5. Underwater Photogrammetric Recording at the Site of Anfeh, Lebanon

The coastal town of Anfeh is located approximately 70 km north of Beirut and 15 km south of Tripoli. It is bordered by the village of Chekka and the Barghoun River to the south; to the north, by the agricultural area of Hraishi and the village of Qalamoun; and to the east by the villages of Barghoun and Zakroun (Fig. 5.1). Anfeh extends westwards to the Mediterranean Sea by a 400-metre long and 120-wide peninsula (Fig. 5.2). The latter, Ras al-Qalaat, is roughly oriented on a NNW-SSE axis while rising some 14 m above mean sea level (MSL). Three rock-cut moats separate it from the mainland at its easternmost limit and are carved on a north-south axis roughly perpendicular to the peninsula (Fig. 5.3) (Chaaya 2016, pp. 281–282; Lawrence 1988, 28; Nordiguian and Voisin 1999, 381; Panayot-Haroun 2015, 399). The coastline north of Ras al-Qalaat is exposed as it offers no lee from the dominant SW winds. It consists of cliffs that drop in places onto a narrow rocky shore that is hazardous for seafaring. Closer to the peninsula, the coastline forms two large well-protected shallow bays—the Nhayreh Bay and the bay of Ras al-Safi—that offer natural havens (Fig. 5.4). Due to coastal urbanisation that started developing in the 1980s, both sides of the Nhayreh Bay are occupied by modern beach resorts. This leaves a narrow space in the bottom of the bay for the present-day modest fishermen’s harbour. To the south of Ras al-Qalaat, the rocky shoreline is low-lying and consists of a small cove with an open bay that are suitable for anchoring and landing places when the northerly winds blow, being in the lee of the promontory.

Recent terrestrial excavation and survey work on the peninsula and its hinterland by the Department of Archaeology and Museology (DAM),² University of Balamand (UoB), suggests an occupation of the promontory of Ras al-Qalaat and of the modern town of Anfeh that extends most likely from the Early Bronze Age to the Ottoman period (Panayot-Haroun 2015, 2016). In May 2016 and in September 2017, DAM³ led a one month detailed recording of underwater cultural material at the site through multi-image photogrammetry. This was preceded by a three week underwater visual survey undertaken in 2013 in the waters around Ras al-Qalaat and off the north and south coastal stretches of the modern town that aimed at assessing the underwater archaeological potential at Anfeh (Fig. 5.5) (Semaan 2016; Semaan et al. 2016).

Considering the logistical constraints in terms of time, and human and financial resources, the 2013 preliminary survey also laid the groundwork for the subsequent underwater photogrammetry surveys. The later surveys mapped submerged relevant archaeological features and artefacts, and generated accurate in situ 3D records of these, while reducing the time and cost of archaeological survey work. Indeed, this low-cost and user-friendly method for acquiring high-resolution datasets is well suited for low-budget research since, traditionally, underwater archaeological research projects are associated with high financial and logistical costs. At Anfeh, these detailed datasets constitute base-line knowledge of its seabed, which is hitherto understudied, and offers first-hand documentation for further research.

An innovative application has been the calculation of volume, and by extrapolation the weight, of in situ stone anchors based on the density of the material associated with each stone type (Table 5.3). Indeed, the standard practice of stone anchor recording or ‘anchorology’, set by pioneer Honor Frost in Mediterranean archaeology (Frost 1997), favours weight as a key typological parameter, which can reflect both the antiquity of anchors and the size of the carrying vessel. Underwater photogrammetry allows a much more accurate calculation of anchor weight, as recently demonstrated by Fulton et al. (2016). Moreover, the site’s submerged archaeological material is vulnerable to looting—thus multi-image photogrammetry allows close monitoring and is a means to mitigate potential losses.

The targeted artefacts and features recorded using multi-image photogrammetry at Anfeh can be divided into four categories:

The authors used the Agisoft Photoscan/Metashape Professional edition® software to record and map underwater material such as masonry blocks and anchors as well as coastal features, mainly the above-mentioned four slipways; and to produce orthogonal projections, sections, plans and drawings as supporting visual documentation in the framework of researching the maritime cultural landscape of Anfeh. One of the future goals is to combine the 3D rendering of anchors and masonry blocks with other digital techniques to produce a virtual tour of the modelled areas for the general public, especially the non-divers, so they gain an appreciation of the underwater cultural heritage at Anfeh (Fig. 5.6). These activities are part of a wider workflow that was inspired by Yamafune et al. (2016). As this still is a work in progress, this chapter will discuss the procedures of underwater photography, photogrammetry, and orthophotos illustrated in the workflow diagram.

During the 2016 and 2017 seasons,⁴ the team used four different underwater photography systems for data collection, in order to compare and optimise results (Table 5.1): (1) GoPro HERO4 Black Edition with its stock underwater housing; (2) Canon PowerShot G15 compact camera, with an underwater Fantasea housing (Fig. 5.7: to the right); (3) Canon EOS 70D DSRL equipped with a Canon 20-mm lens that provides a great depth-of-field while getting closer to the target. The camera was set in an IKELITE Underwater TTL housing, mounted on an aluminium tray with dual quick release handles, and with a modular 8 in. dome with 2.75-in. lens extension. The whole kit was completed with two DS-161 strobes and their light diffusers. The Ikelite housing had a hemispherical glass in front of the lens in order to minimize the refraction error due to the water/glass interface (Drap 2012, 115), and thus generate sharper image corners caused by the use of a wide angle lens (Fig. 5.7: to the left) (Yamafune et al. 2016, 7); and (4) Sony DSC-RX100 compact camera set in a Nauticam Underwater housing working with an INON UWL-H100 wide conversion lens type 2.⁵

A high performance computer is usually required to run multi-image photogrammetry software as it analyses a substantial amount of images (McCarthy 2014, 177). Therefore, the team used a MSI GT80 Titan SLI high-end gaming laptop with a 2.70-GHz Intel Core i7-6820HQ CPU, 32 GB of RAM and Dual NVidia GeForce GTX video card, which successfully dealt with the large photogrammetric datasets.

First, the various targeted areas were marked and delimited: either visually using natural markers on the seabed as references for small surfaces, or with ropes and rods for larger surfaces. The features and artefacts were then cleaned of debris and algae and then tagged. A 1-m scale bar, or sometimes two, were placed next to objects. The scale bars were graded in black and white every 10 cm. Then, a set of clear and well-exposed images was taken of the targeted area from free positions, with a considerable overlap of over 50% between images. As Drap (2012, 114) argues: ‘The key factor of this method is redundancy: each point of measured space must be seen in at least three photographs.’ This is also done to optimize results in Agisoft Photoscan/Metashape as each image is compared to every other image. The settings of the cameras, and particularly the lens zoom, were not changed during data capture.

The series of photos were taken with an estimated field of view of 45–70°, and from different positions while swimming in a circular and/or a zigzag pattern. When the objects and their wider landscape were being recorded, the diver covered the limits of the landscape first. Indeed, several oblique photos were taken around each location/group of artefacts and features to provide a wider coverage. Subsequently and in a continuous mode, the diving photographer captured a series of pictures that covered the object up-close to a distance above object between a range of 0.5–1.5 m with a 360° capture around the object itself (Fig. 5.8).

During data acquisition the GoPro camera did not provide clear high-resolution images. The focal length of its lens is quite small, and as a wide angle fisheye, caused important distortions (Capra et al. 2015, 71).⁶ The authors decided, therefore, to forego capturing images with this system and focused on the other two higher-end systems:

The images collected needed to be processed before they could be used to construct accurate 3D photo models. This was done through using Adobe Lightroom 5 and Photoshop CS5 to edit the colour balance, the contrast, and the brightness properties of the photos. The authors preferred not to greatly modify the colours of the images, in order to stay faithful to the actual state of the object and the landscape.

In the case of the isolated finds, the orthophotos were directly generated from the 3D model. In the case of the larger groups of finds—such as the different groups of anchors at the feet of the southern reef and the large area of masonry blocks on the north-eastern side of Ras al-Qalaat—however, the photomosaics were patched together from the different batches used in generating the photogrammetric models of the sites. In such instances, the batches were close enough to each other and there was sufficient overlap between each group. Therefore, the authors linked these batches by marking common points then aligning the chunks to each other according to these points.

The team was also able to able to export 3D models as 3D PDF format. One of the main advantages of generating such format files is the careful study of the material while reducing the time required for measurements underwater; as well as extrapolating measures and dimensions of artefacts taken at the office from the corresponding 3D PDF files (Tables 5.3 and 5.4).

Several anchors were located during the 2013 and 2016 visual surveys, some of which were left in situ and recorded photographically for 3D photogrammetry. One example is a one-hole triangular anchor (AN.S12.004) of substantial dimensions (Table 5.3), that was located at a depth of 12 m and lodged under the feet of the southern reef of Ras al-Qalaat (Figs. 5.9 and 5.10).

Four groups of anchors, all located scattered at the feet of the southern reef at a depth of 11–12 m below MSL and some 30–80 m from the coast, were surveyed photographically for multi-image photogrammetry (Fig. 5.10). Group 1 is located in Square S11 and includes four small-sized anchors, one of which was two-holed (AN.S11.273), with the rest being one-holed (AN.S11.286–288) (Fig. 5.11). Group 2 is located in Square S11 and includes four single-hole anchors of which two were quite large (AN.S11.292 and AN.S11.293) and the other two modest in size (AN.S11.291 and AN.S11.274) (Fig. 5.12). Group 3 is located in Square T11 and is comprised of seven single-hole anchors (AN.T11.294, 295, 296, 297, 298, 299, and 300) (Fig. 5.12). Group 4 consists of a single-holed anchor along with two stone mill elements (Anchor: AN.T10.146) (Fig. 5.13).

A few isolated blocks were recorded. These are mainly located to the south of Ras al-Qalaat. A rectangular block (AN.U9.240) was recorded in 2013 via GPS at a depth of 10.8 m about 100 m off the southern coast of the Anfeh village. It measures 1.98 × 0.45 × 0.30 m. In 2016, this masonry block was found completely covered under some 40 cm of sediments (Fig. 5.14) (Table 5.4). It was only partially uncovered as no excavation dredge was available.

The master station was positioned on two previously established topographic benchmarks (points 2011 and 2009). Pt.2011 is a geodetic point that had been corrected from the height of 35.87–13.37 m above MSL using the formula by Courbon (2016, 263) ‘Mean Sea level=Ellipsoidal height of the GPS—22.35 m.’ Knowing the absolute coordinate of Pt.2009, the total station was oriented and then other control points (with absolute coordinate and height corrected) were fixed nearer to the features to be recorded such as the masonry blocks in R10, and T8–T9.

The coordinates of most of the submerged artefacts lying at depths greater than 5 m were recorded from the surface with a handheld GPS (Garmin eTrex 10) using small numbered buoys attached to the objects, during calm weather conditions to reduce error margins as much as possible. The accuracy of this GPS position was 3 m at the surface. The depth and orientation of artefacts was recorded using the wrist compass and dive computer. The wrist compass was an Oceanic SWIV calibrated for the Northern Hemisphere. The accuracy of the depth recorded by the dive computer is slightly variable due to the tidal range but it is considered fit for purpose in this case since it does not exceed a maximum of 0.45 m.

Recording of the masonry blocks in R10 (Fig. 5.16) was undertaken using both methods. The shallow part of the masonry blocks were recorded with control points and the total station while the deeper parts were locked into GIS using GPS readings from the surface of a buoy tied to each rebar in calm weather. Depth of the blocks and rebars were taken using the dive computer under water. The authors are aware that such a method is not as accurate as the systematic use of control points. Nevertheless it mitigated potential error margins and improved the orientation and accuracy of the resulting models and orthophotos. It is, however, planned in future fieldwork seasons to return to mapped areas (whenever possible) and lock them within a network of control points, georeferenced using a total station.

In order to assess the internal accuracy of the photogrammetric survey, the authors tested results of digital measurements from 3D models against the classic recording method of taking 2D measurements in situ with a tape measure. All models were scaled in reference to the 1 m scale bar which was placed next to the objects prior to taking photographs. Anchors were measured systematically by both the diver and the archaeologist measuring from the 3D model, according to the maximum dimensions available, i.e. the highest numeric value of the length, width, thickness and hole(s) diameter of an anchor.

Table 5.5 compares the measurements of 17 anchors that were taken on the site by divers alongside corresponding measurements extrapolated from the 3D models of these anchors. The comparisons show generally relative negligible differences of 1–3 cm between the two sets of measurements. Two discrepancies of over 3 cm were ascribed to human error rather than the accuracy of the photogrammetric models.