陽台裝了十二個感測器，作者仍然種死了番茄

My Balcony Has Twelve Sensors and I Still Lost the Tomatoes

My father could tell when a plant needed water by looking at it. He never owned a moisture meter. He'd walk past, brush a leaf, and know.

I moved to the city, started growing on a balcony, and immediately proved that his method did not transfer to a person with a day job. Everything I planted died the first year. Not dramatically. Slowly, which is worse, because you keep thinking you're saving it.

So I did what my generation does. I put sensors on everything.

What the sensors taught me

Twelve sensors. Soil moisture in six pots, temperature, humidity, light, pH. Feeding into a dashboard I built myself. I checked it every morning like a man checking a stock portfolio.

The plants still died. Different ones, but they died.

Here's what I learned, and it took two seasons: a sensor tells you a number, not a decision. The soil moisture reading was 34%. Is that good? Depends on the plant. Depends on the soil mix. Depends on whether it rained last night and the sensor is sitting in a drainage pocket. The number was accurate and useless because it was missing context that only observation provides.

What finally worked was combining them. Sensors for the thing I can't do — notice at 2am that humidity spiked and my balcony has no airflow. Eyes for the thing sensors can't do — see that the lower leaves are yellowing, which no moisture reading will ever tell you.

The technology is genuinely good. It is not a replacement for walking outside.

Where urban agtech is actually winning

Let me be fair about this, because there's real progress in the field, and I don't want to sound like the guy who wants everyone to go back to a hoe.

Greenhouse climate control. Inside a controlled environment, sensors and models are spectacular. This is where the industry is delivering. When you can actuate ventilation, heating, and irrigation automatically and you have enough data for the model to learn the specific microclimate, yields go up substantially.

Aerial crop imaging in cities. Municipalities and community gardens are using drone and satellite data to find underused land, measure heat islands, and target where planting will actually help. That's not a home gardening toy, it's genuine planning infrastructure.

Vertical farms for leafy greens. The economics are brutal and mostly they don't work without subsidy, but for a few crops in the right locations the modeling and automation is real.

What doesn't work for a small grower

The consumer products. Almost all of them.

A $200 automated planter with an app that tells you to add water is a $12 pot with an expensive notification. I've bought two. One stopped syncing after eight months. The other one worked and grew basil about as well as a $5 pot on a windowsill.

The rule I've landed on: if the product automates watering, it is selling convenience to people who will forget. If it automates climate, it might be doing something.

A note on the sensor data itself

One thing worth saying, since I keep telling people to collect it: your sensor data belongs to you.

Most consumer agtech products ship the data to the cloud, aggregate it with everyone else's, and use the aggregate to build a model they own. You provided the observations for free and you get a dashboard. If you're going to run sensors for years, check whether you can export the raw readings, and export them monthly whether you need them or not.

I lost two seasons of data when a company shut down their app. My plants were fine. My history was gone.

The thing I actually recommend

Your phone camera.

Take a photo of the same plant every morning. Same angle, same time. Do it for a season and you will notice things no sensor reports. The leaf that started curling four days before the sensor noticed. The morning the growth spurt happened. The slow decline you were too close to see.

The comparison over time is the whole trick. A sensor gives you a point. A photo series gives you a trend, and gardening is all trends.

My father did this with his memory. He remembered what the plant looked like last Tuesday. I can't do that, so I use a camera roll. Same information, different storage.

Feeding cities

The bigger version of this question is whether any of this helps feed cities. Honestly: not yet, not from balconies. Rooftop and balcony growing is a hobby with good side effects, not a food system.

What does help is using the data to decide where the growing should happen. Which empty lots have the right light. Which neighborhoods have a heat problem that planting would fix. Which schools have grounds that could become gardens.

That's the useful layer. The AI shouldn't be telling you when to water a tomato. It should be telling a city where its tomatoes should be.

Grow or die.