Considering the importance of the Gnalić shipwreck, the project’s demanding underwater research conditions, limited financial resources and consequently small time frame, and the ever-present threat of losing information, the recording process had to be extremely efficient. The documentation process focused on:
Developing an accurate site plan, based on multi-layered information;
Recording the advancement of the excavation during each campaign;
Combining the results of each excavation campaign;
Continuously mapping the spatial distribution of finds before recovery;
Accurate recording of the hull;
Documenting material for future scientific presentation of the research; and
Producing material for public promotion of the project.
The initial Gnalić site recording process in 2012 followed the traditional system of tagging, measuring, drawing, photographing and video recording. Relying on the direct experience of the photogrammetric recording of the Late Roman shipwreck of Pakoštane, Croatia, carried out within the framework of an international project directed by the University of Zadar and Centre National de la Recherche Scientifique (CNRS)—Centre Camille Jullian, France,Footnote 6 and its contribution to the documentation process (Dumas 2012), it was apparent that photogrammetry should not be omitted from the organization of the underwater work. Even with that prior experience, it was challenging to ensure everything was available to guarantee the correct execution of the operation. The first attempt of photogrammetric assessment of the site, during the 2012 trial campaign, however, was extremely encouraging and resulted in photogrammetry becoming an essential part of the recording procedure.
The rapid development of the software, which became readily available and increasingly user friendly, combined with the effort in developing and testing the recording system, resulted in what is reported in the following text.
4.3.1 Trial Campaign 2012
The Gnalić shipwreck excavation restarted in 2012 in the form of a trial campaign, which lasted just 10 working days.Footnote 7 The team consisted of 15 divers, who each spent 30 min twice a day on the seabed. The area chosen for the trial excavation was a transversal cross section of the ship in the western part of the site—an area that had been previously excavated. This choice was influenced by the need to check the state of the preservation of the hull in an area that had been exposed previously and where all artefacts already would have been recovered.
It would be easy to focus on inadequate documentation inherited from the past projects, and the un-systematic assessment of the excavated areas—instead, it should be stressed that the first researchers did an excellent job of preserving the site and its finds for the future. They worked under completely different conditions, without any experience, logistics and expertise in underwater archaeological excavation. Yet, they demonstrated in fieldwork reports the highest level of awareness of the importance of the shipwreck site and a strong desire to systematically study all of its components (Radić Rossi et al. 2013: 70, 73). Therefore, everything accomplished in the restarted Gnalić project carries out respect for what was done in the past and is an attempt to realize the dream of those pioneering researchers.
At the beginning of the campaign, a metal grid, composed of seven squares each 2 × 2 m, was positioned across the site, and the excavation started simultaneously from its northern and southern edge. It should be stressed that it is usually thought that the main function of the grid is accurate documentation. This may have been true during the early development of underwater recording, but today its main function certainly exceeds documentation issues. Based on extensive fieldwork experience, with much effort invested in the training of students and amateur scientific divers, two main purposes significantly justify the positioning of the solid grid over large, delicate surfaces. First, the grid provides solid support for divers—they can rest on it while working without moving finds or disturbing the site regardless of the task they are undertaking (excavating, photographing, filming, sampling, recovering artefacts, etc.). Another important function of the grid is the orientation of the diver, i.e. reducing to a minimum any possible confusion of their assigned work area.
It was observed in 2012, however, that the grid presents a serious obstacle to accurate photogrammetric recording if the target area exceeds the surface of a single square. Therefore, during all the photogrammetric recording operations in the following years the grid was removed from the excavated areas before capturing photos, and repositioned after the operations were completed.
The trial photogrammetric recording in 2012 was led by Pierre Drap from CNRS’ Laboratory of Information Systems and Technology (LSIS) he and his team joined the excavation for 3 days. The more complex methodology and sophisticated computer programs applied at that time are no longer in use (P. Drap, personal communication). This initial photogrammetric recording was mainly targeted to demonstrating what can be relatively quickly and easily obtained through the systematic photographic recording of the excavated area (Fig. 4.2), or bigger areas in relation to the distribution of the surface finds (Fig. 4.3).
As discussed, the main goal of the trial excavation campaign was to check the presence of the cargo items and the state of preservation of the wooden elements of the hull in order to verify the need to restart the project. Therefore, it did not focus on the accurate cleaning and recording of the hull structure, which in any case would have been impossible in just ten working days. Thus, photogrammetry was the best solution for quick data collection and visualization of the situation on the seabed.
4.3.2 Research Campaign 2013
During the 2013 excavation campaign, a team led by José L. Casabán conducted the 3D photogrammetric recording of the remains of the Gnalić shipwreck covering a total surface of 300 m2 at an average depth of 25 m (Fig. 4.4). The methods applied included the trilateration of a network of control points and the photographic coverage of the shipwreck remains (Casabán et al. 2014). The images were later aligned using Agisoft PhotoScan/Metashape to produce a dense point cloud and a mesh, representing the surface of the shipwreck including its artefacts and hull remains. The model was georeferenced using a network of control points and, finally, a texture was added to the model based on the images. An orthophoto of the shipwreck was then generated from the photogrammetric model and imported into AutoCAD Map 3D to trace a site plan, while longitudinal and transversal sections were extracted from the photogrammetric model. The orthophoto plans and sections were integrated into a GIS database to perform spatial analysis.
18.104.22.168 Control Points and Multi-image Coverage of the Site
The first step of the photogrammetric recording of the Gnalić shipwreck was the creation of a network of control points to georeference the photogrammetric model and the orthophoto generated from the multi-image coverage of the site. The georeferencing system, which was applied for the multi-image coverage of the Gnalić shipwreck, was pioneered at the Institute of Nautical Archaeology (INA) excavations of the Classical shipwreck at Tektaş Burnu (Turkey) between 1999 and 2001 (Green et al. 2002: 284, 288–290), the Archaic Greek shipwreck at Pabuç Burnu (Turkey) in 2002–2003 (Polzer 2004: 3–11), and the Phoenician shipwreck at Bajo de la Campana (Spain) in 2007–2011 (Polzer and Casaban 2012: 12–14). This method is based on a series of control points evenly distributed over the site, which also appear in the photographic coverage of the mapping area. The x, y and z coordinates of each point are obtained through 3D trilateration or ‘Direct Survey Measurement’ (DSM), and depth measurements (Bowens 2008: 127–128). In other words, the distances between the points are measured using measuring tapes, while their depths are determined with diving computers. Then, all the linear data acquired in this way is processed on a laptop using Site Recorder SE, a software program which calculates the 3D position of each point based on the linear measurements and depths. The RMS residual for network of control points produced with Site Recorder SE for the 2013 photogrammetric recording of the Gnalić shipwreck was 15 mm, using a distance and depth adjustments of 40 mm and 100 mm (Fig. 4.5). In addition to the control point network, four scale bars were placed on different parts of the mapping area to provide extra measuring references. Finally, six additional distance and depth measurements were taken in specific areas of the site to determine the accuracy of the resulting photogrammetric model. The distances were taken with a measuring tape while the depths were measured using the same dive computer (Suunto Vyper) employed for the network of control points. All these extra measurements were added to the final processing of the photogrammetric model to strengthen the precision of the control point network.
After the control points were positioned on the site, a multi-image photographic survey of the site was conducted ensuring that all the site features, control points, and scale bars showed in the photographs taken. The images required a 60% vertical overlap and 80% horizontal overlap to ensure an optimum photographic coverage of the shipwreck remains.Footnote 8 The multi-image coverage of the different areas of the shipwreck was conducted in 2013 using a Nikon D200 DSLR camera equipped with a single manual strobe light that required calibration under water at the beginning of each dive. A single diver carried out the photographic coverage of the site, following parallel and transversal transects to produce enough overlap between the images. Additional photographs of particular areas of the site were taken from different angles to ensure that each part of the shipwreck was visible from at least two camera locations, and to minimize blind spots which could hinder analysis of archaeological features. A few of these blind spots still occurred, however, since the photographic coverage was mainly conducted perpendicularly to the surface of the hull remains and the time limitations of the project did not allow additional photos to be taken from different angles in all cases. Despite this inconvenience, it was still possible to produce an accurate and detailed orthophoto of the hull remains of the Gnalić shipwreck. The multi-image coverage of the excavation area was conducted several times during the archaeological season to document the different stages of the archaeological work.
The images taken following this method were then processed using Adobe Photoshop to improve their quality by manipulating the image settings such as white balance, exposure, contrast, brightness, and clarity. The image corrections were intended to ensure the best quality of the resulting photogrammetric model and the orthomosaic since the visibility conditions on site were not always ideal, and varied from 1 day to the next depending on the currents and the excavation work.
22.214.171.124 Image Processing, 3D Model, and Orthophoto Generation
After corrections, the images were processed following general workflow tasks in Agisoft PhotoScan/Metashape, until the mesh model of the shipwreck was created. At this stage, the control points that appeared in the photos were plotted manually on the model using the program tools. Then, the x-, y-, and z-coordinates of each control point were loaded into Agisoft PhotoScan/Metashape. This allowed for the georeferencing and optimization of the photogrammetric model using all the coordinates and measurements taken previously, providing a method to check the accuracy of the model. Finally, a texture based on the multi-image coverage of the site was added to the model and a 1:1 orthomosaic of the site was generated (Fig. 4.6).
The georeferenced orthomosaic was imported into AutoCAD Map 3D, a software package which combines the Computer Aid Design (CAD) tools and the main data formats used in Geographic Information Systems (GIS) to trace a 2D site plan directly from the orthomosaic in real scale (Fig. 4.7). In addition, several longitudinal and transversal sections of the hull remains were obtained from the photogrammetric 3D model of the site using software packages such as Rhinoceros or Autodesk Maya.
126.96.36.199 GIS Analysis
Finally, both the georeferenced orthophotos and site plans were integrated into ESRI ArcGIS to manage and to analyse the photogrammetric data combined with other types of information generated in any archaeological project. In this case, the GIS database is used as a tool to produce more rapidly different types of site plans that a trained nautical archaeologist will use to interpret the hull remains (see Steffy 1994: 191–250). In order to analyse the photogrammetric information provided in the orthophotos, the archaeological features and hull components represented in the site plans generated from the orthophotos would be linked to the hull catalogue compiled during the excavation. The catalogue includes the descriptions of the different hull timbers, their dimensions, types of wood, and any other observation and interpretation made by the archaeologists during the excavation of the hull remains.
The GIS database would be used then to perform different types of analyses in order to understand the site formation sequence based on the spatial distribution of hull timbers and related artefacts. In addition, different site plans of the site will be generated based on the data gathered in the timber catalogue. The criteria employed to produce the different set of plans would include the type of hull components (stem, keel, sternpost, keelson, frames, planking, ceiling planking, and other components), dimensions, types of wood used for the construction of the hull, visible scarfs and butts, and other structural components. Separate plans showing the location of scattered small finds, such as nails, will also be produced because they could provide information about hull components that were not preserved.
The analysis of the information provided in the different sets of plans generated with ESRI ArcGIS tools will be used by nautical archaeologists to understand the design and construction sequence of the vessel, including hull modifications or repairs occurred during its operational life, using reverse engineering. This information was later combined with the hull lines obtained from the mesh of the photogrammetric model in order to produce the lines drawings of the hull and, finally, the construction drawings.
4.3.3 Research Campaign 2014
In 2014, the photogrammetric recording on the Gnalić shipwreck site continued. This year the recording team, led by Kotaro Yamafune and Rodrigo Torres, experimented with slightly different methods.
188.8.131.52 Local Coordinate System
At the start of the season, a new local coordinate system was established using control network around the site. While a local coordinate system had been established in 2013, it had been difficult to create secured control points of trilateration because of the shortage of time.
It was decided to create a local control network, in order to make sure all the recording throughout the season could be related to the same x, y, z datum, providing a 4D recording strategy (x, y, z, t) which could accommodate photogrammetry, but also triangulation and grid offset plotting, if necessary. Moreover, once the local coordinate system was re-established in 2014 (Yamafune et al. 2016), it was possible to apply it to 2013 photogrammetric models using/creating the common points, such as tags of timbers that had been retained for both field seasons. The extracted x, y, z, coordinates from the common points of 2014 models were subsequently applied to the 2013 model.
The method used to establish a local coordinate system, according to conventional DSM-techniques (Atkinson et al. 1988; Rule 1989) has been published by Yamafune (2017). Regular fibreglass measuring tapes were employed for distance measurements and a diving computer for depths (10−3 and 10−1 reading resolutions, respectively). Depth measurements of all control points were taken as quickly as possible, with the same dive computer (UWATEC Aladin Ultra) to keep consistency and minimize the effect of tidal variation. Depth readings at the reference datum (D1) were recorded before and after each DSM measurement session, to function as a vertical datum and reference for tide correction throughout the excavation (Fig. 4.8).
Once the statistical (‘Best Fit’) adjustment provided by Site Recorder 4 SE (RMS 0.006 m, in this case) was considered satisfactory, a report was exported in .txt format into MS Excel. A clean x, y, z spreadsheet was then produced, with coordinates from Site Recorder 4 SE adjustment. Since this adjustment produces a network which is correctly scaled but not georeferenced, however, the spreadsheet was transferred to ESRI ArcGIS 10.1 for spatial orientation and coordinates transformation.
To do so, two control points which run across the ship’s axis were chosen (in this specific case, D1 and D5) to act as the excavation baseline. Then a temporary tight reel line from D1 to D5 was laid, and careful magnetic compass bearing was taken. With the help of the Internet (NOAA 2018) the magnetic deviation was calculated based on the inserted date and site coordinates, in order to provide models and orthophoto site plans with true north.
When the network was locked and transformed, six more internal secondary control points (SCP; 12bit water-proofed Agisoft PhotoScan/Metashape coded targets) were added in the system. These SCP were then trilaterated to fit into the primary control network, and x, y, z coordinates were derived for them. The SCP coordinates were used to spatially reference all partial photogrammetry models produced throughout the season, allowing for the overlapping, matching, and scaling of the models and orthophotos. Scale bars were also used to calibrate scaling on each recording/photo-shooting session. Here is where the 4D excavation control should start, plotting artefacts, layers, 3D models, etc. It is important to highlight that the strategy was conceived to allow for in-field processing and on-the-fly excavation feedback.
184.108.40.206 Composite Models
The second experimental approach tested on the 2014 campaign was the development of ‘composite’ photogrammetric models. Operational experience found that the calculation power of the project computer was limited; this meant that when a photogrammetric model of the entire shipwreck site, or a larger area, was created, the available detail in the 3D model deteriorated due to operator reduction of the number of meshes and resolution of textures, in order for the computer to complete the processing. Therefore, to acquire the best results for accurate archaeological information, a large site had to be separated into smaller areas. In other words, if 3D models of the site were created as separate pieces, the number of meshes and the resolution of the 3D models of the entire site could be higher. Moreover, if those 3D model pieces were created separately under the local coordinate system, these pieces could then be merged automatically in other 3D modelling software which has better rendering power. In 2014, the final composite 3D model of the campaign was created from five different model pieces, with each of those pieces created from photogrammetry performed on a different day. These five model pieces were then exported into Autodesk Maya to compose the separate models into one single model (Fig. 4.9). After the 2014 campaign a CG animation was created, and uploaded to YouTube for dissemination. Nevertheless, it is worth noting that since the August 2015, Photoscan v1.2 added a new function called ‘Build tiled model.’ This new function allows computing power to be concentrated on a small area specified by the size of a ‘bounding box’ for building mesh and texture within. In other words, once dense cloud data of the entire site is created, high resolution models of smaller area can be created within the 3D model of the site.
220.127.116.11 GIS Analysis
The third noteworthy approach tested during the 2014 field season was the improved application of GIS with photogrammetry, and its integration into the excavation workflow. Today, this has become a common application of photogrammetry in archaeological projects; however, it was still rare in 2014. The GIS software ESRI ArcGIS was used as an interactive map, updated every day using the photogrammetric orthomosaic and information gathered by the excavators. The main dataset imported into the GIS software as a base map was an orthophotomosaic generated by Agisoft PhotoScan/Metashape. Since the 3D photogrammetric models had already been created under the local coordinate system, an orthomosaic generated from georeferenced 3D models already contained georeferenced information. Whenever a new orthomosaic was generated and imported into ESRI ArcGIS, the maps of the shipwreck site were updated. Additionally, information on artefacts recovered from the site was digitally catalogued, and then linked to the GIS database. Moreover, these site plans and other information were printed out on waterproof paper, which archaeologists brought underwater to execute their assigned tasks faster and more efficiently. This served as a georeferenced database for the project, and it helped the entire excavation process by providing up-to-date information throughout the campaign.
18.104.22.168 Points-Based Deviation Analysis
The final experimental method applied during the 2014 campaign was ‘points-based deviation analysis.’ This concept takes advantage of software such as CloudCompare, which can compare two different point cloud data sets, and generate differences between the two different data sets, showing the results/calculations empirically as quantitative data, or graphically as differences in colours. In 2014, this type of analysis was applied to the repetitive photogrammetric recording of the same areas throughout the 2-month campaign. The dense points cloud data of the area, which contained large barrels and smaller casks, were exported into CloudCompare and the deviation analysis was applied (Fig. 4.10). The software displayed ‘excavated areas’ in colour and provided differences in quantitative data (for instances, how much additional surface had been excavated). Based on the results, it is suggested that stratigraphic recording and analysis could be aided by application of ‘deviation analysis‛. Additionally, this application could be used for multi-year site monitoring; in other words, the ‘deviation analysis’ based on photogrammetry could track changes on underwater sites over time, and could be used for site management plans.
4.3.4 Research Campaigns 2015 and 2016
In 2015 and 2016 systematic photogrammetric recording continued in the same manner. During the two campaigns, the excavation went on along the keel, the exposed ballast, and the area south of the big barrels filled with intense-red hematite. Extremities of the preserved keel area were reached, but the excavation continued in both directions, with the scope of identifying the broken posts.
Unfortunately, the location of the control points placed during the 2014 campaign slightly changed for various reasons. This meant that it was not possible to use these control points to add new reference points for newly exposed areas for photogrammetric recording. For this reason, different methods were applied to create the local coordinate system for 3D photogrammetric models of newly exposed shipwreck areas (Fig. 4.11a, b).
In order to apply the new method, capturing photos for photogrammetry covered slightly larger areas than necessary. Once photogrammetric models of each area were created with intentionally wider capturing areas, the photogrammetric models of the same areas from 2014 campaign were opened. Then, in Agisoft PhotoScan/Metashape, markers were created on exact mutual points on all models; since the 2014 models were already georeferenced, or contained the local coordinate system, the 2014 coordinate system could then be applied to the 2015 and 2016 models. Therefore the 2015 and 2016 photogrammetric models were georeferenced based on the coordinate system of the 2014 campaign. However, this method may be less reliable once excavated areas are extended further. In any case, the new coordinate system was essential for photogrammetric recording in following campaigns.
4.3.5 Mapping the Area of Archaeological Interest in 2017
The 2017 field season continued, following the same research and documentation procedures, in the southern and central area of the hull. According to the system, recently elaborated by Yamafune (2017), the photogrammetric mapping of the whole area of archaeological interest was successfully executed during six consecutive dives, covering the area of 60 × 20 m (Fig. 4.12).
This new methodology does not require direct measurements of control points that was applied during 2013 and 2014 campaigns. The scale bars were placed on the mapping area to scale constrain created 3D models and to allow the application of a 3D CAD software, such as 3D Rhinoceros, to create a local coordinate system.
In order to produce this photogrammetric model, the site was divided into two half-areas, 30 × 20 m each, following the requirements of one photo shooting session. Nine scale bars (five 1-m scale bars and four 0.5-m scale bars), and three coded targets, acting as reference points, were positioned in each half-area. After the successful creation of one half-area model, Agisoft Photoscan/Metashape calculated the residual error of 0.0045 m, based on 0.7-m markers/scale bars, created in Agisoft PhotoScan/Metashape software.
Once the initial model was created, however, it was possible to apply the ‘Optimize camera position’ command on Agisoft PhotoScan/Metashape, in order to fix distortions in the created model. As a result, the residual error of the Gnalić model became 0.0003 m (0.3 mm) in 0.7 m (Fig. 4.13). This theoretically indicates that the possible positional error from one end of the site to the other (around 60 m) is approximately 2.7 cm.
Once the precision of the model was considered satisfactory, the depth measurements of three reference points were taken using a dive computer. When the photogrammetric model was created, scaled, its camera positions optimized, and re-processed, distances between three control points were calculated within Agisoft PhotoScan/Metashape software. Then, a triangle was created in Rhinoceros 3D CAD software, and its position was adjusted based on depths of three reference points. Finally, Rhinoceros provided x, y, and z coordinates of the adjusted position of these local reference points, which were applied to the photogrammetric models in Agisoft PhotoScan/Metashape (Fig. 4.14).
This new methodology has been applied on various underwater archaeological sites, with successful results. It is a great advantage that it does not require much preparatory work to create an accurate local coordinate system for photogrammetric recording, such as DSM. For instance, for recording the Gnalić Shipwreck site in its entirety, one dive to places scale bars and three reference points, and two dives for photography on the western half-area were necessary; followed by one dive to move the scale bars and position the reference points in the eastern half-area, and two more dives to complete the photography. This means that it required just six dives (25 min each), or 150 min of one diver’s time, to cover 60 × 20 m surface, while maintaining accuracy of 2.7 cm residual errors over the whole length of the site (i.e. for 1 m long object possible error is 0.4 mm).