| 28 Jul 2026
An automated snow-removal device for radiometers
Abstract. Snowmelt is strongly influenced by shortwave radiation from the Sun and longwave radiation from the Earth and the atmosphere. These energy fluxes are typically measured in situ using radiometers equipped with pairs of sensors on the bottom and the top of the housing, which measure upwelling and downwelling radiation respectively. However, measuring under snowy conditions is challenging, as the upper lenses tend to become obstructed by snow. Although heating modules can be integrated into radiometers to prevent frost and dew formation on lenses, they are too energy-intensive for off-grid applications and ineffective against substantial snow build-up. This paper presents a novel snow removal device (SRD) designed to automatically brush snow from radiometers lenses with a per-deployment energy cost equivalent to 17 s of heating. A field campaign over an entire winter season (1 Nov 2024–26 Mar 2025) compared a heated and SRD equipped radiometer to a heated-only radiometer, both installed side by side in a snowy forest gap at Montmorency Forest in Quebec, Canada (47°19′ N, 71°10′ W). Two heating strategies were evaluated during this period: fractional heating (cyclic on/off) and continuous heating. Fractional heating proved to induce noise on the data, while continuously heating systematically left snow around the lenses, resulting in increased longwave measurement (+37.20 MJ m–2) and a decreased shortwave measurement (−1.79 MJ m–2) for the winter period. Heating modules alone are insufficient to ensure high-quality data from radiometers in windless, snowy environments; hence solutions such as the SRD should be prioritized.
“An automated snow-removal device for radiometers” by Labelle et al. describes the results of a field campaign during the 2024-2025 winter in Quebec carried out to evaluate the performance of an automatic snow-removal system designed for the CNR4 net radiometer. Two CNR4s were operated, one that was cleaned by the device and another without the device was used as a control. The results demonstrated the system was effective, requiring the equivalent of just 7 hours of CNR4 heating to operate for the entire winter, and the data benchmarked against the control reveal behavior and biases largely consistent with expectations. Snow obstructions in broadband radiometry remains an outstanding problem and this study contributes to addressing it. I have a number of comments, but they are generally minor. Two of the more significant points are as follows: I recommend some reorganizing of Sections 2.3/3.3/4.6 to better situate the study in the relevant literature and build upon it by measuring the transient equilibrium period for the CNR4, then forming recommendations about what a suitable duty cycle for this system might be. I also think some detailed photos and schematics of the SRD are absolutely essential to add to Section 2.5. I think the paper will make a nice contribution to AMT after the authors address my comments.
Comments:
L34: Maybe not the most relevant opening, as this device will not be useful for freezing drizzle/rain and was not tested for water on domes.
Eq. 2: It would make more sense to show the whole calibration equation, of which this is just one of the terms. That said, showing this at all is sort of distracting by shifting focus entirely to pyrgeometers by highlighting a point unimportant for removing snow. Maybe remove it.
L93-96: This is admittedly anecdotal, but in my experience, it is the sun after precipitation events that is responsible for clearing sensors of snow, not wind, which could plausibly even make harder/denser snow on the windward side.
L98-104: I agree that detection is a challenge, but it is also worth stating that mitigation is preferred regardless. Even if you can detect snow on sensors, removal of data during and after precipitation introduces a climatological bias in the record. The obstruction needs to be prevented in the first place, which is significant motivation for your work.
L112-125: Interesting. I was not aware of the Morin et al. paper. It is not in your reference list, so you should make sure to add it. There have been a couple other relevant proposals over the years. These are obscure gray literature, so you can evaluate for yourself whether or not to acknowledge them, but I wanted to make you aware. Persson and Semmer (2010) proposed a methanol wash system and Halliwell, McArthur and colleagues proposed a mechanical housing system that would cover a radiometer in inclement weather (Halliwell et al., 2002). Both designs were built and tested but not widely deployed.
Persson, P.O.G. and Semmer, S. (2010) Impacts of Riming on Arctic Surface Energy Budget Measurements. Autonomous Polar Observing Systems Workshop, Potomac, MD.
Halliwell, D. H., McArthur, B., Abboud, I., and Wu, E. (2002). All-weather cavity measurements at Bratt's Lake. Seventh BSRN Science and Review Workshop, Regina, Saskatchewan, Canada. https://www.wcrp-climate.org/documents/bsrn7rpt.pdf
L156: Was this really 1 min sampling or rather 1 min averages of 1 Hz sampling?
L160-161: Specifically, I think it is a 10 W heater + 5 W ventilator.
L165: “allows to automatically clears” is incorrect grammar: “allows automatic clearing of snow” or something. Also remove the “s” in “details” L166.
L174-177: What sort of battery? Do you have confirmation that the battery will function properly and provide sufficient power (for how long?)?
Section 2.3: Adoption of the fractional heating strategy will have influences beyond the compromise between ice mitigation and power. One of the challenges with heating is that it must be applied continuously and consistently. The thermopile produces a voltage when there is a temperature differential across it. The design intention is that the source of measured voltages is differential heating from the target radiation source (same instrument regardless of LW or SW) but gradients in temperature of the sensor body can also produce thermopile voltages. There is a widely-studied natural version of this phenomenon termed IR-loss (see Dutton et al. 2001 10.1175/1520-0426(2001)018<0297:MOBDSI>2.0.CO;2 and work thereafter) whereby the side of a pyranometer exposed to the sky cools radiatively, producing some spurious negative voltage that presents as a negative solar flux at night. Heater applications are less directly studied, but the principle is similar: whenever you turn a heater on or off there will be a signal response that is measured while the system is transiently equilibrating. Cox et al. (2021, 10.5194/amt-14-1205-2021) shows the equilibration period to be hours, though the tested system was an Eppley, which is significantly more massive than the CNR4, so it’s probably a maximum estimate. Philipona’s (2002, 10.1029/2002JD002396) analysis of Kipp & Zonen pyranometers suggests maybe 10s minutes (e.g., Fig 3) and those would be a more comparable mass, but the equilibration time is not the focus. Ji (2007, 10.1175/JTECH1977.1) and maybe Ji and Tsay (2010) also discuss the effect. It’s also worth noting that operating in a housing with the ventilation off is not the same as operating without a housing (Sanchez et al., 2017, 10.1175/JTECH-D-16-0163.1) and that even the type of fan matters (Michalsky et al. 2017, 10.1175/JTECH-D-16-0224.1). This does not prevent one from duty-cycling a heating/ventilation system, but it does mean at least two things: It is impractical to duty-cycle rapidly and the user should remove data during transient equilibrium.
L216-220: I appreciate that we got to this point. I think it should appear in the introduction.
Section 2.5: We see so little of the device in this paper! All we see are far away photos meant to show snow removal performance rather than the design. Please show close-up photos of the SRD with parts labeled and perhaps a schematic to highlight the mechanism!
L237-239: I don’t understand this. It sounds like it can only clean the very top of the dome.
Figure 2a/L286: Some housings have flat shields and form a “donut” of snow around the dome when the dome is kept clear by the ventilation. The problem is much less on sloped housings. So the BSRN Cold Climates Working Group recommended slanted shields a while back (BSRN 2012, https://epic.awi.de/id/eprint/47763/1/BSRN-12-2012.pdf). Presumably the same logic applies here: perhaps the SRD housing could be modified to have a tapered triangle top like a house to help shed snow?
L307: What causes a failure? If it’s ice obstruction, is there a risk the motor will burn out?
L316: fix grammar: “allowed to discover”
Section 3.3: I’m glad we got to this point, but as I said earlier, I think there is plenty of prior work on this point. Is it possible to quantify the equilibrium time for the CNR4 so that would can make acceptable recommendations for a longer duty cycle with blackouts that could be suitable?
L329-331: This is because the problem isn’t just a poor measurement of the temperature (although it is also that). Even if you do know the temperature of the thermopile precisely (the sigmaT^4 part of the calibration equation), the thermopile is measuring a voltage induced by the temperature gradient through the system (positive or negative depending on whether the heater was just turned on or off). This voltage biases the first term in the V/C portion of the calibration equation. I don’t think you can correct for this with a single-thermistor radiometer, but one version of this is the intention of the dome correction factor in the Albrect and Cox (1977) calibration, but less massive modern radiometers have done away with the second thermistor.
Section 4.3: I’m not sure what this discussion is adding. An increase in LWD for the snow-covered sensor is completely consistent with expectations. Also, the pyrgeometer measurements should not be sensitive to the sensor temperature, because this is accounted for in the calibration (as you discuss in your Eq 2). The problem arises only when the temperature of the senor is not homogeneous and/or the thermistor measuring T_sensor does not actually represent the temperature of the thermopile (two separate but related problems).
A few other questions I had along the way:
Is the SRD adaptable for other radiometer models?
Were there other types of ice (frost, freezing rain/drizzle, rime) during the experiment that affected the radiometers but would not be cleaned by the SRD? How did you deal with this?
How much snow can the SRD clean? It probably does not work to allow 20 cm of snow to accumulate before actuating the arm. So does it operate frequently throughout precipitation events to clean smaller amounts of accumulation incrementally?