ESA Earth Home Missions Data Products Resources Applications
    22-Jul-2014
EO Data Access
How to Apply
How to Access
Use of ATSR-measured ocean skin temperatures in ocean and atmosphere models (Ridley et al.)
Use of ATSR-measured ocean skin temperatures in ocean and atmosphere models
Use of ATSR-measured ocean skin temperatures in ocean a
Services
Site Map
Frequently asked questions
Glossary
Credits
Terms of use
Contact us
Search


 
 
 

Use of ATSR-measured ocean skin temperatures in ocean and atmosphere models

Ian K. Ridley, Sean P. Lawrence, David T. Llewellyn-Jones, Isabelle M. Parkes   Department of Physics and Astronomy, University of Leicester, Leicester, LE1 7RH, UK

email: ikr2 le.ac.uk

Ryuzo Yokoyama, Sumio Tanba, Shinya Oikawa   Department of Computer and Information Science, Iwate University, 4-3-5 Ueda, Morioka, Iwate 020 Japan

Abstract

Ocean and atmosphere models traditionally use bulk ocean temperature records as inputs. However, ATSR measures the radiance at the top of the atmosphere which is the radiometric skin temperature of the ocean surface modified by the atmosphere. The ATSR retrieval algorithm is designed to extract this skin temperature from ATSR radiance. The global coverage of the ATSR ocean skin data set means that it is advantageous to use these data in climate models. In this study, the spatial variability of the ocean skin temperature over small scales (<10m) is examined using in-situ data. The skin temperature variability is compared to the variability of local geophysical parameters. It is found that relatively small changes in wind speed, short wave solar flux, long wave flux and humidity can be expected to increase skin temperature variability. The implications of the spatial variability of ocean skin temperature and the bulk-skin temperature difference in the use of satellite-derived skin temperatures in climate models is discussed.

Keywords: Radiometer, Sea Surface Temperature, Variability, Climate models

Introduction

Ocean temperature measurements have historically been made by recording the sea temperature at depths from a few centimetres to a few meters. Climate models require historical data as inputs and have been "tuned" to use these bulk sea temperature data records (Ma et al, 1994). In the last few decades, radiometric technology has allowed the temperature of the upper 10-20m to be measured. Imaging radiometers mounted on aircraft and satellites have produced spatial resolutions ranging from a few metres to kilometres.

Climate models are sensitive to the temperature of the ocean-atmosphere interface as it forms the upper and lower boundaries in ocean and atmosphere models respectively. It is desirable to use the skin temperature in these models as it is a closer value of the physical temperature at the ocean-atmosphere interface than the bulk temperature. However, the variability of the skin sea surface temperature (SST) under different conditions at varying length scales is not well understood. Differences between the variability of the skin and bulk sea temperatures will require adjustments to climate models if skin temperatures are to be used in this application.

Studies of the bulk-skin temperature difference (the "skin effect") show that it has a typical daytime value of 0.3 K (Schlussel, 1990) for high latitudes. This is a "cool skin" where molecular conduction from a warm ocean to a cooler atmosphere results in a warm bulk temperature. The typically lower air temperatures, higher wind speeds and lower humidity of high latitude regions can result in skin temperatures 1.0 K cooler than the bulk. Equatorial regions with lower wind speeds, high humidity and air temperatures closer to that of the ocean produce smaller values for the skin effect. Under calm conditions with strong solar insolation an unmixed layer can form towards the surface which is up to 1.0K warmer than the bulk temperature below (Schlussel, 1990). These studies have show that local atmospheric conditions can change skin SST over short timescales of the order of seconds (Donlon, 1994).

Several authors (for example Saunders, 1967, Hasse, 1970) have proposed parameterisations of the skin effect in terms of atmospheric and geophysical parameters but some of these models are regionally biased or limited to certain conditions.

The advent of accurate in-situ thermal imaging technology has allowed resolution of skin SST at centimetre scales and enables the spatial variability of skin SST over small scales to be investigated. The Mutsu Bay Experiment in Japan utilised a thermal infrared camera as part of a set of instrumentation designed to measure atmospheric and geophysical parameters at the ocean-atmosphere interface. This study uses the infrared camera data to test the statistical spatial variability of the skin SST in relation to local atmospheric parameters.

Experiment and instrumentation

The field data were acquired during the 1995 Mutsu Bay Experiment off northern Honshu, Japan (Parkes et al, 1997). The research vessel Dai-ni Misago undertook a series of transects in Mutsu Bay (figure 1) during July and August. The vessel was equipped with a meteorological station, a thermal infrared camera, a commercial radiometer to measure downwelling infrared radiation and a salinity-temperature-depth probe to provide bulk SST data.

Figure 1: Location of the Mutsu bay Experiment

Both the infrared camera and commercial radiometer had similar 8-12m wavelength profiles (Figures 2 and 3) to provide accurate correction for the reflected sky component of the observed sea brightness temperature. The camera was mounted at a small angle (23) to the vertical to reduce non-unity emissivity of the sea surface in the infrared (Masuda et. al, 1988). This angle also avoided the problem of direct reflection of the instrument into the field of view which can occur when it is mounted at a near-vertical angle. Both radiometers were mounted on an overhanging platform from the bow to reduce the possibility of reflection from the ship. The platform was 5m above the sea surface producing images 255x207 pixels in size. Pixel size was calculated to between 8 to 9mm with each image covering 2.8 metres of ocean surface.

Figure 2: Response curve for commercial radiometer (Convolved detector and filter profiles, courtesy Tasco Corporation, Japan)

Figure 3: Response curve for thermal infrared camera (courtesy NEC Corporation, Japan)

The thermal camera is manufactured by the NEC Corporation. The camera uses a mirror to focus radiation onto a mercury cadmium telluride detector cooled by liquid nitrogen. Internal calibration is provided by a reference black body. The thermal camera was externally calibrated on shore using a blackbody to correct for variable response across the image (Tamba et al, 1997). Fourier transforms were used to remove "banding" noise in the image.

The commercial radiometer for measuring sky temperature is manufactured by the Tasco Corporation and uses a thermopile as a detector. This was externally calibrated on shore using a blackbody. Donlon et al. (in press) describe the techniques used to calibrate this instrument and the resulting accuracy possible. The effect of direct solar warming on the calibration of the radiometer requires further investigation but, for the purposes of measuring sky radiance, the instrument fulfils the 1.0K accuracy requirement.

The thermal camera is subject to sunglint saturation when the image of the sun is reflected off the sea surface in the field of view. Saturated pixels were masked and removed from the data. In addition, a nearest neighbour test of adjacent pixels was performed to check for lesser levels of sunglint contamination. Adjacent pixels were rejected if their value was outside the standard deviation from the image mean.

The thermal camera brightness temperatures require a correction for reflected sky radiance due to the non- unity emissivity of the sea surface. The emissivity varies with wavelength, look angle and wind speed/surface roughness. In order to extract the SST, the emissivity was integrated over the wavelength response of the radiometers using the values given in Masuda et al, 1988, interpolated for a 23 look angle. This was convolved with the response curves in figures 2 and 3 and used in the Planck function. This is given below:

Where camera and sky are the response functions of the thermal camera and sky-pointing radiometer (Figures 3 and 2), is the sea surface emissivity across the spectral bandwidth of the thermal camera, Tcamera is the observed brightness temperature of the sea surface, Tsky is the observed sky brightness temperature and Tsea is the skin SST.

Test for statistical similarity

Thermal camera data from four transects was processed to produce skin SST data. Each image was acquired at 5 second intervals with the research vessel travelling at 1 to 2 knots. This enabled a more or less continuous transect of the ocean surface to be recorded by the thermal camera. Approximately 6000 images were processed to produce SST.

A one-way analysis of the variance (ANOVA) between images was performed using the F test statistic described in Lindman, 1974. The F test statistic is a measure of the statistical similarity between two populations of data. The F test statistic calculations for populations of equal sizes n are shown in Table 1. The elements of each sample are indexed by the subscript i indicating which sample population there are in, and by j (where j: 0 to n). There are I populations being tested with overall total of N elements. After applications of the sunglint test, not all the images have the same number of pixel elements and the one-way ANOVA for unequal sample sizes from Lindman, 1974 was applied in such cases.

  RS SS df MS F
m   T2/N      
bet (1/n)iti2 RSbet-SSm I-1 SSbet/(I-1) MSbet/MSw
w   RSt-RSbet N-I SSw/(N-I)  
t ijXij2 RSt-SSm N-1    

Table 1: Summary of calculations for one-way F test, equal sample sizes; ti = j Xij; T = i ti = i j Xij (from Lindman, 1974)

The F test has limitations in that the populations being tested must be near-normal, have a near-zero kurtosis and equality of variances. This is a strict criterion when the ANOVA is performed on 2 populations. However, for ANOVAs performed on 3 or more populations the F test is robust with respect to non-normality and equality of variance (Lindman, 1974).

The unequal sample size ANOVA is very sensitive to unequal variances between test populations necessitating the use of the test given in Lindman (1974), p44. The test measures the effect of unequal variances on the F test result. When is near unity, the F test is not affected by unequal variances. F values for 0.9>>1.1 were rejected. The F test was performed on combinations of 3 images from each transect. The maximum difference in time between acquisition of the 3 images in a test combination was 1 hour. A histogram of the F test result is shown in figure 4.

Figure 4: F test results for all 3 image combinations

Overall, 133444 three-image combinations were subjected to the ANOVA test. The F-distribution has a value of 4.61 for numerator degrees of freedom 2 and denominator degrees of freedom 158352. This represents a 99% confidence probability that any F-test values greater than 4.61 represents populations which are statistically dissimilar. 52.4% of the F-test results were greater than the 4.61 F-distribution value. Therefore just over half the 3 image combinations tested can be said to be statistically dissimilar in their spatial variability.

Comparison with atmospheric parameters

An analysis was now performed to compare the F test value with the maximum deviation of corresponding measurements of atmospheric parameters. For each three-image test set, the range of coincident wind speed, humidity, air pressure, longwave sky flux, short wave solar flux and air temperature was found. The mean SST of each image was subtracted from the bulk temperature probe to give a value for the skin effect. The range of the skin effect for each 3 image test set was found.

The absolute value of these ranges for each test set was plotted against that test set's F value for 2 cases: F>4.61 and F4.61. The mean values of the range in each case were tabulated to ascertain if greater variation of a particular atmospheric parameter implies greater spatial variability of skin SST. The results are shown in Table 2 and the corresponding plots for solar flux variability shown in Figures 5 and 6.

Parameter Mean Variation from mean Instrument accuracy
Wind Speed ms-1 0.82 0.66 0.1
Wind Speed ms-1 0.38 0.32 0.1
Solar Flux Wm-2 236 203 1
Solar Flux Wm-2 49 72 1
Air Temp C 0.30 0.28 0.1
Air Temp C 0.07 0.09 0.1
Rel. humidity % 2.06 1.07 0.1
Rel. humidity % 0.57 0.78 0.1
Sky Temp C 1.50 1.07 <1.0
Sky Temp C 0.93 0.79 <1.0
Skin effect C 0.12 0.19 0.2
Skin effect C 0.09 0.08 0.2

Table 2: Mean variation of atmospheric parameters for statistically similar(clear) and dissimilar(shaded) 3-image test sets

Figure 5: Absolute solar flux differences for F4.61

Figure 6: Absolute solar flux differences for F>4.61

In all cases the mean variability of the atmospheric parameter is greater when the F test value is above the 4.61 threshold of statistical dissimilarity. The standard deviations from the mean are a high percentage of that mean. In the domain of F test values above the 4.61 threshold, the regime of the standard deviation includes the mean of the atmospheric parameters for F tests below the 4.61 threshold. This implies that while on average a high variation in say, solar flux, will result in F exceeding 4.61, this is not always the case. This suggests that it may take a particular combination of changes in atmospheric parameters to precipitate a change in skin SST variability.

Attention is drawn to instrumental accuracy limitations in these results. Although this suggests one should be wary of drawing conclusions drawn from the skin effect and air temperature means, the large size of the sample from which the means are drawn must also be considered. The small magnitude of the mean variation in atmospheric parameters reflects the similar conditions under which the data was gathered. Solar flux is the only parameter for which a wide range of data values was recorded.

Conclusions and Further Work

A test to compare the statistical variability of a set of 3 small scale thermal camera skin SST images has been applied. The test has been performed on over 130000 such sets of images temporally separated by no more than 1 hour. 52.4% of these combinations were found to be statistically dissimilar to a 99% confidence probability by an ANOVA method. This suggests that over short time scales of less than one hour there is a relatively high degree of small scale spatial variability in the skin SST.

Comparisons with coincident measurements of atmospheric parameters shows that statistical skin SST variability is linked to variations in wind speed, short wave solar flux, humidity and longwave downwelling sky radiation. The bulk-skin temperature difference and air temperature can be tentatively linked to small scale skin SST variability. However the small range of values for the skin effect and air temperature in the data set used in this study mean that further investigation is required in this area.

Further study is indicated into the relationship between atmospheric parameters and skin SST variability. Data from the Mutsu Bay experiment in which they is a greater variation in atmospheric parameters may form a basis for this. In the longer term the study will be expanded to regional and global in situ data sets combined with ATSR skin SST data.

References

Donlon, C.J., 1994, 'An investigation of the oceanic skin temperature deviation', PhD thesis, University of Southampton, U.K.

Donlon, C.J., Keogh, S., Robinson, I., Ridley, I.K., Sheasby, T., Barton, I., 1997, The use of solid state radiometer instrumentation to determine sea surface skin temperatures, in press.

Hasse, L,. 1970, The sea surface temperature deviation and the heat flow at the air-sea interface, Boundary Layer Meteorology, 1, 368-379.

Lindman, H.R., 1974, Analysis of variance in complex experimental designs, published by W.R. Freeman and Co.

Ma, C.C., Mechoso, C.R., Arakawa, A., Farrara, J.D., 1994, Sensitivity of a coupled ocean-atmosphere model to physical parameterisations, Journal of Climate, 7, 12, 1883-1896.

Masuda, K., Takashima, T., Takayama, Y., 1988, Emissivity of pure and sea waters for the model sea surface in the infrared window regions, Remote Sensing of Environment 24, 319-329

Parkes, I.M., Sheasby, T., Llewellyn-Jones, D.T, Mutlow, C., Nightingale, T., Zavody, A., Yokoyama, R., Tamba, S., Donlon, C, 1997, The Mutsu Bay Experiment - an investigation into the physical processes of the ocean-atmosphere boundary using ATSR data and in situ measurements, in press

Saunders, P.M., 1967, Aerial measurement of sea surface temperatures in the infrared, Journal of Geophysical Research, 72, No. 16, 4109-4117.

Schlussel, P, Emery, W.J., Grassl, H., Mammen, T., 1990, On the bulk-skin temperature difference and its impact on satellite remote sensing of sea surface temperature, Journal of Geophysical Research, 95, No. C8, 13341-13356.

Tamba, S., Oikawa, S., Yokoyama, R., Ridley, I.K., Parkes, I.M., Llewellyn-Jones, D.T., 1997, The relationship between SST measured from ATSR to the spatial and temporal behaviour of the ocean skin temperature as observed with an in situ thermal infrared camera, in press

Acknowledgements

The authors wish to thank the Aomori Aquaculture Research Centre for their continued support and Mr Hosokawa for the use of his vessel the Dai-Ni Misago. The authors also wish to thank Chris Mutlow at Rutherford Appleton Laboratory and Craig Donlon at Southampton University UK for providing instrumentation and support. We acknowledge NASDA, Rutherf