to squeeze extra capacity even before East Side Access opens.

However, the biggest bottleneck in the region is from the west, across the Hudson, and there, present traffic peaks at 24-25 trains per hour on just two tracks. I know of limiting cases in which mainline operations achieve about 30, using moving-block signaling on captive tracks (e.g. the RER A, which shares tracks with nothing else in its central segment), with one example that uses fixed blocks (the shared RER B and D tunnel achieves 32). Here, concrete is unavoidable, so new tunnels are required. In addition, providing service to more points than Penn Station, or Grand Central for commuter lines connected to it, requires new tunnels as well.

However, this new infrastructure should be built economically. The posts I linked to in the initial paragraph of this post provide some ideas, including the use of large-diameter tunnel boring machines to reduce station construction costs, and the use of the existing station cavern at Penn Station. This should be paired with seamless fare and schedule integration, including through-routing, and a fleet replacement plan to get rid of locomotive-hauled trains and replace them with EMUs (electrifying unelectrified branches as needed).

Subject to the requirement for new infrastructure, New York should remember that it’s a major city, and as such, it’s capable of supporting multiple independent commuter lines. Paris has five RER lines, of which only the B and D share tracks, and only between one pair of stations, on top of several major commuter lines disconnected from the RER network. It’s better to keep the map relatively coherent, so that one central trunk will split into several outer branches, but nearly all outer branches will feed into consistent central trunks. (As an example, the London Underground’s deep-level lines’ branching is coherent, while the New York subway’s mostly isn’t, with the E, F, M, and R trains running on what’s technically a branch and then diverging to three different Manhattan trunks.) This simplifies the junctions that need to be built just outside the city core, and also makes the network easier to remember.

The Tunnels

There should be a new pair of tunnels between New Jersey and Penn Station, parallel to and south of the existing tunnels. Those tunnels should then continue to Grand Central. This is the core of ARC Alternative G, which was removed from consideration in the original ARC project for reasons that were never explained adequately (Stephen Smith has been making freedom of information requests for years). However, unlike Alt G, it should not include new railyards in Manhattan, as those belong in areas where land is cheaper, nor should it include a loop for trains from the Erie lines to get to Penn Station.

The lynchpin of the plan is not the tunnels to Penn Station, which are already on the political radar in the form of the Gateway Project, albeit at a large multiple of an acceptable cost, due to such frills as new Penn Station tracks. Rather, it’s a new set of tunnels, meeting at Lower Manhattan in the vicinity of Fulton Street, going in four directions: north to Grand Central, south to Staten Island under Lower New York Bay, northwest to New Jersey via the Erie Railroad’s old Pavonia terminal, and southeast to Brooklyn to the Flatbush Avenue LIRR station. Using a double-O-tube large-diameter TBM, the Fulton Street station should feature cross-platform transfers, large banks of escalators to the street, and, to reduce costs, no station structures outside the tunneled station, putting timetables and ticket-vending machines on the street. All connections should be to Grand Central’s existing station and not the new East Side Access cavern, as the cavern leads only to the LIRR, which is already connected with both Penn Station and Flatbush Avenue. The existing tracks connect to Metro-North, which is not.

A possible additional tunnel in the far future would connect Hoboken with Grand Central’s new cavern, via Union Square. This is only in case the existing lines become congested. Current commute patterns make such congestion very unlikely, but things could change if, as a result of the new capacity, more people choose to live in suburban North Jersey and work in Manhattan.

The Network

There should be five lines running through Manhattan, without any track-sharing between them, and one using East Side Access and terminating in Manhattan. I am going to try using consistent numbering, different from the order I used in my posts on The Transport Politic, in order to group the lines using Penn Station and the lines to Lower Manhattan separately.

Line 1 is the existing mainline. Its inner route goes from Secaucus Junction to Sunnyside Junction, via the existing tunnels to Penn Station. Intercity trains use it (and should continue doing so), but most traffic will always be on commuter rail. Beyond Secaucus, trains can go to either the Northeast Corridor or the Morris and Essex Lines; to simplify junctions, most trains should use the Northeast Corridor (including the Raritan Valley Line, which splits past Newark). Beyond Sunnyside, they can go to the LIRR or the Northeast Corridor; to ensure adequate capacity for intercity trains while still providing service to the eastern Bronx, trains should use a mixture; in the long run, four-tracking everything north of Hell Gate Bridge will be necessary. It may be best to dedicate Port Washington trains to this line. At Penn Station, it uses middle-numbered tracks.

Line 2 uses the new tunnels to Penn Station and Grand Central. Its inner route goes from Secaucus to Penn Station, Grand Central, and Harlem-125th Street, splitting into branches thereafter. Most trains should go to the New Haven Line, since Line 1 could never provide adequate traffic for it; the rest should go to the Hudson Line – see below for Line 3. At the New Jersey end, it should run to a mixture of Northeast Corridor trains (including to the North Jersey Shore and Raritan Valley) and Morris and Essex trains, as required by demand. At Penn Station, it uses low-numbered tracks, potentially just 1-4. I expect it to be the most crowded, because of the service to both primary Midtown Manhattan stations.

Line 3 uses the Empire Connection, realigned tunnels to Penn Station, and the northern pair of the East River Tunnels to reach the LIRR. Its inner route goes from Yonkers to Penn Station and thence to Sunnyside. Intercity trains to Upstate New York use this line, but there are fewer of them than on the Northeast Corridor. Beyond Yonkers it can only go on the Hudson Line, so most Hudson Line trains should use it rather than Line 2. At the LIRR end it should run alongside trains to the East Side Access tunnel; as the splits are far to the east of Sunnyside, it may be prudent to have each branch serve both it and East Side Access, but in either case, timed East Side Access/Line 1 transfers at Sunnyside are necessary. At Penn Station, it uses high-numbered tracks. I expect it to be the least crowded, since to the west it only reaches one commuter line, one whose present traffic is moderate.

Line 4 is the main north-south line, from Staten Island (both the existing Staten Island Railway and the North Shore Branch) through the underwater tunnel to Fulton Street, Grand Central, and Metro-North. North of Harlem-125th Street, it can connect to any line, but I think the Harlem Line is the most coherent, as the only Metro-North line that is not needed for lines that don’t go to Grand Central. I expect it to be very crowded with inner-suburban and outer-urban traffic, as it serves Staten Island and underserved neighborhoods of the Bronx and the suburbs to its immediate north.

Line 5 is the combination of the Erie Lines, and possibly also the Northern Branch and the West Shore Line, and the LIRR’s Atlantic Branch, via Pavonia and Flatbush. For interlocking simplicity, all trains should go to one or two lines beyond Jamaica, ideally the Atlantic and Montauk Lines (the existing turnouts already favor trains from the Brooklyn Atlantic Branch continuing along the branch to Far Rockaway and Long Beach), leaving the Main Line to Line 3 and East Side Access. As there are five possible branches in New Jersey – the Northern Branch, the West Shore Line, the Pascack Valley Line, the Bergen County Line, and the Erie Main Line – frequency would be limited if all were used, so it may be best to choose just three.

Here is an unlabeled map of the five lines, with only inner branches shown; the decision of what station to terminate branches at has nothing to do with the desired service pattern, and is purely illustrative.

A potential Line 6 would take in all Morris and Essex lines, go to Hoboken, cross into Manhattan via a new tunnel with an extra stop at Union Square and then go to Grand Central and East Side Access; as discussed above, it’s omitted due to its very long-term nature.

Penn Station

I have discussed what to do with the Fulton Street Station. Penn Station is more complicated. The easiest thing to do is nothing, beyond new tunnels. There would be many platform tracks, two per access track for Line 2 and more than two for Lines 1 and 3; Line 3 would involve difficult switching moves and slow speeds through the station. Line 1 is the most important priority for allowing intercity trains to serve the stations with few (ideally no) diverging moves at turnouts, to maintain speeds.

To avoid platform congestion, especially on Line 2, more staircases and escalators should be installed. This, however, clutters the narrow platforms.

The second possibility is to pave over tracks to widen the platforms. I vacillate between preferring paving over pairs of tracks to create very wide platforms, and paving over every other track to create wider platforms at which trains can open doors on both sides. Right now I lean toward the former, as it would allow reusing preexisting escalators: the platforms currently have single-direction escalators as they’re too narrow for an adjacent pair of escalators, one per direction, and merging two platforms would be the easiest way to allow wider escalator banks.

Unfortunately, on the line with the biggest platform crunch, Line 2, this would imply a single platform with two tracks serving two tunnel tracks, so that dwell times would limit capacity somewhat. This limit is not too sharp – 24 trains per hour are achieved at through-stations in many cities without additional tracks, with some limiting cases of 30 (such as the moving block signal-equipped RER A) – but it’s still a limit, and requires good timetable adherence departing the station. These are only commuter trains, which run shorter routes than intercity trains, but Line 2 is likely to involve some long-range commuter runs, as far as Trenton or Dover or New Haven. (Since Line 1 is the only one serving local Northeast Corridor stations in the Bronx, it should only get the local trains, while longer-range trains to New Haven should use Line 2.)

The most expansive solution is to rebuild the station’s track level. There is an RPA study Penn Design study in that direction. For optimal passenger usage, the two concourse levels would be replaced by one, and the station’s 21 tracks would be reduced to 12, facing six 15-meter-wide platforms; the platforms’ eastern ends would be shaved slightly, to allow longer curve radii heading from the Lines 2 and 3 tunnels with simple turnouts, each tunnel track turning into two station tracks facing the same platform. In principle, it can be sequenced to shut down parts of the station in succession: first the southern tracks (New Jersey trains would be immediately interlined with Northeast Corridor and LIRR trains for a combination of Lines 1 and 2), then the northern tracks (the LIRR would have East Side Access by then), and finally the remainder of the central tracks. The bulk of the work on the central track could be done in conjunction, first removing the platform between the existing tracks 11 and 12 and then realigning tracks from the center outward.

I want to clarify that I do not support the most expansive solution, as it is likely to cost billions of dollars. It would create a nice Penn Station for train travelers. Those 15-meter platforms could have 6 escalators side by side with not too much obstruction, and 4 with practically none (the widest escalator is 1.6 meters wide outer end to outer end, with 1 meter used for the moving stairs). Reducing the two concourses to one would allow taller ceilings throughout, and redesigns of passageways for maximum passenger throughput. The only problem: it would be extremely expensive.

I bring this up only because the Municipal Arts Society and the RPA have teamed to propose a multi-billion dollar remake of Penn Station above track level, with high aesthetic value and zero transportation value. In addition, Amtrak wants to move its passenger facilities one block west, in the wrong direction, which has negative transportation value. If there has to be a redo of the station, it might as well be one that improves it at track level as well, rather than just making it pretty from the outside.

Phasing and Costs

The ideal phasing is “as soon as money becomes available.” There is a tendency in the US to be overly cautious about everything and chop projects into little pieces, in the name of prudence. It’s always easy to show one’s moderation by chopping a fixed amount of money from every proposal (quintessential moderate Senator Olympia Snowe was famous for this) and by funding many projects by small amounts. These small projects then fail because of reduced network effects or sometimes higher costs due to smaller orders.

The tunnels I proposed in this post sum to about 30 kilometers. These 30 kilometers are objectively difficult to build. The tunnels for Lines 4 and 5 of this proposal go under wide rivers and a bay, and once they reach Manhattan land they have to go under the entire Lower Manhattan subway network. Half a billion dollars per kilometer would be a good deal: Crossrail is more than a billion dollars per double-track tunnel kilometer, assuming there is nothing to build except tunnels (which is far from true), while Crossrail 2’s cost range is $600-850 million per km (see also my first comment in the link). London is a high-construction cost city, but New York is even higher-cost; building a line for London’s costs would be a major achievement for New York.

Bear in mind that Amtrak thinks the Gateway Project alone would be $16 billion. When I propose to build an entire regional rail network for perhaps $20 billion (in 2010 dollars, not year-of-expenditure dollars), based on what it would cost in other cities, I am not taking into account the bloat that leads to high costs in New York. At the per-km costs Amtrak thinks are appropriate for what would be one of the simpler tunneling projects for this system, this is plain unaffordable.

Still, precisely because of the network effects, and because this plan neatly separates branches of the existing commuter rail system, it should be proposed all at once. If it’s expensive then it will also be delayed; it’s better to have six mainline rail tracks under the lower Hudson by midcentury, than to have four and then realize there’s a capacity crunch and six tracks are required after all. Of course it’s best if everything is in place by the late 2020s, on the schedule of the Grand Paris Express. But the point is that longer project latency encourages bigger rather than smaller plans. The Line 2 tunnel, by whatever name, is still the most important priority, but the phasing then becomes “whenever it can be designed and built.”

The lower-end cost I’m proposing is for a project without any frills. It includes a bare minimum touch for Penn Station – simpler interlockings at places and some extra access points, but no more. It includes no Sunnyside decking or other redevelopment, which should be funded separately in any case. (When people build highways, do their projected cost figures ever include the construction of the suburban subdivisions they’d sprout?) It doesn’t include electrification of branches, although that is cheap enough as to be well within the uncertainty in even a first-order estimate. It doesn’t even include rolling stock, although the large preexisting fleet of decent EMUs means there’s no need for immediate fleet replacement as on the MBTA and other diesel-hauled railroads.

The only thing this project does include is more paths for more commuter trains to serve Manhattan and other regional job centers.