This dashboard does not produce its own climate model. It takes published
data from national meteorological agencies and research institutes, and does three things
with it: cuts it to the areas One Acre Fund actually works in, puts it in plain
language, and sets it against what the same conditions did in past years. Every
step that changes a number is listed below, so any figure can be traced back to its source.
Page generated 21 September 2026.
Seasonal forecasts
Precipitation and temperature outlooks
International Research Institute for Climate and Society (IRI),
Columbia University ·
IRI Data Library
- What we take
- The North American Multi-Model Ensemble seasonal forecast
(
ELR_coeffsv2), for each monthly issue and each of the four
three-month target seasons ahead, for both rainfall and temperature.
- What we do
- Calibrate the raw model coefficients into tercile probabilities using Extended
Logistic Regression, apply a light Gaussian smoother (σ = 0.75) to remove
grid noise, mask cells that are climatologically dry for that season, and clip the result
to each country. The map then shows the most likely third and how confident the
forecast is about it. We change no probabilities by hand.
- Reading it
- Deeper colour means a higher chance; white means the forecast has no clear signal.
A plain-language summary and an operational note are generated from the same numbers.
ENSO forecast plume and outlook text
- What we take
- IRI's published plume of model forecasts for Niño 3.4, and the wording of their
monthly outlook statement.
- What we do
- Nothing. These are IRI's own calibrated values, redrawn in the dashboard's
colours so they work offline. We deliberately do not build our own plume from raw model
output: an earlier attempt produced long-lead values several degrees too warm, because
doing it properly means replicating IRI's full per-model bias correction. Since
September 2026 IRI no longer serves the plume numbers directly, so they are read back
from IRI's published figure at refresh; the values are IRI's own to about a hundredth
of a degree.
Ocean observations
Niño 3.4 and the Oceanic Niño Index
NOAA Climate Prediction Center and Physical Sciences Laboratory ·
CPC
- What we take
- The weekly Niño 3.4 sea-surface temperature anomaly and the monthly ONI, exactly as
NOAA publishes them.
- What we do
- Parse and plot them. No smoothing, no re-basing. The ONI is also what we use to label
each past year as El Niño, La Niña or neutral in the historical comparisons, so
our year groupings match NOAA's official classification rather than a definition of our
own.
Pacific sea-surface temperature animation
NOAA/NCEI OISST v2.1, served by NOAA Physical Sciences Laboratory ·
PSL
- What we take
- The daily quarter-degree sea-surface temperature anomaly field, cut to the
Pacific window (110°E–70°W, 40°N–40°S) for the current year.
- What we do
- Average each seven days into a weekly mean, coarsen to half a degree and apply a small
Gaussian smoother so the basin-scale pattern is visible rather than mesoscale speckle,
then render one frame per week. Colour saturates at ±5 °C.
- Cross-check
- Averaging this grid over the Niño 3.4 box reproduces CPC's published weekly index
with a mean difference of 0.03 °C and a correlation of 0.996 across
every week of the year, so the animation agrees with the figures quoted elsewhere on the
dashboard.
Indian Ocean Dipole
- What we take
- The weekly Dipole Mode Index, the difference between western and eastern Indian Ocean
sea-surface temperature.
- What we do
- Parse and plot as published, and use it to split past years by IOD phase
(positive at or above +0.4 °C, negative at or below −0.4 °C).
The weekly record begins in 2008, so older El Niño years are deliberately left off
the IOD comparison chart rather than back-filled from a different dataset.
Column moisture over Kenya
- What we take
- Monthly means of total column water vapour (precipitable water) from 1979, regridded to
0.5° on the way out.
- What we do
- Average October–December over a Kenya box, express each season against the
1991–2020 mean, and colour the seasons by the IOD phase from the Dipole Mode Index
above. The chart on the IOD watch page is rebuilt by hand once a season has ended rather
than on every refresh, because ERA5 is finalised a few months after the fact.
- Attribution
- Generated using Copernicus Climate Change Service information (2026). Neither the
European Commission nor ECMWF is responsible for any use that may be made of it.
Historical rainfall
What past El Niño and La Niña years actually did
TAMSAT v3.1, University of Reading ·
TAMSAT
- What we take
- Satellite-derived rainfall estimates back to 1983, for each country's actual growing
seasons rather than fixed calendar quarters.
- What we do
- Total the rainfall over each growing season, express it as a percentage against the
1991–2020 average for that same season, and average over the One Acre Fund
operating area rather than the whole country. Aggregating over the operating area rather
than the country sharpens the signal noticeably: Malawi's El Niño response goes from
about −1 % country-wide to −7 % over the area we actually farm.
- How each season is grouped
- Seasons are grouped by NOAA's ONI, but over the period each season actually
responds to, not always its own months. Most seasons respond to the Pacific while
they run. The East African March–June rains are different: they respond to the
El Niño or La Niña that peaked just before them, so we group them by that peak
(ONI over November–January) and by the preceding September–November IOD.
A single event therefore covers a run of seasons, and the dashboard names the same event
across all of them. Event strength is the event's peak, which is why the
March–June 2024 Kenyan long rains are labelled part of the very strong 2023–24
El Niño even though the Pacific had already cooled to +0.8 by the time they fell.
- Honest limits
- Most seasons have only 8 to 14 El Niño years on record. The dashboard rates each
season's signal strength openly, and marks the ones where the answer depends on the IOD
as well. Rwanda's and Burundi's March–May rains stay weakly linked to ENSO under
either grouping; they track the Indian Ocean Dipole more than the Pacific.
Other layers
Desert locust observations
- What we take
- Confirmed field observations reported to FAO over the last twelve months.
- What we do
- Plot the ones recording locusts as present, filterable by time window and by
development stage. No modelling or forecasting is applied.
National boundaries and coastlines
- What we do
- The outlines shown here have been modified. They are simplified with a
0.02° tolerance to keep the page small. geoBoundaries outlines exclude large inland
lakes, which is why Lake Malawi, Lake Victoria and Lake Tanganyika appear outside the
national areas. Coastlines on the two explainer graphics come from Natural Earth.
One Acre Fund operating areas
One Acre Fund internal site records
- What we do
- Site coordinates are snapped to a grid of roughly five kilometres, then buffered
by four kilometres and dissolved into a single footprint. Any site falling outside its
country's border is dropped, and flagged in the build log so it can be corrected at
source.
- Why
- The grid snap is deliberate. Without it, a site far from any other stays a lone circle
after the dissolve, and the centre of that circle would give away the site itself. Snapping
first means the published footprint shows only which five-kilometre cell a site sits in.
No individual farmer or site location is published.
Licences
| Source | Licence | Applies to |
|---|
| geoBoundaries | Public domain | Kenya, Burundi |
| geoBoundaries | CC BY 4.0 | Rwanda, Tanzania, Nigeria |
| geoBoundaries | CC BY 3.0 IGO | Zambia |
| geoBoundaries | ODbL 1.0 | Uganda, Malawi, DR Congo, Ethiopia |
| Natural Earth | Public domain | Coastlines on the explainer graphics |
| Copernicus C3S / ECMWF ERA5 | Licence to use Copernicus products | Column moisture over Kenya |
| Plotly.js | MIT, © 2012–2024 Plotly, Inc. | All charts and maps |
Boundary data for the four countries released under the
ODbL
remains available under that licence. Climate and observation data are used under the terms of
the publishing agencies; where an agency asks to be cited, it is credited above and in the
dashboard footer.
How current this is
- Forecasts refresh monthly. IRI issues a new set in the second week of each month,
and the dashboard is rebuilt after it lands. The header always shows which issue is on
screen, and the archive of past issues stays available on the slider.
- Ocean indices refresh with the forecast. Niño 3.4 and the DMI are weekly, so
they are usually a few days behind the present.
- The historical record barely moves. It is only rebuilt when a growing season
completes and adds a year, since nothing else about it changes.
- If a source cannot be reached, the dashboard says so rather than quietly showing
stale data, and the rest of the page is still rebuilt from what did arrive.
