Crop Statistics
See reported yield, production, and harvested area by country and region. Also includes information about how our crop model performs against reported statistics.
Click any area to see its yield history
Using the map
Pick a crop and what to show, then click an area or grid cell to see its history beside the map (below it on smaller screens). Clear area, or Clear cell, in its heading removes it. The map opens on the latest year with data; move the year slider to look back.
- Countries
- All are ticked at first. Untick countries to show only the others; the crop list follows.
- Crop
- Every crop the model grows in at least one ticked country. Where the model grows a crop in two seasons, it is listed for the whole year and for each season.
- Find a place
- Jumps to a town or city, ticks its country and selects the area it falls in.
- Show
- Reported values come from official statistics: HarvestStat for regions, provinces and counties, FAOSTAT for countries.
- Show by
- Admin 1 areas (regions, provinces or counties) or whole countries. Grid cells (about 11 km) exist only for the model, since statistics are reported per area.
- Year
- The year whose values colour the map and are reported for the selected place. Agreement is one number for the whole period, so with agreement shown the year only moves the mark in the place's charts.
- Opacity
- Fade the data layer to see borders and place names underneath.
What the map can show
- Reported yield, harvested area, production
- Values from the statistics, in tonnes per hectare, hectares and tonnes.
- Reported yield anomaly
- How far the reported yield of the year was from its trend.
- Model yield anomaly
- The same for the model, per grid cell or summed over each area.
- Model–reported agreement
- How closely the two anomalies rose and fell together over the years both have.
The model's own yields are never shown in tonnes per hectare: the model is not calibrated to the statistics, so its levels are not comparable with them. What can be compared is which years were better or worse than usual.
Reading the colours
How far a year was from the trend of that place, in standard deviations of its own year-to-year swings. −1 is a clearly poor year, +1 a clearly good one, beyond ±2 is rare. Model and statistics are each measured against their own trend, so they share this scale.
Correlation between the model's and the statistics' anomalies over the years both have, with at least 10 years needed. A value of 1 means they rise and fall together every year, 0 that the model says nothing about which years were good or bad, negative that they tend to disagree.
For reported yield, area and production, darker means more. Colour classes are set from every year of the shown areas, so a colour means the same amount whichever year is selected. Hatched areas report this crop, but not in the selected year.
Reading the panel
Reported values. Values as reported in the statistics. The dashed line is the trend removed before comparing years: a straight line or a gentle curve, whichever fits best, or the average when there is no clear trend.
Yield anomalies. Each series minus its own trend, divided by its typical year-to-year swing. This removes the difference in level between model and statistics, so the lines can be compared year by year: where they move together, the model captures what made that year good or bad. Agreement figures based on fewer than 20 years are marked to read with caution: a few unusual years can then decide the result.
Crops grown here. Harvested area of the main crops here. Model land use is the SPAM 2017 map the model is run with, the same in every year; reported area is the average of the last five reported years.
Grid cells. A grid cell has no statistics of its own, so its model anomalies are drawn against the reported anomalies of the province it lies in.
Why some places are blank or hatched
A blank area has no statistics for the crop at all, or the model grows less than 500 hectares of it there. A hatched area reports the crop, but not in the selected year. Country totals from HarvestStat are shown only for years in which the reporting regions hold at least 75% of the crop's area, so a year reported by a handful of regions is not mistaken for a national figure.
This page sets reported crop statistics since 1994 next to the LPJmL crop model, run with observed weather. It shows what the statistics say about each year, and how well the model captures which years were good or bad. This page describes where the data come from and how the comparison is made.
Reported statistics
Regional figures come from HarvestStat Africa, which compiles official subnational crop statistics. National figures come from FAOSTAT. Both report harvested area and production; yield is production divided by area. Statistics for several crops are summed into the model's crop groups, for example cowpeas and other dry pulses into pulses; the panel lists what a group contains.
In Kenya, maize, beans and sorghum are reported for the long and short rains separately. These appear as first and second season, and the whole-year figure adds them up, but only in years in which every season was reported.
National totals from HarvestStat add up the regions reporting that year. They are shown only for years in which those regions hold at least 75% of the crop's harvested area in the SPAM 2017 crop map, so a year reported by a few regions is not read as a national figure. Weighting regions by the crop's area rather than by their land area matters. For example, in Kenya the five largest counties cover almost half of the country but grow under 1% of its maize.
Cleaning the statistics
Values flagged as outliers by HarvestStat are removed first. Each remaining yield series, one per crop, area and season, is then checked for years far above the series' usual level that also stand well above the years before or after, and for isolated spikes that are not seen for the same crop anywhere else in the country. Finally, yields beyond what a crop plausibly reaches are left out.
The crop model
LPJmL simulates crop growth day by day from weather, soil and atmospheric CO2, for rainfed crops, on a grid of about 11 km. Here it is driven by observed weather from ERA5-Land. The crop map is fixed at SPAM 2017 in every year, so changes over time come from weather, CO2 and nitrogen deposition only.
The model is not calibrated to reported yields, so its yields are never shown in tonnes per hectare. For an area, the model's production is summed over the area's crop land and divided by its crop area. Areas where the model grows less than 500 hectares of a crop have no model values.
Crops
Every crop the model grows in a country is included:
From yields to anomalies
Reported yields rise and fall for reasons the model does not represent, such as new varieties, more fertiliser or changes in where a crop is grown. These act slowly, so each series has its long-term trend removed first: a straight line or a gentle curve (a second-order polynomial), whichever fits significantly better, or simply the average when neither does. What remains is divided by its own standard deviation, so it is measured in typical year-to-year swings.
The model can have a trend too, even with its crop map and management fixed: rising CO2 makes crops grow more, nitrogen deposition changes, and the climate itself has warmed and shifted over the period. In the statistics, these slow changes are mixed with new varieties and fertiliser and cannot be told apart. Model and statistics are therefore treated the same way, each against its own trend, so the comparison rests on which years were good or bad.
Agreement
Agreement is the Pearson correlation between the model's and the statistics' anomalies over the years both have. It is computed only with at least 10 shared years, and is marked to read with caution below 20. For a grid cell, which has no statistics of its own, the cell is compared with the region it lies in.
As a check that the results do not hinge on how trends are removed, the comparison was repeated on year-to-year changes in yield, which need no trend at all. Both give a similar picture.
Limitations
Only rainfed production is simulated, while the statistics include irrigated fields, which matters most for rice and sugarcane. Cassava and sugarcane are harvested over more than one year in practice but simulated as seasonal crops. The model's crop map does not change over time. Reported statistics differ in quality between countries and years, and regions report some years and not others, so a short record can show strong agreement or disagreement by chance.
Data sources
- Subnational statistics
- HarvestStat Africa v1.0 (regions, provinces and counties)
- National statistics
- FAOSTAT crop production
- Observed weather
- ERA5-Land reanalysis
- Crop areas
- SPAM 2017
- Sowing and harvest dates
- FAO crop calendars by agro-ecological zone
- Administrative boundaries
- GADM v4.1