Comparative Performance of SST Algorithms In the Tropical Ocean Using OCTS Data
K. Abdullah, M. Z. MatJafri and N. M. Saleh
Universiti Sains Malaysia
11800 Penang, Malaysia
Tel: 604-6577888 Fax: 604-6579150
E-mail:khirudd@usm.my,
mjafri@usm.my, nasirun@usm.my
A. Bahari
Malaysian Meteorological Service
46667 Petaling Jaya
Selangor, Malaysia
Tel: 603-7569422 Fax: 603-7550964
E-mail:alui@kjc.gov.my
Keywords: algorithm, emissivity, sea surface temperature, transmittance
Abstract
A two-channel algorithm for measuring sea surface temperature (SST) using non-regression coefficients was applied to the Ocean Colour Temperature Scanner (OCTS) data from the Advanced Earth Observation Satellite (ADEOS). The algorithm is based on the fact that the recorded infrared signal at the sensor is composed of three terms, namely, the surface emission, the upwelling radiance emitted by the atmosphere, and the downwelling atmospheric radiance that is reflected at the sea surface. This algorithm requires the atmospheric transmittance values of two thermal bands. In this study the transmittance function for each band was modeled using the MODTRAN code and radiosonde data. The expression of transmittance as a function of zenith view angle was obtained for each channel through regression of the MODTRAN output. The in-situ data (ship collected SST values) were used for verification of results. Contour maps of the in-situ SST values were generated because clouds covered the measured locations. Cloud contaminated pixels were masked out using the standard procedures. The cloud free pixels were extracted and converted to brightness temperature values and then substituted in the proposed algorithm. The appropriate transmittance value for each channel derived earlier was then assigned in the calculation. The correlation coefficients and the root-mean-square deviations between the computed and the ship-collected values were determined. The results were also compared with the results from OCTS multichannel sea surface temperature (MCSST) algorithm. The performance of this algorithm is comparable with the established OCTS algorithm. SST map was created and comparison with the OCTS algorithm generated map was made.
1.0 Introduction
Sea surface temperature (SST) algorithms have been determined by two basic techniques: i) regression between coincident satellite brightness temperatures and surface measurements, and ii) theoretical atmospheric transmission model with a set of representative vertical profiles of atmospheric temperature and absorbing constituents is used (Barton 1995). Various algorithms for performing this task have been developed; these include the single infrared channel method, the multichannel method and the multiangle method. The operational SST algorithms using AVHRR data established their coefficients through regression techniques. Recently, the MCSST algorithm has been developed by NASDA for OCTS data using similar approach.
Application of regression technique requires a sufficient amount of coincident in-situ data with the remotely sensed data. The requirement is not easily achievable in the equatorial region due to unfavorable sky conditions. Therefore in this study, the algorithm that has been derived theoretically which does not require in-situ SSTs for calibration was used. This algorithm is designed to measure SST using thermal images from any satellite. The purpose of this study is to test our developed algorithm using OCTS data sets. The performance of this algorithm was compared with the NASDA MCSST and in-situ data.
2.0 Data Sets
Three OCTS TI level 1B scenes covering the South East Asia region were selected based on availability of cloud free pixels. They were captured (daytime) by ADEOS satellite on 22 March 1997, 21 April 1997 and 13 June 1997. In-situ SST data collected by sea cruises (under the Voluntary Observing Ship Programme of the World Meteorological Organization) on these dates were supplied by the Malaysian Meteorological Service.
3.0 Data Calibration
The conversion from digital number, DN, to radiance, L (mW cm
-2 str
-1 mm
-1) was performed using the information given by NASDA. The radiance values were converted to brightness temperature values, T (K), for band 10, 11 and 12 of OCTS data by using the published conversion table (table IX of Shimada 1998). Analysis of the tabulated values for the temperature range from 273 K to 315 K, showed that the relationship could be accurately described by the polynomial equation
Ti = ao + a1Li +a2Li2 + a3Li3 (1)
The coefficients were determined through regression.
4.0 Cloud Masking Techniques
The thermal infrared bands were used because the near infrared bands contained some noise problems. Three techniques were employed, (a) the infrared threshold test or gross cloud check, (b) the spatial coherence technique, and (c) the channel difference (Saunders and Kriebel, 1988). For each image date, several small sub-scenes (channels 10,11,12) containing clouds and cloud free water pixels were extracted for detailed study using each technique mentioned above. Digital numbers (DN) were used in the analysis instead of brightness temperature values to simplify the data processing. After the data were passed through the series of tests described earlier, the combined cloud masks were finally obtained.
5.0 Sea Surface Temperature Algorithm
Based on the radiative transfer equation in the atmosphere for a cloud free atmosphere under local thermodynamic equilibrium, a single channel algorithm can be derived for measuring sea surface temperature for infrared band i.
Ts =Ti + (Ti-Ta)bi/ai + D(Ti)(1-ai-bi)/ai (2)
where T
i = brightness temperature of band i
T
a= temperature of atmosphere
D(T
i= L
i/[dL/dT(T
i)]
a
i =
eiti
b
i= (1-
ti)(1 +
ti -
eiti)
ei= emissivity of sea water in band i
ti= atmospheric transmittance
By writing the equation for another band j, a split window expression or dual-channel algorithm (Abdullah 1994) for measuring sea surface temperature, T
s can finally be obtained
The algorithm requires two environmental parameters, (i) the atmospheric transmittance and (ii) the emissivity of seawater. The transmittance was modelled using the MODTRAN code. Radiosonde data acquired on the dates of OCTS scenes were used in the MODTRAN calculation for the atmospheric model. For each channel, the view angle from the top of the absorbing atmosphere was varied in steps of 5 degrees from 0 to 65 degrees. The average transmittance values and their corresponding zenith angles were noted. Inspection of the results suggests that the relation can be approximated by the expression
t(q)=to + t1cos(q) + t2cos2(q) t3cos3(q)
(4)
where
q is the satellite zenith angle at the pixel location. The coefficients were determined from regression analysis.
Various emissivity values have been used in SST retrieval algorithms. The earlier algorithms used
e = 1. Singh (1984) assumed 0.98 in his calculation. The spectral dependence of emissivity, where
e4=0.992 and
e5=0.989 was considered by Becker and Li (1990).
