|
|
|
Oceanography
|
Classification of algal bloom types
From remote sensing reflectance
Table 3: Minimum distance classification in the key value space for SeaWiFS
| |
|
Assigned Class(%) |
|
|
| |
|
1 |
2 |
3 |
4 |
5 |
6 |
7 |
8 |
9 |
|
Total
(%) |
Actual
Class |
1 |
100 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
|
100 |
| 2 |
0 |
69.6 |
0 |
0 |
0 |
0 |
0 |
0 |
30.4 |
|
100 |
| 3 |
0 |
0 |
100 |
0 |
0 |
0 |
0 |
0 |
0 |
|
100 |
| 4 |
0 |
0 |
0 |
100 |
0 |
0 |
0 |
0 |
0 |
|
100 |
| 5 |
0 |
0 |
0 |
0 |
100 |
0 |
0 |
0 |
0 |
|
100 |
| 6 |
0 |
0 |
0 |
0 |
0 |
100 |
0 |
0 |
0 |
|
100 |
| 7 |
0 |
0 |
0 |
0 |
0 |
0 |
100 |
0 |
0 |
|
100 |
| 8 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
100 |
0 |
|
100 |
| 9 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
100 |
|
100 |
Average Overall Accuracy 96.6 %
Table 4: Minimum distance classification in the key value space for MERIS
| |
|
Assigned Class(%) |
|
|
| |
|
1 |
2 |
3 |
4 |
5 |
6 |
7 |
8 |
9 |
|
Total
(%) |
Actual
Class |
1 |
100 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
|
100 |
| 2 |
13.0 |
87.0 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
|
100 |
| 3 |
0 |
0 |
100 |
0 |
0 |
0 |
0 |
0 |
0 |
|
100 |
| 4 |
0 |
0 |
0 |
100 |
0 |
0 |
0 |
0 |
0 |
|
100 |
| 5 |
0 |
0 |
0 |
0 |
100 |
0 |
0 |
0 |
0 |
|
100 |
| 6 |
0 |
0 |
0 |
0 |
0 |
100 |
0 |
0 |
0 |
|
100 |
| 7 |
0 |
0 |
0 |
0 |
0 |
0 |
100 |
0 |
0 |
|
100 |
| 8 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
100 |
0 |
|
100 |
| 9 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
100 |
|
100 |
Average Overall Accuracy 98.6 %
Table 5: Minimum distance classification in the normalized spectral value space for SeaWiFS
| |
|
Assigned Class(%) |
|
|
| |
|
1 |
2 |
3 |
4 |
5 |
6 |
7 |
8 |
9 |
|
Total
(%) |
Actual
Class |
1 |
97.5 |
2.5 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
|
100 |
| 2 |
4.35 |
91.3 |
0 |
0 |
0 |
0 |
0 |
0 |
4.35 |
|
100 |
| 3 |
0 |
2.5 |
42.5 |
5.0 |
0 |
0 |
37.5 |
0 |
12.5 |
|
100 |
| 4 |
0 |
20 |
0 |
50 |
20 |
10 |
0 |
0 |
0 |
|
100 |
| 5 |
0 |
0 |
0 |
0 |
77.8 |
22.2 |
0 |
0 |
0 |
|
100 |
| 6 |
0 |
0 |
0 |
0 |
10 |
90 |
0 |
0 |
0 |
|
100 |
| 7 |
0 |
0 |
0 |
0 |
0 |
0 |
100 |
0 |
0 |
|
100 |
| 8 |
0 |
0 |
0 |
0 |
0 |
0 |
50 |
25 |
25 |
|
100 |
| 9 |
0 |
10 |
0 |
0 |
0 |
0 |
0 |
0 |
90 |
|
100 |
Average Overall Accuracy: 73.8%
Table 6: Minimum distance classification in the normalized spectral value space for MERIS
| |
|
Assigned Class(%) |
|
|
| |
|
1 |
2 |
3 |
4 |
5 |
6 |
7 |
8 |
9 |
|
Total
(%) |
Actual
Class |
1 |
100 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
|
100 |
| 2 |
0 |
69.6 |
0 |
0 |
0 |
0 |
0 |
0 |
30.4 |
|
100 |
| 3 |
0 |
0 |
57.5 |
15 |
0 |
0 |
15 |
2.5 |
10 |
|
100 |
| 4 |
20 |
0 |
0 |
60 |
0 |
20 |
0 |
0 |
0 |
|
100 |
| 5 |
0 |
0 |
0 |
0 |
100 |
0 |
0 |
0 |
0 |
|
100 |
| 6 |
0 |
0 |
0 |
0 |
0 |
100 |
0 |
0 |
0 |
|
100 |
| 7 |
0 |
0 |
0 |
0 |
0 |
0 |
100 |
0 |
0 |
|
100 |
| 8 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
100 |
0 |
|
100 |
| 9 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
0 |
100 |
|
100 |
Average Overall Accuracy: 87.5%
The accuracy of minimum distance classification using the normalized spectral reflectance values of the SeaWiFS sensors is only 73.8%. With four additional bands of the MERIS sensor, the accuracy improves to 87.5 % (see Tables 5 and 6). This improvement is expected, as the additional bands are located around the chlorophyll-a absorption band at 670 nm, which helps to discriminate between the algal bloom classes. The transformation of the reflectance spectra into the key value space using a simple matrix multiplication operation improves the classification accuracy to 96.6% and 98.6% for SeaWiFS and MERIS respectively (see Tables 3 and 4). It is noted that for both SeaWiFS and MERIS sensors, the spectra from eight out of nine classes are correctly classified after transformation into the key values. The only class that performs poorly is class 2 (Trichodesmium).
Figure 2: Samples of MERIS (left) and SeaWiFS (right) reflectance spectra of class 2 (Trichodesmium)
Samples of the reflectance spectra of class 2 (Trichodesmium) are shown in Fig. 2. For MERIS, all the class 2 reflectance spectra have similar shapes in the shorter wavelength (<650 nm) region. The variation is greater in the longer wavelength region. Hence, it is expected that some spectra be misclassified. For SeaWiFS, the within class variation is smaller. However, the class 2 spectra shape is very similar to class 9 and hence many spectra are misclassified as class 9.
5. Concluding Remarks
We have presented a technique for classification of algal blooms types from remote sensing reflectance. This technique is based on a linear transformation of the normalized reflectance spectra into a "key value" space. The success of this technique depends on the availability of spectral reflectance signatures of known algal bloom classes. The key vectors required for constructing the transformation matrix are derived from this set of reflectance signatures. The present database of algal bloom signatures used in this study have been accumulated during a 3-year period of water sampling in Singapore waters and in the Manila Bay. The classification technique is tested on the simulated data for the current SeaWiFS and future MERIS ocean color sensors. The simulated data are constructed from in-situ radiance data measured using a handheld spectrometer. Atmospheric effects are not included in the simulation. Hence, it is assumed that atmospheric correction has been done before the classification technique is applied. This study show that the spectral bands of the current SeaWiFS sensor is sufficient for algal bloom classification, while the MERIS sensor provides certain advantages in identifying different algal bloom types.
Acknowledgments
The authors acknowledge support from SPOT Asia Pte. Ltd, in the form of a grant awarded by the Regional Institute of Environmental Technology (RIET) under the Asia-EcoBest Work Programme '99. Some of the data used in this study were obtained during field trips conducted with partial funding from the National Space Development Agency of Japan (NASDA) through a joint NASDA-ESCAP project entitled "National capacity building for sustainable environment and natural resources management through research and studies on the uses of ADEOS data". The authors would like to thank Dr Michael Holmes of the Department of Biological Sciences and the Tropical Marine Science Institute (TMSI), National University of Singapore (NUS); Dr Serena Teo of TMSI, NUS; Dr Karina Gin of the Department of Civil Engineering, NUS; and Dr. I-I Lin (currently at the National Centre for Ocean Research, National Taiwan University) for their contributions in the water sampling field trips in Singapore. The assistance rendered by Prof. Rhodora Azanza and Ms Arlynn Gedaria of the Marine Science Institute, University of philippines at Diliman during the field trips in Manila Bay is gratefully acknowledged.
References
-
Cullen, J. J., A. M. Ciotti, R. F. Davis, and M. R. Lewis (1997), Optical detection and assessment of algal blooms, Limnology and Oceanography, 42(5), 1223-1239.
- Danaher, S. and E. Omongain (1992), Singular value decomposition in multispectral radiometry, Int. J. Remote Sens. 13, 1771-1777.
- Kahru, M. and B. G. Mitchell (1998), Spectral reflectance and absorption of a massive red tide off southern California, Journal of Geophysical Research 103(C10), 21601-21609.
- Liew, S. C., I-I Lin, H. Lim, L. K. Kwoh, M. Holmes, S. Teo, S. T. Koh, and K. Gin (2000), Tropical algal bloom monitoring by sea truth, Paper presented at 28th International Symposium on Remote Sensing of Environment, 27 - 31 March, 2000, Cape Town, South Africa.
- Lin, I-I, S. C. Liew, H. Lim, L. K. Kwoh, M. Holmes, S. Teo, S. T. Koh and K. Gin (1999), Classification of Tropical Algal Bloom Types By Sea Truth Spectral-Radiometric Data, Proc. 20th. Asian Conference for Remote Sensing, 22-25 November 1999, Hong Kong, China.
- Richardson, K. (1997), Harmful or Exceptional Phytoplankton Blooms in the Marine ecosystem. Adv. Mar. Biol. 31, 301-385.
- Sathyendranath, S., F. E. Hoge, T. Platt, and R. N. Swift (1994), Detection of phytoplankton pigments from ocean color: improved algorithms, Applied Optics 33, 1081-1089.
|
|
|
|
|
|
|
