Methodology & Limitations

Telangana EV Charging Opportunity Index · research beta · built 1 August 2026

1. What is computed

Telangana is divided into 21,813 hexagonal cells (H3 resolution 8, about 0.74 km², inside Greater Hyderabad; resolution 7, about 5.16 km², elsewhere), covering 112,291 km². For each cell and each segment:

OpportunityIndex = normalise( Demand − EffectiveSupply ), scaled 0–100.

Both terms are put on a common footing before subtraction: each raw layer, and the supply total, is converted to a z-score, clipped at ±3σ, and rescaled to 0–100. Subtraction therefore happens between two comparable 0–100 quantities, and the result is min-max rescaled to 0–100.

Demand is a weighted sum of normalised layers (z-score, clipped at ±3σ, rescaled 0–100). Effective supply is the distance-decayed sum of existing charger power within a segment-specific radius — 3 km urban and 25 km on corridors for cars, 30 km for trucks.

2. Weights

A layer we could not obtain is never scored as zero — zero would mean "measured, none present". Its weight is removed and the remainder renormalised. The car model therefore runs on 92 of 100 points; trucks on 100 of 100.

3. Sources

4. Station sizing

Gun counts come from an M/M/c queueing model: the smallest number of guns keeping the 90th percentile wait under 10 minutes at peak. Peak hour is taken as 18% of daily sessions.

5. Validation — and what it did not show

Modeled demand was tested against measured electricity consumption for 85 Hyderabad localities.

6. Gap register

6b. The four components

A single 0–100 number hides why a place scores as it does. Strong demand with heavy existing supply and weak demand with none can land on the same index value, so each area read-out now reports four independent measures:

Alongside them we publish the underlying evidence: how many stations sit in the catchment, their combined power, and what share of that power is actually published rather than imputed. "No supply here" and "we cannot see the supply here" are different claims, and the read-out distinguishes them.

Two consistency faults were found and fixed while building this, both of which would have shown a reader contradictory numbers. Min-max scaling of a long-tailed supply distribution squashed the gap to a near-constant (90th percentile = 100), so a cell with five stations nearby still read "supply gap 100/100"; it now uses a rank transform. And the evidence counts had to be put on exactly the same footing as the score — the same catchment radius, and for trucks the same DC ≥ 50 kW filter — or the station count contradicted the gap beside it.

7. Stability — and one place it fails badly

Every weight was re-run at ±20%. The top-25 lists hold 97.0% (cars) and 96.3% (trucks) of their members on average. Rankings are robust to the weight choices — though that measures stability, not correctness.

Weights are not the real uncertainty here. Charger power is imputed for 97.9% of stations, so we re-ran the model with every imputed kW scaled ±50%, leaving measured values untouched:

7b. Resolution normalisation

The grid mixes H3 resolution 7 (~5.3 km²) and resolution 8 (~0.76 km²). All demand layers except one are fixed-radius measures — "how much of X lies within N km of this point" — which are independent of cell size. The exception was EV registrations, an absolute per-cell count, where a 7× larger cell would carry 7× the count for no real reason. That layer is now scored as density per km². Without this correction, cell size leaked into the ranking.

7d. Electric bus corridors

The map shows highways carrying known electric bus services, because e-buses are a captive, high-energy, schedule-predictable load — a corridor already running them is a materially different proposition from an empty one.

Note the difference in charging economics: bus charging is a captive depot-and-terminus load, contracted to an operator, not public forecourt demand from passing traffic. A corridor carrying e-buses signals a counterparty to talk to, not walk-in volume.

7c. Regulatory documents tracked

Beyond the September 2024 guidelines above, the refresh pipeline tracks:

Regulatory coverage requirements answer a different question from commercial opportunity and are deliberately not merged into this index.

8. Reproducibility

The browser calculation and the Python reference implementation are tested against each other on 100 random points and agree exactly. Point queries run entirely in your browser against a precomputed index; no query data is transmitted.