Skip to content

Methodology

Choosing pilot regions

Why Nakuru and Dodoma are the first pilot candidates, and how each region is checked against a second rainfall dataset before its index is trusted.

A region's index is only as good as the rainfall data behind it. Before Mvua trusts a region, its calibration has to hold up on a second, independent dataset. This page lists every region checked so far and the result.

The test

The index methodology calibrates each region on ERA5, a weather reanalysis. To check it, each season is recomputed from CHIRPS, a satellite and rain gauge blend at about 5 km, over the same years and the same season window. Two questions decide the result:

  1. Do the two datasets rise and fall together? The correlation of the season totals, where 1.0 means they move in lockstep.
  2. Do they agree on the droughts? How many of the 6 driest seasons in ERA5 are also among the 6 driest in CHIRPS. These are the seasons the index pays in, so this matters most.

A region is confirmed when the season totals correlate at about 0.7 or more and the two datasets share most of their driest seasons. Only confirmed regions can have their index frozen and be offered in a pilot.

Results

RegionCorrelation of season totalsDriest 6 seasons sharedResult
Nakuru, Kenya0.814 of 6 (1993, 2000, 2008, 2017)Confirmed
Dodoma, Tanzania0.714 of 6 (1996, 1999, 2005, 2016)Confirmed
Lagos, Nigeria0.572 of 6Not confirmed
Anambra, Nigeria0.362 of 6Not confirmed
Delta, Nigeria0.343 of 6Not confirmed
Kano, Nigeria0.18see belowNot confirmed
Katsina, Nigeria0.482 of 6Not confirmed
Sokoto, Nigeria0.402 of 6Not confirmed
Rural Kano state, Nigeria0.131 of 6Not confirmed

Nakuru and Dodoma are confirmed and their indices are frozen on testnet. They are the first pilot candidates.

What the checks found

Kano. The two datasets disagree on which seasons were dry. ERA5's driest Kano season, 2021, which the calibrated index pays in full, is an ordinary season in CHIRPS (16th driest of 34). CHIRPS's driest, 2007, is only 10th driest in ERA5. Satellite greenness agrees with neither, plausibly because of irrigation. With no agreement and no referee, the chance of paying the wrong seasons is too high for a first pilot. Kano still hosts the daily testnet publisher run, which tests operations, not calibration.

Southern Nigeria. Lagos, Anambra and Delta are wet enough that a rainfall shortfall rarely hurts crops, and the datasets disagree there too. These regions are better candidates for a flood or excess rain index later.

North west Nigeria. Katsina, Sokoto and a rural site in Kano state were tested because this drier savanna is where a rainfall shortfall is the main crop risk. None pass. The rural Kano site agrees even less than Kano city, so the city result is not an urban effect. ERA5 reads 23 to 42% below CHIRPS across the region and does not capture the short, intense storms of the Sahel well. A Nigerian pilot would need an index calibrated on CHIRPS or rain gauges, with publishers reporting from the same source.

Dry spells. Both datasets were also compared on the longest run of dry days in each season. Agreement is weak everywhere (at best 0.45, and near zero in the north west), because ERA5 tends to record light drizzle on dry days. A dry days index therefore needs CHIRPS or rain gauges, not ERA5.

Why calibration stays on ERA5. ERA5 reads about 30% below CHIRPS in every region, so the two cannot be mixed. The live publishers report the same ECMWF model family that ERA5 comes from, corroborated by GFS, so calibrating on ERA5 keeps the history and the live data consistent. CHIRPS serves as the independent check.

Next candidates

Regions in other African countries are checked next, using the same test. Nigeria returns as a candidate once a CHIRPS based index and publisher source exist.

Reproduce

CHIRPS seasons are fetched from the public ClimateSERV API and compared with the ERA5 files in mvua-contract/data/history, season by season.