Microgreen sensor monitoring in Bangkok: temperature, humidity, lux, moisture, and what the data actually shows

Key takeaways

A healthy microgreen tray in Bangkok follows a 5-variable fingerprint by day: tray surface temperature within 1.5 degrees Celsius of ambient, canopy humidity dropping from 88 to 95 percent during blackout to 65 to 75 percent once uncovered, lux at canopy level between 8,000 and 15,000 lux from day four onward, substrate moisture holding at 55 to 65 percent volumetric content through mid-cycle, and stem temperature reading 0.5 to 1.5 degrees Celsius below ambient from day four onward. Plant Mood grow log data, 2024.

In 7 of 8 documented mold incidents, tray-level relative humidity exceeded 91% for more than 4 consecutive hours before any visible mold appeared. Visible mold arrived an average of 22 hours after the sensor threshold breach. The intervention window exists. Most growers do not know they have it because they are not measuring the right variable at the right height. Plant Mood grow log data, 2023 to 2024.

Accumulated lux (daily light integral) and temperature integral together predicted microgreen harvest-ready status within plus or minus 0.5 days in 14 of 18 logged Bangkok cycles. Visual checks alone matched the same window in 11 of 18 -- and visual accuracy declined with familiarity while sensor prediction accuracy held flat. Plant Mood grow log data, 2024.

Across a full month of 5-minute interval logging, climate management draw -- fans and dehumidification -- varied by up to 58% between the lowest and highest draw days. Lighting draw varied by less than 3%. Humidity above 87% for more than 6 consecutive hours was the condition shared by the three highest-consumption days. The variable cost in a Bangkok grow room is the climate, not the light. Plant Mood grow log data, 2024.

In April, grow room temperature stayed above 31 degrees Celsius until after 3am on 19 of 30 logged nights. In October, tray-level humidity never dropped below 84% before 6am on any logged night. The grow schedule does not change between seasons. The overnight window does. Sunflower germination averaged 78% in April and 91% in October under matched conditions, with the April deficit tracking directly against overnight heat accumulation. Plant Mood grow log data, 2024.

Tray-level humidity diverged from room-level humidity by an average of 8.4 percentage points across logged mold incidents. A wall sensor reading 80% RH corresponded to 88.4% at the canopy surface. Growers managing mold risk from a room-mounted sensor are reading air that is not in contact with the tray.

TL;DR

The standard advice on microgreen monitoring gives you five separate numbers and tells you to keep each one in range. What the sensor logs show is that you cannot read them separately. Temperature drives transpiration rate, which drives canopy humidity, which determines whether stem temperature can cool below ambient, which tells you whether the moisture-to-lux balance is working. Pull any one variable out of range and at least two others respond within hours. In Bangkok, where ambient temperature runs 6 to 8 degrees Celsius above what most grow guides assume and seasonal humidity swings 25 percentage points between dry season and wet, the interaction between variables is not a footnote. It is the system. This article is what the logs show when you read all five together.

The month I stopped reading variables one at a time

November. A second grow rack had just gone in. Monthly power draw came in 41% higher than the previous month against a 22% increase in lighting load. I pulled the full timestamped log and mapped energy draw against environmental sensor data at 5-minute resolution. The days with the highest draw were not the days with extended photoperiods or additional lighting. They were the days where Bangkok humidity climbed above 87% and stayed there through the afternoon, forcing fan runtime up by an average of 34%. That was the energy question. The same week, a sunflower tray that had looked textbook healthy at 9pm on day five showed white mold across the left half at 7am on day six. I went back to the sensor log and read what had been happening at 9pm: relative humidity at canopy level 93%, substrate moisture 74%, fan off after a tripped switch. The data had been showing a mold event in progress. I was looking at the tray instead of the log.

Both of these happened within the same fortnight. The pattern underneath them was the same: I was reading variables one at a time -- energy draw, or canopy appearance, or a spot humidity check -- and missing what the variables were showing together. The energy log made no sense until it was read against the humidity curve. The mold event made no sense until the moisture and humidity and airflow readings were laid side by side. The sensor data was not incomplete. My reading of it was. That was when the monitoring stack became something I read as a system rather than as five separate numbers.

Why the five variables cannot be read independently in Bangkok conditions

Temperature, humidity, lux, substrate moisture, and stem temperature each affect the others, and the interdependence is what makes multi-variable monitoring worth the overhead rather than just adding measurement complexity.

Temperature is the primary driver. It sets transpiration rate -- the speed at which the plant moves water from root to leaf and releases it into the air. Higher temperature means higher transpiration demand, which pulls moisture from the substrate faster and releases it into the canopy air faster. Both effects are amplified in Bangkok conditions, where ambient regularly exceeds 30 degrees Celsius and seasonal peaks reach 34 to 35 degrees Celsius.

Transpiration rate, combined with canopy density, determines canopy humidity. A dense tray in a warm Bangkok room is a humidity generator. The canopy air layer can run 8 to 12 percentage points above the room sensor reading. That gap is where the mold events live. A room reading of 80% RH corresponds to 88 to 92% at the tray surface in a mature canopy. The wall sensor is reading the wrong air.

Canopy humidity above a threshold suppresses further transpiration. When the air above the leaves is already saturated, the vapour pressure gradient that drives water movement from leaf to air collapses. Transpiration slows. The stem stops cooling below ambient. Stem temperature at or above ambient is the summary signal for this state -- it means the transpiration mechanism has stalled, which has upstream implications for nutrient delivery, calcium transport to leaf tips, and root oxygen status.

Lux drives photosynthesis, which consumes CO2 and produces the sugars that fund growth. But at Bangkok temperatures above 32 degrees Celsius, respiration rate climbs faster than photosynthesis. A tray receiving 10,000 lux at 34 degrees Celsius is photosynthesising at the effective equivalent of 7,000 to 8,000 lux in a 22 degrees Celsius environment, because respiration is burning the gains. Pale, leggy growth that appears under adequate lux in a hot Bangkok room is often misread as a lighting problem. It is a temperature-lux interaction problem.

Energy draw ties all four together. In a Bangkok apartment, lighting draw is fixed. The variable draw is climate management -- the fans and dehumidification devices responding to the temperature and humidity conditions the canopy creates. The afternoon humidity peak that drives the energy spike is itself the output of a warm room, a dense canopy releasing transpiration moisture, and outdoor humidity infiltrating through ventilation gaps. The power meter is reading the consequences of the biological system, not just the environmental inputs.

The lux and temperature interaction produced the clearest misdiagnosis pattern in the grow logs. Two cycles, same fixture, same lux meter reading at canopy: 10,400 lux. One cycle in November at ambient 29 degrees Celsius, one in April at 34 degrees Celsius. November tray: strong colour, cotyledons fully expanded by day eight, fresh weight within expected range. April tray: pale, stems elongated 18% above baseline, fresh weight 22% below November. Same light. Different temperature. Different photosynthetic output. The April tray looked like a lighting problem. It was a temperature problem that made the available lux insufficient. Plant Mood grow log data, 2024.

The VPD -- vapour pressure deficit -- connects temperature and humidity into a single number that describes the air's capacity to pull moisture from the plant surface. At 32 degrees Celsius and 91% RH, VPD is approximately 0.37 kPa. Efficient transpiration requires VPD above 0.8 kPa. Below 0.4 kPa, water movement from leaf to air stalls. The plant stops cooling. Calcium delivery to fast-growing leaf tips drops because calcium moves through the transpiration stream. Tip burn in the final days before harvest, which most growers attribute to substrate calcium deficiency, is almost always a transpiration suppression event caused by high canopy humidity and low VPD. Adding calcium to the substrate does not fix a VPD problem.

The numbers the logs produced

22-hour intervention window before visible mold -- and 58% variance in monthly energy draw driven by humidity, not light

Two findings from the same dataset that point at the same system. Across 8 documented mold incidents, visible mold appeared an average of 22 hours after the sensor threshold breach at tray level -- meaning the data was readable before the loss in every case. Across a full month of 5-minute energy logging, climate management draw varied by 58% between the lowest and highest draw days while lighting draw held within 3%. Both findings became visible only when humidity was logged at tray level, not room level, at high enough resolution to see the curves. Plant Mood grow log data, 2023 to 2024.

8.4 percentage points

Average divergence between room-level humidity and tray-level canopy humidity across logged mold incidents. A wall sensor reading 80% RH corresponded to 88.4% at the canopy surface. At 82% ambient, the tray was already above 90%. Mold risk thresholds written for room-level sensors are not calibrated for the microenvironment where the mold actually establishes. Plant Mood grow log data, 2023 to 2024.

14 of 18 cycles predicted within plus or minus 0.5 days

Harvest prediction accuracy using accumulated daily light integral and temperature integral versus 11 of 18 for visual checks alone. Visual accuracy declined from cycle four onward on familiar varieties as familiarity bias anchored the grower to a fixed mental image. Sensor prediction held flat across all cycles. Plant Mood grow log data, 2024.

1.8-day harvest timing variance between dry season and wet season

The average difference in harvest timing for the same sunflower variety under matched substrate, seed density, and lux targets between April and October cycles. The driver was temperature integral -- dry season cycles accumulated 30 to 40% more degree-hours above the 10-degree base per day than wet season cycles. A fixed calendar harvest schedule spans this variance without explaining it. Plant Mood grow log data, 2024.

34% above baseline fan runtime

Average increase in fan operating hours on the three highest-consumption days of the logged month, all of which shared sustained humidity above 87% for more than 6 hours. On lower-humidity days with identical lighting load and grow density, fan runtime stayed within 8% of baseline. The climate management draw, not the lighting draw, is where the variable cost lives in a Bangkok grow room. Plant Mood grow log data, 2024.

What the logs showed when variables were read together

Across cycles where all five variables were logged simultaneously, the earliest failure signal was almost always substrate moisture on day three -- not appearance, not canopy humidity, not temperature. In 7 of 9 mold-affected trays, substrate moisture exceeded 68% volumetric content on day three or four, 36 to 48 hours before visible mold. At that moisture reading, combined with Bangkok ambient humidity above 70%, the mold event was already in progress below the surface. Day three is the last point at which airflow intervention consistently prevented the outcome. After day five, the intervention required was tray removal. Plant Mood grow log data, 2024.

The 5-variable fingerprint predicts failure before the tray shows it

The grower who checks the tray visually on day three and sees a healthy canopy is reading the correct variable at the wrong resolution. The fingerprint requires the substrate moisture number, not the canopy appearance. Visual inspection and substrate moisture check are not substitutes for each other. They answer different questions about different parts of the tray.

In the harvest prediction dataset, DLI variance between cycles was relatively contained -- target lux was held reasonably consistent and photoperiod was fixed. Temperature variance between seasons was substantially larger, with dry season cycles accumulating 30 to 40% more degree-hours per day than wet season cycles at the same growth stage. When the prediction model ran on DLI alone, residual harvest timing error averaged 1.2 days. When temperature integral was added, residual error dropped to 0.4 days. The temperature variable was carrying most of the predictive weight the DLI variable missed. In a four-shelf rack, the top shelf accumulated 18 to 22% more daily lux-hours than the lower shelf, and reached harvest readiness an average of 0.9 days earlier. A single harvest schedule applied to all shelves was wrong for at least two of them every cycle. Plant Mood grow log data, 2024.

The temperature integral carries more predictive weight than lux for harvest timing in Bangkok

Growers who invest in lux measurement but not temperature logging are instrumenting the less predictively important variable for harvest timing in Bangkok. A cheap temperature logger -- one per shelf -- is more useful for harvest prediction than a more accurate lux meter, if only one additional sensor is being added.

In April, grow room temperature stayed above 31 degrees Celsius until after 3am on 19 of 30 logged nights, with the overnight low arriving at 3:47am average. In October, the overnight low arrived at 2:12am average and reached 27.4 degrees Celsius versus 29.8 in April. But October's advantage in temperature was offset by humidity: tray-level humidity never dropped below 84% before 6am on any logged October night, versus a consistent 71 to 74% overnight low in April. Two risks, two seasons, different intervention windows. Sunflower germination averaged 91% in October and 78% in April under matched conditions, with the April deficit tracking directly against overnight heat accumulation hours. Plant Mood grow log data, 2024.

The overnight window is where seasonal differences in the grow environment are most visible

A grow schedule built in one season and carried unchanged into the other is calibrated for the wrong overnight risk profile. April requires priority airflow in the 6pm to 10pm window to dissipate daytime heat before the overnight begins. October requires priority airflow in the 2am to 5am window when outdoor humidity peaks and indoor conditions follow. Fixed scheduling addresses neither.

Across 30 days of 5-minute interval logging, lighting draw varied by less than 3% day to day. Climate management draw varied by up to 58%. The three highest-draw days shared one condition: relative humidity above 87% for more than 6 consecutive hours. The cheapest 7-day window of the month was not the coolest week -- it was the week with the most stable humidity, which kept fan duty cycles shorter. Reactive dehumidification -- switching on after crossing the 85% threshold -- produced higher runtime per percentage-point humidity reduction than preventive low-duty cycling starting at noon. The shift from reactive to preventive across two weeks of the logged month reduced dehumidifier runtime by 14.6% with equivalent humidity control outcomes. Plant Mood grow log data, 2024.

Energy draw tracks the humidity curve, not the light schedule

The energy optimisation question for a Bangkok grow room is not how efficient the lights are. Lighting is the fixed cost. The variable is climate management. Optimising the light schedule while running reactive humidity control is addressing the wrong side of the equation.

Plant Mood grow log data, 2023 to 2024

The measurements that are not being taken

A wall-mounted hygrometer or a sensor sitting on a shelf above the trays reads the room air. A dense microgreen canopy in the final days before harvest creates its own microclimate -- warmer and more humid than the room -- that the wall sensor cannot capture. Across 8 logged mold incidents, the average divergence between room-level and canopy-level humidity was 8.4 percentage points. A room sensor reading 80% RH may correspond to 88 to 92% at the tray surface. This is not a sensor accuracy problem -- it is a sensor placement problem. Taking a handheld reading at canopy level once per day during the maturation phase costs nothing and produces information the wall sensor cannot. The grower managing mold risk from a wall reading is making decisions based on the wrong air.

Sensor placement: room level versus tray level

An infrared thermometer costs under 500 baht and takes five seconds to use. A stem reading 0.5 to 1.5 degrees Celsius below ambient indicates active transpiration -- the plant is moving water from root to leaf and evaporative cooling is occurring at the leaf surface. Stem temperature at or above ambient means transpiration has suppressed, which could be caused by excess canopy humidity, inadequate substrate moisture, stomatal closure from heat, or root stress. All four failure modes produce the same stem reading. The infrared thermometer does not identify the cause. It tells you in five seconds that something in the balance has shifted, before any visual sign appears. Almost no small-scale grower in Bangkok uses this check. The hardware exists and costs less than a bag of substrate.

Stem temperature as a five-second summary check

Day three is when most growers are watching for germination emergence. Almost nobody measures substrate moisture at day three because the tray has not been watered since seeding and the assumption is that it is fine. In Bangkok conditions, where ambient humidity slows substrate drying during the blackout phase, a correctly seeded tray can still be at 68 to 72% moisture on day three without additional watering. The tray looks healthy. The moisture reading is already at the mold risk threshold. Day three is the last point at which airflow intervention consistently prevented a mold event in our logged data. After day five, the only intervention was removal. The check that costs 30 seconds and changes the outcome is the one that is being skipped.

The day-three substrate moisture check

Most consumer sensor apps display daily or hourly averages. The afternoon humidity spike in Bangkok lasts 2 to 3 hours. The lag between a 3-degree Celsius temperature rise at 1pm and the corresponding power draw step-up was 12 minutes in the logged data -- invisible at hourly resolution, clear at 5-minute intervals. The 22-hour mold warning window compresses into a single daily data point when read at 24-hour resolution. The direction of the humidity line at 2am in October -- still rising -- is the information that changes the overnight airflow decision, and it disappears in a daily average. Coarser logging produces a record. Fine-grained logging produces a decision-making tool.

Logging at the wrong resolution

The lux number most growers remember is from the fixture specification or a measurement taken at the fixture surface. Lux falls off with distance and angle. A fixture rated at 20,000 lux at 15 centimetres may deliver 9,000 lux at a canopy sitting 30 centimetres away. In a four-shelf rack, inverse square law falloff means the top and bottom shelves may differ by 18 to 22% in canopy lux even when the same fixture is overhead. The only lux number that matters for tray health assessment is the one measured with a meter at the canopy surface, not the one on the packaging.

Lux at canopy level versus fixture specification

The things the data shows that the standard advice inverts

Humidity management without airflow produces mold even at moderate humidity levels. Across 8 documented incidents, fan state was the only variable present in every single case -- every incident occurred with airflow off or obstructed. Reducing watering volume without fixing airflow only slows the mold process. The distinguishing variable is not humidity alone: a tray watered at 120% of normal volume with airflow running did not mold. A tray watered at 80% of normal volume with no airflow did. Moisture that evaporates does not become mold. Moisture that sits at the surface in stagnant air does.

Lux management without temperature management produces pale, leggy trays that look like insufficient light. Photosynthetic efficiency per unit of lux drops 15 to 28% as temperature rises above 28 degrees Celsius, depending on variety. A tray receiving 12,000 lux at 34 degrees Celsius is not achieving the photosynthetic output that 12,000 lux produces at 22 degrees Celsius. Raising the lux further without addressing temperature accelerates the respiration-photosynthesis imbalance rather than resolving it.

The experienced grower's visual harvest check becomes less accurate with familiarity, not more. Visual accuracy in our logged data held above average in the first three cycles on any new variety, then declined from cycle four onward as familiarity anchored the grower to a fixed mental image. Confidence increased while fresh weight variance also increased. The sensor prediction did not drift -- it responded only to the current cycle's accumulated inputs, with no memory of previous cycles to bias it.

Energy optimisation focused on lighting efficiency addresses the fixed part of the draw while the variable part -- climate management -- goes unexamined. In the logged month, lighting draw held within 3% across all 30 days. Climate management draw varied by 58%. The October tray that germinated better than the April tray failed in the day-four to day-six window at a higher rate. Better initial conditions produced a more demanding management window, not an easier one.

The coolest week of the logged month was not the cheapest week. It was the second most expensive. It had the lowest average ambient temperature -- 29.8 degrees Celsius -- but the highest humidity variance: a daily swing averaging 24 percentage points between 7am and 3pm. The fans and dehumidifier were responding to threshold crossings, not to averages. A stable high-humidity environment is less expensive to manage than a volatile one that swings across thresholds repeatedly, because threshold-crossing events trigger full-duty responses while stable conditions allow lower duty cycles. Temperature as a headline grow room metric is pointing in the wrong direction for predicting operating cost in Bangkok.

Five pieces of standard advice appear across most microgreen monitoring guides: manage humidity to prevent mold, keep lux high for better growth, use a room sensor to track grow conditions, expect experienced visual harvest checks to outperform guesswork, and add lighting when energy draw is high. Each of these is directionally correct in a temperate climate. In Bangkok, the data inverts or significantly complicates each one.

Building a multi-variable monitoring setup for a Bangkok grow room

Instrument each shelf position separately, not the room

Per-shelf temperature and humidity readings at canopy level, alongside a lux measurement taken at the tray surface. After two cycles, the variance between shelf positions becomes quantified rather than estimated. In most Bangkok grow racks, top-to-bottom DLI and temperature integral variance exceeds 20% -- the threshold at which a single harvest schedule applied to all shelves reliably produces inconsistent outcomes.

Room-level readings average across the gradient that is the actual source of harvest timing variance and mold risk divergence. In a four-shelf Bangkok grow rack, heat stratification and inverse square law light falloff commonly produce 2 to 3 degrees Celsius of vertical temperature gradient and 18 to 22% variation in canopy lux between shelf positions. A single room sensor is accurate for nothing specific and misleading for tray-level decisions.

Add a substrate moisture sensor at tray level for at least one tray per seeding batch

A substrate moisture reading on day three after blackout removal. If it reads above 65%, skip the day-three watering and run airflow for 30 minutes first. If it reads above 68% in ambient humidity above 70%, this is a caution signal for mold onset within 48 hours regardless of canopy appearance.

Substrate moisture is the earliest detectable deviation in a failing tray and the one variable most resistant to accurate visual or tactile assessment. A tray that feels heavy and looks healthy can be at 72% substrate moisture and already in the mold onset window. In Bangkok wet season ambient above 70% RH, the threshold is 68% substrate moisture on day three or four. The sensor does not need to cover every tray -- one reference tray per seeding batch running on the same substrate and watering schedule provides the data point that changes the intervention decision.

Log energy draw per device, not per circuit

Within two weeks: a visible correlation between the afternoon humidity curve and the climate management draw spike. Once that pattern is identified, the optimisation target becomes clear. Preventive dehumidifier scheduling before the 1pm to 4pm Bangkok humidity peak reduces total runtime by approximately 14 to 15% compared to reactive threshold-based switching, with equivalent humidity control outcomes.

A single whole-circuit monitor reports total draw but cannot identify which device is driving the variable component. In a Bangkok grow room, the insight is in the breakdown. Lighting draw is effectively flat. Climate management draw tracks the humidity curve. A smart plug with energy monitoring on the fans and dehumidifier separately, logging at 5-minute intervals, produces the data that connects operating cost to environmental conditions rather than to device assumptions.

Pull the overnight window as a separate data view in both April and October

Two curves -- temperature in April and humidity in October -- showing the direction of travel at midnight. In April, temperature is still falling at midnight, meaning the overnight recovery window has not yet opened. In October, humidity is still rising at midnight, meaning any surface moisture from the evening watering will accumulate through the night rather than dissipate. These two shapes are the information that adjusts the management schedule between seasons.

The grow schedule runs the same in both seasons. The overnight window does not. In April, temperature stays above 31 degrees Celsius until after 3am. In October, tray-level humidity never drops below 84% before dawn. These two different risk profiles require different airflow timing priorities, different watering cutoff times in the evening, and different assessments of what a marginal moisture decision in the afternoon will look like at 6am. A full-day average obscures the curve shapes that make the seasonal difference visible.

Build a variety-specific harvest prediction from three calibration cycles

After three cycles: a predicted harvest window within plus or minus 0.5 days for that variety in Bangkok conditions, with the temperature integral component accounting automatically for seasonal timing variance. The calibration carries across dry and wet seasons without adjustment because the model responds to accumulated inputs rather than calendar days.

Published DLI tables for microgreen harvest timing were built at 20 to 24 degrees Celsius in controlled environments. They do not transfer accurately to a Bangkok apartment in March. Three cycles of logged data -- canopy lux at 30-minute intervals, ambient temperature at 30-minute intervals, harvest timing, and fresh weight -- produce a variety-specific accumulated DLI and degree-hour threshold that is calibrated for the actual conditions rather than adapted from somewhere else. Temperature integral should be capped at 24 degree-hours per hour for any ambient reading above 34 degrees Celsius, where heat stress begins to slow rather than accelerate development.

What the monitoring stack changes about how the grow room runs

Geert Warmenbol

Read the sensor log before looking at the trays, not after. The log tells you what the environment has been doing since the last check. The tray tells you what the environment has done to it by the time you look. In a 22-hour mold onset window, the log is current and the tray is already 12 hours behind. Build the habit of checking the overnight humidity minimum and the day-three moisture reading before the morning visual. The visual confirms what the data already told you, or it flags a divergence worth investigating.

Earlier identification of developing problems and fewer losses that feel unexplained. In our operation, treating the sensor log as the primary morning check rather than the secondary confirmation reduced tray losses in the month following the protocol shift by more than the hardware investment cost.

Separate the humidity monitoring into three distinct readings: room-level ambient from the wall sensor, canopy-level spot check with a handheld sensor taken daily during the maturation phase, and the overnight minimum from the continuous log. Each of these answers a different question. The wall sensor tells you the room baseline. The canopy spot check tells you the mold risk environment the tray is actually in. The overnight minimum tells you whether the previous night provided a dry-out window or compounded the day's moisture accumulation. Managing all three as a single humidity number produces systematic errors.

Elimination of the mold events that occur when the room sensor reads acceptable and the canopy is above the risk threshold. In our logged incident review, every mold event that occurred when the room sensor read below 83% was preceded by a canopy reading above 88% that was never taken. The sensor was correct. The measurement location was wrong.

Set a seasonal airflow schedule adjustment at two fixed points per year: end of March as dry season peaks, and end of September as wet season establishes. In April, the priority window for airflow is 6pm to 10pm -- heat dissipation before the overnight begins. In October, the priority window is 2am to 5am -- humidity suppression during the outdoor peak and the indoor lag that follows it. A fixed schedule calibrated for one season will be structurally wrong for the other and will show up in germination rates and mold frequency before it shows up in any visible symptom.

Season-appropriate overnight management without reactive adjustments after tray failures. The October mold rate in our operation dropped from 4 of 9 cycles to 1 of 9 in the season following the introduction of the late-night airflow priority. No other variable changed.

Use the sensor prediction to set the harvest window, then use the visual check to confirm within it -- not independently. When accumulated DLI and degree-hours reach 90% of the variety-specific threshold, begin daily visual checks. Harvest within 12 hours of the threshold being crossed unless the visual reveals a clear reason to wait. The sensor prediction narrows the window from a vague range to a 24-hour target. The visual confirms within it. Neither is as accurate as both used in sequence. In dry season cycles running warmer than the grower's visual reference temperature, the tray reaches visual maturity before physiological maturity -- it looks done before it is done. The temperature integral catches this. The eye does not.

Fresh weight consistency improvement of 8 to 12% across cycles compared to visual-only timing, with the largest gains in dry season cycles where temperature variance is highest and visual maturity most consistently runs ahead of physiological readiness. Plant Mood grow log data, 2024.

Before scaling the grow operation -- adding racks, expanding canopy density, increasing weekly tray count -- log a baseline month of energy draw and humidity data correlated at 5-minute intervals. Then log the equivalent month after scaling. The difference in climate management draw between the two periods is the actual energy cost of the expansion, which is consistently larger than the lighting draw increase predicts. More canopy means more transpiration, which means sustained higher humidity, which means more fan and dehumidification runtime. The growers who are surprised by the electricity bill after scaling are calculating lighting draw and ignoring the canopy humidity load.

Accurate pre-expansion energy modelling rather than post-expansion bill shock. The baseline-to-post-scaling comparison also identifies whether the existing climate management infrastructure can handle the additional load or whether it will spend the afternoon peak above 87% humidity -- the threshold where mold risk, crop stress, and energy draw all step up simultaneously.

Building the seasonal calibration into the schedule rather than reacting to the first tray failure of the new season. Every grower who has run through both April and October in Bangkok has experienced the point where the protocol that worked last month stops working. The standard response is to treat it as a problem to diagnose after it occurs. The monitoring stack makes it a predictable transition to manage before it does. The airflow schedule, watering cutoffs, and harvest timing thresholds all have seasonal adjustment points that can be set in advance once the overnight sensor profile for each season is logged. The reaction approach costs trays. The adjustment approach costs 20 minutes at the start of each season transition.