China's National Smart Agriculture Action Plan (2024–2028) — the plan that mandates an open-source agricultural model layer

East-Asia (China); national plan with designated provincial pilot (Zhejiang)

Content

《全国智慧农业行动计划（2024—2028年）》, issued by MARA on 23 October 2024 as 农市发〔2024〕4号 and published through the investment-project approval platform (new.tzxm.gov.cn). It is the operative national instrument under MARA’s earlier 关于大力发展智慧农业的指导意见, and it is addressed to every provincial agriculture department, the XPCC agriculture bureau, and named state-farm groups — 北大荒农垦集团有限公司 (Beidahuang) and 广东省农垦总局 (Guangdong Land Reclamation) appear in the addressee list, which is the corpus’s first hard evidence that state-farm enterprises are formal implementation vehicles rather than examples.

Structure and targets. Three actions, eight key tasks, staged as “一年打基础、三年见成效、五年上台阶”:

• By end-2026: smart-agriculture public-service capability “initially formed”; a set of intelligent solutions for large-area yield gains on staple crops; agricultural production informatisation rate ≥30%.
• By end-2028: public-service capability “substantially improved”; whole-chain digital transformation “basically realised” in the pilot regions; informatisation rate ≥32%; 20 or more base model algorithms and general software tools or SaaS services developed and promoted.

The public-service layer — three products.

1. 国家农业农村大数据平台 (National Agricultural and Rural Big Data Platform) — a unified data resource pool with collection, aggregation/governance, and analysis-decision systems; data classification and grading standards, a data resource catalogue and sharing rules; a data management service hub; and “农事直通” (Nongshi Zhitong), a mobile service window for farmers. The plan states that by 2025 the related data-management rules and standards should be “basically established” and by 2028 the platform “basically completed”.
2. 农业农村用地”一张图” (the “one map” of agricultural and rural land use) — a digital base map with thematic layers for cultivated land, permanent basic farmland, high-standard farmland, contracted land, homestead land, cultivated-land planting-purpose control, and 养殖坑塘用地 (aquaculture pond land), giving “every plot a digital file” and supporting applications such as planting-purpose control, disaster prevention, production trusteeship, and precision machinery operations.
3. 智慧农业基础模型 (smart-agriculture base models) — build an agricultural base model and algorithm open platform and establish an open-source software community for agricultural intelligent model algorithms (农业智能模型算法软件开源社区). The named model families are crop growth, animal behaviour and vital-sign recognition, production-management decisions, and facility-environment multi-factor regulation; the named general tools are smart breeding, feed formulation, and farm-management software/SaaS. The plan then says explicitly: accelerate the application of AI large models (人工智能大模型) in agricultural and rural research, production and operations, and management services. Target: platform and model library “basically established” by end-2026; 20+ algorithms/tools/SaaS by end-2028.

The deployment layer — named equipment per production type. The plan prescribes different technology sets per farm type rather than one generic “smart agriculture”:

• 智慧农场 — environment monitoring and regulation, precision water-fertiliser-pesticide management, intelligent plant protection, unmanned inspection and transport, smart machinery.
• 智慧牧场 — individual animal vital-sign monitoring and management, precise environment control, automatic inspection and disinfection, intelligent disease diagnosis, precision-formula feeding, automatic waste collection and cleaning, harmless waste treatment.
• 智慧渔场 — water environment and quality monitoring, automatic aeration, intelligent inspection, intelligent feeding, individual behaviour observation, intelligent fish-disease diagnosis, grading and counting.

A 农（牧、渔）场智慧赋能计划 then offers cooperatives, family farms and agricultural enterprises general software tools, IT training, online technical services and market information — the smallholder-facing arm.

Whole-chain digitalisation. Includes 互联网+农产品出村进城 (products leaving the village for the city), smart upgrading of national origin markets, cold-chain logistics informatisation, industry-chain big-data analysis centres, and quality-safety smart supervision with a national platform, informationised quality-control systems, promotion of electronic commitment-to-compliance certificates (电子承诺达标合格证) and traceability pilots.

Zhejiang as the designated pilot (先行先试). Zhejiang builds the 智慧农业引领区 with provincial special funds; the plan requires linking 浙农码 to 全农码 (linking the provincial code to the national “All-Farmer Code”) and attaining ministry-province data interconnection before other provinces; the 乡村大脑 (rural brain) is iterated into 种植业、畜牧业、渔业”农业产业大脑” (industry brains for crops, livestock and fisheries), with 浙农-series apps. Targets by end-2028: 1,000+ digital agriculture factories and 100 future farms, plus standards and commercially usable hardware/software products. The plan also supports CAS in continuing to develop the “伏羲农场” (Fuxi Farm) model and mandates digital simulation of production processes (soil nutrient inversion, crop simulation, precision weather analysis) toward optimal planting plans.

What this unit is doing in the taxonomy

The policy-mechanism unit for China: it supersedes china-deepening-scan-rural-revitalization.md as the operative document chain, and it is the first unit in the corpus where a state names an open-source model community as public infrastructure. That makes it the Chinese counterpart to the corpus’s open-source thread (agroecology-open-source-correlation.md, the open-CEA and open-source drone units) — same domain, opposite origin: commons-by-mandate rather than commons-by-community.

Distinct from:

• china-deepening-scan-rural-revitalization.md — the umbrella framing drawn largely from consultancy secondary sources; this unit is the primary instrument, in Chinese, with article-level content.
• usda-fy25-26-ai-strategy.md — the US analogue, and structurally different: USDA’s strategy describes priorities and internal capacity, whereas this plan assigns named platform products, numeric informatisation targets, and a provincial pilot.
• eafrd-cap-strategic-plans-digital-agriculture.md — the EU analogue; CAP strategic plans distribute money to member-state choices, whereas this plan dictates the public-service layer centrally and lets provinces build on it.

Why it matters for talks

• The 30% / 32% informatisation targets are the only hard, dated deployment goals any jurisdiction in this corpus has published for agricultural digitalisation.
• The open-source agricultural model community is the surprising item. It reframes the “Chinese agricultural AI” story away from platform companies toward state-provided model infrastructure, and it is directly relevant to the open-source-agrifood question of who builds the shared layer.
• The animal- and fishery-specific equipment lists show a ministry specifying AI use-cases per production type, including fish-disease diagnosis and intelligent feeding — the corpus’s aquaculture gap is a policy-named cell, not an oversight.
• The state-farm addressees (Beidahuang, Guangdong Land Reclamation) are the deployment vehicles for the machinery and unmanned-farm programmes, and belong in any account of who actually operates Chinese agri-AI.

Critical context

• Verification posture. This is a government primary document: the targets are MARA’s own and no implementation audit has been published. Treat the 2026/2028 percentages as commitments, not achievements. activity-status: deployed refers to the plan being in force, not to the targets being met.
• The plan’s own monitoring language is a weakness signal. It establishes a work task-force (专班), annual task lists, tracking and analysis of key tasks, and an expert advisory committee — machinery that exists because implementation across provinces is the known hard part.
• Informatisation ≠ AI. The 30%/32% rate measures information-technology application in production, not model deployment. Quoting it as an “AI adoption rate” would misstate what MARA measures.
• The 2026 target is now testable. Two years have passed since issue; whether the public-service layer exists as described is the honest question for the next cycle.
• 《行动计划》is not the same as the 2026 No. 1 Central Document (china-no1-central-document-2026-ai.md) or the National Agricultural Outlook 2026–2035 already in the corpus. Different instruments, different bindingness.