AEROSOLS Validation Results Over Ocean
Validation of Aerosol Retrieval Over Ocean Using AERONET network
Aerosol optical Thickness Retrieval
Automatic and objective criteria are used to select the data both in sun photometer and satellite database. To help for the analysis, dataset was splitted into two classes (Angstrom exponent smaller than 0.5 or larger than 0.5).
For the validation purpose, mean AOT are calculated from the observations taken within ±30 minutes of the POLDER overpass time (10:30 AM). About 30 AERONET sites were considered to study aerosol retrievals accuracy. Around 70 % are located in the Mediterranean Basin.
Angström Exponent exponent retrieval
For the accuracy estimate on a, presented on Fig. 2 , once again, we keep only data with AOT > 0.08 at 865 nm, which corresponds to AOT (670) > 0.10 for our dataset.
For the class with a larger than 0.5 and AOT (865) larger than 0.08, the r.m.s on a is 0.29. For the class with a smaller than 0.5, r.m.s value is 0.18. Considering, both classes together, leads to a global fit of 0.90 alpha + 0.04, with a quite good correlation (R=0.88, with 0.28 r.ms.). In this case, r.m.s on AOT is 0.04. The results obtained here with the advanced algorithm applied to POLDER-2 data are clearly strongly better than those that were obtained with the former one applied to POLDER-1 where systematic a severe bias was found for alpha (0.65 for slope and + 0.08 for intercept).
Spherical Fine mode AOT retrieval
Fine mode is helpful to distinguish between man-made aerosol and natural aerosol (mainly dominated by coarse mode aerosol such as mineral dust and sea salt). For the analysis of fine mode AOT retrieval, we use the fine mode AERONET AOT that is recalculated from the size distribution (spherical inversion mode) derived from the sky-radiance measurements (Dubovik et al., 2000). For that purpose, at least one retrieved size distribution at time less than two hours from the satellite overpass is required. Moreover, we require that (i) total AOT at 865 nm is larger than 0.08 and (ii) the relative error between the measured and recomputed total AOT is less than 2 % to be included in the validation dataset. Comparison is presented on Fig. 3.
Due to our drastic selection, only 35 very reliable co-located data remain for the match-up. The agreement is of an unprecedented quality, mainly thanks to angular and polarized capabilities. One possible explanation for the observed discrepancies between AERONET and POLDER is the slightly different fine mode definition that could introduce some bias. To compute the fine mode AOT from the retrieved size distribution, AERONET algorithm integrates over particle radius up to 0.6 µm when with POLDER one integrates only between the fine monomodal lognormal size distribution limits. It is clear that the amplitude of this bias depends on the coarse mode potential contamination for particle radius lower than 0.6 µm. Fortunately, in our dataset, few data are associated with low Angstr&omul;m exponents, which limits the contamination, here.
Comparison with MODIS TERRA
Both algorithms use bimodal aerosol size distributions; the ratio between small and large particles is deduced from the spectral measurement behaviour, while the signal level is related to the optical thickness. In specific geometries and using directional and polarized measurements, POLDER is able to detect non spherical aerosols (dust), which are not distinguished by MODIS that considers such particles as large spherical ones.
Two comparisons of the total optical thickness at 550 nm are presented over Indian and Pacific Oceans, only considering cloudiness observations: they exhibit a good correlation between both sensors.