IMPROVING INTER-TIDAL DIGITAL ELEVATION MODELS CONSTRUCTED BY THE WATERLINE TECHNIQUE

ABSTRACT

Improvements to digital elevation models (DEMs) of the inter-tidal zone constructed by the waterline method are discussed. Methods of improving shoreline heighting accuracy by refining the hydrodynamic model include increasing model resolution, assimilating local tide gauge data into the model runs and refining the model bathymetry using an initial DEM. Techniques for improving shoreline delineation on steeper beaches are also considered.

1. INTRODUCTION

This paper discusses improvements to digital elevation models of the inter-tidal zone constructed by the waterline method using remote sensing and hydrodynamic modelling. This involves finding the geocoded positions of the shoreline in a SAR image using image processing techniques, then superimposing on these positions the heights relative to mean sea level predicted using a hydrodynamic tide-surge model. From multiple images obtained over a range of tidal conditions, a set of heighted shorelines can be assembled within the inter-tidal zone, and from this a gridded DEM may be interpolated. The method has been described in (Koopmans and Wang 1995, Mason et al. 1995, 1996, 1997a). It is useful for developing improved tide-surge models, and changes in the DEM over time allow measurement of sediment mass transfers in the inter-tidal zone.

The technique has recently been used to construct an inter-tidal DEM of the Wash, eastern England, from 13 ERS-1 SAR scenes acquired mainly during the winter months of 1992-4 (Mason et al. 1997b). As the height errors produced by the method reduce with beach slope, it is particularly useful in tidal flat regions such as the Wash, with its vast inter-tidal area of 29,770 hectares. A raster DEM was interpolated from the heighted shorelines using universal spatiotemporal block kriging. Comparison of the DEM with beach transect data indicated an average height standard deviation of 22cm over its 60m x 60m blocks. Substantial changes were apparent between this DEM and Ordnance Survey maps (last updated 1977) and 1970 Hydrographic Office charts of the area. An earlier, more intensive study using 33 SAR images over a subset of the complete Wash area (the western Wash) resulted in an average height accuracy of 14cm (Mason et al. 1997a).

The technique has both scientific and non-scientific applications. Potential users include those concerned with coastal defence, coastal environmental management, economic exploitation of the inter-tidal zone and flood forecasting. A survey of potential users and their requirements has been carried out, including how various techniques for constructing inter-tidal DEMs meet these requirements (Davenport et al. 1996). One important requirement identified by many users was the need to achieve a vertical height accuracy of 10cm. If this could be achieved, the low cost of the waterline technique compared with other survey methods (e.g. ground survey, airborne LIDAR) should make it an attractive survey option, particularly for wider, flatter beaches.

In an attempt to achieve this vertical accuracy, two approaches are being studied. On gently sloping beaches such as the Wash, the dominant error source is height inaccuracy in the water elevations produced by the hydrodynamic model. Progress on improving shoreline heighting accuracy by refining the hydrodynamic model is discussed in section 2. On steeper beaches shoreline position error becomes the dominant source of heighting error. Progress on improving shoreline delineation on such beaches is discussed in section 3.

2. IMPROVED SHORELINE HEIGHTING

The hydrodynamic model used to date (the Holderness model) is based on depth-averaged hydrodynamics, and includes the effects of tides and meteorological forcing (Flather 1994). A finite difference method is used to solve the equations on a 1.2km grid covering the east coast of England between Whitby and Cromer. Quadratic laws are assumed for the bottom and surface forcing whilst on open sea boundaries a radiation condition is employed. The model formulation is similar to that employed in storm surge forecast models used operationally in the U.K. to provide coastal flood warnings, and these models provide the open boundary inputs of tide and surge for the Holderness model.

A number of methods for improving shoreline heighting have been studied, including increasing the resolution of the model and assimilating local tide gauge data into the model run (Hill et al. 1997).

One important error source is limited grid resolution which is insufficient to resolve complex channels and inter-tidal areas. In the Holderness model, the grid size of 1.2km square means that errors may be introduced by heighting the shoreline using grid cells which may be several km away from the shoreline. A high resolution model based on a 240m grid has been developed for the Wash and embedded within the Holderness model (Figure 1). The inputs for tidal and surge information along the open boundaries of the Wash model are derived from the Holderness model. This has resulted in significant improvements in the simulation of the drying process in the Wash region.

Figure 1. The Holderness and Wash model grid regions.

A second improvement has been the development of a simple non-optimal scheme to assimilate into the model run local tide gauge data from Whitby on the coast at the northern open boundary of the Holderness model (Figure 1). This has improved the estimation of the surge component. The SAR images used in the method are acquired during conditions of light to moderate wind speed when the surge is not large. However, surges of 10cm to 40cm may still occur, which are significant when the goal is a heighting accuracy of 10cm. The assimilation method involves correcting the Holderness model elevations and currents by functions which depend on the instantaneous difference between the Whitby tide gauge elevation and the corresponding model elevation. Details are given in (Hill et al. 1997). A case study was carried out on the period 4-7 November 1993 and comparisons between the model elevations and tide gauge data were made in order to assess the improvements gained by assimilation, which proved to be significant. Figure 2 shows a comparison of the uncorrected and assimilated model data with the tide gauge data at Cromer, which is to the east of the Wash and at the south-eastern edge of the model grid. A height accuracy of 10cm has been achieved at Cromer for the corrected model elevations in this instance. Since the open boundary inputs to the Wash model are derived from the Holderness model, improvements gained in the assimilation will also translate into better results for the Wash model.

The shoreline heights for the initial Wash DEM were calculated using tidal information from the Holderness

Figure 2. Comparison of sea elevations from the Holderness model with tide gauge data at Cromer using uncorrected and corrected open boundary inputs (after Hill et al. 1997).

model in conjunction with local tide gauge data in a

kriging method (Davenport et al. 1996). A third improvement we intend to implement shortly involves refining the Wash model bathymetry in the inter-tidal zone using the initial Wash DEM, then re-running the model for each of the 13 SAR overpasses to produce improved shoreline heightings and hence a more accurate DEM.

In the future we also intend to develop optimal techniques for assimilating observations into model runs, and to compare these with the simple non-optimal assimilation scheme employed above. An advantage of the optimal methods is their ability to assimilate data from a number of stations within the model region, not simply data adjacent to the model boundary. We intend to use variational methods, which are less computationally expensive than Kalman filters and lend themselves more easily to non-linear models, an important consideration when modelling shallow-water areas.

3. SHORELINE DELINEATION ON

STEEPER BEACHES

Most beaches are steeper than the 1:500 slopes of the Wash, and the method becomes more powerful if it can be extended to these without substantial loss of heighting accuracy. Errors in shoreline position determination arise due to the resolution of the SAR image and in the process of image registration, though both these errors appear to be no more than a single pixel. A further difficulty on steeper beaches is that the radar response of the beach becomes comparable to that from the sea because grain sizes are generally larger and the material drier, thus increasing surface roughness. Indeed the human eye may be unable to discern the sea edge on these beaches.

A study was made of a number of subimages of the beaches around Skegness (Figure 1), where beach slopes are around 1:50. Often there appears to be a thin dark line next to the shoreline which the eye can use to demarcate this (see e.g. Figure 3). But the situation may be confused by narrow black bars running at a narrow angle to the shoreline. Field surveys showed that these black bars were at too high an elevation to be sea; they appear to be caused by flatter areas of fine wet mud and standing water in the hollows between the ridges in the ridge/runnel structure of the beach, which act as specular reflectors of the radar signal away from the satellite.

A further point is that the shoreline appears to be more discernable as a thin dark line in images acquired on ebb and low tides than on flood or high tides. The shorelines in 8 out of 11 flood/high tide images were difficult or impossible to distinguish, whilst in 7 ebb/low tide images the shorelines were largely or entirely distinguishable. The effect is probably due to the sand immediately next to the sea edge being wetter on an ebb than on a flood tide, and thus acting as a specular reflector. This finding has implications for an operational system, in that a fairly high resolution browse facility (not currently supported by ESRIN) is required to decide if, for any particular image, a shoreline is visible and that the image is therefore worth purchasing.

Improved image processing to delineate the shorelines on steeper beaches has also been developed. We have used a multiscale contrast ratio edge detector (Touzi et al. 1988) which combines a window to detect long thin straight edges with the square window previously employed (Mason et al. 1996), and selects the maximum response from either window (after scaling to take account of the fact that windows of different size may have different noise edge false alarm rates). Steeper beaches may often be narrow (100-200m) so that a contrast ratio half-window width of 130m (that used for the square window) may be too large to expect the window simply to contain beach on its landward side when the window is centred on the shoreline, especially at high tide. Most steeper beaches also exhibit shorelines which are fairly straight over distances of up to 0.5 kilometre. The long thin window is applied at angular increments of 5 degrees rather than the 45 degree increments used for the square window.

Figure 3. ERS-1 SAR image of Skegness area at low tide, 2 November 1994, (c.ESA).

For each window type, the maximum response at any angle is chosen. Experiments showed that the shoreline delineator generally gave improved results using the double-window approach, especially in high tide images. However, in order to achieve the maximum edge response using the long thin window, the shoreline must be straight to within 1 pixel over a distance of almost 0.5km. Shorelines are unlikely to be this straight in reality. Future work may be aimed at detecting long fairly straight edges which are allowed to curve slightly, using a simulated annealing approach applied only in the neighbourhood of the shoreline.

Consideration was also given to the possibility of improving the registration process. This would not only reduce the registration error, but also make the method easier to implement, as it involves a substantial amount of data handling. Registration is tedious partly due to the difficulty of finding the same control point in two SAR images because of the presence of speckle. As suggested by Koopmans and Wang (1995), registration error is probably best reduced by deploying a number of corner reflectors in each SAR scene.

4. CONCLUSION

The project aim of achieving 10cm accuracy in shoreline heighting has been shown to be feasible on gently sloping beaches, based on evidence from results at one data station. It has also been shown that judicious scene selection could allow the method to be applied to more steeply sloping beaches, subject to refinement of the shoreline delineation. Further improvements should be gained by the incorporation of the DEM into the model bathymetry, and the use of an optimal assimilation technique to employ data from several locations.

5. ACKNOWLEDGEMENTS

This work was funded under the NERC Land-Ocean Interaction Study (LOIS), the British National Space Centre Earth Observation LINK programme (funding a collaborative project involving ESSC, POL and the National Remote Sensing Centre Ltd.) and NERC contract F60/G6/12. We are grateful to ESA for supplying SAR data under ERS project AO2.UK115. Thanks are also due to the U.K. Meteorological Office for Local Area Model data, and the Environment Agency for beach transect data. This is LOIS publication number 220.

6. REFERENCES

Davenport, I.J., Mason, D.C., Flather, R.A. & Gurney, C., 1996, Foreshore study through shoreline delineation. Proc. EUROPTO Conf. on Microwave Sensing and Synthetic Aperture Radar, Taormina, Italy, 23-26 Sept. 1996, SPIE Vol. 2958, 164-173.

Koopmans B.N. & Y. Wang, 1995, Measurement of land-sea transition from ERS-1 SAR at different phases of tidal water. Netherlands Remote Sensing Board Report 95-20, pp64.

Flather, R.A., 1994, A storm surge prediction model for the northern Bay of Bengal with application to the cyclone disaster in April 1991. J. Physical Oceanography 24(1), 172-190.

Hill, D.C., Flather, R.A., Mason, D.C. & Davenport, I., 1997, Improved tidal modelling for digital elevation models. Proc. COASTAL'97 Conf., La Coruna, Spain, June 1997 (to appear).

Mason, D.C., Davenport, I., Flather, R.A, McCartney, B. & Robinson, G.R., 1995, Construction of an inter-tidal digital elevation model by the 'water-line' method. Geophysical Research Letters, vol 22, no 23, 3187-3190.

Mason, D.C. & Davenport, I., 1996, Accurate and efficient determination of the land-sea boundary in ERS-1 SAR images. IEEE Trans. Geoscience and Remote Sensing, vol 34, no 5, 1243-1253.

Mason, D.C., Davenport, I. & Flather, R.A., 1997a, Interpolation of an inter-tidal digital elevation model from heighted shorelines: a case study in the western Wash. Estuarine, Coastal and Shelf Science (in press).

Mason, D.C., Davenport, I., Flather, R.A. & Gurney, C., 1997b, A digital elevation model of the inter-tidal areas of the Wash produced by the waterline method. Int. J. Remote Sensing (in press).

Touzi, R., Lopes, A. & Bosquet, P., 1988, A statistical and geometrical edge detector for SAR images. IEEE Trans. Geoscience and Remote Sensing, 26(6), 764-773.

Keywords: ESA European Space Agency - Agence spatiale europeenne, observation de la terre, earth observation, satellite remote sensing, teledetection, geophysique, altimetrie, radar, chimique atmospherique, geophysics, altimetry, radar, atmospheric chemistry