AgTech & Controlled Environments · Engineering example

Vertical Farm Environmental Intelligence

Understand why one crop zone or cycle performs differently from another. Connect environmental conditions, equipment behavior, grower actions, and harvest results before changing a recipe or buying another system.

The operating challenge

The recipe may be consistent. The crop experience may not be.

Two racks can share a climate schedule and still deliver different harvests. Air moves differently around each tier; light measured at a fixture is not necessarily light received by the canopy; nutrient readings from a common tank do not describe every root zone. Growers see the result, but the explanation is scattered across controllers, handwritten interventions, crop records, and harvest grades.

Averages hide local differences

A room-level temperature or humidity reading can look acceptable while one tier has poor airflow, a warmer canopy, or a different moisture pattern.

Controls and crop records are separate

Lighting, irrigation, HVAC, dosing, and manual interventions have timestamps, but the crop batch and harvest outcome may live in different systems or spreadsheets.

Every adjustment becomes a guess

If cultivar, growth stage, sensor placement, and equipment state are missing, a lower yield can prompt a recipe change that may not address the real cause.

Why modernize: Make each growing zone and crop cycle comparable enough for a grower to investigate variability, preserve what was changed, and learn whether the change helped. The aim is a better decision record, not automated agronomy by default.

Illustrative operating example

One lettuce tier harvests less than its neighbor.

The zones, crop, and events below are illustrative. They are not a claim about a completed customer installation, a proven cause of crop loss, or a guaranteed yield improvement.

Imagine two vertical-farm tiers growing the same lettuce cultivar under the same approved recipe. The second tier produces less saleable product in several harvests. Its controller schedule appears normal. A grower suspects that light is lower at the actual canopy, but a recently cleaned fan and a changed nutrient probe complicate the story. A single average room reading cannot distinguish among airflow, light delivery, nutrient availability, plant handling, disease, and measurement error.

Define comparable batches.Link every tray to cultivar, seeding and transplant dates, tier, harvest date, and recipe version. Compare the same growth stage and a relevant crop outcome, not just calendar weeks.
Check the measurement.Verify sensor location, calibration and replacement history, timestamps, and missing periods. Compare a canopy-level light reading with the scheduled light output; record any uncertainty before calculating daily light exposure.
Reconstruct the crop experience.Align air temperature, relative humidity, CO₂, light at canopy, irrigation events, nutrient pH and electrical conductivity, fan and HVAC state, and grower observations with each batch and tier.
Investigate a difference.A reviewable comparison shows a recurring lower measured light exposure on the affected tier, but it does not declare light the sole cause. The grower checks fixture output, shading, airflow, plant health, and manual handling before approving any change.
Verify the outcome.Record the inspection, action, and later comparable harvests. If the difference was due to a faulty sensor, that finding stays in the history and prevents a false recipe conclusion.

What the grower would see

Tier B: crop-cycle review requested

Leafy-greens batch group. No setpoint or equipment command has been issued by the analytical layer.

Comparison: Same cultivar, recipe version, and growth stage; saleable harvest is lower across comparable cycles.
Evidence: Canopy light history, air and nutrient conditions, controller events, manual work, and harvest grade are shown with their data-quality status.
Uncertainty: Probe replacement and missing readings are visible. Lower light may contribute, but cause remains unconfirmed until inspected.
Decision: The responsible grower chooses the next inspection or controlled trial; any recipe or equipment change follows existing approval.

Other high-priority situations

The same record can answer different grower questions.

Was the nutrient problem real?

An electrical-conductivity excursion in a recirculating solution could reflect dosing, mixing, water addition, sampling location, or a probe fault. A bounded rule can ask for a fresh measurement and show recent dosing and pH history. An assistant may summarize the sequence with cited records, but it cannot determine nutrient deficiency from EC alone or authorize a dose.

Useful report: batch and reservoir identity, readings and calibration, dosing events, grower finding, and verified follow-up.

Did more lighting create useful output?

A scheduled increase in light changes electricity use and may affect HVAC and dehumidification demand. Measure actual delivered light where the crop grows, compare equivalent batches, and retain energy metering and marketable harvest outcome. A model can help identify a tradeoff only after enough repeatable, well-labeled cycles exist.

Useful report: crop-level light exposure, lighting and climate energy, saleable harvest, and the conditions under which the comparison was made.

What a modernization would connect

Make the crop batch the thread through the system.

The useful unit of analysis is not merely a sensor reading. It is a crop batch in a specific zone and growth stage, with its actual equipment history, human interventions, and harvest result. A first version can read from existing systems while the farm's qualified local controls continue to operate as designed.

Required operating evidence

Batch, cultivar, tray, tier, zone, recipe version, and growth stage
Air temperature, relative humidity, CO₂, and crop-level light when installed
Root-zone or reservoir pH, electrical conductivity, water temperature, flow, and dissolved oxygen where relevant and available
Lighting, HVAC, fan, irrigation, and dosing events, plus manual interventions
Harvest weight, saleable grade, rejects, and recorded reasons

Sensor locations and crop-specific target ranges must be defined locally. Missing measurements remain unknown.

Connected technical layers

Existing climate, lighting, irrigation, and nutrient controllers retain local authority
Read-only acquisition with edge buffering, timestamp and unit checks, and sensor-health flags
A crop-cycle record joins zone, equipment, batch, intervention, and outcome
A comparison view lets growers inspect evidence and log decisions

Any later write-back or automated action requires separate design, limits, testing, and owner approval.

Crop batch and zoneValidated sensor and equipment eventsStage-aligned recordComparable-cycle reviewGrower finding and outcome

Room averages, controller schedules, and crop-level exposure are different things. The system should preserve the source and location of each value rather than blending them into a false single measurement.

Where automation and AI fit

Automate evidence handling first; keep growing decisions accountable.

Deterministic automation

Check sensor freshness and range, detect flatlines, label maintenance and calibration windows, align events to a batch and zone, and flag sustained departures from a grower-approved condition. These rules can make a review queue useful without predicting yield.

Analytical models

After enough comparable cycles and trustworthy outcomes exist, statistical or machine-learning analysis may help distinguish recurring zone patterns and test a bounded harvest forecast. It must be evaluated on later unseen cycles and compared with simpler baselines. Correlation is not proof of cause.

Evidence-linked AI assistance

An approved assistant can retrieve the relevant recipe, intervention, sensor exception, and harvest records; draft a concise question for the grower; and cite the underlying entries. It should say when records conflict or are missing. It does not invent a crop diagnosis or recommend an unreviewed setpoint.

Human and control authority

Growers and qualified technical staff decide crop treatment, maintenance, recipe changes, and trial design. Existing local controllers and environmental limits are separate from the reporting and analytical layer.

How we examine a proposed solution

AI advisory roles challenge the design; growers decide what is valid.

These are Adaptiv Stratum AI advisory roles for scoped planning and analysis. They are not human employees, agronomists engaged on a project, or a claim that named agents operate a customer's farm. Any deployed system has its own reviewed scope and permissions.

Demeter

Frames crop-stage, zone, irrigation, and controlled-environment questions for grower review.

Argus

Challenges sensor placement, calibration, time alignment, and edge data quality.

Helios

Examines energy and resource measures without mistaking lower use for better crop outcomes.

Apollo

Defines comparable-batch metrics, harvest denominators, and reporting lineage.

Hephaestus

Reviews separation between existing controls and proposed monitoring or automation.

Clio and Themis

Challenge source quality, trial design, authority boundaries, and unsupported claims.

What becomes visible

A useful dashboard connects conditions to harvest, not just alarms.

A crop-level view helps a grower investigate today. A program view helps management decide whether the monitoring effort is worthwhile. Both must show the denominator, missing data, recipe changes, and which comparisons are truly comparable.

Per-batch and zone record

Crop identity, zone and tier, stage, and recipe version
Actual environmental history with sensor location, units, and quality flags
Controller events, irrigation and nutrient changes, and human observations
Alert rule or model version, evidence cited, reviewer, and finding
Harvest weight, saleable grade, rejects, and follow-up after intervention

Program-level measures

Valid-data coverage and unresolved sensor exceptions by zone
Time in locally agreed crop-stage bands, where measurements support it
Zone-to-zone and cycle-to-cycle variation for comparable batches
Saleable harvest per area or batch, with cultivar and stage context
Water and metered energy per saleable output, with comparable boundaries
Alert findings, false alarms, response time, and completed corrective trials

A higher yield after a recipe change does not by itself prove the change caused it. A missing sensor cannot be counted as time in target range, and a controller's light schedule is not a canopy-level light measurement.

Operating safeguards

The crop, equipment, and data all need a failure plan.

The exact safeguards depend on crop, facility, water system, local controls, and agreed project scope. The analytical layer should fail visibly while existing approved climate and irrigation control keeps its defined local behavior.

Crop, cultivar, zone, tier, and recipe identity before comparisons
Calibration, location, replacement, and maintenance history for sensors
Freshness, range, flatline, and cross-signal checks; explicit unknown states
Airflow and microclimate checks before relying on room averages
Review of nutrient measurements and dosing before recipe changes
Human approval for material environmental or crop-treatment changes
Clear continuity when network or analytics is unavailable
Recorded trial design and post-change outcome, including negative results

A practical first phase

Start with one crop, two comparable zones, and one decision.

Choose the question.Identify a crop and recurring zone difference where better evidence would change a grower inspection or controlled trial.
Map the record.Confirm batch IDs, zone and tier mapping, controller data, nutrient measurements, grower notes, harvest grade, and who owns each source.
Validate the measurements.Check sensor placement and condition, timestamps, units, missing periods, and the relevant crop stage. Keep the existing controllers unchanged.
Review in observation mode.Compare like batches and expose uncertainty. Let growers inspect candidate differences, record findings, and test any change through their existing process.
Earn the next phase.Decide whether the record improved investigation and whether enough reliable cycles exist for forecasting, deeper automation, or extension to more zones.

Begin with the growing decision

Which crop outcome is hardest to explain today?

A first discussion can start with one crop, two zones, the data you already collect, and the harvest or quality difference you would like to understand.

Discuss a vertical-farm question →

Evidence and further reading

Ground the design in real growing-system evidence.

These primary sources inform the examples. The proposed workflow remains illustrative and is not a case study of an Adaptiv Stratum customer.

University of Arizona: UAg Vertical Farm describes multi-tier lighting, aerial and nutrient measurements, and energy and water monitoring in a research facility.
University of Arizona: Environmental Monitoring and Airflow covers sensing and the importance of climate uniformity.
University of Arizona: Advanced Sensing and Climate Control discusses plant-response sensing and environmental control research.
Cornell Cooperative Extension: Resource Optimization in CEA discusses nutrient-management and lighting-energy tradeoffs.