a. General

The ERA was carried out with ECMWF’s Integrated Forecasting System (IFS) as it was used operationally for most of 1995. An extensive description of that system can be found in Gibson et al. (1997). Its main characteristics are the following:

• intermittent optimum interpolation analysis with 6-h cycling (ECMWF 1992)

• one-dimensional variational (1DVAR) physical retrieval of TOVS cloud cleared radiances (Eyre et al. 1993)

• diabatic, nonlinear normal mode initialization (Wergen 1988).

The model used to determine the background fields in the data assimilation is the ECMWF model cycle CY13R4. Its main characteristics can be summarized as follows:

• global spectral model at resolution T106

• equivalent reduced Gaussian grid of 1.1256° maximum resolution (Hortal and Simmons 1991)

• 31 vertical eta levels (Simmons and Strufing 1981)

• semi-Lagrangian advection scheme (Ritchie et al. 1995)

• model physics

  1. —radiation scheme (Morcrette 1990, 1991)

  2. planetary boundary layer scheme (Louis et al. 1982;Beljaars and Holstlag 1991; Beljaars and Betts 1993; Beljaars 1995a,b)

  3. four-layer land surface scheme (Viterbo and Beljaars 1995)

  4. subgrid-scale orography scheme (Lott and Miller 1997)

  5. mass-flux convection scheme with moisture convergence closure (Tiedtke 1989)

  6. fully prognostic cloud scheme (Tiedtke 1993; Jakob 1994).

b. Clouds in the analysis system

The cloud parameterization used in the ERA system is described in detail in Tiedtke (1993) with further changes discussed in Jakob (1994). The scheme predicts both cloud cover a and cloud condensate l (water/ice) with prognostic equations; that is,

View Expanded

and

View Expanded

where S_a,l and D_a,l are the source and dissipation terms for cloud cover and condensate, respectively. Sources for cloud cover and condensate in the scheme arise from convection, boundary layer turbulence in case of stratocumulus clouds, large-scale lifting, and diabatic cooling. The dissipation terms are determined by evaporation processes due to subsiding motions including both large-scale and cumulus-induced subsidence, diabatic heating, turbulent erosion processes at both cloud top and cloud sides, and in the case of condensate by precipitation processes. The prognostic treatment of cloud cover and the tight coupling of a variety of physical processes to cloud production and destruction are unique and the application of a scheme of that type in long-term data assimilation is unprecedented. It is therefore of great general interest to assess its performance in this context.

The use of prognostic equations for the cloud quantities automatically implies a need for the determination of their initial conditions. The simplest method is to use zero as initial cloud values and let the model develop the clouds in the course of the forecast. However, as shown in Fig. 2, the full development of the global cloud fields requires 10–12 h of forecast time. Although this might be acceptable for medium-range forecasts, it is certainly not for the 6-h forecasts used to provide the first guess for the analysis system. In those the cloud fields would never fully develop and therefore related variables, especially radiative fluxes throughout the atmosphere, would be in serious error (Jakob 1994). Several methods to initialize cloud variables have been proposed (e.g., Huang and Sundqvist 1993). The method adopted at ECMWF is summarized in Fig. 1, which shows a flow diagram of two analysis cycles. At the end of the first guess forecast the two cloud variables are separated from the other prognostic model variables, which undergo changes by both analysis and initialization. The unchanged cloud variables are then passed back into the system as initial conditions for the next first guess forecast. It is worthwhile noting that by using this method the ECMWF system is the only global analysis system and hence the only reanalysis system that uses nonzero initial clouds.

c. Spinup

The main purpose of carrying out analyses with the system described above is to provide initial conditions for numerical weather forecasts. To achieve the best possible forecast, it is necessary to find initial conditions that are closest to the true state of the atmosphere but also in balance with respect to the model equations, in order to avoid the development of fast growing gravity waves, which might quickly destroy the forecast signal. This is usually achieved by applying an initialization procedure, for example, normal-mode initialization (e.g., Wergen 1988), to the analyzed fields. Although this procedure ensures balanced states in a dynamical sense, it is well known that the thermodynamical processes determining the hydrological cycle of the model might undergo a typical evolution in the first few hours of a forecast, which is usually referred to as “model spinup” (Illari 1987).

Because this study uses 6-h forecasts of cloud cover (see section 3), it is necessary to examine whether those are significantly affected by model spinup. The method of finding the initial conditions for clouds outlined above does not in itself guarantee a small spinup because, in the course of the analysis and the initialization, all model fields except the cloud fields undergo modifications. Hence, the balance between the temperature, moisture, and cloud fields that exists at the end of a first guess forecast does not necessarily hold anymore at the beginning of the next forecast. Figure 2 shows the evolution of cloud cover averaged over different geographical regions for an individual 24-h forecast. For comparison the cloud cover of the same forecast starting with zero initial clouds is also shown. It is evident that there is no spinup in cloud cover and hence the use of short-range forecasts of this quantity is justified.

a. ISCCP

ISCCP uses radiances measured by operational weather satellites to deduce a variety of cloud properties. Two channels, 0.6 μm and 11 μm, common to both the geostationary and polar-orbiting satellites are used. A detailed description of the algorithms is given in Rossow and Garder (1993a). From the initial interpretation of the instantaneous radiance fields a variety of different products with varying spatial and temporal resolution are derived (Rossow and Schiffer 1991). In this study the monthly mean cloud cover as provided in the ISCCP-C2 dataset for the months from July 1983 to December 1990 at a horizontal resolution of 2.5° × 2.5° is used.

b. ERA

For the comparison to the satellite data a monthly mean dataset of total cloud cover has been created from the ERA at the same horizontal resolution as ISCCP. Total cloud cover is calculated from the cloud cover predicted for each of the model levels by integrating from the model top using the maximum-random overlap assumption (Morcrette and Fouquart 1986). As shown in section 2, there is no noticable spinup in total cloud cover in the assimilation system. Therefore results of the 6-h first guess forecasts have been used for the comparison, in order to have the best possible representation of the large-scale flow.

a. Global mean

Table 1 shows the average cloud cover for July 1983 to December 1990 for the whole globe and separated into Northern Hemisphere (NH) (90°–20°N), Southern Hemisphere (SH) (20°–90°S), and Tropics (20°N–20°S). The datasets included are the ISCCP-C2 data, ERA, and the operational analysis for which the averaging period is from 1987 to 1990.

The ERA shows very good agreement with the observations with a slight underestimation of global cloudiness mainly resulting from an underestimation of cloud cover in the extratropics; the tropical cloud cover is slightly overestimated. The difference in cloud cover between Northern and Southern Hemispheres is also very well simulated. As a reference the values as given by the operational 24-h forecasts from 1987 to 1990 are included in the table. The then operational diagnostic cloud scheme (Slingo 1987) strongly underestimated cloud cover in all areas and showed no large hemispheric difference.

Figure 3 shows the temporal evolution of the global cloud cover. Again the correspondence between ERA and ISCCP is striking. The interannual variability in cloud cover is well simulated with maxima in 1984 and 1985. However, the ERA cloudiness exhibits a strong annual cycle that is not supported by the observations. The cloudiness in the 1987–1990 operational system shows an even larger annual signal and a strong underestimation of cloudiness. There is a marked increase in cloudiness in the end of the period, which is related to a change to the cloud parameterization in the operational model in 1990. One of the purposes of the reanalysis projects was to eliminate such discontinuities arising from model/assimilation system changes.

b. Zonal means

Figure 4 shows the zonal distribution of cloudiness averaged over the entire period. The first three panels show the annual, July, and January averages. There is fairly good agreement in the Tropics and subtropics, but the extratropical cloud cover between 30° and 60° is underestimated in both hemispheres with an overestimation poleward of 60°. The significance of the differences poleward of 60° is difficult to assess due to difficulties in the ISCCP cloud detection over snow and ice (Rossow and Garder 1993b). Further discussion is therefore concentrated on the areas between 60°N and 60°S.

The fourth panel of Fig. 4 shows the difference in zonal-mean cloud cover between July and January as an estimate of the annual cycle of cloudiness. There is good agreement between ERA and observations except for the NH extratropics, where a summer maximum in cloud cover is not simulated by the analysis. This will be investigated in the next subsection.

Figure 5 shows the time evolution of zonal-mean cloud cover for both ISCCP and ERA. The general distribution of clouds, with high cloud cover in both the tropical and extratropical belts and low cloud cover in the subtropics is well captured in the ERA. The month-by-month variation in cloud cover in the Tropics, which is governed by variations in the location of the intertropical convergence zone (ITCZ) is very well represented, although the overall level of cloud cover is higher than observed. Some interannual variations such as the maximum in cloud cover around 10°S in winter 1986/87, are well captured, although the ERA shows a similar maximum in 1989/90, which is not present in the observations. Subtropical cloudiness appears to be underestimated; seasonal variations are, however, well described with minima in the respective winter season. Extratropical cloudiness is strongly underestimated especially in the bands between 40° and 50°N/S. The annual migration of maximum values of cloudiness from 40°N in late winter to 50°N in summer, which is an annually recurring feature in the NH, is not captured at all in the ERA, which instead shows cloud cover maxima at around 55°N in autumn.

c. Selected areas

To assess the geographic distribution of the cloud cover differences, the spatial distribution of cloud cover averaged over the entire period of the study is shown in Fig. 6. The top two panels show the average cloud cover for ISCCP and ERA, respectively, whereas the bottom panel shows the difference of ERA minus ISCCP. The general spatial patterns agree reasonably well. The underestimation of extratropical cloud cover noted already in the zonal-mean figures is quite evident with the largest errors occuring over the oceans. The good agreement of zonal-mean cloud cover in the subtropics is due to a cancellation of errors, with an underestimation of the cloudiness in the stratocumulus areas off the west coasts of the subtropical continents and an overestimation in the trade cumulus areas. Tropical cloudiness is generally well represented with a slight overestimation over Africa and the Maritime Continent.

To investigate the spatial and temporal evolution of different climatological cloud regimes Hovmoeller diagrams of the mean annual cycle of cloud cover for selected latitude bands are presented. The averaging period is 1984–1990. Cloud cover is averaged over a 10° latitude band and the annual cycle is plotted as a function of longitude. The areas chosen are 50°–40°N representing extratropical cloudiness, 30°–20°N as an example for subtropical cloud regimes, and 10°–0°N as a tropical area.

Figures 7 and 8 show the evolution of cloud cover in the latitude band 50°–40°N for both ISCCP and ERA. The most pronounced feature of the cloudiness at those latitudes is the strong land–sea contrast with continuous cloud cover above 70% over the oceans and considerably lower values over land. There are particularly sharp gradients on the west coast of North America and on the west coast of the Eurasian continent, especially in summer. The land–sea contrast in general is well captured by the ERA although the gradients appear to be weaker mainly due to a general understimation of cloud cover over the oceans by more than 10%, which is further enhanced off the coast of North America in summer and autumn. In the ISCCP there is a pronounced annual cycle in cloudiness over the Pacific with a maximum in summer. This is mainly due to the annual cycle of low stratus clouds (Klein and Hartmann 1993) and is well captured by the ERA. Over the Atlantic the maximum cloud cover occurs in winter. In this region the low stratiform clouds exhibit their summer maximum farther north (Klein and Hartmann 1993), so the influence of baroclinic developments, which peak in winter, dominates the annual cycle. Over western Europe the ERA generally underestimates cloud cover by more than 15% but reproduces well the pronounced annual cycle with a cloud cover minimum in late summer. Over central Eurasia a very strong annual cycle exists with a distinct summer maximum in the observations. Although ERA captures the low winter cloud amounts in this region very well, the summer maximum is underestimated by up to 20%. This behavior could be explained by a lack of convective clouds over land in summer, which form the bulk of the cloudiness for those regions. This may include both fair-weather shallow cumuli and deep convective systems. Over the North American continent the predicted cloud cover matches the observed better although the winter maximum over the eastern part of the region is stronger in the ERA than in ISCCP. Care has to be taken in the interpretation of these differences because of the difficulties in the detection of clouds over snow-covered areas in ISCCP. Revised ISCCP datasets that are currently being produced (Rossow et al. 1996) will include better estimates of cloudiness over snow.

Figures 9 and 10 show the same quantities as Figs. 7 and 8 but for the latitude band 30°–20°N. Those latitudes are dominated by stratocumulus and trade wind cumulus regimes over the oceans, but features like the Indian summer monsoon are also apparent in the observations. Two major deficiencies in the ERA cloudiness are immediately evident. First of all, the trade cumulus regime cloud cover that dominates in summer over both the Pacific and the Atlantic Oceans is overestimated by about 10%–15%. However, the very pronounced continuously large cloud cover off the North American continent, representing mainly stratocumulus clouds, is underestimated in the ERA by about 15%. The cloudiness connected to the Indian summer monsoon is very well simulated with a 10% overestimation of the peak values.

Figures 11 and 12 show the tropical cloud cover evolution averaged over 0°–10°N lat. This region is dominated by the ITCZ for most of the year. Regions of special interest are the western Pacific warm pool, the eastern Pacific, and the African continent. The western Pacific region is characterized by large cloud amounts throughout the year, which ERA generally overestimates by 10%–15%.

The eastern Pacific off the coast of South America is characterized by a strong annual cycle with a marked late summer maximum. Over the continent the annual cycle is of opposite phase with the cloudiness maximum occuring in winter and spring. This leads to a pronounced dipole structure of cloudiness between land and sea, which is most likely caused by annually recurring changes in the Walker circulation. ERA generally captures this dipole structure and the phase of the annual cycles both over land and sea. There is, however, a general overestimation by 10% of the cloudiness minima.

Over the African continent the signal is dominated by the movement of the ITCZ in and out of the region. In summer there is a pronounced maximum in the observations, which is reasonably well captured. Cloud cover over the central part of the continent is, however, overestimated, pointing to too intense convection over land. This is confirmed by a study by Stendel and Arpe (1997), who found an overestimation of precipitation in the ERA over tropical land.

It has been shown so far that the ERA system captures many of the important annual variations in cloud cover. In order to investigate whether the system is also able to capture some of the observed interannual variability, Figs. 13 and 14 show the evolution of the annual-mean cloud cover in the latitude band 0°–10°N from 1984–90. The strongest interannual signal in the Tropics is that of the El Niño phenomenon. A moderate El Niño event occured in 1987 and is clearly visible in the ISCCP observations. The cloud cover maximum normally located in the western Pacific Ocean follows the eastward shift in maximum sea surface temperature (SST) to the date line. The cloud cover in the eastern Pacific is enhanced. This is followed by a cloud cover minimum near the date line in the following year characterized by cold SSTs in the region. Although generally overestimating cloud cover over the western Pacific, the ERA captures both the shift of the cloud cover maximum and the minimum in 1988, indicating that the parameterization schemes that influence the simulation of cloud cover, most prominently the cloud and convection parameterizations, are able to respond reasonably to the SST forcing.

An interesting feature in Fig. 14 is the the continuous reduction in cloud cover over the African continent. This trend has also been noticed in precipitation by Stendel and Arpe (1997). Its reason is unknown since there are no obvious trends in data availability in that region.