I'm Using AI-Controlled Sensors in My Chili Greenhouse. Here's What I've Learned
Last autumn I set up a small automated growing system in my 3x4 metre greenhouse. Nothing fancy by commercial standards — a handful of sensors, a Raspberry Pi, some solenoid valves, and an automated fan — but enough to monitor and control temperature, humidity, soil moisture, and irrigation for about 30 chili plants over winter.
Twelve months later, I have data, opinions, and a few hard-won lessons. Here’s the honest rundown.
The Setup
The hardware:
- 4x DHT22 sensors for temperature and humidity at different heights in the greenhouse
- 6x capacitive soil moisture sensors in different pots and beds
- 1x Raspberry Pi 4 running Home Assistant as the brain
- 2x solenoid valves on drip irrigation lines, controlled via relay switches
- 1x 12V exhaust fan triggered by temperature thresholds
- 1x SwitchBot smart plug controlling a small oil heater for frost protection
Total cost: roughly $350 for the electronics plus about $150 for plumbing and electrical bits.
The software side uses Home Assistant automations: if soil moisture drops below a threshold, open the irrigation valve for X minutes. If temperature exceeds 35°C, run the exhaust fan. If temperature drops below 8°C overnight, turn on the heater.
What Worked
Overnight frost protection
This was the biggest win. Sydney had about a dozen nights below 8°C last winter, and two nights where the overnight low hit 3°C in my area (western suburbs get colder than the coast). The automated heater kicked in exactly when needed, and I didn’t lose a single overwintering plant to cold damage. In previous years, I’d lost 4-5 plants per winter to unexpected cold snaps while I was asleep.
The system used about $45 of electricity for heating over the entire winter. Cheap insurance for plants worth considerably more than that in terms of next-season production.
Soil moisture consistency
Hand-watering is inconsistent. Sometimes I’d overwater, sometimes I’d forget for a few days, and the plants would oscillate between too wet and too dry. The automated drip system maintained soil moisture within a much tighter range. I set the target at 40-55% for most varieties and the system held it there reliably.
The result: fewer blossom end rot issues (which is often linked to inconsistent watering), more even growth, and better fruit set. My overwintered Scotch bonnets produced their first ripe fruit three weeks earlier than usual this season, which I attribute partly to the more consistent watering through winter.
Data and learning
Having twelve months of continuous temperature, humidity, and moisture data has been genuinely educational. I now know that my greenhouse reaches 42°C by 11am on a clear summer day (too hot for even superhots), that humidity drops dramatically when the fan runs, and that my soil moisture drops fastest in the corner bed near the door.
This data has informed physical changes: I added shade cloth over the summer, repositioned two plants to the more consistently moist centre bed, and adjusted my winter watering schedule down (the data showed I was overwatering in June-July when evaporation is minimal).
What Didn’t Work
Over-reliance on soil moisture sensors
Capacitive soil moisture sensors are cheap, but they’re also inconsistent. Readings drift over time as minerals build up on the sensor surface, and different potting mixes give different readings at the same moisture level. I had to recalibrate twice during the year, and one sensor failed completely after eight months.
The lesson: use the sensors as guides, not absolute truth. I now check the actual soil moisture manually once a week and adjust the sensor thresholds if the readings have drifted.
Fan-only cooling in summer
A single exhaust fan is not enough to keep a small greenhouse cool in a Sydney summer. On days above 35°C, the fan could bring the internal temperature down from 45°C to maybe 38°C — still too hot for optimal growing. Several plants showed heat stress (leaf curl, flower drop) during the January heatwave despite the fan running constantly.
Next summer I’m adding evaporative cooling — a wet pad on the intake side with the fan on the exhaust side. Commercial greenhouses have been doing this forever. I should have started there.
Complexity creep
What started as “a few sensors and automations” gradually became a system that I spent more time tweaking and troubleshooting than I would have spent just walking into the greenhouse and checking on things manually. A WiFi dropout at 2am would mean the irrigation didn’t run. A sensor glitch would trigger the fan unnecessarily. The Raspberry Pi needed a restart every few weeks.
There’s an irony in automating a hobby. Part of the joy of growing chilies is the hands-on, daily ritual of checking your plants. When you automate that, you don’t save time — you just replace gardening time with IT troubleshooting time.
Would I Recommend It?
For a hobby grower? Honestly, only if you enjoy the tech side as much as the growing side. If you’re building it because you think it’ll save time, it won’t. If you’re building it because you find sensor data and automation interesting, and you want to learn more about your growing environment, then yes — it’s been worthwhile and educational.
The AI and automation side is getting more accessible, though. The Team400 team recently published some thoughts on how AI is being applied in controlled environment agriculture at commercial scale, and the principles are the same even at my backyard greenhouse level — just with more zeros on the budget.
For the overnight frost protection alone, I’d say the system paid for itself. Everything else was a bonus that I could have achieved with a thermometer, a watering can, and a bit more discipline.
If You Want to Build Your Own
Start small. A single temperature/humidity sensor and one soil moisture sensor connected to a Raspberry Pi or ESP32 gives you monitoring without the complexity of automation. Run it for a few months, look at the data, and then decide if automated irrigation or heating is worth adding.
The Home Assistant community has excellent guides for greenhouse automation, and most of the hardware is available from Australian retailers like Core Electronics or Jaycar.
And keep watering your plants by hand too. The sensors are good, but your eyes and fingers are still better at spotting a thirsty plant than any capacitive probe in a pot.