Investable score · KFL-M 1.6

Which flare sites are worth a first look

Every flare site in the tracker, scored 0 to 100 on the three questions an investor in gas-to-power asks first. Is the flare consistent, so the fuel will be there? Is it close to people without power, so there is demand? Is it secure enough to build and run? A fourth, smaller term asks whether the 1.5 siting screen would keep it. Hover or focus any figure with a dotted underline for a plain-language explainer. Every figure is an estimate, and the score is for triage, not an investment recommendation.

Band A sites

…

Scored sites

…

Median score estimate

…

Gated

…

Figures above are national. Each is derived from the site records below.

How a site is scored

Each component is built from terms normalised to 0 to 100, mostly on log scales so that one very large flare or town does not swamp the rest. The investable score is the weighted mean of the four components. Two hard gates come first: an offshore flare, or one whose gas could not fuel a 0.1 MW generator, scores 0 (its ungated score is still shown). Raw satellite units, bright_ti4 in kelvin and FRP in megawatts, are shown as inputs and normalised before they reach the score.

Bands: A 75 and up, B 65 to 74, C 55 to 64, D under 55. Scores within about 3 points of each other are effectively tied; see the band margin. Full method: methodology section 7c.

Ranked sites

Stress tests

Before publishing, the method was tested against the cases most likely to mislead. The results below are re-run from the committed data by node scripts/build_invest.mjs and published as invest_stress.json.

Where judgement could mislead

• The weights are judgement calls. Consistency 28%, demand 34%, security 26%, viability 12% are not fitted to outcomes; there is no record of built flare-to-power projects to fit them to. The stress tests show the ranking is stable when every weight moves by up to 30%, but a reader with different priorities should re-weight from the published component scores.
• Security looks backwards. It counts recorded attacks since 2015 and ACLED events over the last 12 months. It does not know about community agreements, host-community trusts, or a change in local politics. A clean record is not a guarantee.
• Eight nights of satellite data is short. October is rainy season in the Niger Delta, and cloud reads as "not burning". The persistence term is a lower bound and carries 5.6% of the total score, so a single warm pixel cannot lift a site and a cloudy week cannot sink one. A large flare that FIRMS barely saw is flagged instead.
• The access proxy is state-wide. Every site in a state shares one rural access figure, so the access term separates states, not villages.
• Volume is not capturable gas. The volume term assumes the flared gas could be captured. Gas composition, pressure, the operator's own plans and gas rights decide whether it can be.
• Bands are brittle at the edges. About half the scored sites are within 2 points of a band cut-off. Read the score and the band margin, not the letter alone.

What this is not

Back to the tracker · Investable score method · Mini-grid siting