Malaysian Tiungsat-1 Imagery for Water Quality Mapping


In this study, raw DN values were used as independent variables in our calibration regression analyses. Other forms of water quality algorithms were tested with the data set and their accuracies were compared with that of the proposed algorithm. For each regression model the correlation coefficient, R, and the root-mean-square deviation, RMS, were noted. Many investigators used these parameters in their studies (Keiner and Yan, 1998, Pulliainen et al., 2001). Table 3 shows the comparative performance of the algorithm. The proposed algorithm produced higher correlation coefficient between the predicted and the measured TSS values and lower RMS values compared to other algorithms. With the present data set, the R and RMS values were 0.9256 and 3.6237 mg/l respectively as indicated in Figure 5.

Table 3. Regression results using different forms of algorithms for TSS
Algorithm R RMS
TSS=a0+a1B1+a2B12 0.6334 7.4167
TSS=a0+a1B2+a2B22 0.6305 7.4348
TSS=a0+a1B3+a2B32 0.4098 8.7489
TSS=a0+a1lnB1+a2lnB12 0.6406 7.7235
TSS=a0+a1lnB2+a2lnB22 0.6320 7.5200
TSS=a0+a1lnB3+a2lnB32 0.4083 8.7403
TSS=a0+a1(B1/B3)+a2(B1/B3)2 0.6168 7.5340
TSS=a0+a1(B1/B2)+a2(B1/B2)2 0.3878 8.8225
TSS=a0+a1(B2/B3)+a2(B2/B3)2 0.3302 9.0349
TSS=a0+a1ln(B1/B3)+a2ln(B1/B3)2 0.6134 7.5590
TSS=a0+a1ln(B1/B2)+a2ln(B1/B2)2 0.3962 8.7883
TSS=a0+a1ln(B2/B3)+a2ln(B2/B3)2 0.3528 8.9558
TSS=a0+a1(B1-B3)/B2+ a2((B1-B3)/B2)2 0.6101 7.5842
TSS=a0+a1(B2-B3)/B1+ a2((B2-B3)/B1)2 0.4196 8.6880
TSS=a0+a1(B1-B2)/B3+ a2((B1-B2)/B3)2 0.3818 8.8464
TSS=a0+a1(B2-B1)/(B1+B2)+ a2(B2-B1)/(B1+B2)2 (Waldron M. C. et al.) 0.3969 8.7856
TSS=a0+a1(B2-B3)/(B2+B3)+ a2(B2-B3)/(B2+B3)2 (Waldron M. C. et al.) 0.3511 8.9620
TSS=a0+a1(B1-B3)/(B1+B3)+ a2(B1-B3)/(B1+B3)2 0.6136 7.5582
TSS=a0+a1(B1+B2)/2+ a2((B1+B2)/2)2 (Waldron M. C. et al.) 0.3558 8.9562
TSS=a0+a1(B1+B3)/2+ a2((B1+B3)/2)2(Waldron M. C. et al.) 0.4521 8.5345
TSS=a0+a1(B2+B3)/2+ a2((B2+B3)/2)2(Waldron M. C. et al.) 0.7096 6.7906
TSS=a0+a1B1+a2B2+a3B3+a4B1B2+a5B1B3+a6B2B3+a7B12+a8B22+a9B32 (Proposed) 0.9256 3.6237

Note: B1, B2 and B3 are the digital numbers for red, green and infrared bands respectively

A map of the water quality parameter was then generated using the calibrated proposed algorithm. Then the generated water quality map was geometrically corrected using the cubic convolution method to produce a smoother map. The generated map was filtered using 5 by 5 pixels averaged to remove random noise and then colour-coded for visual interpretation as shown in Figure 6.


Figure 5. Relationship between measured and predicted TSS

The local distribution pattern is shown on the generated TSS map. Higher concentration areas are distributed near the river mouths and the shallow southern region of the channel. The plumes in the river mouths were created by river discharges. In the shallow water, the turbidity might be caused by bottom resuspension due to waves and wind actions. The concentration seems to be relatively lower in the deeper part of the sea. The pattern produced in Figure 6 is consistent with the sea truth data as well as earlier reports (Din, 1995). However, we are unable to establish the validity of the apparent higher concentration on the west coast of Penang Island because of the absence of sea-truth data. Further sea truthing for the area will be done to confirm the values.

Page 4 of 5
| Previous | Next |