Surface-based measurements of broadband shortwave (solar)
and longwave (infrared) radiative fluxes using thermopile radiometers are
made regularly around the globe for scientific and operational environmental
monitoring. The occurrence of ice on sensor windows in cold environments –
whether snow, rime, or frost – is a common problem that is difficult to
prevent as well as difficult to correct in post-processing. The Baseline
Surface Radiation Network (BSRN) community recognizes radiometer icing as a
major outstanding measurement uncertainty. Towards constraining this
uncertainty, the De-Icing Comparison Experiment (D-ICE) was carried out at
the NOAA Atmospheric Baseline Observatory in Utqiaġvik (formerly
Barrow), Alaska, from August 2017 to July 2018. The purpose of D-ICE was to
evaluate existing ventilation and heating technologies developed to mitigate
radiometer icing. D-ICE consisted of 20 pyranometers and 5 pyrgeometers
operating in various ventilator housings alongside operational systems that
are part of NOAA's Barrow BSRN station and the US Department of Energy
Atmospheric Radiation Measurement (ARM) program North Slope of Alaska and
Oliktok Point observatories. To detect icing, radiometers were monitored
continuously using cameras, with a total of more than 1 million images of
radiometer domes archived. Ventilator and ventilator–heater performance
overall was skillful with the average of the systems mitigating ice formation
77 % (many
The US Government retains and the publisher, by accepting the article for publication, acknowledges that the US Government retains a nonexclusive, paid-up, irrevocable, worldwide license to publish or reproduce the published form of this work, or allow others to do so, for US Government purposes.
Radiative fluxes are fundamental environmental observations made regularly
from the earth's surface using thermopile radiometers. In cold climates, ice
from vapor deposition (frost), contact freezing of supercooled droplets
(rime) and accumulation of snow are all commonly observed by station
personnel to obscure sensors, and manual cleaning of sensor domes is a
routine activity. Icing is the source of one of the least constrained
outstanding uncertainties in broadband radiometry in cold climates. For
radiometers mounted facing upwards, ice generally increases the measured
longwave downwelling (LWD) flux because the brightness temperature of the
contaminating ice is typically larger than that of the sky. The relatively
cold background of the sky also facilitates radiative cooling of the sensor
window, which exacerbates icing relative to instruments pointed towards the
ground. Biases can be both negative or positive in affected shortwave
downwelling (SWD) fluxes by attenuation or scattering of incident light,
respectively. The magnitude of the instantaneous errors has been reported to
be up to 80 W m
Recognition of the problem and mitigation attempts have been reported since the earliest era of polar radiometric observations more than five decades ago (e.g., Koerner et al., 1963). Since then, engineering solutions have been pursued by research institutes and industry, largely independently and in parallel. In practice, because the nature of the measurement is sensitive to thermal instabilities within the instruments (e.g., Michalsky et al., 2017), the application of heat as an ice-mitigation technique has limitations. While progress has been made, to this day there is still no agreed-upon approach. The needs of the scientific community also increasingly require high-quality measurements from stations capable of being autonomous for weeks or months at a time. Thus, an automated, low-power solution to the icing problem is sorely needed.
The Baseline Surface Radiation Network (BSRN) (Ohmura et al., 1998; Driemel et al., 2018), under the auspices of the World Meteorological Organization (WMO), is a global network for surface-based radiometric observations; the BSRN is traceable to the world calibration standard, managed using commonly adopted practices and strategically distributed for global coverage. There are six current and former BSRN stations in the Arctic, three in Antarctica and numerous stations at lower latitudes that are located at high elevations and/or experience icing conditions seasonally. In 2008, BSRN established the Cold Climates Issues Working Group (CCIWG) to address uncertainties in cold regions, including icing (Lanconelli et al., 2011). Several BSRN stations affected by icing have reported increased data capture rates using ventilators, including the Sonnblick station in the Austrian Alps (Weisser, 2016) and the Georg von Neumayer station in Antarctica (BSRN, 2016). The US Department of Energy (DoE) Atmospheric Radiation Measurement (ARM) program North Slope of Alaska Radiometer Campaign also reported that high-flow ventilation was a useful technique, but that ice mitigation was further improved when the air was also heated (BSRN, 2012). A consensus in BSRN thus emerged that heating and ventilation are capable of mitigating ice, but the effectiveness and uncertainties remained poorly quantified, and the range of experiences reported by BSRN users indicated that more work was needed to constrain the attributes of effective designs (BSRN, 2016).
To address these objectives, the NOAA Physical Sciences Laboratory (PSL) in partnership with the BSRN-CCIWG and NOAA Global Monitoring Laboratory (GML) carried out the De-Icing Comparison Experiment (D-ICE) to collect data suitable for assessing the influence of icing on the measurements and evaluating the status of ice-mitigation technology. D-ICE was deployed at the GML Barrow Atmospheric Baseline Observatory near Utqiaġvik, Alaska, from August 2017 through June 2018. This location was chosen because a variety of icing conditions characteristic of high latitudes regularly occur there and it is home to two long-term operational stations, one from BSRN (NOAA-GML) the other from DoE-ARM. D-ICE collected new data at the NOAA observatory using a variety of radiometers and housings that have been developed to mitigate the formation of ice or are used in icing environments. The systems were contributed by academic and government research institutions as well as development departments of commercial radiometer vendors and were installed alongside the existing operational suites. The systems were monitored continuously using cameras for the duration of the campaign.
In this paper, we describe D-ICE and associated data sets, which are available for future analyses. These data sets include a 10–15 min resolution classification of the icing status of instruments, quality-controlled versions of the radiometric data with occurrences of icing retained and rejected (Cox, 2020a), and a verified ice-free “best-estimate” (BE) baseline (Cox, 2020b) for comparison produced by the aggregate of the quality-controlled data. We use these data sets to analyze instantaneous and time-averaged biases caused by ice, to calculate ice-mitigation performance statistics for the participating systems, to discern some of the reasons for successful ice mitigation, and to gather insight for interpretation of ice-contaminated data.
D-ICE solicited contributions of radiometric ice-mitigation systems
developed by research institutions and industry manufacturers to be part of
the campaign. In total, 26 systems were included: 21 housing pyranometers
(measuring “global” (hemispheric diffuse
The ice-mitigation strategy used by all contributors was some combination of heating and ventilation, in some cases supplied by separate housings in which radiometers were set and in others integrated into the instruments themselves. This consistency in approach is not surprising. Though other methods have been proposed, such as automated alcohol rinses (e.g., Persson and Semmer, 2010), the use of ventilators for controlling ice is pragmatic because ventilation is already regularly used for maintaining thermal homogeneity in the instrument. However, no specific criteria were given to potential contributors, and D-ICE set up each system as instructed. Several sets of redundant housings were used with different radiometers or with only small modifications (see File S1 in the Supplement). All systems were powered using 12 or 24 VDC except for one 48 VAC heater. All fans were powered by DC, which is less prone to propagation of added uncertainty into the signal (Michalsky et al., 2017), in particular from infrared loss in pyranometers (Dutton et al., 2001).
The instruments were installed on the east end of the GML observatory roof in a single line along a 4.9 m table positioned perpendicular to the predominant wind direction (Fig. 1a) near the BSRN tracker. The purpose of this orientation was to reduce the possibility of instruments being influenced by heat produced from neighboring systems, taking advantage of dominant easterlies characteristic of the site (e.g., Cox et al., 2012). The table was constructed from aluminum with a top consisting of fiberglass resin to electrically isolate the systems. The BSRN global pyranometer was positioned on this table. Refer to File S1 for a complete record of system specifications and S2 for a list of modifications made during the course of the campaign. Individual radiometers are referenced in the text by their serial number and the ventilators by their model number. The positions of the systems are displayed in Fig. 1b, labeled with numbers that are referenced where appropriate and cross-referenced in File S1.
D-ICE data were collected using four Campbell Scientific CR1000 data loggers
in individual logger boxes; most systems were analog, but data were also
logged digitally from seven sensors. Fan speeds and heating current were
logged whenever possible. All data were recorded as 1 min averages of 1 Hz sampling except for the digital systems, which were switched to 0.5 Hz
sampling on 26 October 2017 because lags that occurred in digital
communications at temperatures below
Before deployment in June 2017, the radiometers were calibrated at the NOAA-GML calibration facility in Boulder, Colorado. Per standard procedures, the calibration data were collected without use of the ventilators but did use the same data acquisition system that was later deployed. The digital systems were also included in this procedure for comparison but were not assigned new calibration coefficients because it is impractical to do so. The pre- and post-campaign calibrations (File S3) were found to be within uncertainty for all instruments. The pre-campaign calibration values determined by NOAA-GML are used in the processing of the final data set.
All systems on the D-ICE table were monitored using three 720p low-light (0.1 lx) cameras in heated enclosures. The cameras recorded images every 15 min and were set up such that each captured approximately one-third of the table. They were installed facing west (away from the predominant wind direction). Two 18 W LED flood lights were fixed to poles to illuminate the table for the cameras. The lights were automatic and only on during low-light conditions. The cameras were functional and unobscured by ice for 97.6 % of the campaign. ARM also installed cameras facing the trackers at OLI and NSA with 10 min sampling.
The BSRN and ARM operational systems received their routine daily maintenance procedures. Daily cleaning was performed to remove contaminants such as dust and salt residues, but also ice. Since one of the objectives of D-ICE was to monitor icing it was important to allow icing events to unfold naturally. Therefore, the D-ICE radiometers were cleaned daily only when there was no ice present. Infrequently, in cases when ice persisted on a particular radiometer long after the end of an event, the ice was removed. These dates were 24 October; 14, 22, 25, and 29 (no. 1) January; 7 February; and 14 and 27 March (no. 10). Interestingly, we found that icing can be induced by the very maintenance procedures that are designed to remove it. The use of alcohol (such as ethanol) to clean the domes is common practice and was documented during tests at D-ICE to sometimes result in immediate re-icing of the dome. The precise reasons for this are not known, but it is likely a combination of refreezing meltwater from the ice that is residual being slower to evaporate than the alcohol, and/or atmospheric vapor deposition induced by cooling of the dome from the evaporative process. Complete drying of the dome after cleaning was found to reduce this problem.
In this section we characterize the natural icing events that occurred in the environment surrounding the D-ICE systems to set the context for the types and frequency of icing events to which the ice mitigation systems were subjected. During August and September, the temperatures were persistently above freezing with occasional light snow and frequent rain (Fig. 1c). Significant icing was not observed until a prolonged cold period after 22 October, with only brief frosts prior on 28 September and 10 October. Warm temperatures and rain returned during the first week of November and more winter-like conditions prevailed only in the second half of the month. Autumn 2017 experienced record late freezing of the Beaufort and Chukchi seas (Overland and Wang, 2018), with freezing beginning in earnest north of Utqiaġvik in late November. Because of the predominant onshore flow at Utqiaġvik, autumn temperatures there remain near-freezing until after the sea ice isolates the supply of heat from the ocean (Wendler et al., 2014) and the onset of the snowpack is subsequently delayed in late freeze-up years (Cox et al., 2017). These conditions may have also contributed to the delay in the start of the 2017 icing season.
During the winter, both rime (usually from freezing fog) and frost were
regularly observed in the environment surrounding the D-ICE systems
(distinguished qualitatively from the images), spanning a total of 28.8 and
66.3 d, respectively. Frost events were more common, being identified 108
times compared to 11 rime events, but the duration of individual rime events
was longer. The mean duration of frost events was 0.61 (
During the campaign, 34.9 % of the time that rime or frost was observed to
be present in the vicinity of the D-ICE systems, the station meteorology
indicated that the relative humidity with respect to ice (RHI) was
Here we describe the processing of the data streams, beginning with review and classification of the images in Sect. 2.3.1 and then the radiometric data in Sect. 2.3.2, summarized in Table 1. The processed data streams were then used to produce a BE data set that is the average of the calibrated, bias-corrected, ice-free, and quality-controlled data streams in Sect. 2.2.3, from which uncertainties are derived in Sect. 2.3.4. A second ice-estimate data set was also made that received all of the same treatment except that occurrences of icing were retained for analysis.
List of quality control procedures received by D-ICE instruments. An “X” denotes that the procedure was implemented and “O” indicates that the procedure was not implemented.
The images captured approximately 780 000 views of the D-ICE radiometer domes with an additional 143 000 and 125 000 views captured by ARM at NSA and OLI, respectively. Images were captured of the BSRN global pyranometer and all 25 D-ICE radiometers, but not the instruments mounted on the BSRN tracker. At NSA and OLI, images of the global SWD, DIF, and LWD tracker radiometers were captured, but only limited images of the pyrheliometers were made (see Stuefer et al., 2019). The status of each dome in each image was recorded in a spreadsheet after manual review. Because of the large volume of images, this was done in movie form in 1-month intervals, one radiometer at a time. The radiometer domes were classified as being wet (e.g., raindrops or melted ice and slush); containing frost, rime, or snow accumulation; having accumulation of snow around the domes (but not on the domes); being wet with ethanol (used for cleaning); and (rarely) having “other” contaminants, such as resting birds. Occurrences of rime and frost always took precedence in the classification. For example, in cases when snow and rime simultaneously affected a radiometer, the status of the instrument was recorded as rimed. Note that because the domes are hemispheric, the cameras were blind to some parts of the domes, though this was somewhat alleviated by the fact that the pyranometer domes are transparent and the pyrgeometer domes are relatively small and/or flat. All visible ice regardless of amount or coverage was recorded. Thus, the classification was conservative; a snowflake or thick coating of rime were both flagged as iced. Camera downtime was also indicated.
To increase the robustness of the icing determinations, additional instances
of ice were identified by comparing each of the data streams to the average
of all the data streams and reviewing the images where anomalies were found.
While this procedure successfully identified instances of icing that had
been missed, the number of identifications increased by
While the BSRN instruments that were mounted on the solar tracker were not imaged by the cameras, the tracker instruments provide important information for two reasons: first, the pyrgeometers were shaded, which reduces solar heating of the domes (Alados-Arboledas et al., 1988) and the magnitude of associated corrections that apply to some pyrgeometers (Albrecht and Cox, 1977), and second, because SWD is more accurately represented by the sum (hereafter, “SUM”) of the DIF and DIR due to increased calibration uncertainty in pyranometers from the direct beam at low sun angles (Michalsky et al., 1995). All BSRN data were quality controlled with manual screening and application of the relevant definitive tests described by Long and Shi (2008). The manual screening removed suspect data and shadows from station structures. The BSRN tracker measurements were supplemented where there were missing data by the SUM from the ARM QCRAD value-added product from the neighboring ARM station, which is also based on Long and Shi (2008). The resulting data set was used as an intermediary processing step for two purposes: first, to provide a baseline to aid in identification of shadows on the D-ICE instruments and second to provide a statistical baseline for correcting or validating the aforementioned sources of uncertainty in the D-ICE measurements.
The amount of light pollution from the camera LEDs measured by the
pyranometers was determined empirically for each instrument by comparison to
nighttime periods on 19–20 and 30 September and 1–2 October 2017 when the
illumination was switched off. The calculated biases were then subtracted in post-processing when the lights were on. These biases were small,
ranging from
Light poles, as well as some additional station structures such as nearby
aerosol inlet pipes, were minimal obstructions to the view of the sky by the
radiometers except for episodic appearances of shadows on clear days that
reduced the signal in the pyranometers. The shadows occurred at different
solar azimuth and zenith angles for each pyranometer and were only present
when the sun was unobstructed by clouds. The times when each instrument was
shadowed were identified by a reduction of normalized total irradiance
signal exceeding
Data from each D-ICE radiometer were processed with the same Long and Shi (2008) procedures as the operational systems. This was followed by manual screening. One radiometer (a1571), which is typically operated unventilated, was experimentally set in a ventilator that was later found to shadow its thermopiles. Another (26214) was unlevel, but without the possibility of re-leveling after installation. Both of these radiometers are excluded from the radiometric analyses and BE product but are included in the analysis of the ventilator de-icing performance in Sect. 4.1.
The Long and Shi (2008) “QCRAD” approach is designed to identify outliers
relative to the data stream being screened. It therefore relies on the
assumption that most of the data fall within normal limits and is only
sensitive to data that do not. Figure 2 shows examples of the
“climatological configurable limits” for a D-ICE pyranometer in panel (a)
and a pyrgeometer in panel (b). Since occurrences of icing rarely produce
signals that fall outside the statistical distribution, the spurious data
are not readily captured by outlier-detection methodologies, such as QCRAD.
For example, the second-level threshold for the definitive configurable
QCRAD limit flagged
Examples of climatological configurable limit tests from Long and Shi (2008) for SWD (SN F16305R)
The results from the processing of the D-ICE images were used to flag data contaminated by the presence of ice on the radiometer domes. These data were removed from the BE data set, but a second version with occurrences of icing preserved was needed to calculate the biases caused by the ice. To construct such a data set, only data that had been rejected for failing physically possible limit tests or having been determined to be shadowed were removed, while outliers flagged using other tests that were within physically possible limits when ice was present were retained.
Infrared loss corrections were applied to pyranometers that exhibited nighttime offsets following the method of Dutton et al. (2001) (see Table 1),
though the offsets observed during D-ICE were consistently small (generally
The BE data set was produced by averaging the calibrated, bias-corrected, ice-free, and quality-controlled D-ICE data streams. For LWD, this consisted of all 8 upward-facing pyrgeometers (5 from D-ICE, 2 from NSA and 1 from BSRN). For SWD, this consisted of 17 pyranometers from D-ICE, the global BSRN, and the global NSA. Since the sensitivity of a thermopile is not precisely isotropic, the calibration of global pyranometers is designed to be well-suited for the daily average but prone to varying errors through the day as the incident angle of the direct beam changes. Thus, to produce a BE, the average of the global pyranometers could be used to constrain the SUM from the trackers, or the tracker measurements could be used to bias-correct the average of the global pyranometers. We chose to do the latter because the large number of included data streams produces a data set less prone to discontinuities and noise and importantly was also directly verified as ice-free (Sect. 2.3.1). Thus, the SWD average was bias-corrected as a function of solar zenith angle (SZA) and the diffuse fraction using the SUM; the magnitude of the correction varied between 0.5 % and 3 % depending on the diffuse partitioning.
Uncertainty is estimated empirically as the 1
Uncertainty in SWD is plotted Fig. 3b as a function of SZA. In addition to
absolute units, the uncertainty is also shown in relative units (%). The
BSRN target uncertainty for pyranometers is 2 % (McArthur, 2005), a
condition that is met, and the uncertainty is relatively flat when the SZA
is
To better understand the consequences of icing, Fig. 4 shows a case study for LWD from late January in panel (a) and for SWD on 14–15 April in panel (b). Analysis of these cases, next, is followed in Sect. 3.3 with a more general calculation of biases at the monthly scale.
Case studies of icing for LWD in late January
The LWD time series spans approximately two weeks and shows a range of LWD
typical of the Arctic winter, from
Figure 4b shows an example of a clear-sky day in mid-April that followed frost formation the previous night. The time during which the frost was observed to be growing through deposition ended at approximately 17:00 UTC on 14 April after which the frost sublimated during the day. The black line in the figure shows the SWD BE, and the grey shading is the uncertainty. Five pyranometers that had frost on their domes for at least some of the day are shown by the colored lines. Missing data in the figure are because of shadows.
The biases from the ice are generally positive (up to
To illustrate the influence of ice during diffuse conditions, another
example from 7 April (Fig. 5) shows the transition from a positive bias
(dominated by scattering) to a negative bias (dominated by attenuation) in
F16305R capped with ice due to a transition in lighting. From 20:00 to 20:30 UTC the direct beam is present, being 10 %–30 % of the total irradiance,
during partly cloudy conditions. At that time, a bias of up to
Case study of SWD on 7 April 2018. The solid black line and grey shading are the best estimate and uncertainty; the solid blue line is an iced pyranometer shown in the inset images. The dashed blue line is the bias in the solid blue relative to the solid black, and the yellow line shows the percent of the irradiance contributed by the direct beam: when this value is near zero, the lighting is diffuse under overcast conditions.
These cases demonstrate that errors from ice in SWD can be large and that the sign of the bias is dependent on the amount of coverage of ice on the pyranometer dome, as well as the presence, and likely also the angle, of radiation from the direct beam.
Figure 6 shows monthly mean biases in LWD (panel a) and SWD (panel b) for each radiometer. For each cell, the color indicates the bias associated with frost, rime, snow, or liquid (usually ice melted by heat from the ventilator). The monthly means are also plotted as a time series in panels (c) and (d) with the aggregate means shown by solid lines. Note that the months of July and August include limited amounts of data because of beginning and end dates of the campaign. All bias calculations are corrected to account for differences between individual radiometers and the BE that are associated with calibration uncertainty. The bias calculations are insensitive to the determination of ice occurrences from the images because the average of all conditions, regardless of ice presence, is calculated.
Calculations of biases in monthly means relative to the BE for LWD
As noted in Sect. 3.1, the LWD case study was chosen because it was a
particularly influential event, and a particularly susceptible system was
highlighted. Figure 6 shows that when data are averaged for long periods of
time, the bias in icing of pyrgeometers is actually small. Indeed, only two
radiometers, BSRN and 28507 (having similar configurations and equipment),
exhibit biases that are detectable relative to the average uncertainty
(Fig. 3). The most severely affected month was January when the average
bias was just
SWD icing biases during D-ICE occurred from February through June with a peak in April. This is because biases in SWD depend both on the amount of sunlight and the amount of icing, which have opposing seasonal cycles. The opposition is slightly out of phase such that in autumn there was too little sunlight when the icing first began in earnest, but that both substantial amounts of sunlight and icing co-occurred during spring. Recall from Sect. 2.2 that the beginning of the icing season was late during D-ICE and that in a more typical year at Utqiaġvik some biases may also have been observed in September and October. Note that the calculation includes the average of both negative and positive biases. If the average of the absolute value of the bias is plotted instead (not shown), the biases increase slightly but interpretation is hampered by the fact that noise contributes to the bias calculation rather than canceling out. Nevertheless, the results indicate that biases in pyranometers at D-ICE were dominated by positive perturbations, which is consistent with spurious data being principally tied to a combination of clear skies, low sun angles, and capping in early spring.
To assess ventilator performance, we begin with two qualitative examples
that broadly illustrate the influence that heating and ventilation have in
mitigating ice. The first example is of a freezing fog event that occurred
from 12:30 UTC on 5 January with rime accumulation continuing until about
09:00 UTC on 6 January. The image in Fig. 7a shows the status of the systems
during the event at 19:00 UTC on 5 January. At this time, rime is observable on
the domes of some of the systems while most remain ice-free. Immediately
after these images were taken, the power was deliberately cut to the
ventilators and the radiometers began accumulating ice immediately, being
iced over within 2.5 h (Fig. 7b). The second example began on 21:30 on
9 January when the RHI
From CAM1, a freezing fog event in progress on 5 January and resulting in riming.
To quantify performance over the course of the campaign, Fig. 8
shows a summary of statistics from the systems at D-ICE, NSA, and OLI based
on the classification of the images described in Sect. 2.4.2. The systems
are labeled on the
Ice mitigation performance metric,
Uncertainty in
While no systems were found to be 100 % effective, two-thirds of the
systems, including all those housing pyrgeometers, were effective at least
80 % of the time and 15 out of 34 were effective at least 90 % of the time. The
average was 77 %, but there was also a substantial amount of variability
(
Interestingly, the CMP22 outperformed the SMP22 in the CVF4 by 20 % despite the similarity between the ventilation systems and the radiometers. The only difference was that on 6 January the air intake screen on the CVF4 holding the SMP22 was removed to assess whether clogging by snow and reduced air flow impacted effectiveness. The CMP22 was observed to outperform the SMP22 by 19 % prior to this change and 21 % after, so the difference is not attributable to the presence of the screen and the screen apparently had little impact on effectiveness. We do not know the explanation for the observed difference.
In general, mitigation of ice on pyrgeometers was more effective than pyranometers, even for cases when the systems were otherwise similar (e.g., CVF4 and VEN systems). There are several plausible, but not mutually exclusive, explanations for this: first, the domes of the pyrgeometers are smaller and have a lower profile, and therefore aspirated air may be more easily circulated to the top of the dome; second, the smaller surface area of the dome supports improved conduction of heat, as does the fact that pyrgeometer domes are constructed of silicon, which is more thermally conductive than the quartz pyranometer domes; and third, more speculatively, the outer coatings of the domes may be less prone to accretion of ice.
Successful mitigation of ice is demonstrated by systems in Fig. 8 that were not equipped with heaters. This supports the heuristic within BSRN that ventilation of ambient air alone can be effective. However, it is counter-intuitive because aspiration of saturated air increases rather than decreases deposition rates, specifically resulting in denser, but not necessarily thicker, frost (Kandula, 2011, and references therein).
We examined the properties of the Eppley ventilation system configured
similarly to those in use at the Barrow BSRN station to help elucidate the
attributes that contribute to effectiveness in the absence of heating
elements. The tested system is an Eppley VEN housing a high-flow 80 cfm
(10.3 W) DC fan (Delta Electronics FFB0812EHE) modified with bearings rated
for low temperature; examples of such systems at D-ICE are in positions 6–9
and 24 (Fig. 1b). When the fan is operated outside of the ventilator, the
velocity of the air downstream is
Figure 9a shows a 9 h time series of temperatures collected during D-ICE
in January 2018. The dome temperatures from PIRs 28507 and 34309 were
0.5–0.6
The fan in 34309 was turned off shortly before 20:00 UTC on 8 January for
To better understand the sources of the heating, the experiment was repeated
under controlled conditions in a laboratory in Boulder, Colorado. First, an
FFB0812EHE was placed in a cold chamber without the VEN, and a thermocouple
(Type T; Copper-Constantan) was positioned in the air stream
Next, the experiment was repeated again using an FFB0812EHE installed in a
VEN containing a PIR and having a shield with a 1 mm lift and clay sealing
around the shield edges. Similar to the previous iteration, a 0.39
The heating of the dome by the fan is principally from two sources. The
first is heating of the air moving past the fan motor, which is warmed by
waste energy. This can be calculated by first estimating the amount of waste
heat in watts,
The second source is that the air downstream immediately in contact with the
radiometer necessarily comes to rest and thus undergoes an adiabatic
compression. This topic has been studied extensively for high velocity flows
(e.g., Thompson, 1968; Lenschow, 1972), but less at low velocities. At low
velocities, the properties of the gas can be approximated as ideal.
Therefore, we formulate the problem from the first law of thermodynamics
beginning with the ideal gas law, differentiating as follows:
The results (Fig. 9c) indicate that for the experimental setup in
question, waste heat and adiabatic heating contribute similarly to the
observed temperature increase in the chamber, with
The lab experiments resulted in about half of the total heating that was
observed in the January D-ICE case. D-ICE was carried out at sea level,
whereas the tests in Boulder (
Interestingly, that heating from waste heat increases as air velocity is decreased (Fig. 9c) counters the observed relationship between increasing the lift of the shield and increases in both air velocity measured at the top of the dome and dome temperature. This suggests that the effect of raising the shield is to enhance the circulation of air around the dome.
The De-Icing Comparison Experiment (D-ICE) was carried out in 2017 and 2018
at the NOAA Atmospheric Baseline Observatory in Utqiaġvik, Alaska
(71.3
System performance, defined as the amount of time a radiometer was
classified as iced normalized by the amount of time icing conditions were
present (Eq. 1), was in the mean amongst the systems, a 77 % reduction in
the expected amount of time the systems were iced. Thus, on average the
systems tested during D-ICE were successful in mitigating most ice. Ice was
more effectively mitigated from pyrgeometers than pyranometers. Many systems
housing either type of radiometer were 90 % effective or better, including
some that did not use external heat. Even systems without external heating
elements were observed to have radiometer domes that were warmer than
ambient air by 0.5 to 0.6
Generally, we did not identify significant errors caused by the ventilators, and the nighttime offsets in all systems were small, consistent with Michalsky et al. (2017). One exception was heated ventilators that were susceptible to clogging by snow. These were observed to have small nighttime offsets correlated with wind velocity but not net longwave, the latter being expected for errors from infrared loss (Dutton et al., 2001). Instead, we postulate that blowing snow clogged the ventilators, reducing aspiration and causing differential heating of the radiometer after which during calm winds the heated ventilator unclogged the inlet.
When ice was present on sensors, the instantaneous biases varied but could
be large, up to
Consistent with earlier studies reporting difficulty in distinguishing iced data in post-processing (Lanconelli et al., 2011; Matsui et al., 2012), we find that quality control procedures are poorly suited for detection of iced data because the signal caused by ice is not statistically outside the range of variability in the signal caused by clouds. Therefore, common screening methods (e.g., Long and Shi, 2008) are insufficient. Some of the non-definitive tests proposed by Long and Shi that involve cross-comparison between sensors may be more likely to identify suspect data, but results are dependent on the differential icing characteristics between the sensors. Other tests have been proposed such as comparing the sign and time derivative of the difference between upward and downward LW and SW measurements (van den Broeke et al., 2004; Wang et al., 2018), though these tests rely on similar assumptions. Some studies rely on logbooks from station personnel and thresholds for relative humidity (e.g., Sedlar et al., 2011; Miller et al., 2015, 2017; Persson et al., 2018), but because of the relative infrequency of observer records (e.g., daily) and suspect reliability of RHI as a proxy for icing (Sect. 2.2), these methods also have limitations. D-ICE demonstrates success in quality control by monitoring instrumentation with cameras, but this approach is not always practical. In keeping with van den Broeke et al. (2004), we suggest that time-derivative analysis for ice detection should be further explored. For example, the variability in the iced data in the SWD case (Fig. 4b) is much slower and smoother than would be expected from clouds in a regime not dominated by the diffuse. Thus, development of new algorithms that flag iced data based on time-variant tests might be possible if the regime can be determined to be dominated by the direct beam and can be distinguished from the consequences of instruments being unlevel or expected differences between the global and SUM SWD though the day, both of which can produce structurally similar errors.
Finally, as a baseline for comparison used for analysis, a “best-estimate” data set was produced using a combination of the measurements that were ice-free. Though an unexpected outcome of D-ICE, the number of radiometers, variety of systems, and skillful performance of the systems resulted in production of a verified ice-free data set that is nearly 100 % complete for the duration of the campaign. Empirically based uncertainties were also calculated from the variability amongst the ice-free observations. This data set is uniquely well-characterized in the Arctic and therefore may be suitable for use beyond inquiry related to ice mitigation. For example, D-ICE took place during the Year of Polar Prediction (YOPP) at one of the YOPP “supersite” observatories and the campaign spanned the first wintertime YOPP Special Observing Period (SOP1) during February and March 2018. We therefore propose that the D-ICE best-estimate data products (Cox, 2020b) may be useful for model evaluations, such as the planned YOPP site Model Intercomparison Project (YOPPsiteMIP).
Images collected by ARM as part of DICEXACO are available from the ARM data
archive (
The supplement related to this article is available online at:
CJC and SMM designed, implemented, and led the experiment with input from TU and CNL. CJC, SMM, TU, RB, EH, JW, MK, CNL, AM, and BDT conducted the field programs. CJC led the data analysis and wrote the paper with input from all co-authors.
The authors declare that they have no conflict of interest.
This work is dedicated to the memory of the late Charles N. Long (formerly of CIRES) who first suggested the concept for D-ICE and provided invaluable guidance and support in developing and carrying out the experiment. We appreciate deployment assistance from John Booth (NOAA), Nicholas Lewis (Univ. Colorado), Meghan Helmberger (Univ. Colorado), Christine Schultz (NOAA), Andrew Clarke (NOAA), Amanda Looze (NOAA Pathways intern), and Kevin Olivas (NOAA summer intern); field support from David Oaks (Fairweather LLC), Ben Bishop (Sandia), and Walter Brower (UIC Science, retired); engineering and equipment support from Robert Albee (formerly STC); logistical support from Brian Vasel (NOAA), Jim Mather (PNNL), Mark Ivey (Sandia), Fred Helsel (Sandia), and Martin Stuefer (Univ. Alaska, Fairbanks); useful conversations with Robert Zamora (NOAA, retired), who assisted with adiabatic heating calculations; Ola Persson (CIRES), who suggested that natural de-icing processes vary geographically; Gert König-Langlo (AWI, retired), Jackson Osborn (CIRES), and Matthew Shupe (CIRES); Matthew Martinsen (NOAA), who provided science and logistical guidance; Richard Lataitis (NOAA), who provided an internal review; constructive comments from three anonymous reviewers and facilitation by the editors at AMT; and guidance from members of BSRN's Cold Climates Issues Working Group (CCIWG). The following organizations (points of contact) contributed equipment to the campaign: Delta-T (Dick Jenkins), Kipp & Zonen (Victor Cassella), Hukseflux (Jörgen Konings), Eppley (Tom Kirk), EKO (William Beuttell), PMOD/WRC (Julian Gröbner), Environment and Climate Change Canada (ECCC, Andrew Platt), NOAA, MeteoSwiss (Laurent Vuilleumier), NCAR (Steven Oncley, Steven Semmer, Kurt Knudson), and the Alfred Wegener Institute (Holger Schmithüsen, Bernd Loose). Campaign logistical support was provided by the NOAA Global Monitoring Laboratory (GML). The views expressed in the article do not necessarily represent the views of the DoE or the US Government.
This research was supported by the NOAA Arctic Research Program, the NOAA Physical Sciences Laboratory (PSL), the DoE Atmospheric Systems Research (ASR) program (grant nos. DE-SC0013306 and DE-AC36-08GO283), and the DoE Office of Energy Efficiency and Renewable Energy Solar Energy Technologies Office. This work was authored in part by the National Renewable Energy Laboratory, operated by Alliance for Sustainable Energy, LLC, for the US Department of Energy (DOE) under contract no. DE-AC36-08GO28308.
This paper was edited by Manfred Wendisch and reviewed by three anonymous referees.