COSMO-Coast Tor Vergata Dipartimento di Informatica, Sistemi e Produzione, L Aquila Dipartimento di Architettura ed Urbanistica La Sapienza Dipartimento Ingegneria Civile, Edile ed Ambientale
Introduction Project background Study-area Observables Data
Objectives The goal of the project is to test methodologies based on existing algorithms for the analysis of coastal areas, focusing on the monitoring of coastline evolution at different temporal and spatial scales. The general objective is to test and demonstrate the usefulness of spaceborne EO for coastal studies and management. Particular attention is paid to the exploitation of COSMO-SkyMed data, which shall prove interesting for mapping and monitoring purposes, in view of the constellation s observation capabilities (ground resolution and frequency of acquisitions). Steps to achieve such objective were: Acquisition of satellite data acquired over the coastline of Abruzzo and on specific test-sites within the same area Creation of protocols for information extraction from the data and intercomparison between images acquired at different dates Validation of satellite observations with ground truth Evaluation and visualisation of area variations of the beaches Estimation of the feasibility to monitor the evolution of the coastline with EO data
Problem Since the 90 s an increasingly enlarging area of Abruzzo is reported in coastal erosion. Structures and infrastructures were endangered and defense works were implemented. This resulted in a domino effect calling for additional expenses To date there seem to be no clear overview on status of erosion and effectiveness of those interventions. Most of past studies are at a not optimal scale Only local (punctual) studies at right scale Analysis often limited to a comparison of outcomes from different sources with different (not documented) procedures and with different inherent errors
Study-area 1 Martinsicuro 2 Cerrano 1 2 3 Pescara 4 Ortona 3 4 5 5 Fossacesia 6 San Salvo 6
Data used Type Tot Characteristics Year(s) Coverage COSMO Stripmap 13 HH; 25.36-47.23, ASC- DES, Right 2009, 2010, 2011 Abruzzo COSMO Spotlight 16 HH - VV; 23.86-55.90, ASC-DES, Right Left 2010, 2011 Pescara Ortona Vasto ERS SAR 1 VV; 19.2, DES 1993 Abruzzo ENVISAT ASAR 1 VV; 18.57, DES 2009 Abruzzo ALOS PALSAR 1 HH; 38.7, ASC 2009 Abruzzo IKONOS 4 5 bands 2004, 2007 Pescara Ortona KOMPSAT-2 1 5 bands 2010 Pescara WORLDVIEW-2 1 9 bands 2010 Ortona FORMOSAT-2 2 5 bands 2010 Pescara
In situ data and campaigns
Processing and accuracies estimation Geolocation accuracies Methodologies for Coastline extraction Coastline extraction accuracies/validation
Optical dataset Geolocation Accuracies 1-3 m, depending on location and dataset Image type RMS E (m) RMS N (m) RMS (m) IKONOS 2004 Ortona 0.82* 0.86* 1.19* IKONOS 2004 Pescara 2 1.47 2.48 IKONOS 2007 Ortona 1.01 1.17 1.55 IKONOS 2007 Pescara 2.5 1.88 3.13 KOMPSAT-2 2010 Pescara 2.85 1.99 3.48 FORMOSAT-2 2010 Pescara 1.66 2.92 3.36 WORLDVIEW-2 2010 Ortona 0.9 1.04 1.38 * Residuals estimated on GCP, not on CP: precision and not accuracy estimated in this case
Radar dataset: Geolocation Accuracies Tests with 1 Spotlight image NO DEM SRTM DEM (90 m) High Resolution DEM (2 m) Different SW used NEST PCI Geomatica Comparison with ASI GTC product Denomination No ortho ASI GTC SRTM High Res DEM RMS E (m) 4.165 2.368 3.927 3.082 RMS N (m) 2.697 1.898 2.029 1.878 RMS (m) 4.961 3.034 4.420 3.609
Radar dataset: Geolocation Accuracies Dataset Parameter Before coregistration After coregistration RMS E (m) 1.604 Tests run with the whole dataset: Spotlight: 1-4 m Stripmap : 3-8 m Right: examples of results Spotlight Pescara 06/07/2010 04:01 Spotlight Pescara 07/07/2010 04:01 Spotlight Pescara 07/07/2010 16:54 Spotlight Pescara 09/07/2010 04:07 Spotlight Pescara 10/07/2010 05:13 Spotlight Pescara 10/07/2010 17:00 Spotlight Pescara 14/05/2011 04:46 RMS N (m) 1.265 RMS (m) 2.042 RMS E (m) 2.962 1.224 RMS N (m) 1.922 0.536 RMS (m) 3.530 1.336 RMS E (m) 1.284 0.871 RMS N (m) 2.585 1.347 RMS (m) 2.886 1.604 RMS E (m) 4.012 1.574 RMS N (m) 1.945 1.315 RMS (m) 4.458 2.051 RMS E (m) 1.093 0.453 RMS N (m) 1.167 0.868 RMS (m) 1.598 0.979 RMS E (m) 3.266 0.565 RMS N (m) 1.066 0.681 RMS (m) 3.435 0.884 RMS E (m) 1.596 1.522 RMS N (m) 2.156 1.639 RMS (m) 2.682 2.236
Optical dataset: Coastline extraction Semi-automatic techniques based on object-oriented approach. Application on pansharpened data
Radar processing: the approach used Multi-looking & De-speckling Edge detection Tracing PCNN Standard approach* Input image Gamma Filter Sobel Mean Treshold Robert s PCNN Tracing Output line *For C-band SAR, based on Lee and Jurkevitch (1990) Coastline detection in SAR images, IEEE Trans, vol 28, 4
Pulse Coupled Neural Network (PCNN) Neural Networks algorithms have been shown to be a rather competitive approach for automatic image classification in remote sensing compared to other traditional approaches. PCNN is a relatively novel unsupervised neural network that when applied to image processing, yields a series of binary pulsed signals, each associated to one pixel or to a cluster of pixels. Pulsing Nature of PCNN: the neuron has a Feeding compartment (receiving both an external and a local stimuli), a Linking compartment (only receiving the local stimulus) and an active Threshold value. When the Internal activity becomes larger than the threshold, the neuron fires and the threshold sharply increases. Afterwards, it begins to decay until once again the internal activity becomes larger. 25 20 15 10 5 0 0 2 4 6 8 Epoches 10 12 14 16 18 20 F L U Y
Radar Dataset: Coastline extraction
PCNN results validation Data available: PCNN Extracted coastline from COSMO-SkyMed acquisition Kinematic GPS simultaneous to COSMO-SkyMed acquisition Validation results: Denomination Average (m) σ(m) First set GPS 1.0 0.66 Second set GPS 1.1 0.91
PCNN comparison with the classic technique Data available PCNN Extracted coastline Lee Extracted coastline Manually Extracted coastline Kinematic GPS Comparison results: Vs kinematic GPS (1 Km 2 ) Vs manual (all image) Processing type σ (m) Traditional 17.41 PCNN 3.99
Analysis of the sources of inaccuracy in PCNN
Conclusions Coastline evolution over Pescara test site Additional applications of COSMO on the area Final remarks
Monitoring coastal evolution: Pescara Harbour 1. Spotlight Jul 2010 Spotlight Jul 2011 2. Himage Jul 2010 Himage Jul 2011 3. Rates 2007/11 Rates Jul 2010/11 Trends Erosion Accretion Rates (WLR) m/y 2007-2011 2010-2011 (jul) <-3-3 +3 > +3
Potental COSMO-SkyMed applications Monitoring of harbor activities 24 hrs change detection in Pescara harbor. In blue: boats having left the harbor (July 6), in yellow: newly arrived boats (July 7). Spotlight InSAR couple
Additional COSMO-SkyMed applications Cartographic updating Regional Cartography (scale 1:5000) dates to 2004/2008
Research conclusions PCNN technique validated for X-band coastal retrieval Accuracy shown to be within the range of the pixel size COSMO-Spotlight data seem suitable for coastal studies: Potential for selecting acquisitions corresponding to high/low or 0 tidal height Achievable geolocation accuracy comparable to VHR optical ASI GTC product showed an excellent geolocation accuracy For the test area and the given wind conditions at the time of the acquisitions, HH polarisation resulted preferable to VV (less sensitive to the backscatter of the wavefronts), Incidence angles 30-40 (better contrast in the scene), Right Descending (or Left Ascending) with an orbit as parallel as possible to the coast Over the study-area and especially in flat areas, long (>2 years) observation timeframes are advisable, due to the entity of phenomenon to monitor and meteomarine bias Multi-temporal monitoring (with COSMO and with multi-sensor VHR) demonstrated
Questions? Francesco.palazzo@esa.int