Simo Järvenoja and Laura Rontu
Finnish Meteorological Institute
The radiation scheme of HIRLAM, based on [71990Savijärvi] and documented in [51994Sass et al.], was recently updated to derive and use information about the effective radius of the cloud droplets and ice crystals ([101999Wyser et al.]). In the present scheme all clouds are assumed to consist of water droplets with a predefined size of about 10 mm. The revisions are expected to lead to increased transmissitivity of, especially, high and thin clouds. In addition, some coefficients used in the calculation of atmospheric long-wave radiation below cloud layers, were tuned. In this note a comparison of the results of the present and revised schemes, during a ten-day period in August 1999, are reported.
The HIRLAM radiation scheme was designed to be fast, so only one vertical loop is allowed in both the solar (short-wave, SW) and the thermal (long-wave, LW) part. The SW clear-air global flux is obtained by reducing the top-of-the-atmosphere (TOA) horizontal flux by broadband average ozone absorption (350 DU), water vapour absorption (depending on the scaled precipitable water content u), and Rayleigh scattering in the column. Average aerosol, CO2, and O2 effects are also included. In cloudy air the SW flux is reduced by the total cloud transmissivity [^T], also taking into account multiple reflections between cloudbase and the surface. In a partly cloudy column the clear-sky and cloudy sky results are linearly combined.
The cloud SW transmissivity and absorptivity functions are fits to a two-stream 5-band radiative transfer model derived from [81997Savijärvi et al.] and [21993Hu and Stamnes]. The functions depend on the solar zenith angle q and the modified cloud condensate amount [^M] (in g m-2), which is the vertical integral above the level under consideration of cloud condensate content CCC, multiplied by the relation of the cloud cover C to the maximum cloud cover of the whole column Cmax,
Parameters b10 and b13 are functions of effective droplet radius reff diagnosed from cloud condensate content according to empirical formulae, based on [31994Martin et al.] for water clouds and [41995Ou and Liou] for ice clouds. There are two tunable coefficients influencing on reff, different over oceans and continents. Effective radius for the SW calculations is determined as a weighted mean of values above the level under consideration. Values of all b-parameters are given in [101999Wyser et al.].
Clear-air solar heating due to water vapour absorption, a(u), is obtained by vertical flux convergence, fitting ¶a/¶u from the line-by-line-based broadband a(u) curves of [11986Chou]. In and below clouds the clear-air values are reduced by the cloud transmittance [^T]. In clouds there is also extra heating due to cloud drop absorption, represented by the flux convergence of the absorptivity, ¶[^A]/¶p.
The LW part uses a broadband emissivity scheme in a local isothermal approximation. The water vapour line emissivity is a cubic function of logu; continuum, CO2, and O3 effects are added as extra terms. There can be clouds both above and below the layer in question. Cloud mass absorption coefficient ka,x, where the index x stands for water or ice, depends on the effective radius of water droplets and ice crystals, according to an empirical formula from ECHAM 4 (c.f. [91998Savijärvi and Räisänen]). The effective emissivity e of a cloudy layer is
Furthermore, tuning has been made involving 3 coefficients influencing the cooling rate in a cloudy atmosphere ([61999Sass et al.]) The preliminary tuning is based on a limited amount of data proving that the cooling rate in a cloudy atmosphere is slightly excessive in the reference radiation scheme.
Two different model configurations were used in the parallel tests. Tests were carried out with the HIRLAM 4.6.2 reference system, and are shortly described in the following:
Common features of the systems are summarised in the following:
The CBR turbulence scheme and STRACO precipitation scheme were used in both parallel systems. The forecast model with the radiation updates is computationally about 2% more expensive than the reference version. Figure 1 shows the average mean-sea-level pressure (pmsl) for the 10-day test period 3-12 August 1999. The pressure pattern shows a low pressure area west of the British Isles and a weak low in Nortern Europe, while the gradients over Europe are generally very weak. The weather over Europe was rather warm with occasional showers in many places.
The standard verification scores, obtained by using EWGLAM stations, for mean-sea-level pressure (pmsl) and two-metre temperature (T2m) are shown in Fig. 2. Both systems show a clear diurnal variation in the T2m bias. The small negative daytime bias of the reference system (REF) has practically disappeared in the run with the updated radiation scheme (RAD). An interesting detail is, that the bias is corrected more in the forecasts for the second day (36 hours) than in the forecasts for the first day (12 hours). This may be connected with the cloud-radiation interactions in the model.
On the other hand, the positive night-time bias has increased a little in the RAD run when compared to the REF run. This might be connected with the increased daytime heating. The heat is stored in soil and released during night too effectively, due to the well-known too strong heat conductivity in HIRLAM. Also, there might be problems in the calculation of the atmospheric long-wave radiation.
Thus, the RAD run gives higher temperatures at all times of the day. The higher surface temperatures in the RAD run are reflected in the pmsl scores. The small positive bias in pmsl for the REF run is slightly increasing as the forecast length gets longer. This kind of growing trend has almost disappeared in the RAD run.
Figure 3 shows the systematic difference of the daytime two-metre temperature over the integration area between RAD and REF (RAD-REF) forecasts, based on 36 hour forecasts. The difference over oceans is small, because the sea surface temperature determines the T2m there. The RAD run gives higher temperatures over land areas with maximum differences in Central and Eastern Europe being about 1.5oC. Here, also the difference in the downwelling shortwave radiation at the surface is largest, about 100 Wm-2, as shown in Fig. 4. The value depicted is the difference of one-hour mean fluxes around noon. The corresponding maps based on 12 hour forecasts (not shown) indicate smaller differences between the forecasts. The maximum differences in night-time net long-wave radiation fluxes (not shown) are of the order of ± 20 Wm-2.
Figure 5 depicts the systematic difference in the accumulated convective precipitation between RAD and REF (RAD-REF) 48 h forecasts. The RAD run gives slightly more precipitation over some land areas in Europe. There are also some areas with less precipitation. Differences in stratiform precipitation are negligible (not shown) because most of precipitation in this summer period was of convective nature.
The differences in average cloudiness (not shown) over the period are small. The changes in radiation are mainly connected with the variations of the cloudiness, not with the time-mean values. A time-series of the results of the two experiments at a gridpoint over Poland, in the area of the maximum mean differences, gives an example of the variations (Fig. 6). In most cases the two-metre temperatures and corresponding downwelling short-wave fluxes given by RAD are somewhat greater than those given by REF.
The most striking differences are connected with cases where the differences in the total column cloud condensate path (CCP) are significant. Note that the CCP in defined as a gridpoint average, i.e. it includes the influence of the fractional cloud cover at different model levels. In some cases the fluxes differ even when the CCP values are similar, due to the different vertical distributions of cloudiness or different handling of radiative transfer in clouds. In general the experiment RAD seems to produce slightly lower CCP values than REF.
The radiation fluxes and total cloudiness given by RAD and REF forecasts are compared with observations at a Southern Finland observatory Jokioinen (WMO-number 02963) in a typical case depicted in Fig. 7. The 36 hour forecasts are based on the analysis of 1999080800 and results drawn with an interval of one hour. The daytime two-metre temperature is higher in RAD, especially during the second day. The predicted night-time temperature minima are far from observed in both experiments and evidently connected mainly with other than radiative processes (heat flux from the ground).
The predicted total cloudiness is in this case smaller than observed during the first day and greater during the night and the second day. Signs of cloud breaking are seen in the RAD forecasts valid in the evening of the 8th of August. The total column CCP (not shown) is quite similar in both experiments. The short-wave and net radiation fluxes are overestimated during the first day in both of the experiments. During the second day the short-wave flux flux is underestimated mostly by the REF experiment. The net radiation given by RAD is close to the observed values, partly probably because there are too much clouds and, consequently, the downwelling atmospheric long-wave radiation is overestimated by both model versions. Note also that the observed short-wave and net fluxes are based on data from different independent instruments.
The last examples clearly show the importance of the correct cloud prediction for radiation calculations. They also allow to speculate about the mechanisms of cloud-radiation interactions. It is possible that the somewhat increased downwelling short-wave radiation produced by the revised scheme leads to increased surface temperature ® increased convective activity ® increased precipitation ® decreased cloudiness during the later hours of the forecast. In reality the chain may be not that simple. The things become more complicated also because of the excess heat storage/release in ground, which might transfer the effects of increased short-wave heating even to the following day.
The comparison of HIRLAM forecasts using the present and the updated radiation scheme over the period 3-12 Aug, 1999, showed:
The description of the revised Hirlam radiation scheme (Ch. 2) was prepared in cooperation with Hannu Savijärvi for the HIRLAM-4 scientific paper. Thanks to Bent Hansen Sass for comments to the manuscript and information about tuning of the long-wave part of the scheme. The remarks of Klaus Wyser were useful and led to several corrections in the text.
Figure 7: Comparison between RAD (lines with open circles), REF (lines with filled circles) and observations (lines with crosses). Upper left panel: T2m, lower left panel: total cloudiness (fraction 0...1), upper right panel: downwelling short-wave radiation flux at the surface (Wm-2), lower right panel: net radiation flux at the surface (Wm-2). The 36 hour experiments are based on the 00 UTC analysis, 8 August 1999.