Impact of ERS scatterometer data in the ECMWF 4D-Var assimilation system. Preliminary studies

Abstract

ERS scatterometer data have been assimilated operationally in the ECMWF forecasting system since January 1996. The current data assimilation system makes use of a three-dimensional variational analysis scheme (3D-Var). The impact of scatterometer ocean surface winds, assimilated using 3D-Var, on the forecasts remains limited although larger impacts are found than in the previously operational optimal interpolation analysis. It seems that the fixed error structure assumed by the 3D-Var scheme, especially in the vertical, does not allow the effect of the surface wind observations to propagate up through the atmosphere to upper levels. ECMWF is currently developing a 4D-Var assimilation scheme which extends the 3D-Var by including time as another dimension in the minimisation. This means that observations can be used closer to measurement time, and the structure functions can change dynamically, consistent with the meteorological conditions. It is expected that scatterometer data in particular will have more of a positive impact with 4D-Var. Results will be presented that shows the positive impact of ERS scatterometer data within the 4D-Var assimilation system. This is proven by parallel analysis and forecast cycles carried out with or without scatterometer data. Comparison with the operational runs will also be presented.

Keywords: Variational data assimilation, impact of ERS scatterometer data

Introduction

We will first give a short presentation of the variational method with the focus on difference between 3D-Var and 4D-Var. We will show some results from 3D-Var observation system impact studies related to ERS scatterometer data usage, but the main topic will be to investigate the impact of ERS scatterometer data in 4D-Var.

Variational data assimilation

Figure 1: Data assimilation at ECMWF.

A description of ECMWF's variational code can be found in Courtier et al. 1997 . It is outside the scope of this paper to explain the methods and strategies, here we will shortly mention the differences between 3D-Var and 4D-Var. In 4D-Var a time integration over a six hour period around the synoptic time is performed in order to apply information from observations at the appropriate time, and in order to make the model dynamics and physics devevop vertical structures influenced by the information from the observations.
It has been expected that especially an observation system like ERS scatterometer data could be used more properly in a 4D-Var system because the 10 meter winds can vary quite substantially during a 6 hour period (see Stoffelen and Anderson 1997 ). In 3D-Var all ERS scatterometer data for the 6 hour are treated as being observed at synoptic time, whereas is 4D-Var they are sorted and used in one hourly time slots. The vertical structure functions in 4D-Var can tilt dependent on the state of the atmosphere which should increase the possibility that ERS scatterometer wind information is propagated in the vertical. To really make a substantial impact on the state of the atmosphere it is necessary to impact the middle to high troposphere and not only the layers near the surface.

Figure 2 shows the observations (relevant to the assimilation system) received at ECMWF on a normal day. Surface observations are available over land (SYNOP), from ships (SHIP), and from drifting and moored buoys. Commercial aircrafts measure wind and temperature at approximately 200 hPa level (AIREP, AMDAT, and ACAR). They are an important data source over U.S.A., the North Pacific, and the North Atlantic. Radiosondes (TEMP) and PILOTs are invaluable accurate sources for vertical profiles of wind, temperature and humidity. To some extent profilers have taken over the role of radiosondes and pilots in U.S.A. The remaining observation systems are all satellite-borne. Radiances from polar-orbiting NOAA satellites gives information about the temperature and humidity profiles in the atmosphere (TOVS and SATEM). Geo-stationary satellites are used to track cloud motion which gives information about wind at up to three levels in the troposphere. Finally we have ERS-scatterometer data.

Figure 2: Observations received at ECMWF on 28 April 1996.

Not all the available observations are presented to the analysis system, a thinning is performed to have observations at approximately the same resolution as the analysis scheme. For the ERS scatterometer data around 30000 winds per day are used.
The analysis scheme performs a so-called screening and quality control of data. Observations are compared with the six hour first guess values and grossly wrong values are rejected. For ERS scatterometer data we reject winds with wind-speed higher than 20 m/s, because the CMOD4 transfer function ( Stoffelen and Anderson 1995 ) has only been properly bias corrected in this range. ERS scatterometer data is also rejected if we suspect sea ice is present on the ocean.

Impact of ERS scatterometer data in 3D-Var

Figure 3a: 1000 hPa root mean square errors for a range of 3D-Var observation system impact studies verified against operational analysis. See text for details.

Figure 3b: 500 hPa root mean square errors for a range of 3D-Var observation system impact studies verified against operational analysis. See text for details.

We will summarize the conclusions from these experiments. The impact of removing ERS data (solid versus dashed) is small. On the Northern Hemisphere no impact is seen, but a small impact can be found on the southern hemisphere at 1000 hPa (see figure 3a ) and 500 hPa ( figure 3b ) . When SATOB/SATEM is removed scores are much worse on the Southern Hemisphere for all levels and on the Northern hemisphere at high levels (100 hPa) (not shown). The different behaviour on the Northern and Southern hemisphere reflects the lower observation data density on the Southern Hemisphere. Comparing the no SATEM/SATOB runs with and without ERS data (dashed-dotted versus dotted) no impact is seen on the Northern Hemisphere but a big improvement in scores is found on the Southern Hemisphere. This shows that the 3D-Var analysis system is able to utilize the ERS scatterometer data to improve the analysis and subsequent forecasts in data sparse areas. Previous observation system experiments have shown a similar redundancy among different observation systems. The positive message is that scatterometer data can be utilized to improve the analysis so a greater coverage of the oceans with scatterometer data is expected to be beneficial for analyses and forecasts.

Impact of ERS scatterometer data in 4D-Var

Figure 4a: Average root mean square errors for 1000 hPa geopotential height for Northern and Southern Hemisphere during a 14 days winter period. The curves show 3D-Var and 4D-Var with and without ERS-1 scatterometer data.

Figure 4b: Average root mean square errors for 500 hPa geopotential height for Northern and Southern Hemisphere during a 14 days winter period. The curves show 3D-Var and 4D-Var with and without ERS-1 scatterometer data.

Figure 4c: Average root mean square errors for 200 hPa geopotential height for Northern and Southern Hemisphere during a 14 days winter period. The curves show 3D-Var and 4D-Var with and without ERS-1 scatterometer data.

The main conclusions from these experiments will be summarized here. The direct impact of ERS-scatterometer data in 4D-Var can be seen by comparing solid lines with dashed lines in figures 4a , 4b , 4c . The most pronounced result is the large improvement of 4D-Var performance over the Northern Hemisphere when ERS-scatterometer data is used in the assimilation system. The impact is biggest at 1000 hPa, but almost as big at 500 hPa, and slightly smaller at 200 hPa. It is seen that the improvements in the upper part of the troposphere comes one day later than the impact near the surface. There is a nice degree of consistency in the ERS-scatterometer impact where many of the individual days are improved (not shown). On the Southern hemisphere there is neither a positive nor a negative impact of ERS-scatterometer data in average for the assimilation period.

The impact of ERS-scatterometer data on the 3D-Var assimilations is most cases neutral. The only non neutral impact is day 5 to day 6 over the Northern Hemisphere where we see a small positive impact of he ERS-scatterometer data. This can wholly be attributed to the North Pacific area scores (see figure 7 ).

From figure 4 it is also possible to compare 3D-Var performance with 4D-Var performance even though this is not the main topic of this paper. For the assimilation period 4D-Var was better than 3D-Var on the Northern Hemisphere and slightly worse on the Southern Hemisphere.

Figure 5a shows the average RMS difference in the 3D-Var analyses of the 500 hPa geopotential height field, respectively with and without ERS data for the Northern Hemisphere. Figure 5b shows the similar plot for 4D-Var.

Figure 5a: Root mean square difference between 3D-Var analyses for assimilations with and without ERS-1 scatterometer data during a 14 days winter period. The fields are 500 hPa geopotential height.

Figure 5b: Root mean square difference between 4D-Var analyses for assimilations with and without ERS-1 scatterometer data during a 14 days winter period. The fields are 500 hPa geopotential height.

As can be seen from the figures there is no big difference between analysis differences for 3D-Var and 4D-Var. The both show the biggest impact in North Pacific, and to a lesser extent in the Atlantic. In the Atlantic 4D-Var has a slightly bigger impact than 3D-Var. It is usually not a good thing to have large analysis increments, it is best if the assimilation sticks to the right track and only requires small adjustments every 6 hours.

Figure 6a and figure 6b show the average RMS difference for 5 day forecasts from the data assimilation cycles performed with or without ERS scatterometer data. Figure 6a is for the 3D-Var case and figure 6b is for the 4D-Var case.

Figure 6a: Root mean square difference between 3D-Var 120 hour forecasts for assimilations with and without ERS-1 scatterometer data during a 14 days winter period. The fields are 500 hPa geopotential height.

Figure 6b: Root mean square difference between 4D-Var 120 hour forecasts for assimilations with and without ERS-1 scatterometer data during a 14 days winter period. The fields are 500 hPa geopotential height.

Comparing figure 6a and figure 6b one can see a marked difference where the impact of ERS-scatterometer data is bigger in 4D-Var. In 4D-Var the impact is bigger over the North Pacific than in 3D-Var, but more striking is the influence over North America and along Greenlands east coast in the North Atlantic. It seems like the ERS analysis impact over the North Pacific is propagated by the forecasts over North America and then west of Island along the coast of Greenland. This is supported by an investigation of the atmospheric flow during the two week assimilation period (not shown). We had zonal circulation over U.S.A. and a ridge over Europe for the main part of the period. It is believed that the ERS observations from the Atlantic is of less importance for this assimilation period. Looking on verification scores for specific areas on the Northern Hemisphere (not shown) supports the theory that lows (often intense) developing in the North Pacific in the area south of Japan are better described and "seen" earlier by the assimilation system when ERS-scatterometer winds are available. The ERS-scatterometer coverage is quite good in the North Pacific Area. The analyses and short range forecasts over the North Pacific Area are therefore improved, and this positively affects the medium-range forecasts over U.S.A. and the medium to long range forecasts for the North Atlantic area. If the prevailing circulation pattern would have been more zonal the impact over Europe is expected to have been higher. In this assimilation period there is a neutral ERS impact over Europe. In 3D-Var the impact is generally smaller and does not seem to survive or penetrate as far as is the case for 4D-Var.

In figure 7 we show the scores for the North Pacific area for both 3D-Var and 4D-Var for 1000 hPa and 500 hPa geopotential height. This is the only area where ERS-scatterometer data has a measurable impact on 3D-Var. In 4D-Var the impact of ERS-data is more pronounced, the improvements starts already after day 2-3 and becomes greater for the following forecast days.

Figure 7: Average RMSE for 14 winter forecasts with initial data from 3D-Var and 4D-Var data assimilation experiments with and without use of ERS-scatterometer data. Scores for the North Pacific area verified against operational analyses.

It is seen in figure 7 that the impact is almost the same at 500 hPa and 1000 hPa. For this area this is true for both 3D-Var and 4D-Var. As shown in 4a and 4b it is not generally true that 3D-Var propagates information as well as 4D-Var in the vertical when we look on the Northern Hemisphere area. This supports the view that ERS-scatterometer analysis impact remain more local in the forecasts than what is seen from the 4D-Var results.

A case study

Figure 8: RMSE for the 12 UTC 24 January 1996 forecast with initial data from 4D-Var data assimilation experiments with and without use of ERS-scatterometer data. Scores are for the Southern Hemisphere (20S-90S) verified against operational analyses.

Both at 1000 hPa and 500 hPa the forecasts are improved already from day 2 an onwards. In the medium-range (day 3 to 7) a significantly better forecast is achieved when ERS-scatterometer data is used. This case is specifically selected with the aim to investigate why the ERS-data improves the forecast.

Figure 9a shows the ERS-1 scatterometer data used in the 12 UTC 24 January 1996 analysis. It represents ERS-data from a 6 hour time window around the analysis time. Only observations over ice-free ocean are used.

Figure 9a: ERS-scatterometer data used in the data assimilation on 24 January 1996.

For this case we have more than 4000 ERS-winds. On the average we have 6000 to 7000 ERS "wind observations" per analysis cycle. From figure 9 it is clear that the data coverage is rather limited due to the rather narrow ERS-swaths. It should also be kept in mind that only one level of wind information near the surface is obtained. Nevertheless the scatterometer is a valuable data source which hopefully has become clear from this paper.

In figure 9b we have zoomed in on an area between South America, South Africa and Antactic to show the measured ERS-1 winds based on the CMOD4 transfer function. The analysed mean sea level pressure from the assimilation using ERS-data shows that scatterometer winds really are taken into account when determining the analysis near the surface. In this case two cyclonic circulations are identified with associated troughs and ridges. Also the ERS observations near the South American coast are influencing the analysis.

Figure 9b: Zoom of figure 9a for an area between South America and Africa. The mean sea level pressure analysis from the 4D-Var assimilation with ERS-scatterometer data is also shown.

Figure 10 shows the difference between the analysis for the case study with and without ERS-scatterometer data for four pressure levels in the atmosphere. At 1000 hPa the impact of the scatterometer data is clearly visible. The feature south of Africa can be related to ERS-data from the two previous analysis cycles (not shown). In figure 10 it is seen how the surface data information is propagated in the vertical, even as high as 200 hPa. This is an important feature of 4D-Var which makes it possible to influence the forecasts in a more persistent manner.

Figure 10: Analysis differences for 4D-Var assimilations with and without ERS-scatterometer data for the 24 January 1996. Geopotential height differences for four pressure levels are shown.

We will now look on the forecasts run from 24 January 1996 based on assimilations with or without ERS-scatterometer data. We focus on 500 hPa geopotential height even though a similar behaviour can be seen at other pressure levels. Figure 11a, 11b, and 11c plots the 500 hPa geopotential height contours for the "with ERS" forecasts. In the plots we have also included the difference between the forecasts run from analyses with or without ERS-scatterometer data. The analysis and 1 to 5 day forecasts are shown.

Figure 11a: 24 January 1996 Southern Hemisphere case study.
Top : Difference between analyses based on initial fields from 4D-Var data assimilations with and without ERS-scatterometer data. Bottom : As top but for 24 hour forecasts.

Figure 11b: As 11a but for 48 hour and 72 hour forecasts.

Figure 11c: As 11a but for 96 hour and 120 hour forecasts.

The initial difference in figure 11a is just another way of displaying the results shown in figure 10 . In figure 11a,b,c it is possible to follow how the initial ERS-data induced disturbance develops into a major difference after four forecast days. From the verification scores shown in figure 8 it is obvious that the forecast using ERS-data is better, implying that the developments shown in figure 11 better follows what happened in reality. To investigate this in more detail we have looked at the operational analyses for 500 hPa geopotential height and the conventional observation available in the case study area.

Figure 12a shows the operational 500 hPa analysis for 12 UTC 26 January 1996 for the case study area with TEMP and PILOT height and wind observations used in the analysis. Figures 12b , 12c , and 12d show similar plots for the next three days.

Figure 12a: Operational 500 hPa analysis for 12 UTC 26 January 1996 for an area between South Africa, Australia and Antactic. TEMP and PILOT height and wind observations used in the analysis are shown. Height contours are in units of decameter, observed heights are shown in bold and in units of meter.

Figure 12b: As figure 12a but for 12 UTC 27 January 1996.

Figure 12c: As figure 12a but for 12 UTC 28 January 1996.

Figure 12d: As figure 12a but for 12 UTC 29 January 1996.

It is clear from these analyses that a disturbance similar to the one forecasted in the "with ERS" forecast really did exist in the analyses and was based on observational evidence.

In figure 13 we have shown the 5 day mean sea level pressure forecast from the "with ERS-1" 4D-Var analysis. We have overlaid the surface observations used by the analysis at the validation time, i.e. 12 UTC 29 January 1996. Even though the forecast is not perfect the low pressure system is clearly visible from the observed data, another indication that we can classify the "with ERS-1" 4D-Var analysis was good.

Figure 13: 120 hour forecast from an 4D-Var analysis where ERS-1 scatterometer data has been assimilated. Mean sea level pressure observations from SYNOP, DRIBU, and SHIP for the validation time ( 12 UTC 29 January 1996) are shown.

Conclusions

Data assimilation experiments with ECMWF's 3D-Var and 4D-Var systems have been performed for a two week period in January 1996. To study the impact of ERS-scatterometer data assimilation experiments with or without ERS-data were conducted. The conclusions are that ERS-scatterometer improves the analyses and forecast quite a lot over the Northern Hemisphere especially on the 3-8 day forecast range. The similar studies with the 3D-Var system does not show any real impact on the Northern Hemisphere. On the Southern Hemisphere there in no real impact of ERS-data in this assimilation experiment in either 4D-Var or 3D-Var. This is partly attributed to the fact that it is summer on the Southern Hemisphere causes weaker gradients and fewer intense cyclones.

Acknowledgement

References

Courtier, P., Andersson, E., Heckley, W., Pailleux, J., Vasiljevic, D., Hamrud, M., Hollingsworth, A., Rabier, F. and Fisher, M. , 1997

The ECMWF implementation of three dimensional variational assimilation (3D-Var). Part I: Formulation. Submitted to Q. J. R. Meteorol. Soc.

Stoffelen, A. and Anderson, D., 1997

Ambiguity removal and assimilation of scatterometer data. Q. J. R. Meteorol. Soc., 123, pp. 491-518

Stoffelen, A. and Anderson, D., 1995

The ECMWF contribution to the characterization, interpretation, calibration and validation of ERS-1 scatterometer backscatter measurements and their use in numerical weather prediction models. ECMWF contract rep. to ESA (9097/90/NL/BI). Available from ESA Publ. Div. ESTC, Noordwijk, The Netherlands

Keywords: ESA European Space Agency - Agence spatiale europeenne, observation de la terre, earth observation, satellite remote sensing, teledetection, geophysique, altimetrie, radar, chimique atmospherique, geophysics, altimetry, radar, atmospheric chemistry