Development, deployment and validation of an oceanographic virtual laboratory based on Grid computing David Mera Pérez Santiago de Compostela, Feb. 15 th 2013
Index 1 Context and Motivation 2 Objectives 3 Virtual laboratory development 4 Virtual laboratory validation 5 Results and conclusions
Index 1 Context and Motivation 2 Objectives 3 Virtual laboratory development 4 Virtual laboratory validation 5 Results and conclusions
Context and Motivation Satellite missions for Earth observation increase every year. The study of the ocean requires multidisciplinary teams Distributed computing paradigm. 1/27
Index 1 Context and Motivation 2 Objectives 3 Virtual laboratory development 4 Virtual laboratory validation 5 Results and conclusions
Objectives 1. To develop a user-friendly distributed computational environment based on Grid computing. 2. To develop an oceanographic application to test the Grid environment. - An oil spill automatic detection system based on the analysis of satellite Synthetic Aperture Radar imaging. 2/27
Index 1 Context and Motivation 2 Objectives 3 Virtual laboratory development 4 Virtual laboratory validation 5 Results and conclusions
Retelab User access and registration The access to most of the Grids is not intuitive. - Command line interface. - Digital certicates. - The computer knowledge is mandatory. - The users management is based on les. 3/27
Retelab User access and registration Retelab approach 4/27
Retelab Distributed storage system Based on Metadata - ISO 19115. Integration of visualization tools. - Live Access Server. - Integrated Data Viewer. 5/27
Retelab Job submission and monitoring system Previous job submission systems: - They need interaction with the users. - The interaction decreases simplicity and transparency. Retelab approach: - Grid metascheduler. > To make decisions on behalf of users. > To facilitate the optimal utilization of the Grid resources. > It undertakes the tasks for resource discovery, job scheduling, executing, monitoring and output retrieval. - It was mainly developed by a CESGA researcher. 6/27
Retelab Integration 7/27
Retelab Integration 8/27
Index 1 Context and Motivation 2 Objectives 3 Virtual laboratory development 4 Virtual laboratory validation 5 Results and conclusions
Sentinazos Introduction The international trade is mainly supported by maritime transport. The intensive trac sails along the Exclusive Economic Zones (EEZ) of the countries and generates important pollution problems. Only the 7 % of oil spills come from catastrophes like tanker and oil platform accidents. Sentinazos 35% 45% Accidentes 13% 7% Emanaciones naturales Ríos 9/27
Sentinazos Introduction Synthetic Aperture Radar (a) (b) (c) Vientos de baja intensidad <3 m/s Vientos moderados Vientos fuertes > 15 m/s 10/27
Sentinazos Introduction Synthetic Aperture Radar - Examples 10 30'W 10 W 9 30'W 9 W 8 30'W 44 N 0 5 10 20 30 40 Millas náuticas 43 30'N 43 N 42 30'N 42 N 11/27
Sentinazos Introduction Synthetic Aperture Radar - Oil Spills Total number of spills=1.638 (Mediterraneam sea, 1999) Amorphous Old Spills Angular spills Less fresh spills Very fresh spills 0 100 200 300 400 500 600 Figura: Classication of detected spills in terms of their shapes. 12/27
Sentinazos Goals Hypothesis 1. Is possible to use wind information to segment oil candidates from SAR images? 2. Is the shape analysis relevant to classify the oil candidates? Goal To develop an oil spill automatic detection system focused on the galician coast and based on SAR images. 13/27
Sentinazos Methodology Dataset: a collection of 47 SAR images from the Envisat. - Galician coast. Finisterre Trac Separation Scheme (2007-2011) - Wide Swath Mode - Polarization: vertical-vertical - Coverage: 400 km x 400 km 14/27
Methodology Sentinazos Oil Spill detection system architecture Image calibration SAR images Reprojection Land mask Noise filter Preprocesing Wind data Segmentation Characterization Classification Detection system 15/27
Sentinazos Methodology Segmentation - Establishing the Adaptive Threshold. 16/27
Sentinazos Methodology Segmentation - Applying the Adaptive Threshold Pixel data > 3m/s <=Threshold Wind Speed? Selecting AT Applying AT Intensity? Labeling as oil < 3m/s Labeling as water >Threshold Generating Binary Image SAR images: 8.088 x 6.481 pixels. Incidence Angle Wind Speed Intensity Deleting Blobs <50 pixels (0.3km 2 ) Blob Area? Extracting Blobs Pixel >50 pixels 17/27
Sentinazos Methodology Characterization - The segmented areas are analyzed to get a characteristic vector: > 17 shape characteristics (Ratio area perimeter, 7 Hu moments, Thickness, etc) PCA 5 main components. > 2 physical characteristics related with the pixel intensity values. > 2 contextual characteristics related with the wind speed and the incidence angle. 18/27
Sentinazos Methodology Classication - Clustering of oil spills and look alikes. - Evaluation of the characteristics vector. - Machine learning classiers. > Articial Neural Network > Decision Tree 19/27
Sentinazos Methodology Classication Ratio intensidad < 2,04183 Sí No PCA3 < 9,49121 Sentinazo Sí No Sentinazo Look-alike 20/27
Index 1 Context and Motivation 2 Objectives 3 Virtual laboratory development 4 Virtual laboratory validation 5 Results and conclusions
Results and conclusions Base de datos de candidatos 155 Look-alikes 80 Sentinazos 85% 15% Datos para el entrenamiento 128 66 27 14 Test 70% 15% Entrenamiento 101 52 27 14 Validación Validation set Test set Sentinazos Look-alikes Sentinazos Look-alikes ANN 85,7 % 85,2 % 92,9 % 96,3 % Pruned decision tree 92,9 % 85,2 % 92,9 % 92,6 % 21/27
Results and conclusions A 10 W 9 W 43 N B C 0 5 10 20 30 Millas Náuticas A 10 W a 9 W A 10 W b 9 W 43 N B 43 N B C C 0 5 10 20 30 Millas Náuticas 0 5 10 20 30 Millas Náuticas c d 22/27
Context Objectives Virtual laboratory Validation Results and conclusions Results and conclusions 10 40'W 10 30'W 43 40'N A 0 0,5 1 10 30'W 10 20'W 2 10 10'W 3 Millas Náuticas 10 W 9 50'W 43 N 42 50'N 42 40'N B 0 10 W 2 4 8 12 9 50'W Millas Náuticas 9 40'W 42 10'N C 0 42 N 1 2 4 6 Millas Náuticas 23/27
Results and conclusions 24/27
Results and conclusions 25/27
Results and conclusions Conclusions - Processing time. - The AT could be improved using other wind models. - The oil spills inside of low wind areas are not discovered. - Is the shape relevant? Ongoing work - New wind speed models. - New satellites (Sentinel). - New classiers. - Add contextual data (ships, FTSS, etc). 26/27
Collaborations Retelab - Marine and Food Technological Centre (AZTI Tecnalia) - The Centre of Supercomputing of Galicia (CESGA) - Canarian Institute of Marine Sciences (ICCM) Sentinazos - University of Coruña - Median Engeniering Group (GIM), University of Extremadura - MacDonald Image Lab, Department of Earth, Ocean and Atmospheric Sciences, Florida State University - Spanish Maritime Safety Agency (SASEMAR) 27/27
Thank you!!