PROJECT TAWQ طوق

In shortA ring of bubbles far out at sea holds the jellyfish back, the air goes where the tide pushes water in, and robot boats carry the held jellyfish out to sea.

1. Warning, air onThe early-warning notification switches the ring on before the bloom arrives.
2. HoldRising bubbles from four seabed pipes turn most drifting jellyfish back; a float line with a mesh skirt keeps them in the band.
3. Air follows the tide240 sections and 60 air valves: boosted where water pushes in, off where it flows out. If control is lost, the valves open and every section gets normal air.
4. LiftRobot catamarans lift the held jellyfish, in water, with a slow Archimedes screw into the live well. The screw stops if a turtle, dugong or fish is within 5 m. Designed for live handling; survival rates are measured in the Phase 0 trials.
5. ReleaseThey carry them 8 km offshore and release them; in our model none drift back within 21 days.

In shortThree linked computer models (the water around the site, the bubble curtain, and a 3‑day bloom) turn published site data and our stated assumptions into every result on this page.

In shortEvery number we quote, with how it was calculated and where each input comes from.

3.1 The site

Intake flow: 400 m³/s

The plant draws 400 m³/s of seawater for cooling; we also test 690 m³/s as a conservative bound.Show the calculationHide the calculation

flow = number of units × condenser flow per unit

Input	Value	Source
Condenser cooling water per unit	6,000 m³/min	Published ENEC 2017
Reactor units	4	Published Power Engineering 2021
Design maximum per unit (bound)	172.5 m³/s	Published NEI 2012
Flows used in the models	400 and 690 m³/s	Model output master_params.json

4 × 6,000 m³/min ÷ 60 s/min = 400 m³/s

Bound: 4 × 172.5 m³/s = 690 m³/s

This is the condenser cooling flow. The safety-related essential service water system takes Gulf seawater through its own “essential service water intake structure” (FANR 2012), so it is not part of this flow.

Water speed at the intake mouth: about 0.38 m/s

About 0.38 m/s (estimated: channel width fitted to NMDC's published dredged area). That is about three times the best holding limit of a bubble curtain, about 0.13 m/s (2.8–3.0×), so a curtain at the mouth would hold nothing.Show the calculationHide the calculation

speed = flow ÷ mouth area; mouth area = channel width × depth; ratio = speed ÷ best holding limit

Input	Value	Source
Intake flow	400 m³/s	Published ENEC 2017
Channel width: one average width for the three dredged channels	162 m	Model output final_numbers.json
Depth at the breakwater heads (breakwater toe)	6.5 m	Our assumption inside NMDC's published channel depths of −5.5 to −9.5 m
Total dredged area of the three channels	2.35 km²	Published NMDC
Length of the three channels	14.50 km	Our assumption our estimate from the seabed profile (ETOPO 2022, not survey data); NMDC publishes no channel widths or lengths
Best holding limit (boost, deepest section)	0.126 m/s	Model output system.json

Width fit: 2.35 km² ÷ 14.50 km = 162 m

400 ÷ (162 × 6.5) = 400 ÷ 1,053 m² = 0.380 m/s → about 0.38 m/s

0.380 ÷ 0.126 = 3.0 times

NMDC publishes only the total dredged area, not channel widths, so the width is a fit. The same fit puts the intake channel at 7.0 m deep; with that depth the speed is 0.35 m/s and the ratio 2.8. Either way the mouth is far above what a curtain can hold.

The intake's pull at the ring: about 1.3 cm/s

About 1.3 cm/s at 1.25 km out: small next to the tide, so the tide decides which way the water moves at the ring.Show the calculationHide the calculation

pull at the ring = inward speed caused by the intake alone at the ring radius (site flow model)

Input	Value	Source
Intake's pull at the ring	1.32 cm/s	Model output final_numbers.json
Tide speed at the ring, typical (root mean square)	4.6 cm/s	Model output final_numbers.json
Tide speed at the ring, peak	11.7 cm/s	Model output final_numbers.json

1.32 cm/s from the intake, against a typical 4.6 cm/s and a peak 11.7 cm/s from the tide.

3.2 Holding the bloom

Holding limits and the ×0.85 / ×0.76 factors

The curtain holds against inward water up to 0.081–0.112 m/s with normal air and 0.085–0.126 m/s with boost, depending on depth. These limits include factors of ×0.85 (normal air) and ×0.76 (boost) for a mesh skirt 10 % blocked and a 12‑hour hold.Show the calculationHide the calculation

holding limit = curtain-model limit (clean, short hold) × factor; factor = mesh effect × hold-time effect

Input	Value	Source
Limit, clean curtain, 7 m deep, normal air 0.24 Nm³/min per m, 3.6 h hold	0.1185 m/s	Model output slice_A.json
Same with the mesh skirt 10 % blocked	0.1071 m/s	Model output slice_E.json
Mesh skirt, 3.6 h and 12 h hold	0.1071 → 0.1010 m/s	Model output slice_I.json
Mesh skirt, boost air 0.6 Nm³/min per m, 3.6 h	0.1333 m/s	Model output slice_H.json
Boost limit before the corrections, 7 m deep	0.1664 m/s	Model output system.json
Mesh skirt blocked	10 %	Our assumption fouling rate unknown; measured at the pilot section (step 2)
Hold-time basis	12 h	Our assumption held jellyfish can stay at the curtain for many hours

Normal air: mesh 0.1071 ÷ 0.1185 = 0.904; hold 0.1010 ÷ 0.1071 = 0.943; factor 0.904 × 0.943 = 0.852 → ×0.85

Boost: mesh 0.1333 ÷ 0.1664 = 0.801; factor 0.801 × 0.943 = 0.756 → ×0.76

Limits along the ring (shallowest to deepest section): normal 0.095–0.132 × 0.85 = 0.081–0.112 m/s; boost 0.112–0.166 × 0.76 = 0.085–0.126 m/s

The two factors were typed into the model as 0.85 and 0.76; they reproduce from the curtain-model runs above. With the mesh 40 % blocked the curtain model holds nothing (limit 0.00 m/s), so the mesh must be kept clean.

3.3 Air that follows the tide

• Off: water flowing out 18 %
• Normal air 76 %
• Boost: water pushing in hard 6 %

Share of section-time over a month of tides, all 240 sections, from the air-control model.

Air time shares: off 18 %, normal 76 %, boost 6 %

The controller reads the water 10 minutes ahead and sets each of the 240 sections to off, normal or boost. Over a month of tides the air is off 18 % of the time, normal 76 % and boosted 6 %.Show the calculationHide the calculation

for each section, every 5 minutes: OFF if the forecast inward speed < −0.02 m/s; BOOST if it is above 0.8 × that section's normal-air holding limit; otherwise NORMAL. Boosts are dropped, weakest first, if the total power would exceed the cap of 14.63 MW.

Input	Value	Source
Share of section-time off / normal / boost	18.3 / 76.0 / 5.7 %	Model output system.json
Control thresholds and the power cap	−0.02 m/s; 0.8 × limit; 14.63 MW	Our assumption design choice
Air per metre of pipe: normal (deeper / shallower than 5.5 m), boost	0.24 / 0.4; 0.6 Nm³/min	Our assumption design choice

18.3 % + 76.0 % + 5.7 % = all section-time

Fail-safe: the valves fail open, so a section that loses control gets normal air.

Held per crossing: 86.9 % against 80.4 %

With tide-aware air, 86.9 % of the jellyfish drifting in across the ring meet a section that can hold them at that crossing, against 80.4 % with fixed air. Before any robots, 6.9 % of a design bloom reaches the intake with tide-aware air, against 8.8 % with fixed air.Show the calculationHide the calculation

held share = inward water flux crossing where the speed is below the section's holding limit ÷ all inward flux across the ring. Jellyfish drift with the water, so this is the share of arriving jellyfish held per crossing.

Input	Value	Source
Held per crossing, tide-aware / fixed air	86.9 / 80.4 %	Model output system.json
Share of the bloom reaching the intake, tide-aware / fixed (central case)	6.89 / 8.80 %	Model output summary.json

Held per crossing explains the mechanism only; the share of the bloom reaching the intake is the result that counts.

Air power: 12 % less than fixed air

Tide-aware air uses 12 % less power: 12.8 MW on average against 14.6 MW for fixed air (design basis).Show the calculationHide the calculation

saving = 1 − tide-aware mean power ÷ fixed-air power

Input	Value	Source
Tide-aware air, mean power	12.82 MW	Model output system.json
Fixed air (every section at normal air)	14.63 MW	Model output system.json

1 − 12.82 ÷ 14.63 = 12.4 % → 12 % less

Fixed air runs every section at normal air all the time. That equals the cap the tide-aware controller never exceeds, so the comparison is like with like.

Running power: 14 MW, 0.25 % of the plant's output

The ring draws 14 MW on average while it is on: 0.25 % of the plant's 5,600 MW.Show the calculationHide the calculation

running power = air power × hot-inlet factor × aftercooler factor + main-pipe losses + aftercooler pumps

Input	Value	Source
Tide-aware air power	12.82 MW	Model output system.json
Hot-inlet factor	× 1.02	Our assumption allowance for hot summer air at the blower inlet
Aftercooler pressure-drop factor	× 1.023	Model output aftercooler.json (2.28 % extra power, rounded)
Main-pipe losses, deep / shallow	0.328 / 0.175 MW	Model output air_supply.json
Aftercooler pumps	0.171 MW	Model output aftercooler.json
Plant output, four units	5,600 MW	Published Power Engineering 2021

12.82 × 1.02 × 1.023 + 0.328 + 0.175 + 0.171 = 14.0 MW

14.0 ÷ 5,600 = 0.251 % → 0.25 %

Energy: about 337 MWh a day for the air (345 with robot charging).

Blowers and aftercoolers

44 turbo blowers of 400 kW: 17.6 MW installed, including 2 spares. Seawater aftercoolers cool the air from up to 133 °C to 40 °C for the plastic pipe; their cooling water returns 5 °C warmer, 0.85 m³/s.Show the calculationHide the calculation

installed power = number of blowers × unit size

Input	Value	Source
Blowers: deep only / shallow only / swing / spare	26 / 12 / 4 / 2	Model output cost2_r12.json
Peak demand, including losses	16.8 MW	Model output cost2_r12.json
Air leaving the blowers in summer (45 °C inlet)	133 °C	Model output aftercooler.json
Heat removed; seawater flow	17.4 MW; 0.85 m³/s	Model output aftercooler.json

44 × 400 kW = 17.6 MW

3.4 What reaches the intake

No ring (do nothing)

15.3 %

Net or curtain at the mouth same load: moved, not removed

15.3 %

Ring, fixed air

8.8 %

Ring, tide-aware air

6.9 %

TAWQ: ring + 12 robots

3.8 %

0510152025 %

Share of a 3‑day design bloom reaching the intake. Bar: central case. Whisker: range over all four current cases. 24 tide starts × 12,000 virtual jellyfish. The mouth bar is set equal to doing nothing by argument, not by a model run.

LimitThe ring cuts the total load, not the worst hour: in the central case the average worst hour is about the same (12.1 vs 12.2 t/h) and the single worst hour about doubles (36.1 vs 18.1 t/h). In 17 of 96 tide-start cases the worst hour is above doing nothing.

75 % less reaches the intake; 73 → 18 t/day

With the ring and 12 robots, 3.8 % of a 3‑day design bloom reaches the intake, against 15.3 % with no ring: 75 % less in the central case (about ±8 points, 2 standard errors over 24 tide starts). The load on the screens falls from 73 to 18 t/day.Show the calculationHide the calculation

reduction = 1 − (intake with the ring) ÷ (intake with no ring), both summed over the 24 tide starts

Input	Value	Source
Share reaching the intake, no ring	15.33 %	Model output summary.json
Share reaching the intake, ring + 12 robots	3.79 %	Model output summary.json
Load on the screens, no ring / ring + 12 robots	72.9 / 18.0 t/day	Model output summary.json
Design bloom	5 g/m³: 1,426 t within 4 km (the deck rounds it to about 1,400 t)	Our assumption no published Gulf biomass density was found
Sampling	24 starts × 12,000 jellyfish	Our assumption design choice

1 − 3.79 ÷ 15.33 = 75.3 % → 75 %

72.9 → 18.0 t/day → 73 → 18 t/day

The same jellyfish start positions are used for every configuration in each tide start, so the comparison is paired.

Every configuration and current case

Under steady currents the ring + 12 robots cut the share reaching the intake by 71–97 %; the table shows all 7 configurations in all four current cases.Show the calculationHide the calculation

Swipe sideways to see the whole table.

Configuration	Central	Steady 0	Steady +0.05	Steady −0.05
No ring	15.3 %	21.7 %	12.6 %	12.1 %
Ring, fixed air	8.8 % (43 %)	2.7 % (87 %)	5.2 % (59 %)	7.7 % (36 %)
Ring, tide-aware air	6.9 % (55 %)	1.0 % (95 %)	4.2 % (67 %)	6.7 % (44 %)
Ring + 6 robots	5.1 % (67 %)	0.7 % (97 %)	3.7 % (71 %)	4.4 % (64 %)
Ring + 8 robots	4.5 % (71 %)	0.7 % (97 %)	3.6 % (72 %)	3.9 % (67 %)
Ring + 10 robots	4.2 % (73 %)	0.7 % (97 %)	3.5 % (72 %)	3.8 % (69 %)
Ring + 12 robots	3.8 % (75 %)	0.7 % (97 %)	3.5 % (72 %)	3.5 % (71 %)

Share of the design bloom reaching the intake; in brackets, the reduction against no ring. Central = randomly varying current; steady currents in m/s. Model output summary.json

Ring + 12 robots under steady currents of 0, +0.05 and −0.05 m/s: 96.9 %, 72.2 % and 70.9 % → 71–97 %

Central case (randomly varying current)

75 %

Central case, 16 current meters + forecast

75 %

Steady current 0 m/s

97 %

Steady current +0.05 m/s

72 %

Steady current −0.05 m/s

71 %

Intake at its 690 m³/s design maximum

72 %

Steady 10 m/s offshore wind

86 %

Steady 10 m/s Shamal wind

56 %

Dense bloom, 30 g/m³ worst hour about doubles

60 %

0255075100 %

Reduction against no ring. Central case: about ±8 points (2 standard errors over 24 tide starts). All cases except the steady currents use the central (randomly varying) current. The meters case replaces the model's full knowledge of the current with current meters and a tide forecast.

Bound cases: Shamal wind, offshore wind, 690 m³/s

In a steady 10 m/s Shamal wind the cut is 56 %; with a steady offshore wind 86 %; with the intake at its design maximum of 690 m³/s, 72 %. At 690 m³/s the worst hour is the weak point: it is above doing nothing in 13 of 24 tide starts (mean worst hour 27.3 vs 18.3 t/h).Show the calculationHide the calculation

Input	Value	Source
Reduction, Shamal wind (central current)	55.5 %	Model output summary_wind_se.json
Reduction, offshore wind	86 %	Model output summary_wind_n.json
Reduction, 690 m³/s	72 %	Model output summary_q690.json
Shamal case: a steady wind driving surface water toward the coast	10 m/s	Our assumption a test case; real Shamals are stronger
Real Shamal winds	15–20 m/s	Published Thoppil & Hogan 2010

Real Shamals typically reach 15–20 m/s and last 24–36 hours or 3–5 days (Thoppil & Hogan 2010); they are not modelled, so the real effect may be stronger. Tide starts in which the worst hour with the ring + 12 robots exceeds no ring's (central current case): Shamal wind 0 of 24, offshore wind 3 of 24, 690 m³/s 13 of 24. The bound cases use the same full-knowledge controller as the central case.

Dense bloom: 60 % less, but the worst hour about doubles

In a dense bloom (30 g/m³) the daily load still falls 60 %, from 437 to 173 t/day, but the worst hour about doubles: 148 against 73 t/h. Very dense blooms exceed the robots' capacity.Show the calculationHide the calculation

Input	Value	Source
Dense bloom	30 g/m³: 8,556 t within 4 km	Our assumption a stress test, six times the design bloom
Load, no ring / ring + 12 robots	437 / 173 t/day	Model output summary.json
Worst hour (mean over starts), ring / no ring	148 / 73 t/h	Model output summary.json
Fleet removal rate in a dense bloom	9.4 t/h	Model output fleet_rates.json

1 − 173 ÷ 437 = 60.5 % → 60 %; 148 ÷ 73 = 2.0 times

Sampling noise: about ±8 points

About ±8 points on the 75 % reduction and ±1.3 points on the 3.8 % share (2 standard errors over 24 tide starts). Differences between fleet sizes are much smaller, because every fleet sees the same starts.Show the calculationHide the calculation

share: 2 SE = 2 × standard deviation of the 24 per-start shares ÷ √24; reduction: 2 SE by the jackknife, recomputing the reduction with one start left out at a time, SE = √((n − 1)/n × Σ(rᵢ − r̄)²)

Input	Value	Source
Intake per tide start, no ring and ring + 12 robots (central case)	24 values each	Model output rand_s00–s23.json

Share: ±1.32 → ±1.3 points; reduction: ±7.9 → ±8 points

Paired difference, 12 against 8 robots: +4.4 ± 1.5 points

The noise on the reduction is large because some tide starts send far more of the bloom toward the intake than others. It does not change which fleet is best, because fleets are compared start by start.

Central

7 of 24

Steady 0

0 of 24

Steady +0.05

2 of 24

Steady −0.05

8 of 24

Each square is one tide start, 24 per current case. Filled: with the ring + 12 robots the worst hour exceeds the no-ring worst hour by more than 2 Poisson standard deviations.

Worst-hour disclosure: 17 of 96

In 17 of 96 tide-start cases the single worst hour with the ring is higher than with no ring: a strong tide pushes a held pile through at once. This is our next design iteration.Show the calculationHide the calculation

a case counts if: worst hour with the ring > worst hour with no ring + 2 × √(worst hour with the ring ÷ m) × m, where m is the mass of one virtual jellyfish (2 Poisson standard deviations)

Input	Value	Source
Hourly loads per tide start, ring + 12 robots and no ring	96 cases	Model output rand_, 0_, 0.05_ and -0.05_s00–s23.json
Mass of one virtual jellyfish, design bloom	119 kg	Model output rand_s00.json
Cases counted: Central / Steady 0 / Steady +0.05 / Steady −0.05	7 / 0 / 2 / 8	Model output summary.json
Worst hour, central case: mean / maximum, ring	12.1 / 36.1 t/h	Model output summary.json
Same, no ring	12.2 / 18.1 t/h	Model output summary.json

7 + 0 + 2 + 8 = 17 of 96

The test uses the scatter of the ring's own hourly count; a test on the difference would flag fewer cases (not recomputed).

If the air fails: an estimated 53–72 t

If the air fails when the ring holds the most jellyfish, an estimated 53–72 t reaches the intake over several hours.Show the calculationHide the calculation

pulse = share of released jellyfish that reach the intake × largest held stock

Input	Value	Source
Share reaching the intake; median time	36.2 %; 4.7 h	Model output recapture.json
Largest held stock over all starts, by current case	145.3–200.0 t	Model output summary.json

36.2 % × 145.3–200.0 t = 52.6–72.4 t → an estimated 53–72 t

The 36.2 % comes from animals that leaked through the curtain at high inflow, used as a proxy for a whole held band released at once; it is not a simulated band release.

3.5 Robots

12 robots, about 10 at sea

Each robot works about 7.56 h per charge and then charges for 1.54 h, so it is at sea 0.83 of the time: 12 robots keep about 10 at sea.Show the calculationHide the calculation

working time = trips per charge × trip cycle; charging time = trips per charge × energy per trip ÷ (charger efficiency × charger power) + berthing; availability = working time ÷ (working time + charging time); robots at sea = availability × fleet, rounded

Input	Value	Source
Energy per trip cycle	77.5 kWh	Model output robot_sizing.json
Battery; usable share	300 kWh; 80 %	Our assumption design choice
Trips per charge	3	Model output robot_sizing.json
Trip cycle: collecting + travelling and release	31 + 120 min	Model output robot_sizing.json
Charger; charger efficiency; berthing time	200 kW; 0.9; 0.25 h	Our assumption design choice

Working: 3 × 151 min = 7.56 h; charging: 3 × 77.5 kWh ÷ (0.9 × 200 kW) + 0.25 h = 1.29 + 0.25 = 1.54 h

7.56 ÷ (7.56 + 1.54) = 0.831 → 0.83

0.831 × 12 = 10.0 → 10 at sea

The screw: 2.2 m, 21 rpm, 1 m³/s

A 2.2 m Archimedes screw turning at about 21 rpm lifts 1 m³/s of water, with the jellyfish, into the live well: a slow lift. The screw stops if a turtle, dugong or fish is within 5 m. Designed for live handling; survival rates are measured in the Phase 0 trials.Show the calculationHide the calculation

Input	Value	Source
Diameter; core; flights; pitch	2.2; 1.1; 3; 2.2 m	Our assumption design choice
Incline; length	30°; 4.2 m	Our assumption design choice
Speed needed; limit for this diameter	21.3; 29.6 rpm	Model output phase2_Q400.json
Tip speed; shaft power	2.45 m/s; 17.2 kW	Model output phase2_Q400.json
Flow; lift	1 m³/s; 1.2 m	Our assumption design choice
Intake mouth diameter at 0.5 m/s	1.6 m	Model output robot_sizing.json
Suction speed 5 m from the mouth	0.6 cm/s	Model output robot_sizing.json
Wildlife stop distance (camera and sonar)	5 m	Our assumption design choice; tested at sea in step 2

21.3 rpm, below the 29.6 rpm limit → about 21 rpm

Trip and fleet rate: about 1.9 t, 55 min each way, 7.2 t/h

Each trip carries about 1.9 t, 55 min each way at 2.5 m/s; in a design bloom the fleet removes 7.2 t/h.Show the calculationHide the calculation

leg time = distance ÷ speed, rounded up to whole 5 minutes; fleet rate = robots at sea × rate per robot in the fleet model

Input	Value	Source
Mean distance to the release zone	8.03 km	Model output release_zone_q400.json
Transit speed	2.5 m/s	Our assumption design choice
Trip-sizing stage: load per trip; trip cycle; rate per robot	1.94 t; 151.2 min; 0.78 t/h	Model output robot_sizing.json
Fleet model (12 starts, band model): rate per robot at sea; mean load per trip	0.724 t/h; 1.90 t	Model output fleet_rates.json

8.03 km ÷ 2.5 m/s = 53.5 min → 55 min

10 × 0.724 t/h = 7.24 t/h → 7.2 t/h

The two robot numbers come from different model stages: the earlier trip-sizing stage gives about 1.9 t per 151-minute cycle; the fleet model gives the rate per robot used for the whole fleet. Neither is derived from the other.

The robot hull

A battery catamaran: two hulls 12 m long, each 1.4 m wide with a 0.9 m draft, 6.2 m wide overall.Show the calculationHide the calculation

Input	Value	Source
Hull length; hull width; draft	12; 1.4; 0.9 m	Our assumption design choice
Overall width	6.2 m	Our assumption design choice: leaves a gap between the hulls for the live well and the screw tube

Gap between the hulls: 6.2 − 2 × 1.4 = 3.4 m

3.6 Choosing the fleet

Choosing the fleet: 12 robots, by a fixed decision rule

The decision rule is written into the aggregation code and applied the same way to every fleet. All five fleets are within the 10 % cost band; the 70 % floor leaves 8, 10 and 12 robots tied; the worst-hour tie-break picks 12 (17 cases, against 19 and 21).Show the calculationHide the calculation

1. Metric. For each fleet: annualised cost (middle of the low and high estimates) ÷ tonnes kept out of the intake per design bloom, central case.
2. Tie band. The best is $147k per tonne (ring + 8 robots). Every fleet within 10 % of it, up to $162k per tonne, is tied: all five are.
3. Floor. Keep only fleets with a central reduction of at least 70 %: the ring alone (55.0 %) and 6 robots (67.0 %) drop out.
4. Tie-break. Fewest tide-start cases, over all four current cases, whose worst hour exceeds no ring's by more than 2 Poisson standard deviations: 8 robots 21, 10 robots 19, 12 robots 17. 12 robots.
5. Second tie-break (not needed): smallest failure pulse.

Swipe sideways to see the whole table.

Fleet	Annualised, mid ($M/yr)	Kept out per bloom (t)	$k per tonne	Within 10 %?	Reduction	≥ 70 %?	Worst-hour cases	Failure pulse, central case (t)	Result
Ring, tide-aware air	18.8	120.3	156	yes	55.0 %	no	37	115	below the floor
Ring + 6 robots	21.8	146.5	149	yes	67.0 %	no	24	83	below the floor
Ring + 8 robots	22.8	154.8	147	yes	70.8 %	yes	21	77	tied
Ring + 10 robots	23.7	158.7	150	yes	72.6 %	yes	19	74	tied
Ring + 12 robots	24.7	164.5	150	yes	75.3 %	yes	17	69	picked

Worst-hour cases are counted over all 96 tide-start cases. Reduction = central case. The failure pulse column is the central case: for 12 robots, 69 t is the central case of the 53–72 t failure pulse in 3.4.

Cost per tonne and worst-hour counts: Model output summary.json. Annualised cost: see 3.9. 8 robots are the cheapest per tonne at both the low and the high cost estimate; the tie-break on peak hours picks 12. One model setting, a cap on how fast spare robot capacity removes held jellyfish, does not scale with fleet size; it could shift how 8, 10 and 12 robots compare and is not yet quantified.

3.7 The 8 km release

Release 8 km offshore: none back within 21 days

Released 8 km offshore, none drift back to the ring within 21 days in our model (at 400 m³/s). Released animals that leave the 16 km model area are not tracked further: none leave with no background current (0 %), 83 % leave with the random current and 100 % with a steady ±0.05 m/s current.Show the calculationHide the calculation

share back = released jellyfish that re-enter the ring within 21 days ÷ all released

Swipe sideways to see the whole table.

Release distance	No current	Steady +0.05 m/s	Steady −0.05 m/s	Random current	690 m³/s, no current	Leg time
5 km	100 %	0 % (100 % left)	0 % (100 % left)	0 % (83.3 % left)	100 %	35 min
8 km	0 %	0 % (100 % left)	0 % (100 % left)	0 % (83.3 % left)	13.8 %	55 min
10 km	0 %	0 % (100 % left)	0 % (100 % left)	0 % (83.3 % left)	0 %	70 min

Share back at the ring within 21 days; in brackets, the share that left the 16 km model area and was not tracked further (sources: Model output release_zone_q400.json and Model output release_zone_q690.json). With no background current none leave the area and none return from 8 km. At 5 km with no current, all return by day 21 (median 14 days). At 690 m³/s with no current, 13.8 % return from 8 km. Each case: 12 release times × 200 virtual jellyfish.

3.8 Sensing

Sensing: with 16 current meters the result is still 75 %

In the model the controller knows the current at every section. With 16 current meters and a tide forecast instead, the result is 75 % (74.96 % against 75.27 %, same tide starts).Show the calculationHide the calculation

16 meters spread evenly along the ring read the inward speed with 0.01 m/s noise; a tide-and-intake prediction fills in between them. Same seeds and tide starts as the main runs; results compared start by start.

Input	Value	Source
Intake per tide start, with the meters	24 values	Model output _s16rand_s00–s23.json
Intake per tide start, full knowledge	24 values	Model output rand_s00–s23.json

Reduction 74.96 % against 75.27 %: −0.32 ± 0.53 points (paired 2 SE); 8 starts worse, 10 better, 6 equal

Share reaching 3.84 % (against 3.79 %); 18.2 t/day (against 18.0); tide-aware air alone 54.3 % (against 55.0 %)

The bound cases (Shamal, offshore wind, design-maximum flow) use the same full-knowledge controller.

3.9 Cost

Build cost: $102–252M

$102–252M to build, a class-5 early estimate (±50 %).Show the calculationHide the calculation

CAPEX = direct cost × (1 + engineering and management) × (1 + contingency)

Line	Quantity	$M, low–high
Blowers, 400 kW turbo, one plant on both arms	44 units, 17.6 MW	26.4–52.8
Seawater aftercoolers to 40 °C, pumps	17.4 MW heat	1.6–7.5
Deep main DN1600, ballast SF 1.5	7,692 m	15.2–34.6
Shallow main DN1200, ballast SF 1.5	2,519 m	3.5–8.6
Channel crossing DN1600, buried	250 m	0.7–1.7
Diffusers 4 × DN110	30,767 m	1.0–3.7
60 air-piloted pinch valves DN400, float-line control boxes, DN300 sub-header	60	1.3–3.3
Float line with 1.5 m mesh skirt (passes ≥ 90 % of flow)	7,656 m	1.1–4.6
Moorings, double legs every 50 m	308	1.5–4.6
Current meters (ADCP)	16	0.6–1.3
SCADA, comms, software	1 lot	1.0–3.0
Relocation robots	12	12.0–30.0
Docks, 6 berths each	2	2.4–6.0
Direct cost		68.5–161.6
Engineering and management (EPCM), 15–20 %		10.3–32.3
Contingency, 30 %		23.6–58.2
Build cost (CAPEX)		102.3–252.2

Input	Value	Source
Line items and quantities	13 lines	Model output cost2_r12.json
Unit prices	low–high	Our assumption class-5 unit prices, not verified against quotes

Low: 68.5 × 1.15 × 1.30 = 102.3; high: 161.6 × 1.20 × 1.30 = 252.2 ($M)

→ $102–252M

Running cost: $2.6–13.2M a year

$2.6–13.2M a year: energy for 30–90 days a year on, plus maintenance.Show the calculationHide the calculation

running cost = energy per day × days on × electricity price + maintenance (share of CAPEX)

Input	Value	Source
Energy per day, air and robot charging	345 MWh	Model output cost2_r12.json
Days on per year (low / high)	30–90	Our assumption the ring runs only in the bloom season
Electricity price (low / high)	$54–100/MWh	Published ADDC 2025
Maintenance (low / high)	2–4 % of CAPEX a year	Our assumption planning allowance, not a quote

Low: 344.5 MWh × 30 days × $54/MWh = $0.56M, plus 2 % × $102.3M = $2.05M: total $2.60M

High: 344.5 MWh × 90 days × $100/MWh = $3.10M, plus 4 % × $252.2M = $10.09M: total $13.19M

→ $2.6–13.2M a year

Annualised cost: $12.3–37.2M a year

$12.3–37.2M a year: the build cost spread over the equipment's life at 8 % interest, plus the running cost.Show the calculationHide the calculation

CRF(i, n) = i ÷ (1 − (1 + i)⁻ⁿ); annualised = (CAPEX − robots and docks) × CRF(8 %, 30 yr) + robots and docks × CRF(8 %, 15 yr) + running cost

Input	Value	Source
Interest rate	8 %	Our assumption cost of capital for a utility project
Life: ring and air plant / robots and docks	30 / 15 years	Our assumption robots and docks wear out sooner
Robots and docks, with engineering and contingency	$21.53–56.16M	Model output cost2_r12.json
CAPEX; running cost	$102.3–252.2M; $2.60–13.19M/yr	Model output cost2_r12.json

CRF(8 %, 30) = 0.0888; CRF(8 %, 15) = 0.1168

Low: 80.81 × 0.0888 + 21.53 × 0.1168 + 2.60 = 12.30

High: 196.01 × 0.0888 + 56.16 × 0.1168 + 13.19 = 37.16

→ $12.3–37.2M a year

Break-even: 4–20 lost unit-days a year

TAWQ pays for itself if it avoids 4–20 lost unit-days a year (one reactor unit offline for one day), with electricity valued at $54–100/MWh: Abu Dhabi business tariffs used as a proxy, because Barakah's sale price is not public.Show the calculationHide the calculation

value of one unit-day = unit size × 24 h × price; break-even = annualised cost ÷ value of one unit-day

Input	Value	Source
Unit size	1,400 MW	Published Power Engineering 2021
ADDC business tariffs, 2025	20–36.6 fils/kWh	Published ADDC 2025
Dirham to US dollar	3.6725 AED per US$	Published CBUAE 2014
Annualised cost (low / high)	$12.30–37.16M	Model output summary.json

Price: 20–36.6 fils/kWh = 200–366 AED/MWh; ÷ 3.6725 = $54.46–99.66/MWh → $54–100/MWh

One unit-day: 1,400 MW × 24 h × $54–100/MWh = $1.81–3.36M

Fewest: $12.30M ÷ $3.36M = 3.7; most: $37.16M ÷ $1.81M = 20.5

3.7–20.5 → 4–20 lost unit-days a year

For scale: one bloom at Gravelines took four units offline at once. We do not claim a number of avoided shutdowns.

3.10 Phase 0 sample size

Phase 0 sample size: 100 animals per arm

100 animals per arm are enough to tell 90 % survival from 75 %.Show the calculationHide the calculation

n = (z_1−α/2 √(2 p̄ (1 − p̄)) + z_1−β √(p₁(1 − p₁) + p₂(1 − p₂)))² ÷ (p₁ − p₂)²

Input	Value	Source
Pass mark p₁; rate to rule out p₂	90 %; 75 %	Our assumption the pass gate and the lowest survival we must be able to detect
Two-sided α; power	0.05; 80 %	Our assumption standard choices
Stored sample size	100	Model output phase2_spec.json

n = 99.5 → 100 per arm

In shortWhat the results rest on: which parts are published, which our models computed, which we assumed, and how each will be tested if we are shortlisted.

If shortlisted: three gated steps

1. Phase 0 survival trials (bubbles, mesh, screw, full chain; 100 animals per arm). (indicative: one bloom season, our assumption, since live jellyfish are needed)

   Gate≥ 90 % full-chain survival

   Animals held at the curtain are scored too: bubbles can harm medusae larger than 3 cm (Raskoff et al. 2003), and the field trial of a bubble curtain did not measure injury (Haberlin et al. 2021).

2. 100 m pilot section + 1 robot at sea (holding, sliding, fouling, capture rate, wildlife stop). (indicative: one bloom season)

   Gateholding ≥ modelled limit; mesh ≤ 10 % blocked

3. Full ring + 12 robots, with mooring and storm loads designed.

   GateFANR and EAD approvals

Phase 0 is step 1, the survival trials.

Assumption register

The robots are designed for live handling; survival rates are measured in the Phase 0 trials.

In shortWe tried the obvious answers first; each one fails for a reason our models or simple physics show.

In shortWe re-ran the models from scratch and traced every published number back to the file it came from: the results reproduce.

The convergence check, stated plainly

Before the runs we set a check: the total intake over the first 12 tide starts should agree with all 24 within 5 %. It passes in the central case (no ring +4.8 %, ring + 12 robots +3.4 %). In two steady-current bound cases for 12 robots it does not: −14.9 % at 0 m/s, on a small 0.68 % share, and +5.7 % at −0.05 m/s. Those two steady-current results are less settled than the central case.

This page itself is built by a script from the final result files, and an automatic check compares each number and its rounding with our proposal deck.

In shortThe published sources we used, with the sentence each one supports and working links.

1. ENEC (2026). Challenge statement, Advanced Technology Pioneers 2026, on the ChallengeON platform.

   https://challengeon.atrc.ae/en/challenges/atp2026/pages/enec-challenge-statement?lang=en

   Supports: The problem statement, the scope (after an early-warning notification), the focus species, the data rule and the AI-use rule. Accessed 10 Oct 2026.

   “Problem Statement: HOW MIGHT WE ADVANCE THE SUSTAINABLE MANAGEMENT OF SEASONAL JELLYFISH BLOOMS IN COASTAL ENVIRONMENTS?” “The challenge focuses on sustainable jellyfish management after an early-warning notification.”

2. ENEC (7 Feb 2017). Final Condenser for the Barakah Nuclear Energy Plant… (news release).

   https://enec.ae/news/latest-news/final-condenser-for-the-barakah-nuclear-energy-plant-safely-and-successfully-installed/

   Supports: Condenser cooling-water flow of about 6,000 m³ per minute per unit. Accessed 10 Oct 2026.

   “…will have a volume of 6000 m3 of sea water per minute passing through them…”

3. Power Engineering, R. Walton (7 Apr 2021). Barakah, first nuclear power plant in UAE, starts commercial operations.

   https://www.power-eng.com/nuclear/barakah-first-nuclear-power-plant-in-uae-starts-commercial-operations/

   Supports: Unit size 1.4 GW; four units, 5,600 MW. Accessed 10 Oct 2026.

   “…full connected stage for 1.4-GW Unit 1” “Barakah's four units eventually will generate 5,600 MW at full capacity”

4. Nuclear Engineering International (28 Sep 2012). Building Barakah.

   https://www.neimagazine.com/advanced-reactorsfusion/building-barakah/

   Supports: Design maximum cooling-water flow of 172.5 m³/s per unit (our 690 m³/s bound). Accessed 10 Oct 2026.

   “…with a maximum flow rate of 172.5 m3/second/unit.”

5. FANR (2012). Safety Evaluation Report (summary) of an application for a licence to construct Barakah Units 1 and 2.

   https://www.fanr.gov.ae/en/Documents/Safety%20Evaluation%20Report%20of%20an%20Application%20for%20a%20Licence%20to%20Construct%20Barakah%20Unites%201%20and%202.pdf

   Supports: Safety-related essential service water uses Gulf seawater through its own intake structure, so 400 m³/s is the condenser flow only. Accessed 10 Oct 2026.

   “The ESW system utilizes water from the Arabian Gulf (the plant UHS) to remove heat from plant SSCs.” “…essential service water intake structure HVAC, and circulating water intake structure HVAC systems…”

6. NMDC Group. Barakah NPP marine works (project page).

   https://nmdc-group.com/en/dredging-and-marine/barakah-npp-marine-works/

   Supports: Total dredged area of the three channels, 2.35 km², and their depths; no channel widths are published. Accessed 10 Oct 2026.

   “…dredging of an intake channel for cooling water intake, discharge channel for hot water release and a navigation channel for wharf access.” “The total area dredged was approximately 2.35Km2 at the final depth of the channels, which range from -5.5 MSL to --9.5 [sic] MSL.”

7. NOAA National Centers for Environmental Information (2022). ETOPO 2022 15 Arc-Second Global Relief Model.

   https://doi.org/10.25921/fd45-gt74

   Supports: The seabed depth profile in our site model (one transect, read from this global relief model; not survey data), which sets the ring depths and the channel lengths. Accessed 10 Oct 2026.

8. Nesterov, O. et al. (2023). A numerical assessment of the dispersion of dissolved pollutants in the Arabian Gulf associated with the Barakah nuclear power plant. Ocean Modelling 186, 102274.

   https://doi.org/10.1016/j.ocemod.2023.102274

   Supports: Spread of the non-tidal (background) current, 4.8 cm/s. Accessed 10 Oct 2026.

   “...[RMSE] between the measured current velocities and those reconstructed from the tidal harmonics is 4.8 cm/s with the correlation coefficient of 50.1%. This indicates a relatively large influence of non-tidal forcing.”

9. Thoppil, P. G. and Hogan, P. J. (2010). Persian Gulf response to a wintertime shamal wind event. Deep-Sea Research I 57(8), 946–955.

   https://www7320.nrlssc.navy.mil/pubs/2010/prasad1-2010.pdf

   Supports: Real Shamal winds typically reach 15–20 m/s and last 24–36 hours or 3–5 days. Accessed 10 Oct 2026.

   “Based on duration, there are two types of winter shamal: those which last 24–36 h and those which last for a longer period of 3–5 days. During a shamal, winds typically reach 15–20 m s⁻¹ near the surface during the daytime.”

10. Abu Dhabi Distribution Company (ADDC). Rates and tariffs 2025, business customers.

    https://www.addc.ae/en-US/business/Pages/RatesAndTariffs2025.aspx

    Supports: Business electricity tariffs of 20–36.6 fils/kWh, used as a proxy for the value of lost output. Accessed 10 Oct 2026.

    “Electricity 20 fils For 1 kWh” “27 fils … 36.6 fils For 1 kWh”

11. Central Bank of the UAE. Annual Report 2014, Chapter 5, p. 39.

    https://www.centralbank.ae/media/e44dkkze/cbuae-annual-report-2014-en.pdf

    Supports: The dirham's fixed rate of 3.6725 per US dollar, used to convert ADDC's tariffs to US dollars. Accessed 10 Oct 2026.

    “The fixed peg of the exchange rate of the dirham to the US dollar, at a rate of 3.6725 dirhams per US dollar, means that the CBUAE has limited degree of freedom to set interest rates…”

12. AFP, via L'info durable (13 Aug 2025). Méduses à la centrale nucléaire de Gravelines : un premier réacteur a redémarré (in French).

    https://www.linfodurable.fr/meduses-la-centrale-nucleaire-de-gravelines-un-premier-reacteur-redemarre-52517

    Supports: Jellyfish shut down Gravelines units 2, 3, 4 and 6 on 10–11 Aug 2025. Accessed 10 Oct 2026.

    “Les unités de production n°2, 3 et 4, arrêtées automatiquement depuis dimanche soir pour la même raison” “Le réacteur n°6 a redémarré ce matin à 7H30”

13. Raskoff, K. A., Sommer, F. A., Hamner, W. M. and Cross, K. M. (2003). Collection and culture techniques for gelatinous zooplankton. Biological Bulletin 204(1), 68–80.

    https://doi.org/10.2307/1543497

    Supports: Air bubbles can harm medusae larger than 3 cm, so Phase 0 scores animals held at the curtain. Accessed 10 Oct 2026.

    “Although air bubbles can be helpful in the culture of many small gelatinous animals by increasing water circulation, they can be detrimental to larger adult sizes (>3 cm).”

14. Haberlin, D., McAllen, R. and Doyle, T. K. (2021). Field and flume tank experiments investigating the efficacy of a bubble curtain to keep harmful jellyfish out of finfish pens. Aquaculture 531, 735915.

    https://doi.org/10.1016/j.aquaculture.2020.735915

    Supports: A high-airflow bubble curtain deflected large jellyfish in the field; the study reports no injury or survival data. Accessed 10 Oct 2026.

    “a high air flow 8 m linear bubble curtain set at 5 m depth effectively deflected large compass jellyfish” “increased wave height and increased air flow increased jellyfish transport through the curtain”

In shortHow we used AI, in the words on our proposal deck.

AI use: we used Claude (Anthropic) to help write, debug and run our simulation code, check sources, and draft figures, slides and illustrations. The team set the methodology, made the design decisions, and interpreted the results.

Bubble curtain

A line of seabed pipes releasing air. The rising bubbles drive a surface current that pushes drifting jellyfish back.

Holding limit

The fastest inward water speed a bubble curtain can still hold jellyfish against.

Tide-aware air

Air that is boosted where the tide pushes water in and switched off where it flows out.

Design bloom

Our 3‑day test bloom: 5 g of jellyfish per m³ of water, about 1,400 t within 4 km of the intake. Dense bloom: 30 g/m³.

Central case

The model run with a randomly varying background current. Steady currents of 0 and ±0.05 m/s are the bounds.

Tide start

When the bloom arrives in the tide cycle. We use 24, spread over one spring–neap cycle.

Virtual jellyfish

A computer particle standing for about 119 kg of jellyfish in the design bloom.

Worst hour

The largest load reaching the intake in any single hour of the bloom.

Failure pulse

What reaches the intake if the air fails when the ring holds the most jellyfish.

Unit-day

One reactor unit (1.4 GW) offline for one day.

CAPEX, running cost

The one-off build cost; the yearly cost of energy and maintenance.

Early (class-5) estimate

A cost estimate made before detailed design, here taken as ±50 %.

CRF

Capital recovery factor: turns a build cost into an equal yearly payment over the equipment's life.

2 standard errors

The range the result would likely stay within if the 24 tide starts were drawn again.

Poisson standard deviation

The natural scatter of a random count; used to judge whether a worse hour is more than chance.

Nm³/min

Cubic metres of air per minute at standard conditions.

Current meter (ADCP)

An instrument that measures water speed.

Phase 0

The survival trials: step 1 of our three gated steps if shortlisted. Step 2 is a 100 m pilot section with one robot at sea; step 3 the full ring after approvals.

Live well

The robot's flooded tank: the jellyfish ride in seawater that flows through it.

Spring–neap cycle

The roughly two-week cycle from strong (spring) to weak (neap) tides and back; our tide starts are spread over one cycle (14.77 days).

Shamal

A strong seasonal wind over the Gulf. Real Shamals typically reach 15–20 m/s (Thoppil & Hogan 2010).

Medusa, medusae

The free-swimming, bell-shaped form of a jellyfish; medusae is the plural.

Plume

The column of rising bubbles and water above a bubble pipe; at the surface it spreads into the current that holds jellyfish back.

Aftercooler

A heat exchanger that cools the hot air from the blowers before it enters the plastic pipes; ours use seawater.

σ and τ

The spread (σ, 4.8 cm/s) and the memory (τ, 2 days: how long a pattern tends to last) of the randomly varying background current in the central case.

Jackknife

A way to estimate uncertainty: recompute the result with one tide start left out at a time and see how much it moves.

ENEC

Emirates Nuclear Energy Company, which set this challenge.

ATP

Advanced Technology Pioneers 2026, the competition this entry is for, run on the ChallengeON platform.

FANR

Federal Authority for Nuclear Regulation, the UAE's nuclear regulator.

EAD

Environment Agency – Abu Dhabi.

SCADA

Supervisory control and data acquisition: the computer system that monitors and controls the air plant and valves.

DN

Diameter nominal: the standard size code of a pipe, in millimetres.