This is the engineering summary of an open reference architecture. The paper, its object model as JSON and the model register are free to reuse under CC BY 4.0: https://codeatoms.ai/restricted-crop-monitoring-saudi-arabia/. The operator is described by class, never by name.
An agriculture and earth observation operator in Saudi Arabia enforces a ban on green fodder and limits cultivation to the licensed area, because groundwater is the scarce input. Checking every parcel by visit is impossible. Checking by imagery alone produces flags nobody can act on.
Here is how an 18 month service turns imagery into case files that hold up.
1. Join eight sources into fourteen objects
The national agricultural register, crop and well licences, service records and a mirrored Sentinel-2, Sentinel-1 and Landsat archive enter through one adapter family. Farm holding, field, licence, tasking order, image capture, detection flag, case file and field inspector are among the fourteen objects. A detection means something only once it is joined to the licence of the holding it sits on.
2. Two small open models
RF-DETR, fine tuned per region, finds the crop. Chronos-2, zero shot, forecasts a vegetation index baseline for each parcel, so a field that departs from its baseline stands out. Both are Apache 2.0 and together hold about 0.55 GB of weights.
3. One card, inside the Kingdom
Both run on one 48 GB L40S class inference node in the operator's own facility. Owning it costs about 15,200 dollars over three years, about two thirds of the deepest three year AWS commitment. Any rented option today sits outside the Kingdom and would move the imagery and the register across the border. The card class needs a US export licence, which the paper flags.
4. A person opens every case
Only the flagged minority reaches a named field inspector, who verifies on the ground before a case file opens. The model narrows the search. The inspector makes the finding.
Full design, figures, the cost appendix with every price cited, and the object model: the paper. DOI: https://doi.org/10.5281/zenodo.23186675
Designed on CodeNinja Praxis, the CodeNinja Atoms platform for designing physical AI systems. The object model imports into Hyper Ontology, which turns it into a living system. Load it yourself with the open hyper-ontology loader.
