Where Should Phase 2 Go?

Executive Summary

New York City’s e-bike delivery workforce now numbers in the tens of thousands, and lithium-ion battery fires linked to improperly charged e-bikes have become one of the city’s most consequential fire-safety issues of the decade. In response, NYC DOT piloted five public e-bike charging hubs across Manhattan and Brooklyn in 2024–2025, operated through partnerships with three charging vendors. This report treats that pilot as a starting point rather than an endpoint, and asks a narrower, more actionable question: where should a second phase of hubs go, and does the answer look different if the siting criterion is genuine need rather than ease of installation?

Using NYC DOT’s current bike-lane network, a one-million-trip sample of September 2025 Citi Bike data, the DOT pilot evaluation report, and Census/ACS-derived neighborhood indicators, this analysis builds a transparent, four-factor suitability score — e-bike demand, bike-network infrastructure, delivery/restaurant activity, and equity need — and applies it to the five confirmed pilot sites plus four analyst-identified Phase 2 candidates.

The results are a genuine mixed validation, not a clean confirmation of either the pilot or the critique of it:

• DOT’s own strongest sites, Essex Market and Cooper Square, score highest of all nine locations under an independently constructed model — the demand signal there is real.
• Two pilot sites, Willoughby Street and Brooklyn Army Terminal, score lowest of all nine — consistent with a siting process driven by power and space availability rather than need.
• East Harlem and Jackson Heights, neither of which received a pilot hub, score competitively with confirmed pilot sites, and East Harlem’s equity case is corroborated by an independent data source.
• Flushing has zero Citi Bike presence within the study radius, exposing a structural blind spot in demand proxies built on bike-share data — yet it still scores respectably on delivery activity and equity need alone.
• A stress test of the equity layer found that one candidate’s (Sunset Park’s) income data is not resolved — estimates from four sources range from $48,000 to $96,720 depending on which neighborhood boundary is used, a genuine data-quality finding rather than a flaw to be smoothed over.

The analysis is offered as a recommendation-generation tool, not a final ranking: weights are asserted rather than statistically fit, the sample is nine sites, and half of the equity layer relies on community-district (PUMA) geography rather than exact census tracts. Within those limits, it produces a specific, defensible case for prioritizing East Harlem and Jackson Heights in Phase 2 planning, and for resolving Sunset Park’s data uncertainty before treating it as either a priority or a pass.

1. Background

The problem: e-bike delivery, batteries, and public space

Since roughly 2020, food-delivery e-bikes have gone from a niche convenience to critical urban infrastructure, with tens of thousands of couriers relying on them daily across New York City. Most of that fleet is charged informally — in apartments, stairwells, and delivery-app aggregation hubs — using batteries and chargers of uneven quality. That pattern has been linked to a sharp rise in structural fires involving lithium-ion batteries, prompting the FDNY, City Council, and NYC DOT to treat public, supervised charging infrastructure as a fire-safety intervention as much as a transportation one.

The pilot: “Safer Charging, Safer Deliveries”

NYC DOT’s response was a five-site pilot program, evaluated in the department’s Safer Charging, Safer Deliveries report [1]. The five confirmed locations — Plaza de las Américas (Washington Heights), Cooper Square (East Village), Essex Market (Lower East Side), Willoughby Street (Downtown Brooklyn), and Brooklyn Army Terminal (near Sunset Park) — were operated through three charging-hardware vendors (Swobbee, PopWheels, and Swiftmile). Per the report, site selection prioritized practical feasibility: existing electrical capacity and available curb or plaza space, sited near — but not formally scored against — areas of high restaurant and delivery demand.

That is a reasonable way to launch a five-site pilot quickly. It is a different exercise from asking where a citywide network should expand next, which is the question this report takes up.

Research question

If NYC DOT scales this program beyond five sites, which locations combine the greatest demand, the weakest existing infrastructure, and the greatest need — and how much of that picture is already visible in the pilot’s own results?

2. Data & Methods

Data sources

Four independent data sources were combined. Each is documented in the reference list (Section 7); the table below summarizes scope and role in the model.

Table 0. Data sources, scope, and role in the suitability model.

Suitability model

Following the original project framework, a composite score was computed for nine sites — the five confirmed pilots and four analyst-identified Phase 2 candidates (Sunset Park core, Jackson Heights, Flushing, and East Harlem) — as a weighted sum of four min–max normalized factors, scaled 0–100:

E-bike demand was computed as total Citi Bike trip volume within a 600-meter radius of each site’s coordinates. Bike infrastructure was computed as a weighted count of bike-lane segments within the same radius (protected segments weighted 2×, painted 1.2×, sharrows 0.6×, reflecting relative usefulness for a charging stop). Delivery activity used restaurant-and-food-service employment for each site’s home community district. Equity need combined (inverted) median household income with the foreign-born population share, in equal parts, for the same district. All four factors were min–max normalized across the nine sites before weighting, so scores are relative to this sample, not an absolute citywide scale.

This is a deliberately mixed-precision design: the demand and infrastructure factors use exact point-level geometry, while the equity and delivery factors use community-district aggregates, because point-level census-tract data could not be retrieved within this pass (see Section 4, Limitations). That asymmetry is treated as a finding in its own right, not hidden in the results.

3. Findings

3.1 Baseline: network capacity and mode share

Before assessing individual sites, two citywide baselines frame the analysis. First, current protected bike-lane capacity is unevenly distributed: Manhattan has the highest share of protected (Class I) lanes relative to its network, while Brooklyn and Queens carry the largest total lane volume but a lower protected share (Figure 1). Second, electric bikes are no longer a marginal share of micromobility activity: 70% of the 1,000,000 Citi Bike trips sampled from September 2025 were on e-bikes (Figure 2), confirming that the underlying premise of the pilot — that e-bike charging is now a mainstream infrastructure need — is well supported by current ridership.

3.2 Composite suitability scores

Figure 3 ranks all nine sites by composite suitability score. Essex Market and Cooper Square — both confirmed pilot sites in the Lower East Side / East Village corridor — score highest, at 73.5 and 71.9 respectively, well ahead of every other site. East Harlem (41.1) and Jackson Heights (40.9), neither of which has a pilot hub, score within half a point of each other and close behind Plaza de las Américas (37.2), a confirmed pilot site. Willoughby Street (22.6) and Brooklyn Army Terminal (21.9) score lowest of all nine, despite both being part of the confirmed pilot.

This is best read as a mixed validation. The model does not simply contradict DOT’s siting choices — its two highest-scoring locations are DOT’s own top-performing pilot sites. But it also does not simply confirm them: three of the four lowest scores belong to pilot sites, and two non-piloted neighborhoods outscore three of the five confirmed locations.

3.3 What is actually driving each score

Figure 4 decomposes each site’s score into its four weighted components, and the pattern clarifies why the ranking looks the way it does. Essex Market and Cooper Square score high primarily on e-bike demand and infrastructure — exactly the factors closest to existing Citi Bike and bike-lane data, which also happen to be the corridors Citi Bike itself is most saturated in. Jackson Heights and Flushing, by contrast, score almost entirely on delivery activity and equity need, with the demand and infrastructure bars nearly empty — not because delivery cycling is low there, but because Citi Bike has minimal presence in those neighborhoods and protected bike-lane coverage is genuinely thin (Jackson Heights has just 3 protected-lane segments within 600 meters, versus 115 at Essex Market).

Willoughby Street illustrates the opposite failure mode: it scores moderately on demand and infrastructure (it sits in a well-served, high-ridership part of Downtown Brooklyn) but scores zero on both delivery activity and equity need — its home district has the highest median household income of all nine sites and no restaurant-sector employment among its top three industries. Brooklyn Army Terminal scores low across all four components; it is a large logistics site at the edge of, rather than inside, the Sunset Park commercial corridor it is often associated with.

3.4 A genuine demand-proxy blind spot: Flushing

Flushing (Queens Community District 7) returned zero Citi Bike stations with any trips within the 600-meter study radius of its downtown core — Citi Bike simply does not operate there at meaningful scale. Under this model, that collapses the demand and infrastructure components to zero, and the site’s entire score (37.1) is carried by delivery activity and equity need alone. Restaurant-and-food-service employment in Flushing’s home district is the single largest of any site studied (9,322 workers, the district’s top industry), and the neighborhood has the highest foreign-born share in the sample (58.2%).

3.5 Stress-testing the equity layer

Because the equity factor relies on community-district-level income and demographic data rather than exact site coordinates, two of the sharper findings above — East Harlem’s strong equity case and Sunset Park’s more ambiguous one — were cross-checked against an independent data source, NYU Furman Center’s neighborhood-level profiles [6], which use a different (though still not parcel-exact) geographic boundary than the PUMA data used in the primary model.

East Harlem’s figure held up well: the PUMA-level estimate used in the model ($42,218) and the Furman Center neighborhood estimate ($41,870) differ by less than 1%, and both are far the lowest of any site studied. Its equity case does not depend on which reasonable data source is chosen.

Sunset Park’s did not. Across four sources — the PUMA estimate used here ($89,890, blended with adjacent Windsor Terrace), the Furman Center neighborhood boundary ($96,720), a tighter block-group estimate ($57,669), and a 2014 study of the immigrant commercial core specifically ($48,000, historical) — the range spans nearly $49,000. Every one of these is a defensible “Sunset Park,” and they imply opposite conclusions about whether the neighborhood is an equity priority. The Sunset Park core candidate’s rank in Figure 3 should be read as unresolved, not as a negative finding: it may simply reflect that the wrong boundary was available to draw from in this pass.

3.6 Full results table

Weighted contribution of each component to the final 0–100 score, sorted by total (see Section 2 for the underlying formula and normalization method).

Table 1. Full suitability score decomposition, all nine sites.

4. Limitations

• Sample size: nine sites is sufficient to illustrate the model and surface real findings, but far too small for the component weights to be statistically fit or validated — the 0.35/0.25/0.20/0.20 split is asserted from the original project design, not estimated from data.
• Mixed geographic precision: the demand and infrastructure factors are computed from exact coordinates; the equity and delivery factors are computed from community-district (PUMA) aggregates, which can be materially wrong for any specific point within a large or heterogeneous district, as Section 3.5 demonstrates directly for Sunset Park.
• Demand-proxy validity: Citi Bike ridership is a proxy for general e-bike micromobility, not for delivery-worker e-bike activity specifically, which uses a different fleet, battery chemistry, and trip pattern. Section 3.4 shows this proxy can go to zero in neighborhoods with real, documented delivery activity.
• Delivery-activity proxy: restaurant-and-food-service employment is measured by workers’ place of residence (American Community Survey), not by business location or delivery-order volume, and is a directional indicator rather than a precise count of delivery demand.
• Pilot-site coordinates are approximate: NYC DOT’s evaluation report identifies pilot locations by name and neighborhood, not by published geocode, so this analysis used landmark-level coordinate estimates rather than exact addresses.
• The model cannot yet be causally validated against pilot utilization data, since per-site usage figures were not published in a form usable for this analysis; the comparison in Section 3.2 is against site selection, not measured performance.

5. Recommendations

1. Prioritize East Harlem and Jackson Heights as the strongest near-term Phase 2 candidates: both score competitively with confirmed pilot sites, and East Harlem’s equity case is independently corroborated.
2. Treat Flushing as a priority for further study rather than exclusion: its absence of Citi Bike coverage should be read as a data gap, not a sign of low need, given the district’s exceptional restaurant-employment and foreign-born concentrations.
3. Resolve the Sunset Park income discrepancy before scoring it in either direction — acquire a tract-level or address-level estimate for the specific commercial corridor near 5th/8th Avenues rather than relying on named-neighborhood aggregates.
4. Upgrade the equity and delivery layers to tract-level data before finalizing a Phase 2 site list: replace the PUMA-level layers with U.S. Census Bureau tract-level ACS data, geocoded from each candidate point via the Census Geocoder API — the single highest-value data upgrade identified in this analysis.
5. Refit model weights against real usage if pilot utilization data (sessions per day, unique users, peak hours) becomes available by site, rather than continuing to use the illustrative 0.35/0.25/0.20/0.20 split.

6. Conclusion

The core, defensible claim of this analysis is narrower than “build here next” — it is that demand and equity can be measured well enough, even with public data and modest tooling, to distinguish a strong Phase 2 candidate from a weak one, and to show that the current pilot’s footprint captures some but not all of the strongest locations in the city. Essex Market and Cooper Square earned their place in the pilot on the numbers. East Harlem and Jackson Heights deserve the same scrutiny before Phase 2 is finalized, and Sunset Park deserves better data before it is scored at all.

7. Data Sources & References

• [1] NYC Department of Transportation. Safer Charging, Safer Deliveries: E-Bike Charging Pilot Evaluation. NYC DOT, 2025.
• [2] NYC Department of Transportation. Bike Routes (current and retired segments). NYC Open Data, data.cityofnewyork.us.
• [3] Citi Bike / Lyft. Citi Bike System Data, September 2025 monthly trip file (1,000,000-row sample used in this analysis). citibikenyc.com/system-data.
• [4] U.S. Census Bureau. American Community Survey 5-Year Estimates, via Census Reporter (censusreporter.org), Manhattan/Brooklyn/Queens Community District PUMA profiles.
• [5] DataUSA.io (Deloitte / Datawheel), PUMA-level ACS profiles: median household income, foreign-born share, poverty rate, industry employment by residence, for the community districts referenced in Table 1.
• [6] NYU Furman Center. CoreData.nyc neighborhood profiles, East Harlem and Sunset Park, 2024 estimates.

Method note: all GIS proximity calculations (600-meter site buffers, bike-lane segment matching, Citi Bike station aggregation) and score normalization were performed directly on the source files listed above; no third-party suitability-model software was used.