Predictions of future climate change and its regional and global impacts require that a better understanding of the radiative transfer interactions between clouds, water vapor and precipitation are incorporated into appropriate models. Recent CMIP5 model intercomparisons the Coupled Model Intercomparison Project as described in indicate large variability in ice cloud parameters (for example ice water content) amongst high-latitude models. Shortcomings in ice cloud parameterization impact their representation of radiative effects as well as water cycles and lead to uncertainties in quantifying cloud feedbacks in the context of climate change . High-altitude thin ice clouds consisting of pure ice crystals, which cover between 20 to 40 % of the Earth , can, for example, have opposing effects on the radiative properties of the Earth. A surface-cooling effect ensues when scattering by clouds reduces the solar radiation reaching the Earth's surface (i.e., albedo effect). By contrast, a reduction in the amount of IR energy escaping the Earth–atmosphere system occurs when the upwelling IR radiation emitted by the Earth's surface and lower atmosphere is absorbed by clouds and radiated back downward (i.e., greenhouse effect; ). The macrophysical and microphysical properties of thin ice clouds determine which process dominates and hence determine the net forcing of thin ice clouds on the climate system . The optical properties of thin ice clouds can be represented by extensive parameters such as the optical depth and ice water content as well as intensive parameters such as ice crystal size and shape. In an Arctic environment, the radiative effects of ice clouds are unique because their radiative forcing influence on the energy balance depends on seasonal polar day to polar night variation as well as large-scale processes like the Arctic Oscillation . The advent of active sensors on board satellites (for example CALIPSO/CloudSat) has enabled the application of considerably more resources for polar region ice cloud studies. This permits the evaluation of climate models and satellite cloud climatologies . Long-term ground-based observations, which are also essential for the validation of models and satellite climatology, are, however, limited in their Arctic coverage .

Thermal IR radiometry is a well-known technique for investigating the presence and the emissivity of clouds . Numerous researchers have exploited the thermal IR behavior of the absorption and scattering efficiencies of cloud particles as a means of retrieving CODs and particle effective sizes e.g.,. As cloud altitudes (temperatures) can lead to large uncertainties in this latter technique, proposed using lidar backscatter profiles along with IR radiometry to estimate cloud altitudes and accordingly improve the retrieval accuracy of cloud emissivity. This active/passive technique (called LIRAD for lidar/radiometer method by Platt) has evolved over the years with improvements such as the availability of high-resolution spectrometers . The LIRAD technique is based on spectral radiance/brightness temperature comparisons between measurements and radiative transfer calculations. It performs better in the presence of high thermal contrast and is thus well suited for cloud retrievals . In more recent applications, cloud optical depth, effective radius and ice fraction were retrieved from AERI Atmospheric Emitted Radiance Interferometer; spectral downwelling radiance observations in Antarctic, during the surface heat budget of the Arctic Ocean (SHEBA) campaign and the ARM (Atmospheric Radiation Measurement) Alaska North Slope site respectively. The method was also employed at Eureka (Nunavut, Canada) to retrieve cloud optical depth and cloud microphysical parameters from AERI spectra acquired between 2006 and 2009 . A proposed satellite-based instrument, with the goal of characterizing thin ice clouds in the Arctic using far and thermal infrared channels , was recently tested during an airborne campaign in the High Arctic .

In this paper, we examine how multiband thermal measurements of zenith sky radiance can be used to retrieve what are, as indicated in the early remote sensing literature see,for example,, the most critical extensive and intensive parameters influencing the radiative effects of ice clouds: cloud optical depth (COD) and effective particle diameter (Deff). We propose an application of this LIRAD technique with a relatively simple and inexpensive instrument (less than 10 % of the cost of an AERI) which is well suited to the Arctic environment . Inasmuch as ice particle size is difficult to retrieve from IR radiometry, the Deff component of our retrievals will be focused on a simple discrimination of large and small crystal sizes. This approach was motivated by previously published research that indicated such a discrimination would play a key role in characterizing an important aerosol–cloud interaction process in polar winter, namely precipitative cooling see, for example,. An important aspect in this paper is that our COD and Deff retrievals will be validated using independent lidar and radar retrievals.

Section 2 highlights the importance of studying ice clouds while Sect. 3 is devoted to the description of the study site and instrumentation. Section 4 examines the sensitivity of thermal IR radiometry to key ice cloud parameters. In Sect. 5 we describe and verify the proposed methodology for retrieving COD and Deff using thermal IR radiometry measurements. The results are presented and discussed in Sect. 6 for the 150 thin ice clouds cases we observed in the Arctic.

Water vapor and clouds are a significant climate modeling challenge since they represent major radiative forcing influences, while being the least understood components of the climate system . Much of the recent research has been focused on aerosol–cloud interactive processes involving aerosols acting as ice and water cloud nuclei as well as their subsequent effect on cloud microphysics, precipitation and radiation (see, for example, on recent ARM campaigns, and respectively, on the cloud remote sensing mandate of the A-Train and EarthCARE satellite missions and as part of the NETCARE project). In particular, understanding aerosols and their radiative effects, especially their indirect impacts as cloud condensation nuclei, is of critical importance for climate change models. The indirect effect of aerosols represents a cooling influence (amplitude is subject to large uncertainties) on the global radiative budget Intergovernmental Panel on Climate Change,. The estimated uncertainty of the indirect forcing component (between -0.1 and -1.3 W m-2) is associated with variations in cloud properties and cloud lifetime. In the Arctic, the nature of thin ice clouds can effectively induce an indirect cooling influence given the proper conditions. In terms of the purpose and motivation for this paper, we note that the presence of sulfuric-acid-bearing aerosols (viz., Arctic haze) can significantly increase the size of ice particles (relative to the size of ice particles formed from more pristine, low-acid aerosols or supercooled droplets). This process can cause enhanced precipitation and important cooling effects during the polar winter and could possibly lead to a dehydration greenhouse feedback (DGF) effect, as proposed by .

The small and large ice particles described above are often abbreviated as TIC1 and TIC2 (thin ice cloud, types 1 and 2). Thin ice cloud classification was carried out by using the active techniques of lidar and radar: CALIPSO and CloudSat data were employed to discriminate between TIC1 and TIC2 ice clouds using the CloudSat small-particle sensitivity minimum of approximately 30–40 µm. In this study, we seek to demonstrate that TIC1 and TIC2 discrimination can be determined using zenith-looking IR radiance measurements acquired at the Eureka observatory in the Canadian High Arctic. Figure 1 illustrates lidar and radar backscatter profiles acquired at Eureka for distinct TIC1 and TIC2 cases.

The left-hand lidar profile of Fig. 1 shows a TIC that is largely transparent to cloud radar while the right-hand lidar profile shows a thin ice cloud that is readily detected by the radar. The disparity in radar detectivity enables one to conclude that the former case corresponds to a small-particle TIC1 event while the latter case corresponds to a large-particle TIC2 event.

The presence or absence of thin ice clouds in the winter can lead to significant changes in surface cooling . Given the important radiative influence of ice crystal size and COD it is necessary that these parameters are well characterized in order to improve modeling of their radiative effects and thus their influence within a context of TIC1 and TIC2 clouds.

The effective diameter of atmospheric ice particles is defined by : Deff=3V2A, where A and V are respectively the total projected area and volume of all ice particles per unit surface area in a given atmospheric column . COD is given by COD=∫πQextD24a(D)dD, where Qext is the extinction efficiency (extinction cross section per unit projected-particle area; ), D is the particle diameter and a(D) is the ice particle number density per unit increment in diameter. We note that, while the lidar-derived CODs employed in this article are at 0.532 µm, the IR CODs from our retrieval method were referenced, for convenience, to 0.55 µm (we assume that COD differences between 0.532 and 0.55 µm are negligible within the context of other uncertainties encountered in this study).

The observation site was the Polar Environment Atmospheric Research Laboratory (PEARL) in Eureka, Nunavut (80∘ N, 86∘ W) which is one of the high-latitude stations of the Network for the Detection of Atmospheric Composition Change (NDACC, http://www.ndsc.ncep.noaa.gov/sites/stat_reps/eureka/). This high-Arctic site is located in the northernmost part of the Canadian Arctic Archipelago. It was chosen because of our interest in Arctic ice clouds and to exploit the diverse and complementary inventory of atmospheric instruments listed in Table 1 (i.e., lidar, radar, IR spectrometer and radiosondes), as well as the infrastructure and logistics support for field campaigns.

Detailed descriptions of the AHSRL and MMCR data processing and interpretative techniques can be found in and . A summary of instrument specifications is given in . present a discussion of the AERI performance which is applicable to the present paper. The AERI instrument is known to have a very small warm bias for low-radiance measurements, typically for clear-sky events, of the order of 1 % of the ambient radiance . As the focus of our work is on clouds with COD greater than 0.1, this warm bias in the AERI has only a slight to negligible impact for retrievals involving very thin clouds . The AERI data used in this work have been postprocessed to reduce the uncorrelated random error in the data using principal component analysis .

In this paper, we focus on the potential of using data from a ground-based multiband thermal radiometer, the CIMEL CE-312 developed by CIMEL Inc seefor descriptions of a similar instrument. The six channels of this radiometer correspond to (full width at half maximum) limits of 8.2–8.6, 8.5–8.9, 8.9–9.3, 10.2–10.9, 10.9–11.7 and 11.8–13.2 µm and filter response peak values at 8.4, 8.7, 9.2, 10.7, 11.3 and 12.7 µm. The multiband radiometer is also a robust instrument that, unlike the AERI, does not require a thermally controlled environment. In actual fact, however, we had to simulate the response of this radiometer by convolving the spectral transmittance of each filter with the spectra of the Eureka Polar AERI (P-AERI) instrument (provided by Von Walden at the U. of Idaho and NOAA). The reason for this was that the CIMEL radiometer that we hoped to use was not ready for deployment when we performed the field campaigns.

The spectral sensitivity of longwave (thermal) radiation to the microphysical properties of ice clouds has been investigated for satellite data , airborne data and ground-based sensors e.g.and articles cited above. Previous studies have demonstrated that thermal IR radiometry is relevant in terms of permitting the retrieval of both COD and, to a degree, Deff. The retrieval of the latter parameter permits, in turn, a discrimination of TIC1 and TIC2 clouds. The dependence of thermal IR radiometry on ice particle size is represented by Eq. (2). The extinction efficiency, a measure of particle attenuation (absorption and scattering), depends on particle size, composition and shape as well as wavelength . It is common, in the case of zenith-looking thermal IR (8–14 µm) radiometry of ice clouds, to neglect the scattering portion of Qext (where Qext=Qabs+Qsca), especially for large particles . The result in the presence of a medium such as cloud is extremely simplified radiative transfer that is characterized by a strong forward-scattering phase function: in the limit of a delta-function phase function, all forward-scattered radiation in any given direction of incidence is returned to the incident beam and the only radiance loss is due to absorption see, for example, the delta-function irradiance solution of. and later authors such as indicated that only a small fraction of the zenith-looking downwelling radiation emitted by a cloud was due to scattering (in spite of the fact that Qsca≈Qabs).

We accordingly chose to plot absorption efficiency spectra in Fig. 2 in order to illustrate the spectral sensitivity of this key radiative transfer parameter. The absorption efficiency of TIC1 particles (Deff from 10 to 30 µm) and TIC2 particles (Deff from 35 to 120 µm) for “severely roughened solid column”-type crystals (Fig. 2a and b) were obtained from calculations reported by and . These spectra were then replaced by their mean and standard deviation over the two (TIC1 and TIC2) Deff regimes in order to better appreciate the band to band separability of the TIC1 and TIC2 size classes (Fig. 2c). Prior to computing those means and standard deviations of Fig. 2c, we integrated the individual spectra of Fig. 2a and b across the pass bands of the six channels employed in this study (triangles in Fig. 2c). The Deff ranges employed to define TIC1/TIC2 particles, for the averaging carried out in the creation of Fig. 2c, were, as in , roughly based on their nondetectability to detectability threshold in radar backscatter returns.

This coarse spectral representation obtained for the six band averages and standard deviations enables one to better appreciate the more robust nuances between the two families of spectral curves (especially for the 8.4, 8.7, 9.2 and 12.7 µm channels) and better understand the key discriminatory elements of the classification into TIC1 and TIC2 clouds. It is clear from Fig. 2c that the first four bands offer the greatest potential for discriminating particle size. Figure 2a and b, however, indicate a decreasing sensitivity to increasing particle size as one approaches Deff values in the tens of micrometers.

We simulated the influence of COD and Deff variations, on brightness temperature (Tb) variations using the MODTRAN 4 radiative transfer model . Figure 3 shows simulated Tb variations for the six radiometer channels as a function of COD for fixed Deff and as a function of Deff for fixed COD. The fixed values of Deff and COD (and other independent parameters of the MODTRAN 4 runs) correspond to a reference case with parameters that are defined in Table 2. We chose the input parameters of the reference case as the set of mean parameters obtained by averaging over the parameters of the 150 cloud cases that we employed to provide an empirical validation of our retrieval (see Sect. 6.2 and Table 2 for more details). The curves of Fig. 3 represent an illustrative subset of our inversion lookup table (LUT) that we employed as a means of retrieving COD and Deff from measured values of Tb. The Deff column is biased by the fact that the lidar–radar retrieval is insensitive to TIC1 particles: the Deff reference value was accordingly biased downward to roughly overcome this insensitivity. The value of 50 µm was also the value employed by .

Figure 3a indicates that, at a fixed Deff value of 50 µm, there is a strong and monotonic variation in Tb as a function of COD for all channels. At COD magnitudes greater than 2–3, the Tb values for all channels converge towards an asymptotic ceiling that is the brightness temperature of an opaque representation of the cloud. This clearly shows that the sensitivity of the method decreases progressively as the COD increases beyond 3.

Differences in Tb behavior over a range of Deff values and a fixed COD of 0.5 can be observed in Fig. 3b. For the channels of nominal wavelength less than 10.7 µm, Tb varies monotonically with Deff up to approximately 30 µm, after which the response plateaus to variations of ≈ 1K or less. For the 11.3 and 12.7 µm channels, the responses are nonmonotonic (or considerably less monotonic) for the smaller values of Deff and smoothly decrease with increasing Deff beyond a peak in the 10–20 µm range. This decrease is associated with the relatively large spectral changes seen in the refractive index of ice particles at these larger wavelengths see, for example,.

As discussed above, IR radiance measurements are sensitive to a variety of cloud parameters as well as to the cloud environment. The simulation results in Fig. 4 detail the band-dependent effects of six different parameters by comparing changes in Tb induced by each parameter individually. These were obtained for 1000 appropriately normalized samples of a random number generator with a normal probability distribution, the mean and standard deviation of which was controlled by the six parameter values of Table 2 (COD, Deff, cloud base height, cloud thickness, column-integrated water vapor of the atmosphere, WVC, and particle shape). The particle shape parameter is based on three particle characterizations as defined by : severely roughened solid columns, a general habit mixture involving a set of nine habits and severely roughened aggregate of solid columns. The standard deviations in Tb that result from the variation of the six parameters are computed relative to the reference case defined above (Table 2). We note that there was little sensitivity to the choice of a 50 µm effective diameter for the reference case: changing this typical TIC2 value to a value more representative of TIC1 particles produced differences of less than 1 K in Fig. 4 .

Figure 4 shows that the chosen COD variation had the strongest Tb influence of all of the parameters, especially for the bands at 10.7 and 11.3 µm (as one could infer by referring to Fig. 3a). Changes in Deff (in red) lead to a standard deviation around 2 K for the four first bands as can be qualitatively appreciated by referring to Fig. 3b. Changes in the altitude of the cloud induce a standard deviation up to 5 K. Indeed, because measurements of thermal IR radiometry are sensitive to temperature, a change in altitude causes a Tb difference that is sensitive to the range of temperatures within which the cloud is located. Cloud thickness and WVC are marginally important parameters in terms of the magnitude of the changes induced in Tb. If the altitude and the thickness of the clouds are known from vertical lidar (and radar) profiles, then the magnitude of the altitude and cloud thickness uncertainties of Fig. 4 will fall to levels commensurate with the standard deviations ascribed to the uncertainty in the ice particle shape (≈ 1 K as per the dark blue curve). This latter uncertainty (determined from the habit parameterizations listed in Table 2) is largely inflexible inasmuch as our ability to distinguish ice particle shape from lidar depolarization data is extremely limited. Water vapor content (WVC) in the atmosphere, which remains relatively low during the polar winter at Eureka, has a weak absorption influence on the radiance measurements acquired in the CE-312 bands. Its associated uncertainty is commensurate with the uncertainties due to particle shape and cloud thickness: however integrated WVC is estimated from radiosonde profiles at Eureka and thus its uncertainty can be reduced to levels significantly below the particle shape and cloud thickness uncertainties. These reductions in the uncertainty of nominally known input parameters will be such that the variability of the parameters to be inverted (COD and Deff) is significantly larger than the uncertainty of the known input parameters (for all bands in the case of the COD and at least in the case of the first four bands for Deff).

Our LIRAD objective was performed using LUTs and MODTRAN 4 radiative transfer simulations to parameterize the behavior of the downwelling zenith sky radiance as a function of key input parameters, including COD and Deff. The methodology is represented in the flow chart of Fig. 5. The core of this method, inspired by ground-based retrievals e.g., is to compare thermal IR radiance measurements with LUTs derived from MODTRAN 4 simulations. This inverse problem is solved using the optimal estimation method (OEM; ). The method seeks the state of maximal probability, conditional on the value of the measurements, associated errors and a priori knowledge. This OEM is an efficient inversion method that has already been employed for ice cloud retrievals see, for example,.

The steps of the retrieval method are as follows:

First, knowledge of the cloud environment at the time of a given radiometer measurement is required. Specific input auxiliary data includes pressure, temperature and water vapor profiles from radiosonde data and the effective cloud layer height from lidar backscatter data. The radiosonde parameters are interpolated to the radiometer times while the time of the selected lidar profile is the nearest to the radiometer time. To avoid the issue of interpolating radiosondes over extensively long periods of time, the cases were selected to be as close as possible to radiosonde launch times. Radiosonde humidity sensors are known to be subject to dry bias, especially in dry conditions and could yield relative humidity underestimates of 10 % . The six channels are, however, far less sensitive to WVC than to COD (see Fig. 4) and therefore the bias is expected to be lower in cloudy conditions. Cloud heights are estimated for sustained cloud features where clouds are defined by lidar backscatter coefficients greater than 1.10-6 m-1 sr-1 and a lidar depolarization ratio greater than 20 % (thresholds were inspired by but adapted to a different vertical resolution of our lidar). Upper and lower cloud boundaries are then obtained where four continuous vertical samples of the lidar profile (4×30=120 m for an AHSRL resolution of 30 m) comply with that requirement (preceded by a series of lower, noncloud samples).

We developed a simple inversion technique to retrieve the optical depth and effective particulate diameter of Arctic thin ice clouds from a multibroadband thermal radiometer using lidar and radiosonde measurements as auxiliary inputs to the inversion routine. Specific validation elements were extracted from the combination of lidar and radar data, as well as AERI data. Our retrieval technique was applied to 150 thin ice clouds measured at the PEARL observatory (Nunavut, Canada).

The results of this study demonstrate the potential for retrieving key ice cloud parameters from thermal IR radiometry (with bands between 8 and 13 µm). The COD retrieval algorithm showed good agreement with the validation COD obtained from integrated lidar profiles. The retrievals of Deff showed a marginal correlation for particle sizes restricted to the TIC2 category (a constraint that was driven by the insensitivity of the lidar–radar retrievals to small cloud particles). This is likely due to the weak sensitivity of thermal IR measurements for particles larger than ≈ 100 µm. However, a classification of thin ice clouds in terms of TIC1 and TIC2 classes, using a threshold discrimination of 30 µm results in a significant classification accuracy of 83 % for our passive retrieval algorithm. Further analysis showed that the extinction profile and particle shape had a relatively weak impact on the retrieval results. Comparisons with the MIXCRA algorithm confirm the robustness of the optical depth retrieval.

An important application of our work would be to deploy this technique as part of a network of low-cost, robust instruments to monitor Arctic clouds. Because their occurrence, type and altitude are spatially inhomogeneous according to, we believe that additional ground-based stations would be helpful to broaden our knowledge of arctic ice clouds. Aside from being a ground-based retrieval approach in its own right (in tandem with a lidar system), this method can also be used for comparison with CALIPSO's level 2 products. The CALIPSO remote sensing suite technique employs an on-board imaging IR radiometer and the CALIOP lidar to enable the retrieval of particle size and optical depth across a narrow swath image . Our retrieval, viewed as a CALIPSO validation technique is rendered all the more interesting because of the geographic position of the PEARL site; it is a high-Arctic site that sees frequent thin-cloud events and its position near the maximum latitude of the CALIOP polar orbit ensures that there are frequent overpasses of that sensor package (within a radius of hundreds of kilometers).