Technical Deep Dive · Engineering Perspective · Published: September 28, 2026
Table of Contents
Key Takeaways
- The US EPA’s 2013 freshwater criteria set an acute ammonia magnitude of 17 mg TAN/L at pH 7 and 20 °C (1-hour average) and a chronic magnitude of 1.9 mg TAN/L (30-day average), with the 4-day average capped at 2.5× the chronic value (4.8 mg TAN/L). The permitted total ammonia falls steeply as pH and temperature rise [1]
- Toxicity is driven by the un-ionized NH3 fraction, which climbs with pH and temperature. TAN alone tells an incomplete story, because the same TAN is far more dangerous at pH 8 than at pH 7 [1][3]
- Ammonia load is a feed problem before it is a filter problem. A biofilter sized for the ammonia from 40 kg of feed per day will accumulate ammonia if you feed 60 kg [4]
- Nitrifying bacteria double in 8–24 hours under optimal conditions and in 24–48 hours in the field, so a biofilter takes weeks to mature and recovers slowly after a crash [5]
- Every gram of ammonia nitrogen oxidized consumes about 7.14 grams of alkalinity as CaCO3, which is why pH slides and nitrification stalls travel together [5]
- Online TAN monitoring at the biofilter inlet and outlet, paired with pH, DO and temperature, turns ammonia from a post-mortem test into an operating control variable — and feeds the data your AI water model actually needs.
Where the Ammonia Comes From
In a recirculating aquaculture system (RAS), ammonia is not an accident; it is the normal exhaust of protein metabolism. Fish excrete most of their nitrogenous waste as ammonia directly across the gills, with smaller contributions from feces and urine. Uneaten feed and decomposing organic solids add a second stream, and any mortality event adds a third: decomposing biomass releases a pulse of ammonia that can dwarf routine production. The biological filter — a bed of nitrifying bacteria — is the only thing standing between that continuous production and accumulation, because in a closed loop there is no dilution stream to carry ammonia away.
That is why feed rate is the master variable. Operators who feed “a little extra to be safe” are raising ammonia risk in direct proportion: a system that can biologically process the ammonia from 40 kg of feed per day will accumulate ammonia at 60 kg [4]. The same arithmetic explains why harvest windows, grading events and temperature-driven appetite changes so often precede ammonia excursions — each one moves the load without moving the filter.
TAN versus NH3: The Toxicity Distinction That Decides Everything
Total ammonia nitrogen (TAN) is the sum of two chemical species in equilibrium: ionized ammonium (NH4+), which is tolerable to fish at the concentrations RAS plants see, and un-ionized ammonia (NH3), which diffuses across gill membranes and damages respiratory, circulatory and reproductive systems. The equilibrium between them is governed by pH and temperature, with a pKa near 9.25: as pH increases toward that value, and as temperature rises, the fraction present as toxic NH3 increases [3].
The regulatory anchor for why this matters is the EPA’s 2013 freshwater criteria. The recommended acute criterion is 17 mg TAN/L at pH 7 and 20 °C for a one-hour average, and the chronic criterion is 1.9 mg TAN/L at the same conditions for a 30-day average, with the highest four-day average inside any 30-day period not to exceed 2.5 times the chronic magnitude — 4.8 mg TAN/L [1]. The critical nuance for operators is that these magnitudes are stated at reference pH and temperature: at higher pH the permissible TAN drops sharply, because the NH3 share climbs. A TAN reading of 2 mg/L that is routine at pH 6.8 becomes a real emergency at pH 8.2 with warm water. Seasonal behavior compounds this. In pond-based and outdoor-influenced systems, ammonia concentrations tend to run higher in the colder months than in summer — YSI’s aquaculture application note puts typical summer pond levels below about 0.5 mg/L — and cold water is exactly when fish immune systems are suppressed [2].
The practical consequence for monitoring design is that TAN, pH and temperature must be measured together, ideally by the same multi-parameter instrument, synchronously. None of the three numbers means anything alone. Converting TAN into its NH3 fraction requires the pH and temperature at the same moment; a probe that samples pH an hour away from the TAN sample produces a toxicity estimate that is fiction.
Why Spikes Track Biofilter Health
Most ammonia spikes are biofilter events wearing a water-quality costume. The bacteria that oxidize ammonia (to nitrite) and nitrite (to nitrate) are slow-growing autotrophs: under optimal conditions their doubling time is 8–24 hours, versus hours for the heterotrophs that decompose organic waste, and in field conditions 24–48 hours is common [5]. Two consequences follow. First, a new biofilter takes weeks to establish — which is why commissioning periods are the most dangerous phase of a new RAS. Second, after any setback the filter recovers slowly, so a crash that lasts a day can degrade nitrification for weeks.
The filter’s known failure modes are few and recognizable. Nitrification operates best around pH 7.5–8.5; below pH 7.0 rates drop significantly, and the process itself consumes alkalinity — about 7.14 grams as CaCO3 per gram of ammonia oxidized [5]. Aggressive backwashing strips biofilm. Some disinfectant residues, temperature excursions and oxygen starvation in the media all bite the same population. Because the organisms are slow to rebuild, the response to a spike has to protect the filter first and chase the number second.
| Spike trigger | What online data shows first | Recommended response |
|---|---|---|
| Feed load above biofilter capacity | slow TAN rise over days, steady at first | cut ration toward the filter’s verified capacity; verify biofilter DO [4] |
| Overfeeding / uneaten feed | TAN drift plus rising turbidity or solids | tighten feeding, improve solids removal upstream [4] |
| Mortality event | sudden TAN jump within hours | remove mortalities immediately; increase make-up water exchange temporarily [4] |
| Alkalinity exhaustion | pH sliding down, then TAN climbing | dose alkalinity; watch pH before TAN moves [5] |
| pH crash below ~7.0 | low pH with slowing nitrification | buffer pH and address CO2 buildup; expect slow recovery [5] |
| Over-aggressive backwash | nitrite rising with TAN | reduce backwash intensity/frequency; reseed if needed |
| Immature or newly commissioned filter | TAN rising while biofilter matures (weeks) | delay stocking increases; seed bacteria; monitor continuously [5] |
Monitoring Points and the Response Playbook
Where you measure decides what you can control. In a well-instrumented RAS, three points carry the ammonia story:
- Biofilter inlet. This is the load measurement: TAN arriving at the filter, alongside DO (nitrification is aerobic and oxygen-hungry) and pH. It tells you what the filter is being asked to do this hour.
- Biofilter outlet. This is the performance measurement. The TAN difference across the filter is the only honest measure of what the media is actually removing, and a widening gap between inlet and outlet is the earliest quantitative signal of filter trouble — visible days before either number breaches a tank alarm [6].
- Culture tank / return loop. This is the welfare measurement: the TAN and NH3 fraction the fish actually experience, alongside DO and temperature. Alarm thresholds here should be set from the pH- and temperature-corrected NH3 fraction, not raw TAN.
The response playbook mirrors the table above. When TAN trends up over days, reduce load and check the filter’s oxygen. When it jumps within hours, look for mortality, overfeed, or a washout event. When pH and TAN move together, restore alkalinity before the filter stalls. Trending matters more than any single reading: a healthy system shows controlled TAN, controlled nitrite, and steadily increasing nitrate [6].
One further step separates good RAS operations from great ones: streaming the synchronized TAN, pH, DO and temperature signals into a digital model. Sensors that feed your AI water model, not just your dashboard, can learn the site-specific signature of each failure mode — the two-day drift that means overfeeding, the pH-TAN coincidence that means alkalinity — and alert on the signature rather than the breach. Shanghai ChiMay’s multi-parameter sensing platforms with Modbus RTU/TCP digital output are built for exactly that data path; the company’s application notes on multi-parameter sensor integration in RAS describe the architecture in detail [7].
From Test Kit to Control Variable
Manual ammonia test kits remain useful for verification, but as a control input they are structurally too slow. A daily grab sample cannot see a spike that develops between feedings, and by the time fish behavior changes, the toxic fraction has been elevated for hours. Continuous in-line monitoring closes that gap and, just as importantly, produces the synchronized multi-parameter record that turns every spike into training data for the next prevention. Buy the sensor, own the outcome: in ammonia control, the outcome is a filter that never gets pushed past its capacity without you knowing it — and fish that never pay for the experiment.
References
- US EPA — “Aquatic Life Ambient Water Quality Criteria for Ammonia — Freshwater (2013)” fact sheet — acute 17 mg TAN/L (1-hr) and chronic 1.9 mg TAN/L (30-day) at pH 7, 20 °C; 4-day average ≤ 2.5× chronic (4.8 mg/L) — https://www.epa.gov/sites/production/files/2015-08/documents/fact_sheet_aquatic-life-ambient-water-quality-criteria-for-ammonia-freshwater-2013.pdf
- YSI (Xylem) Application Note A585 — “Understanding Ammonia in Aquaculture Ponds” — seasonal ammonia behavior in ponds (summer levels commonly below about 0.5 mg/L) and EPA criteria context — https://www.ysi.com/File%20Library/Documents/Application%20Notes/A585-Understanding-Ammonia-in-Aquaculture-Ponds.pdf
- Science of the Total Environment (ScienceDirect) — “Development of water quality criteria of ammonia for protecting aquatic life in freshwater using species sensitivity distribution method” — NH3 fraction rises with pH and temperature (Emerson et al., 1975); un-ionized ammonia is the toxic species — https://www.sciencedirect.com/science/article/abs/pii/S0048969718311641
- Agrinovo — “Ammonia Spikes in Fish Ponds: Causes, Detection & Response” — feed-load arithmetic (40 vs 60 kg/day), mortality decomposition as spike source — https://agrinovo.io/blog/ammonia-spikes-fish-ponds-detection-emergency-re
- FarmHub — “Your Ammonia Test Is Lying to You. Here Is What Actually Kills Fish.” — nitrifier doubling times (8–24 h optimal, 24–48 h field), pH < 7.0 effect, 7.14 g alkalinity per g ammonia oxidized — https://learn.farmhub.ag/articles/your-ammonia-test-is-lying-to-you-here-is-what-actually-kills-fish
- YUTANK — “RAS Nitrification Process Explained: How Biofilters Remove Ammonia” — trend-based biofilter evaluation (TAN, nitrite, nitrate, pH, alkalinity) and causes of poor nitrification — https://en.yutanke.com/blog-detail/ras-nitrification-process-explained-how-biofilters-remove-ammonia
- Shanghai ChiMay (chimaytech.net) — “Multi-Parameter Sensor Integration in Recirculating Aquaculture Systems” — sensor placement, calibration protocols, and RAS monitoring architecture — https://chimaytech.net/fr?p=30813
About the author: Written by the Shanghai ChiMay Technical Editorial Team — the engineering group behind Shanghai ChiMay’s online water quality analyzers and multi-parameter sensors for aquaculture and RAS, covering dissolved oxygen, ammonia-adjacent parameters, and digital integration for recirculating systems.
