Dispatches
Essays··8 min read

Jiangsu's Wheat Claims Are Now Settled From Orbit

China has mandated satellite crop-insurance claims at household level, with Jiangsu reporting a 10 per cent gap between remote-sensing estimates and payouts — payments arriving 20 days earlier than before. The figure is a population average. For the smallholder whose loss sits close to a policy trigger, that margin decides whether payment arrives at all; a 20 per cent expense cap will let the satellite have the final word.

A crop-insurance claim is the most personal measurement a government ever takes of a farmer, and China has chosen to take it from orbit, one household at a time.

The choice is written into an implementation plan the Ministry of Finance issued in September 2026 with the agriculture ministry, the National Financial Regulatory Administration and the forestry administration, filed as Caijin [2026] No. 88. For the first time, policies are to be written with the individual farm household as the named party, "以单个农户为主体", and cross-checked against production data. The same text asks insurers to put remote sensing, drones and IoT to work across underwriting, survey, loss assessment and claims, and sets a 2030 target for technology spending of 1% of premiums. On 11 November the Jiangsu branch of PICC Property and Casualty reported what that looks like over a wheat season: satellite yield estimates used to calculate claims household by household across 114 townships in 28 county-level districts, payouts starting 20 days earlier than in past years, and a gap of no more than 10% between the remote-sensing estimate and the amount paid.

I think this is the right direction for China's grain belt and the wrong rule for its smallest farms. The difference sits inside that last figure.

Note from a harvest road

I grew up in cantonments around Delhi, and in April the road out of town ran past wheat being cut by hand and threshed beside the field. Some years I watched an argument that seemed to happen in every village along that road: a farmer, the patwari who kept the land records, and a measuring chain stretched across the stubble. No one in those arguments doubted the wheat had been lost. They argued about where the chain should start.

Every crop-insurance system in the world runs on that quarrel. China's new one moves the chain into a satellite.

Note on the Jiangsu file

Read the Jiangsu figure the way a claims adjuster would. A 10% gap between estimate and payout, across a province's wheat, is a very good number, and it should be set against the old routine. In August, Shanghai Securities News described China Property & Casualty settling wind-lodged wheat in Hua County, Henan, with satellite screening, drone evidence and on-site checks, which cut the assessment cycle from 13–18 days to 8–12. The same report carried the regulator's figures for January to July: 59.7 billion yuan in claims on 123.2 billion yuan of premiums, reaching 20.84 million household-instances. For years speed has been the measure that mattered, and remote sensing delivers it.

But an average error is a property of a population. A claim belongs to one family. The Jiangsu post does not say whether 10% is a mean, a median or a worst case, or how it spread across plot sizes. For a household whose loss sits close to the policy's trigger, a 10% miss decides whether a payment arrives at all.

The post also says the method reduced claim disputes. I believe it, for one season. A farmer paid three weeks early is less inclined to argue about the amount, and in an insurer's report a complaint that was never filed looks exactly like one that never had grounds. Fewer disputes in a good harvest year tell you about speed. They tell you about accuracy only after a bad year, when the amount matters more than the date.

Note from the black-soil farms

The pixel works best where fields are big. In Heilongjiang, Beidahuang Information Co., the data arm of the state farm group, showed visiting reporters on 1 September a platform holding digital records for 600,000 growers, 760,000 machines and 48 million mu of farmland, with a digital ID for every plot, machine and grower. Its "Future Farm" model combines satellite imagery, AI and field sensors to write variable-rate prescription maps; on 15,000 mu of pilots at the 855 and Heshan farms the company reports soybean yields up about 5% and fertiliser use down 5%. Those are the operator's own figures from a press visit, and I would want a season of independent trial data before treating them as settled. The architecture matters more than the yield claim. When every plot already carries a digital ID, "one household, one policy" is a join between two tables.

Now carry the plan south to the North China Plain. A mu is one-fifteenth of a hectare, about 667 square metres. The free imagery most published index-insurance research relies on comes from Sentinel-2, whose finest bands resolve 10 metres. That puts fewer than seven pixels on a mu, and on a narrow strip several of them straddle a field path, a ditch or the neighbour's later-sown wheat. Chinese insurers buy finer domestic and commercial imagery, so the arithmetic is kinder for them than this. It is still the right arithmetic to start from. In our own remote-sensing work at Earthscan, the edge pixel is where every model is least sure of itself, and smallholder China is mostly edges.

Ren Jinzheng, who directs the National Agricultural Insurance Research Center at China Agricultural University, put it more plainly to Shanghai Securities News in September. Small, scattered plots make plot-by-plot assessment expensive, he said; village-wide cover remains practical for smallholders, and one household, one policy should not be rolled out uniformly. He wants it paired with village organisations, sampling, published loss results, expert appraisal and a way to object. Ma Biao of Capital University of Economics and Business, in the same article, warned that faster payment does not guarantee accurate payment and asked for the basis of each assessment to be made public.

Note on the evidence against me

The strongest case against my caution has two parts, and both deserve a straight answer.

The first is fraud. Cao Jingzhi, an assistant general manager at Ping An Property & Casualty, listed the system's familiar failures in that same September article: invented insured subjects, inflated acreage, duplicate policies, "agreement" payouts negotiated between insurer and village, and crude loss assessment. His diagnosis is that the rot starts at underwriting, where basic data on land, farmer and crop is often unclear. He is right. Village-level policies are exactly where negotiated payouts hide, and a household policy matched against land-rights and subsidy registers, which the ministry's text orders agencies to share, closes that door. Precision is partly an anti-corruption tool, and I would not give that up.

The second is evidence. A systematic review in Natural Hazards and Earth System Sciences by Thuy Nguyen and co-authors, published in February 2025, covers 89 index-insurance studies from 34 countries and concludes that satellite vegetation-health indices outperform weather indices and reduce basis risk, particularly at higher resolution. That supports Beijing's direction.

Keep reading and the same paper supports my reservation. China contributed the most studies, 12, yet they relied mostly on non-satellite data. Only two studies anywhere compared hedging performance across spatial resolutions, both on wheat, and the authors ask for more. In one study they summarise, the best satellite indices reached average hedging effectiveness of 38%, against 29% for precipitation contracts and 16% for temperature. A real gain. It still leaves most of a farmer's loss unexplained by the index. The literature says satellites beat rain gauges. It does not yet say they can stand alone as the judge of one smallholder's field.

Note for a county loss committee

So here is my position. Household-level remote sensing should be the default for state farms, consolidated holdings and large cooperatives, where plots are big and records clean. For the smallest holdings it should be evidence, weighed by a person, and never the final number. The ministry's Q&A already allows cities and counties to set up expert appraisal panels and loss-determination committees where conditions allow. Before a county moves its smallholders onto satellite-calculated claims, I would make two things compulsory: a published plot-size floor below which every estimate is confirmed by field sampling, and an annual count, by township, of objections filed and claims revised. Jiangsu has shown the estimates can be fast. Only the objection record can show they are fair.

That position costs money, and the plan pulls the other way. It caps the comprehensive expense ratio at 20%, and Wang Zecai of the Chinese Academy of Fiscal Sciences told People's Daily that household policies will raise grassroots workload before technology brings costs down. Sampling and appeals are workload. An expense cap will push insurers to let the satellite decide. Suo Lingyan of Peking University named the incentive to People's Daily Overseas Edition: the insurer pays for the loss, so it naturally wants to control loss costs. An error band resting on that incentive will not drift toward the farmer by itself.

Last note

125 million: the household-instances China's agricultural insurance covered in 2025, per People's Daily Overseas Edition. Each is now meant to carry its own name, and its own margin of error.


Tarry Singh is the founder and CEO of Real AI, an enterprise AI advisory and deployment firm working with global enterprises on production agent systems, model risk, and AI sovereignty strategy. He also leads Earthscan for Energy AI startup, and is a founding contributor to the EU-funded HCAIM and PANORAIMA programmes for responsible AI education across European universities. He writes at tarrysingh.com.

Cartouche
Jiangsu's Wheat Claims Are Now Settled From Orbit · Dispatches, 21 November 2026 · T. Singh