GISdevelopment.net ---> AARS ---> ACRS 1990 ---> Oceanography

Sea surface temperature estimation by AVHRR spilt window function - A case studies by using Mutsu Bay

Ryuzo Yokoyama, Sumion Tanba
Department of Computer Science
Faculty of Engineering Iwate University
4-3-5 Morioka Iwate Japan 020


By using the match-up data sets of the AVHRR brightness temperatures of NOAA-9 and the buoy sea surface temperatures in Mutsu window functions (SWFs) were evaluated The match ups were carefully screened from the HRPT scenes of all seasons in 1985-1988 The temporal and the spatial coincidences are within 30minutes and one pixel resolution respectively The RMSDs were in the range of 0.69~2.31 Most SWFs accompanied with larger RMDs increased their values due to larger biases The statistics of errors delicately depended upon the coefficients of SWFs which were determined by atmospheric profiles or match up data set used in the derivation .for better SST estimation the SWF should be calibrated by a regional match up data of the investigation.

Introduction
The sea surface temperature (SST) imagery remotely sensed by the Advanced Very High Resolution Radiometer (AVHRR) aboard the NOAA satellite series have been conveniently used in various fields e.g oceanography meteorology fishery etc As the thermal radiation from the sea surface is exposed to atmospheric effects many works have been done to investigation its transmittance mechanism A multiple window method (WMWM) as been developed as one of the most promising error correction algorithms (Anding and Kauth (1970 ) Maul and Sidran (1970) Prabhakara et al. (1974) McMillin and Crosby (1975) deshamps and Phulphin (1980) for the temperature detection the AVHRR is equipped with three spectral bands of Ch 3 (3.5~ 3.9m m) ch.4 (10.5~11.5 m m) and ch.5 (11.5.~12.5m m) which were selected according to the MWM Nowadays a simplified algorithm were selected according method which uses the ch. 4 and the ch.5 brightness temperatures have been popularly used.

Over those various SWfs have been proposed as known in Table -1 The accuracy of the SWf is evaluated by a root mean square of deviations (RMSD) between the sea truth and the estimated SSt The RMSd of an SWF depends upon both truth and the a match up data set used in the validation rest In most cases their RMSDs were claimed in the range of 0.5~1.2 oC.

As the atmospheric profiles are effected by regional weathers and /or climates an SWF obtained at a region might not be applicable to other regions The quality of the match -up data set are directly related to the error statistics in the results .It is necessary to proceed more comparative studies of SWfs by using various data sets on the base of the seasons the geography and the data quality.

This paper is concerned with a comparative study of the fourteen published SWFs by using the match up data sets of Mutsu Bay This is an extended results of Yokoyam a and Tanba (1988) with increased number of match -ups.

Published split window functions.
A general structure of the SST estimation function via the ch.4 and the ch.5 brightness temperatures (X4 and X5= respectively) can be

Y = a X4 + b (X4 - X5) + g---------------------(1)

Where Y is the estimated SST the coefficients of a, band g. Depend upon various factors e.g atmospheric effects air sea interacting effects improper sensor calibration contamination in the optical systematic the SST estimation function contamination in the optical is characterized by the value of a it is theoretically induced to be one from the transmission model of the ch 4 and the ch.5 radiations (Mc.millin) and Crosby) the values can be modified around one by minor adjustments Table 1 shows a list of SWFs published in these a list of SWFs published in these ten years.

Setting up of the match ups.
As shown in figure 1 Mutsu Bay is situated in the northern end of Honshu Japan The Mutsu Bay automatic buoy system is composed of six fixed buoys have been measuring marine environmental items at each every hour on the hour . The temperatures at 1 m depth were used as the sea truth SST in the analysis .Its accuracy is claimed to be ± 0.1.OC in the design specification of the buoy system .

Out of the archived AVHRR data of NOAA-9 received in 1985-1988 at the institute of industrial Science 78 scenes listed in table 2 were picked up as they had cloud free around the bay after applying due their latitude and longitude coordinates The errors were evaluated within one pixel resolution which is about 1.1 Km x 1.1 km at the nadir.

In order to insure the quality of the brightness temperatures in the match up data set we applied several checks as follows
  1. Careful visual check for cloud free and noise free buoy position .


  3. Restriction of the buoy position within 40 degree of the satellite zenith angle.


  5. Homogeneity check of brightness temperatures the standard deviation of the brightness temperatures in 3x3 pixels of a neighborhood buoy position should not exceed 0.2OC,
Finally total of 309 buoy positions were screened out the brightness temperature of ch.4 and ch 5 say X4 and X5 were specified to be mean values in the 3x3 pixel window The recovering procedures of the brightness temperatures from the ch.4 and the ch. 5 AVHRR data were completely followed to those described in Lauritson et. Al (1979) and brown et al (1985).

To the pair of X4 and X5 at each buoy position the buoy SST (y) measured at the nearest time to its overpass was combined to set up a match by using a regression analysis an outlier test was applied to the det of screened match ups there appeared some outliers then the circumstances at their collections were examined by referring to various meteorological observation records Two abnormal cases were found as follows.
  1. Contamination by forecast fire smoke in the summer of 1988 large forecast broke out continuously in eastern Siberia some match ups under the smoke were outliers since the smoke was thin and homogeneous over the bay it was not detected in the former tests.

  2. Strong radioactive cooling condition some match-ups under strong radioactive cooling condition in midnight scenes in autumn were outliers . There might were have been a steep change in the vertical temperature profile between the sea surface of the AVHRR and the 1m depth of the buoy measurement.< /li>< /li>
The causes of the above outliers were evident we included the all scenes to which the outliers belonged and finally got 276 selected match ups .

Form the preparation of the match-ups. The temporal coincidence in each match-up is within 30 minutes since the measurement interval of the buoys is none hour. The spatial coincidence is within one pixel.

Results.
By using the final match up data set the accuracy of SWFs in Table 1 was validated the results are shown figure 2 the RMSD is related to the bias and the scatter as

RMSD2 = s2 + µ2

According to the amount of µ , s and RMSD the SWFs are classified in to the three groups The first is to have small values in both the scatter and the bias. Consequently their RMSD's remained small SWF-5 (Barton (1983) is a typical one it has the smallest RMSD of 0.69oC the situations are similar to the results for SWF-10 ( Barton(1985) ) and SWf-11~13 ( McClain ) et al. (1985) of which RMSds were less than 0.80C .

The second group is such that the SWfs have small scatters but rather larger biases Subsequently their RMSDs become rather large. SWf-1 and SWf-2 9Deschamps and Phulpin (1980)) SWf-7 (McMillan and Crosby (1984)) (SWf-9 (Leewelun-Jones et al. (1984) and SWF-14 ( Mc Clain ) et al. (1985)) belong to this group SWF-1 has a very small scatter of 0.62 but has a large bias of 1.810C the thirds group which includes SWf-3 SWF-4 SWF-4 SWf-6 and SWF-8 is such that SWFs have rather large values for both the scatter and the bias . SWF-3 and SWF-4 were the early results by McMillan et al (1982) but they were revised later to be SWf-11 and SWF-12 which provided smaller RMSDs.

The applications of SWF-3 SWF-11 and SWF-13 are originally specified to daytime data and SWF-3 and SWF-11 are for NDAA -7 and SWF-13 is for NOAA -9 When they were validated by the daytime data only there did not appear remarkable differences in the results on the other hand SWF-4 SWF-12 and SWF-14 are for the nighttime data set but the results were almost some even though were validated by the nighttime data.

Each SWF has Delicate variations in its coefficients and provides various errors statistics It might be difficult to specify completely reasonable SWFs for the Mutsu Bay SSt estimation from the view point of small RMSDs SWF-5 and SWF -12 have provide RMSDs less than 0.80C estimation functions in most cases however larger RMSDs can be reduced to less than 0.80C only by adjustment heir biases that is the slightly sensitive there are various factors to bring larger biases i.g., recovering process of the brightness temperatures at the sea skin and the AVHRRdepth regional dependency of atmosphere profiles deriving process of SWFs and the quality of the match up samples.

Conclusion
The total Mutsu Bay set is composed of 276 match ups collected in all seasons of 1985~1988 and its quality is characterized by
  1. the rigid screening of cloud free and noise free buoy positions.

  2. the excellently temporal and spatial coincidence and

  3. the appropriate exclusion of outliers by referring to the local meteorological records.
The accuracy of the fourteen published SWFs were evaluated by using the test data sets their RMSDswere in the range of 0.69~2.31 0C in most cases however the caused the scatters remained less than 0.8 0C and the larger RMSDs were caused by larger biases In general the Scatter of errors in the published SWFs were robust to the test data sets but their biases were sensitive this means that the RMSDs can be reduce on ly by adjusting the constant term in the SWF .Some SWFs are accompanied with the specifications to the Mustu of applications but they not necessarily consistent with the Mutsu Bay method test data set Although the effectiveness of the spilt window method was confirmed but the coefficients of the SWf should be calibrated by regional match up data set of the application.

Reference
  1. Anding .D. & Kauth D., Remo Sens of Envir ,. 1,217-200 , 1970.

  2. Barton i., Quar. J. of Roy. Metor Soc .109, 365-378 1983.

  3. Barton I., J. of Clim & appl metor 24 508 516 1985

  4. Brown o,.et. al J. of geophy Res. 90 11,667 -11. 6777 1985

  5. Deschamps p,.e.t al. bound Lay Metor 18 131-143 1980

  6. Laurelton let al Q. J of R. Metro Soc, 110, 613,631,1984

  7. Llewellyn-Jones et al M. J of Geophy Res. 78 109-1919-19973.

  8. Maul, G.A., & Sid ran M,. J of Geophy soc 13 439 451 1983

  9. McClain E,. Oceanography from space 73-85 Plenum press 1981.

  10. McClain w,. et..al., III co spar meeting
















Table 1: List of the fourteen published split window functions used in the analysis.
SWF ID. Function (remark) Fundamental data for SWF Derivation Original paper.
SWF-1 Y-1.00X4+2.098(X4 - X5) - 1.280 Monthly mean radiosonade profiles at station K of 45N-16W. Dechamps & Phulpin (1980)
SWF-2 Y-1.00X4+2.626(X4 - X5) - 2.280 Lowtranstandard atmospheric model
SWF-3 Y-1.035X4+3.046(X4 - X5)-1.209(for nighttime data (NOAA-7)) (same comments to those for SWF - 11 ~ SWF-14) Mc-clain of al (1982)
SWF-4 Y-1.076X4+3.168(X4 - X5)-2.310 (for night time data NOAA-7
SWF-5 Y-1.000X4+2.830(X4 - X5)-0.070 36 mid altitude & subtropical maritime radiosondeprofiles Barton (1983)
SWF-6 Y-1.000X4+3.350(X4-X5)+0.3201 Low Tran standard atmospheric model Maul (1983)
SWF7 Y-1.000X4+2.702(X4-X5)-0.582 Regression of 293 match ups in 20~70N and 278~301K Mcmillin &Crosby (1984)
SWF8 Y-1.000X4+2.852(X4-X5)-2.049 39 tropical maritime radiosonde profiles Liewellyn-Jones et al. (1984)
SWF9 Y-1.056X4+2.852(X4-X5+)-2.040 61 north Atlantic maritime radiosonde profiles
SWF10 Y-1.000X4+2.760(X4-X5)-0.420 40 Tropical and subtropical martime radisonde profiles Barton (1985)
SWF11 Y-1.035X4+2.578(X4-X5+)-0.599 Seasonally and geographically diverse set of 59 cloud free maritime radiosonde profiles coefficients are dated are updated the basis of a larger and more seasonally and geographically representation set of close match ups between satellite data and grafting buoy data. Mc-clain et al (1985)
SWF12 Y-1.035X4+2.579(X4-X5)-2.040
SWF13 Y-0.986X4+2.671(X4-X5)+0.525 (for daytime data NOAA7)
SWF14 Y-0.986X4+2.668(X4-X5)+ 0.789 (for night time data NOAA-9)