Science Service System

Summary of Proposal COA1840

TitleInter-tidalmapping in the Tagus estuary, Portugal, based on X-band TerraSARimages(InTiMap)
Investigator Nico, Giovanni - Consiglio Nazionale delle Ricerche (CNR), Istituto per le Applicazioni del Calcolo (IAC)
Team Member
Catalão, João - Universidade de Lisboa, Instituto Dom Luiz (IDL), Faculdade de Ciências

Theobjective of this proposal is to use the high resolution X-bandTerraSAR images to study the inter-tidal flats in the along themargins of the Tagus river. The Tagus estuary, one of the largestones in Europe, has a socio-economic relevance and morphological andsedimentary characteristics that make it an excellent testbed for thestudy of inter-tidal flats. These are the interface between uplandand the deeper portions of the estuary. The inter-tidal flats will emapped by using the detecting the temporal series of "waterline".The waterline detection is, in general, a problem of water and landregions separation. A morphological image processing hierarchicalmethodology in order to detect those regions and, consequently, todetect the correspondent shoreline. In this project we will use amorphological-based segmentation approach for coastline detectiondriven by a waterfall hierarchical algorithm. Themotivation for this project is to contribute to the understanding ofthe morphodynamic response of different estuarine margins typologies.The temporal changes of inter-tidal flats will be mapped by using thehigh-resolution X-band TeraSAR images. Tidal flats will be mappedusing the waterline method. The waterline, defined as the boundarybetween a water body and an exposed tidal flat, is identified basedon the different scattering properties of water and soil surfaces.The waterline method is based on three assumptions: a) that thewaterline represents an equal elevation at the moment of imageacquisition; (b) that the topographic change is negligible during theperiod of data acquisition; and (c) that the absolute elevation ofeach waterline is known. The first assumption allow to generate anintertidal DEM by stacking a series of waterlines observed underdifferent tidal conditions. In this study the waterline elevation isgiven by tide measurements available each hour. The most importantreasons to use SAR images are their well known advantages of beingfree of sunlight and weather conditions, so imaging the earth surfaceat any time independently on weather conditions and cloud cover.Furthermore, TerraSAR-X images have a high spatial resolution (3m instrip-map acquisitions) and a good temporal revisiting cycle (11days) so facilitating the monitoring of rapid changes in the tidalflats. The high-resolution mapping of inter-tidal flats will be andinvaluable tool to study the dynamic morphologic changes that arisefrom high tidal energy and sediment transportation. The study ofsediment budget processes is important in many ecological systems asexplained above. The sediment budget can be estimated only ifhigh-resolution maps of morphological changes are available. Remotesensing, combined with in situ surveying, is an effective tool formonitoring tidal flats. Thisprojectwill lead to the following results: 1)Algorithms and software for the mapping of temporal changes ininter-tidal flats based on the waterline method; 2.)Integration of the software developed in this project, usinghigh-resolution X-band SAR images into the practices for theprotection of inter-tidal flats and coastline environments. Theresults of this project are expected to constitute a valuable toolfor integrated management of land flats and estuaries, in futurepredictions and in the definition of mitigation or adaptationstrategies.

Detailed reportWe resume here the paper Catalao& Nico, 2016. The waterline extraction combined with the tide height is the most used method to generate the intertidal elevation model [1], [2]. The waterline is the instantaneous boundary between the water body and the exposed land at the time of the image acquisition. The height of the waterline is predicted by a hydrodynamic tide model. The intertidal bottom topography model is generated by stacking several waterlines extracted in different tide conditions. The method can be used both with SAR and optical spaceborne systems [3], [4]. The waterline extraction may be considered a threshold problem or a particular image classification problem with two classes [1]. To address this problem several algorithms have been proposed based on adaptive edge threshold [5], interferometric coherence as a measure of the stability of the scatters between two SAR acquisitions [6], [7], wavelet edge detection for decomposition and thresholding [8], using region-based levels sets [9] or k-means clustering [10]. However, the water/land contrast is dependent on several internal factors as the incidence angle and polarization and external ones such as the wind, waves and tides that have influence on backscatter intensity and consequently on the threshold used in these algorithms. Therefore, the accuracy of the extracted waterline depends on the weather and sea waves conditions at the image acquisition time, which may not be comparable along the time frame. This makes it very unlikely that a continuous waterline for each image could be extracted and intensive manual edition is needed. An approach based on the temporal analysis of the scattering properties of the land surface can help to overcome some of the limits of existing methods. In fact, numerous studies have demonstrated the dependence of radar backscatter on surface roughness and on the dielectric proprieties of the surface. In particular, microwave scattering models have been used to characterize the SAR backscattering coefficient, as a function of the local characteristics of the soil moisture [11]. In this project, a new approach based on the temporal analysis of the backscattering coefficient and its correlation with the tide height was developed. The intertidal zone is characterized by a high variability of the backscattering coefficient due to the recurrent and periodic in and out flux of water as the result of ocean tide and river current. The temporal behavior of the backscatter coefficient is modeled by a logistic function with lower and upper limits defined by the water and rough soil backscatter intensity, respectively. The elevation of the resolution cell is associated with the decay of the logistic curve that best fits in least squares sense the ensemble backscatter data. The output is a 3D cloud of points (pixels) that can be used to generate the digital surface model of the intertidal zone. Directly using the temporal behavior of a pixel reduces the need for multiple SAR images processing for waterline extraction and further integration and edition. Furthermore, it reduces the impact of false positives and the problem of choosing the appropriate threshold when applying change-detection algorithms to identify water bodies within images. The intensity variations in a SAR image are related to the surface roughness and dielectric proprieties as well as to the polarimetric proprieties and the local incidence angle of SAR acquisition [11]. Estuarine tidal flats are characterized by low-relief and low-slope, with sand ripples and mud patches and are generally not-vegetated. Twice a day, the tidal-flats are flooded at different levels according to the local tide regime. The exposed terrain surface behaves as a diffuse reflector due the roughness of the sand ripples and mud patches increasing the signal strength and thus the brightness. On the other hand, the water puddles and the remaining water on the surface behave as specular reflectors. The result is a distributed scattering registered on the SAR image as a fuzzy set of pixels on the transition between water and land. Instead of walking across this fuzzy area searching for the land/water separation, as in the waterline method, we propose a pixel-based approach to analyze the temporal variation of the intensity and to estimate the elevation of the pixel and generate the digital surface model of the tidal flat. The temporal window should be relatively short such that the roughness and vegetation do not change. The algorithm is divided into the following three steps: 1) Pre-processing and temporal filtering; 2) Temporal variability assessment; 3) Logistic analysis. The proposed methodology was applied to time-series of TERRASAR-X (X-band) images acquired over Tagus estuary (Lisbon region, Portugal). The Tagus estuary is characterized by an extensive surface of estuarine waters, with tidal flats and saltmarshes on the south margin and manmade structures on the north (Lisbon metropolitan area). The estuary has a semi-diurnal tide regime with maximum amplitude of 3.5 meters. About 40% of the estuarine area is intertidal. A set of 17 HH X-band SAR images acquired by TerraSAR-X (TSX) mission were used in this work. The TSX SAR images were acquired between February and December 2013 with an incident angle of 42.8 degrees and a ground resolution of 3m. The water level was extracted from the ocean tidal model at the Instituto Hidrográfico (IH). The water level is further adjusted to the real water levels measured at the nearby tide gauges and corrected for the sea level rise of 15 cm measured in Cascais since the time of the vertical datum definition (vertical datum Cascais). For simplicity, we have assumed a constant tidal height for all covered area. This is not rigorously true since we must take into account the tidal current and in particular the mixing effect of the river flux into the ocean. The acquisition time of TSX SAR images is at 18:32. It is observed that most of the images are acquired at high tide and only four images are acquired at the lowest tides. This fact reduces the vertical resolution of the derived digital terrain model. The proposed methodology requires SAR images with: 1) a spatial resolution appropriate to map the terrain morphology details of the study area; 2) acquisition times effectively sampling the temporal variation of tight height and 3) precise orbit information to correctly geolocate SAR images at a pixel level. Concerning point 1) X-band SAR images provide the higher spatial resolution even if C-band SAR images can still be useful. The lower and highest elevations of the resulting bathymetric model, 1.2m and 3.7m, respectively, are constrained by the minimum and maximum tide elevations registered in the studied period. It is worth noting that at the Tagus estuary the lowest tide is 0.30m in springtime (March and April). The available TSX dataset does not cover that period and, as a consequence, does not permit to measure the tidal flat minimum elevation at the Tagus estuary. This is not a limitation of the proposed approach but rather of the TSX acquisition time. In fact, the lowest tide occurs around at 10:00 and TSX passage is at 18:30 (ascending) and 6:42 (descending), all local times. For validation proposes a digital elevation model (DEM) and a GPS survey, were used. The DEM covers the area of Seixal Bay and was supplied by Camara Municipal do Seixal. The Seixal Bay DEM is the result of a bathymetric survey with an echo-sounding device performed by the Portuguese Hydrographical Institute in June 2014. The estimated standard deviation of the measurements is 0.18 m. The ship survey cannot completely cover the intertidal range since, even on the high tide, the higher areas cannot be accessed. On the other hand, the ship survey has covered areas that cannot be mapped by SAR derived approaches due to the water depth. The main limitation of this comparison was the lower tide. In fact, the validation was limited to intertidal elevations between 1.2 and 2.7 m. Within this range, a mean difference of -0.20 m has been found with a standard deviation of 0.23 m. The largest differences have been found near the borders of the navigation channel where strong gradients are observed. It seems that there is a geolocation problem between the bathymetric survey and the SAR derived DEM, due to the precision of TSX geolocation. Concerning TSX SAR images we verified that: 1) The spatial resolution is appropriate to map the terrain morphology details. The high spatial resolution of TSX SAR images allows to recover the details of the channel network. 2) Unfortunately, the acquisition times do not sample the temporal variation of tide height. The time acquisition of TSX is 18:30 and the lowest tides are in the morning between 8 and 10 am. 3) The supplied orbit information (Scientific orbits), although enough accurate for most of SAR applications, are not precise enough to correctly geolocate SAR images at a pixel level. References [1] J. Lee, I. Jurkevich, “Coastline detection and tracing in SAR images”, IEEE Trans. Geosci. Remote Sens, 28 (4), 662–668, 1990. [2] D. C. Mason, I. J. Davenport, G. J. Robinson, R. A. Flather, B. S. McCartney. “Construction of an inter-tidal digital elevation model by the ‘Water-Line’ Method”, Geophysical Research Letters, 22(23), 3187–3190, 1995. [3] Joo-Hyung Ryu, Chang-Hwan Kim, Yoon-Kyung Lee, Joong-Sun Won, Seung-Soo Chun, Saro Lee, “Detecting the intertidal morphologic change using satellite data.” Estuarine, Coastal and Shelf Science, 78(4), 623–632, 2008. [4] Bin Zhao, Haiqiang Guo, Yaner Yan, Qing Wang, Bo Li, “A simple waterline approach for tidelands using multi-temporal satellite images: A case study in the Yangtze Delta.” Estuarine, Coastal and Shelf Science, 77(1), 134–142, 2008. [5] D.C. Mason and I. Davenport, “Accurate and Efficient Determination of the Shoreline in ERS-1 SAR images,” IEEE TGRS, 34(5), 1243-1253, 1996. [6] Schwäbisch, M., S. Lehner and W. Norbert, 1998. “Coastline Extraction Using ERS SAR Interferometry”. Proceedingf of the 3rd ERS Symposium On Space at the service of our Environment, Florence, Italy, 17-21 March 1997, ESA SP-414, 3Vols. May 1997. [7] S. Dellepiane, R. De Laurentiis, F. Giordano, “Coastline extraction from SAR images and a method for the evaluation of the coastline precision”, Pattern Recognition Letters, Volume 25(13), 1461–1470, 2004. [8] A. Niedermeier, E. Romaneeßen, S. Lehner, “Detection of coastline in SAR images using wavelet methods”, IEEE Trans. Geosci. Remote Sens, 38 (5), 2270–2281, 2000. [9] M. Silveira and S. Heleno, “Separation between water and land in SAR images using region-based level sets”, IEEE Geoscience and Remote Sensing Letters, 6(3), 471-475, 2009. [10] F. Soares, J. Catalao, G. Nico, “Using k-means and morphological segmentation for intertidal flats recognition”, Proceeding of the IEEE International Geoscience and Remote Sensing Symposium, 764-767, 2012. [11] F.T. Ulaby, F. Kouyate, B. Brisco, “Texture information in SAR images”, IEEE Trans. Geosci. Remote Sens, 24 (2), 235–345, 1986.

Back to list of proposals

© DLR 2004-2016