The B895 flight took place on 13 March 2015 as part of the Cirrus Coupled Cloud–Radiation Experiment (CIRCCREX) campaign, which had the overarching goal of improving our understanding of cirrus cloud microphysics . The main objective of the flight was to simultaneously measure upwelling radiances in both the mid- and far-infrared using the Tropospheric Airborne Fourier Transform Spectrometer (TAFTS) and the Airborne Research Interferometer Evaluation System (ARIES) above a well-characterised cirrus cloud. The flight was performed using the Facility for Airborne Atmospheric Measurements (FAAM) British Aerospace (BAe) 146 aircraft and was based out of Prestwick, Scotland.

Figure  shows the flight track overlaid on top of the L2 COT from the MODIS (Moderate Resolution Imaging Spectroradiometer) Terra overpass at 10:40 UTC. The FAAM aircraft performed three straight and level runs (SLRs) at an altitude of 9.4 km, overflying a decaying band of cirrus cloud associated with an occluded front. We focus on a radiance observation taken at 09:48:39 UTC during the first SLR (09:33 to 09:58 UTC), where the plane was flying in a north-westerly direction. At this time, the atmosphere above the aircraft was cloud-free, limiting the amount of downwelling radiation available for backscatter into the radiometric instrumentation from the underlying cirrus.

The aircraft was also equipped with the Airborne Vertical Atmospheric Profiling System, which released two Vaisala RD94 dropsondes at 09:47:48 and 09:50:42 UTC during the first SLR. Given that these dropsondes use the same humidity sensor as the Vaisala RS92 radiosonde, we estimate the sensor calibration uncertainty was ± 5 % of the measured relative humidity with an absolute offset of ± 0.5 %. The production variability uncertainty was ± 1.5 % above 10 % or ± 3 % for relative humidity below 10 % . However, found discrepancies between simulated and observed microwave brightness temperatures in clear-sky conditions for other flights around the same time as B895 when using the measured dropsonde profiles, and this was attributed to a dry bias in the dropsondes. No equivalent information is available for the temperature sensor, but there is a manufacturer-quoted repeatability of 0.2 K

After completing the other two SLRs, the plane descended through the cloud and measured the microphysics using on-board instrumentation between 10:32 and 11:32 UTC. During these manoeuvres the cirrus remained relatively stationary, with satellite imagery indicating that it had dispersed by 13:00 UTC .

There were a number of instruments on board the FAAM aircraft used to characterise the cirrus cloud that are used to assess the retrievals performed here. These include an elastic backscatter lidar (used to derive a reference for the COT and CTH), a 2-DS probe (used to derive a reference for the CER), a Three-View Cloud Particle Imager (3V-CPI) (used to derive the ice crystal habits in ), a CIP 100, and a holographic cloud probe (HALOHolo) (used to measure the particle sizes). Further details about each of the instruments and their measurements are provided below.

A Leosphere ALS450 355 nm elastic backscatter lidar was installed on the aircraft . This provided information on the cloud vertical extent, ice volume extinction profiles at 355 nm from the range-corrected backscatter profiles following the two-stage process of , and a vertical profile of the particle extinction coefficient using the method described in . Raw lidar profiles were recorded with a vertical resolution of 1.5 m and an integration time of 2 s. The range-corrected backscatter profile measured by the lidar at 09:40:35 UTC (4 s before the radiance observation) is shown in Fig. a alongside the lidar-derived extinction profile (Fig. b) and the temperature and water vapour profiles measured by dropsonde 1 (Fig. c and d). This lidar-derived extinction coefficient profile has been previously used to derive the COT; however, due to poor instrument alignment, there were issues with the depolarisation calibration, which limited the quality of the estimated COT .

To mitigate these issues, we employ a simplified retrieval approach to derive the COT rather than the full vertical extinction profile. This approach considers the power of the attenuated Rayleigh backscatter range-corrected signals below 6 km in cirrus-free (Pref) and cirrus (Patt) conditions. It is assumed that the molecular extinction is constant along the short flight path, and so the signal variation can be attributed to the ice crystals in the cirrus cloud. The presence of cirrus reduces the attenuated Patt relative to Pref through two-way transmission and simplifies the COT retrievals to 1COT355nm=-12lnPattPref, where the factor of 2 accounts for the two-way transmission through the cirrus layer. In this work, Pref is calculated from the average of 1 min of backscatter signals between 4 and 6 km at 09:33 UTC where the atmosphere was cloud-free below the aircraft. Patt represents the backscatter signals from the remainder of the SLR that have then been similarly vertically averaged between 4 and 6 km for each of the 2 s measurement intervals. This means we now state the average lidar-derived COT for the 1 min period surrounding the radiance observation as 0.8 ± 0.1, with the uncertainty representing the standard error in the mean COT within this time period.

The particle size distributions (PSDs) and ice crystal habits were measured using a series of probes that included a 2-DS, a 3V-CPI, a CIP 100, and a HALOHolo . The 2-DS and inlet edge of the 3V-CPI probes were fitted with antishatter tips to reduce a bias towards smaller crystals. presented an analysis of the in situ measurements taken during the B895 flight, with both PSD and habits divided into the different temperature regions shown in Fig. . Each set represents the average of a run that took between approximately 5 and 10 min through sections of the cloud. The ice crystals were found to range from approximately 10 µm (lower limit of the 2-DS probe) to 700 µm. The best-fit distribution for the PSDs was generally found to be a bimodal Gaussian and is used to estimate the CER in Sect. .

The ice crystal habits within the cloud were determined from CPI images of crystals larger than 50 µm using an automatic habit recognition algorithm outlined in . Particles were found to be a mix of predominantly aggregates (∼ 30 %) and droxtals (∼ 25 %), with some rosettes (∼ 15 %) and columns (∼ 10 %) throughout the cloud. This is more similar to the constituents of the GHM model than the SC model, and so we expect the and GHM model to be a better representation of the ice crystal habits within the cirrus cloud.

The RAL Infrared Microwave Sounding (IMS) retrieval scheme simultaneously retrieves vertical profiles of atmospheric temperature and gases, along with surface skin temperature, surface spectral emissivity, and cloud parameters . Recent work has extended it for use on the FSI for clear-sky retrievals , and it has also been used for retrievals of aerosols and in the presence of clouds .

IMS uses the optimal estimation method from to fit an observed spectrum (the measurement vector, y) by iteratively perturbing the retrieval targets (the state vector, x). Estimations of y are calculated from adjusted values of x using a forward model, F(x), which in this case is a radiative transfer model. Prior knowledge of the state is contained in the a priori state vector, xa, with covariance, Sa, representing the vertical variability and correlation of the profile. These are both used to constrain the retrieval. Similarly, the measurement covariance, Sy, represents the uncertainty in the measurement.

Iterations are based on the Levenburg–Marquardt (LM) method : 5xi+1=xi+KiTSy-1K+Sa-1+γi-1×KiTSy-1(y-F(xi))-Sa-1(xi-xa), where i is the iteration, γ is the LM parameter controlling the magnitude of the state vector perturbation and is initially set to 0.001, and Ki is a Jacobian matrix of partial derivatives of the forward model output to elements of the state vector.

The fit optimisation is based on minimising the cost, χ2, with the first and second terms on the right-hand side of Eq. () corresponding to the measurement and state cost, respectively: 6χ2=(y-F(x))TSy-1(y-F(x))+(x-xa)TSa-1(x-xa).

The Radiative Transfer for TOVS v12 (RTTOVv12) fast radiative transfer model is used as the forward model in the IMS scheme. It takes inputs of temperature, water vapour, ozone, carbon dioxide, cloud fraction, cloud effective diameter, and cloud ice water content (IWC) on 101 fixed pressure levels from the surface to the top of the atmosphere.

RTTOVv12 uses the strong Norton–Beer apodisation to simulate FSI spectra covering 5000 channels across the FSI spectral range. Cloudy radiances are calculated using the Chou approximation, where the optical depths associated with atmospheric emission and absorption (referred to here as clear-sky optical depths) are scaled to account for the cloud particle interactions . Both of these are parameterised into coefficients to increase operational speed .

To calculate clear-sky optical depths, we use specialised RTTOVv12 coefficients that have been developed using LBLRTMv12.11 to simulate upwelling FORUM-aircraft spectra at common flying altitudes . For the ice cloud interactions, the absorption coefficient, scattering coefficient, and backscattering parameter used for the Chou approximation have been calculated from the and GHM and SC bulk optical property models for ice water content (IWC) from 0.5E-5 to 0.1 gm-3 and CER from 5–60 µm. The use of the Chou approximation to simulate cloudy radiances significantly reduces the operational speed of the retrieval; however it leads to a positive bias within the far-infrared that increases for smaller crystals .

Within IMS, the ice cloud is modelled as a Gaussian with a standard deviation of 1 km. The visible COT is calculated from the IWC and CER within IMS using the following approximation from : 7COTvis=∫CBHCTH3IWC(z)〈Qe,vis〉4ρiceCERdz, where 〈Qe,vis〉 is the bulk extinction efficiency at visible wavelengths, ρice is the density of ice, and the integral is performed from the cloud base height (CBH) to the CTH. We assume 〈Qe,vis〉=2 as we can assume the average particle is much larger than the wavelength at visible wavelengths.

In this work, we perform simultaneous retrievals of COT, CER, CTH, temperature, and water vapour. The a priori and covariance for the cloud retrieval targets are 9 ± 3 km for CTH and 40 ± 10 µm for CER. COT is stored in log form to prevent it from reducing below zero, and it is assumed there is minimal cloud cover at first, resulting in an a priori of ln(0.01)±10.

We use hourly collocated ERA-5 reanalysis data as the a priori for temperature and water vapour, and the covariance is a two-dimensional matrix derived from the differences between the zonal mean of ERA-5 profiles for 3 d (17 April, 17 July, and 17 October 2013) . In all the retrievals performed in this work, the initial guess is equivalent to the a priori, and the vertical O3, CO2, CH4, and N2 profiles, as well as the surface skin temperature and surface emissivity, are fixed to the values outlined in Sect. .

The measurement covariance (Fig. ) is assumed to be fully correlated and is composed of the NEDT and ARA for a single TAFTS and six ARIES scans (Fig. ), combined with the apodisation uncertainty described in Sect.  (Fig. a).

The retrieval is performed using selected channels rather than for the entire spectral range to minimise operational time and avoid selecting channels with high measurement uncertainty. Figure  shows the 200 channels selected in to optimise retrievals of temperature and water vapour from FORUM-aircraft observations.

The absorption of water vapour dominates the FIR, and as a result, there is a significant overlap in this region between spectral sensitivity to tropospheric water vapour and ice cloud properties. To enable a clearer distinction between habits, 26 additional channels have been included that were not in the initial 200 but are sensitive to ice cloud properties for this specific observation (Fig. ). We anticipate that they are also sensitive to water vapour and temperature.

These additional “cloud-sensitive” channels were selected using a database of FORUM-aircraft spectra simulated using RTTOVv12 that vary in COT (0.1–2), CER (5–60 µm), and ice crystal habit (GHM and SC models) but all have the same temperature, gaseous, and surface profile as in the FORUM-aircraft simulation (Sect. ). The channels not used in the clear-sky retrieval were first filtered based on whether clear-sky residuals between LBLRTMv12.11 and RTTOVv12 were comparable to or exceeded the instrument uncertainty. This was to ensure that the uncertainty associated with the forward model could be directly attributed to its cloud assumptions. For each cloud property, these channels were then filtered based on two conditions. The first was that the largest changes in radiance were detectable above the measurement uncertainty. The second was that the radiance change was not perfectly correlated (i.e. a correlation coefficient of less than 1) to another channel. Channels with the largest change in radiance were selected first.

Figure  shows that 16 MIR and 10 FIR channels were additionally selected to improve the retrieval of ice cloud properties. To highlight the channels that are most sensitive to ice crystal habit, Fig.  shows the brightness temperature (BT) residuals within these channels between the GHM and SC models for five different CERs (25, 30, 35, 40, and 45 µm) and a fixed ice water path (IWP) of ∼ 28 gm-2. We see residuals between the models exceeding the measurement uncertainty between 400–450 cm-1 (up to 1.3 K) and in the 1004.5 cm-1 channel, and so we expect the habit to be most detectable within these channels.

In each set of retrievals, there are four different configurations used: with and without the FIR and assuming the SC or GHM bulk optical property model. These will be referred to using the names outlined in Table  for clarity. Unlike the other retrieval targets, the ice crystal habit is retrieved by examining the spectral fit in certain channels after the optimal estimation retrieval of the other targets has been performed. To do this, we focus on the BT residuals in individual channels and the measurement cost per channel (Jy), which enables comparisons between different configurations. A lower Jy for a given habit or habit mix is indicative that it is a better representation of the true crystal shapes within the cloud. If Jy<1, then the retrieval has fit the spectra on average within the uncertainty. For guidance, an increase of 20 % in Jy is indicative of a 10 % increase in the residual between the observed and fitted spectra, assuming the uncertainty is kept constant.

The uncertainty in the retrieved state vector is calculated from the square root diagonal of the covariance of the retrieved state, Sx, 8Sx=Sa-1+KTSy-1K-1, and is equivalent to 1 standard deviation. This will be referred to as the estimated standard deviation (ESD) in the retrieval.

The averaging kernel (AK) matrix, A, represents the vertical sensitivity of the retrieved state to the true state, x^, and is calculated using Eq. (). 9A=∂x^∂x=GK

In this work, we assume the dropsonde profiles are a fair approximation of the true state, and they are smoothed to the resolution of the retrieval using 10xAK=xa+A(x-xa), where xAK is the AK-treated true state.

The degrees of freedom for signal (DOFS) is calculated from the trace of A and is used to assess the amount of information in the measurement vector.

As described in Sect. , RTTOVv12 uses the Chou approximation to simulate cloudy radiances. This approximation introduces a positive bias into the spectrum within the FIR that is dependent on the cloud parameters. show that for an ice cloud with a COT of 1, CER between 30–40 µm, and CTH of 8 km in a mid-latitude atmosphere, this positive bias peaks at ∼ 1 K at ∼ 410 cm-1. A lower COT and larger CER reduces this bias. Within the MIR, there is a negative bias that is approximately half the FORUM-aircraft uncertainty.

The FORUM-aircraft simulation described in Sect.  was simulated using the LBLDIS radiative transfer model based on the profiles in Fig.  and assuming the GHM model. LBLDIS uses numerical methods to calculate the ice cloud interactions, and therefore by performing a retrieval on the FORUM-aircraft simulation, we can determine if the Chou approximation significantly hinders the retrieval capability as the inputs are known. The simulated spectrum has had Gaussian noise applied based on the instrument uncertainty (Fig. ) and includes the apodisation uncertainty associated with creating the FORUM-aircraft spectra.

Figure a shows the BT residuals associated with the retrieval from the FORUM-aircraft simulation with the four different retrieval configurations. When the correct (GHM) bulk optical property model and all 26 cloud-sensitive channels are used, the retrieved BT between 400–500 cm-1 is, on average, within 0.2 K of the input simulation. Furthermore, the retrieved values in Table  capture the input parameters within 2 ESD. This suggests that despite the Chou approximation's bias in the far-infrared, we can assume that the retrieval scheme can reasonably estimate true values.

All four retrieval configurations also overestimate the BT in the channel at ∼ 730 cm-1. This channel was selected due to its sensitivity to CER despite its increased uncertainty from the FORUM-aircraft apodisation (Fig. a). The weaker fitting in this channel has likely contributed towards the larger CER ESD in Table ; however, the input CER is still captured within 1 ESD for all the retrievals.

Given this, we can now examine the impact of including the FIR in retrievals from the simulation (Fig. a). Using the MIR alone suggests that the SC model is a better representation of the ice crystal habit mix, with a measurement cost 40 % smaller than with the GHM model. This is an incorrect result as the GHM model was used in the simulation. When we include the FIR, the residual associated with the GHM model is up to 0.6 K smaller than its counterpart between 400–450 cm-1, comparable to expected residuals for CER in this range (Fig. ). This is indicative of a better fit and correctly suggests that the GHM model is the better representation of the ice crystals within the simulation. However, this is not significantly reflected in the measurement cost, with only a slightly smaller Jy for the GHM model. This is likely due to the larger BT residual for the GHM than SC habit in the ∼ 730 cm-1 channel countering the reduced residual in the FIR.

Table  shows that including the FIR leads to a closer agreement between all of the retrieved values and the known truth, as well as a significant reduction in the uncertainty associated with the CTH. The MIR + FIR GHM retrieval does show a slight positive bias in the cloud retrieval targets that is not seen with the MIR + FIR SC retrieval as a result of reducing the spectral residual caused by the Chou approximation. However, it is important to recall that the retrieval is assessed using the measurement cost and spectral fit and that the MIR + FIR GHM retrieval still captures the true state within 2 ESD.

As in the previous section, we perform retrievals of COT, CER, CTH, and ice crystal habit with a fixed vertical profile, but now on the radiance observation. Given that the fixed vertical profile is now only an approximation of the true state, we will not explore all the retrieval targets in detail as these are discussed in the context of the full retrieval in Sect. . In this section, we only focus on the retrieved ice crystal habit and BT residuals to help separate the influences of the vertical profile and cloud approximations on the full retrieval in Sect. .

The in situ measurements of the ice crystal habits suggest that the GHM model is a better representation of the ice crystals within the cloud than the SC model (Sect. ). Figure b shows the BT residuals for retrievals from the FORUM-aircraft observation using only the cloud-sensitive channels, and we observe similar trends to the simulated case. The measurement cost per channel using the MIR alone is again slightly larger for the GHM model than the SC model, with retrieved spectra within 0.2 K of each other. When we introduce the FIR channels, Jy is reduced by 13 % using the GHM model. This reduction is driven by the fitting between 450–500 cm-1, where there are differences of up to 0.6 K between the MIR + FIR SC and GHM retrievals. In common with the simulated results from Sect. , this strongly implies that for this case study, observations from the FIR are needed to correctly diagnose the ice crystal habit.

However, the BT residuals within the FIR generally exceed the measurement uncertainty, particularly between 450–500 cm-1, while there is little change in the MIR residuals with and without the FIR. This was not seen in the retrieval from the FORUM-aircraft simulation, which used the same bulk optical property model as the retrieval scheme. This reaffirms the findings in , which highlighted the inability of bulk optical property models to simultaneously match observed radiance signals in the MIR and FIR. Despite this, the bulk optical property models do not prevent the inclusion of the FIR from distinguishing between ice crystal habits.

It is important to note that the temperature and water vapour profile are currently fixed to measurements from the dropsonde. To test that the fixed temperature and water vapour profile is not biasing the habit determination, we now also include the simultaneous retrievals of temperature, and water vapour from the radiance observation.

As discussed in Sect. , the in situ observations of the ice crystals suggest that the GHM model is a better representation of the ice crystals within the cloud. Figure  shows that using the MIR alone does not allow a clear distinction between the GHM and SC models when we only consider the spectral fit. In the cloud-sensitive channels, we see comparable values of Jy for the MIR GHM and MIR SC configurations, similar to the findings in Sect. . This is also true across all 116 MIR channels, which include greater spectral sensitivities to the temperature and water vapour vertical profile. This suggests that while the inclusion of temperature and water vapour in the retrieval does affect the BT residuals, it does not significantly impact our inability to distinguish between habits when we only use the MIR.

When we include the FIR, there is a clearer distinction between habit models than when we only use the MIR, with the Jy for the MIR + FIR GHM configuration 9 % smaller than its SC counterpart in the cloud-sensitive channels. This smaller Jy for the MIR + FIR GHM configuration suggests that the GHM model enables a better spectral fit and so is the more likely representation of the ice crystals within the cloud. This is in line with the in situ measurements.

The addition of temperature and water vapour in the MIR + FIR retrieval has a more notable impact on the spectral distinction between ice crystal habits than when we only use the MIR channels. This is most clearly seen when we compare the BT residuals associated with the MIR + FIR SC and GHM configurations within the FIR (Fig. ). Below 410 cm-1, the BT residuals between the GHM and SC configurations are now comparable. Between 410 and 600 cm-1 the residuals differ by, on average, 0.1 K, which is 0.5 K smaller than when the vertical temperature and water vapour profiles were fixed to the dropsonde measurements (Fig. b). This similarity is reflected by the reduction in percentage difference between Jy in the cloud-sensitive channels for the MIR + FIR GHM and SC configurations by 6 % relative to the retrievals with the fixed vertical profile. Furthermore, when we consider Jy across the clear-sky and cloud-sensitive channels, the percentage difference between Jy for the MIR + FIR SC and GHM configurations is only 5 %. This effect is a result of the differences in the associated retrieved temperature and water vapour profiles and is explored further in Sect. .

The benefits of including the FIR in clear-sky retrievals of upper-tropospheric water vapour have been explored through observations in mid-latitude conditions . Our results indicate that including the FIR increases the DOFS for temperature and water vapour from 2.9 to 3.8 and from 1.9 to 3.7, respectively, due to the additional information contained within the FIR OLR spectrum. Both the MIR and MIR + FIR GHM retrievals generally capture the AK-treated dropsonde profile within 2 ESD (not shown). However, as previously discussed (Sect. ), including temperature and water vapour in the retrieval influences the determination of the ice crystal habit for the retrievals that include the FIR. This is in part due to the increased sensitivity of FIR radiances to the water vapour and temperature profile as reflected in the DOFS.

Figure  shows the retrieved temperature and water vapour profiles using all 226 channels in the MIR + FIR GHM and SC configurations. The temperature retrieval for both habit models is worse than the a priori with residuals of up to 1 K from the AK-treated dropsonde temperature profile (Fig. b). This is in part due to the increased uncertainty in the 15 µm CO2 wings following the FORUM-aircraft apodisation . Despite this, the AK-treated dropsonde temperature profile is within 1 ESD of both retrievals below 6 km and entirely within the GHM retrieval. For the MIR + FIR case, assuming the SC habit reduces the retrieved temperature between 2–8.5 km relative to the GHM case, with differences reaching up to 1.9 K. A reduction is also seen between the SC and GHM cases if only MIR channels are included, but in this case it is smaller, reaching a maximum of 0.4 K (not shown). This reduction in temperature lowers the simulated brightness temperature throughout the channels, particularly between 450–500 cm-1 to balance the changes in the cloud retrieval products that are discussed in Sect. .

Figure c shows that the habit does not affect the retrieved water profile outside of 1 ESD, with the median retrieval bias throughout the vertical only increasing by 2 % from the GHM to SC model. Both retrievals capture the AK-treated dropsonde profile below 4 and above 8 km but struggle at the base of the cloud. Neither retrieval captures the observed dry patch around the cloud base, as they remain close to the a priori.

Figure  shows the retrieved values for the COT and CER in comparison to the measurements of the cloud described in Sect. . The retrieved values are plotted at the midpoint of the retrieved cloud to allow a comparison to the individual CERs derived from the PSDs. While the lidar-derived COT was taken 4 s after the radiance observation, the in situ measurements of the cloud were taken between 10:32 and 11:32 UTC, and so we also include the L2 COT and CER from the MODIS Terra overpass at 10:40 UTC. The MODIS data have been averaged across the spatial range of the in situ measurements and radiance observation (58.7–59.75° N, 4.2–3.3° W).

There is some evidence of systematic behaviour between the COT retrievals shown in Fig. a. For example, the GHM retrievals (purple points) show slightly larger COTs than their corresponding SC retrievals. Similarly, the MIR retrievals (square points) show slightly larger COTs than the MIR + FIR retrievals. However, all four configurations retrieve the COT within the uncertainty in the lidar-derived COT of 0.8 ± 0.1 that was collocated with the radiance observation and are within 1 median absolute deviation of the median MODIS COT taken approximately an hour after (Fig. a). The relative insensitivity of the COT retrieval to the inclusion of the FIR is expected given the strong sensitivity to COT across most of the MIR channels used in the retrieval , and this is reflected in the minimal increase in the DOFS from 0.9 to 1.0 when the FIR is included.

The centre of the cloud measured by the lidar is at ∼ 7.5 km, and we see this captured within the range of all four retrieval configurations, with a consistent DOFS of 1 for CTH across all the retrievals. The inclusion of the FIR (triangle points) slightly shifts the cloud higher by ∼ 0.4 km, but the retrievals still remain within 1 ESD of their MIR counterparts (square points). In contrast, assuming the SC model rather than the GHM model can be seen to slightly lower the retrieved cloud by ∼ 0.4 km, and this is likely to counteract the colder vertical profile that is retrieved with this habit (Fig. ).

For both the retrieved CTH and COT, we see the ESD halve from 0.8 to 0.4 km and 0.06 to 0.03, respectively, with the inclusion of the FIR. This suggests a greater confidence in the state and agrees with previous simulation studies that suggest including the FIR could reduce uncertainties in cloud retrieval targets .

All four retrievals of the CER are between 28 and 38 µm. This is generally lower than the a priori and MODIS CER and tends towards the smaller in situ measurements. There is a notable difference in sensitivity to the inclusion of FIR information depending on the habit selected in the retrieval. For the GHM case the retrieved CER is unchanged at 31 ± 6 µm regardless of whether FIR channels are included, showing good agreement with the weighted mean of the in situ CER measurements (32 ± 14 µm). In contrast, if the SC habit is selected, the retrieved CER increases from 28 ± 4 to 38 ± 5 µm from the MIR only to the MIR + FIR case, with the MIR + FIR SC retrieval remaining closest to the a priori. There is also a slight increase in the CER DOFS from 0.7 to 1.0 when the FIR is included.

Overall, despite the limitations associated with the bulk optical property models and the use of a fast cloud scattering approximation, we are able to retrieve cloud properties that agree with the independent lidar and in situ observations. The addition of FIR information provides a greater constraint on the inferred COT and CTH (reduced retrieval uncertainty) and, importantly, allows an improved discrimination of crystal habit compared to retrievals using only the MIR part of the spectrum.