2. STUDY AREAS Northern Area The area is included into the physiographic unit that goes from the mouth of the Fosso Chi- arore (North re- gional limit) to Cape Linaro (nearby Sant’Agostino). The coast is fringed by par- tially vegetated dunes while to the south is highly atrophic. Within the whole unit grain size is predomi- nantly medium to fine sand, with spots of bioclasts and coarse material transported by rivers (La Monica & Raffi, 1996).The coastal stretch between Fosso Chiarone and Fiume Marta is characterised by continuous low slope (1,2 -1,7 %) sandy beaches with bars parallel to the coastline between 5 m and 10 m of depth (La Monica & Raffi, 1996). Regional scale stud- ies outlined that two subareas can be identified in the north and the south of Punta delle Morelle, where Posidonia oceanic meadow is present. Off shore Punta Morelle the sea floor is irregular due to several bed rock outcrops, riverbeds and intramatte spots in the south (between 10-20 m of depth; La Monica & Raffi, 1996). In this area, many sedimentary de- posits and structures associated to paleo river beds have been observed in front of the river mouths of Chiarone, Tafone, Fiora, Arrone and Marta. From the sedimentological point of view, grain size of superficial sediments increase southward and decrease offshore. Inhomo- geneous distribution of sediments have been observed between Torrente Tafone and Fiora River due to the presence of meander river beds, paleo dunes and silty sediments (Chiocci &La Monica, 1999). Fine sediments characterised the sea floor between Torrente Arrone and Marta River mouths. Southern Area The area is characterized by the presence of a beach-dune system with a trend parallel to the coastline between Capo Portiere and Torre Paola, covering just over 80% of physiographic unit, bounded to the NW and SSE respec- tively, from As- tura river mouth and cape of Monte Circeo. The beach-dune system is the remnants of a barrier Island that separated the coastal plain from the open sea during the Holocene sea level rise. It is about 24 km long, low slope (~ 1%), arched in shape, 80-250 m wide, up to 20 m of height nearby Torre Paola (La Monica & Raffi, 1996, Beachmed-e, 2007a,b; ISPRA, 2008). It is formed by sandy deposits consolidated by specialized vegetation, typical of Mediterra- nean Sea. More dense arboreous and shrub vegetation is present on the land side compare to the sea side, where higher aridity and salinity allows halophyte and xerophilous prolifera- tion and positive feed-back mechanisms of sediment deposition have been occur (Beachmed-e, 2007a). Between Anzio and Cape Circeo, the beach dune system is inter- rupted by several small inlets which connect the coastal lake to the sea and Fiume Astura, but their contribution to coast nourishment is negligible (La Monica & Raffi, 1996) and the coast is suffering of severe erosion. From the morphological point of view a complex bar system has been recognized and its dynamic, even if it is still under investigation, is presented in the present work. Use of airborne LiDAR and hyperspectral data to study the sandy beach morphology along the Lazio region coast (Italy) C. Innocenti (*), A.Taramelli (*), E. Valentini (*), S. Cappucci (**) (*) ISPRA - Italian National Institute for Environmental Protection and Research, Rome, Italy (**); ENEA – National Agency for new technologies, energy and sustainable development 8. References F. L. Chiocci e G. B. La Monica, “Analisi sismostratigrafica della piattaforma con- tinentale”, Il Mare del Lazio–Elementi di oceanografia fisica e chimica, biologia e geologia marina, clima meteomarino, dinamica dei sedimenti ed apporti continentale. Regione Lazio. Tip. Borgia. Roma (1996): 40-61. Bart Deronde et al., “Use of Airborne Hyperspectral Data and Laserscan Data to Study Beach Morphodynamics along the Belgian Coast”, Journal of Coastal Research 225 (2006): 1108-1117. Beachmed-e (2007a) - La gestion stratégique de la défense des littoraux pour un développement soutenable des zones côtières de la Méditerranée. Interac- tions de Posidonia oceanica et Sable avec l’Environnement des Dunes Naturelles. Cahier Technique étendu de Phase A. http://www.beachmed.eu/ Beachmede/SousProjets/POSIDUNE/tabid/99/Default.aspx . Beachmed-e (2007b) - La gestion stratégique de la défense des littoraux pour un développement soutenable des zones côtières de la Méditerranée. Interac- tions de Posidonia oceanica et Sable avec l’Environnement des Dunes Naturelles. Cahier Technique étendu de Phase C. http://www.beachmed.eu/ Beachmede/SousProjets/POSIDUNE/tabid/99/Default.aspx . F. L. Chiocci e G. B. La Monica, “Analisi sismostratigrafica della piattaforma con- tinentale”, Il Mare del Lazio–Elementi di oceanografia fisica e chimica, biologia e geologia marina, clima meteomarino, dinamica dei sedimenti ed apporti continentale. Regione Lazio. Tip. Borgia. Roma (1996): 40-61. V. L. Lucieer, “Spatial uncertainty estimation techniques for shallow coastal sea- bed mapping.” (University of Tasmania, 2007), http:// eprints.utas.edu.au/1919/. J. H Maindonald e J. Braun, Data analysis and graphics using R: an example-based approach (Cambridge Univ Pr, 2007). J. L Myers e A. Well, Research design and statistical analysis (Lawrence Erlbaum, 2003). C. Small, “The Landsat ETM+ spectral mixing space”, Remote Sensing of Envi- ronment 93, n°. 1-2 (2004): 1–17. R. Team, “R: A language and environment for statistical computing”, R Founda- tion for Statistical Computing Vienna Austria ISBN 3, n°. 10 (2008). S. F. Tebbens, S. M. Burroughs, e E. E. Nelson, “Wavelet analysis of shoreline change on the Outer Banks of North Carolina: an example of complexity in the marine sciences”, Proceedings of the National Academy of Sciences 99, n°. 1 (2002): 2554. E. Verfaillie, V. Van Lancker, e M. Van Meirvenne, “Multivariate geostatistics for the predictive modelling of the surficial sand distribution in shelf seas”, Con- tinental Shelf Research 26, n°. 19 (2006): 2454–2468. 1. INTRUDUCTION This research address the multisensors methodology in coastal morphology by a combined use of airborne LiDAR (Light Detection and Ranging) and Hyperspectral MIVIS (Multi- spectral IR and Visible Imaging Spectrometer) data to study a beach-dune system. A phys- ics based approach was applied to MIVIS and LiDAR airborne data, simultaneously ac- quired on 12 May 2009 in order to integrate geomorphological and sedimentological obser- vations into a detailed coastal map of two study areas of the Lazio Region, Tyrrhenian Sea (Central Italy)(Fig. 1-2). 6. MIVIS - Grain size samples Correlations between grain size fraction and radi- ance of the MIVIS bands A preliminary study was conducted on Sa- baudia beach to assess the capacity of MIVIS in identifying the grain size difference in the emerged beach. Twelve samples were collected along 3 different transect few days after the image ac- quisition. To each sediment sample was associated the mean value of the 4 nearest cells, obtaining 12 records with 17 grain size fractions (Tab.) and 102 radiance values measured. The correlation between grain size and radiance was explored by means of both Pearson and Spearman cor- relation coefficients to highlight respectively the linear and non linear dependency (Myers and Well 2003, Maindonald and Braun, 2007). The analysis was con- ducted by means of the sotware R (R Development Core Team, 2010). In Fig. and Fig. are summarized the correlation coeffi- cients with p-value > 0.05 between grain size fractions and MIVIS bands. From graphics is possible to see: • Several correlation seems to be not linear (i.e. 123 μm and 187 μm fractions show significative Pear- son correlation coefficient, but do not show sig- nificative Pearson coefficients) • The wavelengths more correlated with the grain size dimension are comprised between 2.083 μm and 2.39 μm. • Almost all the correlation coefficient are negative. The last observation leads to the conclusion that the ra- diance response is more likely linked to the minera- logical composition of the samples than to the grain size. As example, in Fig. is reported a scatter plot between the 1000 μm fraction and the band 63 by MIVIS. 720000 720000 722000 722000 724000 724000 726000 726000 4672000 4672000 4674000 4674000 4676000 4676000 4678000 4678000 4680000 4680000 Depth Standard Deviation High : 4.2 m Low : 0 m UTM 32 N WGS84 0 20 40 60 80 100 -1.0 0.0 0.5 1.0 1414.21 micron MIVIS Bands Spearman's rho 0 20 40 60 80 100 -1.0 0.0 0.5 1.0 1000 micron MIVIS Bands Spearman's rho 0 20 40 60 80 100 -1.0 0.0 0.5 1.0 707.11 micron MIVIS Bands Spearman's rho 0 20 40 60 80 100 -1.0 0.0 0.5 1.0 250 micron MIVIS Bands Spearman's rho 0 20 40 60 80 100 -1.0 0.0 0.5 1.0 175 micron MIVIS Bands Spearman's rho 0 20 40 60 80 100 -1.0 0.0 0.5 1.0 147 micron MIVIS Bands Spearman's rho 0 20 40 60 80 100 -1.0 0.0 0.5 1.0 125 micron MIVIS Bands Spearman's rho 0 20 40 60 80 100 -1.0 0.0 0.5 1.0 123 micron MIVIS Bands Spearman's rho 0 20 40 60 80 100 -1.0 0.0 0.5 1.0 103 micron MIVIS Bands Spearman's rho 0 20 40 60 80 100 -1.0 0.0 0.5 1.0 87 micron MIVIS Bands Spearman's rho 0 20 40 60 80 100 -1.0 0.0 0.5 1.0 73 micron MIVIS Bands Spearman's rho 0 20 40 60 80 100 -1.0 0.0 0.5 1.0 Medium Sand MIVIS Bands Spearman's rho 0 20 40 60 80 100 -1.0 0.0 0.5 1.0 Fine Sand MIVIS Bands Spearman's rho 0 20 40 60 80 100 -1.0 0.0 0.5 1.0 1414.21 micron MIVIS Bands Pearson's rho 0 20 40 60 80 100 -1.0 0.0 0.5 1.0 1000 micron MIVIS Bands Pearson's rho 0 20 40 60 80 100 -1.0 0.0 0.5 1.0 250 micron MIVIS Bands Pearson's rho 0 20 40 60 80 100 -1.0 0.0 0.5 1.0 175 micron MIVIS Bands Pearson's rho 0 20 40 60 80 100 -1.0 0.0 0.5 1.0 147 micron MIVIS Bands Pearson's rho 0 20 40 60 80 100 -1.0 0.0 0.5 1.0 103 micron MIVIS Bands Pearson's rho 0 20 40 60 80 100 -1.0 0.0 0.5 1.0 73 micron MIVIS Bands Pearson's rho 0 20 40 60 80 100 -1.0 0.0 0.5 1.0 Fine Sand MIVIS Bands Pearson's rho 0.0000 0.0004 0.0008 0.0012 3 4 5 6 Band 63 1000 micron Pearson's rho = -0.63 (p-value = 0.0284) Spearman's rho = -0.6 (p-value = 0.0395) 5. MIVIS - LiDAR By merging MIVIS radiance with LiDAR elevation, the hyperspectral data are associate to bathymetry and two transect of spectral profiles were extracted from vegetated and sandy submerged areas (Fig. 10). Bottom radiance dependence on depth is analyzed in the figure above where the two target shows different trends within the spectrum. Sandy bottom are clearly influenced by depth along all the visible range. For vegetated bottom (dotted lines), more absorption is re- corded if compared with sands. Moreover radiances of vegetation decrease with depth only in the blue region, then from 0.46μm to 0.54μm it seems to be more influ- enced from the spectral variability of vegetation then from the bathymetry. We can assess that there are threshold along the visible spectral range of submerged target that are both spectral and physical: the spectral range from 0.46μm to 0,62μm within the shallow waters from 0 m to – 6 m (Fig. 11 ) is the more suitable for understanding differences within ho- mogeneous target (sands or vegetation) but part of these differences is influenced from bathymetry. 0.000 0.500 1.000 1.500 2.000 2.500 3.000 0.44 0.46 0.48 0.50 0.52 0.54 0.56 0.58 0.60 0.62 0.64 0.66 0.68 0.70 0.72 0.74 0.76 0.78 0.80 0.82 Radiance Wavelenght Sands and vegetation: MIVIS radiance along Lidar DSM - 0.1 m sands - 1.0 m sands - 2.0 m sands - 2.6 m sands - 4.5 m sands - 0.7 m vegetation - 1.0 m vegetation - 2.1 m vegetation - 3.0 m vegetation - 5.1 m vegetation 3. LiDAR In the south area (Sabaudia) two successive LiDAR surveys (February 2008 and May 2009) have been compared: • First of all, the vertical collimation of the two DSMs, calculated on the same grid, was successfully checked on the pavement of a coastal road (mean difference almost zero). • The shoreline was calculated from the two dataset using the contour line + 30 cm, in order to avoid both the major uncertain associated to the land-water interface in the LiDAR surveys and the shoreline changes associated to the tide (Tebbens et al. 2002). • The two shorelines had been intersected by every 13.4 m obtaining 2048 transect in 27.6 km, and in every locations the amount of accretion or erosion was measured. • A new raster representing the elevation changes on the emerged beach was obtained as subtraction of the 2009 DSM from the 2008 DSM. • The two shorelines had been intersected by every 13.4 m obtaining 2048 transect in 27.6 km, and in every locations the amount of accretion or erosion was measured. • A new raster representing the elevation changes on the emerged beach was obtained as subtraction of the 2009 DSM from the 2008 DSM. The shoreline and elevation change analysis seen to fit well each other: where the shoreline in accretion the difference in quote is positive and conversely. Although the results could be biased by extreme meteo-marine events before the surveys, the results show a general erosion of the Sabaudia beach dur- ing the period March 2008 - May 2009, suggested by both the shorelines’ relative positions and the height differences. The mean shoreline erosion is 2.5 m, the longest segment of shoreline in erosion (red dotted line) is 2500 m in length , and the longest segment of shoreline in accretion (green dotted line) is 640 m in length. Even if some accretion was observed (blue areas), the general evolutionary trend is erosive (brown ar- eas). As consequence, the mean height of beach is decreased by nearly 12 cm, equivalent to a loss of about 105000 m^3 of sediment. The LiDAR bathymetry was added to the MIVIS dataset to enhance the capabilities of seabed pattern recognition. In Fig. 10 is pos- sible to see two transects of pixels, respectively on vegetated and sandy seabed. The relative reflectances are joined in Fig. 12. It is possible to notice, that the values of the two datasets (vegetated seabed in green and sediment in red) mix together and is impossible to distinguish the different type of seabed on the basis of reflectance. In Fig. 13 the LiDAR bathymetry has been added to the MIVIS reflectance in a scatter plot, and now the two datasets are well distinguishable. The emerged beach/dune system of the southern part was well described by calculating the slope of the DSM. The foot was placed where the slope of the beach, moving inshore, changes abruptly Fig. 4. The submerged beach have been described by means of the bathym- etry position index (BPI). The BPI has been used in several sedimentological study (Verfaillie 2006) and geomorphological (Lucieer 2007). The BPI permits to distinguish the depth of a DSM cell rela- tively to the surrounding seabed. The BPI has been calculated in Raster Calculator of ArcInfo 9.3 following the expression: BPI = Zgrid − focalmean(Zgrid, circle, r) The submerged beach of the Southern area is characterized by homogenous fine sand and a system of sand bars parallel to the coastline. Be- tween Capo Portiere and Lago di Caprolace sand bars, usually develop within 200 m from the coastline. They are generally placed within the range of 2-3m and 4-5 m of depth and they are overlapped by different second order bed forms (Fig. 4). It is particularly evident nearby Cater- attino (Lake of Sabaudia), where on the crest of the second bar, sand dunes, formed under the effect of NW-SE long shore currents are oriented perpendicularly to the coastline. Different sedimentary structures can be observed in front of Fogliano Lake, where a third order of bar is pre- sent at about 7 meter of depth. Below the average closure depth, at 7-8 m of depth, the sea bed is smooth and regular showing few depressions between Rio Martino, Lago dei Monaci and Lago di Caprolace. Wavelength 0.54μm Radiance Frequency 1.5 2.0 2.5 3.0 0 500 1000 1500 2000 -12 -10 -8 -6 -4 -2 0 1.5 2.0 2.5 3.0 Wavelength 0.54μm Depth Radiance Hawk Eye II Airborne bathymetric LiDAR Acquisition Topographic soundings per second 64000; Bathymetric soundings per second 4000. Nominal depth 3 times secchi disk; Nominal altitude 250 - 500 m; Swath 100 - 330 m Sounding density Hydrographic data: Typical from 1.7 x 1.7 to 3.5 x 3.5 m; Sounding density topographic data: from 4 to 1 points per square meter. Topo-DSM: 2x2 m; Bathy-DSM: 4x4 m. Topo accuracy: Horizontal: ± 0,5 m, Verti- cal: ± 0,15 m; Bathy accuracy: Horizontal: ± 2,5 m, Vertical: ± 0,25 m Date of acquisition: May 2009 Previous Topographic LiDAR Acquisition Soundings per second (kHz):100 – 70; Flight height: 1050 – 1300 m; Resolution: 1.6 – 1.5 point/m^2; Overlap: 30%. Date of acquisition: March 2008 320500 320500 321000 321000 321500 321500 322000 322000 322500 322500 323000 323000 323500 323500 4585000 4585000 4585500 4585500 4586000 4586000 4586500 4586500 4587000 4587000 Shoreline changes Erosion Accretion Elevation changes High : 2 Low : -2 Changes occurred in Sabaudia Beach between March 2008 and May 2009 UTM 33 N WGS84 The standard deviation map, calculated on a moving neighborhood of 3x3 cell, permits to discriminate the sandy seabed (low values) from the rock or seagrass covers (high values). In Fig. 3 an example of std dev map is reported, showing different roughness of the seabed that correspond to sand (blue) and vegetation/bed rock (brow). 7. Summary Several conclusions reached in the present study can be summarized as follow: • The standard deviation of the bathymetry permit to discriminate sandy and flat seabed from rocky or sea- grass covered seabed with a major rugosity (Fig. 3) • Maps derived from the LiDAR DSM, such as slope and BPI, permit to separated the different elements of emerged and submerged beach/dune system. (Fig. 4) • Comparison of LiDAR DSMs from successive sur- vey is an unique instrument to understand the beach dynamics and to quantify the erosion/accretion phe- nomena (Fig. 5) • Emerged vegetation of beach dune system spectral properties are influenced by sandy presence but SMA can resolve both sandy presence and vegetation struc- ture analysis (Fig 7) • Within the interval -3 – -6 m vegetation can be well distinguish from sand by means of SMA. • Joining the LiDAR depth data to the MIVIS radi- ances, the capability to distinguishing different sea- bed covers is increased (Fig. 13) • MIVIS radiances on the sandy beaches seem to be more influenced by the mineralogical content than grain size (Fig. 14) 323500 323500 324000 324000 324500 324500 325000 325000 4584000 4584000 4584500 4584500 4585000 4585000 Slope High : 40 Low : 0 Bathymetric Position Index High : 2 Low : -2 Foot of the dune Morphometry of emerged and submerged beach Fogliano Lake UTM 33 N WGS84 4. MIVIS Hyperspectral MIVIS data has been processed in order to discriminate sand and vegetation spectral properties both on emerged and submerged beach by Spectral Mixing Analysis (SMA). Observed radiance of heterogene- ous mixed pixel is decomposed into its endmembers abundances. On high resolution images as the MIVIS data, pure and non pure (mixed) pixels can be dissected mathematically to discern which compo- nents are represented in a specific area. Fraction maps (Fig. 6 ) are used to describe the composition of single pixels. They can drive dis- crimination of image’s subsets where variability is grouped by spectral properties as well as spatial analysis. The dune boundaries were identified by multispectral data with high spatial resolution such as large scale aerial photography, existing land cover/use map and by field obser- vation. Dune systems subset, within the hyperspectral images were then extracted on the base of LiDAR topography (Fig 7). PCAs for each hyperspectral subset (Tab. 2) of dune were analyzed and the apexes of the multidimen- sional space of the PC1 vs PC2, were used as training set for the Maxi- mum Likelihood classification of vegetated and sandy pixel. The phenological spectral characteristics of vegetation were then evaluated on the base of PCA re- sults of 28 bands (0,43 – 1,55 μm) only for vegetated pixel. Fraction map (Fig. 7) were then generated in or- der to distinguish the structural composition of dune vegetation, that was divided into fraction of: (1) grass- For submerged beach analysis the 102 bands of hyperspectral MIVIS images were resample and only two subset were conserved. The first 20 bands (0.43-0.83 μm) penetrate shallow marine waters up to 10 m of depth (Fig 8) and has been used for sand and vegetation extraction. Second, 10 bands of the thermal wavelengths (8,20—12,7 μm) were useful for masking operation. By a cascade of PCA and multidimensional space generation (Fig 9a), spectral properties of bottom were first characterized and 4 endmember were selected (Fig 9b). Sector Bands Range (μ) Band width (μ) VIS+NIR 1-20 0.433 - 0.833 0.02 NIR 21-28 1.150 - 1.550 0.05 SWIR 29-92 2.000 - 2.500 0.008 TIR 93-102 8.200 - 12.70 0.34-0.54 Date of acquisition: May 2009; Pixel sixe: 2x2 m MIVIS Hyperspectral Scan MIN MAX MEAN Dev.St. Band 1 (yellow‐ sands ) ‐4,04702 2,37380 0,06297 0,38120 Band 2 (cyan ‐ shallow waters) ‐0,16638 1,14203 0,00910 0,05638 Band 3 (green ‐ vegetation) ‐0,77223 2,44271 0,09299 0,21023 and 4 (blue ‐ optically deep water ‐1,40473 3,28916 0,07576 0,31158 Band 5 (RMS error) 0,00000 0,21373 0,00644 0,01430 Linear Spectral Mixing Analysis provided the frac- tion map of sandy and vegetated bottom with lower accuracy then in the case of dunes. The high RMSerror (Tab 3) along bathymetric LiDAR DSM suggested an analysis of water depth influ- ence on the bottom leaving signals. (Fig 10). In submerged beach, spectral variability due to sands and vegetation, can be described not only in terms of spectral profile for each pure or mixed pixel, but also in terms of cluster. By visualizing fraction in color composite R = sands, G = vegeta- tion, B = True Color Image. The first 3m of depth, mixed pixel (more sandy then vegetated) are sparse along the lower PC2 scores. Another cluster of both endmember, sands and vegetation is from –3m to –6m were spectral behavior are really distinct. Over the –6m isobaths, spectral properties of mixed pixel shadows in a single spectral signature that is comparable with the one of optical deep waters. UTM 32 N WGS 84 Fig. 1 Fig. 2 Fig. 3 Fig. 4 Fig. 5 Fig. 6 9. Acknowledgement This research is funded by the Lazio Region within the CAMP Project. A special thanks is addressed to Eng. Paolo Lupino. The topographic LiDAR survey of March 2008, has been provided by National Cartographic Portal. Thanks for the contribution to the achievement of this work goes to Luisa Nicoletti as a co-PI, and to Mateo Conti and all the research an technical staff involved in LIDLAZ project. Fig. 7 Fig. 8 Fig. 9 a b Tab. 1 Tab. 2 Tab. 3 Fig. 10 Fig. 11 Fig. 12 Fig. 13 Fig. 14 Fig. 15 Dimension of the sieves 2000 353.55 125 73 1414.21 250 123 62.5 1000 176.78 103 707.11 175 88.39 500 147 87 Tab. 3