Sea surface effect in sea surface temperature detection
Ryuzo Yokohama, Takashi souma, Sumio tanba
Iwate University, Morioka, Iwate, 020 Japan
Abstract
The sea surface effects in the sea surface temperature (SST) detection by satellite Remote Sensing were investigated by using the match-up data set composed of the NOAA-9/AVHRR brightness temperatures and the sea surface temperature by fixed buoys in Mutsu Bay. By applying the regression analysis to the data set of 390 match-ups and its standard deviation of residues was evaluated to be 0.59C. The meteorological data measured at the buoy were referred to investigate the causes of large resides. Then it was found that large residues appear under the conditions of strong sunshine or strong radiative cooling on calm days. Under the conditions, it was proved by the data of a vertical temperature profiler buoy that large temperature differences between the real sea surface of the brightness temperature and the 1m depth of the sea truth temperature of buoy. By removing the match-ups with larger error, a selected data set with 328 match-ups was prepared and the SST estimation was recalculated. The standard deviation of residues reduced to 0.34C.
Introduction
NOAA/AVHRR is a sensor for the sea surface temperature (SST), which was characteristics: the high radiometric resolution, the wide swath of view, the frequent observation cycle, etc. The atmospheric effect is a dominant disturbance in the SST observation and the split window function method is known as an effective algorithm for the correction. A split-window function (SWF) is described as
Y = X
4 + b (X
4 - X
6) + a (1)
Where X
4 and X
5 mean the brightness temperatures of ch. 4 and ch. 5 respectively. By applying the regression analysis to the match-up data sets, various SST estimation functions have been proposed and their accuracy has been validated in the range of 0.6 ~ 12C (1-8).
This paper is concerned with the investigation of the air-sea interacting effects by using a match-up data set of NOAA-9 and the buoys in Mutsu Bay. It was found by referring to the meteorological data that large residues in the regression analysis appeared when there existed large differences between the air temperature and the buoy SST. The air-sea interacting effect have been suggested as another major factors of errors in the SST estimation.
Figure 1 Geographical location of Mutsu bay and the moored buoy positions.
Figure 2 The scattered diagram of the result of the regression analysis to the total data set.
Figure 3 The scattered diagramof the residue versus(X4-Y)
Figure 4 The scattered diagram of the residue versus (A-Y). A is the air temperature measured on No. 6 buoy.
Figure 5 The scattered diagram of the large residues versus( X4-Y). A large residues is assumed to large if its ablolute values is lager than 1°C.
Figure 6 SSTPB Measurement pole and the positions of temperature detector.
Figure 7 External apperances of SSTPB.
Table 1 Statistics of Data Based In The Regression Analysis
Data used in the analysis
Mutsu Bay is located in the northern end of Honshu, Japan as shown on figure 1. In the bay, there are 6 moored buoys and each buoy measures the water temperature at 1m below the sea surface, which is measured by using PRT detector with the accuracy of 0.1C at every hour. We can use these water temperature as the sea truth SST data.
Out of the NOAA-9/AVHRR data achieved fro 1985 to 1988 at the institute of industrial Science. University of Tokyo, the total number of 110 scenes were picked up as they included cloud-free buoy positions. The capes around the bay were effectively used as the control points in the buoy identification and the spatial errors could be guaranteed within 1 pixel resolution. After careful screening of cloud-free and noise-free buoy positions, the total number of 390 pairs of X4 and X5, were extracted.
A match-up was set up by combining a pair of the brightness temperature with the corresponding buoy SST at the nearest time to the satellite overpass , so that the temporal coincidence in each match-up was 30 minutes. Call this match-up data set the total data set. Its statistics are shown in table 1. The detail process for the data preparation can be referred to Yokoyama and Tanba (9).
Regression function by the total data set
The split window function of eq. (1) is calculated by regression analysis so that eq. (2) is obtained.
Y = X4 + 1.613 (X4 - X5) + 0.914 (2)
Figure 2 shows its scattered diagram. The standard deviation of residues (s) of 0.59C might be very small. Most of the match-ups are distributed along the perfect fit line, but some are accompanies with rather larger residues.
