EDSEnvironmental Decision System

Field-built environmental intelligence

Know what is happening here.
Decide what to do next.

EDS observes changing local conditions, estimates near-term environmental risk and supports reviewable decisions for weather-sensitive operations.

See how EDS works Explore the Eclipse field study
EDS operating model Observation to decision

Local evidence becomes a reviewable decision.

  1. 01 · Observe
    Local conditions

    Temperature · humidity · pressure · light

  2. 02 · Preserve
    History and health

    Timestamped context

  3. 03 · Interpret
    Near-term assessment

    Now · 1 h · 3 h

  4. 04 · Decide
    Review

    Reasons visible

Illustrative operating model. No live data.
Persistent local telemetry Reviewable history + health 1 h and 3 h advisory windows Observable decision state

The decision gap

A regional forecast is context.
The instrument site is the decision.

For astronomical observing, the useful window can change faster than a broad forecast suggests. Humidity can tighten toward the dew point. Local light and pressure can shift. A false positive wastes setup time and energy; a false negative can waste a rare clear interval or leave equipment exposed.

EDS began with a practical need: decide whether to begin, continue or stop an observing session. It also needed to provide earlier warning of conditions that could threaten optics and remote equipment.

01

Set up?

Does the local evidence justify the effort and energy?

02

Continue?

Is the trajectory stable enough to keep observing?

03

Protect?

Is dew or rapidly changing weather becoming a risk?

04

Stop?

Should the operator fail closed rather than assume safety?

Measure locally. Preserve the history. Make the uncertainty visible.

One continuous system

From observation to operating guidance.

A weather station reports conditions; a dashboard displays them. EDS preserves the local history, interprets near-term risk and keeps the resulting decision state reviewable by an operator.

  1. 01

    Observe

    Local temperature, humidity, pressure, ambient light, battery state and signal health.

    Maintained field station
  2. 02

    Preserve

    Timestamped telemetry, latest state, history, aggregates, gaps and staleness.

    Reviewable local record
  3. 03

    Interpret

    Current-condition rules and 1-hour / 3-hour advisory risk windows.

    Confidence stays visible
  4. 04

    Decide

    Operator views, policy state, reasons, events and proposed actions.

    Advisor and dry-run today

The operating loop

Evidence stays attached to the decision.

Each stage preserves enough context to show what changed, what the system inferred, and why a conservative hold may be safer than a confident-looking answer.

01

Every reading has context.

Time, station health and data quality help separate a real environmental change from missing or stale input.

02

History changes the meaning.

Recent trajectories sit beside the latest observation; the Control Center also exposes historical windows and event replay.

03

Rules remain visible.

Thresholds, confidence guardrails and safety states are inspectable rather than compressed into an unexplained score.

04

Uncertainty can block action.

Missing, stale or low-confidence inputs can hold an opening decision instead of silently presenting conditions as safe.

Verified scope

Current capabilities and current limits.

Operational Validated Experimental Expanding
Validated

Persistent local observation

Minute-scale MQTT telemetry, retained station status, bounded offline buffering, historical records and station-health checks.

Validated

History and replay

Latest observations, raw history, aggregates and client-side replay views keep trajectories available for review.

Validated

Observable decision support

Policy, state, reasons, overrides, events and proposed actions are available in advisor or dry-run workflows.

Validated · partial

ASCOM Alpaca conditions

Supported local measurements are available through a partial ObservingConditions surface; wind, rain-rate and cloud-cover properties are not implemented.

Advisory forecast rail

Four domains. Two decision windows.

The 3-hour value represents the highest or worst assessed class within the next three hours. It does not predict an exact event time. Outputs are advisory and trained against rule-derived labels.

Astronomy1 h3 hObserving quality
Rain1 h3 hLikelihood
Dew / fog1 h3 hRisk
Storm1 h3 hEvidence-limited risk
Current technical scope
Experimental

Candidate-model evaluation

Candidate models are evaluated offline and must pass controlled release gates before they can influence operational guidance.

Expanding

Connected-device execution

The safety state machine can propose actions. Direct hardware execution and a verified N.I.N.A. workflow remain future work.

Expanding

More stations, more context

The software can represent multiple stations. A validated distributed physical network and external weather-feed ingestion have not yet been demonstrated.

Where local truth matters

Designed for decisions that happen at a place.

The current evidence base is strongest for one coastal station and weather-sensitive astronomical operations. Other uses require their own sensors, thresholds and validation.

  1. 01

    Astronomical observatories

    Session go/no-go guidance, dew awareness, equipment protection and remote-site context.

  2. 02

    Field research

    Persistent local observations and a reviewable record around an event or experiment.

  3. 03

    Remote installations

    Environmental and station-health context for systems that cannot be watched continuously.

  4. 04

    Independent R&D

    An end-to-end testbed for sensing, data engineering, model governance and inspectable decisions.

Evidence from the field

Two public records. Two different tests of EDS.

One record examines a precisely timed astronomical event. The other examines continuity and repeated local pressure transitions across an extended winter sequence.

01Scientific Field Study

Solar Eclipse 2026

Can EDS preserve and explain a precisely timed environmental event?

Publication cover showing the approved Eclipse-day light trace, the 12-evening control, and the 11 lux sample eight seconds after Espinho predicted maximum
Open the complete scientific figure. This cover is derived from the approved Eclipse light record; complete method remains in the archive.
11 luxRecorded minimum
111 / 111Event-window observations
12Comparison evenings

Timing note The 11 lux reading landed 8 seconds after Espinho predicted maximum.

Explore the Eclipse Archive Launch Replay

02Observational Field Note

When the Pressure Fell

Can EDS preserve repeated local atmospheric transitions across an extended winter sequence?

Publication cover comparing five local peak six-hour pressure falls on one shared zero to thirteen hPa scale; three cross the frozen 10 hPa event screen, Ingrid remains below it, March has no major local signature, and Nils and Oriana are strongest
  1. Ingrid. Below threshold. Peak six-hour pressure fall: -8.855 hPa.
  2. Kristin. Supported. Peak six-hour pressure fall: -12.288 hPa.
  3. Leonardo and Marta. Supported. Peak six-hour pressure fall: -10.568 hPa.
  4. Nils and Oriana. Strongest supported. Peak six-hour pressure fall: -12.911 hPa.
  5. March adverse-weather period. No major local pressure signature. Peak six-hour pressure fall: -2.866 hPa.

The dashed 10 hPa line is the frozen event-screening threshold used by this field note. It is not a national warning, storm-classification, or forecast threshold.

Open the complete scientific figure. Comparison windows and controls remain in the field note.
99.256%Minute-bin coverage
3Supported official sequences
-20.223 hPaLargest local 12 h fall

Acquisition 60-second median interval across the analysed observation record.

Read the Field Note Inspect Data and Method

Field hardware + operations

Built to keep observing when no one is watching the station.

The system joins physical design, embedded firmware, telemetry transport, storage, advisory models, operator tooling and operational policy. Each layer matters only when the chain preserves the meaning and quality of the observation.

Open EDS environmental-station electronics and the installed weather-facing enclosure
Documentary

Station and weather enclosure

Post-event documentary views of the maintained environmental observer, including its electronics enclosure and weather-facing housing. They provide hardware context, not eclipse-day evidence.

Two compact solar panels and the separate custom field power unit used in the documented EDS installation
Documentary

Solar-assisted field power

Two compact solar panels feed a separate custom power-management and distribution unit used by the field installation. The green enclosure belongs to the power subsystem, not to the environmental station.

Sanitized representative Control Center QA capture in a normal refresh state, showing four forecast domains at one-hour and three-hour horizons with current-condition and trend panels; values are fixtures, not live telemetry Representative QA capture · fixture values · not live
Representative

Control Center and advisory output

A sanitized QA capture shows the current operator surface joining 1-hour and 3-hour guidance, current conditions and trends. Its values are representative fixtures, not live station data.

Independent R&D by MDNT

Environmental intelligence should end in a decision someone can inspect.

EDS is engineered by Teófilo Couto as an independent MDNT research and development system spanning station hardware, firmware, data services, model evaluation, operator tooling and field publication.

Contact MDNT Open the Field Archive
Study figure