Read the ocean's pulse · In-situ plankton intelligence

Continuous plankton intelligence
for harmful-bloom & hypoxia
early warning.

OceanPulse pairs autonomous in-water microscopes with on-device classification and multi-horizon forecasting — so drinking-water utilities, aquaculture operators, and monitoring programs see biological water risk before it becomes an operational event.

Request a demo → See the instruments
Sample alert packet · illustrative 2026-04-22T02:14Z · schema v3 · instr OP-PSC-014
Event
HAB onset · Karenia brevis
Risk
0.91 ± 0.06
Horizon
72 h
Drivers
cell density ↑ · SST ↑ · wind shear ↓
Action
PAC dose review · intake A monitoring
Example of an alert payload an operator receives. Values shown are synthetic and shipped here only to illustrate structure — not a live feed.
By the numbers

Designed for
continuous duty.

Operating envelope for the OceanPulse instrument family. Throughput and accuracy figures are design targets validated against UMCES/NOAA reference imagery; site-by-site numbers will publish in our validation note as pilots mature.

Throughput
900K
Specimens classified per instrument, per 24 h — design target
Accuracy
2.34%
Mean error on key-point morphology vs. UMCES reference set
Lead time
72h
Forecast horizon, scored on historical bloom events
Coverage
40+
Phyto + zoo taxa identifiable at genus or species level
The instruments

Two scopes,
one signal.

The PlanktonScope family captures the two halves of the planktonic biosphere — the photosynthetic floor of the ocean, and the animal layer that grazes it. Deploy them together for a complete picture of your water column, or independently for targeted programs.

PlanktonScope · Phytoplankton

P‑Scope

Continuous microscopy of the photosynthesizing floor of the food web — diatoms, dinoflagellates, cyanobacteria. Designed to flag harmful bloom precursors days before chlorophyll-a alone would catch them.

Magnification×10–×40
Throughput~900K img / 24h
Cell range2 µm – 500 µm
IlluminationPulsed LED
DeploymentDock · Buoy · Lab
Power12V · 18 W
Spec sheet Pricing
PlanktonScope · Zooplankton

Z‑Scope

Larger field of view, deeper depth of focus — engineered for the animal layer. Tracks copepods, larvae, and gelatinous zooplankton with key-point morphology so size shifts and condition signals aren't lost in the noise.

Magnification×2–×10
Throughput~600K img / 24h
Body range200 µm – 25 mm
IlluminationStroboscopic
DeploymentTow · Mooring · Cage
Power24V · 32 W
Spec sheet Pricing
Solutions

Two product packages.
One biological intelligence layer.

The instruments are the foundation. The packages are how operators actually buy them — bundled with the right classifier, decision layer, and service plan for the job at hand. Start with operational alerts at the pen or intake; expand into compliance evidence for offshore projects when the wedge is ready.

First wedge · pilot phase Aquaculture & coastal infrastructure

ReefWatch

Real-time plankton and particle risk monitoring for the operators watching daily water quality — salmon pens, shellfish leases, drinking-water intakes, and coastal hatcheries.

Built for
  • Aquaculture operations — pen-by-pen risk scoring
  • Drinking-water utilities — cyanotoxin precursor flagging
  • Shellfish growers — closure-window forecasting
  • Coastal hatcheries — HAB & hypoxia early warning
Decisions it powers
  • HAB onset risk & species-specific precursor flags
  • Hypoxia & DO trend forecasts before stress conditions
  • Daily pen / intake / station decision summary
  • Auditable record for state regulators & insurers
Request a pilot → See validation protocol
Roadmap · with offshore-energy partners Offshore energy & ocean infrastructure

EnviroGrid

Compliance-grade ecological monitoring for the projects putting turbines, cables, intakes, and platforms in the water — built for permitting, baseline campaigns, and post-construction evidence.

Built for
  • Offshore wind — baseline & post-construction monitoring
  • Wave / tidal / current / OTEC — permitting evidence at test sites
  • Desalination & seawater intakes — biofouling & clogging risk
  • Coastal observatories & marine labs — persistent biodiversity records
Decisions it powers
  • Pre-/post-deployment biological baselines
  • Particle-plume detection during construction & maintenance
  • Larval-abundance proxy for fisheries impact
  • Compliance-ready datasets for environmental impact assessments
Express interest → See integration surface
Product packages are commercial framings of the same OceanPulse stack. Hardware, classifier, and decision layer are shared; what differs is the target outputs, support model, and reporting cadence.
How it works

From seawater
to signal.

Four stages run continuously and locally — even when the link to shore is intermittent. Buffered acquisition means no gap in the record; staged inference means the forecast updates the moment new evidence arrives.

01

Sample

Seawater flows continuously through the imaging chamber. No nets, no jars, no shift work — the sample is the column itself.

02

Image

Pulse-illumination freezes specimens mid-motion. Pulsed LED + fast exposure cuts turbidity and preserves fragile gelatinous animals.

03

Classify

On-device CV identifies taxa, counts abundance, and measures key-point morphology in real time at the edge.

04

Forecast

Plankton signals fuse with sensor and weather data to score risk at 6h, 24h, 72h, and 7-day horizons — with explainable drivers.

The platform

One pane of glass.
From edge to alert.

Everything the instruments produce — images, classifications, time-series, forecasts — lands in a single role-based interface. Agency-grade, NOAA OneStop-compatible metadata and versioned datasets keep the science reproducible. Webhooks and API keep your operations talking.

Tiered alerting

Multi-signal early warning.

Static thresholds, rate-of-change, and ensemble risk scoring gate every alert. Each notification ships with the top contributing drivers — so on-call teams know what to do, not just what's happening.

Critical
HAB onset · Karenia brevis Risk 0.91 · 72h horizon · Tampa Bay intake A
02:14
Warning
Hypoxia trend · DO 4.1 mg/L Falling 0.6 mg/L per 24h · Pen 7
06:48
Advisory
Diatom rate-of-change Skeletonema +34% wk-over-wk · Station C-3
11:03
Illustrative — demo data, not live telemetry
Multi-horizon forecasting

6h. 24h. 72h. 7d.

Bloom and toxin probability at four horizons, every horizon shipped with confidence intervals and the drivers behind it. Validation framework runs against historical events at every site.

HAB risk · 7d peak FRI · 0.90 ± 0.06
Real-time AI classification

Forty-plus taxa, identified at the edge.

A single specimen passes through the lens and lands as structured data within seconds — name, confidence, key-point measurements, and a thumbnail filed in the lab archive.

Skeletonema costatum
p 0.94
Calanus finmarchicus
p 0.91
Pseudo-nitzschia
p 0.88
Acartia tonsa
p 0.86
Karenia brevis
p 0.82
Oikopleura dioica
p 0.79
Data integrity

Agency-grade by default.

Every image, classification, and reading carries timestamp, geolocation, depth, instrument ID, and calibration state. Versioned datasets, schema-checked ingestion, drift monitoring — so the science is reproducible the moment it lands.

timestamp geolocation depth instrument_id calibration drift_flag quality_pass schema_v3
Sensor fusion

Plankton, in context.

Biology gets meaning when it lands beside chemistry, weather, and hydrodynamics. Pull in your existing CTD, sondes, and satellite chlorophyll — OceanPulse aligns it all in time and space.

Temperature18.1 °C
Salinity32.4 PSU
Dissolved O₂6.4 mg/L
pH8.07
Turbidity4.1 NTU
Chlorophyll-a9.4 µg/L
The category

The biological
intelligence layer.

Beyond water-quality sensors. Beyond sample microscopy. OceanPulse is a deployed instrument, AI classification, forecasting, and alerting platform for plankton-driven water risk — built for operators who cannot wait for lab confirmation.

Traditional approach OceanPulse
Chlorophyll & turbidity proxies Taxa, abundance, morphology, images
Bottle samples · lab queues Continuous in-situ microscopy
Regional HAB awareness Site-specific risk scoring
Single sensor feeds Plankton + chemistry + weather
Retrospective reports Multi-horizon operational alerts
Performance figures shown across this site are pilot or benchmark values and vary by site, configuration, taxa, turbidity, and deployment mode. OceanPulse is in pilot phase; "field-proven" claims publish only after third-party acceptance review.
Evidence

What we'll
publish first.

Before "field-proven" goes anywhere on this site, three things publish: the validation protocol we're scoring against, the design targets we're scoring, and the deployment list scoring them. Below is a snapshot of each — built on UMCES/NOAA reference imagery and our deployment pipeline.

01 · Validation protocol

Five tests, four weeks, scored at every site.

Throughput
≥ 800K img / 24h
rolling 96h continuous capture · UMCES illumination profile
Classification accuracy
F1 ≥ 0.85 across HAB taxa
balanced eval set · double-blind ground truth
Morphology error
≤ 3% on key-point measurements
synthetic + UMCES reference · per-keypoint MAE
Drift
< 5% accuracy delta over 30d
weekly snapshot vs. fixed eval set · automatic flag
Downtime
≥ 99.0% capture uptime
instrument-side watchdog · independent shore log
02 · Throughput · target vs. measured

Where we are, against where we'll commit.

Phytoplankton (P-Scope)
740 measured· 900 target· img·k / 24h
82% of target
lab loop · UMCES illumination
Zooplankton (Z-Scope)
510 measured· 600 target· img·k / 24h
85% of target
tow tank · stroboscopic
Edge inference latency
280 measured· 220 target· ms / image
100% of target
median · Jetson Orin Nano
Forecast horizon
48 measured· 72 target· h
67% of target
scored on 2024–2025 events
Lab + reference-loop figures. Site-by-site numbers will publish in the validation note as pilots commission. Source imagery: UMCES CBL reference set and synthetic loops; not a live operating record.
03 · Taxonomy coverage · v3 classifier

Forty-plus taxa, scored honestly.

GroupTaxonTypeF1RecallEval status
Phyto Skeletonema costatum Diatom · chain 0.94 0.97 In reference set
Phyto Pseudo-nitzschia spp. Diatom · pennate 0.88 0.92 In reference set
Phyto Karenia brevis Dinoflagellate · HAB 0.82 0.89 In reference set
Phyto Microcystis aeruginosa Cyanobacteria · HAB 0.79 0.86 In reference set
Phyto Alexandrium fundyense Dinoflagellate · HAB 0.74 0.83 Limited reference
Zoo Calanus finmarchicus Copepod 0.91 0.95 In reference set
Zoo Acartia tonsa Copepod 0.86 0.92 In reference set
Zoo Oikopleura dioica Larvacean 0.79 0.85 In reference set
Zoo Euphausia pacifica Krill 0.88 0.93 In reference set
Zoo Mnemiopsis leidyi Ctenophore 0.71 0.80 Limited reference
Showing 10 of 42 trained taxa. Full coverage table publishes with the validation note. F1 / recall scored against held-out UMCES + open reference imagery. Not scored on live customer data.
04 · Deployment status

Where we are, in plain text.

CB-01Site survey
Chesapeake Bay · Choptank
Drinking water2026-Q2
CB-02MOU drafted
Chesapeake Bay · Patapsco
Research2026-Q2
TB-01Pilot proposal
Tampa Bay
HAB monitoring2026-Q3
PS-03Site survey
Puget Sound · pen 7
Aquaculture2026-Q3
ME-04Letter of intent
Gulf of Maine · Cobscook
Aquaculture2026-Q3
BC-01Reference data
Bering Sea · canyon
Researchvalidated
Six prospective deployments tracked publicly. No site is "live in production" as of v2026.05; we'll move cards to a Live state only after acceptance review.
Trust layer

The boring
that matters.

Every reading carries the metadata it takes to defend itself. Every model carries the version and confidence it was scored at. Every alert carries the drivers behind it. None of it is implicit.

Metadata schema · OneStop-compatible

Every observation is self-defending.

Conforms to the NOAA OneStop ISO-19115 profile. Schema-checked at ingestion; rejected payloads route to a quarantine bucket with a human-readable reason. 

{
  "instrument_id":   "OP-PSC-014",
  "deployment":      "CB-01 · Choptank · pier",
  "captured_at":     "2026-04-22T02:14:07.213Z",
  "geo":             { "lat": 38.9784, "lon": -76.4922 },
  "depth_m":         4.2,
  "calibration":     { "version": "cal-2026-03-11", "passed": true },
  "model":           { "version": "v3.2", "weights_sha": "8af1c2…" },
  "classification":  { "taxon": "Karenia brevis", "p": 0.91, "ci95": [0.85, 0.95] },
  "morphology":      { "length_um": 32.6, "width_um": 18.1, "kp_mae": 0.7 },
  "drivers":         ["cell_density↑", "sst↑", "wind_shear↓"],
  "quality":         { "schema_v": "3", "drift_flag": false, "qa_pass": true },
  "license":         "CC-BY-4.0 derived — see /docs/licensing"
}
Model versioning

You always know which model called the alert.

v3.2 current 2026-04-09 + Microcystis · drift retrain
v3.1 2026-02-18 + key-point head · zoo body length
v3.0 2025-12-04 phyto + zoo unified backbone
v2.4 2025-09-22 Pseudo-nitzschia recall pass
Uncertainty policy

Confidence is published, not implied.

  • Per-class probability + 95% CI on every classification.
  • Forecast horizons publish bands, not point estimates.
  • Below-threshold detections route to active-learning queue, not to alerts.
  • "Probable" / "Possible" / "Indeterminate" labels are reserved for ambiguous low-confidence frames.
Drift monitoring

The eval set runs weekly. The instrument knows.

w-12F1 baseline 0.93 · band ±0.04now
QA workflow

Human in the loop, by design.

01Captureinstrument · sampled by frame
→
02Auto-classifyedge model · v3.2
→
03Confidence gate≥ 0.80 → publish · < 0.80 → review
→
04Active learningreview queue · weekly relabel
→
05Retrain · scoremonthly · gated by validation eval
→
06Promote · publishrelease notes + diff vs. prior
Integration & security

Talks to your stack.
Owned by you.

OceanPulse runs in your VPC, on-prem, or hosted — your choice. The platform exposes versioned REST + webhook surfaces, an S3-compatible image archive, and offline-first capture so a flaky link doesn't lose science.

API · v3 · OpenAPI

REST + webhooks · stable since v3.0.

GET /observations time-bounded query · paginated · filterable by taxon / site
GET /observations/{id} full payload · image URL · metadata · model version
GET /forecasts/{site} current forecast bundle · 6h / 24h / 72h / 7d
GET /instruments fleet roster · health · last-seen · firmware
POST /webhooks subscribe to alerts · classifications · drift events
POST /labels submit ground-truth labels · feeds active-learning queue
Deployment

Three places it can live.

Hosted
AWS us-east / us-west · SOC 2 in progress · single-tenant available
VPC
deploy into your AWS / GCP / Azure account · Terraform module
On-prem
air-gapped option · ARM + x86 builds · USB / sneakernet model updates
Offline mode

Capture survives the link.

Every instrument carries 30 days of local storage and a buffered upload queue. When the link returns, the record fills back in — with the original timestamps, not the upload time.

28 days ↳ 6 days offline · zero data loss
Access control

Roles & audit, agency-grade.

Admindeployment · billing · all data
Operatoralerts · forecasts · their sites only
Scientistraw imagery · labels · model versions
Auditorread-only · audit log · export
Audit log

Who did what, when.

02:14alert.publishHAB onset · Karenia brevis · CB-01
02:14alert.deliverwebhook · cwa-ops · 200 OK
02:17alert.acki.okafor@cwa.gov
06:48alert.publishHypoxia trend · pen 7 · PS-03
09:02label.submit12 labels · scientist · CB-01
09:30model.deployv3.2 · validation pass
Use cases

Built for the
people on call.

Marine science, water utility ops, and aquaculture move on different clocks. OceanPulse adapts the same underlying signal into role-specific dashboards — so the alert that lands in your inbox is the alert you can act on.

in ReefWatch

Drinking water utilities

Forecast intake bloom risk and probable treatment stress before acute events.

  • Cyanotoxin precursor detection
  • Operational adjustments hours ahead of an event
  • Auditable record for state regulators
Talk to a specialist
in ReefWatch

Aquaculture operations

Combined hypoxia and HAB warning so feeding, aeration, and harvest decisions land before a loss event.

  • Pen-by-pen risk scoring
  • Operational decision support validated in pilot deployments
  • Open API into farm management systems
Talk to a specialist
in EnviroGrid

Research & monitoring

Continuous, high-resolution biodiversity records — including the species that nets damage and bottles miss.

  • Long-term taxonomic baselines
  • Active-learning labeling workflow
  • NOAA OneStop-compatible metadata
Talk to a specialist
Pilot status

What we'll publish,
and when.

OceanPulse is in pilot phase. We're committed to publishing a validation note, deployment list, and acceptance review before any claim of "field-proven" appears on this site. Below is what we intend to make public, and the cadence we're holding ourselves to.

Active pilot deployments
in negotiation
stations across drinking-water, aquaculture, and research programs
Validation framework
v1.0 draft
throughput, classification accuracy, drift, downtime — published before commercial launch
Reference dataset
UMCES + NOAA imagery
open imagery used to score classifier accuracy across phyto + zoo taxa
Independent acceptance
scheduled
third-party deployment review with public summary
Operating in good faith with prospective customers, agencies, and reviewers. Materials available under NDA today; all of the above will be public at v1.0.
Get started

See it on your water.

Tell us about your site, your concern, and your timeline. A solutions engineer will get back inside one business day with a deployment sketch and a quote.

↳ Pilot deployments under NDA ↳ NOAA OneStop-compatible metadata ↳ On-prem option
No marketing list. Reply within 1 business day. See privacy & data handling.
Privacy & data handling

How we treat
what you send us.

Pilot-phase summary. A full privacy policy publishes alongside our v1.0 commercial launch; this section covers how OceanPulse handles the data flowing through this site and through pilot deployments today.

Demo & pilot requests
scope only
Used to size your deployment and reply directly. Never added to a marketing list, shared with third parties, or sold.
Pilot deployment data
under NDA
Images, classifications, and metadata captured during pilot deployments are held under the deployment NDA. Customer data is never used to train shared models without explicit consent.
Cookies & analytics
essential only
This site uses only essential cookies. No third-party trackers, no ad networks, no cross-site identifiers.
Data access & deletion
on request
For data deletion, access, or correction requests, reach us via the demo request form. Responses within one business day.
OceanPulse is operated by sigaihealth (the company behind oceanpulse.ai). For specific questions about data handling under a pilot agreement, contact your solutions engineer directly.