home

ELDORADO: the  surface radiation system

Dirk Meetschen, Bart van den Hurk, Felix Ament and Matthias Drusch
last update: 17 September 2002

System overview

The ELDAS radiation system ELDORADO is designed to create databases of surface downwelling longwave and shortwave radiation from a combined use of radiative transfer code from the ECMWF global model and METEOSAT observations. A future version of ELDORADO will make use of Meteosat Second Generation.

Surface downward radiation fluxes are created by the ECMWF radiative transfer code applied to given atmospheric profiles of temperature, humidity, cloud cover and liquid and ice water content, for a given land surface albedo. In a number of consecutive steps the profiles of cloud cover and liquid water content are modified, in order to optimize the correspondence between (a) modelled Top-of-Atmosphere (TOA) net shortwave radiation and METEOSAT observations, and (b) modelled total cloud cover and the cloud cover derived from METEOSAT using the MetClock algorithm (Feijt and de Valk, 2001). It is assumed that the radiative transfer code provides an accurate estimate of the surface radiation fluxes provided it is forced to reproduce the TOA quantities. This assumption is only partially valid, and further validation of the fluxes in the near future has yet to be carried out. 

ELDORADO is currently being applied to generate fluxes for the European area at 3-hourly interval between 1 October 1999 and 31 December 2000. Below is a more detailed description of the steps taken in ELDORADO, and a brief example showing the main features is included.

System components and processing steps

ELDORADO contains a series of steps: The main components of the system are the ECMWF analysis archive, a limited area model carrying the ECMWF radiative transfer code (RACMO), METEOSAT radiance measurements, and MetClock cloud cover data. METEOSAT and MetClock data are not available above 67 degrees North and the area east of the line connecting (35 N, 22 E) - (67 N, 40 E). This excludes the South-east corner of the ELDAS domain from use of ELDORADO radiation data.

Creation of background cloud fields

The ECMWF analysis archive contains fields of water vapour and temperature at 6-hrly time intervals, but no analyses of the prognostic quantities cloud cover and liquid water content. These quantities are created/modified in the model code as a result of advection and cloud formation processes. To create a first guess of cloud cover and liquid water profile, short-term forecasts of RACMO are started from interpolated ECMWF analyses every 6 hours, and cloud cover and liquid water profiles are passed between these forecast cycles as passive (non-analysed) variables. To save computer recourses, this short term forecast cycle is operated on a rotated lat-lon grid with 0.4 degree resolution. At 3-hourly intervals atmospheric fields including cloud cover and liquid water profiles are interpolated from this forecast cycle to the finer ELDORADO resolution and archived.

First guess radiative transfer

From the interpolated atmospheric fields a single time step is integrated with the RACMO model containing the ECMWF radiative transfer package from the parameterization scheme operated in ERA40 (cycle 23R4). The model is run at a rotated lat-lon grid at spatial resolution of 0.2 degrees (31 vertical levels). The spatial cover of the domain is not identical to the ELDAS grid, to avoid significant redundance in high latitude areas. A monthly climatology or tropospheric aerosols is prescribed.

From this first guess radiative transfer calculation the TOA radiation fluxes and total cloud cover are archived.

Recalibration of METEOSAT radiances

METEOSAT radiances are first converted into broadband shortwave hemispherically integrated fluxes following Arino et al (1992). This accounts for spectral and angular integration of radiances into an energy flux. However, large uncertainties remain present, in particular with respect to the spectral response of the underlying land surface, the poor angular sampling of the METEOSAT sensor, and distortion by the sub-optimal viewing geometry of high latitude areas. As a pragmatic solution to this, a manual recalibration of METEOSAT reflectance is carried out by comparing first guess output from RACMO with METEOSAT data for grid boxes where both RACMO and MetClock assume cloudfree conditions. The average ratio between RACMO and METEOSAT TOA reflectance is calculated per month and per UTC time interval. Monthly maps of this ratio are used to recalibrate the cloud-free METEOSAT reflectance, and these correction factors (thus specified per time interval, per month and per grid box) are applied for all days in the corresponding month. Note that the influence of cloud cover on the calibration is not taken into account. Correction factors are usually between 1 and 1.1 for noon over land areas, and around 1 over oceans.

Relocation of grid boxes

The next step is to shuffle the cloud field in the first guess simulation modestly, in order to continue the optimization using background cloud profiles with radiative properties already closely matching METEOSAT observations. METEOSAT radiances are transferred into (recalibrated) fluxes as described above, and interpolated to the ELDORADO grid. For each grid point location, a search is carried out in 21 times 21 grid boxes surrounding the target. Foreach of these 441 grid points a cost function is constructed, measuring:
  • the distance to the target grid point 
  • the difference between the METEOSAT TOA reflectance at the target location and the RACMO TOA reflectance 
  • the difference between the surface albedo of the target grid point and the albedo of the RACMO gridpoint 
This cost function is used to find the grid point in the direct neighborhood matching the METEOSAT reflectance closest, but within an acceptable spatial range. The cloud cover and liquid water profile of the best matching grid point is copied to the target location. This procedure allows to create or remove clouds in grid points that were originally cloud free or cloud, respectively.

Optimizing cloud cover

RACMO total cloud cover is calculated from a cloud cover profile using a maximum/random overlap assumption (maximum overlap in neighbouring cloudy levels, random overlap between levels with cloud free separation in between). The total cloud cover in RACMO is adjusted to create a total cloud cover that is compatible with the MetClock cloud cover product. MetClock calculates a cloud mask by using thresholding techniques involving METEOSAT visible and infrared channels. Temperature thresholds are evaluated against first guess surface temperature estimates from the HIRLAM forecast model (see Feijt and de Valk, 2001 for details). For each ELDORADO grid box an average MetClock cloud cover estimate is constructed from 15 surrounding MetClock pixels (compatible with the METEOSAT resolution).

The RACMO cloud cover profile is scaled with a factor (uniform with height) that results in a total cloud cover closest to the MetClock product. This scaling factor is found by recalculating the total cloud cover corresponding to a perturbed cloud cover profile (90% of the original cloud cover in each layer), and linearly interpolating between the original and perturbed cloud cover results. Tests have shown that a single linear interpolation is usually sufficient to find the MetClock cloud cover. The non-linear relation between scaling factor and total cloud cover is usually small enough so that a second iteration can be avoided. 

Optimizing liquid water content

The vertical distribution of the first guess liquid water profile is not altered in ELDORADO. But the total liquid water content is modified by a scaling factor similar to the cloud cover optimization. Here the scaling factor is determined by minimizing the difference between RACMO en METEOSAT TOA reflectance, the same quantity that was used for the gridbox relocation procedure. For all gridboxes where liquid water is present somewhere in the column and where TOA net radiation exceeds a threshold of 50 W/m2, a linear minimization is performed by interpolating between the original and a perturbed liquid water profile. The relation between the scaling factor and TOA reflectance is fairly non-linear in a wide range of liquid water content, and thus a second iteration is applied in this step. 

Postprocessing

A final postprocessing step is the storage of the surface downward fluxes that result from the RACMO calculations with modified cloud cover and liquid water profiles. The fluxes are interpolated to the ELDAS grid, and optionally a flag will be added to indicate the origin of the fluxes. This may either be the first guess from RACMO (in case no METEOSAT and MetClock data were available), or the product of any step in between, depending on the availability of satellite data. 

Examples

Convergence of TOA fluxes and cloud cover

Shown is an example for 5 October 1999, 12 UTC. The following links display cloud cover and TOA fluxes maps for  The example clearly shows that TOA variables can be successfully reproduced with the ELDORADO system, exept for cloud free areas. There the RACMO profiles will not be modified and biases already present in the first guess remain present. Furthermore, the purpose of ELDORADO is to generate surface fluxes instead of TOA fluxes, and the quality of these is briefly explored in the next example.

Time series of surface radiation in The Netherlands

The routine surface radiation network in The Netherlands was used to carry out a brief surface radiation validation. Shown is a time series of daily radiation bias and rms score of both the first guess suite and the ELDORADO final result. It is clearly seen that
  • The ELDORADO suite results in a strong reduction of both the bias and rms score of downward surface radiation in The Netherlands. For instance, the average bias in April/May is reduced from 37 to 11 W/m2 
  • The first guess bias is maybe affected by a bug in the operation of the ELDORADO system, causing the cloud cover and liquid water fields erroneously to be reset to zero values at 0 UTC every day. This may induce too low cloud cover in the morning due to a small spin-up, giving rise to overestimated surface radiation 
  • Some spurious events in the first guess are not all removed by the ELDORADO system. For instance the strong bias in early May 2000 is still present in the ELDORADO suite. In this particular case low MetClock cloud cover is combined with low values of observed surface radiation. Other strong peaks are caused by absence of satellite data, or failure to move appropriate cloudy/cloud-free model gridboxes to the target locations. Systematic analysis of similar cases has yet to be carried out. 
Further systematic bias may be induced by 
  • errors in the (re)calibration of the METEOSAT data 
  • errors in the assumed radiative transfer, due to wrongly specified aerosols or cloud optical properties 
  • sampling errors induced by the 3hrly time step of the ELDORADO system as opposed to the continuous sampling of the surface network.