Print Version

Chapter VIII: Real and simulated WV imagery

Table of Contents


Simulated satellite images are produced by ECMWF on a regular basis out to five days. These images are generated from the model forecast data using the same radiative transfer algorithm (RTTOV) as in the operational data assimilation.

Simulated satellite images have the same units, format and resolution as the MSG satellite data that forecasters are familiar with. They show WV brightness temperatures as if they were observed from MSG satellites, including the same viewing angle and projection. These images can be easily compared with real IR and WV images for that time.

Despite having the same resolution, real and simulated WV images can be distinguished quite easily. As seen in figure 1, a side-by-side comparison of real and simulated WV 6.2 µm channels shows that the simulated WV image is much smoother and reveals fewer details. This comparison also demonstrates the richness of atmospheric processes depicted in real WV imagery, which model analysis cannot simulate.

before after

Figure 1a and 1b: Simulated (background) and real (overlay) WV 6.2 µm satellite image from 18 February 2013 at 12:00 UTC.

This chapter will focus on three ways in which simulated WV images can deviate from real ones:

  • Displacements
  • Patterns and sub-structures
  • Greyscales

Real and simulated WV images are hardly ever similar. Nevertheless, it is important to identify synoptically relevant differences; small deviations do not need to be taken into consideration.

When differences between observation and model output are noticed by comparing WV imagery, it is important to know that the implications are valid only for the area in question, and most probably only for the upper troposphere. In no way they can be generalized for the whole model domain.

When comparing real and simulated WV images, one should always use the latest model run.

A. Displacements

Features in simulated WV images are displaced relative to those in real WV images when both images show the same pattern but in different geographic positions. This might be an indication that the model physics have basically grasped the synoptic situation, but either not in the correct place or not at the correct time step.

Displacements for example may occur when the centre of an upper level low is located on different positions when comparing both data sources. For a rapidly propagating cold front on the other hand, a time shift may be the manifest reason for differences in the position of the frontal zone. In practical work, the comparison with a WV loop can be of help to discriminate between time shift and misallocation. When the position of the WV feature in simulated WV imagery lies in the path of propagation, the model lagging or being too fast is a more likely explanation for the discrepancies.

Figure 2 shows the typical case of the model lagging behind the real development for a warm front located over the Benelux and western part of Germany. Comparison between MSG WV channel and the corresponding simulated WV image from ECMWF shows the propagation of the warm front much further to the south-east than model data would suggest.

The same time shift is visible for the cold front's passage over southern Ireland. While in reality the cold front cloud band had passed the island by 06:00 UTC, model data still places it over southern Ireland.

before after

Figure 2a and 2b: Simulated (background) and real (overlay) WV 6.2 µm satellite image from 22 December 2012 at 06:00 UTC.

A comparison of real and simulated WV images clearly shows the time shift of the model data compared to the actual situation.

The second example (figure 3) shows a pronounced upper level low (ULL) over western France. The WV satellite image from 18:00 UTC on 11 February 2013 shows the dark centre of the ULL inland near Bordeaux, while the 500 hPa geopotential isolines show a pressure minimum off-shore and more to the north. When comparing the simulated WV image of the same time stamp with model 500 hPa height, a better match is visible.

before after

Figure 3a and 3b: Simulated (background) and real (overlay) WV 6.2 µm satellite image from 11 February 2013 at 18:00 UTC.

B. Structures and sub-structures

In some cases, the model data may not be consistent with observations. This is most likely to happen over data-scarce regions like oceans, or in cases of very rapid developments, which were not taken into account at the model initialisation. The cases that are not visible in simulated WV images (while real WV imagery depicts them clearly) can be divided into two categories:

  • Large structures (e.g. upper level lows)
  • Sub-structures (e.g. frontal waves)

Also, it may be the case that simulated WV imagery shows features that do not exist in real images, as will be shown in the example below.

Example from 13 February 2013, 12:00 UTC

Figure 4a shows a real WV 6.2 µm channel image. A WV-eddy can be seen over the Atlantic at 56N and 38W (upper left image, red arrow). This feature of the upper troposphere cannot be detected in the simulated WV image for the same date (upper right image, figure 4b). This eddy is not contained in the model data of this specific model run. By contrast, the simulated WV image shows a pronounced bulge at the rear side of a cold front (green circle). This bulge is not present in real WV imagery.

The existence of this cloud bulge is also reflected in the model parameters. The comparison of real versus simulated WV imagery, combined with the field of relative vorticity at 300 hPa, shows a bend of the vorticity field where the cloud bulge of the simulated image is located (lower left image, figure 4c). This reflects the internal coherence of the model fields and explains why all other prognostic fields deviate from reality from there on.

a) b)
c) d)

Figure 4a 4d: WV 6.2 µm images from 23 December 2012, 12:00 UTC
a) Real WV 6.2 µm image
b) Simulated WV 6.2 µm image
c) Real WV 6.2 µm image with positive vorticity and isotachs at 300 hPa
d) Simulated WV 6.2 µm image with positive vorticity and isotachs at 300 hPa

Example from 23 December 2012, 12:00 UTC

The WV image from 23 December 2012 (12:00 UTC) depicts a typical situation where a frontal sub-structure is not reflected in the model prognostic fields. An upper wave (UW) has formed over the British Isles, appearing as a light bulge within the cold front (figure 5a, red arrow). There is no such feature in the simulated WV image for the same date; instead, the rear side of the front shows a uniform curvature without extra bulges. Consequently, all other NWP parameters, like relative vorticity at 300 hPa or isotachs, follow the simulated WV image's idealized front line (figure 5c and 5d).

a) b)
c) d)

Figure 5a 5d: WV 6.2 µm images from 23 December 2012, 12:00 UTC
a) Real WV 6.2 µm image
b) Simulated WV 6.2 µm image
c) Real WV 6.2 µm image and positive relative vorticity at 300 hPa
d) Simulated WV 6.2 µm image and positive relative vorticity at 300 hPa

The above example clearly shows the shortcomings of prognostic model fields for frontal substructures compared to the real situation. While the large scale atmospheric currents are well reflected in ECMWF model fields, differences between forecast and actual data become visible in smaller scale features when comparing simulated against real satellite data. Comparing the two data sources in the same format allows a rapid cross-checking of the model’s coherence with observed data.


Find synoptic relevant differences from real VW image in the simulated WV image.


SIMULATED: Click on the image below in the region where you identify synoptic relevant differences:

C. Differences in grey shades

As discussed in the previous chapters, the water vapour content of the upper troposphere mainly depends on two factors: vertical and horizontal transport of humidity. The different shades of black and white in WV images correspond to the humidity content. Deficiencies in the model simulated horizontal WV transport result in differences between real and simulated WV imagery as discussed in section A and B. When vertical humidity transport is not correctly reflected by model physics, this will also result in differences of the grey shades between both data sources.

In summer season, vertical humidity transport is mostly convective. While most models process atmospheric instability quite accurately, they often fail to pin down the precise location of convective vertical transport. This may result in grey shade differences between real and simulated satellite images (figure 6).

before after

Figure 6a and 6b: Simulated (background) and real (overlay) WV 6.2 µm satellite image from 14 July 2011 at 12:00 UTC.

Differences in the images caused by over- or underestimated vertical humidity transport in the model are also frequent in frontal areas. This could be due to left exit regions not being correctly located or the front proving more (or less) active than predicted. The next images show an excellent example of underestimated vertical transport (figure 7a and 7b).

a) b)

Figure 7a and 7b: Real and simulated MSG WV 6.2 µm satellite image from 23 December 2012 at 18:00 UTC. Showalter index (yellow lines) superimposed to simulated image.

Figure 7a and 7b show a comma-like structure formed in the lee of the Alps. Although this comma lies within an anticyclonic ridge in the model, the development of the high cloud structures seen in both IR and WV images indicates cyclogenesis on the lee side of the mountains. Model fields at 18:00 UTC do not resolve this process in its whole extent, just showing some lee cloudiness south of the Alpine crest.