Atmospheric Ice
As published in the GEWEX News from February 1998

To a large extent, the climate of Earth is governed by the reservoirs of water on the planet and the exchanges of water and the associated exchanges of heat between these reservoirs. A quantitative understanding of the way water is exchanged between these reservoirs in the form of the so-called hydrological cycle is viewed as crucial for understanding climate and climate change. Processes that govern the way water is exchanged between its three phases in the atmosphere, the smallest of the reservoirs, are particularly critical to understanding climate change. In one way or another these atmospheric processes underpin the important feedback mechanisms that are thought to govern the response of global climate to greenhouse gas forcing. For example, water vapor is the principal greenhouse gas (e.g. Chahine, 1992) and the absorption and emission of infrared radiation by water vapor in air is the mechanism of the water-vapor feedback. Water in the form of precipitation sized particles is an important source of energy fueling circulation systems, is the fundamental supply of fresh water to life on Earth and is an integral part of snow - sea-ice and vegetation albedo feedbacks. Water in the form of cloud droplets significantly modulates the radiative budget of the planet (e.g. Wielicki et al., 1996) leading to poorly understood but perhaps the largest of all climate feedbacks.

The most uncertain contribution to the atmospheric reservoir of water is the portion that resides in the atmosphere in the form of ice. The purpose of this article is provide a number of illustrations of the importance of ice clouds on both climatology and weather forecasts. We do this using the operational model of the European Centre for Medium range Weather Forecasts (ECMWF) and explore the sensitivity of this model to the way ice clouds are parameterized in the model.

Our ability to estimate the amount of water in the atmosphere is limited by a lack of observations, and, for the most part, these observations are only column-integrated quantities. Programs like GVaP (ref here) provide us with reasonable estimates of total column water vapor (perhaps to the level of 5% under certain circumstances). However, we are much less certain about how this water vapor is distributed vertically. This information is unfortunately important for at least two reasons. The presence of low amounts of water vapor high in the atmosphere has a significant influence on the emission of radiation to space and thus a disproportionate influence on water vapor feedback (e.g. Lindzen, 1990). Knowledge of the vertical distribution of water vapor is also required to understand and predict the evolution and maintenance of clouds –knowledge that is an essential ingredient of cloud feedback problem. Our current ability to partition water vapor vertically using present satellite data is limited to a few layers (3-4 at most) and even then accuracies of layered water vapor amounts are no better than 30 % (e.g. Engelen and Stephens, 1998).

Quantitative estimates of condensate, including cloud water, cloud ice and precipitation are also limited to column integrated quantities. Programs like GPCP (ref here) provide us with estimates of global precipitation (at the surface) based on microwave and infrared emission measurements using approaches that are limited in application. Satellite data when composited into monthly mean values are probably no more accurate than 10-20% level (this is the goal of the Tropical Rainfall measurement Mission, Simpson et al., 1988). Cloud liquid water path estimates, on the whole, have been limited to microwave emission measurements over oceans and are only accurate to the 20-30% level when clouds are not precipitating (Greenwald et al., 1995). Global determinations of ice content are even less direct and are the most uncertain of all. Current methods to determine this quantity rely substantially on indirect a priori relationships of one form or another. An example is provided in the work of Lin and Rossow (1996) who use the relation between visible optical depth, ice water path and particle size. The accuracy of this method is not known but estimated to be about a factor of 2. Unfortunately, none of the approved satellite observing systems that are expected to be launched in the coming few years, including operational systems under NPOESS, will provide measures of ice cloud content that are likely to improve on existing crude estimates. Millimetric radar observations also provide information about the ice content of clouds with an accuracy approaching the 30% level when other observations are combined with the radar. Combinations of active and passive observations, like that described in Evans et al. (1997), are perhaps represent the most promising method available for providing quantitative information about the ice content of clouds.

A new cloud parameterization (Tiedtke, 1993) was introduced into the ECMWF operational forecast model in 1995 (Jakob, 1994). The scheme is based on two prognostic equations for cloud cover and cloud condensate (i.e. water and ice) and its introduction has led to a number of improvements in the model including cloud forecasts and model climatology. One of the more sensitive aspects of the new scheme concerns the various assumptions associated with !the treatment of cloud ice. In many models this treatment assumes the fraction of condensate assumed to exist as water (f_l) rather than ice is simply defined by the temperature of the volume of cloud. In reality the relationship between f_l and temperature is very complex. The nature of such relationships has important consequences not only for climate model simulations, as pointed out in the study of Senior and Mitchell (1993), but also has direct influence on forecasts produced by the operational model. The impact of different forms of the f_l – T relationship on a forecast is illustrated in Figs. 1 and 2. Figure 1 shows the 500 hPa analyses for a 4 day period from 11 February 1997 to 14 February 1997. In the sequence a trough can be seen to deepen as it progresses eastward over the eastern USA and into the Atlantic. This is a typical scenario whereby the strong lower tropospheric temperature gradients between the continent and ocean lead to strong synoptic developments.

Fig. 1 The 500 hPa analyses for a 4 day period from 11 February 1997 (top left) to 14 February 1997 (bottom right). In this sequence a trough deepens as it progresses eastward over the eastern USA and into the Atlantic. Note two cross points that define a segment shown bottom left running north-west to south-east from the great lakes of the USA to the Atlantic Ocean.

Figure 1 shows results of forecasts from two versions of the operational model. One version, labeled 'control', uses the new cloud scheme with a relation that assumes f_l=1 at T=0 C and decreases monotonically to f_l=0 at T=-23 C. Forecasts labeled 'mod' are obtained using the new scheme with a form of f_l that treats all clouds colder than 0^oC as being all ice. Shown in the upper two panels is the vertical distribution of clouds forecast by the control version of the model (upper panel), the mod version of the model (middle panel) and the mod-minus-control temperature differences (bottom panel). The cross sections are defined by the latitude and longitude as given and correspond to that segment indicated in Fig. 1 by the two cross-points. Significant amounts of low cloud were forecast over the land area of the segment by the control model. This cloudiness was substantially reduced in the modified forecast as a result of the assumption that more of these clouds contain ice, and individual ice particles being more massive than individual water droplets, fall from the atmosphere faster. This fall-out of condensate reduces the cloud cover over land which in turn substantially impacts the temperature forecasts in that region (bottom panel). The significant result is that the boundary layer of the modified forecast is much warmer than the boundary layer forecast by the control model (by more than 3 degrees). Conversely, the surface temperature is colder for the mod-forecast. Both effects can be attributed to changes associated with the radiative heating by cloud. In the case of the control, a warmer surface temperature is maintained by the presence of the low cloud and the associated emission of longwave radiation to the surface by these clouds. This cloud, in turn, radiatively cools the lower atmosphere and produces significantly colder atmospheric temperatures. The mod-case with warmer boundary layer temperatures significantly reduces the low-level baroclinity and underdevelopment of the trough occurs in the resulting (poor) forecast.

Fig. 2 Vertical cross section of cloud amount for the control (top panel) and mod forecasts (middle panel) the mod-minus-control temperature difference lower panel for section indicated on Fig. 1.

Fig. 3 A profile of relative humidity averaged over DJF 1987/1988 and averaged over the region of the tropical pacific (15N to 5S, 130 to 180E). Shown are Radiosonde observations from TOGA COARE (Lin and Johnson, 1996).

Fig. 3 The solid line is the profile obtained from a new ECMWF prognostic scheme; the dashed line is from the previous version of the model with the diagnostic scheme.

One of the most noticeable changes caused by the introduction of the new cloud scheme was evident in the humidity structure of the tropical atmosphere. Figure 5 is a three month average (DJF 87/88) of the profile of relative humidity over the Western and Central Pacific (15N to 5S, 130 to 180E) taken from T63L31 integrations of the global ECMWF model with the new prognostic cloud scheme. The profile derived with the new scheme is given by the solid blue curve. The difference between this profile and that obtained using the old diagnostic cloud scheme which was operational before April 1995 and shown by the dashed blue curve is substantial. The decrease in relative humidity just above 200 hPa and the increase in a thick layer between 300 and 700 hPa are due to the strong coupling of the clouds to the convection scheme. Instead of evaporating all detrained condensate it is now able to precipitate into lower layers providing a source of water vapor on evaporation increasing the relative humidity in this broad layer. It is worthwhile noting, that the humidity structure below 300 hPa using the new scheme is in much better agreement with TOGA/COARE observations of Lin and Johnson (1996) also shown for reference. Within the climate context, changes of humidity profile of the type shown are profound. This result serves to illustrate an important point – that the water vapor budget of the climatically sensitive region of the upper troposphere is greatly influenced by the amount of condensate (in the form of ice) and the vertical distribution of this condensate. Clearly observations of the ice water content of clouds and the vertical distribution of these clouds are required to verify these more advanced cloud prognostic schemes and model predictions of water vapor.

Another reason upper tropospheric clouds are important to climate is because they are high and thus cold. Because these clouds are cold, they emit much less radiation to space (referred to as outgoing-longwave-radiation, OLR) than either low clouds or the surrounding clear sky. This decreased emission is sometimes described as the greenhouse effect of these clouds and the importance of the enhanced greenhouse effect of these high cold clouds is well recognized. The treatment of ice clouds and the way ice settles in the atmosphere, not only affect the humidity profile, but also determines their ice content (Jakob and Morcrette, 1995). The radiation balance at the top of the model atmosphere (TOA), in turn, is governed by the ice content of clouds and the way ice is distributed in the vertical. Figure 4 shows the difference between longwave radiation emitted to space observed by Earth Radiation Budget Experiment (ERBE, see Wielicki et al., 1996) minus that obtained by the model. For both cases, the OLR is averaged over the DJF 1987/88 season. The upper panel is the difference between ERBE and the model with the diagnostic version of the cloud scheme and the lower two panels show differences between ERBE and the new scheme with different assumptions about the fall speed of ice particles. The results of Fig. 4 show substantial differences between ERBE OLR and model OLR derived with the old scheme. Although the new scheme provides significantly better estimates of the OLR, the degree of agreement is a product of tuning the assumptions of the parameterization.

Fig. 4 The difference between OLR observed by Earth Radiation Budget Experiment (ERBE, see Wielicki et al., 1996) minus that obtained by the model for the three month period DJF 1987/88. The upper panel is the difference between ERBE and the model with the diagnostic version of the cloud scheme and the lower two panels show differences between ERBE and the new scheme with different representation of fall speeds of ice crystals.

Figure 5 is an example of how data from one particular case study has been used to test the cloud scheme. Mace et al., (1998) provide an analysis demonstrating how composite data available from a continuously operating surface radar can also be used to help verify the model forecasts. The case study used to produce Fig. 5 is the cirrus cloud observed on November, 26^th 1991 over Coffeyville, Kansas, as part of the FIRE II IFO (see the FIRE II special issue of J. Atmos. Sci., Dec, 1995 for a number of papers describing results of analyses of FIRE II data). The occurrence of cirrus observed on November 26^th over Coffeyville was well forecast by the operational model using short-range forecasts of the ECMWF model at T213L31 resolution (Klein and Morcrette, 1997). Forecasts were run with two different numerical versions of the ice settling formulation as mentioned above. The upper panel shows time-height radar reflectivity data obtained from the NOAA ERL 35 GHz radar (e.g. Matrosov et al., 1995) converted to ice water content.

The bottom two panels are equivalent time height cross sections of ice water content obtained by the operational model but at a much coarser resolution than the radar data. The difference in ice water content between the two versions is large and matches to radar data of this type help to constrain the parameterization.

Fig. 5 The upper panel is the time-height radar reflectivity data obtained from the NOAA ERL 35 GHz radar (e.g. Matrosov et al., 1995) converted to ice water content. The bottom two panels are equivalent time height cross sections of ice water content obtained by the operational model at a much coarser resolution.

It is evident from the examples shown above how changes in the treatment of clouds in large-scale models profoundly influence quantities that determine the heat and moisture budgets of the planet. In particular, the radiation and moisture budgets are grossly affected by the way ice clouds are treated in models. Unfortunately, our ability to verify whether or not new and physically improved cloud schemes like that now implemented at ECMWF adequately represent real clouds cannot be properly determined at this time. The limited amount of data such as described above are invaluable but even these data lack quantitative accuracy in the sense that ice water content obtained directly from radar data alone has significant uncertainties. However, even these data with the attendant uncertainties if collected from a satellite orbiting Earth would provide a major advancement on our understanding and prediction not only of ice clouds but also of upper tropospheric water vapor. The lack of quantitative global observations of ice content is now deemed to be one of the highest observing needs by the cloud-climate modeling community (WMO, 1995). Sensors that lead to improved methods to observe ice content such as that described by Evans et al. (1997) need to be flown on satellites to obtain observations in climatically critical areas of the globe. Furthermore, maintenance of long-term observations like those now becoming available as part of ARM is an essential part of this observing philosophy. The combination of these surface observations with new observations from satellites, together with their integration in advanced assimilation/forecast systems, represents a sensible approach to advance our knowledge about these complex but important clouds.