Two alert emails went out, on 3 and 5 August 2026, each reporting five newly crossed thresholds. Every threshold now active is a drought threshold, in seven regions, including one where the site itself classifies the hazard as flooding. Those percentages were not measurements. Every one of them came from a hard-coded function of a single Pacific sea surface temperature index, with no rainfall observation anywhere in the calculation. We are retracting them, showing exactly how they were produced, and setting out what the observational record actually says about drought right now.
This site exists to correct the coverage rather than repeat it. That obligation does not stop at our own front door, so this post starts with what we got wrong.
Our alert system emailed the subscriber list when internal thresholds were crossed. It fired twice, on 3 August and 5 August 2026, each time with the subject line "5 New Threshold(s) Crossed", to six addresses on the list, of which five delivered and one was an internal test address that bounced. For drought, the message body read, for each region:
Current El Niño conditions have pushed drought risk to [N]% for [region].
That sentence claims two things. It claims a current condition, and it claims El Niño caused it. Neither was established by anything in the calculation.
The alert state stored in our database shows what is currently set. Every active flag is a drought flag. Seven regions sit at the 60 percent rung, and two of them, Indonesia and Southeast Asia and the Peru and Ecuador coast, also sit at 80 percent. None of the ENSO-strength or analog thresholds is active, so everything that has gone out recently came from the ladder below.
Each region carried a hard-coded step function of one variable: the Oceanic Niño Index. The value it is currently reading, from our own cache, is +1.39°C for May-July 2026. For South Asia, in full:
tele: (a) => a >= 1.5 ? 92 : a >= 1.0 ? 78 : a >= 0.5 ? 62 : 0
ONI crosses +1.0, the function returns 78, the email reports 78 percent. Nothing about South Asian rainfall enters it. The same structure applied to all seven regions, so a single Pacific index crossing one threshold changed seven regional percentages at the same instant. These are the values the ladder holds at +1.39°C; they reached subscribers across the two emails as each rung flipped.
| Region | Percentage emailed | Rung that produced it | Any rainfall data used? |
|---|---|---|---|
| Indonesia & SE Asia | 80% | ONI ≥ 1.0 → 80 | No |
| Peru & Ecuador coast | 80% | ONI ≥ 1.0 → 80 | No |
| South Asia / India | 78% | ONI ≥ 1.0 → 78 | No |
| Eastern Australia | 75% | ONI ≥ 1.0 → 75 | No |
| Northeast Brazil | 73% | ONI ≥ 1.0 → 73 | No |
| Northern China | 68% | ONI ≥ 1.0 → 68 | No |
| Southern Africa | 65% | ONI ≥ 1.0 → 65 | No |
IAMElNino.com analysis. We cannot reconstruct where any individual rung came from. There is no citation in the code, no baseline period, no dataset, and no definition of what "drought risk" means: risk of what deficit, over what area, during what months, measured against what climatology. A number that cannot be reproduced from a stated method and a named dataset is not a measurement, whatever it looks like in an email.
One region makes the point on its own. The entry for the Peru and Ecuador coast is internally identified as peru_flood, and this site's dashboard classifies it as a flood region, because the canonical El Niño signal on that coast is heavy rainfall, not drought. But the alert's label and message strings are hard-coded to say "Drought Risk" for every member of the array. So subscribers were told that El Niño had pushed drought risk to 80 percent on a coastline where a strong El Niño is associated with flooding. Nothing in the code noticed, because nothing in the code was looking at the region.
Three further problems sit alongside it. The threshold was evaluated using ONI. CPC now uses RONI for official ENSO monitoring and prediction and verifies its official probabilities against it. For May-July 2026, CPC reports ONI at +1.4°C and RONI at +1.0°C. The alert also carried an "1877 analog score" produced by a weighted formula containing literal constants that are not derived from anything. And the drought percentages were static between rungs, so they could not respond to rainfall at all, in either direction.
This is the part that should have stopped the alert before it sent. The site already fetches observed rainfall from CHIRPS via ClimateSERV, for the same regions. Here is that feed against the percentages we emailed.
| Region | Alert said | Observed rainfall anomaly | Feed's own label |
|---|---|---|---|
| Indonesia & SE Asia | 80% drought risk | +39% (wetter) | NEAR NORMAL |
| Eastern Australia | 75% drought risk | +267% (wetter) | NEAR NORMAL |
| South Asia / India | 78% drought risk | +7% (wetter) | NEAR NORMAL |
| Northern China | 68% drought risk | +16,142% (see below) | NEAR NORMAL |
| Northeast Brazil | 73% drought risk | −94% (drier) | NEAR NORMAL |
| Southern Africa | 65% drought risk | −93% (drier) | NEAR NORMAL |
| Peru & Ecuador coast | 80% drought risk | −92% (drier) | NEAR NORMAL |
The obvious reading is that the alert contradicted our own observations. That reading is too generous to us, because the second column is not trustworthy either.
IAMElNino.com analysis. So the honest summary is worse than a contradiction between a good number and a bad one. We had two drought numbers on this site and neither was measuring drought. One was a function of a Pacific index. The other was a mislabelled, stale, partly undefined rainfall series. The on-page alert system has been removed from this website, and on 7 August 2026 the scheduled trigger on the backend service that sent the emails was switched off, so it cannot fire again. Subscribe and unsubscribe still work. The drought panel on the dashboard now carries a note stating that its Model Risk figure is a teleconnection estimate derived from ONI rather than an observation of current rainfall; until it is either rebuilt on observed rainfall or removed, treat it as under repair.
Deleting bad numbers is only half the job. There is a real and well-documented food-security problem in Southern Africa, alongside a forecast of elevated drought risk during the coming rainy season, and both come from an agency that does this properly. FEWS NET's Southern Africa Food Security Outlook, covering June 2026 to January 2027, is built on country-level analysis of harvests, prices, livelihoods and rainfall rather than on an index.
Its central finding on El Niño is a forecast about a season that has not started:
Read the tense. Southern Africa's main rains begin in October. The El Niño drought signal there is a forecast for a season that is still two months away. Our alert reported it as a percentage of current risk in early August.
An index-driven alert cannot represent any of the following, all of which are in the same report:
IAMElNino.com analysis. Seven Southern African countries in this Outlook carry a published population-in-need range. Summing the endpoints gives approximately 23.1 to 26.5 million people over the projection period. That is a crude construction: FEWS NET publishes per-country ranges and no regional total, and adding endpoints of independent ranges overstates the true spread. Treat it as an order of magnitude, not a figure. That figure will circulate. It should not be attributed to El Niño.
| Country | People in need (FEWS NET range) | Primary drivers FEWS NET names |
|---|---|---|
| DR Congo | 17 to 17.99 million | Conflict, displacement, Ebola outbreak |
| Madagascar | 1.5 to 1.99 million | Early stock depletion, cyclones, delayed 2026/27 rains |
| Malawi | 1.5 to 1.99 million | Below-average southern harvest, prices; need is down year on year |
| Mozambique | 1.5 to 1.99 million | Conflict in Cabo Delgado and Nampula, poor 2026 harvest |
| Angola | 1.0 to 1.49 million | Below-average harvest, terms of trade, lean season |
| Zambia | 500,000 to 749,999 | Seasonal lean period, localised shocks |
| Lesotho | 100,000 to 249,000 | Early stock depletion, waterlogging damage, remittances |
DR Congo alone is 17 to 18 million of that total, about 74 percent of the low end, and FEWS NET names conflict, mass displacement and an Ebola outbreak as the drivers there, not rainfall. Strip DR Congo out and the remaining six countries total approximately 6.1 to 8.5 million on the same crude basis. Even that residual is driven by a mix of conflict, prices, cyclones and harvest timing, with El Niño appearing mainly as an expected influence on a rainy season that begins in October.
This is the same failure mode as our alert, one step up: taking a real aggregate and attaching a single cause to it. We have written before about how a large impact number acquires a cause it was never measuring.
We are not rebuilding regional alerts until each one can state all four of these. Publishing the standard so it can be held against us:
Until then, the honest product is a link to FEWS NET, national meteorological services, the SADC Climate Services Centre and NOAA CPC's African Desk products, which do this work with the data and the caveats intact.
We emailed seven drought percentages that were rungs on a hard-coded ladder keyed to one Pacific index, described them as current conditions caused by El Niño, and did so while our own rainfall feed showed several of those regions wetter than normal and was itself mislabelled and stale. Both numbers have been withdrawn. The alert UI is gone from the site, the scheduled service behind the emails was switched off on 7 August, and the dashboard drought panel now says plainly that its Model Risk figure is not an observation.
The real Southern African story is more useful than the one we sent. Harvests in 2026 were mixed rather than uniformly poor, Zambia set a maize record, Malawi's need has fallen year on year, and the El Niño drought signal is a forecast for a rainy season that starts in October. A risk percentage in August could not have told you any of that, which is the point.