1. Scientific and Environmental Need
Ocean alkalinity enhancement is one of the few carbon dioxide removal approaches with a mechanism grounded in equilibrium chemistry rather than in engineered permanence. Adding alkalinity to seawater shifts the carbonate system, converting dissolved carbon dioxide to bicarbonate and drawing atmospheric carbon dioxide into the surface ocean. The physical chemistry is not disputed. What is unresolved is the measurement problem: quantifying how much carbon a given alkalinity addition actually removed, in an ocean that mixes, advects and dilutes the intervention faster than any conventional sampling platform can follow it.
That measurement problem is now the binding constraint on the field. As of October 2025, offtake agreements had secured 578,000 tonnes of marine carbon dioxide removal, and only 0.3% of that contracted volume had been formally issued as credits, with issuance concentrated in a small number of ocean alkalinity enhancement projects (AlliedOffsets, Marine-CDR: Market Landscape, Momentum, and Challenges, 22 October 2025). Fifty-six developers are active in a sector that has absorbed $209 million of private investment. Verification, not chemistry and not capital, is what is failing.
The scale of the sampling deficit has been measured directly. In September 2023 a team led by the Woods Hole Oceanographic Institution released a rhodamine-traced patch in federal waters south of Martha’s Vineyard and tracked it for 36 hours from the R/V Connecticut. The patch displaced approximately 14.8 km south-southeast over that period. Underway sampling at intervals of ten minutes or longer corresponds, as the authors note, to one sample every 1,200 metres of vessel track. Tracer concentration fell from a peak near 720 ppb to 4 to 5 ppb above a 0.1 ppb baseline within 36 hours. And for the simulated 20-tonne sodium hydroxide addition analysed in the same study, only 10% of the potential carbon removal had occurred by the time monitoring ended (Subhas et al., Biogeosciences, 2025).
The authors identify the constraint without qualification: “Dispersion and dilution of water masses, both horizontally and vertically, will be a critical factor, as will the ability to track the intervention through space and time. This is the central challenge of in-water MRV.” They recommend sampling above 1 Hz and monitoring campaigns extending far beyond the duration a research vessel charter can support.
The consequence is a mismatch between the timescale of the science and the timescale of the instrument. Carbon uptake proceeds over weeks to months as air-sea gas exchange equilibrates the treated water mass, while ship-based campaigns operate over days. A 27.4-metre coastal-class research vessel bills $10,937 per 24-hour day at 2025 rates (University of Connecticut Department of Marine Sciences, effective July 2024), so the platform capable of the measurement cannot economically remain on station for the duration of the process it measures. Ocean alkalinity enhancement additionally counteracts local ocean acidification, and its deployment at climatically meaningful scale is contingent on a verification method regulators, registries and coastal stakeholders can independently trust. Verification is a precondition for the field, not an accessory to it.
2. State of the Art
Four research communities hold the components of a solution, and each has published its own ceiling.
Autonomous carbonate-system sensing
The Ocean Technology and Engineering group at the National Oceanography Centre, Southampton (Socratis Loucaides, Matthew Mowlem, Allison Schaap, Stathys Papadimitriou) has developed microfluidic lab-on-chip analysers that perform spectrophotometric titration for total alkalinity and spectrophotometric pH determination in situ. The alkalinity analyser achieves precision and accuracy better than 5 µmol kg−1 against certified reference materials (Schaap et al., ACS Sensors, 2025). The group then integrated both sensors onto the Autosub Long-Range autonomous underwater vehicle for an eleven-day Celtic Sea deployment, the first characterisation of the marine carbonate system from autonomous subsurface measurements (Hammermeister et al., Environmental Science & Technology, 2025). The methodological ceiling is not analytical. The vehicle executed a pre-planned transect, and the resulting offsets between sensor readings and validating shipboard co-samples ranged from one minute to 85 hours. The authors explicitly caution that special care is required when applying their validation treatment “in waters with high spatial biogeochemical variability”, which is a precise description of a dispersing alkalinity plume.
Autonomous fixed-point monitoring of an active intervention
A group at Dalhousie University (Vincent Sieben, Douglas Wallace) deployed an autonomous lab-on-chip alkalinity analyser beside a live magnesium hydroxide dosing operation, holding station 60 metres from the discharge for 40 days (Zabihihesari et al., Communications Engineering, 2026). The instrument resolved tidal coherence structure before and during dosing and detected cumulative alkalinity retention. The ceiling here is geometric rather than analytical: an Eulerian point measurement cannot reconstruct the three-dimensional, time-varying alkalinity field that a crediting protocol requires.
Autonomous feature tracking on long-endurance vehicles
The Monterey Bay Aquarium Research Institute (Yanwu Zhang, Brian Kieft, Brett Hobson, James Bellingham, Christopher Scholin) has demonstrated the control primitive this problem needs, applied to different variables. Their long-range autonomous underwater vehicle, carrying a third-generation Environmental Sample Processor, autonomously tracked and sampled the deep chlorophyll maximum inside a cyclonic eddy in the North Pacific Subtropical Gyre for four days without surfacing, holding vertical position by locking onto the isotherm corresponding to the chlorophyll peak and flying tight circles while drifting with the eddy current, producing a quasi-Lagrangian time series (Zhang et al., IEEE Journal of Oceanic Engineering, 2020). The same group subsequently demonstrated autonomous tracking of salinity-intrusion fronts on the same platform (Zhang et al., 2022), extending the primitive from a stratified layer to a laterally propagating boundary. The ceiling is the observable: chlorophyll fluorescence and conductivity are continuous, low-power, high-rate measurements. No published control law has been closed around a carbonate-system observable that arrives once every ten minutes.
Learned adaptive sampling policies
The reinforcement-learning literature on plume tracing has produced policies that outperform the canonical lawnmower survey in coverage efficiency, including a double deep Q-network formulation for subsurface plume source tracing (Wu et al., Robotics, 2024). The authors state their scope precisely: the approach was tested in numerical simulation and on a ground vehicle. This body of work has not been flown on an autonomous underwater vehicle in water, and it has not been evaluated against a crediting objective.
The field’s structure is therefore the finding. Analytical chemistry groups have solved the sensing. Ocean modelling groups (Galen McKinley at Columbia, the [C]Worthy collaboration) have built the transport and air-sea flux models that convert an observed alkalinity field into a carbon accounting statement. Marine autonomy groups have solved feature-following on production hardware. No group has joined the second capability to the first, and Section 6 examines why.
Scope of the novelty claim
Stated precisely, the claim is that no published deployment has used a carbonate-system observable as the onboard state variable in a closed-loop adaptive survey of an intentional alkalinity-addition plume. Three adjacent literatures are named here because a reviewer will raise them and because the distinction is substantive rather than cosmetic.
Sub-seabed carbon storage leak monitoring. STEMM-CCS (Goldeneye, North Sea, 2019) and QICS (Scotland, 2012) integrated novel and commercial pH and pCO2 sensors with landers, AUVs and ROVs for high-resolution surveys of a controlled sub-seabed CO2 release, resolving minimum pH of 7.965 and maximum pCO2 of 942.1 µatm within 8 m of the bubble stream, corresponding to 16.4% [H+] and 125.6% deviations from baseline. This is the closest prior art. It differs in objective: leak monitoring answers whether and roughly where a release is occurring under a regulatory obligation, whereas verification answers how much was removed, to a stated uncertainty, for credit issuance. A detection objective is satisfied by finding the anomaly; a quantification objective is satisfied only by bounding an integral, and the two produce different trajectories.
Hydrothermal and chemical plume tracing. Autonomous loops have been closed on chemical observables for two decades, in the Farrell and Li plume-tracing lineage and in nested-survey vent localisation with ABE and Sentry. The observable is generally redox potential, turbidity or methane rather than carbonate, and the objective is source localisation rather than anomaly quantification.
Autonomous carbonate observing platforms. BGC-Argo floats and gliders carrying solid-state pH sensors, and Saildrone uncrewed surface vehicles carrying NOAA PMEL’s ASVCO2 system, collect carbonate-system data at scale, including a 22,000 km, 196-day autonomous circumnavigation of Antarctica producing direct air-sea CO2 flux measurements (Sutton et al., Geophysical Research Letters, 2021). These are Lagrangian drift or pre-planned trajectories; the carbonate measurement is payload, not state variable.
Confirming or refuting this scoping is an explicit first-quarter task of the proposed programme, and the claim is stated at the layer it survives at rather than at the broadest layer available.
3. Foundational Research
Subhas AV, Rheuban JE, Wang ZA, McCorkle DC, Michel APM, Marx L, Dean CL, Morkeski K, Hayden MG, Burkitt-Gray M, et al. (2025). “A tracer study for the development of in-water monitoring, reporting, and verification (MRV) of ship-based ocean alkalinity enhancement.” Biogeosciences, 22, 5511–5534. DOI: 10.5194/bg-22-5511-2025.
Methodology: 56 kg of powdered rhodamine WT dye in 1,000 L of solution was dispersed over 75 minutes at 0.2 L s−1 from the R/V Connecticut (27.4 m) in federal waters south of Martha’s Vineyard (41.530668° N, −70.645629° W) on 1 to 3 September 2023. The patch was then monitored for 36 hours using underway surface sampling at intervals of ten minutes or less (rhodamine to 0.5 Hz), nine CTD casts at roughly four-hour spacing, four Lagrangian drifters of which two carried rhodamine fluorometers logging at one-minute intervals, and three high-resolution satellite images from Planet Labs SkySat and PlanetScope. A hypothetical 20-tonne sodium hydroxide addition was superimposed on the observed dye field using a TA:RT ratio of 8.9 µmol kg−1 ppb−1 to simulate the carbon signal an actual intervention would produce.
Results: 214 paired measurements of temperature, salinity, fCO2, total alkalinity and rhodamine were collected, 168 of them in-patch (78%) and 46 baseline (22%). Rhodamine fell from a ~720 ppb peak during dispersal (58 ppb mean) to 4–5 ppb above a ~0.1 ppb baseline at 36 hours, with visual signal loss at ~26 hours. The patch displaced ~14.8 km south-southeast. Detection limits for a real intervention were established at >10 µmol kg−1 for total alkalinity, >0.01 units for pH and >10 µatm for fCO2; substituting a dynamic baseline for a static one reduced baseline variability by 60% for total alkalinity and 25% for fCO2. The simulated intervention produced, at 36 hours, an fCO2 increase of 4 µatm, a pH decrease of 0.004 units and a DIC increase of 1.8 µmol kg−1, representing 10% of total potential carbon dioxide removal.
Significance for the proposed work: this is the quantitative specification of the problem, produced by the group that holds the only United States EPA permit for open-ocean alkalinity enhancement. It establishes three parameters our system must satisfy. The vehicle must remain with a water mass that translates ~10 km per day. It must resolve anomalies at the 10 µmol kg−1 level against a baseline that must itself be sampled to be useful. And it must persist far beyond 36 hours, because the majority of the carbon signal has not yet formed at that point. The authors’ own recommendations, sampling above 1 Hz and campaigns of much longer duration, are jointly unsatisfiable by a crewed vessel and jointly satisfiable by a long-endurance autonomous platform that decides where to go.
Hammermeister EM, Papadimitriou S, Arundell M, Ludgate J, Schaap A, Mowlem MC, Fowell SE, Chaney E, Loucaides S (2025). “New Capability in Autonomous Ocean Carbon Observations Using the Autosub Long-Range AUV Equipped with Novel pH and Total Alkalinity Sensors.” Environmental Science & Technology, 59(14), 7129–7144. DOI: 10.1021/acs.est.4c10139. PMID: 40168248.
Methodology: microfluidic lab-on-chip sensors for pH, total alkalinity and dissolved inorganic carbon were integrated onto the Autosub Long-Range (ALR6000 variant, 3.6 m length, 750 kg nominal weight, 6,000 m depth rating, 60 W hotel load, ~10-day endurance, 550 km range at 0.6 m s−1), deployed in the Celtic Sea from 19 to 30 March 2022 across water depths of 100 to 3,000 m on shelf and deep transects. Sensor output was validated against 156 discrete CTD co-samples.
Results: 947 in situ pH measurements at one per 7.5 minutes (precision <0.001; accuracy 0.003 ± 0.022 against validation seawater; combined standard uncertainty ±0.010), 423 total alkalinity measurements at one per 10 minutes (precision and accuracy better than 5 µmol kg−1; estimated uncertainty ±7 µmol kg−1), and 251 DIC measurements at one per 15 minutes. Mean total alkalinity residuals against co-samples were 1 µmol kg−1 (σ = 4, n = 191) on the shelf transect and 2 µmol kg−1 (σ = 5, n = 129) on the deep transect; derived DIC residuals were 4 to 7 µmol kg−1 and derived pCO2 residuals 10 to 17 µatm. The DIC prototype failed during the mission through gas-exchange-unit failure and calibration error and was flagged unreliable. Sensor-to-validation temporal offsets ranged from 1 minute to 15 hours on the shelf transect (mean 6 h) and 5 minutes to 85 hours on the deep transect (mean 40 h).
Significance for the proposed work: this removes the sensing question from the risk register. Registry-grade total alkalinity measurement already rides a long-endurance AUV with an energy budget that permits ten-day missions, and its accuracy comfortably exceeds the 10 µmol kg−1 detection requirement Subhas et al. established. Two constraints carry directly into our design. The 10-minute alkalinity cadence is the fundamental limit on how fast the credit-relevant observable can inform a control decision, and the authors’ own documented offsets demonstrate that a pre-planned trajectory cannot guarantee measurement co-location in a spatially variable field.
Zabihihesari A, Burt W, Sonnichsen C, Motahari S, Whitworth A, Izett R, Fradette C, Wallace D, Sieben V (2026). “High frequency in situ total alkalinity measurement for monitoring ocean alkalinity enhancement field trials.” Communications Engineering, 5. DOI: 10.1038/s44172-026-00665-w. PMID: 42020707. PMCID: PMC13323400.
Methodology: the first field deployment of an autonomous lab-on-chip total alkalinity analyser during an active ocean alkalinity enhancement trial using magnesium hydroxide slurry. The analyser was co-deployed with pH, salinity and temperature sensors at a fixed position 60 m from the discharge point and operated continuously for 40 days in 2023, with onboard certified reference material measurements interleaved for drift control.
Results: 314 total alkalinity measurements and 52 onboard certified reference material measurements from approximately 3,300 optical readings over 40 days. High-frequency records revealed stronger semi-diurnal tidal coherence before dosing, followed by reduced coherence and more variable phase relationships as dosing progressed. Total alkalinity relative to a baseline alkalinity-salinity relationship increased by approximately 40 µmol kg−1 after roughly 210 tonnes of alkaline addition and did not return to baseline between dosing intervals, indicating a system memory effect with cumulative alkalinity retention.
Significance for the proposed work: two findings transfer. First, autonomous alkalinity instrumentation survives 40 days of unattended operation beside a working discharge with onboard reference-material quality control, which is the endurance and traceability profile a crediting campaign requires. Second, the cumulative retention result is a warning about sampling design: alkalinity did not relax to baseline between dosing events, so a monitoring strategy that assumes a clean pre-dose baseline will misattribute the signal. This is a direct argument for adaptive sampling that allocates effort between the patch and a dynamically re-estimated baseline, which is the objective function our planner optimises.
Zhang Y, Kieft B, Hobson BW, Ryan JP, Barone B, Preston CM, et al. (2020). “Autonomous Tracking and Sampling of the Deep Chlorophyll Maximum Layer in an Open-Ocean Eddy by a Long-Range Autonomous Underwater Vehicle.” IEEE Journal of Oceanic Engineering, 45(4), 1308–1321. DOI: 10.1109/JOE.2019.2920217.
Methodology: in March and April 2018 autonomous underwater and surface vehicles were deployed into a cyclonic eddy in the North Pacific Subtropical Gyre to study microbial community variability in the deep chlorophyll maximum. One long-range AUV carrying a third-generation Environmental Sample Processor autonomously tracked and sampled the layer. Rather than servoing on the chlorophyll fluorescence peak directly, the vehicle maintained vertical position by locking onto the isotherm corresponding to the chlorophyll peak, and ran tight circular patterns while drifting with the eddy current to hold a quasi-Lagrangian reference frame.
Results: continuous autonomous tracking and sampling of the layer for four days without surfacing, producing a quasi-Lagrangian time series of the same water mass rather than a spatial transect through changing water masses.
Significance for the proposed work: this is the single closest precedent for the capability we propose, and it contributes both an existence proof and a design principle. The existence proof is that multi-day autonomous feature-following with onboard sampling is production capability on this vehicle class. The design principle is the isotherm lock. The team chose to control on a fast, low-noise proxy and to use the expensive target observable to interpret rather than to steer. That is precisely the architecture an alkalinity plume demands, where pH, salinity and temperature update continuously while total alkalinity updates every ten minutes, and it is the reason our proposed inner control loop servos on the pH-salinity anomaly with alkalinity used to update the proxy relationship.
Zhang Y, Yoder N, Kieft B, Kukulya A, Hobson BW, Ryan S, Gawarkiewicz GG (2022). “Autonomous Tracking of Salinity-Intrusion Fronts by a Long-Range Autonomous Underwater Vehicle.” IEEE Journal of Oceanic Engineering, 47(4), 950–958. DOI: 10.1109/JOE.2022.3146584.
The same vehicle class was tasked to detect and follow a laterally propagating salinity-intrusion front, extending autonomous feature following from a horizontally stratified layer to a moving lateral boundary. An alkalinity patch is a lateral, advecting, diluting structure, so this is the closer geometric analogue of the two MBARI results. Together the pair establishes that the tracking primitive generalises across feature topologies on operational hardware, which allows us to treat autonomy as an integration and objective-design risk rather than a fundamental controls research risk.
Schaap A, Papadimitriou S, Mawji E, Walk J, Hammermeister E, Mowlem M, Loucaides S (2025). “Autonomous Sensor for In Situ Measurements of Total Alkalinity in the Ocean.” ACS Sensors, 10, 795–803. DOI: 10.1021/acssensors.4c02349.
The instrument paper for the analyser used in both deployments above: a microfluidic lab-on-chip device performing spectrophotometric titration in situ, characterised against certified reference materials, achieving precision and accuracy better than 5 µmol kg−1. It is included for a reason that is not analytical. The analyser is a laboratory-fabricated assembly, and every alkalinity measurement referenced in this brief was produced by a small number of hand-built units held by two research groups. A programme running simultaneous campaigns requires tens of units at controlled tolerance with a documented calibration and traceability record, because credits are audited. That is a design-for-manufacturability problem, and it is why manufacturing engineering appears here from month one rather than at the translation stage.
Wu Z, Wang S, Shao X, Liu F, Bao Z (2024). “Adaptive Path Planning for Subsurface Plume Tracing with an Autonomous Underwater Vehicle.” Robotics, 13(9), 132. DOI: 10.3390/robotics13090132.
A Double Deep Q-Network agent learns a plume-source-tracing policy by interacting with the environment, with the optimal heading extracted from a deep network mapping. The learned policy was compared against the canonical lawnmower path used in practice and favoured for large-scale exploration. The authors state that the approach “was tested by numerical simulation and on a real ground vehicle.” This establishes both that learned policies outperform fixed survey patterns on plume problems and that the literature stops short of in-water AUV validation. It also identifies the formulation gap we intend to close: source-tracing rewards finding a source, while verification rewards reducing uncertainty in a quantity a registry will price. The objective function, not the learning algorithm, is the research contribution.
4. Competitive Landscape
This landscape has three layers and conflating them produces a competitor count that will not survive review. The sensor layer is crowded: Sunburst Sensors, Sea-Bird (SeaFET), ProOceanus, 4H-Jena, ANB Sensors, and Clearwater Sensors, the National Oceanography Centre lab-on-chip spinout whose product line is the instrument this programme depends on. The platform layer is crowded and well capitalised: Saildrone, Teledyne Webb (Slocum), Kongsberg (HUGIN), Ocean Infinity, Bedrock Ocean, Sofar Ocean and Open Ocean Robotics. The decision layer is empty, and that is where this proposal sits.
Two entrants sit closest. Open Ocean Robotics (Victoria, British Columbia; approximately $7.22 million raised, including a CA$2.8 million round in October 2024) builds solar-powered uncrewed surface vessels and has released a second-generation platform aimed at marine carbon dioxide removal monitoring. atdepth MRV (Cambridge, Massachusetts; an MIT spin-off funded through ARPA-E SEA-CO2 at $2,524,964) provides GPU-accelerated ocean and biogeochemical modelling to removal operators. Both partner on the Ocean Frontier Institute’s SCALE MRV project at Dalhousie University. Saildrone is a genuine carbonate-observation operator rather than an adjacent one, having carried NOAA PMEL’s ASVCO2 system on a 22,000 km Antarctic circumnavigation, and its limit is architectural rather than analytical: its vehicles are surface platforms executing operator-specified missions.
From a research standpoint the relevant observation is capability topology rather than market share. Platform vendors sell vehicles that execute missions specified by an operator; modelling vendors sell inference over data whose sampling design they did not choose. Neither addresses how a vehicle should decide where to sample next in order to minimise uncertainty in a carbon accounting statement. The platform products are also surface-based, while an alkalinity plume disperses vertically as well as horizontally, so subsurface structure is unobserved by either.
The most consequential entry is not a competitor but a threat to the premise. [C]Worthy, a non-profit ARPA-E-funded at $3,884,825, is building the open modelling framework for model-based mCDR quantification, and the Isometric protocol states that quantification currently relies on biogeochemical ocean models. If registries lean further toward modelled quantification, demand for in-water measurement weakens. The counter is that the same protocol requires quantification of model skill through data-model comparison, so observations bound what a model may claim; but this is the demand-side assumption the programme should test early rather than assume, and Section 8 carries it as a named risk. Removal operators including Planetary Technologies, Ebb Carbon, Vesta, Captura and Equatic run in-house MRV and could internalise this layer; Running Tide’s 2024 wind-down is the cautionary case in the same set.
Commercial solutions do not solve the research problem for a further reason. Every deployed verification workflow to date, including the EPA-permitted LOC-NESS trial, depends on a rhodamine dye tracer to locate the treated water mass. Dye is an excellent research instrument. It is not a viable basis for routine commercial verification, because repeated dye release at operational dosing cadence carries its own permitting burden under the same statute that governs the alkalinity release. A verification method that works on the carbonate signal alone is a research requirement before it is a product requirement.
5. Addressable Scope and Public Benefit
Federal research programmes increasingly require applicants to show a route from funded science to deployed benefit. The route here is unusually direct, because the science being funded is the thing currently blocking deployment.
What the crediting standard demands. The Isometric registry’s Ocean Alkalinity Enhancement protocol requires characterisation across three spatio-temporal regimes: the mixing zone over seconds to hours, the coastal domain over hours to years and potentially several hundred square kilometres, and the open-ocean domain at basin scale over months to years. It requires construction of a time-variable three-dimensional alkalinity forcing function, quantification of model skill through data-model comparison, and it sets a materiality threshold of 5% for the totality of omissions, errors and mis-statements. This specification, rather than any vendor roadmap, defines the observational workload.
Bottom-up cost of the present method. Using the Subhas campaign as the measured unit: 36 hours of vessel day-rate at $10,937 per day is approximately $16,400, excluding personnel, sample analysis and instrumentation, so the true campaign cost is higher. The simulated intervention was 20 tonnes of sodium hydroxide. At 40 g mol−1 that is 5.0 × 105 moles of alkalinity; applying a conversion of 0.8 moles CO2 per mole of added alkalinity gives an uptake potential of approximately 17.6 tonnes of CO2. Potential is not removal: the realised fraction is governed by plume dilution, residence time and air-sea gas exchange rate, and the study measured 10% realised at 36 hours. Roughly 1.76 tonnes of uptake were therefore verified for $16,400 of ship time, or approximately $9,300 per verified tonne, against an implied $271.68 per tonne from the largest disclosed offtake ($31.3 million from Frontier buyers to Planetary Technologies for 115,211 tonnes delivered 2026 to 2030). That is the unit price of a single named transaction rather than a market mean.
Two qualifiers travel with the figure wherever it is used. Research campaigns are deliberately over-instrumented and purchase process understanding rather than credits, so the ratio bounds the current state rather than forecasting steady-state cost. And the gap between uptake potential and realised removal is itself the scientific argument for this programme: dilution, residence time and gas-exchange rate are none of them observable from a ship snapshot, and all require sustained spatially resolved tracking of an advecting patch.
Why verification cost does not collapse at scale. The immediate objection is that the same 1.5 days of ship time spread over a 10,000-tonne deployment costs under $2 per tonne, dissolving the problem. It does not, for three reasons that are physical rather than economic. Dilution drives the anomaly toward the detection floor as the treated domain grows, so a larger deployment presents a fainter signal. Sampling requirements scale with advected area and residence time rather than with tonnage, since a patch from a large dose still translates roughly 10 km per day and still has to be followed. And registry protocols specify spatial and temporal coverage across defined regimes rather than a fixed sample count, so the observational obligation grows with the domain. Verification effort tracks the geometry of the water mass, not the mass of reagent added to it.
Serviceable scope and its honest limits. Applying a 20% verification share to the 578,000 tonnes already contracted at $271.68 per tonne implies roughly $31 million of embedded verification spend across deliveries running to 2030. This is below $100 million and is flagged as a small market at present scale. The nearer and larger pool is federal: $36 million through ARPA-E SEA-CO2 and $24.3 million through NOAA’s FY23 NOPP awards, detailed in Section 7. Verification demand then scales linearly with verified volume, so each additional 1 million tonnes per year of verified removal adds approximately $54 million per year of in-water verification demand at the same share.
Top-down cross-check. MarketsandMarkets sizes the autonomous underwater vehicle market at $3.13 billion in 2025 rising to $4.64 billion by 2030, an 8.2% CAGR, with unit volume growing from 997 to 1,424 vehicles (Autonomous Underwater Vehicle (AUV) Market, November 2025). Verification services are a small fraction of that, confirming that platform supply is not the constraint; verification capability is.
Payment and crediting pathway. No CPT or HCPCS analogue applies, since this is not a reimbursed clinical service. The equivalent rail is the carbon registry protocol: Isometric published the first dedicated ocean alkalinity enhancement protocol in 2024 and has issued the first verified credits under it, to Planetary Technologies and to CREW Carbon. Verification is a project expense recovered from credit revenue, so the $271.68 per tonne price and the 5% materiality threshold jointly define the cost and accuracy envelope any method must satisfy.
Dual-use and public benefit. A vehicle that autonomously characterises carbonate-system anomalies also serves ocean acidification monitoring, water-quality assessment near industrial and municipal outfalls, and leak detection around sub-seabed carbon storage, all regulated monitoring obligations rather than voluntary markets.
6. Research Gap and HHA Contribution
What has not been done. No published work closes a control loop around a carbonate-system observable. The specific absence is precise rather than general: sensing is validated on an endurance AUV (Hammermeister et al., 2025), feature-following autonomy is validated on the same vehicle class for thermal and haline features (Zhang et al., 2020, 2022), learned adaptive sampling is validated in simulation and on a ground vehicle (Wu et al., 2024), and the measurement requirement is quantified (Subhas et al., 2025). Nobody has built the estimator and planner that connect them, and nobody has formulated the sampling objective in terms of the quantity a crediting protocol prices.
The precise technical gap. Three problems stand between the published results and a deployable method.
The first is an observability-rate mismatch. Total alkalinity arrives once per ten minutes; at 0.6 m s−1 the vehicle travels approximately 360 metres between samples. A controller servoing directly on alkalinity acts on information that is already spatially stale. The second is objective mis-specification. Existing adaptive-sampling formulations maximise information gain or minimise time-to-source; a verification campaign must minimise the posterior uncertainty in an integrated alkalinity anomaly under a 5% materiality constraint, which is a different objective and yields different trajectories, including trajectories that leave the patch to characterise baseline. The third is baseline non-stationarity. The Dalhousie 40-day record showed alkalinity does not relax to baseline between dosing intervals, and the WHOI campaign showed a dynamic baseline reduces alkalinity variability by 60%, so baseline estimation is part of the control problem rather than a pre-mission calibration step.
What the HHA research programme would do differently. The methodology, not merely the goal, is what distinguishes the proposed work.
We propose a two-loop architecture. The inner loop is a proxy-locked patch-following controller, adopting the MBARI isotherm-lock principle: the vehicle servos on the fast pH-salinity anomaly, while each ten-minute alkalinity measurement performs a recursive update of the estimated mapping between that anomaly and true alkalinity, implemented as a state-space filter over the proxy coefficients. The outer loop is an informative path planner over a Gaussian-process representation of the alkalinity anomaly field, warm-started with an advection-diffusion prior derived from an operational circulation model, selecting each subsequent leg to maximise expected reduction in the posterior variance of the integrated anomaly rather than in the field as a whole. Baseline sampling enters the same objective naturally, because when the in-patch anomaly approaches the detection floor, the largest available variance reduction comes from constraining the baseline.
Validation proceeds against a defined counterfactual. Every field trial flies both the adaptive policy and a matched pre-planned transect over the same traced release, so the deliverable is not a demonstration but a measured difference in verification uncertainty per unit of vehicle time. Simulation-based pre-training uses ensembles of advection-diffusion realisations conditioned on the published Subhas dispersal statistics, so the policy is exposed to the observed 14.8 km per 36 hours translation and the observed dilution trajectory before it ever enters water.
Why HHA is positioned to do this. The gap maps onto the team by component. The estimator and planner design, the simulation environment and the evaluation methodology are machine-learning systems problems and map to Haedar Hadi. The carbonate-system state estimation, the plume-dispersion prior, the uncertainty propagation into a carbon accounting statement and the experimental design for traced-release trials are physical-science and experimental-design problems and map to Hass Dhia. The instrument-production problem, which is the reason this capability cannot scale past a single campaign without intervention, maps to Ahmed.
Why the originating groups have not closed it. The reasons are structural rather than technical, and they are visible in the funding record. ARPA-E’s SEA-CO2 programme committed $36 million across eleven projects, and every one of them is a sensor or a model. Not one funded the decision layer that determines where a sensor should be. Analytical chemistry groups build instruments and do not maintain autonomy stacks; their deliverable is a characterised sensor. Ocean modelling groups consume observations rather than commissioning them; a model that requests its own data is outside the funded scope. MBARI holds the autonomy capability but is a research institution with no commercialisation mandate and no incentive to specialise a general-purpose vehicle for a single application. And the removal operators are capital-constrained companies whose engineering effort goes to the dosing system that produces the tonne, not to the instrument that proves it. The gap is an integration problem sitting in the space between four funded communities, which is exactly the class of problem a small interdisciplinary team can close and a large single-discipline group cannot.
7. Comparable Funded Projects
ARPA-E SEA-CO2 (Sensing Exports of Anthropogenic Carbon through Ocean Observation). US Department of Energy. $36 million across 11 projects in 8 states, announced 26 October 2023.
Individual awards: University of Colorado, $5,904,233, broadband-laser optical underwater sensors for dissolved carbon; Woods Hole Oceanographic Institution, $4,802,245, natural thorium decay sensors for particulate organic carbon flux, and $3,738,960, integrated system-on-chip sensor consolidating commercial sensor capability at reduced power and cost; GE Research, $4,274,658, multi-kilometre fibre-optic cables for chemical ocean carbon parameters; [C]Worthy, $3,884,825, community framework for model construction and data integration; atdepth MRV, $2,524,964, GPU-based ocean modelling; Bigelow Laboratory, $2,279,867; University of Pittsburgh, $2,274,859, buoy-based optical fibre pH and CO2 sensing; Pacific Northwest National Laboratory, $2,080,715, OAE efficacy modelling and mesocosms; University of Texas at Austin, $2,034,903, acoustic sensor networks for seagrass carbon; University of Utah, $2,004,554, micro-optical seafloor probes.
Relation to this opportunity: the portfolio funds sensing on one side and modelling on the other and funds nothing that decides where the sensing should occur, which is the strongest available evidence for the gap. A proposal targeting the decision layer is complementary to every award listed rather than competitive with any, and increases the return on sensors DOE has already financed.
NOAA Ocean Acidification Program, FY23 NOPP marine CDR awards. $24.3 million across 17 projects with partners at 47 institutions, announced 7 September 2023, of which $14.36 million came from the Inflation Reduction Act under NOAA Integrated Ocean Observing System priorities.
Named projects include Dennis McGillicuddy (Woods Hole Oceanographic Institution), “Multiscale observing system simulation experiments for iron fertilization in the Southern Ocean, Equatorial Pacific, and Northeast Pacific”, co-funded by NOAA Global Ocean Monitoring and Observing, the Ocean Acidification Program and the National Science Foundation; and Galen McKinley (Columbia University), “Data requirements for quantifying natural variability and the background ocean carbon sink in mCDR models.” Andreas Andersson (Scripps Institution of Oceanography) received $1.45 million for seaweed-farm carbon capture and ocean acidification mitigation, and Andrew Dickson (Scripps) just under $1 million for a flue-gas-to-alkalinity conversion system.
Relation to this opportunity: two named awards are observing-system design and data-requirements projects rather than instrument projects, establishing that NOAA already treats what-to-measure-and-where as a fundable problem. McGillicuddy’s observing-system simulation methodology is the natural evaluation framework for a sampling policy, and McKinley’s work defines the accuracy targets it would be optimised against. Both are collaboration targets as much as precedents.
EPA MPRSA Research Permit, LOC-NESS Wilkinson Basin Study. Woods Hole Oceanographic Institution. Permit issued April 2025 under the Marine Protection, Research, and Sanctuaries Act; docket EPA-HQ-OW-2024-0189; field trial conducted August 2025 in federal waters of Wilkinson Basin, Gulf of Maine.
The trial released 50% sodium hydroxide solution with a water tracer dye and monitored the patch with ships, autonomous gliders, long-range AUVs, drifters, satellite imagery and shore-based sensors over approximately five days, reporting strong correlation between modelled and observed dispersal. An earlier phase off Martha’s Vineyard was withdrawn following stakeholder engagement including fishing organisations.
Relation to this opportunity: this is the regulatory and operational precedent our field programme would follow, and it demonstrates the limit we propose to remove. The campaign already used long-range AUVs, so the gap is not vehicle availability; the vehicles executed operator-specified plans and the patch was located by dye. Our contribution is the layer that makes those vehicles self-tasking and, ultimately, tracer-independent.
Frontier advance market commitment to Planetary Technologies. $31.3 million for 115,211 tonnes of carbon dioxide removal delivered between 2026 and 2030, announced August 2025.
Relation to this opportunity: private rather than public, included because it sets the price per tonne against which verification cost is judged, and because Planetary holds the first registry-verified ocean alkalinity enhancement credits. It is the clearest evidence that demand is real and already contracted.
Taken together the pattern is consistent and is the argument for funding this work: federal agencies are funding the instruments and the models, private capital is funding the removal, and the observational decision layer that connects them is unfunded while 99.7% of contracted tonnes remain unissued.
8. Opportunity Assessment
Technology readiness
The integrated system is TRL 4, with a subsystem evidence chain. Autonomous carbonate sensing on a long-endurance AUV is TRL 5 to 6, validated in the relevant environment over eleven days with alkalinity residuals of 1 to 2 µmol kg−1 against 156 co-samples. Autonomous feature-following on the same vehicle class is TRL 6, on four days of continuous quasi-Lagrangian tracking plus published front tracking. Fixed-point registry-relevant alkalinity monitoring is TRL 6, on a 40-day deployment beside a live discharge. Learned adaptive plume tracing is TRL 3, simulation and ground vehicle only. The integrated carbonate-adaptive planner exists at no readiness level, which sets the system as a whole at TRL 4. TRL 5 is defined here as the planner flown on an instrumented AUV against a traced release in open water, evaluated against a matched pre-planned transect over the same event.
Technical risks, framed as research questions
Can a control loop close around an observable that updates every ten minutes? At 0.6 m s−1 the vehicle travels ~360 m between alkalinity samples. The hypothesis is that a fast proxy carries sufficient information, with alkalinity used to update the proxy relationship rather than to steer. Go/no-go at month 9: if proxy-guided tracking in simulation does not hold the vehicle within the patch for a materially greater fraction of mission time than a matched lawnmower transect over identical dispersal realisations, the architecture reverts to a multi-vehicle formation trading onboard inference for spatial coverage.
Does adaptive sampling reduce credit-relevant uncertainty, or only improve coverage? These are not the same objective, and the distinction is the scientific contribution. The measured endpoint is posterior variance of the integrated alkalinity anomaly per unit of vehicle time, not area covered.
Can verification work without a dye tracer? Every campaign to date has located the patch with rhodamine. The question is whether the carbonate signature alone, at the 10 µmol kg−1 threshold Subhas et al. established, supports patch discrimination against a dynamically estimated baseline. Trials carry dye as ground truth while withholding the dye channel from the policy, which makes the question answerable rather than arguable.
Does the instrument scale, and does its reagent budget bound the policy? Whether the hand-built microfluidic assembly holds tolerance across a production run at the required reagent-handling and optical-path precision is an open manufacturing question, addressed in parallel from month one. There is also a controls consequence: the analyser performs a wet-chemical titration and its sample budget is finite, 423 measurements across an eleven-day AUV deployment and 314 across a 40-day mooring. A greedy policy exhausts the instrument before the campaign ends, so sample budget enters the planner as a hard constraint and is a further argument for spending the expensive measurement on estimator correction rather than on steering.
Does registry acceptance of model-based quantification undercut the premise? The Isometric protocol states that quantification of atmospheric CO2 removal currently relies on biogeochemical ocean models. If protocol evolution leans further that way, demand for in-water measurement weakens. The bounding argument is that the same protocol requires quantification of model skill through data-model comparison, so observations constrain what a model may claim, and the 5% materiality threshold applies regardless of method. This is nonetheless the demand-side assumption most worth testing early, and it is scheduled as a first-quarter task alongside the novelty screen.
Regulatory pathway
This is not an FDA-regulated system and no 510(k), De Novo or PMA pathway applies. In United States waters the governing regime is the Marine Protection, Research, and Sanctuaries Act, administered through the EPA ocean dumping programme, under which the released material rather than the measurement platform is the permitted object; the LOC-NESS permit of April 2025 is the operative precedent and its year-long review sets a realistic expectation for permitting timeline on any field trial involving a release. Internationally, the London Protocol governs marine geoengineering placement. Vehicle operations fall under Coast Guard navigation rules for uncrewed systems. Any trial involving a release would be conducted in partnership with a permit-holding institution rather than on an HHA-held permit, and trials that observe an existing permitted release require no separate release authorisation.
Regulatory burden functions as a moat rather than a barrier for the verification layer specifically: the permitting record shows in-water evidence quality is what a review scrutinises and what stakeholder objections target, so better monitoring strengthens every subsequent application in a public, precedential process.
Locked versus adaptive algorithm
The distinction the FDA formalised for medical AI has a direct analogue here and is the most consequential design decision in the system. A planner that continues learning at sea produces a verification dataset whose sampling design is not the design an auditor reviewed, and under a 5% materiality threshold an unauditable sampling policy is a defect rather than a capability. The architecture we propose keeps the planning policy locked in the credit-relevant path: trained ashore, versioned, frozen for the duration of a monitored campaign, and logged such that the full decision trajectory can be replayed and justified. Adaptation is confined to parameters declared in advance with stated bounds, specifically the proxy-relationship coefficients that must update as water masses change. This mirrors the FDA Predetermined Change Control Plan framework, which specifies in advance what may change, how it will be validated and what performance envelope is acceptable. Carbon registries have no equivalent framework today, and adopting PCCP discipline voluntarily is both the fastest route through verification audit and a concrete standards-shaping contribution while these protocols are still being written.
First six months
Simulation environment built from published dispersal statistics; proxy estimator implemented against the Celtic Sea and Dalhousie datasets; the credit-relevant uncertainty objective specified against a registry protocol; first planner-versus-lawnmower comparison in simulation; first design-for-manufacturability assessment of the analyser; and collaboration discussions opened with a permit-holding institution. Section 10 gives the full milestone timeline.
9. Team Fit
Hass Dhia — Co-Principal Investigator. MS Biomedical Sciences (Wayne State University School of Medicine), medical school background including service as an anatomy teaching assistant, and an AI infrastructure architect with a physical-sciences foundation spanning chemistry, thermodynamics and fluid dynamics. Mapping to components: carbonate-system state estimation and the propagation of measurement uncertainty into a carbon accounting statement; the plume-dispersion prior that warm-starts the planner; experimental design for traced-release validation trials, including the dye-as-ground-truth-but-withheld-from-policy design that makes the tracer-independence question falsifiable; and overall system architecture for the sensing and decision stack. The methodological through-line from his training is quantitative physiological measurement under uncertainty, which is the same estimation problem this system solves in a different medium.
Haedar Hadi — Lead Principal Investigator. MS Computer Science, Boston University, Information Systems focus, with cloud and database architecture experience. Mapping to components: the machine-learning core of the outer loop, comprising the Gaussian-process field representation and the informative path planner optimised against posterior variance in the integrated anomaly; the recursive estimator that fuses continuous pH and salinity against ten-minute alkalinity; the simulation and training infrastructure for policy pre-training over advection-diffusion ensembles; and the evaluation methodology and benchmark design. The benchmark is itself a research deliverable, since no standard currently exists for scoring a verification sampling policy, and benchmark design against a defined counterfactual is the discipline his evaluation background contributes.
Ahmed — Director of Manufacturing, Key Team Member. Mapping to components: design for manufacturability of the lab-on-chip alkalinity analyser, tolerance analysis on optical path length and reagent handling, production scaling from hand-built units to a controlled run, and the quality system and calibration traceability record required for instruments whose output supports audited credits.
The lab-to-production bridge. Most research proposals end at “it works in the lab.” This proposal includes explicit DFM milestones at every phase, ensuring that prototype decisions consider production scaling, tolerance analysis, and quality systems from day one. This addresses the valley of death between TRL 4-5 prototypes and TRL 7+ deployable systems — the gap where most funded research stalls. In this programme the bridge is not a translational afterthought but a technical precondition: every alkalinity measurement in the cited literature came from a small number of hand-built microfluidic analysers, so a verification capability that cannot reproduce those instruments at tolerance cannot run more than one campaign at a time regardless of how good its autonomy is.
Capability gaps and how funding addresses them. The team does not hold a vehicle, a sea-going operations capability, or an EPA release permit, and does not propose to acquire them. Funds are budgeted for vehicle time through a partner institution operating long-range AUV assets, for a subaward to a carbonate-chemistry laboratory for reference-material analysis and sensor cross-calibration, and for observational participation in a permitted release conducted by a permit-holding institution. This positions HHA as the computational and manufacturing contributor to existing field programmes rather than as a competitor to them. A three-person team claiming it will outperform a $36 million eleven-project federal sensor portfolio on analytical chemistry would not be credible; a team supplying the decision layer that portfolio did not fund, and the manufacturable instrument its outputs require in order to leave the laboratory, is.
10. Recommended Next Steps
Target funder programmes
NOAA Ocean Acidification Program and the National Oceanographic Partnership Program, marine CDR research solicitations. The closest fit. The FY23 round funded observing-system design and data-requirements work explicitly, at $24.3 million across 17 projects, and the programme has continued to solicit in marine carbon dioxide removal research and development. Target request $1.2 million to $2.0 million over 24 to 36 months, consistent with the per-project scale of that round.
ARPA-E, SEA-CO2 follow-on or OPEN solicitation. SEA-CO2 awards ranged from $2.00 million to $5.90 million with a median near $2.5 million. The positioning is deliberate and should be stated plainly in any submission: the programme funded sensing and modelling and did not fund the tasking layer between them, and this proposal raises the return on instruments the agency has already financed. Target request $2.0 million to $3.0 million over 36 months.
NSF, National Robotics Initiative successor programmes and Ocean Technology and Interdisciplinary Coordination. The informative-path-planning and estimation contributions are fundable as robotics research on their own terms, with marine carbon as the motivating testbed. NSF co-funded two of the NOAA marine CDR awards, so the cross-directorate precedent exists. Target request $1.0 million to $1.5 million over 36 months.
DOE Office of Fossil Energy and Carbon Management, and SBIR/STTR Phase I. The measurement, monitoring and verification portfolio for carbon storage covers technically adjacent sub-seabed monitoring. An SBIR Phase I at approximately $200,000 to $300,000 is the fastest non-dilutive route to a funded simulation-stage result, with a Phase II path tied to a field trial. Carbon to Sea Initiative and Ocean Visions also run targeted ocean alkalinity enhancement solicitations covering MRV, on faster cycles than federal programmes.
Estimated funding range
Based on the comparable awards in Section 7, a credible first programme is $1.5 million to $3.0 million over 24 to 36 months. The SEA-CO2 median of approximately $2.5 million and the NOAA per-project scale of roughly $1.0 million to $1.5 million bracket the range. A staged approach is recommended: an SBIR Phase I or philanthropic grant to fund the simulation-stage result, converted into a federal proposal once the counterfactual comparison against a lawnmower baseline has been measured.
Proposed 24-month milestone timeline
- M1–3 R&D: Simulation environment built from published dispersal statistics (14.8 km per 36 h translation, observed dilution trajectory). Proxy estimator implemented against Celtic Sea and Dalhousie datasets. Credit-relevant uncertainty objective specified against a registry protocol.
- M1–3 Manufacturing (Ahmed, parallel track from month one): Design-for-manufacturability assessment of the lab-on-chip alkalinity analyser. Tolerance budget for optical path length and reagent metering. Bill of materials and supplier assessment.
- M4–9 R&D: Planner trained and evaluated in simulation against matched lawnmower baselines over identical dispersal realisations. Go/no-go at M9: proxy-guided tracking must hold the vehicle in-patch for a materially greater fraction of mission time than the matched baseline, or the architecture reverts to a multi-vehicle formation.
- M6–12 Regulatory and partnership (parallel track): Collaboration agreement with a permit-holding institution for observational access to a planned permitted release. Data-sharing and publication terms agreed. No independent release permit sought.
- M9–15 R&D: Hardware-in-the-loop integration on a partner vehicle. Tracer-independence evaluation with dye present as ground truth and withheld from the policy input. Onboard compute and power budget validated against the 60 W hotel-load envelope.
- M10–18 Manufacturing (Ahmed): Pilot production run of analyser units at controlled tolerance. Calibration and traceability record designed to audit standard. Process documentation and inspection plan.
- M15–21 R&D: Field trial against a traced release, flying the adaptive policy and a matched pre-planned transect over the same event. Primary endpoint: posterior variance of the integrated alkalinity anomaly per unit vehicle time, adaptive versus baseline.
- M18–24 Regulatory and standards (parallel track): Locked-policy audit package with full decision-trajectory replay, structured on the PCCP model. Submission to a registry as a proposed MRV method annex, and results published with the simulation environment and benchmark released openly.
- M21–24 Programme: Phase 2 proposal for multi-vehicle campaigns and extended-duration monitoring. TRL 5 declared on the field-trial evidence, or the gap documented and the follow-on scoped against it.