EVALUATION OF GLOBAL NUMERICAL WEATHER PREDICTION ANALYSES AND FORECASTS USING DMSP SPECIAL SENSOR MICROWAVE IMAGER RETRIEVALS .2. ANALYSESFORECASTS INTERCOMPARSION WITH SSM/I RETRIEVALS/

Satellite retrievals of environmental parameters for October 1993 were used to validate the Canadian Meteorological Center (CMC) analysis/fo recast system and an improved version of the forecast model (including a prognostic equation for cloud water). In part 1 [Deblonde and Wagne ur, this issure], SSM/I (special sensor microwave imager) retrievals u sing different algorithms were intercompared for each environmental pa rameter (column integrated water vapor (IWV), near-surface oceanic win d speed (SWS), liquid water path (LWP) and surface rain rate (SRR). He re a subset of the SSM/I retrieval algorithms that performed best and other observation data sets are used to evaluate the hydrological cycl e of the analysis/forecast system on a monthly timescale. Over the glo bal oceans the agreement between analyzed IWV, forecasted IWV, and SSM /I retrievals was quite good (R similar to 0.96). However, where Humsa t (GOES) retrieved water vapor profiles were assimilated, significant biases were observed in large areas (up to similar to 15 kgm(-2)). Sig nificant differences were found between analyzed SWS and SSM/I-retriev ed SWS. In the tropics, zonally averaged analyzed SWS was underestimat ed by up to 1.4 ms(-1) and in the southern hemisphere midlatitude diff erences of similar to 1 ms(-1) were found. Forecasted cloud fraction f rom the improved model was compared with that observed using the Humsa t retrieval system. Over the open oceans, modeled cloud fraction was o verestimated by 8%, suggesting that cloud formation was too active. By comparing SSM/I LWP retrievals with those forecasted it was possible to identify mispositioning of cloud systems that were associated with noted biases in the humidity analysis. It was also shown that the choi ce of function to obtain the liquid phase component of the modeled clo ud water modified considerably the magnitude of the monthly mean LWP, particularly in the midlatitudes. The differences in patterns between precipitation short-range forecasts (accumulated over a month) obtaine d with the operational and improved forecast models were considerably smaller than the differences between each of these forecasted fields a nd SSM/I retrievals or long-term climatologies.