The impact of satellite scatterometer data on a global NWP model

Atsushi Nomura and Yoshihiko Tahara^*
Numerical Prediction Division, Japan Meteorological Agency

Performance of a forecast of a operational Numerical Weather Prediction (NWP) model depends not only on the model's ability but also on accuracy of an initial field of the forecast. The initial field is produced by a four dimensional data assimilation (4DDA), whose performance strongly depends on quantity and quality of observations used. If there is no reliable datum over an area, a guess field which is forecasted by the model cannot be corrected; accuracy of the analyzed field over there decreases.

Present global observation network is mainly maintained by conventional observing systems such as synoptic land surface stations, voluntary ship observations, drifting and moored buoys, radiosonde observation networks, and meteorological satellite observations such as cloud wind vectors by geostationary satellite and TOVS radiances by NOAA satellites. The conventional observations are not homogeneous: dense over the land and sparse over the ocean. The problem of data sparseness over the ocean is serious. Lower predictability of forecast models over the southern hemisphere comparing the northern hemisphere is thought to be due to the data sparseness. Satellite observations moderate the situation but their contribution is not enough so far.

Satellite's sea surface observations have a possibility to revise the situation. SSM/I soundings by DMSP satellites have provided sea surface wind speed information. Some studies show benefit of the data (Atlas and Bloom, 1989; Atlas 1992; Nomura 1992). However their impact is small because their information is only wind speed; there is no information for wind direction.

Scatterometer observations provide sea surface wind vectors. Scatterometer data by ESA/ERSs' operationally provides global sea surface wind observations. Impacts of the ERS wind observations on NWP models are confirmed (e.g. Atlas 1982; Duffy and Atlas 1986; Anderson et al 1991; Hoffman 1993) and they are used in operational models in some NWP centers. However its data swath (about 500km) is too narrow to get a remarkable impact on NWP models.

NSCAT (NASA scatterometer onboard NASDA/ADEOS), which also provided sea surface wind observations globally, would have given a great benefit to solve the problem. The swath of NSCAT(600kmx2) is about twice from that of ERS-2 and simultaneous use of ERS and NSCAT data could cover whole globe for 12 hours.

Since some improvements can be expected in operational forecasts by using NSCAT data, JMA had planed to introduce them in the operational NWP system and had been preparing for it. Unfortunately the sudden death of ADEOS brought to the halt of the plan before the actual use. However other similar data, such as SeaWinds by QSCAT and ADEOS-II, are expected to be supplied to JMA near feature. Therefore we have continued developing technologies of the best usage of scatterometer data. This paper presents the results of the studies.

NSCAT data were processed near real time by NOAA/NESDIS and delivered to several operational NWP centers including JMA; they contributed to the operational global meteorological observing system (World Weather Watch).

Its wide data coverage was the most important character of NSCAT for the contribution to the global observing system; NSCAT could cover almost the whole globe for twelve hours' observation. As mentioned in the introduction, data sparseness over the ocean is a major problem for the system. Figure 1(a) shows data coverage over the ocean by conventional observations (voluntary ships, moored and drifting ships and marine platforms) for twelve hours. They are relatively rich along coastlines and major sea-lanes, however many and wide data sparse regions can be seen especially over the southern hemisphere.

ERS observed sea surface wind vectors operationally and delivered to meteorological centers via GTS. However since width of its swath is narrow, as shown in Fig. 1(b), the data coverage is not sufficient. Swaths of NSCAT were about twice as that of ERS. As shown in Fig. 1(c), their data covered almost the whole globe for twelve hours. Areas of no observation in Fig. 1(c) could be reduced by composing other operational data shown in Fig. 1(a) and 1(b). As a result of the composition, data coverage became excellent as shown in Fig. 1(d).

Fig. 1. (a) shows a data coverage map of conventional marine observations during twelve hours (from 21UTC May 13 1997 to 09UTC May 14). They are by voluntary ships, moored and drifting buoys and marine platforms. Total number of observations is 5079. (b) and (c) show data coverage maps of ERS-2 and NSCAT sea surface wind observations respectively. The piriod is the same as that of (a). Total number of ERS-2 data is 70108 and that of NSCAT data is 225967. (d) shows a composite map of (a), (b) and (c).

NSCAT observed wind speed with high accuracy, however ambiguity of wind direction remains. It is a common problem for sea surface wind data by scatterometers, because it comes from the principle of the sounding system. From scatterometers' observations wind speed can be determined as a certain value at an observation cell but there are multi solutions for wind direction. The wind direction can be normally determined through some filtering processes. However it is very difficult to select the proper one for some cases. The problem is serious because (1) differences among the solutions are very large; they are 90 or 180 degrees and (2) the wrong selection of it for a cell is possible to affect succeeding filtering process; all wind directions of the surrounding cells are resolved to be wrong directions.

Figures 2 shows wind vectors by NSCAT over an area of southern hemisphere and an estimated surface pressure field (6 hour forecasts). Wind directions of data in the left plate are selected these of highest priorities among the solutions retrieved by NESDIS. Lower one-third of observations on the left side swath are consistent with the estimated surface pressure field but upper two-third are completely wrong. There can be also recognized in the right side swath that a small data cluster whose wind data are inconsistent against the field. Considering the estimated surface pressure field and the other data, almost all of them are thought to be wrong.

Fig. 2. NSCAT wind data over a part of Indian Ocean (28S - 50S, 45E - 67E). Contours represent estimated surface pressure field. Wind directions are selected the highest priority detemined by NESDIS in the left plate. In the right plate, they are determined considering the estimated surface pressure field.

Wind directions were re-selected to fulfil the consistencies among the estimated surface pressure field and surrounding data and plotted in the right plate of Fig. 2. There remain some doubtful data yet but almost all data are quite well. To use the scatterometer effectively in the data assimilations, this ambiguity removal is essential. If the completely wrong data were used in analyses, quality of the analyzed field would be largely degraded.

NSCAT data represent fine structures of wind field over the ocean; weather states for them can be easily imagined from them. However it is not so easy to represent the weather states on analyzed fields through data assimilations. The reason is that the sea surface winds are the phenomena at the bottom of atmosphere and complicated boundary layer physics prevents to transmit effects of the surface wind data to upper layers inside a model.

To get more impact from the data, we try to estimate surface pressures from surface winds and use them in data assimilations. The principle idea of the scheme is as follows: a gradient field of surface pressure over a satellite swath can be estimated by dense surface wind data. To calculate gradient field of surface pressure, the geostrophic relation can be used with some modifications for an effect of surface frictions. The gradient field can be easily converted to surface pressure field by calibration using conventional marine observations. The scheme is almost same as that of Brown's (Brown 1995) and Hsu's (Hsu et al. 1997) but much simplified in surface friction effect.

Equations to estimate gradient of surface pressure from surface winds are:

$\begin{displaymath} {\rm grad} P_{\rm NS}~=~- 2 \rho \Omega \sin \phi \cdot W \cos \alpha \end{displaymath}$

(1)

$\begin{displaymath} {\rm grad} P_{\rm EW}~=~2 \rho \Omega \sin \phi \cdot W \sin \alpha\end{displaymath}$

(2)

where ${\rm grad} P_{\rm NS}$ is the gradient for north-south direction at a point and ${\rm grad} P_{\rm EW}$ is for east-west direction. $\rho$ is density of the atmosphere which is ;

$\begin{displaymath} \rho = P / RT\end{displaymath}$

(3)

where P is surface pressure, R is the gas constant and T is temperature. $\Omega$ is coriolis parameter and $\phi$ is latitude of the point. W is wind speed and $\alpha$ is a wind direction modified by surface friction :

Fig. 3. Schematic diagram of the surface pressure retrieval technique from surface wind observations.

The statistics of the retrieval against the first guess (6 hour forecast) is shown in Fig. 4. The period of the data used is about 1 month (from May 1 to 29 May 1997). Distribution of estimated surface pressure values is consistent with the first guess quite well.

Fig. 4. Distribution of retrieved surface pressure values and surface pressures of first guess at the data points. Dark line represents for retrieved data and pale line represents for the first guess. The latter line is almost above the latter line.

Surface pressure data directly affect mass field and their impact on a data assimilation is much stronger than that of surface winds. However fine structures are often smoothed by the conversion. This character of the scheme is not appropriate for the use in fine mesh models whose main aim is to represent such fine structure of atmosphere. To use the data more effectively in a fine mesh model, variational analysis model with high performance boundary layer physics is requested.

Impact study of satellite scatterometer data for a global NWP model was conducted using ADEOS/NSCAT sea surface wind observations and ERS-2 ones.

6 hourly intermittent global data assimilation system, which is the same as the operational one, was used. The forecast model was a reduced version (T63L30) of the operational global spectral model (GSM9603) and the analysis scheme was three dimensional multivariate optimum interpolation. NSCAT data used were wind vectors (level 2 data) processed by NOAA/NESDIS. They were transmitted to JMA via NASDA (EOC) at near real time. For ERS-2 winds, data processed by EUMESAT and also delivered in near real time via GTS were used. However processed wind vectors were not used but calculated wind vectors from sigma-0 using the CMOD4 function were used. Data assimilations were started from 00UTC May 1 1997 and continued to 12UTC May 29. Three experiments were conducted as listed on Table 1.

**Table 1:** Variation of the experiments. O denotes that the data were used in the experiment and X denotes no.
Experiment	NSCAT	ERS-2 wind
Control	X	X	the same as operational
ERS+NSCAT	O	O	using all scatterometer data
ERS	X	O	using ERS-2 winds only

Positive impact for model's performance by using NSCAT data is obvious over the southern hemisphere. Figure 5 shows a comparison of 5 days forecasts around Patagonia between the Control run and the ERS+NSCAT run. Analysis shows that a low pressure of 995hPa is at just off of the east coast of Patagonia. In the Control run, its position is forecasted at 800km westward from the coast and it is too developed; the size is too big and its center pressure is 980hPa. In the forecast of the ERS+NSCAT run the low pressure can be seen on the coast and its size is almost the same as the analysis. Comparing these forecasts, performance of the ERS+NSCAT run is much better than Control run in this case.

Fig. 5. Comparison of 5 days forecast around Patagonia (south part of South America) between ERS+NSCAT run and Control run. Left plate is 5 days forecast of ERS+NSCAT run and middle one is that of Control run. Right plate shows analysis at the forecast time (00UTC May 18 1997).

Figure 6(b) shows mean anomaly correlations of forecasted 500hPa geopotential height field over the southern hemisphere (from 20S to 60S). The anomaly correlation is a score that evaluates agreement of patterns between analyzed field and forecasted field. The higher of the value is the better of the forecast performance. 100% means the perfect, over 80% means good forecast and over 60% means useful one. The figure shows that the score drops to 80% at 3.4 days for the ERS+NSCAT run but 3.0 days for the Control run. It means that period for good forecasts improved for 0.4 days by using the scatterometer data. Even using only ERS winds for scatterometer data, positive impact which is smaller than that of the ERS+NSCAT run can also be recognized in Fig. 6(b). It probably means that due to sparseness of observations over the southern hemisphere, positive impact of adding the other data can be easily obtained. The impact of using the scatterometer data over the northern hemisphere was also recognized but smaller than that over the southern hemisphere. Figure 6(a) shows the same as Fig. 6(b) but over the northern hemisphere. Apparent improvement can be seen after five days forecasts. Forecast day when the score drops to 60% is 7 days for the ERS+NSCAT run and 6.7 days for the Control run. Useful forecast period improves for 0.3 days by using the scatterometer data. Improvement using only ERS wind data was not so large than that over the southern hemisphere. It means that since data coverage over the northern hemisphere is relatively good, impact of using another data is not so large than the case of the southern hemisphere.

$\begin{figure} \centering (a) \hspace*{13cm} \\ \includegraphics [height=7.0cm,... ...} \\ \includegraphics [height=7.0cm,width=8.5cm]{z5AnomSH.eps} \\ \end{figure}$

Fig. 6. Mean anomaly correlation of forecasted 500hPa geopotential height against initialized analysis over (a) the northerm hemisphere(20N - 60N) and (b) the southern hemisphere(20S - 60S). The forecasts are carried out at every 00 and 12UTC from May 11th to 29th 1997. Pale line shows Control run and solid dark line shows ERS+NSCAT run, dash line shows ERS run.

To investigate features of the improvement, score of each forecast is monitored. Scores of four days forecasts are plotted on Fig. 7 both for northern hemisphere (left plate) and southern hemisphere (right plate). Differences between the Control run and the ERS+NSCAT run is very small over the northern hemisphere. In the southern hemisphere, they are relatively large. However we can recognized a trend that they are small for high score cases but large for low score cases; the lowest forecast score is 58% for ERS+NSCAT run but 43% for control run. Considering that degraded analyses extremely make worse forecast performances, the fact shows that initial field created by data assimilations becomes more stable using the scatterometer data which cover widely over the southern hemisphere.

$\begin{figure} \centering \includegraphics [height=6.5cm,width=15cm]{z5AnmDay.eps2} \\ \end{figure}$

Fig. 7. Scattering diagrams of forecast scores of 4 days forecasts between Control run and ERS+NSCAT run for 500hPa geopotential height. Left plate shows those over the norther hemisphere(20N - 60N) and right plate shows over the northerm hemisphere(20S - 60S).

NSCAT data provided very nice coverage for sea surface wind observations. They played important role in global observation network. Impact study using a global NWP model apparently shows improvement of forecast performance by use of the data. Next generation scatterometer data SeaWinds are expected to be provide much better observations both for data coverage and performances.

Anderson, D., A. Hollingsworth, S. Uppala and P. Woiceshyn, 1991 : A study of the use of scatterometer data in the European Centre for Medium-Range Weather Forecasts operational analysis-forecast model. 2: Data Impact. J. Geophys. Res., 96, 2635-2648.

Atlas, R., P. M. Woiceshyn, S. Peteherych, and M. G. Wurtele, 1982 : Analysis of satellite scatterometer data and its impact on weather forecasting. Oceans, 82, 415-420.

Atlas, R. and S. C. Bloom, 1989 : Global surface wind vectors resulting from the assimilation of satellite wind speed data in atmospheric general circulation models. Oceans '89 Proceedings, 160-265.

Atlas, R., 1992 : Impact of satellite Wind data on Ocean Surface Analysis and Numerical Weather Prediction. ADEOS/NSCAT Science Team Meeting. Nov. 12-14, 1992, Hakone, Japan.

Brown, R. A., 1995 : Revelations of Surface Pressure fields from ERS-1, Proceedings, ADEOS/NSCAT Science Working Team Meeting, NASDA, 214-220

Duffy D. G., and R. Atlas, 1986 : The impact of SEASAT-A scatterometer data on the numerical prediction of the Queen Elizabeth II storm. J. Geophys. Res., 91, 2241-2248.

Hoffmann, R. N., 1993 : A preliminary study of the impact of the ERS 1 C-band scatterometer wind data on the ECMWF global data assimilation system. J. Geophys. Res., 98(C6), 10233-10244.

Hsu, C. S., M. G. Wurtele, G. F. Cunningham and P. M. Woiceshyn, 1997 : Construction of Marine Surface Pressure Fields from Scatterometer Winds Alone. J. Appl. Metero., 36, 1249-1261.

Nomura, A., 1992 : Impact of SSM/I wind data on a Numerical Weather Prediction model. ADEOS/NSCAT Science Team Meeting, Nov. 12-14, 1992, Hakone, Japan.