Irrigation monitoring: a 7-day sensor and approval plan

Overview

# Irrigation monitoring: a 7-day sensor and approval plan

## Goal and required inputs

Produce an auditable sensor-to-field registry, a quality-aware monitoring view and a human decision log. Bring your real device's installation/calibration instructions, field identity, measurement depth, sampling interval, clock/time zone and an authorized operator. Define the farm's locally approved decision procedure separately. VWC (%) and water potential (kPa) are different quantities; this plan supplies no conversion, threshold, water volume or runtime.

## Procedure

1. Choose Adapt to my program to create a private farm program; edit days and approved inputs, preview selected fields/start date before creating tasks, and inspect the five tasks: mapping day 0, inspection day 1, baseline day 2, decision day 4, review day 7.
2. Match each real probe to a field, depth and documented unit. Confirm the displayed time and sampling interval against the device.
3. Inspect installation, power and connection using the manufacturer's procedure. Mark missing, flat or implausible readings for investigation.
4. Show quality flags with the signal. Break a chart line at missing VWC; keep suspect readings identifiable rather than deleting them.
5. Record an authorized person's decision and rationale. A review task is not an irrigation order. Use an existing local operating procedure only if an actual action is approved.

## Deliverables and acceptance criteria

Create private `probe-registry.csv`: `sensor_id,field_id,model,depth,depth_unit,quantity,measurement_unit,timezone,sampling_interval,installation_check`. Create `decision-log.csv`: `decision_id,field_id,decision_time_with_timezone,evidence_window,data_quality,decision,approver_alias,rationale,actual_action_reference`.
Accept the monitoring view when every signal resolves to one registry entry, units/time zones are explicit, missing values remain empty/null and suspect flags survive export. Accept a decision record only with an approver and reason; 'defer pending inspection' is a valid recorded outcome. Keep sensor faults separate from agronomic judgments.

## Worked example and next action

Download [synthetic sensors](/hub/agridigi/synthetic-sensor-timeseries): two fictional fields, 72 hourly records each, 138 ok/4 missing/2 suspect overall. `sensor-a-017` has missing VWC; `sensor-a-035` has the intentionally suspect value 99. Check that both states survive a CSV/JSON round trip and chart export. These values cannot justify a real watering decision. Close the cycle with a fault list and a separately approved next monitoring visit, not a savings claim.

## Privacy, provenance and license

Keep real device IDs, locations, credentials and named staff out of public examples. Use the synthetic fixture for parser demonstrations. AgriDigi authored this operational template; manufacturer and local instructions remain controlling inputs. Original text/tasks are [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/): credit AgriDigi, link the original and license, and describe adaptations. No automated-control safety, water savings, yield or chemical recommendation is asserted.

Action plan

Crop: Crop independent · Duration: 7 days

1. Create the device–field–unit registry

   Day 0

   Record each real device's ID/model, field, depth, quantity/unit, time zone and sampling interval from its documentation. Deliver probe-registry.csv plus the manufacturer's instruction reference. Accept only unambiguous mappings; VWC and kPa are not interchangeable. Keep synthetic devices separate and do not expose credentials or real locations publicly.

2. Inspect installation and isolate data-quality faults

   Day 1

   Use the manufacturer's procedure to check placement, power, connection and calibration status. Compare device and displayed times; flag missing, flat or unexpected readings. Deliver a fault list with owner and next check. Accept when missing readings remain empty/null and faults are explicit; a missing value is not zero moisture.

3. Build a quality-aware monitoring baseline

   Day 2

   Display signal time and quality together; break missing-value chart segments and retain suspect markers through export. Deliver a saved view and a CSV/JSON quality check. Accept when units/time zone are visible and source flags survive. Locally authorized experts determine any real threshold; this example supplies none and must not enable automatic control.

4. Record an authorized human decision and its evidence

   Day 4

   Have the authorized person consider current field observations, reliable readings and the farm's approved method. Record decision time, quality, approver and rationale in decision-log.csv, including any deferral. Accept only a documented decision; this task does not order irrigation or prescribe amount/runtime. Approved actions follow the real local procedure.

5. Review outages, faults and decisions separately

   Day 7

   Summarize interruptions and unresolved faults separately from human decisions. Confirm all four missing and two suspect synthetic records remain identifiable in test exports. Deliver an exception summary and approved next monitoring task. Accept when each unresolved fault has an owner and no synthetic signal is reported as real performance or savings.