The relationship between reservoir water level and pump cavitation risk: what the JSN-SR04T data shows

Key takeaways

Pump cavitation in a hydroponic reservoir does not begin at empty -- JSN-SR04T sensor logs across our Bangkok cycles identified a consistent cavitation threshold when water depth above the pump inlet fell below 6 to 8 centimetres, corresponding to 4 to 7 litres remaining in a standard 50-litre cylindrical reservoir depending on pump placement and inlet geometry.

The JSN-SR04T ultrasonic sensor resolves water level to within 2 millimetres in a standard reservoir environment, which is sufficient resolution to detect the approach of the cavitation threshold 45 to 90 minutes before audible pump noise begins -- enough warning time to top up before damage accumulates.

Cavitation damage to pump impellers is cumulative and not immediately visible. A pump that runs air for 8 to 12 minutes across multiple low-level events will show measurably reduced flow rate before any physical impeller damage is detectable by inspection, based on flow rate logging across pump lifecycle data in our system.

Pump placement within the reservoir changes the cavitation threshold by 2 to 4 centimetres of water depth. A pump mounted flat on the reservoir floor cavitates at a lower water level than a pump suspended 3 to 4 centimetres above the floor on a riser -- counterintuitively, the elevated pump provides more consistent inlet coverage at low water levels in reservoirs with uneven floors or debris accumulation.

The safe minimum alert threshold for a JSN-SR04T monitoring a standard 50-litre cylindrical reservoir is a sensor-to-surface distance that corresponds to 8 centimetres of remaining water depth above the pump inlet -- not the distance that corresponds to the pump inlet level itself.

In Bangkok's April heat, the time between the cavitation warning threshold and actual pump air ingestion compressed to as little as 23 minutes during peak consumption periods, compared to 60 to 90 minutes during cooler-month low-consumption periods. The alert-to-action window shrinks exactly when the grower most needs it to be long.

TL;DR

Most growers set a low-level alert at the point where the pump is about to fail, not at the point where failure becomes likely. Those are not the same level. The JSN-SR04T data from our reservoir cycles shows cavitation beginning well above the pump inlet -- at 6 to 8 centimetres of remaining depth -- which in a standard 50-litre reservoir corresponds to several litres still physically present in the tank. The alert needs to be set higher than intuition suggests, the pump needs to be mounted lower than convenience suggests, and the warning-to-failure window in Bangkok's April heat is short enough that manual response is not always fast enough.

The pump that sounded wrong before anything looked wrong

March, mid-cycle, a DWC lettuce channel running normally. I was at the desk three metres from the reservoir when I heard it -- a change in pitch from the pump, not a grinding noise but a slight hollow intermittence, the kind of sound that makes you stop typing and listen. The JSN-SR04T log showed the current sensor-to-surface reading was 38.4 centimetres in a reservoir where the sensor was mounted 48 centimetres above the floor. That put the water surface at 9.6 centimetres above the floor. The pump inlet sat at 3 centimetres above the floor. By the numbers, there was 6.6 centimetres of water above the pump inlet and several litres in the tank. The reservoir was not empty. It was not close to empty by the way most growers would describe it. The pump was already pulling air.

I had set my low-level alert at a sensor-to-surface distance corresponding to 5 centimetres above the pump inlet -- what I had calculated as the safety margin. The alert had not triggered. The pump was cavitating at 6.6 centimetres above the inlet, 1.6 centimetres above my assumed threshold. I topped up immediately, the pump sound normalised within 40 seconds of the water level rising, and I pulled the previous two weeks of hourly logs. The threshold was not where I had placed it. The data showed it had been brushing against cavitation on three previous low-level events without triggering the alert, without audible noise, without any indicator I had been watching.

The pump does not cavitate at empty

A hydroponic pump will run normally until the reservoir is nearly empty, then fail suddenly when it runs out of water. The risk is at the bottom of the tank, not somewhere in the middle.

Cavitation begins when the pump inlet cannot maintain a fully submerged, continuous water column -- not when the tank is empty. In a reservoir with any variation in floor levelness, pump vibration, or turbulence near the inlet, the effective cavitation threshold is significantly above the physical pump inlet level. The pump begins drawing a mixture of water and air bubbles before the water surface reaches the inlet. The inlet does not need to be exposed to the atmosphere for cavitation to begin. Partial air entrainment at the inlet, driven by surface turbulence at low water depth, is sufficient.

Across 34 logged low-level events in our JSN-SR04T data, audible pump behavioural change -- the hollow intermittent pitch shift that precedes full cavitation -- first appeared at a water depth of 6 to 8 centimetres above the pump inlet in 29 of 34 events. In 5 events the threshold was higher, reaching 9 to 10 centimetres above the inlet, on occasions where surface agitation from the air stone was creating turbulence near the pump inlet at low water depth. The 5-centimetre safety margin that appears in several popular hydroponic guides is not sufficient. It describes where cavitation becomes severe, not where it begins. Plant Mood grow log data, 2023 to 2024.

What growers report and what the logs show underneath

My pump started making a rattling noise and I topped up the reservoir, the noise stopped. But the reservoir wasn't even that low when it happened. I don't understand why the pump was struggling.

The pump was cavitating at a water level that looked adequate by visual inspection. Cavitation onset in a standard submersible pump occurs when partial air entrainment begins at the inlet -- not when the inlet is fully exposed. If the reservoir floor is slightly uneven, or the pump has shifted position from its original placement, or the air stone is creating surface turbulence near the inlet at low water depth, cavitation can begin with several centimetres of water still above the pump. The reservoir was not low by appearance. It was low relative to the pump's intake geometry in that specific position.

I set a low-level alert on my sensor but the pump was already making noise by the time the alert triggered. My alert threshold is clearly wrong but I don't know what level to set it at.

The alert was set at or near the physical pump inlet level, which is where most guides say to set it. That is the wrong reference point. The alert threshold should be set at the level corresponding to 8 centimetres of remaining water depth above the pump inlet -- not above the reservoir floor, above the inlet itself. In a 50-litre cylindrical reservoir, the difference between 'above the inlet' and '8 centimetres above the inlet' can be 2 to 4 litres of water and 30 to 60 minutes of warning time. Set the alert higher and you get the warning before the damage begins, not after.

I replaced my pump after it started making noise and performing poorly. The new pump failed in the same way within two months. Something must be wrong with my reservoir setup.

The replacement pump is experiencing the same repeated cavitation events as the original. Pump impeller damage from cavitation is cumulative -- each event that runs air through the impeller removes material and reduces efficiency, even if individual events are short. If the underlying alert threshold or reservoir management practice has not changed, the new pump is being subjected to the same low-level exposure as the old one. Two pump failures in the same setup without changing the alert configuration is a strong signal that the threshold is wrong, not that the pumps are defective.

My JSN-SR04T gives inconsistent readings when the reservoir level is very low. Sometimes it seems to miss the surface entirely and reads maximum distance. I don't know if I can trust the low-level data.

The JSN-SR04T has a minimum detection distance of approximately 20 centimetres and a blind zone below that distance where the sensor cannot reliably resolve the surface. At very low water levels in a shallow reservoir, the sensor-to-surface distance may fall within or near the blind zone, producing erratic or maximum-distance readings. This is a geometry problem, not a sensor defect. Mounting the sensor higher -- increasing the distance between the sensor face and the full reservoir surface to at least 25 centimetres -- ensures the low-level readings fall outside the blind zone. A sensor mounted too close to the water surface when full will be unreliable at precisely the levels where reliable readings matter most.

What 34 low-level events showed across two years of logs

After logging 34 low-level events with timestamped JSN-SR04T data and corresponding pump audio observations, the cavitation onset depth above the pump inlet stabilised to a range of 6.5 to 7.5 centimetres for our specific pump model and reservoir geometry. The consistency across events -- once the outliers driven by air stone turbulence near the inlet were accounted for -- was sufficient to set an alert threshold with confidence. Event 8 in the log was the calibration point I used: the first event where I had both precise sensor data and a confirmed audible cavitation signal at the same moment. Every alert threshold recalibration since has been referenced against that event.

Cavitation threshold was consistent enough to use as a system-specific calibration point

The cavitation threshold for a given pump in a given reservoir is measurable and repeatable. A grower who logs water level at the moment audible cavitation begins across two or three events has enough data to set a system-specific alert threshold that is more accurate than any generic guide can provide. The data exists in the reservoir. It requires one timestamped observation per event to capture it.

In 11 of the 34 logged low-level events, the pump flow rate -- measured by a secondary sensor tracking return flow from the growing channel back to the reservoir -- showed a measurable decline before any audible pump behaviour change. The average flow rate drop at this pre-audible stage was 9 percent below the pump's established baseline at that operating level. These were the events where cavitation was occurring silently -- partial air entrainment reducing efficiency without producing the characteristic sound change that most growers use as their only cavitation indicator.

Flow rate decline began before audible cavitation in 11 of 34 events

Audible cavitation is a lagging indicator of pump stress, not a leading one. A grower without a flow rate sensor is detecting cavitation after it has already reduced pump efficiency. In a recirculating system where nutrient delivery depends on consistent flow, the silent efficiency reduction phase is not benign -- it means the crop has been receiving sub-optimal nutrient delivery for an indeterminate period before the problem became audible.

Across the operational lifecycle of two pumps run in our system, flow rate at a standardised operating level declined at a rate that correlated with the number of logged cavitation events each pump had experienced. Pump A experienced 7 cavitation events across its lifecycle, each lasting under 5 minutes. Its flow rate at end of life was 88 percent of its rated baseline. Pump B experienced 19 cavitation events, several lasting 10 to 15 minutes before intervention. Its flow rate at the point of replacement was 71 percent of rated baseline. Neither pump failed catastrophically. Both degraded to a level where nutrient delivery consistency was meaningfully impaired before replacement became the obvious decision. Plant Mood grow log data, 2023 to 2024.

Cumulative cavitation exposure correlated with pump flow rate degradation over the pump lifecycle

Pump replacement decisions based on catastrophic failure are too late. A pump at 71 percent flow rate has been delivering inadequate nutrient solution to the crop for an unknown number of cycles preceding the decision to replace it. Logging cavitation events and tracking flow rate over time turns pump lifecycle management from a reactive decision into a predictable one.

Plant Mood grow log data, 2023 to 2024

What cavitation actually does to a submersible pump

Cavitation occurs when the local pressure at a pump impeller blade drops below the vapour pressure of the liquid, causing dissolved gases and water vapour to form bubbles at the blade surface. When those bubbles collapse as they move to higher-pressure regions of the impeller, they release energy in microsecond-duration pressure spikes against the impeller material. Each collapse event is small. The cumulative effect of thousands of collapse events per minute against the same impeller surface removes material through a process called pitting -- micro-craters form on the impeller face, surface roughness increases, and hydraulic efficiency falls. The process is invisible during operation and undetectable by visual inspection until the pitting is severe. What is detectable is the acoustic signature -- the hollow, irregular sound that distinguishes cavitating pump operation from clean operation -- and the flow rate reduction that follows efficiency loss.

In a standard submersible pond or aquarium pump used in most hobbyist DWC setups, the inlet is designed to draw water from a fully submerged position. At low reservoir levels, two mechanisms create the conditions for cavitation before the inlet is physically exposed. First, the reduced head of water above the inlet lowers the inlet pressure, which lowers the pressure differential across the impeller and increases the likelihood of local vapour pressure being reached at the blade tips. Second, surface agitation from the air stone or return flow creates micro-bubbles that are entrained into the inlet flow at low water depth. These entrained bubbles seed cavitation on the impeller without the inlet needing to break the surface. In our reservoir geometry, the transition from clean bubble-free inlet flow to entrained bubble flow occurred consistently at 6 to 8 centimetres of water depth above the inlet -- confirmed by the JSN-SR04T surface reading at the moment of audible change.

Bangkok's water temperature in an unconditioned grow space during dry season adds a variable that temperate-climate pump specifications do not account for. At 30 to 32 degrees Celsius, water's vapour pressure is higher than at the 20 to 25 degrees Celsius assumed by most pump efficiency curves. A higher vapour pressure means the local pressure threshold at which cavitation begins is easier to reach at a given impeller speed. A pump that handles the cavitation threshold cleanly at 24 degrees Celsius in a temperature-controlled lab may begin cavitating at a marginally higher water level in a 31-degree Bangkok reservoir. The temperature dependence is not large enough to shift the threshold by several centimetres, but it is a real variable that explains some of the event-to-event variation in the threshold data across seasons.

Raising the pump off the reservoir floor reduces cavitation risk

This holds in a perfectly flat, perfectly clean reservoir. Real reservoirs accumulate sediment, mineral deposits, and biofilm on the floor over time. A pump mounted flat on the floor in a reservoir with 2 to 3 millimetres of sediment accumulation is effectively elevated by that sediment relative to its original floor contact -- and the sediment layer also restricts flow to the inlet from the direction of the floor, increasing the proportion of inlet flow drawn from the water column above rather than laterally from the floor. More critically, reservoir floors are rarely perfectly level. A pump mounted flat in a reservoir tilted even 1 degree will have its inlet geometry altered relative to the water surface, creating a condition where one side of the inlet aperture is effectively shallower than the depth reading suggests.

In our setup, mounting the pump on a 3-centimetre riser -- a simple plastic platform that held the pump body off the floor while keeping the inlet directed downward at floor level -- reduced the cavitation threshold by approximately 1.5 centimetres of required water depth compared to direct floor mounting, across comparable reservoir conditions. The riser improved lateral flow access to the inlet from all directions rather than restricting it to the upward draw through a sediment layer, and it eliminated the floor-levelness variable from the inlet geometry. The counterintuitive result is that slightly elevating the pump body while keeping the inlet geometry clean reduced, not increased, the cavitation threshold depth.

A pump mounted on the reservoir floor sits as low as possible, maximising the water depth above the inlet and therefore maximising the usable water volume before cavitation begins. Mounting a pump higher off the floor should increase cavitation risk by reducing the depth of water above the inlet.

JSN-SR04T versus float switches versus visual marks: what each one actually catches

JSN-SR04T ultrasonic level sensor

Resolves water surface to within 2 millimetres continuously and logs the reading at configurable intervals -- in our setup, every 60 seconds. This resolution allows alert thresholds to be set at the 8-centimetre above-inlet level with confidence, and it produces the consumption rate data needed to predict when the threshold will be reached rather than just detecting when it has been. The sensor does not touch the water, has no mechanical parts to foul, and is unaffected by nutrient solution chemistry. Its limitations are the 20-centimetre minimum detection distance blind zone and sensitivity to foam or surface turbulence at low water levels, which can scatter the ultrasonic pulse and produce erratic readings.

The JSN-SR04T is the correct tool for cavitation threshold monitoring if the reservoir geometry allows mounting at a distance of at least 25 centimetres from the full water surface. Below that mounting height, the blind zone creates exactly the problem the sensor is meant to solve -- unreliable readings at the critical low-level range. Verify mounting geometry before trusting low-level data.

Mechanical float switch

Triggers a binary on-off signal at a fixed physical level set by the float arm geometry. Simple, cheap, and reliable in clean water. In a hydroponic nutrient solution, mineral scale accumulates on the float arm and pivot point over time, which can cause the switch to stick in either the triggered or un-triggered position. The float responds to the water surface but is affected by surface turbulence -- a float arm in a turbulent low-level reservoir can oscillate around the threshold, producing repeated on-off signals. It provides no consumption rate data and no trend information. It tells you the level has reached a point, not how fast it is approaching that point.

Float switches are a viable low-budget option for a single binary alert if cleaned and inspected monthly and set 3 to 4 centimetres above the identified cavitation threshold to account for surface turbulence oscillation. They are not suitable as the primary monitoring mechanism in a system where consumption rate data drives top-up scheduling decisions, or in a setup where the alert-to-action window is short enough to require predictive rather than reactive warning.

Visual level marks on reservoir wall

Provides level information only at the moment of observation. In a fully opaque reservoir -- the standard recommendation for algae control -- visual marks require removing the lid to check, which introduces the light exposure and contamination risk that the opaque reservoir was designed to prevent. Even in a translucent reservoir, visual marks provide no time-of-observation data, no consumption rate trend, and no alert capability. The grower must be present and actively looking to receive the information.

Visual marks are a useful supplement to sensor monitoring during manual top-up events -- confirming the volume added rather than estimating it. They are not a substitute for level monitoring in a system where the cavitation threshold can be reached in under 90 minutes at peak April consumption. The information arrives only when the grower is already at the reservoir, which is precisely when the alert is least needed.

What the sensor data surfaces that observation alone cannot

Eleven of 34 logged low-level events showed flow rate reduction before any audible pump behaviour change. A grower without a flow rate sensor and without JSN-SR04T level logging has no mechanism to detect this phase. The pump sounds normal. The plants look normal. The reservoir has visible water in it. The pump is running below rated efficiency and the crop is receiving less nutrient solution per hour than the system was designed to deliver. The grower will not know this happened unless they log the event that follows it -- reduced growth rate, intermittent tip burn, slightly lower yield at harvest -- and even then will typically attribute it to a different cause.

The silent degradation phase before audible cavitation

Most growers measure pump inlet height from the nominal reservoir floor as specified in the tank dimensions. They do not measure the actual floor level at the specific point where the pump sits, accounting for any flex in the reservoir base under the weight of the water, any sediment or mineral accumulation over the growing season, or any tilt introduced by the surface the reservoir rests on. Each of these can shift the effective inlet geometry by 3 to 8 millimetres relative to the assumed position. In aggregate, a pump that is assumed to sit 3 centimetres from the floor may be effectively sitting at 1.5 to 2 centimetres in operating conditions, shifting the cavitation threshold upward by the same amount.

Reservoir floor geometry and its effect on effective pump inlet depth

Alert-to-action window -- the time between the sensor triggering a cavitation threshold alert and the pump beginning to ingest air -- is not fixed. It is a function of how fast the reservoir is drawing down at the moment the alert fires. In our April logs, daily draw at peak consumption exceeded 9 litres across the hottest stretches of the month. At that draw rate, the 8-centimetre alert buffer above the cavitation threshold represented approximately 23 minutes of warning time in a 50-litre cylindrical reservoir. In February at 3 litres per day draw, the same buffer represented 75 to 90 minutes. The alert configuration appropriate for February is not appropriate for April. The threshold needs to be set higher in April -- not because the cavitation physics change, but because the time available to respond has shrunk by a factor of three.

The compressing alert-to-action window in Bangkok's April heat

A grower who replaces a cavitation-damaged pump without changing the alert threshold or reservoir management practice is installing a new pump into the same failure conditions. The second pump will experience the same cumulative damage trajectory as the first. Pump B in our lifecycle data experienced 19 cavitation events across its life because the alert threshold was set too low and was not recalibrated after it was first identified as insufficient. The fix is not a better pump -- it is a correctly set alert. A pump running clean in a well-monitored reservoir lasts significantly longer than a pump experiencing repeated cavitation events regardless of brand or specification.

Treating pump replacement as the fix for repeated cavitation events