Mumbai Traffic Network Planning · Baseline Report

Should we build it? Testing road fixes on the Western Express Highway.

A digital model of the Dahisar–Bandra corridor that predicts how traffic responds to infrastructure changes — before construction.

The bottom line

On the same traffic demand, the model tested four interventions. Only widening the worst link helped. A new bypass backfired (+6% total travel time — the Braess paradox). Closing the highway link was worst (+19%). A stalled vehicle steadily worsened delay as more piled up. These conclusions held under every uncertainty setting we tried — so the advice is trustworthy even though the exact numbers still need peak-hour calibration.

1The problem

Mumbai's roads are saturated, and every fix is expensive. But roads don't behave simply: you can't tell drivers which route to take. Build a new road and everyone piles onto it; widen one spot and the jam moves down the street. Occasionally a brand-new road makes traffic worse — a real, proven effect called the Braess paradox.

So "will this project help?" can't be answered by intuition. This tool answers it by simulation: build a digital twin of the corridor, pour in realistic demand, let simulated drivers each pick their fastest route (as real drivers do), and measure the congestion. Then change one thing and re-simulate to see the difference.

2How it works

Five layers, each feeding the next — from raw map data up to the maps and charts in this report.

5Reporting — maps, comparison charts, this report

4Scenarios — change the network, re-simulate, compare

3Assignment — drivers pick fastest routes until nobody can do better

2Demand — how many trips run between each area in the peak hour

1Network — the road map: lanes, speed, capacity per segment

0Data — OpenStreetMap (roads) + TomTom (live speeds)

Road network of the WEH corridor — The corridor's real road map from OpenStreetMap. The red spine is the Western Express Highway; orange/yellow are main arterials; grey is the local street fabric.

Real congestion, measured live

We sample real vehicle speeds along the highway from the TomTom traffic API and measure how much slower than free-flow each point is. This grounds the model in reality.

Live congestion snapshot along the WEH — free-flowing slowing congested — a real snapshot. The Dahisar–Goregaon stretch and Bandra approach slow down even off-peak.

3The model, in three ideas

Drivers are selfish. Each picks the route fastest for them; the model shuffles routes until no one can do better — a stable equilibrium.

Roads slow as they fill. Near-empty roads run free; over capacity they crawl. A standard engineering curve captures this.

A stalled vehicle steals extra road. Traffic swerves around it, losing far more capacity than the car's size — our custom addition.

The headline measure is Total System Travel Time — the total person-hours the corridor spends driving. Lower is better; every scenario is judged by how it moves this number.

4Results

The model finds the real bottleneck on its own

Run on today's network, the highway itself lights up red (over capacity) — exactly the bottleneck every commuter knows. Reproducing reality without being told to is the key validation.

Base-case congestion map, WEH in red — Base case. under capacity → over capacity. The WEH spine is saturated end to end.

Every case, simulated and compared

Intervention	Total travel time	What it means
A · Widen worst link	▼ improves	Helps, but the jam partly relocates
B · Add a bypass	▲ +6% worse	Braess paradox — the new road backfires
C · Close the link	▲ +19% worse	Losing the bottleneck link is very costly
D · Stalled vehicles (1→3)	▲ +1% → +8%	Each breakdown compounds the delay

Bar charts comparing all scenarios — Left: total travel time per case (green improves, red worsens). Right: the same as a percentage change from today.

Side-by-side congestion maps for each scenario — The corridor under each intervention. Note how closing the link (third panel) pushes red and orange congestion out onto the parallel arterials.

Does the advice survive uncertainty?

Our demand and speed inputs are approximate, so we re-ran every case under five different settings (higher/lower traffic, different congestion assumptions, capacity caps). The directional conclusions never flip — only the exact percentages move.

Robustness of each intervention across five settings — Each intervention's impact across five settings. Widening is always the only improvement; stalled vehicles always hurt; closure is always ~20% worse; the bypass always backfires. **The planning advice is stable.**

5What we assume — and the one real gap

This is a baseline: it proves the machinery works and gives stable directional advice. It is not yet planning-grade, and we are explicit about why:

Demand is synthetic — a plausible estimate, not a travel survey. Absolute volumes are approximate.
Fixed demand before/after a project — ignores that new roads attract new trips (to be added later).
Instant rerouting — right for planning questions, but understates the short-term chaos of a live incident.
Some lane counts guessed — only 16% are tagged in the open map data.

The one thing that would most improve accuracy: a single working-day peak-hour speed reading (Mon–Fri, 8–10 AM or 6–8 PM). Our three snapshots so far are all holiday/off-peak and too "flat" to fully calibrate the congestion curve. One good peak reading unlocks real calibration.

6Possible next steps

Collect one working-day peak reading → unlock real calibration (highest value, least effort).
Calibrate demand to real volumes using the TomTom "origin–destination from travel-times" method — the biggest accuracy jump.
Add induced demand so project benefits aren't overstated.
Verify lane counts on the highway from satellite imagery.
Real census wards for the demand zones.
Microsimulation for true incident dynamics and signal timing, once the base model is trusted.

Baseline status: modelling pipeline complete end to end (network → demand → equilibrium → scenarios → reporting). Results are structurally sound; absolute numbers await peak-hour calibration. Pilot corridor: Western Express Highway, Dahisar–Bandra. Data: OpenStreetMap & TomTom. Method: static User-Equilibrium traffic assignment.