High Temporal and Spatial Resolution Animations of Winds Observed with the NSCAT Scatterometer


Mark A. Bourassa*, David M. Legler, James J. O'Brien, James N. Stricherz, and Jiraporn Whalley
Center for Ocean-Atmospheric Prediction Studies (COAPS), Florida State University

* Corresponding Author address:

Mark A. Bourassa
Center for Ocean-Atmospheric Prediction Studies (COAPS)
Florida State University
Tallahassee, FL 32306-2840
Email: bourassa@coaps.fsu.edu
Phone: (850) 644-6923

1. INTRODUCTION

A methodology is presented for producing high temporal and spatial resolution gridded fields of surface winds from NSCAT scatterometer wind observations. Daily fields of 0.5 (~50 km) spatial resolution are produced from satellite observations of the winds over water. There is considerable call for daily marine wind products for forecasting, research, and commercial activities such as fishing and offshore oil drilling. Traditional wind observations are sparse and the output of general circulation models (GCMs) has been questionable. In recent years, space-borne scatterometer (Seasat, ERS-1/2, NSCAT) observations of wind speed and direction over the ocean surface have greatly enhanced the coverage and observational density of marine winds. The processing of scatterometer data is relatively easy (i.e., rapid and cost effective) in comparison to other remote sensing techniques such as synthetic aperture radar (SAR), and has much better coverage than winds derived from altimetry. The NSCAT scatterometer is an active microwave sensor that covers approximately 77% of the ice-free ocean in one day, and 97% in two days. A new technique is developed to fill the gaps in the fields of daily observations.

These daily winds are used in animations showing moving wind vectors, and the evolution of divergence and vorticity fields. Objectives of this project include production of the animations for many regions of the globe, and the display of these animations on a public website (http://www.coaps.fsu.edu/nscat). The animations show surface airflows that have not been identifiable from previous in situ or satellite observations. A few examples include the surface structure of a mid-latitude storm, and the extent of cool air flowing from a mountain gap near Tehauntepec, Mexico.

2. BACKGROUND

Daily wind products are used to study a wide range of meteorological and oceanographical phenomena with time scales from days to seasons. Observations over oceans are typically sparse in both space and time. Consequently, the initialization fields from GCMs (e.g. ECMWF and NCEP winds) have often been utilized as surface winds. Alternatively, fields have been developed from long term averages (at least monthly) of ship and buoy observations (Hellerman and Rosenstien 1983; Legler et al. 1989). In both cases the spatial resolution is course, typically 2 to 5 degrees. Better temporal and spatial resolutions are desirable.

2.1 Previous fields of scatterometer winds

Spaceborne scatterometers provide a high spatial resolution (~25 km) within their observation swaths, but no observations outside the swaths. The time taken to completely cover the world's oceans has been limited by the choice of orbits and operational restrictions. For example, without operational restrictions, the ERS-1/2 scatterometers cover the globe in approximately six days. This time scale is typical of synoptic events, suggesting that it would be impractical to use these scatterometers to thoroughly monitor events of comparable or smaller time scales. Furthermore, the sampling time required to remove sampling biases related to the orbital pattern has been found to be greater than one month for the ERS scatterometers (Zeng and Levy 1995). The construction of daily surface wind fields, based on oceanic observations, has not been practical prior to the period of NSCAT observations.

2.2 NSCAT wind fields

NSCAT wind observations cover most of the Earth in one day, and have nearly 100% coverage in two days. The advantages over ERS scatterometers are duo swaths (one on each side of the orbital trajectory), nearly continual operation (no mission constraints), and a lack of dropout. Additional advantages are improved accuracy in wind speed and direction (Bourassa et al. 1996; Bourassa et al 1997), and superior ambiguity selection.

3. METHODOLOGY

The production of daily wind fields, appropriate for forcing ocean models, requires that gaps in the coverage are filled by accurate estimates of the wind. The wind fields must also be smooth at these edges of gaps, otherwise the winds could excite spurious Kelvin waves. Our solution to this problem is to use observations from closely related times. The observations closer in time to the day of interest are weighted much more heavily than those further away in time. This approach retains the dominance of winds observed on the day in question, and it also allows for a relatively smooth transition into regions where there were no observations.

3.1 Binning and weighting

The bin size for the winds was chosen to include sufficient observations such that incorrectly selected ambiguities were unlikely to have a large influence. For the 50 km resolution NSCAT wind product, which is used in this study, this condition is satisfied with a 1 x 1 bin size. The 25 km resolution NSCAT product has a similar accuracy in ambiguity selection (M. Freilich, personal communication, 1997); therefore, the bin size could be reduce to 0.5 resolution when the higher resolution data is used. The lower limit on resolution is that of the NSCAT observations, which is approximately 25 km.

The weighting procedure is analogous to a weighted vector average of one, two, four, and eight day centered averages. Each of these averaging periods is centered on 12Z of the day in question. The weighting mechanism is designed to favor observations in the short averaging periods. The eight day averaging period was sufficiently long that there are observations in each grid box despite occasional 12 to 24 hour gaps in the observations. The effective averaging window is then reduced through a weighted average of the eight day () and four day () fields. The effective averaging window is then further reduced by averaging the product () with two day and one day averages. The key equations are

, (1)

, (2)

, (3)

, (4)

, (5)

where the numerical subscript is the number of days used in determining the average wind vector , and n is the number of observations in cell i, j. The superscript '*' indicates that observations from longer time periods have been included in the wind field. Increasing the value of the weighting parameter (b) sharpens the fit to at the expense of smooth fields near swath edges. We have found a value of =3 to be sufficiently large that the field is dominated by winds observed on the day of interest, and sufficiently small that the field is reasonably smooth where swaths intersect.

Fig. 1. NSCAT based wind vectors for Nov. 15, 1996.

It will be shown that there is usually little loss of information from daily fields. This point is illustrated in two examples of the differences between the filled winds () and mean winds for one day (). One case is a tropical wind field (Fig. 1), and the other is a 'worst case' test with a strong and rapidly moving cyclone (Fig. 2). In both cases, rms differences are calculated using only the points where there were observations in the one day average (). The first case has an rms difference of 1.2 m s-1 in the East-West wind component (u), 1.1 m s-1 in the North-South wind component (v), and 1.7 m s-1 in speed. There is little spatial bias in these differences. These differences are similar to the observational uncertainty in wind speed (Bourassa et al. 1997).

The 'worst case' example has an rms difference of 2.9 m s-1 in the East-West wind component (u), 2.5 m s-1 in the North-South wind component (v), 3.8 m s-1 in speed. These differences are biased such that the largest values occur near the center of circulation. They are associated with the timing of satellite passes and the motion of the center of low pressure. The result is a blurred image of rapidly translating systems. In this extreme case, there is an underestimation of the cyclones wind speed by approximately 33% of the daily average. A more elaborate weighting system is required to reproduce instantaneous snapshots rather than daily averaged fields. Daily averages can be enhanced through direct minimization and generalized cross validation (Legler et al., 1998).

Fig. 2. NSCAT based wind vectors for Sept. 29, 1996.

3.2 Smoothing

The strengths of advantages and disadvantages of smoothing depend on the application of the wind field. The unsmoothed product has disorganized winds in regions of low wind speeds; however, it retained the fine scale structure of low pressure systems and fronts. Smoothing was found to be useful for determining mean wind directions in areas of low wind speeds. These are the areas where incorrect ambiguity selection is most likely, as well as areas with the lowest signal to noise ratios. However, studies of tight circulations, gap flow, and fronts are best undertaken without further smoothing of the fields. The binning applied in the above averaging process applies sufficient smoothing that fields of vorticity and divergence are smooth. Hence, additional smoothing is not considered to be necessary.

4. ANIMATIONS

The above wind fields were used to produce animations of the winds and related fields. The winds are shown with moving vectors. The motion of the vectors is Lagrangian, and the vector length indicates the wind speed. The vector positions are calculated by interpolating the daily wind fields to one hour time steps, and integrating with a fourth order Runge-Kutta method. The Runge-Kutta technique uses an adaptive time step, with a first guess of ten minutes. The time interval between animation frames is dependent on the highest wind speeds; however, we have found a time step of three hours to be effective. The animations show one week of wind fields.

Fig. 3 NCEP wind vectors for Nov. 15, 1996.

Horizontal divergence for surface air is calculated at the grid points as a second order finite difference. The divergence field is bi-linearly interpolated to the central position of each moving vector. Divergence are shown by the color of the wind vector. Red vectors indicate convergent flow, and blue vectors indicate divergent flow. This coloring scheme helps highlight convergence zones, fronts, and structure within strong cyclones.

The relative vorticity field is also calculated as a second order finite difference. The background colors (over water) are contours of vorticity. Vorticity indicates rotation: positive values indicate clockwise (anticyclonic) rotation, and negative values indicate counter clockwise (cyclonic) rotation. The vorticity contours highlight the central position of storms, and help show the extent of flow modifications due to orographic features.

5. RESULTS

The advantages of scatterometer observations are shown in a comparison of wind fields generated by this technique to National Center for Environmental Prediction (NCEP) daily wind fields. The animations are examined to find new and interesting features.

Fig. 4. NCEP wind vectors for Sept. 29, 1996.

5.1 Comparison to NCEP fields

The NCEP wind fields (Figs. 3 and 4) can be compared to the fields generated by this filling technique (Figs. 1 and 2). The NCEP fields have a resolution of only 2 latitude and 2 longitude. The NCEP winds capture most of the large scale features in Fig. 1; however, the position of the cyclone is too far North, the wind direction in the western Gulf of Mexico has too large an easterly component, and the strong winds flowing through the Isthmus of Tehauntepec are too weak. The NCEP field tends to be too smooth in the data sparse regions (i.e., the bulk of the Pacific ocean shown).

The NCEP winds around North Sea (Fig. 4) are dominated by the strong cyclone North of Scotland, which appears at the same latitude and longitude in the NSCAT wind field (Fig. 2). The main difference between the figures is the flow over the North Sea. The NCEP wind pattern is cyclonic about the low pressure center; whereas the NSCAT observations show that the wind fields curve more to the right, and flow towards the Baltic Sea (South of Scandinavia). The differences in this data rich region are smaller than the difference in Equatorial and Southern latitudes.

5.2 New findings

5.2.1 Gap Flow

Small spatial scale and short time scale features of wind fields caused by gap flow through the Isthmus of Tehauntepec have not been examined. Previous studies have used sparse wind observations from ships, or indirect methods such as sea surface temperature (SST) patterns (Roden 1961; Clarke 1988). The winds have been assumed to follow an inertial trajectory. However, the animations show that there is considerable short term variation, and highly non-inertial winds (Fig. 1) can exist for eight days. Animations of wind field showed one example of winds flowing South through the Isthmus of Tehauntepec, turning to there left, flowing through passes near Lake Nicaragua, and entering the Caribbean Sea. This type of flow pattern had not been seen prior to the NSCAT observation period. It is readily apparent the non-inertial flow is induced by a pressure gradient related to a Caribbean cyclone. The animations allow the study of the evolution of this wind pattern.

5.2.1 Cyclone Evolution

It has been recognized for some time that the movement and evolution of centers of low pressure is more complicated than the common conceptual model of radially symmetric flow. The vorticity and divergence information in the animations clearly show frontal structures associated with low pressure systems. The animations are the first observation based demonstration of the evolution of a cyclone's surface winds. The 'center' of low pressure can fill at one location while deepening at nearby location within cyclones. In most cases this shift of location would appear as a short lived change in the cyclone's central motion.

5.2.1 Southern Ocean Cyclones

There has long been questions regarding the evolution of southern hemisphere anti-cyclones (storms) that enter the southern ocean surrounding Antarctica (Bromwich et al. 1992). Observations of wind over southern hemisphere waters had been sparse prior to satellite coverage. The storm systems move rapidly, consequently, they could not easily be studied without very good spatial and temporal coverage. The animations show storms moving from the southern mid-latitudes into the southern ocean, as well as what appear to be storms moving off the ice sheets and into the southern ocean. The complexity of these wind patterns was suspected prior to NSCAT, and can now be studied with these wind fields.

5.3 Educational tool

The animations can be an excellent tool for demonstrating the atmospheric dynamics in a manner that is enthralling and easy to follow. Concepts such as high and low pressure systems, fronts, convergence zones, and gap flow can easily be conveyed at levels appropriate for a wide range of backgrounds.

6. CONCLUSIONS

The new methodology for filling data void in scatterometer observations is accurate and easily implemented for NSCAT observations. High resolution (1.0 x 1.0, and 0.5 x 0.5) daily wind fields were created. The averaging process makes acceptably smooth fields and retains the vast majority of synoptic scale details. The mean difference between observed and averaged speeds was typically less than 2 m s-1, although the difference could be up to 5 m s-1 for the 'worst case' example of a rapidly translating cyclone. The wind fields are more accurate and detailed than those available from the NCEP reanalysis.

These fields were used to produce moving vector animations. The NSCAT animations are excellent as educational tools, and as a first look product for research. The animations demonstrate observed atmospheric circulation and evolution with a clarity that is unprecedented.

Acknowledgments. We thank Kathy Verzone and Al Davis for their contributions developing the moving vector code. Funding for this project is from the NASA JPL NSCAT Project (contracts 957649 and 980646). COAPS receives its base funding from ONR's Secretary of Navy Grant to Dr. James J. O'Brien.

REFERENCES

Bourassa, M. A., M. H. Freilich, D. M. Legler, W. T. Liu, and J. J. O'Brien, 1997: NSCAT Really Worked: A First Comparison of Satellite Winds to Quality Research Ship Observations. EOS, accepted.

Bromwich, D. H., J. F. Carrasco, and C. R. Stearns, 1992: Satellite observations of katabatic-wind propagation for great distances across the Ross Ice Shelf. Mon. Wea. Rev., 120, 1940 - 1949.

Clarke, A. J., 1988: Inertial wind path and sea surface temperature patterns near the Gulf of Tehuantepec and Gulf of Papagayo. J. Geophys. Res. 93(C12), 15,491, 15,501.

Hellerman, S. and M. Rosenstien, 1983: Normal monthly wind stress over the world oceans with error estimates, J. Phys. Oceanogr., 13, 1093 - 1104.

Legler, D. M., M. A. Bourassa, A. D. Rao, and J. J. O'Brien, 1998: NSCAT surface wind fields using optimally tuned direct minimization techniques. Ninth Conference on interaction of the sea and atmosphere, Pheonix, AZ, American Meteorological Society, in press.

Legler, D. M., I. M. Navon, and J. J. O'Brien, 1989: Objective analysis of pseudo-stress over the Indian Ocean using a direct minimization approach. Mon. Wea. Rev., 122, 1632 - 1639.

Roden, G. I., 1961: On the wind-driven circulation in the Gulf of Tehuantepec and its effect upon surface temperatures. Geofis. Int., 1, 55, 72.

Zeng, L., and G. Levy, 1995: Space and time aliasing structure in monthly mean polar-orbiting satellite data. J. Geophys. Res. 100(D3), 5133 - 5142.