Predicting reservoir top-up day from EC depletion slope and level sensor data

Key takeaways

EC depletion slope, measured as the daily drop in mS/cm over a rolling 3-day window, predicts reservoir top-up timing within plus or minus one day during stable growth phases. Accuracy drops sharply at growth transitions, heat events, and canopy density changes. Plant Mood grow log data, 2024.

Water level drop rate and EC depletion slope are not the same signal. A reservoir can lose water volume at a steady rate while EC drops erratically, or hold water level while EC falls faster than expected. Using both together produces a more reliable top-up prediction than either variable alone.

The formula breaks most often when plants transition from vegetative to pre-harvest growth acceleration. EC uptake spikes 30 to 50 percent above the established slope in the 4 to 7 days before harvest maturity in leafy greens, making the final top-up interval shorter than the model predicts.

In Bangkok's wet season, ambient humidity above 80 percent suppresses transpiration enough to reduce water level drop rate by 20 to 35 percent without changing EC depletion rate proportionally. A formula calibrated during the dry season will overestimate reservoir longevity during the wet season.

A formula can replace the volume check and the timing estimate. It cannot replace the smell check, the root inspection, or the visual read of canopy behaviour. These carry information the sensors do not capture.

First-year growers running a 10 to 20 tray hydroponic operation without any automated logging will structurally top up too late in the final week of each cycle because EC and water level both fall fastest when the plants are largest and the checks feel least urgent.

The top-up I was three days late on

It was the third week of a lettuce cycle. Everything had been tracking well enough that I had relaxed my check schedule. The canopy was full, the plants looked strong, and the previous two weeks had been predictable. I had been topping up every five days like clockwork and I assumed week three would be the same.

It was not. When I finally lifted the lid on day six, the water level was critically low. The EC had climbed to 3.1 mS/cm, nearly double the target range for that stage, because the plants had been pulling water faster than nutrients in the final push before maturity, concentrating what was left in the reservoir. The outer leaves on the lower channels were showing tip burn. The root tips had gone from white to a dull cream. The crop was not lost but the final week yield was reduced and the texture was off. Bitter. The kind of thing a restaurant buyer notices.

I had the level sensor data. I had been logging EC manually every other day. I had everything I needed to have caught this four days earlier. What I did not have was a way to turn those two numbers into a forward-looking answer instead of a backward-looking record.

That cycle was the one that made me want a formula. Not to replace checking, but to stop relying on the assumption that last week's pattern would hold this week.

The formula is useful. It is not sufficient.

A top-up prediction formula built from EC depletion slope and water level drop rate is genuinely worth building and worth using. It will catch the timing problem I described more often than a fixed check schedule will. For a one-person operation running multiple towers, it is a real force multiplier because it turns two streams of data you are already logging into an actionable number rather than a log you review after something goes wrong.

But I do not think it replaces the daily check, and I want to be clear about why. The formula models what the plant has been doing. It does not know what the plant is about to do. And plants in a Bangkok apartment in the final week of a growth cycle do not behave the way they did in week two. EC uptake accelerates. Transpiration rate changes. Root zone behaviour shifts. The slope you calculated from days 7 through 14 is not the slope that governs days 15 through 21.

The growers who will over-rely on this formula are the ones who build it, see it work for two or three cycles, and then stop checking because the model has been right. That is exactly when the model will miss, because the conditions that made it accurate were stable conditions, and stability in a Bangkok hydroponic grow is not the default state.

The safety margin argument works for timing. It does not work for quality. A formula tells you when to top up. It does not tell you that the reservoir smells slightly different than it did on day 3, or that the root tips have yellowed in a way that suggests the EC was already too high for 36 hours before you got there. Those observations change what you do when you top up, not just when. A grower who tops up on day 4 instead of day 5 but arrives with no context about what the system experienced is still flying partially blind. The formula narrows the window. It does not replace the read.

The reasonable counterargument is that a formula with a built-in safety margin, say predicting top-up one day earlier than the model suggests, handles the growth acceleration problem adequately. If the model says day 5, you act on day 4. The buffer absorbs the variability.

How wrong a fixed schedule gets it

The range of actual top-up intervals across a single 21-day lettuce grow cycle in our Bangkok operation, when measured from EC data rather than assumed from a fixed schedule. The interval in week one averaged 6.8 days. In week three it averaged 3.9 days. A grower using a fixed 5-day top-up schedule based on early-cycle behaviour would be topping up correctly in week one and significantly late in week three, which is when the crop is most sensitive to EC concentration stress. Plant Mood grow log data, 2024.

30 to 50 percent

The increase in EC uptake rate observed in the 4 to 7 days before harvest maturity in leafy greens, relative to the midcycle baseline slope. This acceleration is the primary reason a formula calibrated on week-two data underestimates final-week consumption. The plant is not behaving abnormally. It is behaving exactly as it should. The model just was not built for that stage.

20 to 35 percent

The reduction in water level drop rate during Bangkok wet-season weeks when ambient humidity holds above 80 percent. Transpiration slows when the vapour pressure deficit collapses. A formula that does not account for ambient humidity will overestimate how fast the reservoir is being depleted by transpiration, producing a top-up prediction that is earlier than reality. Plant Mood grow log data, 2024.

The fraction of our formula-predicted top-up dates that were accurate within one day across all growth phases, without seasonal or growth-stage correction factors applied. With a rolling 3-day EC slope and current water level drop rate, the formula produced a useful prediction in stable midcycle conditions and a less reliable one at both ends of the growth cycle and across season transitions.

How EC slope and water level combine into a prediction

The logic of the formula is straightforward. EC depletion slope is the rate at which the reservoir loses nutrient concentration over time, measured in mS/cm per day. Water level drop rate is the rate at which total reservoir volume decreases, measured in litres per day. Together they describe two boundaries: how long before the EC falls below the minimum threshold for plant growth, and how long before the physical water volume becomes insufficient for the pump to circulate properly.

The top-up is triggered by whichever boundary is reached first. In practice, during early and midcycle growth, water volume is usually the binding constraint. The reservoir drops below the minimum operating level before EC falls below target. In final-week growth, EC concentration becomes the more relevant limit, because plants are pulling water faster than nutrients and concentrating the remaining solution, pushing EC upward toward stress range rather than downward toward depletion.

The formula in its basic form takes current EC, subtracts the minimum target EC, divides by the 3-day rolling slope of EC change, and produces a day count. It runs the same calculation for water level: current volume minus minimum operating volume, divided by the daily drop rate, producing a second day count. The earlier of the two is the predicted top-up day.

What the formula cannot encode is the reason for the slope. A slope of minus 0.2 mS/cm per day produced by a healthy plant in week two and a slope of minus 0.2 mS/cm per day produced by a root zone problem suppressing uptake in week three look identical in the data. The prediction is the same. The correct response is not.

In a June 2024 cycle, the formula predicted top-up on day 17 based on a 3-day EC slope of minus 0.18 mS/cm per day and a water level drop of 1.9 litres per day. The actual top-up was needed on day 14 because pre-harvest growth acceleration kicked in on day 12 and the slope steepened to minus 0.31 mS/cm per day over the following 48 hours. The formula had no way to anticipate the acceleration because it was calculated before it began. The canopy told the story earlier: the leaves had visibly darkened and the stem thickness on new growth had increased, both signs of accelerating vegetative growth rate that preceded the EC acceleration by about a day.

A more robust version of the formula incorporates a growth-stage correction factor applied from day 14 onward in a standard 21-day leafy green cycle: multiply the EC slope by 1.35 to account for the expected pre-harvest acceleration. This is not a published standard. It is a correction factor derived from our own cycle data and should be calibrated against your own crop and system before relying on it. The underlying principle, that EC uptake rate is not constant across the growth cycle and the final week systematically underperforms the midcycle prediction, is consistent enough across our logs to treat as a planning assumption rather than an exception.

What five cycles of tracking actually showed

Across five cycles logged with both continuous level sensor data and twice-daily EC readings, the formula predicted top-up timing within one day in 14 of 19 midcycle intervals, defined as days 7 through 14 of a 21-day cycle. In the first-week intervals, accuracy dropped because EC slope had not stabilised from the initial fill. In the final-week intervals, accuracy dropped because of pre-harvest acceleration. The crop variety, lettuce versus pak choi versus basil, had less effect on formula accuracy than the growth phase.

Formula accuracy was phase-dependent, not crop-dependent

The formula is most reliable in the window where growers are least likely to miss a top-up anyway, the predictable midcycle stretch. It is least reliable at the phase transitions where missing a top-up has the most consequence for yield quality. Build the correction factors for those phases explicitly rather than assuming the midcycle slope holds.

In two cycles that included multi-day periods with afternoon reservoir temperatures above 30 degrees Celsius, water level dropped faster than the EC slope predicted it should. The plants were transpiring at an elevated rate, pulling water faster than nutrients, which also concentrated the EC upward rather than allowing it to fall. The formula, which was treating both variables as parallel indicators of plant uptake, was reading two different underlying events: thermal transpiration stress on the water side and suppressed nutrient uptake on the EC side.

Water level drop rate and EC slope decoupled during heat events

Water level and EC are not interchangeable uptake signals. When they decouple, something has changed in the system beyond normal plant growth, almost always a temperature event in Bangkok conditions. Using both together exposes the decoupling. Using either alone hides it.

In the two cycles where the formula missed top-up timing by three or more days, both misses were preceded by observable changes at the tray level that were visible before the slope shifted enough to change the prediction. In one case, the root tips had developed a faint yellow-green tint indicating early algae pressure, which was suppressing uptake and holding EC artificially flat. In the second, the reservoir had a faint earthy smell on day 12 that preceded visible EC and level changes by two days. Neither of these signals appears in sensor data.

The smell and root check caught what the formula missed, twice

The formula's failure modes are not random. They are caused by biological events that leave physical traces before they change the numbers. A grower who is checking the formula prediction but not checking the roots and smelling the reservoir is removing the early warning layer that the sensors cannot provide.

Plant Mood grow log data, 2024

More data points make the formula less stable, not more

In practice, high-frequency EC logging introduces noise that makes the slope less stable rather than more. EC in an active hydroponic reservoir fluctuates slightly across the day based on transpiration rate, photosynthesis cycle, and temperature-driven changes in ion mobility. A reading taken at 08:00 will differ from one taken at 14:00 and again from one taken at 19:00, not because uptake has changed dramatically but because the pool of ions in the water is in constant local flux.

When you calculate a slope from twice-daily readings over three days, those intraday fluctuations introduce variance into the slope estimate that a once-daily reading at a consistent time does not. The slope calculated from six readings taken at variable points in the daily cycle is noisier than the slope calculated from three readings taken at the same time each morning.

The fix is not fewer data points but consistent timing. A once-daily EC reading taken at the same time each day, ideally at a consistent point in the light cycle such as one hour after lights-on, produces a cleaner slope than twice-daily readings taken whenever the grower has time. Consistency of timing matters more than frequency of measurement for slope accuracy.

When we compared slope variance between once-daily consistent-time EC logs and twice-daily variable-time logs across three cycles, the once-daily logs produced slope estimates with lower day-to-day variance and better top-up prediction accuracy. The twice-daily logs produced more data that was paradoxically less useful for prediction purposes. Plant Mood grow log data, 2024.

Logging EC more frequently, daily instead of every other day, or even twice daily, should produce a more accurate depletion slope and therefore a more reliable top-up prediction. More data points mean a more precise slope calculation.

EC slope alone is not a nutrient depletion signal

A declining EC slope tells you the plants are consuming nutrients and the reservoir needs topping up with fresh solution. The steeper the slope, the faster the plants are feeding.

EC slope is a net signal, not a pure uptake signal. EC declines when plants absorb ions faster than water evaporates. EC rises when water leaves the system faster than ions are absorbed, or when evaporation from an uncovered reservoir surface concentrates the remaining solution. EC holds flat when uptake and loss are balanced, or when root zone dysfunction has suppressed uptake entirely.

A grower who reads a declining EC slope as evidence of healthy plant feeding and a flat or rising EC as evidence of a well-balanced reservoir is interpreting the signal correctly in one common scenario and incorrectly in several others. Rising EC in a Bangkok grow during afternoon heat is almost never evidence of nutrient balance. It is usually evidence of thermal transpiration stress concentrating the solution while suppressed root function reduces ion uptake.

The depletion slope predicts when the reservoir needs intervention. It does not tell you what kind of intervention. A fresh nutrient top-up is the correct response to genuine EC depletion from healthy uptake. It is the wrong response to EC concentration from thermal stress. Both scenarios can produce the same slope value at a point in time, and the treatment for one makes the other worse.

In two documented cycles at Plant Mood, EC was rising on the morning check in a pattern that looked like straightforward concentration from evaporation. Standard interpretation would have been to add plain water to bring EC back down. Root and smell inspection on both occasions revealed early-stage root zone stress. Adding plain water to a thermally stressed reservoir with compromised roots would have further diluted dissolved oxygen without resolving the root cause. In both cases the correct intervention was thermal management first, not reservoir chemistry adjustment. Plant Mood grow log data, 2024.

Where this is going in the next few years

Growth-stage-aware top-up prediction will become standard in consumer hydroponic controllers within three years

The limitation of current EC-slope formulas is that they treat the growth cycle as a single continuous slope when it is actually two or three distinct phases with different uptake rates. The data to build phase-aware models exists in every logged grow operation. The barrier is not knowledge, it is implementation. As small-scale hydroponic controllers become more capable, the jump from a flat EC threshold alert to a growth-phase-corrected depletion forecast is a small engineering step with a meaningful grower benefit. Several hardware startups are already working in this direction. The first controller to ship this feature will make fixed-schedule top-up management look as dated as pH testing strips.

A commercial hydroponic controller available by 2027 that applies different EC depletion slope multipliers based on detected growth stage, either from day-count or from a canopy sensor, rather than using a single continuous slope for the entire cycle.

Bangkok-specific humidity correction factors will separate serious tropical hydroponic growers from those using temperate-calibrated formulas

The 20 to 35 percent reduction in water level drop rate during wet-season high-humidity periods is large enough to make a temperate-calibrated top-up formula consistently wrong for four to five months of the year in Bangkok. Growers who are logging their own data will discover this pattern and build local correction factors. Growers using default formula parameters from guides written in cooler climates will continue to find their predictions drifting during wet season without understanding why. The gap between these two groups will become visible in yield consistency data before it becomes visible in any published resource on tropical hydroponics.

Published grow logs from Bangkok or equivalent tropical apartment operations showing wet-season versus dry-season water level drop rate variance of less than 10 percent, which would indicate the humidity effect is smaller than our cycles suggest. If such data emerges across multiple operations, the correction factor would need revisiting.

The grower who eliminates daily checks entirely based on a sensor formula will lose a crop to an undetected biological event within six months

This is less a prediction than a structural observation. The failure modes of the EC slope and level sensor formula are biological: algae pressure suppressing uptake, root zone oxygen stress changing ion absorption, early pathogen colonisation altering the chemistry of the reservoir. None of these events change the EC slope or level sensor readings in a way that is distinguishable from normal plant behaviour until the biological event is already well established. A formula that is never cross-checked with a physical inspection removes the only detection layer that catches these events early. The six-month window is based on the average cycle count at which growers in stable setups encounter their first biological disruption.

A documented case of a grower running a purely formula-driven top-up schedule without physical checks for more than six months across multiple crops, with no biological crop loss events. This would require both the absence of biological disruption and rigorous documentation, which makes it a high bar. If it exists, it would genuinely change the argument.