|
Calibration of Optical Satellite Data for Coastal Bathymetric Mapping
Raquib Ahmed, Chandan K. Roy, Shitangshu K. Paul * Department of Geography and Environmental Studies, Rajshahi University, Rajshahi 6205, BANGLADESH. Email: raquib@librabd.net *Asian Institute of Technology, Bangkok, Thailand. ABSTRACT Recent introduction of satellite survey has opened up the possibility of the use of optical light for water depth measurement as an alternative method. The water penetration capability of shorter weave length visible channel, such as blue generates radiance that reflects submarine albedo. Calibration using information of energy attenuation due to water column depth and back scattering due to suspended loads in the bay water help to create a 3D model of submarine shelf areas of the Bangladesh coast. The result shows a close similarity to that of sonic bathymetric chart, except the areas where presence of suspended sediment is too high, such as in the upper estuary. In addition to this cheaper and quicker mapping method, the study is also important to track the rapid development of near-coastal offshore lands in the shelf region due to deposition of fluvial sediments that unpredictably generate bumps in the water surge and accelerates devastation. The study interpolated selected sound data points to generated a 3D surface and was refereed to compare the calibrated Landsat blue signal that helped develop a model for image-based submarine surface. The collected sample of bay-waters from several locations determined the impact of sediments in energy scattering and was used in signal calibration algorithm. INTRODUCTION There is about 750 km long coast line bounded by a continental shelf of about 150 km wide (about 112,500 km2). Very close to the country's total area. The shelf region bears good number of creeks, troughs, highs as well as several turbid currents arrive through the rivers and delta channels. The really huge water-carried sediments are transported by the world's two great rivers, the Ganges and the Brahmaputra. The amount and the location of sediment deposition are varied with seasonal change and turbidity direction. So, the island erosion, accretion, appearance of new land and sediment deposition on the near-shore continental shelf region is particularly significant as they directly influence the densely populated coastal districts. There has always been a change of the submarine floor near the coast that significantly influenced coastal navigation and the routing of water surge dynamics generated by frequent coastal cyclones. During any approaching water surge rolling over the continental shelf it gains a sudden extra height and falls on the coast lands which is the normal system. In addition, when the approaching rolling water surge meets local submarine humps the existing water height gains a further rise in water surge height. These local submarine humps are often found to change their position and are sometimes untraced due to irregular bathymetric survey. This inundates the lands more than the expected estimation and carries a potential for damage. Existing coastal bathymetric system largely depends on sound charts supplied by Bangladesh Inland Water Transport Authority (BIWTA) and the Bangladesh Navy using specialized water vessels. Scientifically, this is very time consuming and labor intensive procedure and involves huge organizational and instrumental investment. The last bathymetric survey conducted by BIWTA was about 24 years ago in 1980 (Department of Hydrology, BIWTA, Nov 1979- Jan1980), and it was sufficiently extensive. But however, there has always been a search for an alternative method for existing echo sound bathymetric survey. Present attempt was targeted to develop an alternative methodology of the sound based bathymetric survey and test its usability. Satellite optical data was the principle ingredient. The higher spectrum visible energy sensed by the regularly orbiting satellites was found to be potentially useable for this purpose. This new method inhibits two limitations; (i) one is the signal attenuation due to water column depth, and the other is (ii) the back scattering due to suspended sediment loads. This is particularly prominent in Bangladesh coast. Here lies the main justification of this research that attracted the researchers to investigate whether might be the method that could be adopted for Bangladesh coast. The Landsat satellites 7 ETM+ data was found suitable due to its 16 day revisit period. So, naturally it was found suitable to detect continuous submarine relief change. The information could be used for water surge damage prediction and coastal navigation. The most important aspect of this sensor was its spectral width of 0.45 to 0.515 µm - common standard for the spectral width of blue light. Essentially, two aspects were to be investigated in this research; (i) the processing algorithm of the satellite digital data and mapping submarine relief and, (ii) calibration of the reflectance signals due to scattering and energy decay in the water. There were several similar attempts but none were found to have all-out acceptable result, and were rather focused to specific areas to match with particular local characteristics. The attempts were concentrated around some particular problems such as signal attenuation effect, effect of background variation and back scattering. Among few a notable one could be referred was on mapping benthic habitats and bathymetry near the Lee Stocking Island of the Bahamas (Louchard, E.M., Pamela Reid, R. and Carol Stephens, F., 2003). The depth in the study area was not more than 10 meters. Landsat MSS data was used to identify sea grass areas, where bathymetry was an influencing factor (referring signal attenuation). Compensation to correct error due to low light availability was by the use of a portable hyperspectral imager for low light spectroscopy. The other interesting work on bathymetric charting was done at the Penang Strait in Malaysia where the signal reflectance data were corrected (and compared too) using a sound signal generator on board a small boat (Abdullah, K, MatJafri, M.Z. and Din, Z.B.). The correlation between the two parameters were used as a model to calibrate the deviation aided with signal changes due to multi-band channel differences. Philpot (Philpot, W., http://www.opl.ucsb.edu/hycode.html) indicated that when the water type and bottom reflectance were uniform over the study area, the bathymetric mapping algorithm could be a straightforward one -a one-variable problem which requires minimum ground information data. But the idea offered a limited application as the physical content of water differs from region to region. Edwards, A. indicated that the depth and habitat mapping at the Caicos Bank in the Caribbean was done using differential signal attenuation rate of different signal channels and the back scatter due to suspended matters (UNESCO, 1999). The differences in the signal decay were used to develop a model supplemented with several ground information locations collected through scuba diving. This yielded a good result as the suspended sediment was at a minimum level which is a problem in many other parts of the world, particularly in estuarine regions. The theoretical basis of the present investigation was thus, organized around a slightly different direction - (i) to study the differential presence (and of characteristics) of suspended loads in sea water, and (ii) and check the artificially generated model with the existing sound data (as reference). DATA AND MATERIALS The two important and relevant data are the BIWTA echo sound chart of the Bay of Bengal and the raw satellite digital data of Landsat ETM+ (20 Jan 2001) of the Bay of Bengal. The last BIWTA echo sound survey of the Bay of Bengal was carried out in 1980. Recently Bangladesh Navy began to maintain a similar bathymetric survey program of its own. It takes about a year to cover such a wide water - from about 150 N up to the coast. The bay is very turbulant during about 9 months, from March to November due to high weave. The bay is also known for frequent visit of severe tropical cyclones. For such a wide area, the lower spatial resolution (even up to 1000m) impacts little to view the features. Rather, higher spectral resolution is found useful to separate different features distinctly. Temporal resolution is useful in examining time series analysis - which is particularly important for the present research. The table below gives comparative characteristics of some satellite sensors for visible parts. ETM+ channel blue having spectrum width of 0.45 µm to 0.52 µm was found to be the most suitable. Among other visible spectrums the blue has the maximum water penetration capacity of up to 20m (Lillesand and Kiefer, 2002) due to its shorter wave length but susceptible to back scattering (Raileigh's effect) due to the presence of smaller suspended particles. Also, availability of Landsat data is easier and cheaper than all others. For main bathymetric modeling data of Landsat ETM+ blue channel (single) of Dec 25th 2001 was used. This was the time when under water visibility was found to be maximum in the Bay of Bengal. The raw data was atmospherically and radiometrically corrected from its source. Ground verification was done by scuba diving at 7 locations around the St. Martin's Island (the study area). Photographic visibility of about 20m (maximum in some points) was found in late January. For the correction of back scattering in the water column 2 liters of water samples were collected for pre-selected 10 locations sampled at every 10 km from the Bangladesh coast along 900 05' east longitude. To reach the pre-selected locations in the sea hand GPS (Magellan 2000XL) on board a small mechanized boat was used. The in built error rate of GPS signal reception was ±30 m, although in most of the time it was possible to connect up to 8 satellites. The ideal time would have been 21 Dec. The satellite (the data of which was used) passed on 21 Dec, 2002. Due to hostile sea condition the ground verification survey was fixed on 25th Dec, 2003. The tide condition was similar to that of the original time. To traverse the 70 km route (140 km in round trip) to deep sea it took about 26 hours in a mechanized boat as the vessel movement was influenced by tide direction and high waves.
Ref: Lillesand and Kiefer, 2002. DATA PROCESSING AND RESULT Sound chart generated 3D surface: The data processing algorithm was basically focused on comparing the result to that of the sound chart. The experiment's target was to establish an alternative useable method. The data processing flow chart shows the logical flow of the experiment (Figure 1). The available BIWTA sound data of the Bay of Bengal covers a region from the Meghna River estuary to about 20.00 S that includes the continental shelf, slope and a small part of the basin. The BIWTA survey area also includes a very deep trench called 'Swatch of No Ground' at the south-western part. But most of the area was within 20 m depth - i.e., within the possible range of optical light penetration. The echo-sound data locations were of course quite sufficient and uniformly distributed to give a very reasonable view of the bathymetry. However, only the north-western part (Figure 2) of the BIWTA survey area was included in the present analysis due to the limitation of image processing speed in the computer. The point data were vectorized, resampled and interpolated systematically. Software Cartalinx was used to digitize points and to resample those. The RMS error was kept low (Hagan, J., User's Guide, Cartalinx, 1998). The 3 dimensional surface was generated by the interpolated data using the software IDRISI windows version 2. The resolution, i.e., row and column length was reduced (original c5148 x r4308, reduced c718 x r854) for processing and analysis. The reduction of pixel resolution did not affect as the study region was quite big (about 32,400 km2) and the bottom relief was found not to be have abrupt irregularity. To have better observational effect the 3D surface of the bottom relief was divided into 8 subsets (Figure 3A and 3B). To get a smoother effect the 3D image/s were filtered in 5X5 pixel window by low pass filter. The effect of the view was found to be dramatic in case of the south-western part that includes the 'Swatch of No Ground'. But however, an overall gradual increase of depth was found in the southern part, i.e., after about 30 km to the south. The satellite image shows a sharp drop of radiance, approximately after about 100 km south of the coast at 900 east longitude which indicates the beginning of the continental slope. It should be noted that penetration of even the upper portion of visible electromagnetic spectrum is limited up to 20.8 m only even in clear water. Simple physical observation also shows that the water was found significantly turbid with suspended sediments near the coast and it decreases southward and become significantly clear after about 100 km towards south.  Figure 01: Work flow chart.  Figure 02: Study area.  Figure 03(a): Study area divided into 8 frames.  Figure 03(b): Three dimensional view of the sea bed morphology. View of the frame no. 6. Digital image processing and signal calibration: The main problem appeared during geo-registration and removal of the back scattered part of the radiance in the water. It was an advantage that the image was atmospherically corrected from its source. As spectral characteristic was the prime concern than all other satellite and sensor features, the visible portion of 0.45 to 0.52 µm would be best for bathymetric measurement, although this spectral portion is, on the other hand, susceptible to significant back scattering (Releigh's). Other optical channels such as green, red, near IR, middle IR, upper IR and thermal (both emissive and reflective) were initially excluded. The main reason was the noticeably gradual increase of signal attenuation in increasing water column depth. That means, in the spectral region of 0.76 to 0.90 (Near IR), the EMR saturation will be close to 100 percent in a depth of a meter (BILKO, UNESCO, 1999). The band combination 1,2,3, shows a visual distribution of the suspended sediment. Practically, this indicates the pattern of back scattering effect. There is of course a straightforward method to identify depth calculating radiance loss in different channels calibrated by the set signal attenuation models (BILKO, UNESCO, 1999) that requires very little local information. But the suspended load prohibits the application of this method in case of the Bay of Bengal coast. The alternative approach chosen here was to depend on the removal of back scattering effect using local bathymetric information (provided by the BIWTA sound chart) that was found sufficient. It is theoretically accepted that drop of digital numbers will have a negative relation with the increase of water depth. That is, if the depth is more, the DN value will be proportionately low at that point. A model was developed by using a statistical calculation of generating a regression model between the sound chart generated 3D surface and the blue channel of satellite image (Eastman, R.J., User's Guide, IDRISI,1992). Principally, the satellite image, which presents DN values that are just the reflectance of the light depends on how far it penetrated into the water (theoretically) and maintains a relation with the depth of water (so, it is a dependent variable). This relation is unknown and should correspond to that of the sound data (the independent variable). The relation could be written as: Y = a + bx; where 'a' is the intercept and 'b' is the slope of the dependent variable Y. The equation itself is a mathematical expression of the line. In this case the equation should result as: Y = 22.292295 + 0.164983 X. Where r = 0.9138 coefficient of det (r2) = 83.50% s.d. of X (Sx) = 13.5644894 s.d. of Y (Sy) = 2.4490030 s.e. of estimate = 0.9948812 s.e. of beta = 0.0003717 t stat for r or beta = 443.8184509 t stat for beta <> 1 = -2246.2719727 n = 38915 apparent df = 38913 In effect this equation says that it can predict depth at any location when the dependent variable (the satellite image in this case) is multiplied by +0.313199 and add 110.819099 to the result. This was the model. Before this, the effect due to suspended load and water column depth has to be removed. Measurement of back scattering and signal calibration: In the limited scope two liters of water samples were collected from every 10 km interval both from surface and from 10' (3.33m approximately) under the surface of the sea (Figure 15-16). The justification of the methodology was that as the arrival of the sediments from the north is closely similar (if not uniform due to different amount of water discharge) all along the coast, there will be an approximately similar decrease of suspended loads as distance increases from the coast southward ( Figure 4). FCC image presents the view of suspended sediment distribution in the upper Bay of Bengal. The collected waters were examined in laboratory to determine the amount of sediments and their average size for each of the locations both for the surface and 10' under water (Figure 17, 18, 19, 20). Both the graphs show a clear decrease in the amount of sediments as well as in the average of the suspended sediments towards deeper sea. The general tendency was the decrease in size of sediments and in lower proportion of suspended loads in amount per unit of water as the depth increases. This rate of decrease of size and amount could be used to detect the extra reflectance due to back scattering in the water. There is no easy alternative method to adjust this data. However, amount of sediment per unit of water was found most convenient for the purpose. Realistically, sediment size has a clear positive relation with the amount of suspended sediments (Figure 5A, 5B).  Figure 04: Distribution pattern of the suspended sediments in the study area. Water collection sample locations are also shown in the image using dots.  Figure 05(a): Pattern of the distribution of the amount of suspended load in sea water.  Figure 05(b): Pattern of the distribution of the suspended load size in sea water. A difference of radiance was identified between the water sample collection at a depth of 3.3 m in the coast and at a known clear water reservoir of similar depth. This difference, theoretically speaks the extra radiance (that may be considered as the amount of back scatter) of the specific water sample collection locations at 3.3 m depth in the coast. This difference of radiance should be equal to that of the amount of suspended loads at that point. Referring this as a standard, the other locations could be checked about their specific amount of back scattering. Subtraction of these differences of DN from the actual digital numbers of the respective locations could give the corrected radiance. This is explained in the Table 02. The attenuation of radiance was corrected using the standard norm, that is the signal at blue range
Source: Researchers' own field investigation. would be saturated completely at 28.8 m depth in clear water. The actual depth of the water sample collection locations were known from the BIWTA sound chart. So, the rate of attenuation at all these locations were easily calculated referring the standard. Finally, the amount of attenuation was subtracted from the back scatter corrected radiance of the locations. The total decay (absorption + scatter) was subtracted from the actual available radiance of the water sample collection locations. Now the image may be considered as corrected and free from the effect of back scatter and signal attenuation. Simulated submarine relief and bathymetric mapping: Theoretically, the corrected satellite image and the image of the BIWTA generated 3D model must show a negative relation, i.e., where depth in the BIWTA image is more, the DN number in the TM image would be lower. This relation has been shown in the regression model where TM image was put on dependent axis and the BIWTA image on the independent (Figure 6). Now the value of the relation line's interception and height could be fitted in to the image to convert the image's DN to simple bathymetric figures in meters. The regression does not show a perfect relation but was found to be reasonably acceptable. This may be due to the presence of very low number of ground verifications. The newly converted DN values in the resultant image was used to generate a 3D model that represents the submarine relief of the upper Bay of Bengal. But to remove noises and sharp speckles in the images it was run in a 5X5 pixel window for mean pass filter. Finally a set of contour lines were prepared from the 3D image for mapping purposes. The relief map shows (Figure 7) some minor but significant rises very close to the coast, particularly to the south of Bhola and of Kuakata. The speedy approaching water surge during any coastal cyclone from the south/south-west might gain an extra height and inundate more than usual expected height, which should be due to the presence of small humps detected in the image. The accuracy may be viewed from the some comparisons. Lines in the Figure 8 were compared using BIWTA image and the simulated bathymetric image. Most of the images were found to give a close similarity expect the last two, which might be due to very deep water (Figures 9, 10, 11, 12, 13, 14), which is beyond the limit of light's water penetration capacity.  Figure 06: Relation between image and the actual depth.  Figure 07: Simulated sea floor relief generated from satellite image of frame 6.  Figure 08: Location of cross section lines used to compare result accuracy.  Figure 09: Cross section A B.  Figure 10: Cross section C D.  Figure 11: Cross section E F.  Figure 12: Cross section G H.  Figure 13: Cross section I J.  Figure 14: Cross section K L. CONCLUSION The calibration of the signals influencing depth measurement was done based on only seven sample locations in a straight line which is of course insufficient, but it was found to be a path indicating. The best would have been if to collect water samples from a uniform locations across the Bay of Bengal at 10 km apart in both the directions - which was not possible in the present study and was left open for any future study. This would perhaps, give a better calibration. Although sedimentation process is not as fast as to change topography every month, but certainly six monthly observations might give a significant change in the mapping of the bathymetry - which might be very useful for navigational purposes and water surge damage prediction. The methodology is meant for only shallow coastal shelf region - and hence might also be useful for prediction and continuous monitoring of the emergence of new lands in the estuarine areas. It is possible to perform the simulation every 16 days, the satellite's revisit period.  Figure 15: Collection of sea water.  Figure 16:The vessel used for water collection.  Figure 17:Microscopic view of the suspended sediment at water collection location 21°44/ N 90°05/ E. Magnified 900x.  Figure 18:Microscopic view of the suspended sediment at water collection location 21°40/ N 90°05/ E. Magnified 900x.  Figure 19:Microscopic view of the suspended sediment at water collection location 21°24/ N 90°04/ E. Magnified 900x.  Figure 20:Microscopic view of the suspended sediment at water collection location 21°20/ N 90°04/ E. Magnified 900x. Acknowledgement: The authors are grateful to the Center of Environment and GIS of Dhaka and Dr. Riaz Khan, its Executive Director for the satellite data received, and to Dr. M. Munsur Rahman of Bangladesh Council for Scientific and Industrial Research, Rajshahi for their laboratory support. REFERENCES
Page 1 of 1
|