Thermal Watch.

The need.

What you want to know: what is running hot, or has stopped running cool?

A transpiring leaf runs cooler than the air because evaporation carries heat away; when it stops, the plant is telling you something hours before it wilts. A fevered animal, a smoldering bale, and a failing breaker all announce themselves the same way: as heat, before they are visible any other way. Thermal is the sense that hears that announcement.

The one rule of thermal.

A thermal reading does not measure the plant; it measures an energy balance, and only one term of that balance is about the plant. Leaf temperature moves with the radiation load, the air, the humidity, and the airflow before it says anything about stress. So the rule: thermal is only legible next to air temperature and humidity measured in the same place at the same time, which is exactly what the sensor fleet on the rest of this site already produces. That is why thermal belongs on this shelf rather than as a gadget: your existing readings are what convert the picture into a signal. The full argument, including why an image beats a better thermometer, lives at the Crop Water Stress Index.

The ladder: three rungs, honestly labeled.

• The spot ($10 to 20). One temperature from one fixed aim, continuously. The compost surface, the udder line at the stanchion, a patch of canopy against its air. Mind the cone: at arm's length the wide variant averages a dinner-plate circle (leaf, gap, and floor all melted together), while the narrow variant reads a single leaf. Pick the cone for the target.
• The coarse grid ($40 to 70). A 32×24 array: 768 temperatures, not a picture. It cannot tell leaf from pot from floor, so it only works where the entire view is one thing, framed tight over a single bench of closed canopy. In that one role it is excellent and nearly free, and published research has used exactly this part for continuous stress monitoring indoors.
• The phone clip ($200 to 350). 256×192 real thermal pixels, sensitive to a few hundredths of a degree, riding the phone already making your rounds. Enough resolution to keep only the leaf pixels and discard the rest, which is precisely the ability that makes cheap thermal trustworthy. A bonus the community has documented: these clips present themselves as ordinary USB cameras, so the same module plugs into a Raspberry Pi as a fixed thermal eye, with open-source tools reading the thermal stream. One part, two roles. We have not verified that ourselves yet; when we have, this sentence gets stronger.

The config: the parts.

The reference leaves: the $15 build that makes it all honest.

The stress index needs two baselines: how cool a fully watered canopy would run right now, and how warm a fully stressed one would. Instead of computing them from theory, build them. Two thin metal discs, painted matte green so they absorb light like a leaf. One stays wet from a small wick and reservoir: it evaporates flat out, a leaf with all the water it wants. The other is coated with petroleum jelly so it cannot evaporate: a leaf with its pores shut. A contact temperature probe bonds to the back of each (the job that probe is genuinely right for), reporting both baselines continuously as ordinary oat-ods readings.

Then any thermal reading of the canopy converts on the spot: the index is where the canopy sits between the wet reference and the dry one. Because all three surfaces hang under the same lights, the heat the lights add cancels out of the comparison, which solves a scheduling problem that otherwise spoils everything: it makes the lights-on reading a person takes on a normal walk-through valid. The full design and reasoning live on the CWSI page. We have not built this pair yet; the design follows published research on reference surfaces, and our build notes will land here.

The Station Card here.

Plain honesty first: a thermal sensor cannot see the Station Card. Printed markers, scale bars, and color patches are visible-light objects; to a thermal eye the card is a card-shaped patch of whatever temperature the room is. So on this one approach the card's jobs are carried differently. Identity and geometry ride the paired visible frame: the phone clip shoots thermal and visible together, and a fixed clip-on-Pi station mounts a regular camera beside the thermal one, so the card sits in the visible frame and the pair registers as one record. The reference job passes to the wet and dry leaves, which are to thermal what the card's color patches are to color: the known things in view that make the unknown things readable. For the fixed spot and grid sensors, which see no card and pair with no camera, identity lives in the station's configuration, the same as any temperature probe on this site.

The Station Card The reference leaves (CWSI)

Rules that save you from plausible garbage.

• Never shoot thermal through glazing. Greenhouse poly, glass, and tent windows are opaque to heat radiation. From outside you are measuring the wall, and the numbers will look perfectly plausible. Every thermal eye goes inside the envelope.
• Let it warm up. Uncooled thermal sensors drift for the first minutes after power-on. Give a fixed sensor a warm-up before trusting it, and treat the first reading after a cold start with suspicion.
• Watch what is in the view. Wet leaves, shiny pots, and metal read wrongly to a thermal eye. Frame to canopy, or measure knowing what fraction of the view is not leaf.
• Trust differences, not absolutes. Affordable thermal is accurate to a degree or two in absolute terms and much better than that at telling two things in the same frame apart. Canopy minus reference, animal minus herd: the differences are where the signal is.

The software.