1 of 59

Extended Outlook Crop Yield Conditions

Sep 14, 2026

Famine Early Warning Systems Network

2 of 59

Key takeaways

2

El Niño Forecast: The El Niño event is likely to be very strong and accompanied by a positive Indian Ocean Dipole, which makes impacts in East Africa and Southern Africa more likely. Following the El Niño, a La Niña is very likely to develop in 2027

Western Kenya maize, 2026: Using earth observations through early September, 31 of 47 counties have yields below the 2015–2025 mean; 13 are more than 20% below. The 14 North Rift / western highland counties average −24.0% (median −24.2%), range −36.5% (Trans Nzoia) to −6.7 % (Bomet).

Eastern East Africa Short Rains 2026/27: Compared to the August forecast, the likelihood of the most extreme precipitation has decreased. Based on these seasonal climate forecasts and analog years, maize and sorghum production in Somalia is likely to be normal to below-normal along parts of the Juba and Shabelle rivers, but above normal everywhere else and above normal at a regional scale unless floods of the magnitude observed in 1997/98 occur

Eastern East Africa Long Rains 2027: Too early to say, many of the potential impacts will depend on how long the El Niño event persists

3 of 59

Key takeaways

3

Southern Africa 2026/2027: Based on seasonal precipitation forecasts, 2026/2027 maize yields in Southern Africa are most likely to be below normal in Madagascar, Zimbabwe, South Africa, and Southern and Central Zambia. The impacts in Malawi and Northern Zambia are less certain. Regional total production of maize and sorghum during past El Niño events has been ~10-15% below normal. Groundnut and common bean averaged production deficits of 10-15% below trend in Zimbabwe, South Africa, Lesotho. Sweet potato production deficits were 20% below trend in Zimbabwe and cassava production deficits were 12% below trend in Malawi

Ethiopia Meher 2026 season: Crop production in Afar and eastern Amhara are expected to be significantly below normal in many zones. At a national level in official crop production statistics data, production of wheat, sorghum, and maize was significantly below trend in the analog year of 1997 but near-normal in 2015.

Central Asia 2026/2027: El Niño events tend to increase wheat yields in Central Asia for wheat, rice, and maize. For Afghanistan, El Niño events improve rainfed wheat production more than irrigated wheat with total production 0-10% above normal during El Niño events

4 of 59

Methods

4

5 of 59

Methods description

5

Forecasts are presented as one of four methods: yield forecasts based on the forecast for the El Niño Southern Oscillation (ENSO-based forecast), yield forecasts based on seasonal precipitation forecasts (Climate forecast-based yield forecasts), yield forecasts based on monitoring the progression of the growing season using remote sensing (remote sensing-based forecasts) or analyses of years that had climate conditions similar to what might be expected this year (analog year analyses). Each type of analysis is noted in the slide title.

ENSO-based global forecast: These are long lead time forecasts made up to a year ahead of harvest. They are based on forecasts of the El Niño Southern Oscillation (ENSO) only. These are probabilistic forecasts for the likelihood that crop yields will fall into the bottom tercile of the yield anomaly distribution. They are screened for skill because ENSO does not affect all countries

Climate forecast-based yield forecasts: These forecasts are made just before the beginning of the cropping season and into the early portion of the cropping season. They are based on seasonal multi-model precipitation forecasts from NOAA PSL S2S multi-model dataset

Remote sensing-based forecasts: These are short lead time forecasts made a few months before harvest. These forecasts use remote sensing information of e.g. soil moisture, precipitation, and vegetative health to train statistical models using subnational crop yield statistics. The statistical models then forecast end-of-season yields based on the progression of the growing season to date. Two different models (CAPE from UCSB and GEOCIF from UMD) are producing these forecast for select countries.

Analog year analyses: These are the least certain outlooks. They provide information where we don’t have remote sensing-based forecasts. These are not formal forecasts, so once environmental monitoring data during the mid-to-late portion of the growing season becomes avaialble (e.g. vegetative health, soil moisture, precipitation, etc.), that information should be prioritized.

Full methods for forecast models can be found here

6 of 59

ENSO Outlook

6

7 of 59

The El Niño will be very strong

7

The ENSO forecast indicates that at its peak intensity, this El Niño will be very strong (> 2.5C)

Based on the NOAA probabilities from July forecasts, the El Niño has a ~75% chance of being stronger than any event since at least 1950

The strength assessment from NOAA, which uses a single model, is largely consistent with the assessment from the IRI multi-model plume forecast from July

The NOAA forecast uses a relative ENSO index that accounts for climate change and should be considered more reliable. Likely intensity is 2.25-3C

8 of 59

A strongly positive Indian Ocean Dipole is likely

Along with the very strong El Niño, a positive Indian Ocean Dipole is likely, and may be a strong event similar to 1997

The Indian Ocean Dipole may affect the outlook for crop yields in the upcoming wet short-rains season for Somalia, due to the potential for flooding

8

9 of 59

Following El Niño, a La Niña event is very likely to develop

9

In agreement with the official

NOAA ENSO forecast,

NOAA-PSL Model-analog

ENSO forecasts indicate

that El Niño conditions are

likely to persist through

boreal spring

The NOAA-PSL Model-analog ENSO forecasts indicate that a transition to La Niña is likely to occur in July-September 2027, with a 60-100% likelihood, depending on the model ensemble used

This is in line with academic literature of when we have long-range predictability and the frequency with which La Niña events follow El Niño events (see e.g. Lenssen et al. 2024)

10 of 59

Global ENSO-based

Maize and wheat yield forecast

10

11 of 59

ENSO-based maize yield forecast

11

The developing El Niño is associated with a high probability of poor maize yields throughout Southern Africa, particularly in Zimbabwe and South Africa, with less certain outcomes in Malawi

The El Niño event increases the likelihood of good yields in

Southeast South America

12 of 59

ENSO-based wheat yield forecast

12

The forecast El Niño increases the odds of poor wheat yields in India and Morocco for 2027 harvests

Note that wheat in India is irrigated and so even years in the bottom ⅓ of events (lower tercile) have relatively stable yields

Wheat yields in Iran and the Southern US Great Plains are likely to be above-normal

13 of 59

Western Kenya June-September 2026/27

long rains maize yields

Remote sensing-based forecast

13

14 of 59

Remote sensing-based forecasts: GEOCIF

GEOCIF Forecast

Contact: Ritvik Sahajpal, ritvik@umd.edu

14

NDVI anomaly Evaporative stress anomaly Maize yield anomaly

31 of 47 counties are below the 2015–2025 mean; 13 are more than 20 % below. The 14 North Rift / western highland counties average −24.0 % (median −24.2 %), range −36.5 % (Trans Nzoia) to −6.7 % (Bomet).

These conditions arose from a decline in water availability during key portions of the growing season first in April then in June

15 of 59

Eastern East Africa Short Rains

OND 2026

2026/2027 Sorghum and maize yields

El Niño analog year analysis

15

16 of 59

Important indicators of the 2026 Short Rains/Deyr season

Historical analogs for OND 2026:

1982, 1994, 1997, 2006, 2015, 2019, 2023

Based on September analysis

As of September 14th, observed values indicate a positive IOD index (+0.65) during the past week. This was driven by warmth in the western IOD region, consistent with the OND 2026 forecast. Compared to August, models using September initial conditions slightly reduced the forecast strength for strong positive IOD, in line with neutral IOD as of September 6th. The OND 2026 positive IOD forecast ranks ~midway between the high 1997, 2019, 2023 and the moderate 2006, 1994 values, thus leading to 1994 becoming a 7th year in the previously 6-year analog set.

Forecasts of record-high, very warm western Indian Ocean IOD SST, and very strong El Nino are driving the OND 2026 IOD and RONI forecasts, making the outlook situation extreme and with no perfect analog. A lack of earlier IOD development veers 2026 farther from the IOD-driven 2019 season. 2015 similarly had a western IO-driven IOD, but 2026 is forecast to be much warmer. Ingredients for above-normal OND rainfall, especially in eastern areas, will likely be present. Anomalous westward moisture transports are possible as El Nino and the Indian Ocean SST gradient strengthens, while the very warm SST could promote heavy rain events. September model rainfall forecasts became less confident since August, consistent with IOD.

Positive IOD analogs

El Niño analogs

Years that fell within a 95% confidence interval around the current SST forecasts for RONI and IOD (right), and IOD and western Indian Ocean (next slide) were selected. The multimodel ensemble mean (NMME and ECMWF models) SST forecast values were first unbiased relative to 1991-2025 observations, using empirical quantile matching. This analysis (August) allows the unbiased current forecast to extend beyond the historical observations, using additive delta correction (observed maximum value + difference between current and previous forecast maximum).

This plot shows 1982, 1997, and 2015 as closest years based on El Niño, and 1997, 2006, 2019, and 2023 based on positive IOD, using the OND observed and forecast data. The large red triangle shows the unbiased 2026 forecast. The small red triangles show the 95% confidence interval, determined from the standard deviation of forecast errors. Points are colored by observed eastern East Africa region CHIRPS v3 rainfall tercile.

17 of 59

Rainfall anomalies: OND 1982, 1994, 1997, 2006, 2015, 2019, 2023

Data: CHIRPS v3.0 Analogs: OND 1982, 1994, 1997, 2006, 2015, 2019, 2023

These analog rainfall anomaly maps show historical OND rainfall outcomes during the identified seven analog years.

Dark green colors:

Rainfall totals were 100 mm to 300+ mm higher than average

OND Rainfall Anomaly (mm)

-300

+300

18 of 59

Wet outlook is consistent with September model rainfall forecasts, though odds have declined since August

Compared to the August forecast, the likelihood of the most extreme precipitation has decreased

August forecast

Current forecast (September)

NOAA PSL C3S Monthly to Seasonal Forecasts

19 of 59

Maize yields are likely to be below normal in riverine areas but above normal elsewhere

Compared to the August forecast, the likelihood of the most extreme precipitation has decreased

Based on seasonal climate forecasts, maize yields in Somalia are likely to be below normal in parts of Somalia along the Juba and Shabelle, but above normal everywhere else

19

20 of 59

Maize analog year short rains crop outcomes

20

Maize yields and production are over 30% below normal in areas along the Juba and Shabelle river during past analog years, although overall production and yields are above normal as favorable conditions elsewhere more than offset riverine losses

As the likelihood of a 1997/98 type event decreases, the likelihood that overall production will be above normal increases

21 of 59

Maize analog year short rains crop outcomes

21

Yield Area

22 of 59

Sorghum analog year short rains crop outcomes

22

As the likelihood of a 1997/98 type event decreases, the likelihood that overall production will be above normal increases

Sorghum yields and production are likely to be normal to above normal unless precipitation as extreme as in 1997/98 occurs

23 of 59

Sorghum analog year short rains crop outcomes

23

Yield Area

24 of 59

Ethiopia Kiremt 2026

Sorghum, wheat, and maize outlook

El Niño analog year analysis

24

25 of 59

Ethiopia analog years

25

Severe June/July deficit precip. analogs - 1987, 2015

A similarly strong, early developing El Niño also occured in 1997

Note that 1997 began with a wetter June before the July and August precipitation deficits began (see next slide)

From the CHC blog: https://blog.chc.ucsb.edu/?p=2049

26 of 59

Ethiopia analog years

26

Precipitation deficits were severe in July in all years

Precipitation deficits in August 2026 were less severe in the highlands than in 2015 or 1997

NDVI in Aug 2026 looks similar to that in 2016

Sensor discontinuity means that we do not have an NDVI comparison from 1997 to 2015

2026 2015

August NDVI

27 of 59

Ethiopia crop production anomalies

27

In official crop statistics, production of wheat, sorghum, and maize was further below trend in 1997 than in 2015 (no data for 1987). In 2015 many locations show near-normal production in official crop statistics

Given the similarity of precipitation deficits in 1997 and 2015, it’s not clear why the crop production would be near-normal in 2015 but it is possible

As August precipitation improved NDVI in Amhara and western Oromia, it’s possible that crop outcomes end somewhere between 1997/98 and 2015/16

Production of wheat, sorghum, and maize in 1997 was about -10 to 25%, -20 to 30%, and -20 to 30% relative to trend depending on the detrending method used

Maize Wheat Sorghum

1997 2015

28 of 59

Ethiopia wheat yield, area, production anomalies

28

Production deficits come primarily from strongly reduced area in 1997, with only moderate reductions in national-scale yield

29 of 59

Ethiopia maize yield, area, production anomalies

29

For maize as well, production deficits come primarily from strongly reduced area in 1997, with only moderate reductions in national-scale yield

30 of 59

Ethiopia sorghum yield, area, production anomalies

30

For sorghum, production deficits come from reductions in both area and yield roughly equally in 1997

31 of 59

Southern Africa 2026/2027 season

Climate forecast-based yield forecasts

El Niño analog year analysis

31

32 of 59

Climate forecast-based yield forecasts

Based on seasonal precipitation forecasts, maize yields in Southern Africa are most likely to be below normal in Madagascar, Zimbabwe, South Africa, and Southern and Central Zambia.

Predictors and target: Season-total and grain-fill precipitation forecasts from the NOAA PSL S2S multi-model dataset. The target is admin-level yield from HarvestStat Africa, which has been detrended while making sure there is no leakage.

Model: One pooled OLS per country × crop × season.

yield anomaly = β₀ + β₁·(z-scored seasonal rainfall forecast) + β₂·(z-scored grain-fill rainfall forecast) + error

32

33 of 59

Southern Africa Maize Yields Likely to be Below Normal

During historical

El Niño events, South African maize yields have been up to 40% below normal.

The last two very strong El Niños (2015, 1997) resulted in maize yield declines of ~20%.

South Africa may see worse outcomes in stronger El Niño events.

33

Current Forecast Historical El Niño Outcomes

34 of 59

2027 maize yields are often below normal in Southern Africa during El Niño years

During El Niño years, maize yields tend to be below-normal in Zimbabwe, South Africa, Malawi, and southern Zambia with regional total yields being -15% below normal on average

Given the developing El Niño, we can now forecast that maize yields in Southern Africa for the 2027 harvest are likely to be below normal, particularly in Zimbabwe and South Africa

Maize yields during El Niño years Maize yield forecast

35 of 59

2027 maize yields are often below normal in Southern Africa during El Niño years

The yield analysis is consistent with observed below-normal NDVI in Southern Africa during El Niño years

Teleconnections to Zimbabwe and South Africa are most consistent

Teleconnections to Zambia, Northern Mozambique, and Malawi depend on the event and are less reliable

Nov-Apr Vegetation health

36 of 59

2027 production of most crops are below normal in Southern Africa during El Niño years

Groundnut and common bean are both typically below normal during El Niño years, with average production deficits of 10-15% below trend in Zimbabwe, South Africa, Lesotho, but less severe deficits further north

Sweet potato production is typically strongly below normal in Zimbabwe, with average production deficits of 20%, but near normal in other countries

Cassava production is 12% below trend in Malawi on average in El Niño years, but near normal elsewhere

Groundnut Common bean

Sweet potato Cassava

37 of 59

2027 production of most crops are below normal in Southern Africa during El Niño years

Groundnut production is below normal in most countries during El Niño years, with deficits often being over 20% below trend in the worst years

38 of 59

2027 production of most crops are below normal in Southern Africa during El Niño years

Sweet potato production is most strongly below normal in Mozambique and Zimbabwe, with deficits average production deficits of 20% in ZW and 7% in MZ, but potential deficits exceeding 25% in MZ and deficits exceeding 50% in ZW

39 of 59

Central Asia wheat

2026/2027

Oct - Feb main season

40 of 59

Central Asia yield outlook: wheat yields during El Niño years

El Niño events tend to lead to above-normal precipitation in Central Asia, which in turn leads to above-normal wheat yield outcomes in Afghanistan and Iran

Oct- Dec Jan - Mar

41 of 59

Central Asia yield outlook: wheat yields during El Niño years

El Niño events tend to lead to improved wheat yields in Afghanistan, especially for rainfed wheat

Total wheat yield anomalies tend to be +0-10% in Afghanistan

(Shukla et al., 2024)

42 of 59

Global ENSO-based forecasts

42

43 of 59

ENSO-based crop yield forecast methods

43

Crop yields are detrended using a low-frequency filter equivalent to a 7 year running mean

Crop yield anomalies (deviations from the trend) are grouped into terciles

The frequency of lower-tericle outcomes conditioned on ENSO is combined with the probability of each state of ENSO using the NOAA PSL model-analog forecasts

For details, see Anderson et al. (2024) https://www.nature.com/articles/s41467-024-51555-8

44 of 59

Remote sensing-based forecasts:

UMD Global Earth Observations for Crop Yield Forecasting (GEOCIF) Model

44

45 of 59

Remote sensing-based forecasts: GEOCIF

1. Process EO Data

3. Create CIDs

Crop mask

2. Extract EO time-series

4. Train ML Model

5. Evaluate Model and Generate Results

We use a gradient boosting algorithm to predict crop yields based on CIDs
Gradient boosted models iteratively refines its predictions by focusing on the residuals
We divide the administrative regions into clusters using a nearest neighbor model and run a Generalized Additive Model to account for extreme event scenarios

45

46 of 59

Remote sensing-based forecasts: GEOCIF

Climatic factors explain 20%-52% of variance in crop yield anomalies globally (Lobell et al., 2007, Ray et al., 2015, Vogel et al., 2019)
An IPCC framework, Climatic Impact-Driver (CID) is a value that describes the state of the climate system for a specific location and at a given point of time
We compute CIDs representing two types: Mean, Extreme; and also seven categories: Drought, Heat, Temperature, Rain, Compound, Snow and Cold

46

47 of 59

Remote sensing-based forecasts:

UCSB CAPE Model

47

48 of 59