Proof of Drought

For Fiji's sugarcane farmers

Drought insurance that pays itself.

Three independent rainfall records, from Europe, NASA and satellites, watch every cane district. When they agree the rain has failed, every insured farmer is paid automatically, without filing a claim.

Ba, 1997–98 Press play to watch the season, month by month

Your cane farm

Ba

Mobile walletF$0
  1. No alerts yet.
    WetNormalDrySevere droughtPaid
    Paid to youF$0
    To all 96 test farmersF$0

    01The 1998 drought

    8 months earlier.

    That's how much sooner Fiji's cane farmers would have been paid in 1998: January, against a disaster declared in September.

    What actually happened

    • Sep 1997The drought begins in Fiji's west
    • Sep 1998A year later, Fiji declares a disaster
    • Meanwhile90% of people in the worst-hit west on food and water rations

    With Proof of Drought

    • Jan 1998F$300 per farmer in 7 of 8 districts, in time to save stock and replant
    • Apr 1998The rest: F$1,000 per farmer in every cane district

    From rain to phone, step by step

    1. Three records watch the rain

      Copernicus ERA5 (Europe) and NASA MERRA-2 are weather models fed with satellite, weather-station and balloon measurements. CHIRPS is satellite rainfall checked against rain gauges.

    2. Three oracles post the readings

      Each posts its record's reading, an AI check on the data and a seal of the raw numbers. In this prototype we run all three. In a real scheme each would be a different organisation, for example a weather service, a university and a reinsurer.

    3. The contract decides and pays

      Two of the three records must agree. Then it pays everyone in the district at once. Here it uses test money; a real scheme would hold a digital Fiji dollar.

    4. The money reaches the phone

      A licensed payout partner would turn it into M-PAiSA or MyCash mobile money and send an SMS. Not built yet: it needs a partner and approval from the Reserve Bank of Fiji.

    Source: UN OCHA, Fiji Drought Situation Report No. 2, 16 December 1998. Payout dates: this project's test on real Copernicus and NASA data. Grower count: about 10,200 active cane farmers (Prime Minister Sitiveni Rabuka, reported by the Fiji Times, October 2025).

    02Track record

    Would it have worked? We ran it on 45 years of real weather.

    Each square is one season since 1982. Green: the cover paid out. Red dot: Fiji's cane harvest really fell that year.

    Paid a lotPaid a littlePaid nothingHarvest fell 10%+Cyclone or flood

    What would you have received?

    Every payout the rules would have made for F$1,000 of cover since 1982.

    03Can it be cheated?

    You've hacked the European climate feed. Fake a drought.

    Ba, 2010–11: a normal, wet season. Report less rain than really fell, then try to claim the payout.

    Slide left to pretend less rain fell than really did. Less rain makes it look like a drought.

    1

    AI check

    2

    Two other records

    3

    Public seal

    04Right now · the 2026–27 season

    A 1998-strength El Niño is under way. This is what the contract sees today.

    The last time the Pacific looked like this, Fiji's cane belt went eleven months without enough rain. These are today's Copernicus and NASA readings, ranked exactly as the contract will rank them.

    AI forecast for this season

    60%

    chance a cane district gets a drought payout this season, against 17% in a normal year.

    How we know

    The price

    F$380

    per F$1,000 of cover this season. The price follows the El Niño forecast, so nobody can game it by buying only in risky years.

    How we know

    Pacific Ocean warmth (Niño-3.4), September

    +2.8°C2026
    +1.8°C1997, before the 1998 drought

    Degrees warmer than normal, January to September. Source: NOAA Climate Prediction Center, which expects a strong El Niño to last into early 2027.

    05For insurers and funders

    Does the pool survive? 45 years at Fiji's scale.

    About 10,200 cane growers with F$1,000 of cover each, every season since 1983, paid by the same rules and priced by the El Niño forecast made before each season began.

    Money in the poolPaid out that season

    06Check the data

    Check the data yourself.

    Every reading the contract used carries a digital seal of the raw data behind it. Download that data straight from Copernicus, NASA and the CHIRPS satellite archive, right now, and see if it matches.

    Ba, November 1997 – April 1998

    Real

    45 years of daily rainfall and soil moisture for Fiji's 8 cane districts from Copernicus ERA5, NASA POWER and the CHIRPS satellite record, and today's readings, downloaded live · the El Niño record (NOAA) · Fiji's cane harvests (FAOSTAT) · the 1998 timeline (UN OCHA) · the smart contract, running on a real Ethereum virtual machine in your browser and on the public Sepolia test network, with this season's forecast sealed on it

    Simulated

    The 96 farmers, their wallets and the money are test versions · district points are approximate · the payout rules and pricing margin are this project's proposal; a real scheme would set them with an insurer and the regulator

    Built for COP31

    COP31 meets in Antalya in November 2026. Its Pre-COP ministers met in October in Nadi, Fiji, one of the eight districts this cover protects. It builds on parametric insurance already working in the Pacific (PCRIC, UNDP's PICAP).

    What we haven't solved yet

    • Payouts can miss the harvest. Fiji only publishes cane yields for the whole country, so we can't check district payouts against district harvests. In 1989–90 and 2004–05 the cover paid in a dry season that the cane survived.
    • El Niño years are expensive. The fair price can be more than three times the normal one. Sponsors could cover the difference, but no sponsor has signed up.
    • Nobody in Fiji has reviewed this yet. Not farmers, the Sugar Cane Growers Fund, the Fiji Meteorological Service or an insurer. Their input would change the rules.
    • We run every oracle, with test money. The contract runs on a public test network. Real oracles need independent operators.
    • A disagreement slows the payout down. If Copernicus and NASA disagree, the decision waits for CHIRPS, which comes out about three weeks after each month. That happened at about 1 in 9 past checkpoints. CHIRPS's quicker preliminary version could cut that to days.
    • The AI is deliberately small. A check for faked data and a one-input forecast that is only a little better than the long-run average.

    Satellite basemap: Sentinel-2 cloudless – https://s2maps.eu by EOX IT Services GmbH (Contains modified Copernicus Sentinel data 2024), CC BY-NC-SA 4.0. Contains modified Copernicus Climate Change Service information 2026 (ERA5, via Open-Meteo, CC BY 4.0). NASA POWER data obtained from the NASA Langley Research Center POWER Project funded through the NASA Earth Science / Applied Science Program. Niño-3.4: NOAA Climate Prediction Center. CHIRPS: Funk et al. (2015), UCSB Climate Hazards Center, via NASA/USAID SERVIR ClimateSERV. Yields: FAOSTAT.

    Built by Tia · IEEE ClimateChain Global Hackathon 2026, Climate Data & Environmental Monitoring · Source code