bullet  EUMETSAT   bullet  SAFs   bullet   NWP SAF   bullet  Monitoring Reports   bullet  AMV Monitoring Report  bullet   Feb 2009

Diurnal Variation - Feb 2009

Back to investigations index page

Hovmoeller Conclusions

Conclusions
General
  • The Hovmoeller pressure plots binned by time of day provide a useful tool for detecting diurnal signal in the O-B statistics.
  • The diurnal signal, where present, tends to be more prominent over land than over sea.
  • Temporal resolution of some AMV datasets is not high enough to reliably resolve diurnal pattern (e.g. MTSAT AMVs every 6 hours).
  • Caution is required in interpretting spikes in statistics values where data volumes are low (for example GOES-12 WV TR bias plot).
  • For the more significant features seen in the plots further investigation has been carried out to confirm that the diurnal signal is a systematic diurnal effect recurring on several/all days in the period. In some cases it may be due to an isolated bad event. There may be subtle impacts on the visible and IR 3.8 results due to changes in the geographic coverage as the day-night boundary crosses the area.
Meteosat-9 (centred 0°, local midnight 0 UTC)
  • IR 10.8 NH winds show a signal at 400 hPa (Figure 1). At this level there is a slow (negative) bias of roughly -1 to -2 m/s between 1800 - 0600 UTC. The bias then increases to a fast (positive) bias during daylight hours, eventually peaking around mid-day at +3 m/s which also coincides with a peak in MVD. If we filter by surface type we can see the that the signal is stronger for winds over land which peak at a bias of +5 m/s (Figure 2) and much less discernible over the sea (Figure 3). The feature is also seen, although to a lesser extent, in the Tropics and for the WV 7.3 channel.


    Figure 1. Meteosat-9 IR 10.8 NH, all surface types.


    Figure 2. Meteosat-9 IR 10.8 NH, observations over land only.


    Figure 3. Meteosat-9 IR 10.8 NH, observations over sea only.

    Figure 4 shows the IR 10.8 plot for the NH over land, but filtered by height assignment method. The marked bias is primarily linked with winds assigned CO2 slicing heights.

    We can also plot as a function of latitude to more precisely narrow down the location of the bias (Figure 5). The bias is most prominent between 20N-30N which roughly delimits the central belt of the Sahara and parts of Saudi Arabia. A specific case on 16 February was investigated further. The fast bias relates to AMVs produced by tracking cloud associated with the sub-tropical jet. These fast AMVs (> 60 m/s) are put above 300 hPa at 0 UTC, but closer to 350-450 hPa at 12 UTC. The most probable explanation for the daytime fast bias is that the CO2 slicing approach is putting the AMVs too low during daytime hours (peaking at 12 UTC). This could point towards an inadequate representation of the large diurnal temperature range of the desert surface, in turn leading to erroneous height assignments.



    Figure 4. Meteosat-9 IR 10.8 NH for CO2 slicing (left), EBBT (centre) and WV intercept (right) HA methods over land.




    Figure 5. Meteosat-9 IR 10.8 over land using CO2 slicing HA method for mid-level (left) and high-level (centre) winds. Also, satellite image highlighting 20-30N for approximate Meteosat-9 land coverage (right).




    Figure 6. Visualisation plot showing observation pressure for the QU00 (left) and QU12 (right) runs on the 16th February over a section of the Sahara (-10W, 20E, 20N, 30N).


  • There is no signal for IR10.8 winds in the SH and for the WV 6.2 channel at any latitude.
  • The HRVIS plots show some diurnal variation e.g. fast bias in the morning, which is most pronounced over land. Figure 7 is a similar type of plot but binned by latitude. The most prominent feature is a fast bias and high mean vector difference around 20N, which persists throughout the day. This fast bias is most prominent during the winter months when the jet crosses this area (see Feature 2.7 in the 3rd analysis of the NWP SAF AMV monitoring) and is probably linked to AMVs being assigned too low in the atmosphere. Elsewhere, there is a tendency towards a small fast bias in the morning and a small slow bias in the evening over land, which is also present in the VIS 0.8 plots. The mean vector difference tends to be higher at the beginning of the day.


    Figure 7. Hovmoeller plot for Meteosat-9 HRVIS winds over land binned by latitude.


  • The distribution of IR and WV winds is relatively uniform throughout the day, except in the Tropics where there is a peak in the number of high levels winds and reduced number of mid level winds between ~1200-2000 UTC.
Meteosat-8 rapid scan (centred 9.5°E, local midnight 2320 UTC)
  • IR10.8 over land shows a small increase in bias at 400 hPa and 700 hPa around local midday. However, if we bin the data by latitude the apparent signal which occurs at ~50°N is only evident for a few days around the 15th and is not systematic throughout the month.
  • The VIS 0.8 channel exhibits a small fast bias towards the end of daylight over sea at 17-19 UTC. This is based on very few observations so may not be reliable.
Meteosat-7 (centred 57.5°E, local midnight 2010 UTC)
  • There is a reduced number of IR and WV winds in the 18-19 UTC and 20-21 UTC time bins as shown in Figure 8 for the IR case. This is related to image quality problems near local midnight during the eclipse season (solar stray light entering the radiometer). The outages start on 12 February and continue until the end of March.



    Figure 8. Meteosat-7 NH IR plot showing a reduced number of winds around local midnight.

    We might expect some drop-off in quality of the AMVs due to solar stray light problems near local midnight (see Bormann et al., 2003 (JAM, 42, 1035-1042)). A couple of possible signals are seen in the Hovmoeller plots. (1) The NH IR winds show a larger slow bias and MVD at high levels (~250 hPa) in the two time-slots near local midnight. However, this was found to be linked to a bigger slow bias at the start of the month rather than a problem specifically in those time-slots. (2) The IR SH plot shows poorer statistics in the last time-slot (2230 data). Time-series plots were also examined to check for problems in the time-slots either side of those removed at EUMETSAT (i.e. 1800 UTC and 2230 UTC) and to check before and after the eclipse outages were enforced (i.e. before 12 Feb and beginning of April). No significant impact on the statistics is seen even when no QI filtering is applied. This is encouraging, although it should be noted that it doesn't rule out the possibility of some localised impacts on AMV quality which may be masked in group statistics.
  • There is a peak in the number of SH WV winds around 1100 UTC. This coincides with a slightly reduced bias and MVD around 200-300 hPa.
  • VIS winds show an increased MVD in the Tropics at 750-800 hPa towards the end of daylight.
MTSAT-1R (centred 140°E, local midnight 1440 UTC)
  • No diurnal pattern seen, but limited temporal resolution.
GOES-12 (centred 75°W, local midnight 0500 UTC)
  • No diurnal pattern seen in IR and WV.
  • NH VIS winds show a fast bias and larger MVD around 650hPa at the end of daylight. By contrast the TR VIS winds over sea show an improving bias and MVD around 700 hPa as the day progresses.
GOES-11 (centred 135°W, local midnight 0900 UTC)
  • No diurnal pattern seen, but limited temporal resolution (every 3 hours, but some additional gaps due to delays in data receipt because of limited capacity on the Washington GTS link).