IoT x Plants x BoM

Tl;DR

Intro

Velxio has been a recent surprise to me, when doing a small come back to IoT and electronics.

The big win with ESPHome for these sensor projects: the entire DHT22+MQTT setup that took a main.py, boot.py, DHT22.py and manual error handling is replaced by ~15 lines of YAML.

MQTT x IoT

Current Setup: MQTT DHT22 PGSQL

From this section and these scripts:

#mosquitto_sub -h 192.168.1.2 -t "pico/#" -v #if this is receiving data already
git clone https://github.com/JAlcocerT/RPi
cd ./RPi/Z_MicroControllers/RPiPicoW/picow-dht-webapp

#docker ps | grep emqx
#cd ./RPi/Z_SelfHosting/pgsql
docker run -d --name timescaledb \
    -e POSTGRES_USER=pico \
    -e POSTGRES_PASSWORD=pico \
    -e POSTGRES_DB=sensors \
    -p 5432:5432 \
    timescale/timescaledb:latest-pg16

#docker ps | grep timescaledb
docker exec -it timescaledb psql -U pico -d sensors

See this file for the full commands (we need to make a one time setup at TimescaleDB):

CREATE TABLE readings (
    ts          TIMESTAMPTZ NOT NULL DEFAULT NOW(),
    topic       TEXT        NOT NULL,
    value       DOUBLE PRECISION NOT NULL
);

SELECT create_hypertable('readings', 'ts');
-- .............

As we need one script to push the data, and another (FastPI) to push it via websockets…

we better use tmux: similarly for those lenghty blender runs

#tmux ls
cd ./home/jalcocert/RPi/Z_MicroControllers/RPiPicoW/picow-dht-webapp
tmux new-session -d -s mqtt "uv run mqtt_to_db.py"
tmux new-session -d -s webapp "uv run uvicorn main:app --host 0.0.0.0 --port 8077"
#tmux attach -t mqtt #see the logs flowing CTRL B + D to go out w/stopping it
#docker exec -it timescaledb psql -U pico -d sensors -c "SELECT * FROM readings ORDER 
#BY ts DESC LIMIT 5;" 

By the time you are reading this, there is already one version that combines DHT22 from a PicoW and a DHT11 from a ESP32:

cd ./home/jalcocert/RPi/Z_MicroControllers/dht-webapp

From T and H to VPD

VPD (vapor preassure deficit) is slightly more complex than just a “subtraction” of humidity from temperature.

It represents the difference (the deficit) between how much moisture the air can hold when saturated and how much it is currently holding.

While it is based on temperature and humidity, the math involves the Saturated Vapor Pressure ($SVP$), which is a non-linear relationship—air’s ability to hold water increases exponentially as it gets hotter.

1. The Core Components

To calculate $VPD$, you need to find two values:

1. $SVP$ (Saturated Vapor Pressure): The maximum amount of water vapor the air can hold at a specific temperature.
2. $AVP$ (Actual Vapor Pressure): The amount of water vapor currently in the air.

The formula is: $$VPD = SVP - AVP$$

2. The Detailed Formulas

To calculate this in your microcontroller code, you usually use the Tetens Equation to find the pressure in kilopascals (kPa).

Step A: Find $SVP$

Using the air temperature ($T$ in °Celsius):

$$SVP = 0.61078 \times \exp\left(\frac{17.27 \times T}{T + 237.3}\right)$$

Step B: Find $AVP$

Once you have $SVP$, you use your Relative Humidity ($RH$ as a percentage) to find the actual pressure:

$$AVP = SVP \times \left(\frac{RH}{100}\right)$$

Step C: Calculate $VPD$

$$VPD = SVP \times \left(1 - \frac{RH}{100}\right)$$

3. Why This “Beats” Standard Humidity

Standard Relative Humidity ($RH$) is misleading for a greenhouse. For example:

• At 15°C and 70% RH, the $VPD$ is approximately 0.51 kPa.
• At 30°C and 70% RH, the $VPD$ is approximately 1.27 kPa.

Even though the “Humidity” is 70% in both cases, the plants in the second scenario are losing water more than twice as fast.

If you only controlled for 70% RH, your plants might dry out or stop growing when it gets hot because the “thirst” of the air changed.

4. Implementing in your PID

If you use $VPD$ as your “Process Variable” (your input) for your PID loop:

1. Input: Your code calculates $VPD$ every second using the formulas above.

2. Setpoint: You set a target $VPD$ (usually between 0.8 and 1.2 kPa for most plants).

3. Output: The PID opens the window if the $VPD$ gets too high (air too dry) or closes it/turns on a mister if it gets too low (air too stagnant).

4. The “Leaf Temp” Nuance

In professional systems, engineers don’t use air temperature for $SVP$; they use Leaf Temperature (measured with an infrared sensor).

Because plants transpire, they are usually $1$–$2$°C cooler than the air.

If you want the “Ultimate Greenhouse,” measuring the leaf temp gives you the most accurate $VPD$ possible.

Since you’re looking at outside vs. inside temp anyway, calculating $VPD$ is just a few extra lines of math in your Python or C++ code.

Particularly, now you have the VPD version: with just cosmetic changes at the dashboarding side

#tmux ls
#cd ./home/jalcocert/RPi/Z_MicroControllers/dht-webapp
cd ./home/jalcocert/RPi/Z_MicroControllers/RPiPicoW/picow-dht-webapp-vpd
tmux new-session -d -s mqtt "uv run mqtt_to_db.py"
tmux new-session -d -s webapp "uv run uvicorn main:app --host 0.0.0.0 --port 8077"

MQTT to Sources

We have seen towards timescaleDB

A good mental model is: Prometheus wants metrics, Elastic wants documents, TimescaleDB wants rows, and MQTT is just the transport.

For MQTT to Prometheus, it’s usually not hard if your data is already a numeric measurement and the topic structure is stable. The common pattern is an MQTT-to-Prometheus exporter or a small bridge service that subscribes to MQTT and exposes metrics on an HTTP endpoint for Prometheus to scrape.

For TimescaleDB to Prometheus, it’s less natural. Prometheus is pull-based and expects current metric values, not arbitrary historical SQL rows, so you normally export a current aggregate or a derived gauge rather than “syncing” the whole database. If the data starts in TimescaleDB, a custom exporter or scheduled job is usually the cleanest bridge.

For MQTT to Elasticsearch, it’s also manageable, but the approach is different: you usually use Logstash, Elastic Agent, a custom consumer, or an ingest pipeline to transform messages into documents before indexing them. Elasticsearch is much better at storing event records and searchable history than at serving as a live metrics endpoint.

MQTT x DHT x Prometheus

MQTT x DHT x Elastic

MQTT x Alerts

1. Initially I thought about webhooks

Big Data Tech Talk

A tech talk that goes from embeded systems, to data engineering and to visualizations?

If someone would have let me know that id be doing these sessions so easily:

#cd selfhosted-landing\y2026-tech-talks\4-iot-to-bigdata
#html :)

I mean…so hard, bc Im not using Slidev, not python-pptx nor just html for presentations.

cd .\tech-talk\sensor-to-dashboard
#npx skills add heygen-com/hyperframe
npx hyperframes render --quality high
#npx hyperframes render --fps 30 --quality high --output renders/jalcocertech-dhtwebapp.mp4

Conclusions

Whats next from here?

Maybe…the 3D print thing to make a custom build with the solar panel mounted properly on top of the plants?

And…designed as a code with CadQuery, of course :)

Yea…that can be an interesting upgrade

FAQ

The BoM for the project

I thought initially about doing it with gravity + a NC solenoid valve.

Buuut…some of them need more preassure than gravity will do for me at my use case

So water pump it is

This is the final, definitive “Safety-First” Bill of Materials for your indoor 5L reservoir project.

I’ve updated it to include the specific pump, the protection components, and the connectors needed to make this “plug-and-play” and house-safe.

1. The “Brain” (Logic & Time)

2. The “Muscle” (Water & Power)

3. The “Translator” (Voltage Regulation)

4. The “Safety Trio” (Protection)

5. Hardware & Connectors (The “Glue”)

Total Estimated Cost: ~$62.00

Critical Assembly Reminders:

• The Multimeter Test: Before you connect the ESP32, you must use a multimeter to set the Buck Converter output to exactly 5.0V.

• The Stripe on the Diode: Make sure the silver stripe on the 1N4007 diode is facing the Positive (Red) wire of the pump.

• The Common Ground: Ensure the Battery (-), Buck Converter IN(-), MOSFET Source, and ESP32 GND all share one single, solid connection point.

• Siphoning Check: Place your 5L bottle lower than your plant pot. This way, if the pump stops, gravity pulls the water back into the bottle rather than letting it continue to drip onto the floor.

This list covers everything you need to build a system that is safe for your house, reliable for your plants, and ready for vacations.

How much solar is enough for the ESP32?

I checked recently how much the consumption of both micro-c was as per how much time they send data to TimeScaleDB before killing a small battery.

And…it was not that much: ~12/24h

Now the questions is different: is deep sleep + internal clock + a solar panel enough to power an ESP32 directly during sun hours?

Yep, this is in preparation for the upcoming watering setup :)

Is the sun hitting hard? x MLX90614

As seen at the SVP section, the leaf temperature matters.

We are so lucky that we have IR sensors that can measure exactly that:

• https://jalcocert.github.io/RPi/posts/rpi-iot-MLX90614/

Wireguard (wg-easy) Server

A Pi can also get you started with IoT and sensors