Use of A-train satellite observations (CALIPSO–PARASOL) to evaluate tropical cloud properties in the LMDZ5 GCM

Abstract

The evaluation of key cloud properties such as cloud cover, vertical profile and optical depth as well as the analysis of their intercorrelation lead to greater confidence in climate change projections. In addition, the comparison between observations and parameterizations of clouds in climate models is improved by using collocated and instantaneous data of cloud properties. Simultaneous and independent observations of the cloud cover and its three-dimensional structure at high spatial and temporal resolutions are made possible by the new space-borne multi-instruments observations collected with the A-train. The cloud cover and its vertical structure observed by CALIPSO and the visible directional reflectance (a surrogate for the cloud optical depth) observed by PARASOL, are used to evaluate the representation of cloudiness in two versions of the atmospheric component of the IPSL-CM5 climate model (LMDZ5). A model-to-satellite approach, applying the CFMIP Observation Simulation Package (COSP), is used to allow a quantitative comparison between model results and observations. The representation of clouds in the two model versions is first evaluated using monthly mean data. This classical approach reveals biases of different magnitudes in the two model versions. These biases consist of (1) an underestimation of cloud cover associated to an overestimation of cloud optical depth, (2) an underestimation of low- and mid-level tropical clouds and (3) an overestimation of high clouds. The difference in the magnitude of these biases between the two model versions clearly highlights the improvement of the amount of boundary layer clouds, the improvement of the properties of high-level clouds, and the improvement of the simulated mid-level clouds in the tropics in LMDZ5B compared to LMDZ5A, due to the new convective, boundary layer, and cloud parametrizations implemented in LMDZ5B. The correlation between instantaneous cloud properties allows for a process-oriented evaluation of tropical oceanic clouds. This process-oriented evaluation shows that the cloud population characterized by intermediate values of cloud cover and cloud reflectance can be split in two groups of clouds when using monthly mean values of cloud cover and cloud reflectance: one group with low to intermediate values of the cloud cover, and one group with cloud cover close to one. The precise determination of cloud height allows us to focus on specific types of clouds (i.e. boundary layer clouds, high clouds, low-level clouds with no clouds above). For low-level clouds over the tropical oceans, the relationship between instantaneous values of the cloud cover and of the cloud reflectance reveals a major bias in the simulated liquid water content for both model versions. The origin of this bias is identified and possible improvements, such as considering the sub-grid heterogeneity of cloud properties, are investigated using sensitivity experiments. In summary, the analysis of the relationship between different instantaneous and collocated variables allows for process-oriented evaluations. These evaluations may in turn help to improve model parameterizations, and may also help to bridge the gap between model evaluation and model development.

Appendices

Appendix 1: Spatio-temporal sampling of CALIPSO and PARASOL observations

1.1 Temporal resolution

The CALIPSO and PARASOL instruments follow the same sun-synchronous A-train orbit, which passes over each location twice a day at about 1:30 AM and 1:30 PM local solar time. Since PARASOL collects measurements during daytime, only the daytime CALIPSO data are considered. The two instruments fly over the same orbit so they document the same cloud parcel simultaneously at about 1:30 AM local solar time.

The incomplete sampling of the diurnal cycle has a negligible impact on the results (less than 1 %) (Chepfer et al. 2008).

1.2 Spatial resolution

A PARASOL pixel (6 × 6 km) is much larger than a CALIOP/CALIPSO pixel (330 m along-track, 75 m cross-track). One value of the directional reflectance is associated to at least 18 lidar profiles. To overcome these differences, the CALIOP cloud cover and the PARASOL reflectance are processed independently on a statistical basis, and then compared to daily mean values on a 2° × 2° grid (several hundreds of km²). To test the impact of the sampling over seasonal mean results on a 2° × 2° grid, two PARASOL reflectance datasets have been built in the same viewing direction (θ_v = 27°,ϕ_s − ϕ_v = 320°). The first dataset includes all reflectance values measured by PARASOL, and the second dataset includes only the reflectance measured along the CALIPSO ground track. The maximum distance between a PARASOL and a CALIOP pixel in the first dataset is 50 km. The number of measurements is about 30 % lower in the second dataset compared to the first dataset. Maps of 2° × 2° mean directional reflectances and variances are similar for both datasets (not shown), although the second one is noisier, thus suggesting that both PARASOL datasets (collocated or not with CALIOP) can be analyzed. The similarity between the two datasets also shows that the few PARASOL pixels collocated with CALIOP (6 × 6 km²) are representative of all PARASOL pixels included in the 2° × 2° grid cell.

Similarly, it is reasonable to consider that the CALIOP dataset (even with a 330 m × 75 m resolution), when averaged over several months, is statistically representative of the monthly/seasonal cloud cover within a GCM grid cell.

Appendix 2: Sensitivity of the PARASOL monodirectional reflectance to the atmosphere’s composition

2.1 Optical properties

The cloud particle optical properties (e.g. single scattering albedo, scattering phase function, and extinction coefficient) depend on the wavelength, on the particle size and on its shape. Since the absorption phenomena is negligible in ice and water at 864 nm (Warren 1984; Hale and Querry, 1973), the single scattering albedo is close to one regardless of the size and shape of the particles. Since the radius of cloud particles is always larger than the wavelength considered here, the scattering phase function is sensitive to the particle shape, but not highly sensitive to the droplet size. A spherical shape assumption, which is typical of liquid water computed with the Mie theory, is used. A non-spherical shape, which is typical (Chepfer et al. 2002) of ice crystals whose optical properties are computed with Geometric Optic enhanced with Finite Differential Time Domain (Yang et al. 2000; 2001) is also used. As shown in Fig. 10a, their scattering phase functions differ significantly for scattering angles close to backscattering (180°), haloes (22° and 44°) and rainbow (140°), and also between 90° and 130°, which corresponds to the viewing angle and to the solar zenith angle selected for PARASOL data in the tropics. Complementary computations (not shown) indicate that the scattering phase function at this wavelength depends less on the particle size than on its shape. On the contrary, the particle extinction coefficient directly depends on the particle size: it is proportional to the scattering efficiency (close to 2, since the particles are larger than the wavelength) multiplied by the particle cross section, which is expressed as a function of the particle size.

2.2 Radiative transfer computations

The directional reflectance is computed using a doubling-adding radiative transfer code (De Haan et al. 1986). The cloud particles optical properties, such as the single scattering albedo and the truncated scattering phase function developed in Legendre polynomial, are introduced in the radiative transfer code. The Rayleigh scattering is also considered in the computation, even though its contribution to the total directional reflectance is small (τ is about 0.013 for the whole atmospheric column). Since the viewing direction studied is off-glitter, the ocean is described as a lambertian surface, with a constant plane albedo of 0.03. The directional reflectance is then computed as in Chepfer et al. (2002) for various cloud optical depths and solar zenith angles.

Figure 10b shows that changes of reflectance values due to solar zenith angle variations are less than 0.1 in the tropical regions (30°S–30°N, 18° < θ_s < 60°) for a given phase function. It reaches a maximum of 0.15 between the ITCZ and the higher observable latitudes (θ_s > 60°). Variations of the latitudinal reflectance that are larger than 0.15 (0.1 in the Tropics) cannot therefore be attributed to variations of θ_s. These variations are due to changes in the atmosphere composition (clouds). The sensitivity of the reflectance to the cloud particles scattering phase function is maximum at high latitudes/high solar zenith angle (0.13) and slightly reduces in the tropics (0.1).

2.3 PARASOL simulator

The PARASOL simulator is initiated with the mixing ratios of in-cloud liquid and ice water content in each model grid cell. These mixing ratios are then converted into sub-grid mixing ratios using SCOPS. In each subcolumn, the total cloud optical depth (τ_tot) is the sum of the optical depth of the ice (τ_tot_ice) and liquid (τ_tot_liq) in the subcolumn. These are computed assuming that the cloud particles are spherical, with a radius equal to the simulated effective radius. For five solar zenith angles (θ_s = 0°, 20°, 40°, 60° and 80°) and given the total cloud optical depth, two directional reflectance values are then computed for each day and for each solar zenith angle assuming that the cloud is entirely composed of liquid water (Refl_liq) or entirely composed of ice water (Refl_ice). These reflectance values are derived from a bilinear interpolation over pre-calculated look-up tables, which contain results of radiative transfer computations (“Appendix 2”) for the cloud particle’s shape assumption (spherical and non spherical) used in the model. The subgrid directional reflectance is then computed as follow: Refl = (Refl_liq*τ_tot_liq + Refl_ice*τ _tot_ice)/τ _tot. The directional reflectance obtained for each subgrid is then averaged over each GCM grid cell, for each day and for each θ_s. After the simulations have been performed, the five monodirectional reflectances corresponding to the five solar zenith angles from the simulator’s outputs are used to linearly interpolate the monodirectional reflectance depending on the monthly mean value of the solar zenith angle at each grid point. The simulated monodirectional reflectance can then be directly compared to the observations (Fig. 9).

Fig. 9

a Scattering phase function for spherical and non-spherical particles. b Monodirectional reflectance simulated as a function of the solar zenith angle for spherical and non-spherical particles in the viewing direction (θ_v = 27° θ_v = 320°)

Appendix 3: Traditional global monthly mean evaluation of cloud properties

3.1 Cloud cover

On average, cloud cover is underestimated over the tropical regions and is broadly consistent with observations in the mid and high latitudes (Fig. 10). In the Tropical Western Pacific, along the ITCZ and the SPCZ, the cloud cover simulated by LMDZ5A is about 60–70 % whereas observations indicate a cloud cover ranging from 80 to 100 %. In regions where the cloud cover is low, such as in the trade wind cumulus region, observations indicate a cloud fraction between 40 and 60 % whereas the simulated cloud cover is only about 20 to 50 %. Although LMDZ5B underestimates the averaged cloud cover in the tropics, the bias is reduced by a factor of 2 compared to LMDZ5A. The improvement is very significant in almost all fully overcast regions (e.g. warm-pool, east Pacific, and Atlantic), even with an overestimated cloud cover.

Fig. 10

Geographical distribution of the total mean cloud cover over the ocean averaged over the period 2007–2008 a observed with CALIPSO-GOCCP during day time, b simulated with LMDZ5A and the lidar simulator, c simulated with LMDZ5B and the lidar simulator; d zonal mean of the same quantity observed (red line) and simulated (LMDZ5A: black line and LMDZ5B: black dotted line)

3.2 Cloud vertical profile

The zonal mean vertical distribution of the observed CALIPSO-GOCCP cloud fraction clearly highlights the well-known connetions between the cloud characteristics and the large circulation of the atmosphere (Fig. 11a). The altitude of the higher clouds follows the tropopause height, and decreases from the equator to the poles. The LMDZ5A model with the lidar simulator produces a cloud fraction of high-level clouds that is too large almost everywhere and the altitude of these clouds is too high, in particular over the polar region of the southern hemisphere (Fig. 11b). In the tropics, the cloud fraction at low and middle altitudes is strongly underestimated in LMDZ5A. Although this feature is amplified by the masking effect of high clouds on the lidar signal (thick high level clouds, with typical Cτ > 3, attenuate the signal and mask low- and mid-level clouds that might exist below them), this underestimation occurs with the cloud cover simulated by the model (i.e. without using the lidar simulator, cf Chepfer et al. 2008). At higher latitudes, the model cannot simulate the large vertical extent of the frontal clouds associated with storms. Instead, it simulates two separate groups of low- and high-level clouds. This zonal mean vertical distribution of clouds is improved in LMDZ5B (Fig. 11c). In the tropics, boundary level clouds are simulated, although too low and too concentrated in one single layer. At middle and high latitudes, the model almost simulates the continuous vertical structure of the cloud fraction.

Fig. 11

Zonal mean cloud fraction profile averaged over the period 2007–2008, a observed from CALIPSO-GOCCP, b simulated with LMDZ5A and the lidar simulator and c simulated with LMDZ5B and the lidar simulator

3.3 Cloud reflectance

In the trade wind regions, the observed cloud reflectance (typical value of 0.15) is only slightly higher than the clear-sky value (approximately 0.03), indicating that clouds are optically thin. This is not the case in the two model versions (Fig. 12b, c). They strongly overestimate the cloud reflectance almost everywhere, in particular over the subtropical oceans and in the mid and high latitudes. The models versions cannot reproduce the contrast between the higher values (≈0.3) of cloud reflectance observed along the ITCZ and the Eastern Pacific ocean and the lower values (<0.2) over the tropical trade wind cumulus region. They both simulate high cloud reflectances (>0.2) over the tropics. On average, the cloud reflectance, and therefore the cloud optical thickness, simulated by the models over the ocean is too high almost everywhere.

Fig. 12

Same as Fig. 10 for the monodirectional cloud reflectance over ocean, on average for the period 2007–2008, a observed with PARASOL, b simulated with LMDZ5A and the PARASOL simulator, c simulated with LMDZ5B and the PARASOL simulator; d zonal mean of the same quantity observed (red line) and simulated (LMDZ5A: black line and LMDZ5B: black dotted line)