Founded in 300 BC by the Seleucids, the city of Antioch was famous in ancient times for its sophisticated and efficient system of water supply (Fig. 2) (Kondoleon, 2000). Drinking water mainly came from karsts overhanging the plateau of Daphne, 6 km south of the city. The aqueduct system visible today was described by the Byzantine chronicler Malalas (243, cf. Dindorf, 1831). Following an earthquake in 37 AD, Caligula built baths and an aqueduct supplying them with water from Daphne springs. Malalas (275–276, cf. Dindorf, 1831) also writes that after the earthquake in 115 AD, Trajan and Hadrian also built baths with their associated aqueduct. The subsequent fate of the aqueducts is unknown, although one of the two aqueducts may have been used at the end of the 12th century according to Johannes Phocas’ testimony (Wilber, 1938). Both aqueducts (fig. 2) were identified during archaeological surveys in the 1930s. One aqueduct (AQ1), supplying the North and the centre of the city, was dated to Caligula's reign, while the other (AQ2), located at a higher elevation and better designed, seems to date back to Trajan and supplied the southern part of the city (Wilber, 1938). It should be noted that those surveys were not extensive and have not been updated since then. During a field mission in July 2012, we identified two other portions of aqueduct (AQ4 and AQ5), which seem to belong to a third system, higher than the two lines described above.

Thanks to an extensive and almost continuous documentation since the Roman period, we know that Antioch and its surroundings have been repeatedly struck by many earthquakes (Fig. 1). According to ancient sources, the city was struck by no more than nine significant earthquakes between 0 and 500 CE (Ambraseys, 2009; Guidoboni et al., 1994). These successive earthquakes, together with the effects of time, led to several repairs that are still visible on the aqueducts.

The different building materials analysed in this study were sampled in July 2012. Twelve bricks were sampled on four different portions of aqueduct identified in the city. AQ1 and AQ2 belong to the two systems described above, while AQ4 and AQ5 are part of a third higher system. Nine bricks come from the first building stages, while three were sampled on repaired structures, which enabled us to compare the materials used in different building stages, and to check if the restoring bricks present peculiar features. The name, location, size and building context of each brick are summarised in Table 1.

The different bricks were first compared on the basis of macroscopic features (Fig. 4), such as the size, colour, and texture. Then, in order to study more precisely their manufacturing process, thin sections were made at the Laboratory of Geology of the École Normale Supérieure (Paris) and characterized under an optical microscope Leica DMRXP, equipped with a camera Leica DFC420. The nature of different components observed was also studied with an FEG Zeiss Sigma scanning electronic microscope (Table 2).

SEM picture mosaics were realised on most of the bricks, with a magnification of 500 and a resolution of 1024 × 768 pixels. On the SEM images, the pores are the only areas coloured in black, which makes it easy to compute then 2D-porosity using the JMicroVision software, (Roduit, 2007). To check the reliability of this method, 3D-porosity measurements were also carried out concurrently on two bricks (AQ4-4 and AQ5-2), using the triple-weight method (NF EN 1936, 2007 norm).

For the granulometry study, the brick thin sections AQ1-1, AQ2-4, AQ2-5, AQ2-8, AQ2-9b, AQ4-1, AQ5-1, and AQ5-2 were fully photographed in polarised light with a magnification of 1.6 and a resolution of 2592 × 1944 pixels. The pictures were then processed using the JMicroVision software. The area and equivalent circular diameter of the grains are then estimated. The percentage of area covered by the grains is then plotted in diagrams (Figs. 5 and 6).

In order to constrain the age of production of the different bricks, magnetic palaeo intensities measurements were carried out on eight samples (AQ1-1, AQ2-1, AQ2-5, AQ2-9b, AQ2-10, AQ4-1, AQ5-1, and AQ5-2) at the Saint-Maur-des-Fossés observatory, in collaboration with Yves Gallet and Maxime Le Goff (IPGP, Paris). These measurements were done with two triaxe magnetometers, whose operating principle is described in Le Goff and Gallet (2004) (Table 3). Low-field thermomagnetic curves acquired with a KLY Agico equipped with a CS2 oven up to 580 °C on AQ1-1, AQ2-5, AQ2-9b, AQ2-10, AQ4-1, AQ4-4, AQ5-1, and AQ5-2 allowed us to gain insights into the magnetic mineralogy and stability of the magnetic phases during thermal treatments. In addition, the composition, size and texture of ferri/ferromagnetic phases were described using the BSE mode during SEM observations on AQ1-1, AQ2-5, AQ2-9b, and AQ2-10 thin sections.

Charcoal fragments were extracted from the mortar of samples AQ1-1, AQ2-9, AQ2-10, and AQ4-4, and dated by Aeon Laboratories (aeonlaboratories.com) with the ¹⁴C method, to constrain the building age of the different architectural units. The radiocarbon ages were converted into calendar ages with the INTCAL09 calibration (Heaton et al., 2009) using Oxcal program (Ramsey, 2009).

All the bricks show their own characteristics in terms of texture, which may reflect different qualities of fabric (Fig. 4). AQ2-1, AQ2-2, and AQ4-1, have a rather homogeneous granulometry of the temper and show a good quality of clay-temper mixing. On the contrary, AQ2-4, AQ2-8 and AQ2-10 contain various sizes of grains and the mixing was done less careful, as showed by the presence of several clay lumps without temper. The other bricks present intermediate qualities of fabric. Fig. 5 presents the grain area percentages estimated by image processing. Most of the bricks have a value close to 30%, but some bricks, like AQ1-1, AQ4-1 and AQ5-2, exhibit lower values, while AQ2-8 shows remarkably high grain content, larger than 40%.

The mineralogy of the temper is reported in Table 2. It is composed of various fossils, such as nummulites (from Eocene limestones), and isolated minerals and fragments of rocks, which can be identified. Most of them come from basalts and gabbros. All the bricks exhibit quite similar mineralogical contents. The temper includes rounded and angular fragments, the latter often having very acute and reflex angles.

The 2D-porosity results are shown in Fig. 5. The 3D-porosity measurements carried out on AQ4-4 and AQ5-2 provided values of 28% and 16%, respectively that confirm the 2D-porosity analysis done on the photograph. The 2D analysis seems therefore to give reliable porosity estimates. The 2D-porosity values are quite similar among the bricks. Still, AQ2-4, AQ4-1 and AQ5-1 exhibit higher values, closer to 20%, while AQ2-1 and AQ2-2 show lower ones, below 10%.

Typical granulometry diagrams are shown in Fig. 6. The curves can be divided into two parts. The first one includes the smaller grains that clearly follow a Gaussian pattern, while the biggest grains lie above the Gaussian curve. It is noticeable that the grain diameter for which the grain area percentage is the highest is the same for almost all the bricks, around 40 μm, except AQ2-9b, which contains more grains of a lower diameter, close to 20 μm. The coarser grains in most of the bricks follow a regular pattern that looks like a parabola (e.g., AQ2-4). Several points do not fit on this parabola, between 200 and 700 μm, and can represent a significant area, as in AQ2-9b (fig. 6). The average diameter of these outliers is also different between the bricks. The maximum grain diameter for all the bricks does not exceed 3 mm, and it is even lower than 1 mm for AQ2-9b and AQ4-4.

Over the 24 samples measured, 6 did not provide results that fit the quality criteria, such as described in Gallet and Le Goff (2006). AQ2-5 and AQ4-1 showed anomalous results that can be explained by an unsuitable mineralogy, with an important amount of large titanomagnetites. All suitable results are displayed in Table 3. All the samples provided similar intensities, with a mean of 57.6 ± 1.7 μT. The obtained intensity values are too close, considering the associated uncertainties, to really distinguish subgroups among the bricks.

Thanks to the reference curve established for ceramics from the Nishapur site in Iran, at the same latitude as Antioch (Genevey and Gallet, 2013), but 1700 km more to the east (Fig. 7), the measured intensities can be converted into ages. As the reference curve is not established in Antioch, some errors may be generated, in particular a systematic time lag due to a westward drift of the geomagnetic secular variation has been recently under discussion (Aubert et al., 2013; Livermore et al., 2013). The time interval that we obtain with this correlation ranges from the 5th century BCE to the 1st century CE, which covers the construction age of AQ1 estimated by the archaeologists (see Section 2.2.1). Another less probable possibility would be later than the 4th century CE. As the bricks were likely produced during the Roman period, after the 1st century BCE, the production age is mostly constrained by the lower limit of the intensity interval, toward the mid-1st century CE.

Three charcoals extracted from the mortars of AQ2-9 and AQ2-10 provided similar ages of 0 ± 50 year CE (2σ uncertainty) that would match a construction during Caligula's reign. A second charcoal from AQ2-9 mortar gives a much older age, between 500 and 400 BCE, which could be due to the use of old wood in the construction. The charcoal from AQ1-1 mortar provided an age of 800 ± 100 year. The last charcoal, from AQ4-4 mortar, gives a quasi-modern age, certainly dubious as the sample was too light to be analysed properly.

The bricks AQ2-1 to AQ2-5, sampled in the arch of AQ2, are all very similar on many aspects. They were produced using the same manufacturing process; they certainly belong to the same production, and were probably made by the same ceramist. The bricks of this coherent group present variability for some studied parameters, which give an idea of the heterogeneity inside one set. It is noticeable that the porosity is quite uneven within this group, and that their mineral content may not be strictly similar. Therefore, the porosity and the occurrence of some peculiar minerals/rocks will not be used to discriminate the different productions. By contrast, the amount of temper, the granulometry, and the colour are very homogeneous within this group, and may certainly be used as relevant parameters.

A first group of bricks (AQ2-1 to AQ2-5) is characterised by a high proportion (30%) of temper, a selection of middle-sized grains (around 300 μm), a 3-mm sieving, and a rather oxidising firing (type A). AQ2-10 and probably AQ5-1 can be added to this group A. A second group (type B) can be identified, which could include AQ1-1 and AQ4-1, and is characterised by a very small amount of temper, fine grains (around 200 μm), a 3-mm sieving, and an oxidising firing. Finally, a third group (type C) differs from the type A because it presents no grain above 1 mm, and could include AQ2-9b and AQ4-4. They do not show a similar distribution of fine grains, though the peak of the Gaussian curve occurs at 40 μm for AQ4-4 like for the other bricks, against 20 μm for AQ2-9b. So, these two bricks seem to have followed the same manufacturing process, but they differ by their clay resource and may not be from the same workshop, AQ4-4 having a similar clay type to the other bricks contrary to AQ2-9b. There remain three isolated bricks, AQ2-8, AQ2-9c, and AQ5-2, all different. The three bricks share the particularity of having been fired under a reducing atmosphere, as shown by their brownish colour, but does not present similar grain size distribution: while AQ2-8 has a huge amount (over 40%) of unsorted temper, AQ5-2 contains coarse temper in very small proportion (20%).