Study of Sea Surface Temperature (SST) using Landsat-7 ETM (In Comparison with Sea Surface Temperature of NOAA-12 AVHRR) Bambang Trisakti*), Sayidah Sulma*) and Syarif Budhiman*) **)
Remote Sensing Application and Technology Development Center – LAPAN e-mail:
[email protected]
Landsat-7 ETM has wide range of electromagnetic wavelength band, including visible, infrared and thermal bands. The thermal bands (band 61 and 62) of Landsat-7 ETM can be used to detect thermal radiation released from objects on the earth surface. This study has been conducted to develop new Sea Surface Temperature (SST) algorithm for Landsat-7 ETM in comparing with SST of NOAA-12 AVHRR. Digital numbers of Landsat data were converted to effective temperatures based on radiance value, and then the effective temperatures were compared to SST generated from NOAA data using field data calibrated algorithm. Sample of both data were done randomly in several locations. In each location, SST of NOAA was taken from value of 1 pixel size, on other hand, effective temperature of Landsat was taken by calculating mean value of several pixels represented same area as 1 pixel size of NOAA. To reduce effect of seasonal change to SST distribution, 5 data sets of Landsat and NOAA for April, May, June, July and October were sampled. Correlation of effective temperature – SST, and pattern of SST distribution in different season were also analyzed. The results show that Effective temperature generated from Landsat data has polynomial correlation with SST generated from NOAA data, where band 62 of Landsat is looked more effective than band 61 for detecting SST. Furthermore, SST distributions of Landsat generated using new algorithm have been proven to be able to represent SST condition in different season. Keywords: Sea Surface Temperature, Thermal band, Landsat-7 ETM, NOAA-12 AVHRR 1. Introduction Information concerning Sea Surface Temperature (SST) is needed in the assessment of remote sensing for fisheries application, for example: potential fishing zone, and site selection for marine culture (grouper, snapper, seaweed, and pearls). The information of SST is used to identify some marine phenomenon (e.g.: upwelling, front, and eddies) (Nontji, 1987), which are normally these areas are rich of nutrient. SST is also used as indicator of the environment required for the living of some marine biota. In the assessment of potential area for marine culture, some physical parameters with relatively high accuracy of coastal waters are needed (e.g.: SST, Total Suspended Matter (TSM), and Chlorophyll). This information, for example, can be derived from Landsat data which has 30 meters spatial resolution for visible, NIR (Near Infrared), SWIR (Short Wave Infrared) bands, and 60 meters spatial resolution for thermal bands. Therefore, it is need to do the assessment of SST from high spatial resolution, such as Landsat data. The study of SST distribution model mostly has done using low to moderate spatial resolution satellites data, such as NOAA, Fengyun, and MODIS. Because, the result always used for global scale application, such as: potential fishing zone, global warming change analysis, Elnino prediction, etc. A lot of papers and reports have been written about algorithm model for SST using NOAA data, most common algorithm known are algorithm model by McMillin and Crosby (Pellegrini and Penrose, 1986;
Goda, 1993; McClain, 1981 cited in Hasyim et. al, 1996). Hasyim et. al. (1986) reported that algorithm model by McMillin and Crosby can represent the condition of SST distribution in Indonesian waters. The study of SST algorithm model derived from Landsat-5 TM, has been conducted by some researcher (e.g. Mujito et al. (1997) in East Kalimantan), but mostly done based on correlation between insitu measurements and digital number value, using just single time acquisition. Therefore, it is difficult to adopt the model for another location with different time acquisition. This research is conducted to develop SST algorithm model from Landsat-7 ETM (Enhanced Thematic Mapper) in correlation with NOAA-12 AVHRR calibrated data. Correlation done using several set of Landsat and NOAA data, which represent the different condition of waters and different monsoon situation, thus the result can be apply to different condition of waters and different season. 2. Method This research used 5 set of Landsat-7 ETM data and NOAA-12 AVHRR data for April, May, June, July and October as shown in Table 1. All data cover part of Situbondo Regency (northern part of Java Island, path/row 117/65) and part of Banyuwangi Regency (southern part of Java Island, path/row 117/66), thus the research can represent different condition between northern waters of Java Island (relatively still waters) and southern part of Java Island (more dynamics waters).
1
Table1. Landsat-7 ETM and NOAA-12 AVHRR Data Data type Landsat-7 ETM NOAA-12 AVHRR
Time acquisition 26 April 2002, 28 May 2002, 29 June 2002, 6 July 2002, 3 October 2002 26 April 2002, 27 May 2002, 29 June 2002, 7 July 2002, 2 October 2002
Image processing started with geometric and radiometric correction. Radiometric correction done by converted the digital number value in Landsat thermal band (band 61 and band 62) to radiance value, and then changed it to effective temperature value according to Landsat-7 ETM handbook. Method to convert digital number value to effective temperature value is shown in equation 1 and equation 2.
value was derived from counting the average of TLandsat for several pixel in the same region (area) with 1 (one) NOAA pixel. The training sample was done for all data set with total of 100 sample points. The result of SST algorithm model from the correlation between TLandsat and SSTNOAA was used to map SST distribution from Landsat data which represent different location and season (time acquisition).
Equation 1: Lλ = ((LMAXλ - LMINλ)/(DNMAX-DNMIN)) * (DN-DNMIN) + LMINλ
Equation 2: TLandsat = K2/ ln((K1/ Lλ)+1) – 273
where, Lλ
:
DN LMINλ
: :
LMAXλ
:
DNMIN
:
DNMAX TLandsat K1
: : :
K2
:
Spectral radiance watts/(meter squared * ster * µm) Digital Number Spectral radiance which is correlate with DNMIN watts/(meter squared * ster * µm) Spectral radiance which is correlate with DNMAX watts/(meter squared * ster * µm) Minimum value of DN (1 (LPGS Product) or 0 (NLAPS Product)) Maximum value of DN = 255 Effective temperature (Celsius) Calibration constant 1 watts/(meter squared * ster * µm) = 666.09 Calibration constant 2 watts/(meter squared * ster * µm) = 1282.71
Value of LMINλ, LMAXλ, DNMIN and DNMAX can be obtained from header file information (meta data) that come with the Landsat data. The NOAA data converted to SST value using McMillin and Crosby method (Pellegrini dan Penrose, 1986), which has been calibrated with insitu data in Madura Strait waters, Situbondo Regency, shown in equation 3: Equation 3: SSTNOAA = 0.522*[Tw4 + 2.702*(Tw4-Tw5)-273 – 0.582]-13.68
where, SST Tw4 Tw5
: : :
Sea Surface Temperature (in Celsius) Emisivity temperature in Band 4 Emisivity temperature in Band 5
Afterward, the comparison of TLandsat and SSTNOAA distribution pattern was conducted. Training sample was done for both values randomly in different location using 5 set Landsat and NOAA data with relatively same time acquisition (Table 1). For each sample, SSTNOAA value was derived from 1 pixel of NOAA data (1.1 km spatial resolution), and TLandsat
3. Result and Discussion Figure 1 shows TLandsat (Effective Temperature) distribution pattern using band 61, band 62, and the distribution pattern of SSTNOAA for the northern and southern waters of Java and Bali on May 28 th 2004. It is shown that TLandsat for band 61 and band 62 has lower value compare to SSTNOAA, but it has the same distribution pattern with SSTNOAA. High temperature was found in the northern waters of Bali, while lower temperature was found in the southern waters of Java and Bali. To test the level of effectiveness of band 61 and band 62, some training sample was collected from TLandsat61, TLandsat62 and SSTNOAA in 20 locations. The result shows that TLandsat62 value was higher than the value of TLandsat61 with average differences was 0.04 C; moreover the correlation between TLandsat62 and SSTNOAA was higher than the correlation between TLandsat61 and SSTNOAA. The differences of TLandsat for both bands (band 61 and band 62) was not too significant, but from the results can be concluded that band 62 has higher sensitivity sensor and the SST distribution pattern was similar to SSTNOAA. The development of SST algorithm model was conducted using the statistical correlation between TLandsat62 and SSTNOAA. Total of 100 locations of training samples was collected randomly at different location and season using 5 set Landsat and NOAA data as shown in Table 1. SSTNOAA value collected from value of 1 pixel of NOAA (1.1 km spatial resolution) and TLandsat value collected by counting the average of TLandsat62 for several pixels that cover the same region (area) with 1 pixel of NOAA data. Figure 2 shows the correlation between TLandsat62 and SSTNOAA from the sample. The value of TLandsat62 will increase with the increasing of SSTNOAA value. It is polynomial relationship with R2 = 0.65. The algorithm is shown in equation 4.
2
b)
a)
c)
230C
28 0C
a) TLandsat of band 61 b) TLandsat of band 62 c) SSTNOAA
28.50C
32 0C
Figure 1. Distribution pattern of TLandsat (Effective Temperature) of band 61, band 62, and SSTNOAA on June 29th 2002.
32
SST of NOAA (Celcius)
31
30
29
y = 0.0684x3 - 5.3082x2 + 137.59x - 1161 R2 = 0.65
28
27 23
24
25
26
27
28
Effective Temperature of band 62 (Celcius)
Figure 2. Correlation between TLandsat62 and SSTNOAA
3
240C Figure 3.
Pattern of SST distribution of Landsat-7 ETM (Left) and NOAA-12 AVHRR (Right) in October 2002
240C Figure 4.
320C
320C
Pattern of SST distribution of Landsat-7 ETM (Left) and NOAA-12 AVHRR (Right) in June 2002
Equation 4: SST = 0.0684 T3 – 5.3082 T2 + 137.59 T – 1161.2 where, SST = SSTNOAA ≈ SST actual (C) T = TLandsat62 (C) After that, this algorithm was tested in Landsat data which represent different location and season. Figure 3 shows distributon of SST (using new algorithm model) on October for Landsat-7 ETM and NOAA-12 AVHRR. It shows that distribution of SST from both satellites has relatively similar
pattern. On October, the northern waters of Bali has higher SST in the range of 29 – 31 C, while the southern waters of Java has relatively low SST in the range of 26 – 28 C. This condition caused by upwelling process that occurred in southern waters of Java every east monsoon. Figure 4 shows SST distribution in northern waters of Java and Bali in June. The distribution pattern of SST for both data is relatively similar, where in this water, the high temperature differences are not found. The different condition is found in southern waters. The northern waters of Java relatively sheltered and did not have the water
4
mass mixing with the differences of temperature was high. SST of northern waters of Java in the range of 29 – 31 C, and this temperature does not change significantly in entire year. 4. Conclusion The study of SST algorithm model for Landsat-7 data has been conducted using the method approached to SST derived from calibrated NOAA data. The result shows that TLandsat of band 61 and band 62 has similar distribution pattern with SSTNOAA, but Tlandsat62 is more effective rather than Tlandsat61 to measure SST distribution. The result of SST algorithm model is a polynomial relationship from the correlation between TLandsat62 and SSTNOAA, and this model can be applied to obtain SST distribution from data with different seasonal condition in northern and southern waters of Java and Bali. This method is an alternative method in the case of lack of insitu data samples in the same time of data Landsat satellite acquisition, or no higher spatial resolution satellite data than Landsat and already calibrated. REFERENCES
Goda, H.H. 1993. Remote Sensing for Fisheries in India. Asian-Pacific Remote Sensing Journal Vol. 5 No. 2. Hasyim, B.; Khairul Amri and Maryani Hartuti. 1996. Pemanfaatan Data Penginderaan Jauh NOAA-AVHRR untuk Pengamatan Pola Arus Laut dan daerah Potensi Penangkapan Ikan. Kumpulan Makalah Seminar Maritim Indonesia 1996. Jakarta. (In Indonesian) Mujito; M. Husen; Heru Riyanto; Adji Gatot Tjiptono; Suliantara; R.K. Risdianto dan Sudiarto, 1997. Evaluasi Penginderaan Jauh untuk Studi Dasar Lingkungan Wilayah Kerja UNOCAL Indonesia Company Kalimantan Timur. Bidang Litbangtek Eksplorasi. Pusat Penelitian dan Pengembangan Teknologi Minyak dan Gas Bumi. LEMIGAS. Jakarta. (In Indonesian) Nontji, A. 1987. Laut Nusantara. Penerbit Djambatan. Jakarta. (In Indonesian) Pellegrini, J.J. dan I.D. Penrose. 1986. Comparison on Ship Based Satellite AVHRR Estimates of Sea Surface Temperature. Proceeding 1st Australian AVHRR Conference. Perth, Australia.
5
