REMOTE-SENSING OF INTEGRATED CLOUD LIQUID WATER - DEVELOPMENT OF ALGORITHMS AND QUALITY-CONTROL

Algorithms are developed to infer integrated cloud liquid water path ( LWP) over the oceans from spaceborne and ground-based passive microwav e measurements. These algorithms are built from simulated observations , which are calculated with a radiative transfer model applied to a se t of about 10,000 atmospheric profiles obtained from the European Cent re for Medium-Range Weather Forecasts forecast model. In this model th e liquid water content is computed from a prognostic cloud scheme. A m ultilinear regression is applied to functions of simulated brightness temperatures (log linear form) and LWP to derive the algorithm coeffic ients. The retrieval accuracy based on the regression analysis includi ng instrumental noise is 0.0257 and 0.0345 kg m(-2) for the DMSP speci al sensor microwave imager (SSM/I) and the ERS1 along-track scanning r adiometer/microwave (ATSR/M), respectively, and 0.0308 kg m(-2) for th e ground-based radiometer. It is shown that the log linear form is ade quate to transform the nonlinear problem into a quasi-linear problem f or LWP below 0.8 kg m(-2). The coherence of the global approach is ver ified through the validation of total perceptible water (TPW) algorith ms developed in a way similar to LWP algorithms. The LWP retrievals fr om the algorithm for the ground-based radiometer are in good agreement with retrievals from airborne measurements performed in the vicinity of the radiometer. A coherence test is performed for ATSR/M, benefitin g from the coincident infrared images obtained from an infrared radiom eter (ATSR/IR) aboard the same platform to select clear-air areas. Reg ardless of the slight mean bias of the inferred LWP due to inaccurate calibration, there is no anomalous dependency upon latitude, i.e., upo n high water vapor contents in the tropics and strong winds in the hig h latitudes. The results of the algorithm for SSM/I are compared with a Meteosat cloud classification. When the classification detects the o cean surface, the algorithm systematically retrieves contents close to zero. The retrievals for other classes (i.e., low stratiform clouds, medium clouds) are consistent with the Meteosat data; retrievals in th e presence of low stratiform clouds appear more realistic than values provided by some already published algorithms. It is also shown that u p to 0.8 kg m(-2) the log linear regression approach has a quality of the same order as a variational method, which requires much more compu tation time.