Photogrammetric Methods for Monitoring Cliffs with Low Retreat Rate

Transcription

1 Journal of Coastal Research SI ICS2009 (Proceedings) Portugal ISSN Photogrammetric Methods for Monitoring Cliffs with Low Retreat Rate P. Redweik, R. Matildes, F. Marques and L. Santos Dept. of Geographical Engineering, Geophysics Faculty of Sciences, University and Energy, Faculty of Sciences, University of of Lisbon, 1749 Lisbon, Portugal, Lisbon, 1749 Lisbon, Portugal, predweik(at)fc.ul.pt rita.matildes(at)oniduo.pt, marialeonorsantos(at)gmail.com Dept. of Geology, Centre of Geology, Faculty of Sciences, University of Lisbon, 1749 Lisbon, Portugal, fsmarques(at)fc.ul.pt ABSTRACT REDWEIK, P., MATILDES, R., MARQUES, F. and SANTOS, L., Photogrammetric methods for monitoring cliffs with low retreat rate. Journal of Coastal Research, SI 56 (Proceedings of the 10th International Coastal Symposium), Lisbon, Portugal, ISSN Sea cliff retreat is dominated by mass movements which are a source of natural hazard. Hazard assessment requires complete inventories of cliff failures, usually based on multi-temporal aerial surveys by means of simplified or photogrammetric methods. However, low retreat rate cliffs still pose problems, because parts of the cliff top contour remain unchanged during the monitoring period. The inherent difficulties of a multi-temporal processing of aerial photos (different flight directions and scales, insufficient ground control and poor radiometric quality) require a refined photogrammetric assessment in order to overcome discrepancies such as false mass movements that can not be included in the inventories. Because cliff retreat is discontinuous in space and time, better sampling of the process requires the widening of the monitoring time window, by use of older aerial photos, which involve further problems as unavailable camera calibration and photo distortions. To address these problems, the 13km long sea cliffs of Burgau-Lagos coast (Southwest Algarve, Portugal) were studied using digital photogrammetry methods, which involved aerotriangulation, stereo plotting of the cliff top, ridges and toe and automatic generation of digital terrain models. Inconsistencies in the results for the different epochs were solved by systematic stereo photo interpretation, supported by oblique aerial photos. All relevant results were organised in a Geographic Information System designed for the purpose allowing future and more elaborated analysis. ADITIONAL INDEX WORDS: Cliff failures; Landslides, Photogrammetry, GIS, Coastal management INTRODUCTION Slope mass movements, including rock falls and different types of landslides, are the dominant process of sea cliff retreat (TRENHAILE, 1987; SUNAMURA, 1992), a significant source of natural hazard and a constraint for human activities. In fact, the increasing human use of cliffy coastal areas made cliff instability phenomena a problem with growing importance in several areas of the world. Natural hazard prevention and mitigation request accurate coastal monitoring, in order to support coastal management policies based on scientific approaches. By analogy with landslide hazard assessment (VARNES, 1984, GUZZETTI et al., 2005), a complete sea cliff hazard assessment requires the determination of spatial, time and magnitude components of cliff instabilities, which are usually derived from the analysis of systematic inventories of past events, based on aerial photographs and historical maps. Such an inventory has been compiled for the Algarve cliffs through photo interpretation and measurements on aerial photos of several epochs (methods in MARQUES, 2006) and extensive field work (MARQUES, 1997). Although providing a rich data base over a 142 km long coastline, the used methods did not enable the desired degree of objectivity and measuring accuracy, and do not allow easy update. To validate and improve the existing inventory, a digital photogrammetric study was made in one section of the southern coast of Algarve. The selected section corresponds to the coast between Burgau and Lagos, which is a typical low retreat rate cliff dominated coastal segment, with computed average cliff retreat rates from 10-2 to 10-3 m/year (MARQUES, 1997). The 13 km long cliffs are composed by sections of alternating beds of Cretaceous marly limestone and marls, marls, and sandstones, and of Miocene weak calcarenites, heavily affected by karst features, which are partially filled with Plio-Pleistocene sandy deposits. Cliff height varies from 20 m to more than 100 m and cliff face slope varies between circa 50º to near vertical with base notches and locally overhanging portions. Due to the very irregular time distribution of cliff retreat events and for a better sampling of the process, it is necessary to use the wider time window possible, implying the use of older aerial surveys. These tend to be problematic in terms of quality and geometric information necessary for proper photogrammetric processing. To set up an accurate and precise methodology and to provide a scientifically based support to land management, photogrammetry provided very adequate solutions with the additional advantage of allowing control and quality assessment of the produced data. METHODOLOGY For this study a set of 11 flights was collected dating from 1947 to 2007 and covering completely the coastal section. After careful analysis, three flights were chosen according to the criteria: widest time window, best ground control, best scale or GSD (Ground 1577

2 Photogrammetric methods for monitoring cliffs with low retreat rate Sample Distance) and best radiometry. Table 1 shows the main characteristics of the chosen surveys and the workflow is illustrated in Figure 1. Spatial orientation of all flights had to be determined by aerotriangulation, using the specific software BLUH ( based on ground control points defined for the INAG 2002 survey. The aerotriangulation results enabled stereo plotting and generation of digital elevation models (DEM) for the three epochs. These were produced using automatic stereo correlation of the images supported by stereo plotted break lines of the top, ridges and toe of the cliffs. All photogrammetric procedures that precede aerotriangulation and DEM generation were conducted in the photogrammetric software LISA ( Both creation of epipolar pairs of photos and the stereoplotting of elements were conducted in PCI Geomatics Geomatica 10 software (Orthoengine module). Using ArcGIS 9.2 (ESRI) difference DEMs between each pair of epochs were derived, revealing the height changes in the coast and their distribution in planimetry. A thorough interpretation of the results allowed separating the changes due to human activity from those corresponding to actual landslides in the coast. Geomorphologic and dimensional parameters could then be obtained in order to quantify each cliff retreat event. Table 1: Technical features of the used aerial surveys (MATILDES et al, 2008) Name - Date SPLAL-1952 INAG-2002 IGP-2007 Scale/ GSD(m) 1:18000/ :8000/ :44000/ 0.50 Strips Images Focal length (mm) PROJECT DEVELOPMENT Spatial Orientation of Aerial surveys The used aerial surveys presented different original formats. The original film of the SPLAL-1952 flight was not available, so it was necessary to scan the existing paper prints. The INAG-2002 flight was obtained by a film camera and had to be scanned for this study. As for the IGP-2007 flight, it was originally digital, obtained by a DMC Intergraph digital aerial camera. For the INAG-2002 flight there was a set of 30 well documented ground control points (GCP) that could be used in the aerotriangulation. As no ground control for the other flights was available, 80 well distributed common points between the three surveys had to be identified and collected. This task was especially hard to accomplish for the SPLAL-1952 flight, since the landscape changed considerably in this region in a 55 year long period. Camera information, fundamental for a precise geometric handling of each aerial image, was not available for the older flight. Thus, a pseudo camera had to be defined for this flight, based on the focal length of mm printed on the pictures and on the mean values of the photo coordinates of fiducial marks that have been measured in all 56 available SPLAL scanned photos. Strip geometry was poor and half of the photos showed water in 70 to 90% of its area, causing serious problems in aerotriangulation, which were solved by using a dense network of Figure 1. Flowchart of the study process. ground control points. This process was repeated iteratively adding further control points and redesigning ground control network until the residuals on ground control points were approximately 1 GSD. As an accuracy indicator for the results of the aerotriangulation, the RMSE values for the control points (CPs) are presented in table 2. These values were satisfactory considering the characteristics of the used aerial surveys and the purpose in view. Table 2: Statistical Results for the aerotriangulations (REDWEIK et al, 2008) INAG-2002 SPLAL-1952 IGP-2007 RMSE X ±0.109 m ±0.800 m ±0.327 m RMSE Y ±0.117 m ±0.670 m ±0.369 m RMSE Z ±0.258 m ±1.362 m ±0.841 m Sigma 0 18 µm 22.5 µm 37.8 µm CPs GSD 0.18 m 0.41 m 0.50 m Generation of Digital Elevation Models The generation of detailed digital elevation models of the region of interest in each epoch was the keystone of this project. In fact, when supported by an adequate set of break lines, a DEM generated from aerial photos by image correlation presents a high level of detail and can have as much resolution as the image itself. For a comparative study, however, the mesh dimension must be chosen according to the least geometric resolution (greater GSD) of the images involved and to the least planimetric accuracy resulting from the spatial orientation by aerotriangulation. Thus, the effect of a possible planimetric error in one DEM can be 1578

3 Redweik et al attenuated in order to avoid erroneous conclusions about elevation changes when considering two different DEMs of the same spot. Although a reference DEM with 20 cm resolution was generated for the INAG-2002 flight, the mesh dimension of 1 m was chosen for the comparison DEMs. Landslide detection in DDEMs A typical behaviour of cliff failures that occur in this region in the Cretaceous rock cliffs is exemplified in Figure 2. The lost material at the top of the cliff builds a deposit at the base which is afterwards removed by beach dynamic processes. a) b) c) Figure 2. Schematic section of one cliff failure (Marques, 1997). Based on this fact an algorithm was developed in order to detect this type of occurrences automatically. Two difference DEMs (DDEM) were built: DDEM [ ] and DDEM [ ]. The resulting models had the same mesh dimension as the DEMs, each cell representing the height change in its position. Positive values revealed an increase in height, negative values a decrease, from the older to the newer epochs. For visualization, a chromostereographic color ramp was applied, enabling an automatic detection of possible cliff instabilities and a 3D perception of the relief changes. This was very useful for visual interpretation of the changes in its general context. A narrow buffer was applied along the coast line in the chromostereographic representation. Possible landslides appear enhanced as small areas of lost material, sometimes neighbored by similar small areas of gained material (Figure 3). Figure 4. Example of detected landslide: a) result of the DDEM [ ]; b) classification of areas of lost and gained volume. Cliff top and toe of 2002 in white and 1952 in black; c) oblique photo of Praia do Canavial. Cliff failure assessment To define the horizontal areas lost at cliff top the stereo plotted lines corresponding to cliff top in each epoch were superimposed in the XY plane. Except for vertical cliffs that suffer retreat, there is, in general, no geometric coincidence (Figure 5) between lost areas at cliff top and areas of lost volume highlighted by the DDEMs. Although the latter are very important for the event detection and for the displaced volume calculation, the first are the most relevant for hazard assessment and definition of set back lines, and had to be determined in a separate step of the process. Also the affected cliff top length and width were measured in this step. Figure 3. DDEM and possible landslides (original in color). Validation of detected ocurrences The cliff failure candidates were analyzed one by one (Figure 4- a and -b) in order to confirm its existence and enabling the determination of the number of events occurred within each time interval. To quantify lost areas at cliff top, volumes of displaced material, affected cliff top length and width, a thorough analysis was made supported by additional information such as oblique aerial photos (Figure 4c) and orthophotos. Figure 5. Limit definition for area of lost volume and for lost area at cliff top (vertical cross section). Displaced volumes of rocks were obtained by a sequence of operations on the DDEM. In each area of lost or of gained volume, the pixel area on the ground is multiplied by each pixel value (=height change) yielding an elementary volume. The sum of all elementary volumes within an area results in the displaced volume. Assuming no failed values exist, an equivalent result is obtained by multiplying the area on the ground by the mean value of all pixels within the area. 1579

4 Photogrammetric methods for monitoring cliffs with low retreat rate The former calculation is done separately for areas of lost and of gained volumes. The difference between both values referring to the same landslide indicates how much material has been removed from or brought to a particular location, by external actions. Accessing the information The relevance of this kind of study becomes visible first when the involved information is well organized in order to allow both a flexible access to users and the generation of new data through spatial analysis operations. Therefore, a geographic information system (GIS) had to be designed containing several layers of information sharing the same geographic reference: a 1:2000 scale contour map; a 1: scale geological map; buildings and roads at 1:2000; orthophotos dating from A set of partial results, such as stereo plotted cliff top and toe and DEMs corresponding to each analyzed epoch, is attached to the GIS allowing a spatial-temporal analysis of the data collected so far and the integration of new data in the future. All the cliff failures are located and have an associated set of attributes (Figure 6) including date or time interval of occurrence, affected cliff length, lost area at cliff top, displaced volume, lithology, geology and type of landslide. New detected occurrences can be easily integrated in the database. Figure 6. Example of information automatically available when choosing a determined landslide: geometric and qualitative information, and the correspondent oblique photo, allowing a better analysis of the event. Orthophoto on the background. Figure 7. Detected landslides between 1947 and 2007 in the Burgau (W) - Lagos (NE) cliffy coastal section. 1580

5 Redweik et al. and 1991, horizontal area lost 3368m 2 ) are much more conservative than those produced with photogrammetry (122 cliff failures between 1947 and 1991, horizontal area lost 10234m 2 ), which produced mean cliff retreat rates 2 to 3 times higher than those previously proposed and enabled the identification and characterization of a much larger number of retreat events. There are also substantial differences in the cliff retreat pattern mainly due to the occurrence of an exceptionally large cliff failure at Praia do Canavial (Figure 4), which lead to a very high increase of the sub-section retreat rate. This strongly stresses out the irregular patterns of time and space frequency of cliff failures, indicating the need for wide space and time retreat event sampling windows. Figure 8. Cumulative plot of horizontal area lost at the cliff top by cliff failures ( ) with cliff length. The linear regression slope of data for sections with different lithology corresponds to the average cliff retreat rate for the 60-years study period, in meters. The mean cliff retreat rate is obtained dividing the slope of the best fit lines by the number of years studied (60). Figure 9. Cumulative plot of horizontal area lost at the cliff top ( , MARQUES, 1997) with cliff length, with linear regressions of data for sections with different lithology. (Compare with Figure 8). RESULTS The present study enabled the detection and characterization of 137 cliff failures that occurred between 1947 and 2007 along the 13 km long cliffs of the studied coastal section (Figure 7), which caused the loss of m 2 of horizontal area at the cliffs top. The cliff failures identified correspond to 3 main types: planar slides (58%) mainly in Cretaceous alternating limestone and marls; toppling failures (17%) mainly in Miocene calcarenites; slumps (15%) in Plio-pleistocene silty sands that infill the karst in the Miocene rocks. The remaining 10% correspond to complex movements, rockfalls and not determined cases. The space distribution of cliff failures is quite irregular (Figure 8), not changing substantially the patterns defined with the former inventory (Marques, 1997, Figure 9). However, the comparison of the two inventory data sets shows that the results obtained by photointerpretation (Marques, 1997; 44 cliff failures between 1947 CONCLUSIONS The photogrammetric methods used in this study, and specially the strategies to obtain reliable information from old aerial photos, provided an accurate characterization of the cliffs and supported a reliable compilation of a systematic inventory of cliff failures occurred in a 60 years period, which is a considerably large time sampling window of the cliff retreat phenomena. These methods provided a confirmation and enhancement of a previous inventory made with simplified methods, providing a more detailed picture of the space and time distribution of cliff failures, but without changing the general pattern of the studied phenomena. The adequate use of these methods requires a considerable degree of expertise, of working time, of resources for proper digitalization of film based aerial surveys, software and hardware, and, for reliable compilation of an inventory data set a final systematic checking of each cliff retreat event is still required. The specific topic of DDEM generation for reliable assessment of volumetric data is still in development and may provide a significant improvement of the outputs of these methods in the particular topic of low cliff retreat rate monitoring. LITERATURE CITED GUZZETTI, F., REICHENBACH, P., CARDINALI, M., GALLI, M., ARDIZZONE, F., Probabilistic landslide hazard assessment at the basin scale. Geomorphology, 72, pp MARQUES, F.M.S.F., The sea cliffs of the coast of Algarve. Dynamics, processes and mechanisms. University of Lisbon, Ph.D. thesis, 556 p. MARQUES, F.M.S.F., A simple method for the measurement of cliff retreat from aerial photographs. Zeitschrift für Geomorphologie - Supplementbände, 144, p MATILDES, R., REDWEIK, P., MARQUES, F., SANTOS, L., Detecção e medição de recuos nas arribas no litoral do Algarve para integração em SIG. X Encontro de Utilizadores de Informação Geográfica (Oeiras, Portugal), pp REDWEIK, P., MARQUES, F., MATILDES, R., A strategy for detection and measurement of the cliff retreat in the coast of Algarve (Portugal) by photogrammetry. EARSeL eproceedings, 7 (2), pp SUNAMURA, T., Geomorphology of Rocky Coasts. Wiley, New York, 302 pp. TRENHAILE, A.S., Geomorphology of Rock Coasts. Clarendon Press, Oxford, 384 pp. VARNES, D.J., Landslide Hazard Zonation: a Review of Principles and Practice. UNESCO Press, Paris, 63pp. 1581