Skip to content
No. 035Going aheadClimate and parametric insuranceUpdated Nov 15, 2026

How often does the drought index behind Africa's sovereign drought insurance (the African Risk Capacity's Africa RiskView model, built on the WRSI crop water-balance index) miss real droughts or flag false ones, and which inputs (rainfall dataset, crop season length, planting rule) cause the misses?

Good it would do
3 / 5
Can AI help
3 / 5
How neglected
3 / 5
Can we check it
3 / 5

Why we're going ahead

Accepted on a narrowed question: how much an ARV-style WRSI result swings with rainfall product, season length and planting rule, and whether any setting beats a rainfall baseline against held-out yields, as a pre-registered Burkina Faso and Malawi pilot. The fragility is documented (a fivefold Mauritania 2021 payout spread by rainfall product) and FCDO's evaluators asked for an external review, but no user has committed and outcome data are thin where the index failed.

Why it's stuck

The model is run by the African Risk Capacity (ARC) and customised country by country with governments, and the only public study of how much its output depends on its inputs that I found is a World Bank note from February 2017. It covered only maize in Malawi, millet in Niger and rangeland in Mauritania, called its own results preliminary, and recommended that 'the variability due to the WRSI index itself is major, and a quantification of its domain of applicability might be required'.

What we'd try

Re-run the WRSI water-balance index (the method is published, and FEWS NET distributes its GeoWRSI software free) for every ARC member country and season from about 2001 to 2025. Run it as a large ensemble over five or more public rainfall products (CHIRPS, TAMSAT, ARC2, RFE2, ERA5), a range of crop season lengths, and different planting-date rules: many thousands of gridded runs.

Source: https://www.opml.co.uk/sites/default/files/2025-02/arc-ia1-evaluation-report.pdf

Do you know the answer?

One reply from someone in the field can settle what our AI can't.

Members can read the full record for this problem, check its sources and comment. Open the member view →