Sandra Heleno

Separation Between Water and Land in SAR Images Using Region-Based Level Sets

This letter presents a method for the separation between land and water in synthetic aperture radar (SAR) amplitude images. The proposed technique uses region-based level sets and adopts a mixture of lognormal densities as the probabilistic model for the pixel intensities in both water and land classes. The expectation-maximization algorithm is used to estimate the probability density functions for each class. Experimental results with real SAR images of riverbeds, flood extent areas, and shorelines demonstrate the good performance of the proposed algorithm compared with state-of-the-art approaches.

Multiparameter monitoring of Fogo Island, Cape Verde, for volcanic risk mitigation

Fogo Island in the Cape Verde Archipelago (North Atlantic) is a stratovolcano of nearly conical shape that rises 2829 m above sea level and V6000 m above the surrounding seafloor. With a population of 40 000, the island has known intense historical volcanic activity since AD 1500, with an average interval between eruptions of the order of 20 years. Twentieth-century rates were more subdued, with only two flank eruptions in 1951 and 1995. Following the 1995 eruption, increased awareness of the volcanic hazard affecting the population of the island led to the deployment of the permanent VIGIL Network. Seismographic stations (both broadband and short-period), tiltmeters and a CO2 sensor where installed in Fogo, together with a telemetry infrastructure to allow remote real-time monitoring. A broadband seismographic station was installed in neighbour Brava Island. The operation of the network was complemented by the introduction of routine geodetic and microgravity surveying and the operation of an automatic meteorological station. In this paper, we describe the methodology adopted to monitor the volcanic activity, combining real-time data analysis (volcanotectonic and volcanic earthquakes, volcanic tremor and tilt) with repeated surveying at intervals of several months (GPS, microgravity). Examples of data from the first years of operation are presented. In particular, the data pertaining to a period of anomalous activity in September^October 2000 are discussed, in the context of the risk mitigation strategy currently being developed.

Water/land segmentation in SAR images using level sets

This paper presents a method for the separation between land and water in SAR amplitude images. The proposed technique uses region based level sets and adopts a mixture of log-normal densities as the probabilistic model for the pixel intensities in both the water and the land regions. The expectation-maximization (EM) algorithm is used to estimate the probability density functions in each region. Experimental results with real SAR data are provided to illustrate the performance of the proposed algorithm.

Classification of water regions in SAR images using level sets and non-parametric density estimation

This paper presents a semi-supervised algorithm for the classification of water regions in SAR images. The proposed technique is based on region based level sets and non-parametric estimation of the probability density function (PDF) of the pixel intensities. The level set framework allows automatic topology adaptation and provides the regularization while the PDFs are estimated in each region using Parzen windows. Using non-parametric density estimation gives the method the flexibility to be used with different kinds of SAR data. To illustrate the performance of the proposed algorithm, the method is applied to the problems of river mapping and coastline extraction in real amplitude SAR images.

Segmentation of SAR images using textons

In this paper we investigate the use of the well known textons method [1] for the segmentation of SAR images. Two approaches were tested: using the MR8 filter bank and using only the pixel intensities. The K-NN classification algorithm and the SVM algorithm with both Linear and GHI kernels were used as classifiers. Results obtained with real amplitude SAR images for the separation between water and land demonstrate that the texton method is appropriate for the segmentation of SAR images.

Methodology to Combine Topography and Bathymetry Data Sets for Hydrodynamic Simulations: Case of Tagus River

AbstractRecent techniques for acquiring elevation, such as the advanced synthetic aperture radar (SAR) interferometric techniques, enable the creation of very detailed models in a short time. However, because of its incapacity to penetrate the water, the collection of bathymetric information in water-covered areas must be performed with other techniques. Thus, data compatibility is a key factor for two-dimensional (2D) hydrodynamic simulations because a single elevation model is required in general. Such operation is often challenging, because it is not simply obtainable by merging the data sets or replacing the main river channel information with the bathymetry. This study presents a geographic information system (GIS)–based methodology to merge the bathymetry and topography of a river and validates it with 2D flood event simulations. Therefore, the floodplain topography data sets, including emerged fluvial features and profile-shaped bathymetric surveys covering only limited sections of the river channe...

Using simplified bathymetry and SAR imagery in the validation of a hydraulic model for the Tagus River floodplain

ABSTRACT Araújo, M.A.V.C.; Pestana, R.; Matias, M.; Roque, D.; Trigo-Teixeira, A., and Heleno, S., 2016. Using simplified bathymetry and SAR imagery in the validation of a hydraulic model for the Tagus River floodplain. In: Vila-Concejo, A.; Bruce, E.; Kennedy, D.M., and McCarroll, R.J. (eds.), Proceedings of the 14th International Coastal Symposium (Sydney, Australia). Journal of Coastal Research, Special Issue, No. 75, pp. 13 - 17. Coconut Creek (Florida), ISSN 0749-0208. This work presents several approaches in the validation of the hydrodynamic model Tuflow on the simulation of flood extents and water levels, based on satellite SAR imagery. A methodology that uses a simplified bathymetry in the river main course is employed, which proves to be reliable and accurate for high-flow events. This was made possible as the digital terrain model was acquired in a dry period, accounting for large dry areas in the river bed, avoiding in this way the need of expensive river bathymetry surveys. Also, two methods are applied to the SAR imagery to extract the flood boundaries: visual interpretation followed by manual delimitation and an object-based algorithm approach. The hydraulic model is tested on a reach of the Tagus River, Portugal, where the largest flood inundation areas occur, using a historical flood event to verify its robustness and reliability. The accuracy of model prediction is done through comparisons of water levels at a hydrometric station and the determination of commission and omission errors of flood extent, between the reference SAR image and the predicted inundation. It was concluded that the methodology followed in this work is well suited for the hydraulic model validation.

Texton-Based Ensemble Classification of Landslide Source and Transport Areas in VHR Imagery

The assessment of sediments produced, displaced, and deposited by landslides is important for hazard evaluation and mitigation. However, existing methods for landslide identification seldom address the effective separation of their internal constituents (source and transport). This letter presents a methodology to classify these constituents in very high resolution remotely sensed images. It is based on an ensemble of Texton classifiers using spectral and textural information. An experimental strategy is devised to evaluate different ensembles of features. An overall accuracy of above 90% is obtained in a cross-validation procedure in GeoEye-1 images from test sites in Madeira Island.

Assessment of supervised methods for mapping rainfall induced landslides in VHR images

In this work we develop and compare three different supervised approaches for semi-automatic mapping of landslides, including the separation of landslide source and transport areas, using a single GeoEye-1 image acquired after a rainfall-induced landslide event in Madeira Island. The methodologies cover object-based classification using support vector machine (SVM) algorithms; pixel-based classification using textons; and object-based classification with a rule-set framework. The assessment was made by comparison of the results obtained in the validation areas with the ground-truth landslide mapping. In what concerns landslide recognition, the results of the object-based and pixel-based machine-learning approaches have higher accuracy when compared with the rule-set method. The object-based SVM approach achieves false positive rate FPR=20% and false negative rate FNR=18% for landslide area detection, while the pixel-based texton method displays even higher accuracy (FPR=19% and FNR=9%) although at higher computational cost and slower execution. In what concerns internal mapping of landslide source areas, the three methods show lower but still reasonably good performance, in particular in the sunnier east-facing slopes.