POLDER 2 - CLOUD PROPERTIES: Cloud Optical Thickness
CLOUD OPTICAL THICKNESS
In the previous algorithm described in Buriez et al. (1997), all clouds were assumed to be composed of water droplets with an effective radius of 10 µm. The water droplet model was found to be inadequate for ice clouds (Parol et al., 1999). Moreover, the ice fractal polycrystal model used in the ISCCP algorithms (Rossow et al., WMO/TD-No. 737, 1996) was found to be much less appropriate than the IHM (Inhomogeneous Hexagonal Monocrystal) model (Doutriaux-Boucher et al., 2000). This IHM model gives satisfactory results with both total and polarized reflectance measurements (C.-Labonnote et al., 2000; C.-Labonnote et al., 2001). The IHM model is now applied to pixels labeled as "ice" in the cloud phase algorithm. The liquid water clouds are treated with a droplet model which effective radius is 11 µm over ocean and 9 µm over land, in agreement with POLDER polarization measurements (Bréon and Colzy, 2000).
Nevertheless, a major default remains in the optical thickness retrieval, due to the limitations of the plane-parallel assumption (Buriez et al., 2001). The retrieved directional values of optical thickness are thus averaged using angle-weighting functions. In this way, the most reliable viewing directions are taken into account. The angle-weighting process is based on the statistical analysis of a large set of ADEOS 1 - POLDER data and should compensate the induced theoretical approximations used in the algorithm.
Under these assumptions, the optical thickness of a cloudy superpixel (composed of 3 x 3 pixels) is derived by two different techniques:
- For each cloudy pixel, the optical thickness is retrieved from radiance measurements. The linear mean is then computed at the superpixel scale.
- The 3 x 3 measured radiances are first spatially averaged and then the radiative mean optical thickness is retrieved.
Practically this is done for each viewing direction. Then, the retrieved directional values are angle-weighted averaged.
The comparison of the two means (linear and radiative) is indicative of the spatial variability of the observed cloud optical thickness. For partially cloud covered superpixels, the linear mean optical thickness TAU corresponds to the only cloudy part of the scene. Thus the radiative mean optical thickness TAU* has to be compared to CC x TAU, where CC is the cloud cover fraction. Practically, in our products we keep TAU and = 1 - TAU* / (CC x TAU). Note that TAU is directly related to the mean condensed cloud water content while is similar to the cloud inhomogeneity factor used by Rossow et al. (J. Climate, 15, 557-585, 2002) from ISCCP data.
The cloud spherical albedo is also calculated as in our previous algorithm. For a given microphysics model, the cloud spherical albedo (defined over black surface with no atmosphere) is a one-to-one function of the cloud optical thickness. The radiative mean optical thickness is thus very close to the optical thickness associated to the mean spherical albedo as calculated in the ISCCP algorithm and in the previous POLDER algorithm. It is also close to the logarithmic mean optical thickness used in the MODIS algorithm. For practical purposes, the directional values of the retrieved cloud spherical albedo are preserved in our products.
© CNES Copyrights 2002