Gayana 68(2) supl. t.I. Proc. : 174-179, 2004 ISSN 0717-652X

SURFACE CURRENT MAPPING OFF CALIFORNIA WITH RADIOMETRY AND ALTIMETRY

W. Emery, D. Matthews, R. Crocker & D. Baldwin

CCAR, Univ of Colorado Boulder, CO, 80309

ABSTRACT

Surface coastal currents off California are a good example of the complex variability in space and time of currents in an ocean basin eastern boundary. Conventional oceanographic measurements are unable to resolve these relatively small scale variations but by employing both satellite imagery and satellite altimetry we are able to resolve this complex structure and its variations in space and time. Moreover using historical satellite imagery we are able to extend this study back into the past. Our emphasis will be on resolving these variations and later studies will relate these variations to various forcing functions.

INTRODUCTION

Conventional methods of mapping ocean surface currents have unable to resolve the complex space/time structure of surface currents in a region like the California Current off southern California. These conventional methods convolve space and time sampling and alias higher frequency oscillations into the lower frequency/wavenumber aspects of the surface current field. Modern numerical models (Batteen, 1997; Haney et al., 2001) are now capable of resolving these complexities but are therefore in need of measurements that can be used for initialization, assimilation and verification.

Only satellite methods have the potential for resolving the space/time structure of the surface currents in this area. Satellite altimetry (Strub and James, 2000) has been explored as a measure of these currents but it is not clear that the poor cross track sampling of the present altimeter configuration is able to resolve the mesoscale structure that dominates the California Current.

We will introduce a method to produce high-spatial resolution surface currents from sequential infrared and ocean color images that can then be merged with the altimeter derived geostrophic surface currents. This image method and its combination with altimetry appears able to resolve the complex structure of the mesoscale components of the California Current. We will demonstrate how the infrared and ocean color image pairs yield very similar current maps and how these two current fields are better able to resolve the wavenumber structure of the California Current than is the present constellation of altimeters. The ageostrophic currents that are the differences between the geostrophic altimeter currents and those derived from the imagery are primarily wind driven in this strongly upwelling driven region.

Data and Processing

There are three sources of data for this study: a. Advanced Very High Resolution Radiometer (AVHRR) channel 4 (11 µ) imagery, b. ocean color imagery from the SeaWIFS satellite and the MODerate-resolution Imaging Spectrometer (MODIS) instruments on NASA's morning TERRA and afternoon AQUA satellites, and altimeter data from NASA's TOPEX/Poseidon satellites, Jason-1 satellites and the ERS-2 satellite. Most of the data will be from 2002 and 2003. Up to 10 AVHRR images per day were used while about 6 ocean color images per day were used.

Computing Surface Currents

The tool used for the routine computation of surface currents from all imagery was the Maximum Cross Correlation (MCC) technique (Emery et al., 1986, Ninnis et al., 1986). The MCC method is demonstrated here in Fig. 1, which shows two sequential images. The template window in the first image on the left defines the pattern that must be found in the later second image on the right. The template window must be large enough to make the cross correlation computed from the pixels in the window statistically significant.

In the second image the template window is moved stepwise around the larger search window to locate the pattern by the maximum cross correlation. The search window must be large enough to accommodate the largest velocity expected in the area. Once found the maximum cross correlation defines the end point of the velocity vector from the center of the template window in the first image.

Figure 1: MCC method; template window in the first image is located by the maximum cross correlation in the later second image.

Essential to this method is the accurate geo-registration of both images. Any errors in geo-registration will translate into erroneous MCC surface currents. Since we have to process a great many images we developed an automatic image navigation technique (Emery et al, 2002) that is used to accurately geo-locate all of the images.

Another critical processing step is the conservative removal of all cloud contaminated pixels. Here we use thermal cutoff thresholds in the infrared channels along with some temporal persistence filters. We are certain that we remove all of the clouds. To be certain we are also eliminating a certain amount of good image data with the clouds.

Where we find coincident "holes" in image one and image two we are able to compute the surface currents. Thus, a single current field from a single image pair will look like a patch-work quilt. We then composite the vector fields over time, usually 3 to 10 days. It is important to point out that we are compositing vectors and not images. To composite images would smooth out the gradients that we are trying to follow.

We also filter the final composite vector field to produce a smooth vector field. We first eliminate vectors that are the result of lower correlation values. We now reject all vectors with correlations lower than 0.7. We then require vectors to be consistent with their neighbors within 2 pixels. This removes large and errand vectors which often crop up. The resultant vector field is then remapped using overlapping template window boxes to produce a higher resolution surface current map.

Finally for merging with the altimeter data the MCC surface currents are fit to a stream function (thus requiring that they be geostrophic and have no divergence) and mapped with an optimum interpolation (OI; Wilkin et al., 2001). This same OI is used to merge the MCC vectors with those from the altimeters.

One might think that it should be equivalent whether one uses a SST product or a single channel brightness temperature. That turns out to not be the case as the cross correlations for an SST product are seen to be significantly smaller than the coincident cross correlations from a single channel 4 (11 µ) brightness temperature image pair. This is likely because most SST algorithms involve a "split-window" which is the difference between two very similar infrared channels, which would have the effect of increasing the pattern noise in the algorithm.

Using Ocean Color

In principle ocean color should behave similarly to the infrared imagery as the ocean color surface features will be advected much as are the infrared patterns. To test this we computed matching ocean color and infrared derived coincident surface current maps. An example is presented here in Fig. 2, which shows a surface current field computed from a pair of ocean color images overlain on one of the ocean color images. We should point out that for ocean color we used a chlorophyll product rather than any single channel. This chlorophyll product was found to give better and more consistent results than any single channel. For MODIS the chlorophyll was already calculated and available as images while for SeaWIFS we had to calculate the chlorophyll (Strub et al., 1997). Since we are only interested in the chlorophyll patterns and not in absolute values it is not important if the chlorophyll algorithms gives different biases. This is actually a 10-day composite of ocean color derived surface currents rather than a current field from any single image pair.

Figure 2: 10-day composite MCC surface currents computed from ocean color superimposed on a chlorophyll image.

This field can be compared with the vectors in Fig. 3, which is a 10-day composite from sequential infrared images composited over the same 10-day period as Fig. 2. The vector are then superimposed on one of the component infrared images. Note the very different infrared pattern in Fig. 4 versus the chlorophyll pattern in Fig. 3. It was found that often the ocean color imagery contained gradients in regions where the infrared gradients were weak. Thus, in some ways the use of ocean color and infrared imagery can be considered as complimentary in the computation of MCC surface currents.

To evaluate how well longer term averages of ocean color and infrared vectors compared we computed all of the vectors for both ocean color and AVHRR for 2003. To save space we will present only a few examples. In Fig. 5 we present the average surface current for Oct.., 2003 with vectors computed from ocean color imagery in red and vectors computed from sequential infrared imagery in black.

The least cloud cover in this region is in the fall as demonstrated clearly here in the mean MCC current map for Oct. This map shows all of the complex structure of this region consisting of both near circular eddies of both signs and a southward offshore mean flow. To the northwest a pair of vortices is a common feature of this area which generates a lot of eddies many of which are matched as these two. This pair is also part of the mean southward California Current. A much smaller cyclonic eddy is found inshore of the mean flow to the south of this eddy pair.

Figure 4: 10-day composite MCC currents computed from sequential infrared AVHRR images

Figure 5: Mean Oct., 2003 MCC currents with ocean color (red) and infrared (black)

Comparing MCC Currents with Satellite Altimetry Derived Currents

One of the main limitations of the MCC method described above is the fact that cloud cover obscures the Earth's surface making it impossible to compute ocean surface currents where cloud cover is present. In many cases the cloud cover either moves off or disperses over time scales short relative to the surface current mapping and temporal composites are able to recover the important surface current structure.

Persistent cloud cover, however, will lead to real problems in terms of sampling biases that will make it impossible to map currents evenly across any one region. Another method of computing surface currents from satellite data are geostrophic surface currents calculated from satellite altimeter measurements. These measurements are all weather and do not depend on cloud cover making it possible to compute these surface currents even in the presence of persistent cloud cover.

A disadvantage to the altimeter sampling is its limited spatial resolution due to the relatively large space between ground tracks (about 50 km for the present "tandem" mission). Another limitation is the fact that altimeter surface currents are only the geostrophic component of the surface currents unlike the MCC currents, which represent some combination of both geostrophic, and ageostrophic surface current components. Thus to combine MCC currents with those from altimeters we fit the MCC currents to a stream function which restricts the MCC currents to divergenceless and thus geostrophic flow. In this way we combine geostrophic currents from MCC and altimetry.

A graphical demonstration of the agreement between altimeter currents and those computed from AVHRR infrared imagery is shown here in Fig. 6 which has the altimeter vectors overlaid in black on the field of MCC vectors for a 3-day composite in Sept. of 2003 on the left. On the right we show the OI map for this AVHRR vector field with the same altimeter vectors overlaid in black. Both versions show excellent agreement between the geostrophic vectors of the altimeter and the MCC velocities both in raw and mapped forms. A similar agreement was found for ocean color imagery.

Figure 6: MCC currents (red) with altimeter currents (black) overlaid on; left., filtered MCC currents from AVHRR images; right, OI mapped MCC currents for Sept. 13-16, 2003.

Comparison Statistics

There is a question of whether or not the statistics of the respective fields are the same. We decided to examine wavenumber spectra of all surface velocity fields as a way to characterize the surface velocities in terms of their wavenumber behavior. Here we will refer to alongshore and cross-shore vector components and spectral axes. We will compare wavenumber spectra for ocean color alone MCC vectors with infrared only MCC vectors and altimeter geostrophic currents. Rather than present the 8 possible wavenumber spectral triplets we will restrict our attention to one example to demonstrate the salient features of these spectral fields.

Typical is the cross-shore (x) spectrum of alongshore (V) velocity as given here in Fig. 7. This spectrum shows similar low wavenumber spectral peaks for both ocean color and AVHRR vectors with similar spectral energy values. The altimeter spectral values while having a peak at the same wavenumber are significantly lower than the other two suggesting the inability of the altimeter sampling to resolve much of the mesoscale surface current field. At wavelengths shorter than about 96 km all of the spectra have dropped to almost zero. The strong peak at about 240 km wavelength is quite interesting as it suggests the size of the mesoscale features in this region. This size is consistent with earlier numerical model studies (Ikeda et al.,1984) and other observations (Chereskin et al, 2000).

Figure 7: Cross-shore (x) wavenumber spectra of alongshore (V) velocity.

DISCUSSION AND CONCLUSIONS

It is clear from the many comparisons both of surface vector fields and wavenumber spectra that infrared and ocean color satellite imagery can be used together with the MCC method to map mesoscale surface currents in the ocean. In fact it appears that often the ocean color vectors compliment the infrared vectors with respect to spatial coverage. In other words ocean color gradients are often located in areas where the infrared temperature gradients are weak.

In coastal regions such as the California Current spatial resolution such as that available with the MCC current mapping method are needed to resolve the complex space/time variations of the mesoscsale surface currents that dominate this region.

ACKNOWLEDGEMENTS:

This work was supported by NASA Earth Sciences Physical Oceanography Program as part of its altimeter science team, Eric Lindstrom program manager.

REFERENCES

Batteen, M.L., 1997: Wind-forced modeling studies of currents, meanders and eddies in the California Current System. J. Geophys. Res., 102, 985-1,010.         [ Links ]

Chelton, D.B., 1984: Seasonal variability of alongshore geostrophic velocity off central California. J. Geophys. Res., 89, 3, 486-4, 373. [         [ Links ]1]

Chereskin, T.K., M.Y. Morris, P.P. Niiler, P.M. Kosro, R.L. Smith, S.R. Ramp, C.A. Collins & D.L. Musgrave, 2000: Spatial and temporal characteristics of the mesoscale circulation of the California Current from eddy-resolving moored and shipboard measurements. J. Geophys. Res., 105, 1,245-1,269. [         [ Links ]2]

Emery, W.J., A.C. Thomas, M.J. Collins, W.R. Crawford & D.L. Mackas, 1986: An objective procedure to compute advection from sequential infrared satellite images. J.Geophys. Res., 91 (color issue), 12,865-12,879. [         [ Links ]3]

Emery, W.J., D. Baldwin & D. Matthews, 2003: Maximum Cross Correlation Automatic Satellite Image Navigation and Attitude Corrections for Open Ocean Image Navigation, Trans. Geosci. Rem. Sens., 41 ,1 , 33 - 42. [         [ Links ]4]

Haney, R.L., R.A. Hale, & D.E. Dietrich, 2001: Offshore propagation of eddy kinetic energy in the California Current. J. Geophys. Res., 106, 11,709-11,718. [         [ Links ]5]

Ikeda, M., W.J. Emer y & L.A. Mysak, 1984: Seasonal variability of the meanders in the California Current system off Vancouver Island. J. Geophys. Res., 89, 3487-3505. [         [ Links ]6]

Strub, P.T. & C.James, 2000: Altimeter-derived variability of surface velocities in the California Current System: 2. Seasonal circulation and eddy statistics. Deep-Sea Res. II, 47, 831-870. [         [ Links ]7]