Simulations vs. observations of in-cloud icing - sensitivity to the parameterization of cold cloud microphysics

In cold climates, in-cloud icing by supercooled droplets is a major hazard for many human activities, e.g., aviation, power transmission lines, wind turbines and telecommunication towers. Consequently, accurate forecasts of such events are of great interest. Until recently, such forecasts used crude empirical approaches, e.g., linking icing occurrence to simulated profiles of relative humidity. Quite recently, within the European COST727 action, a significant potential for accurate forecasting of ground-level icing, based directly on NWP model output, was demonstrated for several cases of in-cloud icing at various locations in Europe, where icing observations were available. A large sensitivity was found to model resolution and the treatment of cloud microphysics (Fikke et al., 2008). In this study, several European icing events have been studied in more detail, paying particular attention to the formulation of the cloud microphysics in the NWP model. The model of choice is the non-hydrostatic WRF model, in which several cloud microphysics schemes are available. We have used three schemes: a) the simple and computationally efficient Ferrier ETA scheme (EGCP1), which has only one prognostic variable for the mass of total condensed water; b) the more detailed Thompson scheme (Thompson et al., 2004; 2008), having 5 prognostic hydrometeor species (cloud water, cloud ice, rain, snow and graupel), and a prognostic equation for ice crystal number; c) the full two-moment Morrison scheme (Morrison et al., 2005a; 2005b), with prognostic equations for both mass and number of 5 species: cloud water, cloud ice, rain, snow and graupel. The first part of the study deals with several icing events in Northern Finland, for which accurate in situ measurements of supercooled water by a rotating multicylinder have been carried out on a mountain top (Mt. Ylläs). In order to adequately resolve the terrain, the horizontal grid spacing in the innermost domain must be 1 km or less. In the comparison of model results and measurements, we investigate local terrain effects, terrain blending, and how to extract and interpret information on the cloud liquid water and the cloud droplet size from the NWP model. For the eight cases considered, the predicted LWC is sensitive to both the horizontal grid spacing and to the choice of cloud microphysics scheme. Most evident is the sensitivity to the horizontal grid spacing, the lowest mean absolute error (MAE) being obtained by using a horizontal grid spacing of 0.333 km. The comparison of the cloud microphysics schemes shows that we get the lowest MAE of 0.083 g m-3 by applying the Thompson scheme. The corresponding errors for the Morrison and the Ferrier schemes are 0.128 g m-3 and 0.136 g m-3, respectively. Detailed ongoing studies investigate the reasons behind these differences. When coarser grid spacing is used (dx = 3 km), the model skill decreases dramatically, and the MAE values rise to 0.186 g m-3, 0.201 g m-3 and 0.261 g m-3 for the Thompson, Morrison and the Ferrier schemes, respectively. The resolution dependency seems to be closely related to a reduction of the mountain elevation in the model, and hence weaker terrain-induced vertical motions. The second part of the study deals with 4 recent cases of aircraft icing in southern Norway, one of which almost had fatal consequences. Only the Ferrer ETA scheme and the Thompson scheme were run for these cases. Since direct observations of supercooled water were not available, the model results are validated using observed soundings, weather radar output and synoptic observations. In the first case, from 14 September 2005, the remnants of the tropical cyclone Maria brought extremely humid and warm air towards southern Norway. Unexpectedly, the aircraft encountered severe icing at altitudes between 3000 and 4300 m, and heavy rain at -10°C. The Thompson scheme simulates significant supercooled water all the way up to 7000 m, in a zone behind the advancing cold front. Conversely, in the simpler microphysics scheme, the condensate is quickly converted to snow, and very little supercooled water is simulated. In another case from 14 December 2008, warm, moist air over a surface inversion caused freezing drizzle over Oslo and surroundings, causing problems for air traffic. Also in this case, the simple scheme unrealistically produces widespread snow, effectively depleting the supercooled cloud water, while the more detailed Thompson scheme simulates a cloud of supercooled liquid water of about 0.5 g m-3 around 1000 m elevation, with significant freezing drizzle underneath. For the other two cases, the results gave qualitatively similar results to those of the former two cases. Among other things, the formulation of the size distribution for snow seems to be crucial for the success of the Thompson scheme in these cases. Fikke, S. M., J. E. Kristjánsson, and B. E. K. Nygaard, 2008: Modern meteorology and atmospheric icing. Chapter 1 (pp. 1-29) in “Atmospheric icing of power networks”, M. Farzaneh (ed.), Springer, 381 pp, ISBN 978-1-4020-8530-7. Morrison, H., J. A. Curry, V. I. Khvorostyanov, 2005a: A new double-moment microphysics parameterization for application in cloud and climate models. Part I: Description. J. Atmos. Sci., 62, 1665-1677. Morrison, H., J. A. Curry, M. D. Shupe, and P. Zuidema, 2005b: A new double-moment microphysics parameterization for application in cloud and climate models. Part II: Single-column modeling of Arctic clouds. J. Atmos. Sci., 62, 1678-1693. Thompson, G., R. M. Rasmussen, and K. Manning, 2004: Explicit forecasts of winter precipitation using an improved bulk microphysics scheme. Part I: Description and sensitivity analysis. Mon. Wea. Rev., 132, 519-542. Thompson, G., P. R. Field, R. M. Rasmussen, and W. D. Hall, 2008: Explicit forecasts of winter precipitation using an improved bulk microphysics scheme. Part II: Implementation of a snow parameterization. Mon. Wea. Rev., 136, 5095-5115.