Testing general circulation model parameterizations against observations is traditionally done by comparing simulated and observed, time-averaged quantities, such as monthly mean cloud cover, evaluated on a stationary grid. This approach ignores the dynamical aspects of clouds, such as their lifecycle characteristics, spatial coverage, temporal duration, and internal variability. In this study, a complementary Lagrangian approach to the validation of modeled tropical cloudiness is explored. An automated cloud detection and tracking algorithm is used to observe and track overcast decks of cloud in a consecutive set of hourly METEOSAT-5 images and the National Center for Atmospheric Research Community Climate Model version 3 (NCAR CCM3). The algorithm is applied to the deep convective cloud systems of the tropical Indian Ocean region during a 49 day period of the 1999 winter monsoon season. Observations of precipitation are taken from the Tropical Rainfall Measuring Mission (TRMM) satellite in addition to a METEOSAT-5 infrared rainfall technique that is calibrated using the TRMM data.

Clouds, defined as overcast decks, are observed spanning spatial scales from 25 km^2 to greater than 10^7 km^2, as well as temporal scales from 1 hour to greater than 100 hours. Semi-permanent decks of anvil and cirrus cloud, with numerous regions of deep convection embedded within, dominate total cloud cover. Bridging between convective centers within the deck by cirrus cloud, particularly during the suppressed portion of the diurnal cycle of convection may help to maintain the integrity of semi-permanent overcast decks over long time scales. At scales greater than 10^6 km^2 the size distribution of simulated clouds is biased such that the dominant scale of clouds is several million square-kilometers larger than the dominant scale of observed clouds. Virtually all of the simulated precipitation occurs at rain rates lower than 2 mm hr^-1, while as much as 25% of observed precipitation occurs at rain rates higher than 2 mm hr-1. Precipitation associated with deep convection in observed semi-permanent cloud systems is organized into more localized mesoscale structures of adjacent convective cells attached to stratiform precipitation regions. A separate analysis of TRMM data (Wilcox and Ramanathan 2001) determined that such structures can exceed the size of grid cells in coarse-grid global models and have area-averaged rain rates up to and exceeding 2 mm hr^-1. These mesoscale convective systems are responsible for the extreme, episodic precipitation events that are not parameterized in the model. The simulated cloud systems gently precipitate throughout their duration and everywhere within their boundaries. The model lacks a process that acts to organize the convective cells within fewer grid cells, in addition to a representation of the observed stratiform precipitation structures. A modification to CCM3 is tested that is intended to account for the evaporation of upper-level precipitation in mid-level mesoscale downdrafts. The modification results in only a slight change in domain-averaged precipitation. However, it causes a significant shift in the distribution of precipitation toward higher rain rates that is more consistent with the distribution of TRMM observed rain rates. The modification demonstrates the sensitivity of the model to one important component of mesoscale organized convection.