1. The aim of FASTER (FAst-physics System TEstbed and Research) is to evaluate and improve the parameterizations of fast physics (involving clouds, precipitation, aerosol) in numerical models using ARM measurements. One objective within FASTER is to evaluate model representations of fast physics with long-term continuous cloud observations using an NWP testbed. This approach was successful in the European Cloudnet project. NWP model data (NCEP, ECMWF, etc.) is routinely output at ARM sites, and model evaluation can potentially be achieved in quasi-real time. In this poster we outline various ways of evaluating the parametrization of cloud fraction in NWP models. 1. Overview 2. Observed and NWP model cloud fraction 6. Variability 7. Cloud fraction forecasts: Has there been any improvement in the skill scores? Observations How well do NWP models forecast cloud fraction? Over the SGP site, models tend to show more skill in the mid-troposphere than in the boundary-layer. Met Office global model has much lower skill for high cloud fraction amounts. It is known that this model has difficulty in filling the grid-box completely with cloud. NB. Not all models are shown with the same forecast leadtime! Seasonal Diurnal cycle All models underpredict cloud fraction throughout the year. Cloud fraction vs. omega at 500 mb Rather than ‘tune’ the model to give the correct mean, we want to know why the mean model cloud fraction is low. Split the mean cloud fraction into two components: How often is cloud present (above a given threshold)? When cloud is present, how much cloud is there? Investigate the PDF of cloud fraction in the mid-level. ECMWF has non-radiative ‘snow’ and does not produce enough grid-boxes which are full of cloud. At SGP, cloud fraction skill scores drop substantially in summer, even for ERA Interim Re-Analyses. In some years, at the height of summer, forecasts are often no better than persistence. There appears to be no improvement in the cloud fraction forecast over time. At SGP, any small improvement is likely to dominated by the variability from year to year. Establishment of an NWP testbed using ARM Data Ewan O’Connor 1,2 (e.j.oconnor@reading.ac.uk)e.j.oconnor@reading.ac.uk Robin Hogan 1 Yangang Liu 3 1 Department of Meteorology, University of Reading, UK 2 Finnish Meteorological Institute, Helsinki, Finland 3 Brookhaven National Laboratory At SGP, models tend to underestimate the mean cloud fraction, especially in the mid- levels. Note: Météo France adjusted their cloud fraction parametrization in 2006. Their mean model cloud fraction is now similar to the other models shown here. 3. Mean cloud fraction The observed cloud fraction is determined from Doppler radar and lidar data and is calculated on each model grid. Model cloud fraction can be prognostic or diagnostic. Below is an example of one month of data for June 2004 over ARM SGP. NCEP GFS model ECMWF model ERA Interim Analyses Met Office Global Model 4. Cloud fraction skill scores 5. Forecast degradation How does the model forecast degrade over time? Is mid-level cloud more susceptible? Does the model have issues with spin-up? In addition, boundary layer cloud during the day is not forecast often enough Not enough cloud forecast in anti- cylconic conditions, especially low-level boundary-layer cloud.