Airborne observations of far-infrared upwelling radiance in the Arctic

The first airborne measurements of the FarInfraRed Radiometer (FIRR) were performed in April 2015 during the panarctic NETCARE campaign. Vertical profiles of spectral upwelling radiance in the range 8–50 μm were measured in clear and cloudy conditions from the surface up to 6 km. The clear sky profiles highlight the strong dependence of radiative fluxes to the temperature inversion typical of the Arctic. Measurements acquired for total column water vapour from 1.5 to 10.5 mm also underline the sensitivity of the far-infrared greenhouse effect to specific humidity. The cloudy cases show that optically thin ice clouds increase the cooling rate of the atmosphere, making them important pieces of the Arctic energy balance. One such cloud exhibited a very complex spatial structure, characterized by large horizontal heterogeneities at the kilometre scale. This emphasizes the difficulty of obtaining representative cloud observations with airborne measurements but also points out how challenging it is to model polar clouds radiative effects. These radiance measurements were successfully compared to simulations, suggesting that state-of-the-art radiative transfer models are suited to study the cold and dry Arctic atmosphere. Although FIRR in situ performances compare well to its laboratory performances, complementary simulations show that upgrading the FIRR radiometric resolution would greatly increase its sensitivity to atmospheric and cloud properties. Improved instrument temperature stability in flight and expected technological progress should help meet this objective. The campaign overall highlights the potential for airborne far-infrared radiometry and constitutes a relevant reference for future similar studies dedicated to the Arctic and for the development of spaceborne instruments.

the instrument. Although the FIRR has a nominal field of view of 6 • corresponding to a 20 pixels diameter area on the sensor, here only a 15 pixel diameter area is used to avoid the small vignetting on the edges of the illuminated area. This corresponds to a field of view of 4.5 • , which translates into a footprint of 7.8 m at a 100 m distance, and 470 m at 6000 m. Since the temperature aboard the unpressurized cabin quickly varied between approximately 0 and 15 • C, the ambient blackbody (ABB) was maintained at 15 • C, while the hot blackbody (HBB) was set to 45 or 50 • C. These correspond to BB nominal temperatures 5 in flight but some experiments were performed with different BB temperatures depending on the environmental constraints, which is not problematic since the instrument's response is linear in this range of temperature (Libois et al., 2016). One FIRR measurement sequence lasts 210 s, during which approximately 40 s are used to actually take measurements and 170 s are spent rotating the filter wheel and the scene selection mirror. A sequence consists of two calibration sequences (one on the ABB and one on the HBB) followed by 3 scene sequences, each sequence corresponding to one complete rotation of the filter wheel 10 that measures all 9 filters in approximately 40 s. For each spectral band, 100 frames are acquired at 120 Hz and then averaged to provide a single 2-D image. One spectral measurement thus corresponds to a 0.8 s long acquisition and no supplementary temporal average is performed, highlighting the potential for fast scanning compared to interferometers that usually require averaging over several spectra to achieve comparably high performances (e.g. Mlynczak et al., 2006). Such acquisition rate is essential when looking at heterogeneous or quickly moving targets, as is the case from an aircraft or satellite view. It is the 15 main advantage of trading spectral resolution for higher signal levels. Note, though, that measurements in successive spectral bands are offset temporally, hence spatially, which has to be borne in mind at the stage of data interpretation. In this study, the FIRR is not used as an imager, hence the data presented here correspond to averages over the selected area of 193 pixels.
In this configuration, the radiometric resolution of the FIRR in laboratory conditions is essentially limited by detector noise and is about 0.015 W m −2 sr −1 . This corresponds to noise equivalent temperature differences of 0.1 − 0.35 K for the range of 20 temperatures investigated in this study. The radiometric resolution is nearly constant for the 7 bands ranging from 7.9 to 22.5 µm because the absorptivity of the gold black coating is spectrally uniform and the filters all have similar maximum transmittances.
It is approximately 30% less for the filters 22.5 − 27.5 µm and 30 − 50 µm, because of limited filter transmittance for the band 22.5 − 27.5 µm, and reduced package window transmittance for the band 30 − 50 µm. Such performances compare well with similar airborne spectroradiometers (e.g. Emery et al., 2014) and satellite sensors (e.g. MODIS). 25 A critical issue during the campaign was the temperature stability of the instrument in operation. Indeed, the first flights were characterized by excessively noisy measurements, especially in the 30 − 50 µm channel. This noise was due to excessive air circulation within the chimney, cooling down very quickly the calibration enclosure and the filters. In particular, the metallic mesh filter 30 − 50 µm has a very low thermal capacity and its temperature significantly changed in less than 1 s, making the acquired data unusable. A float-zone silicone window was available that could be placed at the entrance of the instrument, 30 but we decided not to use it since its limited transmittance of 30% in the FIR drastically reduced signal level. This issue was fixed on 13 April by partially closing the rolling door in flight to prevent cold air flow from entering the inlet chimney, without impacting the field of view (Fig. 2). For previous flights, the calibration procedure detailed in Libois et al. (2016), that takes Transmittance (%) FIR MIR 7.9 -9.5 µm 10 -12 µm 12 -14 µm 17 -18.5 µm 18. 5 -20.5 µm 17.25 -19.75 µm 20.5 -22.5 µm 22.5 -27.5 µm 30 -50 µm  advantage of non illuminated pixels of the detector to remove the background signal, ensured good quality data for all bands except the 30 − 50 µm.

Other measurements
Polar 6 was equipped with a large set of sensors and instruments but only those relevant for the present study are mentioned below. Air temperature was recorded with an accuracy of 0.3 K by an AIMMS-20 manufactured by Aventech Research Inc.
H 2 O is 39.1 ppmv or 2.5 %, whichever is greater. Broadband longwave (LW) radiation was measured with Kipp & Zonen CGR-4 pyrgeometers installed below and above the aircraft (Ehrlich and Wendisch, 2015). These sensors have uncertainties of a few W m −2 . Nadir brightness temperature in the range 9.6−11.5 µm was measured by a Heitronics KT19.85 II with a field of view of 2 • and an accuracy of 0.5 K. A number of probes also provided qualitative information about the presence of cloud particles. 5 Total and liquid water content were measured with a Nevzorov probe (Korolev et al., 1998). An FSSP-300 particle probe was used to measure particle size distributions from 0.3 to 20 µm from which cloud presence can be deduced (e.g. Ström et al., 2003). A PMS 2D-C imaging probe was supposed to detect larger particles, but the images were obscured due to a problem with the true air speed used in the image re-construction, preventing accurate retrieval of particle size distribution. Practically, this sensor was mostly used to assess the presence of large cloud particles, but did not provide quantitative information about 10 particle shape or size. A sun-photometer specially designed for Polar 6 (SPTA model by Dr. Schulz & Partner GmbH) was mounted on top of the aircraft and continuously tracked direct solar radiation in 10 spectral bands in the range 360 − 1060 nm.
From these spectral measurements, the atmospheric optical depth was deduced and further processed with the SDA method (O'Neill et al., 2003) to retrieve the contributions of the fine (aerosols) and coarse (mainly cloud and precipitation) mode components. In addition to these particle measurements, black carbon concentration was estimated to give an indication on the 15 level of pollution of the investigated air masses. To this end, ambient air was sampled with an inlet mounted above the cockpit of Polar 6, and a Single Particle Soot Photometer (SP2 by Droplet Measurement Technologies, Boulder, Colorado) was used to evaluate the mass of individual refractive black carbon particles per volume of air (Schwarz et al., 2006), from which the mass for particles within the size range 75 − 700 nm was deduced. High resolution nadir pictures taken at 15 s intervals also provided valuable information about the surface and the presence of clouds.

Selected flights
For the present study, 5 vertical profiles taken during 5 different flights were selected. These flights, whose trajectories are shown in Fig. 3 on 7, 11, 13, 20 and 21 April. All profiles were measured above snow-covered sea ice, which ensured that the surface was homogeneous contrary to flights performed above patches of snow and tundra or over areas of mixed sea ice and open water. 25 All the investigated flights except 7 April were taken close to a track of the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations satellite (CALIPSO , Winker et al., 2003). Images taken by MODIS and the associated cloud products (Platnick et al., 2003) are also used to investigate cloud conditions above the aircraft. The 5 profiles were acquired in distinct atmospheric conditions, thus providing valuable samples of Arctic conditions in early spring. April 7 to 13 flights corresponded to typical conditions of the high Arctic cold season, with low temperatures and a pronounced inversion, while the conditions 30 near Inuvik were more representative of subarctic spring, with near-melting temperatures at the surface and denser clouds typically found in the mid-latitudes. Some ice clouds were encountered on 7 April flight, but the more typical polar optically thin ice cloud was probed on 13 April near Eureka. The three other flights exhibited clear sky conditions below the aircraft.

Radiative transfer simulations
One objective of the study was to perform radiative closure experiments by comparing FIRR measurements with radiative transfer simulations based on thermodynamical and microphysical profiles recorded by the instruments aboard Polar 6. Here Aerosols are approximated to the standard rural profile with a visibility of 23 km which is consistent with the presence of Arctic haze during the campaign. Multiple scattering is computed with DISORT (Stamnes et al., 1988) using 16 streams, and the band model is at 1 cm −1 spectral resolution. The model atmosphere has 75 levels from the surface to 30 km, with a resolution of 0.1 km near the surface stretching to 0.7 km at the top. In addition to radiances, MODTRAN was used to compute Jacobians through finite differences (Garand et al., 2001).
Temperature and humidity profiles were interpolated from the in-situ measurements up to the maximum flying altitude.
Above, they were taken from the closest ERA-Interim reanalysis (Dee et al., 2011), the latter being offset to ensure vertical continuity. Ozone profiles for the whole column were also taken from ERA-Interim. Snow surface temperature was obtained 5 from the KT19 observations assuming a uniform spectral response of the instrument and a spectrally flat surface emissivity of 0.995 in the range 9.6 − 11.5 µm. All simulated clouds in this study are ice clouds defined by their optical thickness τ and particle effective diameter d eff . Their single scattering properties are calculated after the parameterization of Yang et al. (2005) for cirrus clouds. Cloud geometrical characteristics were deduced from the combination of in situ observations. Optical thickness and effective cloud particle diameter were not directly measured. For 7 April, both quantities were tuned to minimize 10 the deviation from measurements. For 13 April, the particle effective diameter was taken from DARDAR satellite product (Delanoë and Hogan, 2010) and simulations were performed for various optical depths.

Results
In this section, the FIRR radiometric performances are first analyzed based on experiments performed on the ground and during one flight. The five case studies are then analyzed in detail and the vertical profiles of radiance acquired in clear sky and cloudy 15 conditions are compared to radiative transfer simulations.

FIRR radiometric performances in airborne configuration
The FIRR performances were investigated through laboratory and ground-based experiments by Libois et al. (2016). They estimated a radiometric resolution around 0.015 W m −2 sr −1 and an absolute error of 0.02 W m −2 sr −1 , again slightly dependent on the channel considered. In airborne configuration, the environmental conditions were more demanding due to 20 cold ambient temperature and quick background temperature variations. The FIRR performances for this specific setup are thus estimated from two experiments for which the environmental conditions were similar to nominal airborne operation, except the scene was more constant than in operation. Firstly, the brightness temperature of the snow surface below the aircraft was measured on Eureka runway on 12 April, while Polar 6 was parked without the propellers running. The ambient temperature was around -32 • C, the ABB was at -9.5 • C and the HBB at 20 • C. Secondly, measurements taken on the closed rolling door just 25 before landing on 11 April were analyzed. For this case, the ABB was at 15 • C and the HBB at 45 • C.
The experiment on snow consisted of 10 consecutive measurement sequences covering 30 min, so that 30 radiances were recorded for each spectral band. For all bands, the radiance increased continuously throughout the experiment, which was attributed to an increase of snow temperature. To remove this effect and focus on the resolution of the measurement only, the radiance series were first detrended, and the standard deviation of the residual was then computed. The latter does not exceed 30 0.012 W m −2 sr −1 . The experiment performed on the rolling door consisted of 5 consecutive sequences, and the standard deviation of the signal was larger, reaching 0.021 W m −2 sr −1 . Figure 4a shows the corresponding brightness temperatures for both experiments, highlighting a temperature resolution around 0.1 K above snow and 0.2 K above the rolling door. Although the environmental conditions are slightly different in flight, these results provide a valuable reference and show that the installation of the instrument in the aircraft did not affect its performances.
To further investigate the reduced radiometric resolution observed in flight, Fig. 4b shows the sequence of brightness temperatures recorded on the rolling door. A recurrent pattern is observed within a sequence of 3 consecutive measurements, with 5 the first temperature generally larger than the following ones. We interpret this as the signature of fast and complex temperature variations of the skin temperature of the filters, that cannot be removed through the calibration procedure. We attempted to use the numerous temperature sensors embedded in the calibration enclosure and in the filter wheel to reconstruct the filters actual temperature, but this proved unsuccessful. Without any indication of whether any of the 3 consecutive points is the best, we simply conclude that this thermal instability results in an additive noise of approximate amplitude 0.2 K in worst conditions. 10 This leaves room for future improvement of the instrument. The operational resolution of the FIRR nevertheless remains well below 0.5 K, which is still satisfactory and comparable to temperature measurements performed aboard Polar 6. This issue had not been noticed by Libois et al. (2016), most likely because in their study ambient temperature was closer to the internal temperature of the FIRR, limiting the range of filter temperature variations.

15
The profiles on 11, 20 and 21 April were all taken in clear sky conditions, but the total columns of water vapor were very different. These flights are specifically used to investigate the impact of temperature and humidity variations on the measured profiles of spectral radiances.

11 April
The ascent started at 19:02 UTC and at 19:52 UTC Polar 6 reached the maximum altitude of 5.56 km, where it stayed for 20 4 min. On its way up it also levelled at 2.75 km for 7 min. The surface temperature retrieved from the KT19 was -32.6 • C while a maximum of -24 • C was observed in the atmospheric temperature profile between 1 to 2 km (Fig. 5a). The whole atmosphere was undersaturated with respect to ice, except near the surface. The total column water vapor was 1.5 mm, with 1.4 mm below 5.56 km. No clouds were observed and the Aqua MODIS image taken at 18:45 UTC shows that no clouds were present above either. FIRR brightness temperature profiles show interesting features (Fig. 5b), with the temperature inversion more 25 obvious for the longer wavelengths for which the atmosphere is more opaque. To further illustrate this differential sensitivity to the temperature profile, Fig. 6 shows the penetration depth of each channel as a function of altitude. The channels that penetrate the least are sensitive to the conditions closest below the aircraft. As expected, the brightness temperature in the highly transparent atmospheric window (10 − 12 µm) is essentially constant with height since it is insensitive to the properties of the atmosphere. The slight increase of 0.5 K from the surface to the top is also observed in KT19 records and is probably the Brightness temperature ( space an atmosphere with a strong temperature inversion. The peaks in the shorter wavelengths channels around 4 km were found to visually correspond to variations of sea ice characteristics. They could be due to thinner and warmer sea ice or finer snow with higher emissivity (Chen et al., 2014). Since all individual measurements were used, the vertical resolution is close to 200 m. However, the instability along 3 measurements is noticeable, for e.g. the 18.5 − 20.5 µm channel below 2 km. Besides this instrumental noise, part of the observed signal variation might be due to horizontal inhomogeneity, especially when the 5 aircraft roll reaches up to 20 • in turns.
The vertical profile of upwelling broadband LW radiation also highlights the temperature inversion, with a maximum around 2 km, similar to the FIR channels of the FIRR (Fig. 5c). LW fluxes have been simulated with MODTRAN and are also shown. The simulated and measured profiles are in close agreement above 2 km, with a root mean square deviation (RMSD) of 0.35 W m −2 . Such a value is consistent with the accuracy provided by the manufacturer and the absolute uncertainty of 10 2 W m −2 suggested by Marty (2003) for such sensors. This is very satisfactory for a sensor sensitive only up to 42 µm while a significant part of the energy lies beyond, and considering that the calibration was done above 2 • C. This agreement gives high confidence in the atmospheric profile measurements, but also in the aerosols modelled in MODTRAN, because errors in aerosol profiles could result in discrepancies of several W m −2 (Sauvage et al., 1999). Regarding the upper extrapolated part of the atmosphere, comparison of measured and simulated downwelling LW fluxes (not shown) are also in reasonably good agreement, which gives confidence in the ERA-Interim fields. Close to the surface, measurements show an unexpected peaked minimum. Although the origin of this peak is not fully understood, we believe this is an instrumental artifact resulting from the 5 strong temperature gradient near the surface, and the sensor not being at thermal equilibrium (Ehrlich and Wendisch, 2015).
This hypothesis is supported by the fact that data taken on the way down just before starting the ascent show a peak in the opposite direction.
MODTRAN was also used to simulate FIRR brightness temperatures (Fig. 5b). The measured profiles for all channels are well simulated, with a mean bias and RMSD below 0.2 K. The agreement in the window bands confirms that no clouds were 10 present below the aircraft. FIR simulations provide strong validation of the radiative transfer model, resulting in a satisfactory radiative closure in clear-sky conditions. The spectral brightness temperatures are compared at the two altitudes where multiple measurements were taken. Figure 7 shows the average measured brightness temperatures at 2.75 and 5.56 km, and the corresponding simulations. The spectral RMSD is below 0.15 K at both altitudes, which is very satisfying, given that MODTRAN user's manual suggests that the model accuracy is 1 K. The variability of the measurements at each step is below 0.4 K which 15 is consistent with the results of Fig. 4b. In addition, most deviations between observations and simulations are within the range of uncertainties due to uncertainties of the temperature and relative humidity measurements.
Overall, the simulations reproduce well the observations, which validates to some extent the radiative transfer code configuration and the implemented snow emissivity. However, such measurements can hardly be used for model improvement.
As pointed out by Mlynczak et al. (2016), the inherent uncertainties related to the atmospheric measurements and radiative

20 and 21 April
Both flights took place in the vicinity of Inuvik and showed relatively warm conditions and above freezing temperatures at the 5 inversion level (Figs. 8a and c). The cloud probes suggested that no clouds were present, which is consistent with the relative humidity profiles. For 20 April flight, a moist layer typical of long range transport was found, that peaked near 2.5 km at about 85% humidity with respect to water. Above 3.5 km, this layer was topped with drier air associated with weak air subsidence.
Above 3.8 km, the air was very whitish, and the FSSP-300 and sun-photometer indicate increased level of aerosols. Likewise, SP2 measurements showed increasing amounts of black carbon with altitude, exceeding 0.1 µg m −3 , which is indicative of a 10 polluted air mass. Similar conditions were encountered on 21 April, except that the polluted layer was located above 2.6 km, which again coincided with a drop of relative humidity. Sun-photometer data suggest the presence of high altitude clouds with optical depth around 0.2, but characterized by large variability. Those clouds were not accounted for in the simulations.
The vertical profiles of brightness temperatures are similar for both flights (Figs. 8b and d). Again, the window channels show very weak variations, which is characteristic of clear sky conditions. On the contrary, FIR channels are characterized by 15 rapid variations near the surface and a larger lapse rate at higher altitude compared to the 11 April flight. These features are due to a sharper temperature inversion and a reduced transparency of the atmosphere (the column water vapor below 5.4 km are 10.3 mm and 10.5 mm, respectively). The difference between the conditions encountered on 11 and 20 April is further illustrated in Fig. 9. It shows the high spectral resolution brightness temperature simulated by MODTRAN at 6 km altitude  The simulated brightness temperatures in the atmospheric window are in good agreement with observations, but deviations exceeding measurement uncertainties are found in the FIR channels for the upper part of the profile. The largest discrepancies are obtained in the 30 − 50 µm band, with measurements being approximately 1.5 K warmer than the simulations. In fact, the air transmittance in this channel is so low that a significant part of the signal comes from the air contained in the 56 cm-long 10 chimney just below the instrument, rather than from the atmosphere below. This artifact was noticed by Mlynczak et al. (2016).
Using their correction (eq. 1), we find that air at -5 • C and 50 % relative humidity in the chimney can increase the apparent to explain such differences. Errors in water vapor measurements are also unlikely, because independent measurements taken by distinct instruments aboard Polar 6 show differences less than 20%, while only an increase larger than 50% could explain the observed differences. In addition, water vapor measurements along track did not show significant variability, so that spatial 10 variability of water vapor can be ruled out. Only the incursion of a wet air mass below the aircraft before the end of the ascent could explain such a discrepancy between observations and simulations. In such case the water vapor profile used in the simulation would not correspond to the actual profile at the time of the measurement, but this is unlikely given that it was observed on two different flights. Adding an optically thin cloud between 6 and 9 km altitude did not improve the simulations either. Given the verified accuracy of the FIRR, we hypothesize that the differences are the consequence of the observed haze similar to water vapor in the FIR. This question is left to future work, where hyperspectral measurements would certainly help investigating the detailed response. It should nevertheless be borne in mind that in these particular cases the greenhouse effect is underestimated in MODTRAN simulations, which can lead to significant deviations on the atmospheric and surface energy budgets.

Cloudy cases 5
Flights performed on 7 and 13 April are used to assess the radiative impact of optically thin ice clouds in the FIR. They also highlight the difficulty to compare in situ observations to radiative transfer simulations due to high variability of the cloud microphysics.

7 April
During this flight west of Alert, singular atmospheric conditions were encountered. Near the surface, a saturated layer was 10 found up to 1.1 km where a cloud was present, as detected by the Nevzorov and 2D-C probes. Another cloud was found above 4 km, that extended up to the maximum flying altitude of 6 km. In between, the atmosphere was very dry. The temperature profile had a complex signature near the surface, where a double temperature inversion was observed (Fig, 11), probably due to radiative cooling at top of the near-surface cloud. Observed FIRR brightness temperatures are consistent with the atmospheric profile. In the clear sky region, the profiles are similar to that of 11 April. In clouds, brightness temperature varies more rapidly 15 with altitude, as a consequence of increased absorption and scattering in all channels. Consequently, all brightness temperatures samples at 5.7 km are contained in a narrow 1.5 K range.
Since CALIPSO does not cover such high latitudes, we do not have supplementary information regarding the clouds properties. The profile of relative humidity suggests that the cloud was initiated above 5 km in saturated air with respect to ice, and below ice particles were precipitating, without saturating the air. For the MODTRAN simulations, the particle effective 20 diameter was set to 75 µm, with relatively large particles consistently seen by the 2D-C probe, but missed by the FSSP-300.
We then tuned the optical depth to 0.5 for the near-surface cloud layer and 1.0 for the upper layer cloud. This set of cloud properties produces brightness temperatures profiles in agreement with the measurements. The brightness temperature difference between 7.9−9.5 µm and 10−12 µm channels is larger in the model than in the observations yet, which suggests an imperfect definition of aerosol and haze profiles.

13 April
The best case of optically thin ice cloud was observed during 13 April flight. A vertical profile was taken during the descent between 18:15 and 19:12 UTC. The temperature profile was fairly typical of Arctic winter conditions, with an inversion at 1.3 km and surface temperature around -25 • C (Fig. 12a). A tenuous cloud layer was found below 1 km and a much thicker cloud was observed between 2.2 and 5 km according to the combination of 2D-C and FSSP-300 probes. These two instruments, along 30 with the relative humidity profile, suggest that ice particles formed above 3 km but large precipitating crystals were observed 7.9-9.5µm 10-12µm 12-14µm 17-18.5µm 18.5-20.5µm 20.5-22.5µm 22.5-27.5µm Measurements Simulations Figure 11. Vertical profiles of (a) temperature and relative humidity, and (b) measured and simulated FIRR brightness temperatures for 7 April flight. Shaded areas in panel (a) indicate the presence of clouds. The optical thickness and particle effective diameter used for the simulations are also indicated. The dashed lines correspond to the ERA-Interim profiles used for the simulations above maximum flying altitude.
down to 2.2 km. This cloud is similar to a TIC-2B type from the classification of Grenier et al. (2009). The FIRR brightness temperatures are characterized by high vertical variability, especially above 3 km (Fig.12b). This variability is identical for all bands, suggesting that it is due to actual scene variations. The excellent match between KT19 measurements and the 10 − 12 µm channel confirms that observed variations are not instrumental artifacts (Fig. 12c). Instead, they are attributed to cloud horizontal variability. This hypothesis is supported by the sun-photometer data that show highly varying optical depth upper cloud optical depth τ was varied from 0.5 to 5 in the calculations. Figure 12c shows that the range 0.5−5 reproduces quite well the observed variability of brightness temperature. We infer that at small scale cloud variability is extremely high, which is unexpected from satellite data on the large scale for this type of cloud (Grenier et al., 2009). To further investigate the spatial variability, MODIS cloud products at 18:09 UTC were analyzed. In particular, the cloud optical depth and cloud top altitude, shown in Fig. 13, are very instructive. At the scale of Polar 6 spiral, the cloud optical depth is indeed highly variable, ranging 5 from nearly clear sky to values exceeding 5. The cloud top altitude also shows that the probed cloud with top at 5 km was very localized in the most SE section of the spiral. Interestingly, these spatial features are consistent with FIRR observations. In fact, the difference between the temperature measured by the 10 − 12 µm channel and the simulation with τ = 2 (indicated by the color of the trajectory in Figs. 13a and b) is minimum near the area corresponding to the high altitude cloud, which suggests that the cloud there has an optical depth larger than 2. The difference is larger elsewhere, meaning that FIRR senses warmer 10 temperatures corresponding to either a thinner or lower cloud. The variations of the brightness temperature difference are more evident in Fig. 13c, that shows the time series of the difference along with the MODIS estimates of cloud characteristics.
Observed FIRR spatial variability is thus consistent with the presence of a cloud of optical depth around 4 in the SE bound of the trajectory that extends up to 5 km. Elsewhere on the trajectory the atmosphere ranges from clear to low-altitude clouds. The latter also seem to be variable, resulting in slight variations of brightness temperature in the window channels near the surface.
This case illustrates the complexity of atmospheric radiative transfer in heterogeneous conditions. It also shows that the FIRR is responding consistently with variations in clouds conditions from a nadir view similar to a satellite view.
The five case studies investigated in the previous section provided a valuable insight on FIRR performances from an airborne nadir configuration, and on the FIR characteristics of the Arctic atmosphere in clear and cloudy conditions. To further explore the dependence of FIRR measurements on atmospheric profiles, a series of radiative transfer simulations are performed. The results are then discussed in the framework of TICFIRE, with the intent to improve the data quality in future similar airborne 5 campaigns.

Sensitivity to temperature, humidity and cloud properties
In order to extend the interpretation of the data acquired during the NETCARE campaign, the Jacobians of the top of atmosphere (TOA) brightness temperature with respect to temperature and humidity were computed for 11 April simulations (Fig. 14). The Jacobian at a given atmospheric level is the difference in simulated TOA brightness temperature resulting from 10 an increase of 1 K (1% specific humidity) of the temperature (relative humidity) at this level. The temperature Jacobians show that the 30 − 50 µm channel is mostly sensitive to atmospheric layers below 500 hPa (above ∼ 5 km), which explains why this channel was not very useful at lower altitude during the campaign. The shorter FIR wavelengths are sensitive to lower layers of the atmosphere, and window channels are almost insensitive to the atmosphere temperature. These Jacobians also suggest that the 3 channels between 17 and 20 µm are very similar, making them somewhat redundant in such atmospheric conditions.

15
Comparing the absolute values of the Jacobians to the FIRR resolution gives a lower estimate of the vertical resolution the FIRR could reach for profiles retrieval applications. Given the radiometric resolution of the FIRR is about 0.2 K, temperature variations of 0.2 K are detectable with a vertical resolution of 100 to 200 hPa in FIR bands. Regarding the FIRR sensitivity to variations in relative humidity, Fig. 14b shows that the 30 − 50 µm band is the most sensitive, as expected due to the water vapor absorption spectrum. Humidity variations of 5 % for a 100 hPa thick layer above 600 hPa should produce a detectable 20 signal for all FIR bands, highlighting the potential of the FIRR for probing humidity profiles in such cold and dry conditions.
Note that the Jacobians are positive around the temperature inversion, which is a feature typical of polar conditions. Negative values are consistent with the fact that increasing water vapor increases the greenhouse effect due to the atmosphere and hence decreases radiation at TOA.
To complement this sensitivity analysis, an ice cloud was inserted between 2 and 6 km in the same atmosphere, and the 25 relative humidity with respect to ice correspondingly set to 100 %. Starting from a reference cloud, its optical depth and particle effective diameter were varied. Figure 15 shows that TOA FIR brightness temperatures are very sensitive to cloud optical depth, with variations up to 5 K between clear sky conditions and τ = 5. The FIRR resolution approximately converts into a 0.2 resolution in terms of optical depth. The same exercise with varying optical depth shows that for small particles FIR channels are very sensitive to particle size. However, the sensitivity quickly decreases for largers sizes, which is consistent with 30 the findings of Yang (2003) and Baran (2007), who suggested a sensitivity up to 100 µm effective dimensions. This sensitivity is directly related to the crystal shape and size distribution assumed for this study, which correspond to cirrus clouds. Although the results above are qualitatively robust, using another ice cloud parameterization could have resulted in different values (e.g. 7.9-9.5µm 10-12µm 12-14µm 17-18.5µm 18.5-20.5µm 17.25-19.75µm 20.5-22.5µm 22.5-27.5µm 30-50µm Figure 14. (a) Temperature and (b) humidity Jacobians for the TOA brightness temperature for 11 April atmospheric profile. For humidity, variations are in % of the specific humidity. Baran, 2007). In particular, Arctic clouds characterized by rapid crystal growth in high supersaturation conditions may actually have shallower particle size distributions (Jouan et al., 2012) and exhibit more sensitivity to particle size.

Recommendations for future operation
The preceding results are now discussed in the framework of planning the TICFIRE satellite mission and in view of future airborne campaigns with the FIRR or similar instruments. First of all, one advantage of using uncooled microbolometers is 5 the possibility to have an imager, as will be the case for TICFIRE. In this study, the FIRR was not used as an imager, though, because it has a much narrower field of view than TICFIRE satellite configuration. However it is worth exploring how the accuracy of the measurements would decay if spatial averaging was skipped. To this end, the spectral brightness temperature shown in Fig. 7 is computed again from FIRR measurements, except that spatial averaging is made on 1 (no averaging), 4, 9 or 193 pixels. Nominal data processing is optimized for 193 pixels and could not be applied to a single pixel (Libois et al., 2016), 10 so that the procedure was slightly changed to ensure that the same calibration is applied independently of the number of pixels averaged. The results are shown in Fig. 16. As expected, spatial averaging improves the repeatability of the measurement, but averaging over 9 pixels already provides a resolution close to 193 pixels. The absolute values are very consistent, with differences less than 0.5 K if more than 1 pixel are used. The remaining differences can be attributed to instrument errors, but scene spatial heterogeneities can not be ruled out. This suggests that the present study is relevant to verify the performances of the future TICFIRE satellite instrument, whose precision could be increased through spatial averaging over neighbour pixels.
It is worth pointing out that the NETCARE campaign was not dedicated solely to radiation measurements. Probing ice clouds was one of the objectives, but not the only one. In addition, few clouds were encountered during the campaign and 5 days with too many clouds prevented aircraft operations for safety reasons. Overall the dataset is still modest and further campaigns in the Arctic winter remain necessary, in particular to complete a radiative closure in cloudy conditions, which was not possible here due to lack of quantitative information about clouds properties. Such campaigns should be dedicated to the radiative properties of ice clouds in order to maximize the scientific success of this research topic (e.g. CIRCCREX, Fox, 2015). During the NETCARE campaign, the FIRR was supposed to have a zenith view to allow net fluxes computation and associated cooling rates, but shortly before the campaign started this configuration proved to be incompatible in terms of safety. In the future, combining nadir and zenith views as in Mlynczak et al. (2011) Figure 16. FIRR spectral brightness temperatures at 5.56 km as in Fig. 7, except that measurements were averaged over a varying number of pixels, from 1 to 193. Error bars indicate measurement variability along 4 consecutive measurements. For each spectral band the 4 corresponding error bars are slightly displaced horizontally for sake of clarity.
instrument radiometric resolution is essential to further constrain radiative transfer simulations and cloud properties retrievals.
This can be achieved by improving the environmental conditions of the FIRR within the aircraft, paying more attention to temperature stability. Adding an insulating window to prevent air circulation around the instrument or increasing the pressure inside the instrument to ensure constant outflow from the aircraft would minimize temperature variations. Note that these recommendations are linked to the fact that Polar 6 cabin is unpressurized and other constraints should be thought of in the 5 case of a pressurized aircraft. Complementary zenith and nadir observations would also be extremely valuable in order to compute cooling rates and sample the whole atmospheric profile.
At the instrument level, the FIRR is the first prototype and improvements are expected from technological developments of uncooled microbolometers, but optimization in the analogical-numerical converter and absence of the detector window in space could already increase the current resolution by a factor of 3 to 5. Likewise, increasing acquisition rate by using a faster 10 filter wheel and scene selection motor would reduce the acquisition time of a sequence by one order of magnitude, thus limiting temperature variations in between calibrations. It would also ensure that measurements in all channels are taken on the same target, which was not always the case during the campaign above leads or through highly heterogeneous ice clouds. Such technical developments are already considered and will be mandatory for the satellite version of the instrument which requires acquisition times around 1 s for a complete scene sequence.

Conclusions
The first airborne campaign of the FIRR took place in the Arctic in the framework of the NETCARE aircraft campaign. It was a great opportunity to study the radiative properties of the early spring Arctic atmosphere, and highlighted the importance of water vapor and ice clouds in this remote environment. Vertical profiles of brightness temperature acquired in clear sky and cloudy conditions provided a strong observational constraint on the radiative properties. At the same time, they increased the limited amount of observations available in the far-infrared, especially in such remote regions. These observations also provided valuable knowledge about the FIRR instrument, which can be used to improve operation and development in view of the TICFIRE satellite mission. This campaign showed that the current state-of-the-art radiative transfer models are well suited 5 for the Arctic and confirm that instrument resolution is better than the uncertainties inherent to the radiative transfer formulation and input observations. They also show that aerosols can significantly impact the radiative budget of the atmosphere, thus implying that a detailed characterization of the aerosols and haze is necessary to refine radiative closure experiments.
Although the FIRR behaved very well during the campaign with respect to its nominal performances, the latter could be improved for accurate retrievals of atmospheric and cloud characteristics. The campaign proved that ice clouds in the Arctic are 10 hard to probe, as much for reasons of safety as for their complexity and their high heterogeneity. As a consequence, measured ice clouds spectral signature could not be compared to simulations with sufficiently well-constrained cloud properties. Such airborne campaigns should be replicated to improve our understanding of ice cloud formation and radiative properties in polar regions. Accordingly, they should be dedicated to radiation and combine cloud microphysical observations with various radiation sensors. Such studies are necessary to continue improving our knowledge of ice cloud formation and its parameterization for their technical help during airborne operation. We thank Alexei Korolev (ECCC) for providing Nevzorov and 2D-C probes data and Gerit Birnbaum (AWI) for helping with the processing of AWI radiation sensors. Norm O'Neill (Université de Sherbrooke) accommodated the involvement of L. Ivanescu to the campaign. Canadian Network for the Detection of Atmospheric Change (CANDAC), to which L. Ivanescu is affiliated, provided logistic support during the stay at Eureka Weather Station. The Institut National d'Optique (INO) provided technical support for the FIRR before and during the campaign. We are grateful to Marcus Dejmek and Daniel Gratton (CSA) for providing logistic 5 resources for the FIRR. We are indebted to Mike Harwood (ECCC) for his direction of the instrument integration and support in the field.