As an Amazon Associate, Akermin earns from qualifying purchases. Product links on this page are affiliate links.
A thermopile pyranometer absorbs energy across essentially the whole terrestrial solar spectrum, roughly 285 to 2800 nm, and converts the heating into a voltage, so it measures total irradiance directly. A silicon photodiode only responds to photons energetic enough to cross the silicon bandgap, which cuts its sensitivity off above about 1100 nm and leaves roughly a quarter of the solar energy invisible to it. The silicon device compensates with a calibration factor that is correct under one reference spectrum and drifts away from correct whenever the real spectrum differs, which is what the thermopile adds: a reading that does not depend on the sky behaving as the calibration assumed.
A thermopile sees the whole spectrum, a photodiode sees a slice
The two sensing principles are unrelated. A thermopile is a thermal device: a black absorbing surface under a glass dome takes in radiation of any wavelength it can absorb, warms slightly, and a stack of thermocouple junctions reports the temperature difference between that surface and the instrument body as a small voltage. Nothing in that chain cares what colour the light was. Its spectral flatness comes from the absorber being black and the dome being transparent across a wide band.
A photodiode is a quantum device. A photon either carries enough energy to lift an electron across the bandgap or it does not, and in silicon the cutoff falls near 1100 nm. Longer wavelength infrared passes through without generating any signal at all. Within the band it does respond to, the response is not flat either: sensitivity climbs towards the near infrared and falls away in the blue, so the instrument weights the spectrum rather than integrating it evenly.
That weighting is corrected by a single calibration constant chosen so the output agrees with a reference under a defined spectrum. The constant is a number, not a function, so it cannot track a spectrum that shifts. When cloud scatters out the infrared, or a low sun angle lengthens the atmospheric path, or humidity absorbs specific bands, the true proportion of energy inside the silicon window changes and the fixed constant is no longer the right one.
Classification, levelling and the slow response you pay for
What the thermopile really adds is not just spectral coverage but a published error budget. Instruments of this class are graded against a standard that specifies separate limits for each way the reading can be wrong: the zero offset caused by the dome radiating to a cold sky, the zero offset caused by changing ambient temperature, non stability over a year of use, non linearity across the range, directional or cosine response error, spectral selectivity, temperature dependence and tilt response. A class designation is a statement that every one of those terms was bounded, which is a different kind of claim from a single accuracy percentage.
The price of that is paid in handling. A thermopile has thermal mass, so it settles over seconds rather than milliseconds and smooths out exactly the fast fluctuations a broken sky produces. It has to be levelled, because the quantity is defined on a horizontal plane. The dome has to be clean, since dust and dew both intercept radiation before the absorber sees it, and dew on a cold morning is a routine problem rather than an unusual one. Serious installations add ventilation and heating to deal with that, and a desiccant to keep the inside of the dome dry.
When the narrow silicon response is the right answer
If the question is how much solar energy is arriving at this site, the thermopile is the correct instrument and the silicon device is an approximation. If the question is whether this module is performing as it should, the answer flips. A silicon reference device responds to the same slice of the spectrum a silicon module responds to, so when the spectrum shifts, the sensor and the module shift together and much of the error cancels in the ratio between them. Using a spectrally flat instrument to judge a spectrally selective converter introduces a mismatch that then has to be corrected away.
Speed matters for the same reason. A module reaches its new operating point within milliseconds of a cloud edge passing, and an electrical measurement captures that instantly. A fast photodiode follows it; a thermopile reports an average over the preceding seconds. Pairing a fast electrical reading with a slow irradiance reading manufactures an error that looks like a performance problem. Ruggedness, tolerance of being held at an angle against a module and indifference to levelling all point the same way for field work.
| Property | Thermopile pyranometer | Silicon photodiode sensor |
|---|---|---|
| Spectral coverage | Broad, roughly 285 to 2800 nm | Cut off near 1100 nm, weighted within the band |
| Response time | Seconds, set by thermal mass | Milliseconds |
| Error specification | Separate limits for offsets, cosine, spectral and tilt terms | Usually a single accuracy figure against a reference |
| Installation demands | Levelling, clean dome, desiccant, often ventilation | Point it at the surface of interest and keep it clean |
| Best suited to | Reporting the solar resource at a site | Judging a silicon module against the light it received |
Every detector only answers for the physics it uses
The spectral blindness of a photodiode is one instance of a rule that runs through every sensing instrument: a detector reports on the interaction it was built around and is silent about everything else. The clearest illustration is an instrument that carries two unrelated sensors in one package. The Aranet4 carbon dioxide and radiation sensor bundle states that one unit tracks carbon dioxide, temperature, humidity and atmospheric pressure while a separate unit measures ionising radiation across gamma, beta and x-rays with Geiger counter functionality, both with electronic ink displays and long battery life. Neither half of the bundle publishes an accuracy figure for its reading, so for that the entry is: not stated in the listing bullets. Two devices, because no single sensing element responds to both an absorbing gas and an ionising particle. Pairing them in a bundle is an admission of exactly that.
The same logic sorts gas sensing technologies. A non dispersive infrared element measures absorption in one narrow band chosen for the target molecule and is essentially blind to gases that do not absorb there. A metal oxide element responds to a broad family of reducing gases and cannot tell you which one it found. Neither is better in the abstract, which is the argument laid out in gas detector sensor types and in more detail for one of them in semiconductor gas sensors.
A published error band is the other half of the comparison, and it is what separates a sensor from an indicator. The GoveeLife 2.0 hygrometer and thermometer set states a Swiss made sensing element with accuracy of plus or minus 0.3 degrees Celsius and plus or minus 3 percent relative humidity, a refresh every two seconds, trend data covering the past twenty days and export covering the past two years. Whether those figures are the right ones for a given job is a separate question from whether they exist, and an instrument that publishes none cannot be assessed at all.
Placement is the third term. The Autopilot carbon dioxide monitor and controller states a remote sensor probe on a 15 foot data cable with a photocell, specifically so the sensing point can be chosen independently of where the controller hangs, along with a customisable set point and deadband and a trend chart recording per minute, hour, day and week. That cable exists because the reading belongs to the location of the sensor rather than to the instrument, which is the point made at length in why sensor placement can change your measurements.
Matching the sensor class to the question you asked
Write down the question before choosing between the two. If the output is a figure for the energy resource at a location, or anything that will be compared against a published climatological record, the broad spectrum instrument with a class designation is the only defensible choice and its upkeep is part of the commitment. If the output is a performance ratio for silicon modules, a matched reference device is more accurate for that purpose and far easier to carry.
Whichever class you choose, the calibration has a life. Thermopile instruments drift as domes degrade and desiccant exhausts, photodiodes drift as filters and encapsulants yellow under ultraviolet, and neither announces the fact. That slow movement away from truth is the subject of what sensor drift is, and the broader pattern of sensing elements as consumables rather than permanent parts is set out in why sensors do not last forever.
Frequently Asked Questions
Does a silicon sensor read high or low compared with a thermopile?
It depends on the spectrum at the moment of the reading, which is the whole difficulty. A silicon device is calibrated to agree under a reference spectrum, so it reads correctly there and departs when conditions shift. Overcast skies, low sun angles and high water vapour all redistribute energy relative to that reference, pushing the silicon reading either side of the thermopile value.
Why does a pyranometer need levelling?
Because it is defined to measure the irradiance falling on a horizontal surface over a full hemisphere, and tilting it changes the geometry of what it collects. A small levelling error produces a systematic offset that varies through the day as the sun moves, which is harder to spot than a constant one. Most instruments carry a spirit level and adjustable feet for this reason.
What is a zero offset on a thermopile?
It is a reading the instrument produces when there is no solar radiation at all. The dome radiates heat to a cold night sky and cools below the detector, producing a small negative signal, and a changing ambient temperature produces another. Both are specified in the classification standard because they set the floor below which the instrument cannot distinguish signal from artefact.
Can a silicon reference cell replace a pyranometer for photovoltaic work?
For judging module performance it is often the better device, because a reference cell built from the same technology responds to the same part of the spectrum the module does, so spectral mismatch largely cancels. For reporting the solar resource at a site, it cannot, because the quantity being reported is total irradiance and the silicon device does not see all of it.
Why does response time matter if irradiance changes slowly?
Because it does not change slowly under broken cloud. A thermopile has thermal mass and takes seconds to settle, so during a cloud edge it reports a smoothed version of what the module actually experienced. Pairing a slow sensor reading with a fast electrical reading creates a mismatch that no later arithmetic can undo.
Do these sensors drift over time?
Both do, by different mechanisms. Thermopile instruments change slowly through dome degradation, desiccant exhaustion and detector ageing, and the classification standard puts a limit on that non stability. Photodiodes shift as their filters and encapsulants yellow under ultraviolet exposure. Either way the calibration has a shelf life and the instrument does not announce when it has passed it.