Sector · Agriculture and earth observation
Physical AI for agriculture and earth observation
Farms, irrigation and land use are watched by sensors, satellites and aircraft, but the readings, the licences and the field records usually sit in separate systems. These are the questions engineers ask first, answered from 4 published CodeNinja Atoms reference architectures.
- What does a physical AI system for agriculture and earth observation look like?
- Which AI models can an agriculture and earth observation operator run on its own hardware?
- How much compute and hardware does AI in agriculture and earth observation need?
- Is it cheaper to own AI hardware or rent cloud GPUs in agriculture and earth observation?
- What ontology or object model does an agriculture and earth observation AI system need?
- Who approves the decisions an AI system makes in agriculture and earth observation?
What does a physical AI system for agriculture and earth observation look like?
A complete physical AI design for agriculture and earth observation names what to sense, which existing systems to join, the object model that joins them, the models and hardware, the three-year cost and the person who approves every action. CodeNinja Atoms has published 4 such reference architectures for agriculture and earth observation, each free to reuse under CC BY 4.0.
| Design | Country | What it does |
|---|---|---|
| Field Ledger | Pakistan | one dashboard that joins a farm's live sensor feeds, its historical datasets and a big data and analytics repository into one object model, so any farm, crop cycle or season can be drilled into, exported and reported on under role-based access. |
| Baseline | United States | one summer flight of 3-inch 4-band orthoimagery and USGS Quality Level 1 LiDAR, with every tile, point cloud and elevation model bound into a site ontology, so the next season starts from a comparison instead of rediscovery. |
| Fodder Watch | Saudi Arabia | an 18-month earth observation service that screens every parcel for restricted green fodder and cultivation beyond the licensed area, checks each detection against the holding's licence, and turns the confirmed ones into case files an inspector can act on. |
| A Self-Improving AI System for Canal Irrigation Water Allocation in Pakistan | Pakistan | An open reference architecture for a self-improving AI system in physical AI, built for a provincial irrigation department in Pakistan that takes its share of river water at barrages and head regulators and delivers it through a main canal, distributaries and minors to outlets farmers share by rotation, where owning the inference hardware that runs it costs about 636,000 US dollars over three years. When a head regulator reading drops below its indent for three days at peak demand, the design joins the telemetry, the indent and release ledgers, the rotation schedule, the gauge reader's diary, the complaint register and the revenue crop records into one entitlement versus delivery argument, so the divisional engineer sees which reach is silted, which branch is over drawing and which gate adjustment cured the matching shortfall last season. Every model runs on the department's own hardware inside Pakistan under the inter provincial water accord, canal and drainage law, the national data rules for farmer records and export control on the GPU node. |
Which AI models can an agriculture and earth observation operator run on its own hardware?
Each published agriculture and earth observation design names its models and why, and every model is open-weight or no model is used at all, so the operator can run it on hardware it owns.
| Design | Models |
|---|---|
| Field Ledger | None: the analytics are deterministic aggregations, drill-down and reporting |
| Baseline | None: vegetation analysis stays with the operator's own analysts |
| Fodder Watch | RF-DETR fine-tuned per region and Chronos-2 zero-shot, both Apache-2.0, about 0.55 GB of weights together |
| A Self-Improving AI System for Canal Irrigation Water Allocation in Pakistan | Frontier class open weights for the three drafting agents (shortfall, indent and dispute) and the command centre, self hosted for sovereignty; the operator names GLM 5.3 and the plan holds its FP8 arithmetic, but no weights are committed until the register verifies a model live; embedding, OCR and forecasting model classes for retrieval, handwritten Urdu and English ledgers and shortfall detection; no third party model service, nothing trained into weights |
| Design | Choice | What was picked | Why |
|---|---|---|---|
| Field Ledger | Model | Zero models committed; a forecaster over sensor time series stays open work gated on Milestone 1 evidence | The requirement commits aggregations, drill-down and reporting, and no model was verified for this run |
| Field Ledger | Hardware class | No compute procured; all compute runs on operator-provided servers, storage and connectivity | The operator provides the hosting infrastructure and the sensors already generate the data, so no field compute or inference server is bought |
| Field Ledger | Sizing rule | Database and buffer sizing set against the recorded server specification at the Milestone 1 kick-off, re-estimated on measured sensor volumes before full-farm integration | Sensor formats, volumes and refresh rates are settled by the prototype, not assumed at bid time |
| Field Ledger | Sensing | The operator's existing sensor devices, read-only; no cameras and no new field hardware | The world is sensed by farm rather than watched, and the engagement installs no equipment |
| Field Ledger | Pattern | System of Context, with the ontology as a versioned schema and mapping projection inside the database | Farm, field, crop cycle, sensor device and dataset objects answer the cross-system questions no single source can |
| Field Ledger | Ground | Open-source PostgreSQL with PostGIS, on operator premises in Pakistan, under permissive and copyleft open-source licenses | The open-source database mandate and the source code and intellectual property transfer make the operator the owner of every artefact |
| Baseline | Learned model | None; the model register is empty by decision | No agent or model layer was wanted; vegetation analysis stays with the operator's own staff in its own tools |
| Baseline | License position | No model license exists to hold or trigger | With zero weights, ownership runs through the professional services agreement, not through a license |
| Baseline | Positioning class | GNSS guidance and correction service, with network RTK or an owned base station | Quality Level 1 accuracy is decided by this chain, so it is specified by baselines, correction source and test evidence |
| Baseline | Acquisition platform | Manned aircraft with a 4-band red, green, blue and near-infrared large-format camera | Unmanned data will not be considered, so the acquisition class is fixed by the requirement |
| Baseline | LiDAR sensing | Quality Level 1 sensor at a minimum 8 pulses per square meter | The 16 pulses per square meter option is priced separately and exercisable at the operator's option |
| Baseline | Pattern the design stands on | System of Context, with the ontology as a projection over the systems of record | Deliverables carry flight mission, sensor and processing provenance, so analysis joins to units, missions and seasons rather than to tiles |
| Baseline | Ground it runs on | The operator's ArcGIS Enterprise environment in the United States | Files are publish-ready, and publication stays with the operator's own GIS staff inside its own boundary |
| Fodder Watch | Detector | RF-DETR, Apache-2.0, fine-tuned per region, BF16, about 61 to 68 MB at 16 bit | No field-of-use restriction, so the operator owns the fine-tuned weights outright |
| Fodder Watch | Forecaster | Chronos-2, Apache-2.0, zero-shot, about 120M parameters, FP32, about 0.48 GB | Thousands of independent per-parcel index series with no fine-tuning dependency |
| Fodder Watch | Checkpoint exclusions | RF-DETR XL and 2XL under Roboflow's Platform Model License, excluded | Field-of-use terms would compromise in-Kingdom weight ownership |
| Fodder Watch | Inference node | 48 GB PCIe GPU class (L40S class), air-cooled, in the sovereign facility | Footprint rule: combined weights under 0.55 GB leave batch and cache headroom |
| Fodder Watch | Positioning | RTKLIB class GNSS with optional site base station, field tablets at 3 to 5 m grade | Arrival coordinate checked against the flagged polygon, so phantom visits file as not-reached |
| Fodder Watch | Optical sensing | Sentinel-2 multispectral, 10 m, 5-day revisit | Resolves center pivots for crop class without a foreign processing dependency |
| Fodder Watch | Radar sensing | Sentinel-1 C-band SAR | Cloud-proof complement; dust and haze, not cloud, are the optical channel's real attacker |
| Fodder Watch | Archive sensing | Landsat 8/9, 8-day combined, archive reaching back decades | History for per-parcel baseline series |
| Fodder Watch | Cloud evidence | Sentinel-2 scene classification and cloud probability layer | Treated as evidence, not truth, in the screening rule |
| Fodder Watch | Aerial sensing | national aerial survey products | Rounds over the sedimentary shelf, matching the operator's existing method |
| Fodder Watch | Field truth | GNSS field tablets on inspector visits | Confirmed and refuted visits become the labelled corpus for each season's retraining |
| Fodder Watch | Pattern: triage | Per-parcel signature time series with traffic-light triage | Only the flagged minority reaches a human inspector |
| A Self-Improving AI System for Canal Irrigation Water Allocation in Pakistan | Reasoning model (frontier work surface) | Frontier class open weights, self hosted for sovereignty, Kimi class; the operator names GLM 5.3 and the plan holds its arithmetic (published FP8 weights of about 756 GB on one 141 GB class node of eight GPUs with about 280 GB of KV cache headroom), but no weights are committed until the register verifies a model live | The operator holds the weights and no third party model service ever touches canal data; no open weight model was verified live on this run, so the row stands as a class and the GPU count follows only when a verified model stands in the register |
| A Self-Improving AI System for Canal Irrigation Water Allocation in Pakistan | Embedding model class | Embedding models for retrieval, by role rather than name, on the 48 GB PCIe class site server | Project ledger text and telemetry annotations into the retrieval space the three drafting agents search; licence terms attach to the register row when a named model at a named version stands |
| A Self-Improving AI System for Canal Irrigation Water Allocation in Pakistan | OCR model class | OCR models for the handwritten Urdu and English paper ledgers, on the 48 GB PCIe class site server | Every batch passes a human verification stage with a 2 percent failure gate before anything enters the object model, because a wrong entitlement baseline would corrupt every downstream comparison; OCR corrections are a verification stage, not a training set |
| A Self-Improving AI System for Canal Irrigation Water Allocation in Pakistan | Forecast model class | Forecasting models for shortfall detection, on the 48 GB PCIe class site server | Shortfall detection must land inside one telemetry cadence so a developing head shortfall is visible on the first of the three days it takes to breach; modest, continuous work that must survive independently of whatever the frontier layer becomes |
| A Self-Improving AI System for Canal Irrigation Water Allocation in Pakistan | Frontier hardware | 141 GB HBM GPU class, one node of eight GPUs (about 1,128 GB of aggregate memory) | The held arithmetic fits about 756 GB of FP8 weights with about 280 GB of KV cache headroom; the node sits above the US advanced computing export threshold for Pakistan, so the export licence determination runs first in phase one with an in country procurement route or colocation fallback named before any order |
| A Self-Improving AI System for Canal Irrigation Water Allocation in Pakistan | Site server hardware | 48 GB PCIe class site server beside the frontier node, in the circle server room | Ledger digitisation and shortfall detection need modest, always on compute chosen by class; no edge tier, nothing served at a canal head in phase one |
| A Self-Improving AI System for Canal Irrigation Water Allocation in Pakistan | Frontier footprint sizing rule | Frontier model GPU footprint sizing rule, vLLM and SGLang serving runtimes | Open serving runtimes sized against the weights and the KV cache ceiling on hardware the operator owns |
| A Self-Improving AI System for Canal Irrigation Water Allocation in Pakistan | Node count rule | GLM 5.3 node count by GPU class: one 141 GB class node at FP8 | Chosen against GLM 5.3 at BF16 on two nodes or a mid size model; conditional until the register verifies |
| A Self-Improving AI System for Canal Irrigation Water Allocation in Pakistan | Serving runtime | vLLM and SGLang | Serve the FP8 weights within the KV cache ceiling the sizing reserves once a verified model stands |
| A Self-Improving AI System for Canal Irrigation Water Allocation in Pakistan | Registry | Harbor, air gapped artifact and model registry | Model and container updates enter as staged, versioned artifacts, verified before promotion; a failed or suspect update leaves the previous version serving |
| A Self-Improving AI System for Canal Irrigation Water Allocation in Pakistan | Sensing | Canal head flow depth and velocity telemetry, inter provincial point telemetry, route based gauge reader readings, complaint register entries, digitisation verification checks | Live flow at the heads, the manual baseline in the field, and the human check that gates OCR into the ontology; no cameras and no new field hardware |
| A Self-Improving AI System for Canal Irrigation Water Allocation in Pakistan | Pattern: memory | Memory of confirmed causes and accepted adjustments, nothing trained into weights | The operator requires improvement every rotation without training a model on the fly; only what a named engineer confirms at the decision compounds, captured then and not reconstructed |
| A Self-Improving AI System for Canal Irrigation Water Allocation in Pakistan | Pattern: validation | Manual and telemetry baselines taken together in the first month and reconciled in the open | A comparable irrigation automation program found its computed efficiency fell once continuous measurement exposed manual under measurement, so no improvement is claimed before the baselines agree |
| A Self-Improving AI System for Canal Irrigation Water Allocation in Pakistan | Pattern: ontology first | The governed canal network ontology reconciled against one real week of telemetry and ledgers before any agent drafts | Phase one's exit gate is the ontology itself, proven on one main canal for one cropping season before extension to the other canals |
| A Self-Improving AI System for Canal Irrigation Water Allocation in Pakistan | Pattern: boundary | The operator's own hardware in the circle server room, Pakistan; read only adapters; no gate actuation; no third party model service | Telemetry and ledger data never leave the network, open weights on owned hardware, and every physical action keeps its existing owner under canal and drainage law and the department's gate procedures |
How much compute and hardware does AI in agriculture and earth observation need?
The compute follows from the models: the published agriculture and earth observation designs size it as follows, from no new hardware to a full GPU node.
| Design | Part | The design |
|---|---|---|
| Field Ledger | Stack | Open-source PostgreSQL with PostGIS on the operator's own servers; source code and full intellectual property handed over |
| Baseline | Ground | The operator's ArcGIS Enterprise; nothing runs in a cloud the operator does not control |
| Baseline | Acquisition | One manned flight within a week either side of 1 July; 3-inch 4-band (red, green, blue, near infrared) orthoimagery and Quality Level 1 LiDAR at a minimum of 8 pulses per square meter |
| Fodder Watch | Compute | One 48 GB L40S-class inference node inside the Kingdom; the card class needs a US export licence |
| A Self-Improving AI System for Canal Irrigation Water Allocation in Pakistan | Compute | One node of eight 141 GB HBM cards (about 1,128 GB) in the department's circle server room, the held arithmetic for about 756 GB of FP8 weights, beside a 48 GB PCIe class site server for the small model tier; no edge tier, nothing served at a canal head. The node sits above the US advanced computing export threshold for a D:2, D:3 and D:4 destination, so the export licence determination runs first in phase one with an in country procurement route or colocation fallback named before any order |
Is it cheaper to own AI hardware or rent cloud GPUs in agriculture and earth observation?
Each agriculture and earth observation design prices three years of ownership in its Appendix A, with every price cited, against renting the same capacity from a cloud region at its deepest three-year commitment where hardware is bought.
| Design | Line | Three years |
|---|---|---|
| Field Ledger | Three-year cost | No hardware line to price: the cost is the integration and software work (Appendix A) |
| Baseline | Three-year cost | No compute to price; the acquisition is priced by the survey firm against the operator's own task table (Appendix A) |
| Fodder Watch | Three-year cost | About 15,200 dollars to own the node, about two thirds the deepest three-year AWS commitment, and renting would move the data outside the Kingdom (Appendix A) |
| A Self-Improving AI System for Canal Irrigation Water Allocation in Pakistan | Three-year cost | About 636,000 US dollars to own the frontier node and the site server for three years (a range of 606,000 to 662,000: servers, support and power at the national uniform tariff), against about 773,000 to rent the same eight 141 GB cards and one 48 GB card on the deepest three year commitment in AWS Asia Pacific (Mumbai), so ownership costs about four fifths of it, and about 2.05 million on demand, against which ownership costs a little under a third. Renting is a price reference only, because AWS lists no region in Pakistan and every rented option would carry telemetry and ledgers across the border; the frontier footprint itself stands only when the register verifies a model (Appendix A) |
What ontology or object model does an agriculture and earth observation AI system need?
The object model is the part that makes the system an ontology rather than a pipeline: typed objects for the things in the agriculture and earth observation operation, with properties, status values and typed links. Every published design ships its object model as hyper-ontology/1 JSON that loads into Hyper Ontology.
| Design | Size | Objects | Download |
|---|---|---|---|
| Field Ledger | 12 objects, 12 links | Farm Site, Field, Crop Cycle, Sensor Device, Sensor Reading, Historical Dataset, Publication, Dashboard User, Access Role, Alert, Report, Data Source Connection | objects.json |
| Baseline | 16 objects, 14 links | Mapping Site, Land Unit, Watercourse, Dam, Flight Mission, 4-Band Aerial Camera System, QL1 LiDAR Sensor System, GNSS/IMU Georeferencing Chain, Multispectral Orthoimagery Product, Classified LAS Point Cloud, Bare-Earth and Highest-Hit DEMs, QA/QC Accuracy Report, Professional Services Agreement, Monthly Itemized Invoice, Contractor Project Manager, Operator Project Manager | objects.json |
| Fodder Watch | 14 objects, 14 links | Farm holding, Centre pivot / cultivated field, Farm enterprise / large farmer, Crop licence (wheat / seasonal fodder), Water source (well) use licence, Green fodder ban control, Agricultural fuel / electricity service condition record, Imagery tasking order, Satellite image capture, Processed imagery product, Restricted-crop detection flag, Violation case file, Field inspector, Restricted-crop classifier | objects.json |
| A Self-Improving AI System for Canal Irrigation Water Allocation in Pakistan | 16 objects, 13 links | Barrage, Head regulator, Reach, Gate, Outlet, Main canal, Distributary, Rotation, Shortfall, Season, Indent, Release, Complaint, Crop area, Telemetry reading, Gauge reading | objects.json |
Who approves the decisions an AI system makes in agriculture and earth observation?
In every published agriculture and earth observation design a named person makes the decision that changes the physical world; the system prepares it.
| Design | Human control |
|---|---|
| Field Ledger | Every alert is acknowledged by a named user under a role the operator assigns |
| Baseline | The operator's project manager accepts each deliverable against the QA/QC accuracy report |
| Fodder Watch | Every flag is verified on the ground by a named field inspector before a case file opens |
| A Self-Improving AI System for Canal Irrigation Water Allocation in Pakistan | Agents draft and people decide, by name: the controlling engineer confirms or rejects the drafted cause and correction before any gate instruction is drafted, the indent draft is approved before it goes to the national authority, and the dispute account is drafted from cited records for the officer to send; the gate stays a hand operation on telephone instructions, and every confirmed cause and accepted adjustment is kept as memory the next season reads |
Every answer on this page is drawn from the papers linked in it: each paper's At a glance table, model register and cost appendix. Full text for agents: llms-full.txt. Designed on Praxis; object models load into Hyper Ontology.
Want one of these designs for your own operation? Talk to us.