Evaluation against CALIPSO lidar observations of the multi‐geostationary cloud cover and type dataset assembled in the framework of the Megha‐Tropiques mission

To support the Megha‐Tropiques space mission, cloud mask and cloud type classification are needed at high spatial and time resolutions over the tropical belt for water vapour and precipitation analysis. For this purpose, visible and infrared radiance data from geostationary satellites ( GEO ) are used with a single algorithm initially developed by SAFNWC (Satellite Application Facility for Nowcasting) for Meteosat Second Generation. This algorithm has been adapted by SAFNWC to the spectral characteristics and field of view of each satellite. Retrieved cloud cover characteristics (cloud mask, classification and top pressure) are evaluated over four months in summer of 2009 against CALIOP lidar observations from the CALIPSO polar‐orbiting satellite. To better identify atmospheric and instrumental issues, separate analyses are performed over land and ocean, for 1:30 a.m. and 1:30 p.m. CALIPSO overpasses and for each GEO . Both mean cloud‐cover occurrence and instantaneous cloud‐cover statistics are compared. We found that each classification has specific features, which depend on observed cloud regimes and instrument capabilities. Most important, a common behaviour of the GEOs against CALIOP depending on cloud type is observed. We found that GEO cloud occurrence is lower by about 10% than for CALIOP , with the largest biases over land during daytime and the smallest over ocean during daytime. Further detailed analysis reveals specific discrepancies in the retrieved cloud types. As expected, high‐level clouds are detected more frequently by the lidar. We show that, over ocean when the optical thickness of detected high‐level clouds is limited to greater than 0.1 in the comparisons, multi‐spectral radiometry performs very similarly. However, the most significant difference is attributed to non‐detection of low‐level clouds that are often broken, which causes a reduction of up to 20% in low‐level cloud fraction and even 30% in some regions. Other significant differences are seen over land, where mid‐level clouds are not detected or are misclassified.