Yield Forecast
Forecasts, updated monthly, of how this season's crop yields compare with normal, including how the season has gone so far and the range of possible yields at harvest.
Yield compared with past years
Brown is below the usual yield, green above it
Select a crop to load the forecast
Using the map
Pick countries and a crop, then click a coloured cell or area to see its forecast beside the map (below it on smaller screens). Clear cell, or Clear area, in its heading removes it.
- Countries
- One country is ticked at first. Tick more to compare them on one map; the crop list follows.
- Crop
- Crops forecast this month in at least one ticked country. Crops grown twice a year are listed per season.
- Find a place
- Jumps to a town or city, ticks its country and selects the unit it falls in.
- Show by
- Each grid cell (about 11 km), or crop-area weighted averages over Admin 2 areas (districts), Admin 1 areas (regions, provinces or counties) or whole countries.
- Colour by skill
- Colour the cells by how skilful the forecast has been there in past years, instead of by this season's forecast. Grid cells only.
- Opacity
- Fade the forecast layer to see borders and place names underneath.
The forecast is updated every month with the latest seasonal weather forecast, and the date of the last update is shown under the title.
Reading the colours
Colour is the percentile of the forecast within that place's own past 30 years: brown a low year for that place, green a high one.
Percentile rather than percent difference, because year-to-year variability differs widely between places: a 10% shortfall is a bad year in one place and a normal one in another, while the 20th percentile means the same everywhere.
The numbers
- Percentile
- Where this season's expected yield (the average of the 51 forecast members) falls among the past 30 years of this place. Around 10 puts it among the lowest years, 50 is a typical year and 90 among the highest.
- Anomaly
- How far the expected yield is from the average of the past 30 years for this crop and place, in percent. -15% means 15% below that average.
- Below / near / above
- Share of the 51 forecast members in the lowest, middle and highest third of the past 30 years. 60 / 30 / 10 means most members point to a below-normal yield.
- Skill (CRPSS)
- How much better the forecast did in past years than simply expecting a repeat of the previous five seasons. 0 is no better, 0.4 means 40% better. Only places with positive skill are shown.
- Area covered
- Share of the crop area behind an area value. Cells without a forecast or without skill drop out.
Percentile and anomaly can point in different directions where a place has a few very high years: they raise the average without changing the ranking.
Reading the distribution
Rose is the 51 forecast members, blue the 30 reference years, both as percent difference from the reference average (the dashed line at 0). A narrow rose curve means the members agree, as late in a season, when most of the weather has already happened.
Read the anomaly against the historical variability given just above the chart: the typical size of a year's departure from the average. A 20% shortfall where yields normally vary by 5% is unusual; where they vary by 40%, it is not.
Reading the season so far
"Season so far" shows how the crop has developed from sowing up to the last day of observed weather, the end of the month before the forecast. It appears only while a season is under way. Nothing in it is forecast. The rose line is this season; the blue dashed line and band are the normal and usual range (5-95%) of the past 30 seasons at the same number of days after sowing.
Biomass to date is how much the crop has grown so far; 100% is what a normal season reaches by harvest. From mid-season on, a crop that is behind rarely catches up fully.
Water needs met is the share of the crop's water demand met since sowing. It shows water stress early, before it shows in growth. Where nearly all water needs are met every year, 81% can be the lowest in 30 years, so read the percentage next to the percentile.
The two can disagree: where rain is usually ample, cloud or temperature often limit growth more than water does. Where water is scarce they usually move together.
Illustrative example
Biomass to date
Water needs met
- Late May and June, a dry spell: water needs met drops below its usual range and biomass starts to fall behind.
- From July, rain returns: water needs met recovers, slowly, because it counts everything since sowing.
- Biomass stays behind: growth lost in the dry spell is not made up. What it costs in yield depends on when it hit; stress around flowering costs most.
How the indicators are built. LPJmL runs
daily on ERA5-Land weather up to the last observed day, from a 31 December state of its
continuous ERA5-Land historical run. For each crop (rainfed) and grid cell the season runs
from the sowing day (sdate) to harvest, the first day the crop's NPP and water
demand fall to zero.
- Biomass to date: daily crop net primary production (
pft_npp, g C/m²) summed from sowing, converted to dry matter (carbon is 45% of dry matter). - Water needs met:
pft_water_supplysummed from sowing divided bypft_water_demandsummed from sowing. Demand is the transpiration the canopy would reach that day without water stress; supply is what the roots can take up from the soil. - Reference: the past 30 seasons of the same run, compared on the same day after sowing. Areas are crop-area weighted means of their cells (SPAM 2017), shown when at least half the crop area has a season under way. Biomass of crops with a significant national trend is moved to the current year, separately for each day of the season.
Why some places are blank
Blank: the crop is not grown there.
Hatched: the crop is grown but has no forecast this month, because its harvest falls outside the forecast window, because the model already harvested it before this month, or because the forecast has had no skill there in past years. This is also why the crop list changes from month to month.
Areas need at least half their crop area covered by forecast cells, so one skilful cell cannot stand for a whole district.
Each month, a crop model is run with the latest seasonal weather forecasts to estimate how this season's yields will compare with a normal year, and with observed weather to show how the season has gone so far. This page describes how the forecasts on this site are made. It is updated as the service develops.
How a forecast is made
Up to the start of the forecast, the crop model is driven by observed weather from ERA5-Land. From then on it is driven by the ECMWF SEAS5 seasonal forecast, which looks seven months ahead. SEAS5 is first corrected for its systematic errors and brought to the model's grid of about 11 km, so that its weather matches what is typically observed in each place.
SEAS5 provides 51 equally likely versions of the coming months, and the crop model is run with each. The 51 resulting yields at harvest form the forecast, and how widely they spread shows how certain it is.
The crop model
LPJmL simulates crop growth day by day from weather, soil and atmospheric CO2, for rainfed crops. Nutrients are assumed not to limit growth, so differences between years reflect the weather. Each crop matures once it has received a set amount of warmth, derived from recent temperatures over its usual growing period.
The model is not calibrated to reported yields. Yields are therefore shown relative to the model's own normal, never in tonnes per hectare.
Crops and growing seasons
The crops covered make up at least 1% of a country's harvested area, and are shown where at least 10 hectares are grown. Sowing and harvest dates come from FAO crop calendars by agro-ecological zone, checked with farmers and extension officers in each country. Crops grown twice a year are modelled as two seasons where subnational statistics report both; elsewhere only the main season is modelled.
- Ghana
- Maize, rice, sorghum, millet, groundnuts, soybean, beans, other pulses, yams and taro
- Kenya
- Maize, beans and sorghum (each in two seasons), pigeon peas, other pulses, millet, wheat and potatoes
- Zimbabwe
- Maize, sorghum, millet, groundnuts, cotton, soybean, sunflower, beans, other pulses, wheat and sweet potatoes
On the map, wheat appears as temperate cereals, potatoes as temperate roots, and other pulses such as cowpeas and Bambara beans as pulses. Not every crop has a forecast every month, since that depends on its growing season and on how skilful the forecast is at that time of year.
What "normal" means
Normal is what the same model simulates with observed weather over 1994-2023. Where yields show a long-term trend, for example from rising CO2, past years are adjusted to today's level so that they are compared fairly with this season.
How skill is measured
The forecast was re-run for every month of 1994-2023 as if issued at the time, and compared with the yields simulated with observed weather. This shows how well it predicts the model's yields, not reported yields.
Skill is measured with the fair continuous ranked probability skill score (CRPSS). It tells how much closer the forecast came than simply expecting a repeat of the previous five seasons. Only places where the forecast did better are shown. Skill is generally weak around sowing and grows as the season advances (Saldivia Gonzatti et al., 2026).
Limitations
Only rainfed production is simulated, and only annual crops are covered. Perennial and multi-year crops such as sugarcane and cassava are not, and some important crops such as cocoa, tea and tobacco cannot be represented by the model. Sowing dates are fixed for each agro-ecological zone and do not follow how farmers adapt to the season. Where crops grow relies on a single crop map (SPAM 2017).
Data sources
- Seasonal weather forecasts
- ECMWF SEAS5
- Observed weather
- ERA5-Land reanalysis
- Sowing and harvest dates
- FAO crop calendars by agro-ecological zone, validated by local experts
- Crop areas
- SPAM 2017
- Soil
- USDA-NRCS soil texture; ISIMIP3a topsoil pH
- Nitrogen deposition
- ISIMIP3b
- Atmospheric CO2
- NOAA/EPA observational records
- Lakes
- Global Lakes and Wetlands Database
- Administrative boundaries
- GADM v4.1
Reference
The forecasting approach and its skill over 1994-2023 are published in:
Saldivia Gonzatti, I., Siderius, C., Paparrizos, S. and Biemans, H. (2026). Process-based seasonal crop yield forecasting for early warning in sub-Saharan Africa. Environmental Research Letters, 21(18), 184030. doi:10.1088/1748-9326/aea797