A RIGOROUS METHOD TO RETRIEVE TSM CONCENTRATIONS FROM MULTI-TEMPORAL SPOT IMAGES IN HIGHLY - TURBID COASTAL WATERS

Size: px
Start display at page:

Download "A RIGOROUS METHOD TO RETRIEVE TSM CONCENTRATIONS FROM MULTI-TEMPORAL SPOT IMAGES IN HIGHLY - TURBID COASTAL WATERS"

Transcription

1 A RIGOROUS METHOD TO RETRIEVE TSM CONCENTRATIONS FROM MULTI-TEMPORAL SPOT IMAGES IN HIGHLY - TURBID COASTAL WATERS Hans van der Woerd and Reinold Pasterkamp Institute for Environmental Studies (IVM), Vrije Universiteit De Boelelaan 1087, 1081 HV Amsterdam, The Netherlands Phone No Fax No Address: hans.van.der.woerd@ivm.vu.nl Abstract Remote sensing as part of a monitoring system can keep track of temporal and spatial changes in an important ecological parameter, the Total Suspended Matter (TSM) concentration. However, proper comparison of series of multi-temporal satellite maps requires an excellent inter-calibration. Due to a lack of appropriate historical data of atmospheric conditions, this is often hard to achieve. This paper present ways to pin down the atmospheric conditions at the time of satellite overpass, by close inspection of dark pixels and semi-invariant bright pixels in the image. The method is exemplified with a case study of the access channel to the port of Surabaya, Java, Indonesia. Introduction The world's coastal waters often are used intensively for economic activities like fishery, tourism and (heavy) transport, especially near river outflows and estuaries. These activities can impose a heavy burden on the commonly rich and diverse ecosystems. One of the key waterquality parameters is the sediment load of the waters, which mainly determines the underwater light availability. In this paper we present some results of a study to describe the complex patterns of suspended matter in the Strait of Madura, prior to intended dredging activities near the Surabaya harbour access channel. The TSM behaviour in the Strait of Madura is mainly determined by 4 factors: 1. Morphology (depth and type of sediments); The Strait of Madura has a shipping lane which is has a depth between 10m and 20m. The shipping lane is surrounded by shallow areas, which are partially dry during ebb. River sediments that have been deposited in these areas formed these shallow areas. At the northern entrance of the Strait the depth increases from 4m till 22m within a range of 2km only. This ridge can be seen as the northern boundary of the area, which separates the Strait of Madura from Java Sea.

2 2. Tidal forces that erode and transport sediments; It should be noted that most of Java Sea and especially the entrance of Strait of Madura in the East has a mixed tide that is prevailing diurnal. The maximal difference between flood and ebb is about 2m, whereas the minimal difference is roughly 40cm. 3. Wind that leads to bed erosion by wind-induced waves and that affects transport; In winter, the monsoon wind blows from west and is parallel to the northern coast of Java. This wind may bring river sediments (river loads for sediments are high in these period due to rainfall) from the Solo river to Madura Straight. Part of these river sediments will deposit in the area, another part will deposit elsewhere 4. Rivers that put sediments into the Java Sea and Strait of Madura. The area has important sediment import from two rivers: The Bengawan Solo river in the North and the Brantas river in the South. Sediment import is largest during the rainfall period from end of November till the beginning of April. For the monitoring of this highly dynamic system, multi-temporal remote sensing images are an important source of information. The remote sensing images provide a synoptic view of the entire coastal area and information of all seasons, thus allowing the understanding of seasonal variation without the need of measuring the year around. Several authors have shown that it is possible to map the TSM concentration of turbid inland and coastal waters using satellite remote sensing instruments with high spatial resolution such as SPOT or LANDSAT. In this paper we present some results of the study on the conditions of flow and sediment transport in the narrow Madura and Surabaya strait by a combination of field measurements and remote sensing techniques. Methodology and approach For the Surabaya study 7 SPOT images were selected, covering a range of relevant seasonal and tidal situations (see Table 1). Based on the experience with the Indonesian coastal water project Banjar masin (Dekker et al., 1999) and the RESTWAQ 2 projects (Vos et al., 1998) it was decided to follow the analytical method to retrieve TSM concentrations from the selected SPOT images. In the analytical approach the inherent optical properties and apparent optical properties are used to model the reflectance and apparent optical properties and vice versa. The water constituents are expressed in their specific (per unit measure) absorption and backscatter coefficients. Subsequently an algorithm for estimating suspended sediment from satellite images can be developed. (See Dekker et. al, 2001). In the analytical approach physical relations are derived between water quality parameters, the underwater light field and the remotely sensed measurements. The inherent optical properties are physically related to the subsurface irradiance reflectance R(0-) which is the key parameter linking the properties to the remotely sensed irradiance data.

3 Table 1 Date, geometry and atmospheric parameters for all remote-sensing images. ID date time (UTC +7) θ0 f aerosol model atmospheric composition horizontal visibility 1 June 22, : maritime tropical 45 km 2 July 12, : maritime tropical 30 km 3 April 19, : maritime tropical 30 km 4 May 20, : rural tropical 18 km 5 December 30, : maritime mid.lat. winter 12 km 6 April 1, : maritime tropical 17 km 7 September 30, : maritime tropical, cirrus 8 km Several workers have investigated the relation between R(0-) and the inherent optical properties for ocean, coastal and inland water systems (Gordon et al., 1975; Kirk, 1991; 1994, Dekker et al., 1994). Dekker et al., (1994) found that Equation 1 is the most appropriate optical model for turbid waters (all parameters are spectral: this dependency was omitted from the equation): bb R(0 ) = f a + b Where R(0-) A b b f b is the spectral subsurface irradiance reflectance (dimensionless) is the total absorption is the total backscattering is a coefficient depending on solar elevation. Equation 1 Parameters derived from the field samples Water samples were taken on February 26 th 2000 at eight locations along the Surabaya Access channel and near the port of Surabaya in December The samples were frozen and transported to The Netherlands where they were analysed at the Institute of Environmental Studies. The sampling locations are indicated in figure 1. Total suspended matter concentrations (TSM) ranged from 6 to 61 g m -3. No correlation was found between the TSM concentrations and the distance to the ocean. This can be caused by the turbulent flow patterns, especially with strong winds and currents and patchiness in the TSM concentrations.

4 Figure 1. The 8 locations sampled at February 26th, 2000 indicated on a false-colour composite of the image of 04/01/99. The concentration of organic material was low, with an average TCHL of 4.1 mg m -3. The absorption by algal pigments turned out to have relative small effect on the spectral reflectance. The concentrations of aquatic humus, expressed as the absorption at 440 nm, ranged between 0.1 and 0.5 m -1 (average of 0.25) and an average exponential function with slope 0.01 (fitted from 350 to 550 nm). Referring to the in-situ measurements performed in the study for the deepening of the access channel to the port of Banjar Massin (Dekker et al., 1999), it was found that there is significant difference for ocean and non-ocean waters. Within this project a more extensive field campaign was launched where ocean and non-ocean waters (coastal waters, coastal

5 waters and rivers) were analysed on the specific scattering coefficient. For a wavelength of 650 nm, an average specific scattering coefficient of ~0.08 (m 2 g -1 ) was found to be typical for ocean waters, and a significantly higher specific absorption efficiency, ~0.23 (m 2 g -1 ) was found to be typical for non-ocean waters (including rivers, tidal flats, and coastal waters). Comparing those values to the values measured within the access channel to Surabaya, we see that again some samples (2, 3 and 6) resemble typical ocean waters, and the others are more representative for non-ocean waters. For this study we used the average set of Specific Inherent Optical properties (SIOP) composed of the samples 1, 4, 5, 7 and 8 to model the (more turbid) coastal waters. Samples 2, 3 and 6 were used to model the clear ocean waters. Using the optical model for turbid waters, parameterised with the specific inherent optical properties measured in the lab for measurement locations typical for the more turbid waters, the relationship between TSM concentration and reflectance was modelled and shown in figure 2. The green band has a reasonable relationship of an increasing reflectance with increasing TSM concentration. However, the dynamic range (the range of reflectance covered by the highest and lowest concentrations) is not as high as in the red band. This band is also still sensitive to the absorption by dissolved organic material and phytoplankton pigments. In the red band the relationship of reflectance with increasing suspended matter concentrations is the highest. Influence of absorption by dissolved organic matter and the absorption by phytoplankton pigments is also much less than in the red. The choice between this band and the nearby infrared band (not plotted here) would be difficult, however, the sensitivity of the satellite detectors is less in the nearby infrared for low reflecting targets such as water, thus the red band is most suitable. Figure 2. Relation ship between the total suspended matter concentration and the subsurface reflectance. The reflectance saturates for high TSM concentrations.

6 Based on this analysis a choice for the red band (SPOT Band 2) is most appropriate for use in an algorithm for estimating suspended matter in these waters. The relationship between TSM concentration and reflectance in band 2 ( R 2 ) can exactly be described by the function given in equation 2 (Pasterkamp and Van der Woerd, 2002) TSM = ef er 2 er 4 2 ef 2 The coefficients e K e 1 4, derived from the convolution of the spectral properties of the water constituents and the SPOT spectral response curves, are given in Table 2. Table 2 Coefficients describing the relationship between R(0-) in band 2 and TSM concentration. SPOT Sensor HRV 1 HRV 2 HRV 1 HRV 2 e e e e The modelled data indicate that the sensitivity of the applied algorithm becomes less at increasing concentrations. Especially beyond about 200 (g m -3 ) the associated increase in reflectance becomes minimal. This so-called saturation implies that concentrations above this level cannot be distinguished as the reflectance than becomes almost constant. The Dark and Bright Pixel correction Before calculation of TSM concentration can take place, the signal has to be corrected for atmospheric effects. When looking to a waterbody from a satellite, the signal is very much influenced by the intermediate atmosphere. Important processes are 1. the diffuse transmittance from the incoming solar irradiance to the earth surface, and the transmittance from the water-leaving radiance back trough the atmosphere to the sensor, 2. the contribution of photons scattered once or more in the atmosphere and reaching the sensor without having interacted with the water. The radiance originating from the latter contribution, the so-called path radiance can be up to 90% of the total radiance received at the sensor, depending (amongst others) on the solar elevation, reflectance of the water, and aerosol loading of the atmosphere. Note that the wavelength dependence of these processes has to be taken into account. Because each remote sensing image was taken under different atmospheric conditions and under different viewing and illumination geometries, at-sensor radiances cannot be compared

7 between images. Thus to have a starting point for the multi-temporal analysis of RS images, atmospheric correction is a necessary step to take. Moreover, the analytical inversion models for retrieval of TSM concentrations need sub-surface reflectance as input. There are several approaches to perform the atmospheric correction for SPOT or LANDSAT images. The most straightforward approach is to do (multiple) reflectance measurements in the field, simultaneously with the satellite overpass. By comparing the reflectance s with the at-sensor radiance, values for transmission and path-radiance can easily be calculated. Another approach is to use measurements aerosol type and loading in the atmosphere at the time of satellite overpass and then calculate the atmospheric correction parameters from an radiative transfer model. The main drawback of both methods is that a dedicated field cruise is necessary simultaneous with the satellite overpass. Apart from the logistics, scientific expertise and funding this requires, this is simply not possible when dealing with 'historical' RS images. Several studies (Dekker et al., 1999; Pasterkamp et al., 2000) have shown that it is still achievable to get a handle on the atmospheric correction without the need for simultaneous measurement. This main starting point of this approach is to get as much as possible information out of the image itself by predicting surface reflectance s in the images for specific targets. The targets can be on land or parts of the water under investigation. Targets on land While in practise it is impossible to predict absolute reflectance values of land targets, those targets can still be used to compare several images where the target is visible. Care must be taken however that the reflectance of the target doesn't change over the time span between the images. Land targets were not used in this study. Water targets Targets on water can provide valuable information on absolute reflectance values. Especially when studying estuaries where the difference in sediment loadings can be very large. Optical models can be used to predict the reflectance s of very dark (ocean) pixels and very bright (turbid) pixels, provided that these models are calibrated with the regional specific inherent optical properties, measured during a (non-simultaneous) dedicated field cruise. The SIOP must be known for the dark (ocean-type) pixel and for the bright (turbid coastal/estuarine) water type. The reflectance of the dark pixel is calculated by the forward optical model with typical ocean water quality parameters. Note that the dark pixel is thus NOT assumed to be totally black. Certainly the variability in the ocean's water quality will put an error bar on the calculated reflectance. The darkest ocean pixels were estimated to have a spectrum coincident with a concentration of suspended sediment of approximately 3 6 (g m -3 ), corresponding to a reflectance in band 2 of 0.01 to 0.02 This range is acceptable compared to the wide range of reflectance s expected in the complex coastal water (1% up to 25%). The calculation of the reflectance of the bright pixels is more complicated, because the exact amount of sediment loading is uncertain, which makes direct use of the forward model unfeasible. Fortuitously the phenomenon of saturation can be used here to get a readily accurate assessment of the reflectance. The saturation was shown in Figure 2 as a flattening of the reflectance for very high sediment concentrations, approximately larger than 100 g m -3. The saturation can also be observed in

8 the image when analysing transects from low to high sediment concentrations. The transect, shown in Figure 3, is approximately 6.7 km long and starts approximately 2.5 km upstream. Sediment concentrations in the river are expected to be very high (>200 g m -3 ) because of the transport of sediments from the river, and resuspension of the sediments from the delta plain. Referring to Figure 4, reflectance s in band 2 are thus expected to be between 0.22 and 0.24 near the river mouth. From the radiance development along the transect, it can be concluded that for this image the saturation occurs at approximately 105 Wm -2 sr -1 µm -1. Using MODTRAN 3 software the values for the aerosol type, the atmosphere composition and for horizontal sight were adjusted by trial and error until a suitable atmospheric composition was found that explained the at-sensor radiances for the observed bright and dark pixels given the predicted target reflectance. Having fixed the atmospheric composition, the whole image was processed accordingly to subsurface irradiance reflectance. The inverse model was than applied to yield total suspended matter concentrations. transect Bengawan Solo Figure 3. Sampled transect (orange dashes) along the outflow of the Bengawan Solo river

9 Figure 4. Radiance development and identified saturation level in the transect along the outflow of the Bengawan Solo river. Results The above presented dark and bright pixel correction was applied to all images to derive the atmospheric correction. The images were processed to thematic maps of TSM, put in a standard ArcInfo frame (See Figure 5). This image represents an example of a low amplitude tidal condition. The effect of the rivers is pronounced due to the fact that river sediments still are transported into open sea as a result of (the now ceasing) rainfall period As the river discharge is high, the outflow at the Solo river mouth in the North West is hardly hampered by the high sea levels. This permits a wide plume of sediment rich water extending seawards for several kilometres. The strong tidal currents in the Surabaya strait introduce the entrance of relatively clear (low TSM values) marine waters from the East. In this images wind effects are probably small. There was no clear correlation found between the images and the tide. On the other hand, some typical phenomena were observed that were repeated for the various seasons. These phenomena were found back for both years (1997, 1999) and were further confirmed by additional quick looks from the SPOT Dali catalogue at Toulouse for 1998.

10 Figure 5. TSM map for 1 st of April Conclusions The historical databases of the LANDSAT and SPOT sensors enable long-term analysis of tidal current and water turbidity. With the methodology described in this paper it is possible to analyse a series of observations in a strict and physical correct way. Of course the atmospheric correction will have a limited accuracy. The restricted number of aerosol models in the radiative transfer code software, and the unavailability of ground truth data for the images mainly cause this limited accuracy. The overall accuracy of atmospheric corrected images is estimated to be within 1% (absolute value) subsurface irradiance reflectance (Pasterkamp and Van der Woerd, 2002). Using the approximated exponential relationship between total suspended matter and the reflectance, this corresponds to maximal 20% (relative value) error in the retrieved total suspended matter below 100 g m -3. Baring in mind that local campaigns are difficult to organise, expensive and have limited information content and coverage of this complex estuary system, the above methodology is relatively cheap and reliable.

11 Acknowledgement This research was carried out as part of the Surabaya Access Channel study under the direction of Ballast Nedam, The Netherlands. We like to thank drs. H. Berghuis for collecting the water samples in the Strait of Madura. We also like to thank dr. A. Dekker for discussions on IOP s and dr. R. Vos for his views on the sediment regime and influence of Monsoon winds. References Dekker, A.G., Peters, S.W.M., Vos, R. and Rijkeboer, M Remote Sensing fro inland water quality detection and monitoring: State-of-the-art application in Friesland waters. In A. van Dijk and M.G. Bos (eds), GIS and remote Sensing Techniques in Land- and Watermanagement, Kluwer Academic Publishers, The Netherlands, pp Dekker, A.G., Peters, S.W.M., Rijkeboer M. and Berghuis, H Analytical processing of multitemporal SPOT and Landsat images for estuarine management in Kalimantan Indonesia. In: G.J.A.Nieuwenhuis, R.Vaughan, and M.Molenaar, editors. 18th EARSeL Symposium on operational remote sensing for sustainable development, A.A.Balkema, Rotterdam,The Netherlands, p Dekker, A.G., H.J. Hoogenboom, L.M. Goddijn, and T.J.M. Malthus. (1994) The relationship between spectral reflectance, absorption and backscattering for four inland water types. 6th Intern. Symp. on Physical Measurements and Signatures in Remote Sensing, , CNES, France, Val d'isere, p Gordon, H.R., Brown, O.B. and Jacobs, M.M Computed relationships between the inherent and apparent optical properties of a flat homogeneous ocean. Appl. Opt., Vol. 14, No. 2, p Kirk, J.T.O Volume scattering function, average cosines, and the underwater light field. Limnol. Oceanogr., Vol. 36, No. 3, p Kirk, J.T.O Light & photosynthesis in aquatic ecosystems, 2 ed., 509 p. Pasterkamp, R. and Van der Woerd, H.J A general approach to map turbid estuaries with remote sensing, Remote Sensing of the Environment, submitted. Pasterkamp, R., Peters, S.W.M. and Dekker, A.G Environmental Baseline Mapping of Total Suspended Matter Concentrations in the Western Scheldt Estuary Supported by the use of SPOT Images. 6th International Conference on Remote Sensing for Marine and Coastal Environments, Charleston, SC, USA, p Vos, R.J., Villars, M., Roozekrans, H., Peters, S.W.M.and Raaphorst, W. van RESTWAQ2, Part 1. Integrated monitoring of total suspended matter in the Dutch Coastal Zone. Delft: Netherlands Remote Sensing Board, 44 pp.