Method in brief
Three ways to weigh a harvest without a scale
Kazakhstan's spring wheat grows on the northern steppe. Four oblasts, Kostanay, Akmola, North Kazakhstan and Pavlodar, hold 83 percent of the country's wheat area. We examined every 10-kilometre patch of them.
The crop model calculates how much wheat each patch could grow given its soil and that year's daily rain and heat. Its level is set once, from high-resolution satellite yield maps for 2023 and 2024.
The satellite canopy measures how green the cropland stays from May to August. A greener, longer season means a bigger crop, so comparing each summer's greenness with past harvests turns it into a yield estimate.
The records show what others saw: USDA's attaché in Astana, the customs of the countries that trade with Kazakhstan, and the government's own monthly grain stocks.
The grain belt
Share of national wheat area, average 2018-2022
12010 to 2024
In ordinary years, the numbers agree
Before testing a near-record, a method has to pass an easier test: does it match the official figures when nothing unusual is going on?
It does. Landsat satellites have photographed every wheat field in the grain belt each summer since before 2010. Their greenness, turned into yield without ever seeing that year's official figure, lands within 2 centners per hectare of the official grain-belt yield in 14 of the 15 years from 2010 to 2024, through the droughts of 2010 and 2012 and the good harvest of 2024. The one exception is 2011, the year of Kazakhstan's record grain harvest: officially 17.3 c/ha, against 12.7 c/ha from space.
A centner per hectare (c/ha) is 100 kilograms of grain per hectare.
Grain-belt wheat yield, c/ha
Official figure and Landsat estimate, 2010-2024; each year estimated without its own official figure
Satellite and official figures move together in poor years and good, except in the record year 2011.
2Harvest 2025
The fields looked like 2024
2024 was a wet, mild season and a good crop. In 2025 the crop model, fed with that year's rain and temperature, gives exactly the same yield as in 2024. Satellite images show the fields stayed slightly less green over the summer than in 2024, which translates into a yield 4 percent lower.
Both independent measures say 2025 repeated 2024. The official figure says the yield jumped by a fifth.
From 2024 to 2025: what changed?
Grain-belt wheat yield, c/ha
Only the official figure rises.
3Winter 2025/26
A near-record the fields did not see
The statistics bureau reported 19.3 million tonnes of wheat for 2025, with a grain-belt yield of 16.8 c/ha. The crop model puts it at 14.1 c/ha and the satellite canopy at 14.0 c/ha.
The gap is 2.7 c/ha. Across the belt's 10.1 million hectares of harvested wheat, that is 2.7 million tonnes: one tonne in every seven of the official harvest.
The official 2025 wheat harvest, 19.3 Mt
Split by what independent evidence supports
4March 2026
USDA counts fewer fields
USDA's agricultural attaché in Astana makes its own estimate of the crop. In March 2026 it put 2025 wheat at 18.0 million tonnes and stated its doubts plainly.
"Post and industry sources believe this figure is overstated based on observed yields in the field."USDA Foreign Agricultural Service, Astana, 2 March 2026
The attaché also finds 0.74 million hectares less harvested wheat than the official 12.20 million, 6 percent fewer fields.
The official count has 740,000 more hectares than USDA finds
Harvested wheat area, 2025. Each square is 100,000 hectares
5The 2025/26 season
If the crop was smaller, how did exports hold up?
Kazakhstan exported 11.6 million tonnes of wheat in the 2025/26 season, the most since 2011. A crop 2.7 Mt smaller would need grain from somewhere else. Part of it came from Russia.
Wheat imports tripled to 1.5 million tonnes. USDA's attaché reports the imports from September to February were nearly all Russian feed wheat, milled into feed flour and shipped to China as a Kazakh product. Feed-flour exports to China grew more than fifty-fold in three years.
Kazakh feed flour shipped to China, thousand tonnes
Calendar years, as reported by Kazakhstan
Russian wheat in, Kazakh feed flour out.
6March 2026
China shuts the door
China buys Kazakh feed flour because it is Kazakh. In March 2026 Beijing made every feed-flour mill re-register, approving only 27 of 66 at first, and then banned Russian wheat from Kazakh feed flour altogether.
Kazakhstan's own government had already acknowledged the practice: in November 2024 the Prime Minister told parliament that the country's wheat import ban had restricted the re-export of Russian grain to Uzbekistan and China.
The route was closed, not redirected. Russia's own data show China bought only about 50,000 tonnes of Russian wheat directly from January to July 2026.
How the route was closed
Key decisions, 2024-2026
7Since 2022
Grain that crosses without being counted
Recorded imports are only part of the flow. The road border with Russia has no weighbridges, and a Kazakh label pays: shipping Russian wheat to Uzbekistan in transit costs about $50 a tonne more than shipping Kazakh wheat.
Kazakhstan's Grain Union estimated that actual Russian wheat imports in 2022 exceeded 2.5 million tonnes, against 1.6 million recorded. In September 2026 the financial monitoring agency completed a case against five suspects over 130,000 tonnes passed off as Kazakh. It is one case among many: industry estimates for 2022 and 2023 put the unrecorded flow at 1 to 2 million tonnes a year.
Why it happens, and how much
Freight to Uzbekistan, and Russian wheat into Kazakhstan
8August 2026
Where the shadow grain sits
Every month the statistics bureau publishes grain stocks by holder. Stocks in licensed warehouses are backed by digital receipts and match the receipt platform. Stocks that farms report on their own are checked by no one.
On 1 August 2026, at the end of the season, the bureau counted 5.0 Mt of wheat. USDA and the Grain Union put the carryover at 2 to 3 Mt. The difference is almost exactly what farms reported holding, a line that has more than doubled since 2022.
Remove farm-reported stocks and the count matches USDA
Wheat stocks at the end of the 2025/26 season, million tonnes
Farm-reported wheat stocks on 1 August, million tonnes
9Harvest 2026
A hot July and a smaller crop
July 2026 was 2 to 4 °C hotter than July 2025 across the belt, and Akmola and Pavlodar received a third less rain from May to August. The crop peaked greener than in any year since 2010, then faded early.
Model and satellite canopy agree that belt yields fell 6 to 11 percent, most where the rain failed. That points to about 15 to 16 million tonnes of wheat.
Yields fell most where the summer rain failed
Modelled wheat yield, change from 2025 to 2026
10September 2026
The official yield is still above the evidence
On 29 September, with 83 percent of the area harvested, the agriculture ministry reported an interim grain yield of 14.7 c/ha and forecast about 23 million tonnes of grain, below 2025.
Weather and canopy support 13.4 to 13.8 c/ha. The ministry's figure sits 0.9 to 1.3 c/ha higher: 1.5 to 2 Mt across 16.3 million hectares of grain. The gap is narrower than in 2025, but it has not closed.
Interim yield, 29 September 2026, c/ha
Bunker weight
What it adds up to
Five findings, in the order they happened
Official wheat beyond what weather and canopy support, after a 2024 season in which the measures agreed.
USDA's attaché finds 0.74 million fewer hectares of harvested wheat than the official count.
Wheat imported, mostly Russian; until March 2026 nearly all of it went into feed flour for China.
Unverified farm-reported stocks: the whole excess of the official count over USDA and the Grain Union.
Belt yields below 2025 on independent evidence, about 15 to 16 Mt of wheat; the ministry's interim yield runs 1.5 to 2 Mt higher.
What to watch: the statistics bureau publishes the final 2026 harvest in early 2027. The evidence from the fields points to 15 to 16 Mt of wheat. The further the final figure sits above that range, the more of the harvest exists only on paper.
Sources and method
Crop model
GeoAquaCrop, the open-source regional toolchain for FAO's AquaCrop built by Seyed Hossein Hosseini, Josias Láng-Ritter, Chris Bowden, Henrikki Tenkanen and Tim Foster, run with AquaCrop-OSPy at 0.1° for Kostanay, Akmola, North Kazakhstan and Pavlodar; AgERA5 daily weather (Copernicus); SoilGrids soils; 30 m wheat maps for crop area. Sowing 20 May, 105-day season. One belt-wide yield level fitted to 30 m satellite yield maps for 2023-2024; GeoAquaCrop's optional correction to official statistics was not used. Model yields converted to 14% moisture.Satellite canopy
MODIS MOD13Q1 NDVI over ESA WorldCover cropland, 2010-2026, oblast means weighted by wheat area. Canopy-implied yield from the 2010-2024 relationship between season-integrated NDVI and official belt yield (r = 0.77), so 2025 and 2026 are estimated out of sample.Landsat fields
Landsat 5, 7, 8 and 9 (Collection 2 surface reflectance, cloud and shadow masked, Landsat 5 and 7 harmonised to Landsat 8), 2010-2026: peak NDVI from 15 June to 31 July for 64,119 wheat fields drawn from the 30 m wheat map 2024 within cadastral parcels. Oblast index: the greenest 60% of each oblast's wheat-field area. Yield from the 2010-2024 link between that index and official oblast yields, each year estimated with its own year left out; 2025 and 2026 are out of sample.Validation
- Landsat greenness against official belt yields, 2010-2024, each year estimated without its own official figure: average miss 1.1 c/ha, r = 0.78. The 2025 gap, 3.1 c/ha, is the second largest in 16 years, after 2011.
- Crop model against official belt yields: average miss 0.8 c/ha in 2018-2024, the years used to set it, and 1.9 c/ha in 2012-2017, which it never saw; it overstates the 2012 and 2013 droughts. For 2025 it gives 14.1 c/ha, in line with Landsat (13.7) and MODIS (14.0).
- Crop model against official district yields, 45 to 58 districts a year, 2016-2020: r = 0.75 once year-to-year swings are removed (0.63 to 0.86 by year).
- Satellite canopy, 2025: official yield above the canopy estimate in all four oblasts, by 1.8 to 3.0 c/ha. In 2024, the greenest season since 2010, the canopy estimate sat above the official yield, so the method does not under-read good harvests.
- Satellite canopy by district, 45 to 58 districts a year, 2016-2020: greenness ranks districts as official district yields do (r = 0.67; 0.79 for peak greenness), on a par with the crop model. Calibrated on those districts alone, it matches the official 2024 belt yield within 0.1 c/ha and puts 2025 at 13.5 c/ha, 3.3 c/ha below the official figure. 42 of 58 districts were less green in 2025 than in 2024.
Records
- Bureau of National Statistics: yields, harvested area, monthly grain stocks
- USDA PSD; FAS Astana reports KZ2024-0006, KZ2025-0004, KZ2026-0003, KZ2026-0007, KZ2026-0008
- UN Comtrade: wheat, flour, bran, feed preparations
- Grain Union of Kazakhstan, market reviews 2013-2023 and 2026 statements
- Ministry of Agriculture; AFM; APK News, ElDala, Inbusiness.kz, Kursiv, Zakon.kz, Business FM, 2024-2026