A new method to reduce noise on satellite sea surface temperature observations
Jacob Lorentsen Høyer * and Jun She,
Danish Meteorological
Institute, Lyngbyvej 100, 2100
Abstract
Statistical
comparisons between coinciding satellite and in situ sea surface temperature
observations (SST) observations show that satellite errors tend to be more
spatially correlated than temporally. These findings have been used to perform
tests where in situ observations of high quality (e.g. bouys ) are utilized to
perform a real time correction of the satellite observations over a large area.
The method has been tested in the
Keywords:
Satellite SST observations, error
reduction, operational products,
1. Introduction
Observations of sea surface temperature (SST) from passive infrared satellite observations have become one of the most widely used satellite product for oceanic and meteorological applications. The errors of the SST products are typically in the range of 0.5 0C in open ocean and about 0.7 0C for coastal applications. The two major error contributions - when the satellite observations are compared to in situ observations at a few meters depth - arise from the atmospheric correction and from the difference between the skin temperature at the very top of the surface(~10my) layer and at 1-2 meters depth. In the recent years, several methods have been presented to reduced the errors on the satellite observations and thus to make the satellite observations more representative to the in situ observations (e.g., Horrocks et al., 2003, Castro et al., 2003). The methods use in general auxiliary data such as wind and heat flux to correct the for the vertical temperature variations in the upper 1-2 meters. However, as the processes in the upper ocean are very complicated, the existing methods has not been able to fully account for the vertical temperature differences (Castro et al., 2003).
In this study, we propose a new method of
correcting the satellite SST observations. The method is based upon the
findings that the differences between satellite an in situ observations are
correlated in space over several hundred kilometers. By using a few accurate
real time buoy observations, we are thus able to correct real time satellite
observations over a much larger area. The method has been applied to the
2. In situ and satellite observations
A large amount of in situ observations (~200000) from year 2001 have been gathered from the North Sea/Baltic Sea within the ODON project. The observation database comprises observations from fixed platforms, ships of opportunity, Ferrybox and from scientific cruises.
Two different satellite products are used
for this study. The first products is the Ocean and Sea Ice SAF (O&SI-SAF)
product, MNOR, which has a resolution of 2 km in the North Sea/Baltic Sea region
and provide data from the NOAA 16 and 17 (Brisson, 2001). The other satellite
product is produced by BSH in
3. Error correlations
Spatial and temporal correlations have been calculated from both in situ and satellite observations. The data set have been averaged into 10x10 km grids and one average per day prior to the calculation. The correlations have only been calculated from grid points where both in situ and satellite observations are available to make the results intercomparable and to exclude differences in the results due to sampling issues. The average spatial and temporal correlations are shown in Figure 1.
The higher spatial correlations of the satellite observations compared to the in situ observations indicate that the spatial noise on the satellite observations has a much larger scale than on the in situ observations. Conversely, the differences between the temporal satellite and in situ correlations indicate that the that the noise on the satellite observations -which is higher than on the in situ observations – is not very well correlated from one day to the next. These findings are essential for the construction of a method that uses real time in situ observations to reduce the differencies between the satellite and the in situ observations. Note that several processes, such as atmospheric effects, cool skin layer and diurnal thermoclines can account for these differences and that we only focus on the combined effect.
4. Correction method
Assuming that the real time buoy observations of SST are well calibrated, we use the difference between satellite SST and bouy SST in a local region to correct a much larger field. In this way we are able to correct for some of the spatially correlated errors on the satellite SSTs. The correction is performed as:
Where the tilde represents the corrected observations and the correction term is calculated using coinciding satellite and buoy observations at x0and y0.
5. Results
The performance of the method to reduce noise on satellite observations, was tested using data from the three buoys shown in Figure 2. In situ observations, not used for the corrections, were used to validate the satellite SST observations before and after the correction was applied. The RMS results are shown in Figure 3 for each buoy as a function of the radius of the correction area. All three figures show that there is a significant potential for reducing the errors on the satellite observations. The satellite RMS errors are consistently reduced for all the areas. The largest effect in reducing the RMS errors is seen when the EMS is being used for the correction, where the errors are reduced by up to 0.20C close to the buoy.
6. Conclusions
Comparisons of empirical correlations from satellite and in situ observations showed large spatial correlations of the noise on the satellite observations. This implies that satellite data over a large area can be corrected using a limited number of accurate in situ observations. Observations from 3 buoys in the North Sea and in the Danish Straits have been used to test the improvements of the method. The method reduced satellite RMS errors with up to 0.2 0C over regions of several hundreds of kilometers, demonstrating the large potential for operational applications.
Ongoing and future work include the application of this method to satellite observations all over the North Sea and to incorporate the spatial scale of the errors, to estimate corrections based upon a combination of several buoys.
Acknowledgements
This work is supported by the European Commission under the Fifth Framework Programme. The project is ODON (Optimal Design of Observational Networks), Contract No. EVK3-2002-00082.
References
Brisson, A., P. Le Borgne, A. Marsouin (2001). Ocean & Sea Ice SAF, North Atlantic Regional Sea Surface Temperature, O&SI SAF Product Manual version 1.1. Meteo--France/DP/CMS, France.
Castro, S.L., G. A. Wick, W. J. Emery (2003). Further refinements to models for the skin-bulk sea surface temperature difference, J. Geophys. Res., Vol. 108, C12, doi:10.1029/2002JC001641.
Horrocks, L. A., B. Candy, T. J. Nightingale, R.W. Saunders, A. O'Carroll, and A.R. Harris (2003). Parameterizations of the ocean skin effect and implications for satellite-based measurements of sea surface temperature, J. Geophys. Res., Vol. 108, C3, doi:10.1029/2002JC001503.
Høyer, J.L. and J. She (2004). Validation of satellite SST products for the North Sea-Baltic Sea region. Technical Report 04-11, Danish Meteorological Institute.
Høyer, J. L. and J. She (2005). Optimal
interpolation of sea surface temperature for the
Loewe, P. (1996). Surface temperatures of
the
Figure 1: Spatial (left) and temporal (right)
correlations calculated from satellite and in situ observations. All results
are calculated from grid points where both satellite and in situ observations
are available
Figure 2: Positions of buoys used to test the
correction method: Drogden (DRO), Deutsche Bucht (DB) and the
Figure 3: RMS errors of the satellite observations
during year 2001, as a function of distance from the buoy. Blue indicate the
original satellite error and red shows the RMS errors of the corrected fields.