Validating the Validation: The Influence of Liquid Water Distribution in Clouds on the Intercomparison of Satellite and Surface Observations

The intercomparison of LWP retrievals from observations by a geostationary satellite imager [Spinning Enhanced Visible and Infrared Imager (SEVIRI) on board Meteosat Second Generation (MSG)] and a ground-based microwave (MW) radiometer is examined in the context of the inhomogeneity of overcast cloudy skies. Although the influence of cloud inhomogeneity on satellite observations has received much attention, relatively little is known about its impact on validation studies. Given SEVIRI’s large field of view (3 km × 6 km for northern Europe), especially when compared to the narrow width of the radiometer tracks (100–200 m), cloud inhomogeneity may be expected to significantly affect the satellite retrieval validation. This paper quantifies the various validation uncertainties resulting from cloud inhomogeneities and proposes an approach to minimize these uncertainties. The study is performed by simulating both satellite and ground-based observations through resampling a set of high-resolution (100 m) cloud fields that are derived from 1 km × 1 km Moderate Resolution Imaging Spectroradiometer (MODIS) observations. The authors’ technique for generating realistic high-resolution LWP fields preserves the information present in the original observations and creates extra LWP variation at smaller-length scales by considering clouds as simple fractals. The authors believe that this is a new technique for creating high-resolution LWP fields. Validation errors resulting from cloud inhomogeneity can be classified in two groups. The first group relates entirely to the retrieval process for satellite observations and includes the well-known plane-parallel bias as well as field-of-view mismatches between different channels used in the retrieval. The second group relates to differences in the scene observed by satellite and ground-based sensors. This includes systematic shifts in the observed scene resulting from viewing conditions (parallax effect), offsets between satellite images and ground sites, and different fields of view. Results indicate that the plane-parallel bias for the authors’ sample of 604 clouds has a median value of −3.3 g m−2. All other error contributions appear to be random and have no biases. For individual observations, the parallax effect easily dominates the total error budget for sites that are observed under large viewing angles (e.g., northern Europe). The authors show that this error may be partly compensated by using information about cloud-top heights and by spatially interpolating among an array of SEVIRI pixels to obtain the best estimate of the satellite-retrieved LWP value over the ground site. Optimal intercomparison of satellite and ground-based observations is also possible by matching the track length of the ground observations to the imager’s pixel size in the wind direction. Thus, one surprising conclusion is that the LWP errors resulting from the second group (scene differences) are significantly larger than those resulting from the first group (satellite retrieval), even after corrections have been applied. Smaller satellite pixels do not alleviate the problem but rather aggravate it, unless the parallax error is corrected. Temporal or spatial averages of observations may be used to reduce the random errors, but the statistical properties of such aggregates are, at the moment, not obvious for reasons that will be discussed. Calibration errors are not considered in the present study.