Donovan Nitrate Destroyer

A carbon-fed U-path denitrator after Donovan Joannes's design, built for this tank's 750 L system. Water runs down an aerobic column where dosed carbon burns off dissolved oxygen, crosses at the base, then up an anaerobic column where nitrate is reduced and leaves as nitrogen gas. Fed by its own channel on the same reef system's SmartDoser, over a stepper-driven Kamoer peristaltic pump. It replaced the tank-wide vinegar dosing described on the Acropora page, which stopped on 21 October, two days after the reactor started.

Live model and test results: DND model →

Every reactor and tank test, the fitted nitrate production, where the tank's nitrate will level off at each feed flow, and the safe carbon band for each flow (too little brings nitrite back, too much makes hydrogen sulfide). It refits itself within about 20 minutes of each logged test or setting change. This page is the build and the story; the model page has the current numbers.

Design figures for this build
Feed flowCarbonResidenceN₂ produced
60 mL/min (11.5% of daily turnover, ~86 L/day)5–15 mL/day, 8% vinegar at the inlet~3.1 h (~11.0 L void volume at 60 mL/min)~300 mL/day at 1.5 g NO₃/day reduced

Plumbing

SUMP ceramic rings ceramic rings crossover: open pipe, no media O₂ exhausted below here AEROBICANAEROBIC DOWNUP FEED, PERISTALTIC 29 mL/min water in, ~5 cm below top VINEGAR 8% 21 mL/day peristaltic carbon through the top cap check valve inoculation port: syringe valve shut except when inoculating or dosing IRON ferrous gluconate peristaltic check valve iron in, just below brewing airlock = bubble counter vent, top above the reactor T out, ~10 cm below top rises a few cm past the T 25 mm PVC downpipe, open top: can't run full, can't siphon air gap capped tee stub below each column
As built. Water enters the side of the aerobic leg about 5 cm below the top, the iron solution through its own port just below that, behind a check valve, and the vinegar through the top cap, behind a check valve. Vinegar and iron each have their own small peristaltic pump. After that check valve a T with a valve and a syringe serves as the inoculation port; the valve stays shut except when inoculating or dosing something else. The outlet leaves the anaerobic leg about 10 cm below the top, so the small head difference drives the U once the pump has filled the first column. Donovan's original uses a small aquarium pump with an adjustable valve at this point; a peristaltic does the same job at a rate that doesn't sag as the media loads. From the outlet the line rises to a T: the gas goes up a vent topped above the reactor and out through a brewing airlock, which doubles as a bubble counter; the water rises a few centimetres more and drops into an open-topped 25 mm PVC downpipe with an air gap to the sump, which cannot run full or siphon (why it ended up like this). Both legs are filled with ceramic rings; the crossover at the base carries no media -- it's open pipe, shown unfilled. Building it from tees leaves a short capped stub under each column. The service manifold and flush valves described in the next section were planned but never built.

After Donovan Joannes's Nitrate Destroyer. The aerobic/anaerobic split and the inlet/outlet level offset follow his design; his graded media (bio rings and balls, pumice, crushed coral) became ceramic rings in both legs here; flow, carbon and gas figures are worked for this 750 L system.

Flushing a clog: planned, not built

None of this was built. The service manifold, valves V1–V7 and the 3-way valve below were the planned design; the reactor as it runs has none of them (see the drawing above). It stays here as the option to fit if the bed ever clogs. The valve numbers refer to that plan, not to the drawing.

The service manifold tees off the pressure side of the return pump through a gate valve. With V2 and V3 shut it is a dead leg, so normal running loses nothing; opening either one puts return pressure across the reactor, throttled back to something the beds survive.

The two directions are not redundant -- each one lifts the bed the other packs down:

  • Forward flush (V2) runs down the aerobic leg and up the anaerobic, fluidising the anaerobic bed.
  • Backflush (V3) runs down the anaerobic leg and up the aerobic, fluidising the aerobic bed -- which is where detritus actually collects, because it is the first thing the water meets.
Valve positions by mode
ModeV1 feedV2 fwdV3 backV4 outletV5 topV6 aer. stubV7 anaer. stub3-way
Runopenshutshutopenshutshutshutsump
Forward flushshutopenshutopenshutshutshutwaste
Backflushshutshutopenshutopenshutshut—
Sludge drainshutshutshutshutopenopenopen—
Fill & ventopenshutshutopencrackshutshutwaste

V5 earns its place three times over: it is the backflush exit, the air bleed when first filling, and the air inlet without which the stub drains will not run. Drain one stub at a time and smell each separately -- a sulfide note from the anaerobic side is the earliest warning the reactor has gone too far, well before a test kit shows anything.

Doing it

  1. Set the gate valve once, by trial, to 2–4 L/min -- roughly 40–80× the feed rate. That lifts a packed bed; much more ejects media into the bucket.
  2. Purge the manifold first. Open V2 and V5 with V1 shut and the stagnant water in the dead leg goes straight up and out to the bucket without crossing a bed.
  3. 30–60 seconds per direction, 1–4 L to waste. Watch the bucket: stop when it runs clear.
  4. Reopen V1 last and confirm the air gap is running before you walk away.

What it costs

  • It sets the biology back. Flush water is oxygenated, nitrate-rich tank water. A flush aerates the anaerobic leg and stalls denitrification -- expect nitrate to flatten for several days and then resume. Flush when throughput genuinely drops, never on a schedule.
  • It disturbs the bed. A hard flush shifts the rings and strips biofilm off them. Use the lowest flow that restores flow.
  • A peristaltic hides the clog. Positive displacement means flow does not sag as the bed loads -- pressure rises instead, until a fitting or the tube gives. So fit a cheap gauge or a relief tee on the feed line; do not expect to notice a clog as a flow drop.
  • Media needs retaining. A coarse screen under each top cap and a strainer on the outlet port, or a flush carries media out of the column.
  • A capped tee stub is a sulfide trap; a valved one is an asset. The dead legs below the crossover hold stagnant, anoxic, carbon-rich water with organics settling into it -- and with no flow there is no nitrate resupply, so once the nitrate in that pocket is used up the bacteria move on to sulfate. Fit a valve, not a plain cap, and drain them to a bucket on the same schedule you check the effluent. Keep them short: 10cm is enough to catch solids, and nothing above that earns its keep.
  • Check valves matter here. One on the vinegar line and one on the manifold riser. Without them, backflush pressure pushes into the dosing head, and reactor water -- low ORP, possibly sulfidic -- can siphon into the return line when the pump stops.

What happens along the path

NO₃ = 0 → sulfate reduction → H₂S O₂ NO₃ AEROBIC, down ANAEROBIC, up inlet base crossover outlet high zero
Indicative, not measured. The shape is what matters: nitrate barely moves until oxygen is gone, then falls through the second leg -- and must still be above zero at the outlet.

Bacteria use oxygen first because it yields more energy than nitrate. So the aerobic leg isn't wasted volume -- it is what makes the anaerobic leg anaerobic, by spending carbon to strip the O₂ the incoming water carries.

That is also why carbon enters once, at the top. It has two jobs in sequence, and the same dose does both.

The outlet reading is the control signal for the whole reactor. Nitrate still present at the outlet means the bed ran out of carbon before it ran out of nitrate -- which is the safe ordering. Nitrate at zero means carbon is left over with nothing to reduce, and sulfate is next in line.

If effluent nitrate reaches zero: raise flow first, cut carbon second.

Where nitrate settles

10 ppm @ 87 5 ppm @ 231 diminishing returns 2015 105 100200 300400 feed flow, mL/min tank NO₃, ppm
At 1.0 g NO₃/day production and a 2 ppm effluent. Going from 15 to 10 ppm costs 35 mL/min; going from 10 to 5 costs another 144.

tank NO₃ ≈ effluent + (production ÷ flow)

FlowSettles atResidenceCarbon
52 mL/min~15 ppm3.5 h~14 mL/d
87 mL/min~10 ppm2.1 h~22 mL/d
174 mL/min~6 ppm63 min~33 mL/d
231 mL/min~5 ppm48 min~39 mL/d

The pump is never the constraint -- 231 mL/min is 58% of the Kamoer's range. Residence time is. At 5 ppm the bed has 48 minutes to strip oxygen and reach a 2 ppm effluent, against the ~3.5 hours this geometry gives at the design flow. If it can't, removal per pass drops and you need yet more flow, which shortens residence again. More volume is the only way out of that spiral -- see Sizing below.

Carbon nearly triples too, and mostly for oxygen rather than nitrate: at 231 mL/min roughly 27 of the 39 mL goes on deoxygenating 333 L/day.

Don't pick a target yet. Every row rests on that 1.0 g/day production figure, which is inferred rather than measured. Run Donovan's 52 mL/min for a fortnight: flow × (influent − effluent) gives the real removal rate, and the tank's trend against it gives production. Then the target is arithmetic.

Sizing

The figure that decides everything is easy to get wrong, and an earlier version of this page did: residence time is the water volume divided by the flow, not the pipe volume. Media fills roughly half the tube in each column -- but the base crossover between them carries no media at all, confirmed on the real build. So a system holding 19.4L of gravel, rings and open pipe holds about 11L of water, not the roughly 10L a uniform 50% estimate across the whole bulk would suggest. The empty crossover holds more water per litre of bulk than a packed column does, not less.

residence = (column bulk × 50% void) + (crossover bulk × 100% void), ÷ flow

For 100mm pipe (110mm OD, 103.6mm bore) in the columns, with the 0.3m base crossover counted separately at 100% void:

Two towersBulkWaterAt 52 mL/min (~15 ppm)At 87 mL/min (~10 ppm)
1.0 m each19.4 L11.0 L3.5 h2.1 h
1.25 m each23.6 L13.1 L4.2 h2.5 h
1.5 m each27.8 L15.2 L4.9 h2.9 h

The 0.3m crossover is worth 2.5L of bulk, about 13% of a 1m build -- and since none of it is packed, all 2.5L is water, not half of it. It is still horizontal, and an empty horizontal pipe still channels in its own way: flow takes the shortest path across rather than using the full cross-section, so treat the crossover's contact value as near zero even though its hydraulic volume counts in full. It is also why the base drain sits at that exact point.

The two numbers pull against each other

Equilibrium is set by flow; residence is set by volume. Raising flow to settle the tank lower shortens the contact time that makes the settling possible, so a given volume quietly fixes how low the tank can go.

  • 1 m towers reach ~15 ppm comfortably -- 3.5 h at the design flow, squarely inside the 1–4 h band carbon-fed denitrators work in.
  • 10 ppm asks for 87 mL/min, leaving 2.1 h. Workable, with a bit more margin than the open-crossover correction first suggested. An extra 25cm per tower adds further headroom and changes nothing else about the build.
  • Below ~6 ppm the geometry runs out before the pump does -- under an hour of contact to strip oxygen and reduce nitrate. More flow then removes less per pass, which is the spiral that only more volume escapes.

If you can choose the pipe

Volume goes with the square of the bore, so diameter buys far more than height per euro and per metre of headroom. Same 1m towers, same shelf:

Pipe (OD)BoreBulk, 2 towersAt 52 mL/minAt 87 mL/minL/D
110 mm103.6 mm16.9 L2.7 h1.6 h9.7
125 mm118.6 mm22.1 L3.5 h2.1 h8.4
160 mm152.0 mm36.3 L5.8 h3.5 h6.6

Keep the length-to-diameter ratio above about 5. Below that a packed bed starts channelling and the extra volume stops earning its keep -- which is the same reason four towers in series is the wrong way to add capacity. A four-leg path puts a downflow leg deep in the anaerobic zone, where the nitrogen it generates would rise against the flow. If you need more than two legs, run two U-paths in parallel, each with its own balancing valve, so every gas-producing leg still runs upward.

Packing is worth a whole tower

Void fraction moves residence as much as half a metre of pipe does, and it costs nothing to get right. The crossover's 2.5L is fixed regardless of packing choice -- it isn't packed at all -- so it's added on top of whichever column void fraction you end up with, not scaled by it.

Column mediaVoidColumn water+ crossover (2.5L, fixed)At 52 mL/min
Dense or tamped down40%6.8 L9.3 L3.0 h
Mixed, typical50%8.5 L11.0 L3.5 h
Open bio-balls and rings65%11.0 L13.5 L4.3 h

Donovan's media schedule already leans open in the aerobic leg -- large rings and balls rather than fines -- because that leg only has to burn oxygen. This build went further: ceramic rings in both legs, and none in the crossover. Either way, pour the media in loosely and let it settle, and resist both tamping and anything sand-like.

Operating figures

750 L system, NO₃ 42 ppm → ~10 ppm target.

ParameterValueBasis
Feed flow60 mL/minconfirmed by direct timed-volume measurement under real load, not bench spec
Start-up flow, week 130 mL/min~5 mL per 10 s, then work up
Carbon dose5–15 mL/daylean on purpose -- full reduction would take ~30 mL
Media volume~19 L2 × 1 m of 100 mm, plus a 0.3 m crossover (crossover carries no media)
Residence time~3.1 h~11.0 L of water ÷ 60 mL/min
Nitrogen produced~300 mL/dayat 1.5 g NO₃/day reduced
Driverspeed 110 @ mstep 256SERVO42C → 64.45 RPM, measured directly -- see note below

The pump's own bench spec doesn't hold under real load. Measured mL/rev under this build's actual suction and discharge resistance came out speed-dependent -- 0.885 mL/rev at low speed, rising to 1.054 at high speed -- not the constant 1.422 mL/rev a free-discharge bench test gave. Lower speed delivering less per revolution than higher speed is backwards from the usual assumption that slower rotation means more complete tube recovery; it was real and repeatable across multiple timed runs, not measurement noise. Trust a timed-volume measurement against the actual installed system over any catalogue mL/rev figure.

These were the planned figures. What actually runs, and why it differs, is in the commissioning log below: 30 mL/min of feed and 21 mL/day of vinegar as of 16 November 2026. Current figures are on the model page.

Rules the drawing encodes

  • The gas-producing leg runs upward. At 52 mL/min the water moves about 0.011 cm/s while bubbles rise at ~20 cm/s, so no practical flow carries gas downward. Nitrogen is only made in the anaerobic leg, and that leg exits at its top.
  • No high points in the return. Gas collects at any peak until it forms an air lock that stops flow -- and stopped flow is the hydrogen sulfide condition.
  • Keep carbon lean. The bed must run out of carbon before it runs out of nitrate. That ordering is the entire safety margin.
  • Positive displacement on the feed. As the media loads, a valve-throttled centrifugal pump's flow falls -- drifting toward the hazard. A peristaltic holds its rate.
  • Flush before returning effluent. After any stoppage, send the first volume to waste rather than into the display.

Commissioning

  1. Fill, purge air, and run water only with no carbon. Confirm flow by timing output into a measuring cylinder -- that measured figure is what you set the pump by from then on.
  2. Day 3: begin carbon at 5 mL/day at the inlet, seeding bacteria at the same time. Leave the display vinegar dose running unchanged; the tower removes little until it colonises.
  3. Watch for nitrogen bubbles at the air gap. Their appearance is the first evidence of denitrification, usually one to three weeks in.
  4. Test effluent nitrate. It should fall below tank level but stay above zero. If it reaches zero, raise flow before cutting carbon. In practice: if your nitrate kit works by zinc reduction (the Hanna HI782 does), nitrite in the effluent reads as nitrate, and the effluent cannot be tested for nitrate at all while nitrite is present. Test effluent nitrite instead, diluted — see the log below.
  5. Only once the tower demonstrably works, taper the display vinegar dose and move that carbon to the inlet.

Two side effects to anticipate

Both arrive before the nitrate drop does.

Alkalinity rises

Denitrification returns about 3.57 g CaCO₃ per g NO₃-N reduced -- intrinsic to the reduction reaction itself, not dependent on the media it runs over. At meaningful removal that offsets a real fraction of this system's 0.44 dKH/day demand, so the NaOH dose has to come down. Test alkalinity every two or three days once bubbles appear. This build runs ceramic rings throughout rather than Donovan's original crushed coral, so expect somewhat less than his design shows -- ceramic adds none of the slow CaCO₃ dissolution a coral-media build gets on top of the reaction's own output, just the reaction-driven portion above.

Phosphate falls

Biofilm growth consumes phosphorus. With PO₃ already near the 0.05 ppm floor, a stall will look like the reactor failing rather than a nutrient running out. If phosphate bottoms out, denitrification stops -- raise the phosphate dose before chasing anything else.

Commissioning log, 19 October – 18 November 2026

What actually happened when the reactor went online, including what I got wrong. As of 16 November: feed 30 mL/min, carbon 21 mL/day of 8% vinegar, and denitrification is complete: no measurable nitrite in the effluent, and effluent nitrate 0.5–1 ppm against about 29 in the tank. For the four weeks before 15 November, about a tenth of the nitrate kept leaving as nitrite. 15 November covers how that ended, 16 November how close the reactor then ran to the sulfide side and what moved it back.

The current numbers are on the DND model page, fitted to this tank's own tests and refitted automatically as new ones come in.

Weeks one and two: the design flow outran the carbon

The reactor started on about 19 October at Donovan's 52 mL/min with 12 mL/day of vinegar. Undiluted, the effluent nitrite test went darker than the chart's top colour, and the tank's nitrate didn't move. The electron budget explains it: at that flow, stripping the feed water's oxygen in the first leg took roughly half the carbon, and what was left could take the incoming nitrate only as far as nitrite. For those ten days the reactor was a nitrite factory: nitrite went back to the tank, where the biofilter turned it into nitrate again. On 29 October the feed came down to 29 mL/min (timed into a cylinder, twice) and carbon went to 14 mL/day. The next day the first properly diluted test read 3.0 mg/L of nitrite: about 89% of the nitrate was now going all the way to nitrogen.

Lesson: start at a flow your carbon can actually match, around half the design flow here, and raise both together.

Measuring the effluent

  • Nitrate is unmeasurable while nitrite is present on a zinc-reduction kit: it reads nitrite as nitrate (×1.348 by weight). The same correction applies to the tank's own nitrate whenever the tank shows nitrite.
  • Nitrite, diluted, single step. With the Tropic Marin nitrite test, 1 mL of effluent in 14, 29 or 59 mL of RODI (15×, 30×, 60×) brings the reading onto the chart's most distinguishable swatches, 0.05–0.2. At 15×, 1.5 and 3.0 mg/L fall on the same swatch; 30× separates them. A two-stage 100× dilution read implausibly low and was discarded: errors multiply across stages.
  • 10× turned out to be the useful step for this effluent (1.5 mL in 13.5 mL of RODI): at 30× the reading sat between the chart’s two lowest colours. Read exactly at the stated time, too. The colour keeps darkening afterwards; a 30× vial that read about 0.1 on time looked like 0.25–0.3 later.
  • The effluent is not steady. Carbon goes in as 1 mL pulses and the bed behaves like plug flow, so a sample taken just after a dose need not match one taken just before. Test at the same time of day.

Gas: an independent check that needs no test kit

A clear section of outlet tubing showed small bubbles from the first days, with the feed line itself bubble-free and no sulfide smell, so the gas was forming in the bed. Caught under a water-filled inverted vial in the sump, it came to about 10.7 mL/hour on 5 November and 9.8 mL/hour on 6 November, against about 11.8 mL/hour expected for 90% conversion at this flow (some nitrogen always leaves dissolved, and some of the gas is CO₂). Two very different measurements, one chemical and one volumetric, agree.

The outlet plumbing needed rework to get there:

  • At 29 mL/min the water in a 6 mm tube moves 1–2 cm/s, too slow to drag bubbles along. In any horizontal or downward run they stall and stick to the wall. The vial then reads low, and gas collects into pockets.
  • A full downward run siphons. With the outlet end submerged in the measuring vial, a vent at the high point pulled air in. A check valve on the vent made it worse: the suction then acted on the reactor itself, where any leaky fitting could admit air into the anaerobic leg. Even with the end in free air, a 6.4 mm (3/8" RO) drop could still run full.
  • What works: rise from the outlet to a T; a vent upward from the T, topped higher than the reactor, with a brewing airlock on it; the outlet rising a few centimetres more after the T (that small head is what pushes the gas through the airlock's water seal instead of out of the outlet); then into an open-topped 25 mm PVC downpipe with an air gap, which cannot run full and cannot siphon at any flow this reactor will see. The airlock then doubles as a bubble counter: a daily glance confirms the bed is working.

Carbon: more did not finish the job

FromCarbon, mL/dayEffluent nitrite, mg/L
29 Oct14 (about 13.6 delivered)3.0, 2.25, 3.0
1 Nov162.25, 2.25, 1.5, 1.5
4 Nov172.25, 2.25
5 Nov18—
6 Nov192.25, 2.25
7 Nov20about 2 (30× and 60×, next morning)
8 Nov20, iron trial starts1.75, 1.75 (10×, 10 and 11 Nov)
13 Nov211.75 (13 Nov, before the change); about 0.1 (15 Nov)

A carbon model with no net biofilm growth, calibrated on the 14 mL/day reading, predicted zero effluent nitrite from about 14.8 mL/day. It was wrong. Carbon went up 45% and the effluent stayed in the same band, one swatch step either way. No cloudiness in the display and no slime at the outlet means the extra carbon was being consumed, not passing through. Something other than the amount of carbon is limiting the last step, nitrite to nitrogen. The model still serves for the nitrate balance below, which rests on measured flow and measured nitrate; it no longer sets the carbon dose.

Candidates for the last tenth

Written while the problem was open; it closed on 15 November (below). Kept as it was, because the list is what anyone stuck at the same point will need.

  • Iron. One of the two enzymes that reduce nitrite contains iron, a growing bed needs more of it than a mature one, and the trace product this tank uses supplies very little. A trial started on 8 November: an acidified ferrous gluconate solution, about 29 µg of iron a day in ten 1 mL doses, from its own pump, through a check valve, into its own port on the aerobic leg just below the water feed. Since 11 November that pump turns at the same speed as the carbon pump and fires with every second carbon pulse, so iron and carbon enter the bed together. That figure is my own order-of-magnitude estimate of what the bed can use while it grows, not a published protocol; even if all of it reached the display it would be about 120 times below the manufacturer's maximum daily iron dose for the tank. An ICP sample went to the lab on 9 November as the baseline.
  • Carbon pulses. Packets of water that pass between doses get little carbon; nitrate reduction takes what there is and nitrite reduction comes up short. Smaller, more frequent doses would be the fix.
  • Gas pockets in the bed, taking volume out of play.
  • Too few bacteria for the second step. For the first ten days almost nothing reached nitrogen, so the bacteria that do the second step had little to live on; they have only been properly fed since 29 October.
  • Ruled out so far: missed doses (every carbon dose logged as delivered), temperature at 18–20 °C (municipal plants denitrify completely in colder water), molybdenum (dosed at standard), phosphate (0.33 ppm), salinity, light on the reactor (it is dark).

15 November: the last tenth closed

Two days after carbon went from 20 to 21 mL/day, the effluent nitrite test at 10× dilution read the chart's lowest colour, 0.01. That puts the effluent at about 0.1 mg/L, down from 1.75 in three readings over the previous five days. With nitrite gone, a nitrate test on the effluent finally means something: 1–2.5 ppm (Aquaforest kit), against 33.4 in the tank the day before.

  • What the reactor now removes: about 31 ppm across the bed × 41.8 L/day of feed = roughly 1.3 g of nitrate a day, or about 1.75 ppm/day of the 750 L. That is no more nitrate per day than before; it was already about nine-tenths of the way. The difference is that none of it now goes back to the tank as nitrite to be turned into nitrate again.
  • Why it happened now, I can't say for certain. Three things coincided: the fifth one-step carbon increase, day seven of the iron trial, and plain time. The literature on acetate-limited denitrification reports that the nitrite step takes several weeks to recover once carbon is adequate, and the bed had been properly fed for about two and a half weeks. Four earlier carbon steps of the same size did nothing visible, so carbon alone is the least likely explanation. Changing one thing at a time didn't help here, because time kept passing between the changes.
  • The reactor now runs at its edge. With nitrate in the effluent close to zero, every extra millilitre of carbon has nothing left to reduce except sulfate, which makes hydrogen sulfide. At a fixed flow, carbon from here only goes down, as the tank's nitrate falls. Calibrated on this point, the dose at 29 mL/min is about 5.9 + 0.49 × (tank NO₃ − 3) mL/day of 8% vinegar: 21 at 33 ppm, 14 at 20, 9 at 10. Both numbers scale with flow, because more water brings more oxygen and more nitrate. Fitted on this point, only about 64% of the carbon goes to respiration; the rest builds biofilm.
  • Next step: flow, not carbon. This is the self-protecting loop the forums describe: climb carbon until the effluent nitrite reaches zero, then raise the flow so fresh nitrate restores the margin. Flow is also the only thing that lowers the level the tank's nitrate settles at: fitted to the tests since 31 October, the tank makes about 1.4 ppm of nitrate a day (1.2–1.6), so at 29 mL/min it levels off near 27 ppm, and at the 52 mL/min design flow near 16.
  • The iron trial continues to its planned end around 22 November, unchanged, so that stopping it is the next single change. The ICP baseline from 9 November is due back this week.

16 November: complete, at the edge, then more flow

The next day's effluent tests (Tropic Marin kit): nitrite undetectable, nitrate 0.5–1 ppm, no smell of hydrogen sulfide. The tank read 28.9 ppm nitrate (Hanna) and 0.1–0.15 nitrite. So the bed was taking out about 28 of the 29 ppm it was fed, and nitrite was gone for the second test in a row.

  • Compared with the literature. Nitrite-free at a net 2.0–2.1 g of vinegar carbon per g of nitrate-nitrogen removed (after the carbon that burns the feed's oxygen), which is about 5.4–5.7 g COD. That is within the 3–6 that comparable reactors need; see the literature, in one unit below. Corrected 17 November: this said the threshold was the same as the 2:1 of Oh & Silverstein (1999), but that paper's abstract does not say whether its ratio counts carbon or COD, so the comparison did not hold. The model never doses below this reactor's own measured floor.
  • Complete, but at the edge. Effluent nitrate below 1 ppm means the end of the bed is running out of nitrate. Past that point the spare carbon goes to sulfate, which makes hydrogen sulfide. So there are two edges: too little carbon brings nitrite back, too much makes sulfide. At 29 mL/min and this tank's nitrate the band was about 19.3 to 21.4 mL/day of vinegar, and the reactor ran at 21, a margin of 0.4. Narrow, and it narrows by itself: each ppm the tank's nitrate falls moves the whole band down by about half a millilitre.
  • The fix is flow, not less carbon. The feed went from 29 to 30.1 mL/min (pump speed 54 → 56), carbon unchanged. More water brings more nitrate, which puts nitrate back at the end of the bed, and it removes a little more from the tank. The band moved to 20.0–22.2 and the margin to 1.2 mL/day, back in the middle. Effluent nitrate should settle near 2 ppm; the next tests will tell.
  • Now watched continuously. Home Assistant recalculates the band from every logged test and from the doser's live settings, and warns when the margin drops below 0.5 mL/day or the dose falls under the nitrite floor. The model page shows the band for every feed flow.

17 November: more flow, same carbon, nitrite back

About 21 hours after the feed went from 29 to 30 mL/min with carbon unchanged at 21 mL/day, the effluent nitrite test read more than 1 mg/L. A timed catch at the outlet confirmed the new flow: exactly 30 mL in 60 s. The feed went back to 29 mL/min the same evening.

  • What happened. 3.5% more water with the same vinegar is 3.5% less carbon per litre. The net C:N fell from 2.13 to about 1.99, just under the lowest value at which this reactor had tested clean (2.01). The step that suffers first is the last one, nitrite to nitrogen: missing a few percent of the electrons is enough to leave 1–2 mg/L of nitrite.
  • What I got wrong. The band of 16 November (20.0–22.2 mL/day, margin 1.2) assumed that a small carbon shortfall would leave a little nitrate in the effluent. It left nitrite instead, so the real nitrite edge sat right at the carbon the reactor was already getting. The forum rule “raise the flow so fresh nitrate restores the margin” only works with carbon to spare; at the edge it swaps the sulfide risk for nitrite.
  • How uncertain the band is. Worked out test by test, the share of carbon that goes to respiration ranges from 58 to 67%. That is about ±1.5 mL/day on both edges, wider than the gap between them, and the sulfide edge has never been tested: it is a calculation. Until more tests narrow it, the rule is empirical. Keep the vinegar per litre of feed at least where the effluent last tested clean (21 mL per 41.8 L at about 29 ppm tank nitrate); raise carbon before flow, and lower flow before carbon.
  • Expected recovery: a day or two. The bacteria that reduce nitrite went short for one day, not for weeks; the slow recovery in the literature followed long starvation. The next effluent test, nitrite and nitrate together, will show it.

18 November: back at 29 mL/min, clean in a day

About 24 hours after the feed went back from 30 to 29 mL/min, still with 21 mL/day of vinegar, the effluent read 0.05 mg/L nitrite and 0.5–1 mg/L nitrate (both Tropic Marin kits, nitrate read as 0.9). With about 29 ppm going in, that is roughly 97% of the nitrate removed and nitrite at about 0.2% of the load: the same picture as 16 November, before the flow step.

  • Faster than expected. I said a day or two; it took one. The nitrite-reducing bacteria had been short of carbon for a day, not weeks, and picked up as soon as the carbon per litre was back.
  • What it shows. At 21 mL/day of vinegar, 29 mL/min is complete and 30 mL/min is not. The nitrite edge for this carbon lies between those two flows: 0.50 mL of vinegar per litre of feed (21 mL over 41.8 L) is enough, 0.49 (over 43.2 L) is not. One test each way, so a narrow but not a precise edge.
  • Next. Hold 29 mL/min and 21 mL/day. More flow only comes after more carbon, in that order, and with an effluent test after each step.

The literature, in one unit

Most studies give the carbon a reactor needs as COD, chemical oxygen demand: the oxygen it would take to burn the carbon completely, which counts the electrons it can give. One gram of acetate carbon is 2.67 g COD, and a millilitre of 8% vinegar about 85 mg.

  • Stoichiometry: 2.86 g COD per g nitrate-N. Nitrate to nitrogen gas takes five electrons per nitrogen atom. That is the floor if no carbon went anywhere else.
  • In practice 3–6, because part of the carbon builds bacteria and part burns the feed's oxygen [1]. Acetate-fed biofilters ran without nitrite from a C/N of 3.0–3.2, and acetate gave the fastest denitrification of the carbon sources tested [2] (the abstract does not say whether that ratio counts carbon or COD).
  • Too little carbon leaves nitrite. In acetate-limited activated sludge, at C:N 1:1 more than half the nitrate stopped at nitrite; balance returned at 2:1, but only after several weeks [3]. That abstract does not define its ratio either, so it cannot be compared directly with the numbers here.
  • Too much makes sulfide, in seawater. A denitrification unit on a closed marine fish system worked best at 1 g of organic carbon per g of nitrate-N: less left nitrite, more produced sulfides [4]. The same two edges this reactor has.
  • Acetate is the safer carbon. Denitrifying marine recirculating-system water, acetate kept sulfide at 0.002–0.008 mg/L, where ethanol let it reach 0.26 [5]. Vinegar was the right choice.
  • This reactor: nitrite in the effluent up to a net 1.99 g carbon per g nitrate-N (about 5.3 g COD), clean at 2.01 and 2.13 (5.4–5.7 g COD). That is inside the practical 3–6, but two to two and a half times the carbon of the seawater optimum in [4]. Where the rest goes, a biofilm still growing or more oxygen than the 7 mg/L I assume, I don't know yet.

References. I read the abstracts, not the full papers.

  1. van Rijn J., Tal Y., Schreier H.J. (2006). Denitrification in recirculating systems: theory and applications. Aquacultural Engineering 34(3):364–376. doi:10.1016/j.aquaeng.2005.04.004
  2. Rocher V. et al. (2015). Nitrite accumulation during denitrification depends on the carbon quality and quantity in wastewater treatment with biofilters. Environmental Science and Pollution Research 22(13):10179–10188. doi:10.1007/s11356-015-4196-1
  3. Oh J., Silverstein J. (1999). Acetate limitation and nitrite accumulation during denitrification. Journal of Environmental Engineering 125(3):234–242. doi:10.1061/(ASCE)0733-9372(1999)125:3(234)
  4. Sauthier N., Grasmick A., Blancheton J.P. (1998). Biological denitrification applied to a marine closed aquaculture system. Water Research 32(6):1932–1938. doi:10.1016/S0043-1354(97)00406-5
  5. Letelier-Gordo C.O., Huang X., Aalto S.L., Pedersen P.B. (2020). Activated sludge denitrification in marine recirculating aquaculture system effluent using external and internal carbon sources. Aquacultural Engineering 90:102096. doi:10.1016/j.aquaeng.2020.102096

The tank side

  • Nitrate 37.5 ppm on 31 October, 32.5 on 6 November, 31.7 on 9 November, 33.4 on 14 November, 28.9 on 16 November. Single readings scatter by a ppm or two; the trend is down. Fitted over all of them, the tank makes about 1.3 ppm of nitrate a day (1.1–1.5). That fit spans the weeks after the tank-wide vinegar stopped on 21 October, while the bacteria it fed died back, so production may still be settling. At the current 29 mL/min nitrate should level off near 25 ppm (near 24 at 30). Only more flow lowers that level; carbon cannot. The model keeps this up to date.
  • Nitrite in the tank stayed between 0.05 and 0.15. On 16 November it was still 0.1–0.15 with none leaving the reactor, so that is the tank's own level, not the reactor's.
  • Alkalinity rose from 6.7 to 7.3 dKH in four days once removal started, as the side-effect section above predicts. The NaOH dose came down 16% and alkalinity has held at 7.3 since (11 and 15 November).
  • Phosphate went from 0.44 to 0.33 in a week, then held (0.34 on 14 November). Stopping the phosphate dose explains only about a third of that; the rest is biofilm uptake. The floor warning above still stands.
  • Bacterial supplements stopped on 1 November. An established bed gains little from inoculation, and the product in use reportedly contains a sulfur-reducing genus and an acetate-eating methanogen, harmless while nitrate persists through the bed but unwelcome at exactly the near-zero edge this reactor is being steered toward.

What I would tell myself on day one

  1. Match carbon to flow from the start. Begin at a low flow and climb both together.
  2. Measure effluent nitrite, diluted in one step. Check what your nitrate kit actually measures before trusting it on effluent.
  3. Route the outlet up to a vent, then into an open drop. Never a full downward tube.
  4. Count the gas. It is the cheapest independent proof that the reactor works.
  5. Write down every change with its time, and change one thing at a time. Most of what this log could conclude, it could conclude because of that.
  6. Expect the second step, nitrite to nitrogen, to lag the first by weeks. Judge changes over several days, not overnight.
  7. Once the effluent reaches zero, every further gain comes from flow, but carbon has to lead: raise the vinegar first, then the flow, in small steps, testing effluent nitrite after each. Raising the flow alone, by 3.5%, brought nitrite back within a day. At a fixed flow, carbon only comes down.
  8. Know both edges of the carbon band: below it nitrite, above it sulfide. Near complete removal they are close together, and a small shortfall shows up as nitrite, not as nitrate.

Same ownership stance as the rest of this site's build writeups: this is my own tank, my own hardware, run and modified at my own risk. If any of this is useful background for your own build, the same applies to you.