Depositional and diagenetic history of travertine deposited within the Anio Novus aqueduct of ancient Rome

A plentiful source of fresh drinking water, combined with construction of an efficient and reliable water supply and distribution system, have been the fundamental requirements needed to sustain large centers of civilization throughout human history (Scarborough, 1991). Ancient Rome still serves as the pinnacle example of water supply infrastructure, where the largest and most complex water delivery systems yet envisaged by humankind enabled population density to reach unprecedented levels that rival those of modern-day urbanization (Morley, 1996). Beginning in 312 B.C., 11 aqueducts were built to service Rome, which in the end had 502 km of channels that were constructed either below ground or elevated above ground atop arcades and embankments depending on local topography and elevation (Hodge, 2002a). These sprawling aqueduct systems were maintained at considerable expense for more than a millennium, which testifies to their importance to Rome's elite (Coates-Stephens, 1998; Staccioli, 2007). Eventual expansion of the aqueduct systems throughout the Roman Empire provided long-distance water supply technology to Europe, North Africa, and the Near East (Hodge, 2002a). In fact, Roman aqueducts were so effective and durable that their basic design remained in widespread use until the Industrial Revolution, and some Roman aqueducts are still in use today (Keenan-Jones, 2013).

Ancient Rome brought unprecedented societal integration and urban and economic development to an area that was large enough to be comprised of a dozen modern-day nation-states. One consequence was the dissemination of aqueduct technology throughout this area, which resulted in the construction of more than 2300 aqueducts (C. Passchier, 2018, ROMAQ: The Atlas Project of Roman Aqueducts, https://www.romaq.org/; accessed September 2020). Extensive aqueduct record keeping, completed during the Republican and especially high Imperial periods, has survived and provides detailed insight into the functioning of the Eternal City's aqueduct system. In particular, in A.D. 97, Sextus Julius Frontinus was appointed Rome's water commissioner and compiled detailed (albeit sometimes internally inconsistent and inaccurate) written descriptions (Frontinus, 2004) of the aqueducts and associated water distribution pipelines that supported an estimated 600,000–1,000,000 people (Ashby, 1935; Keenan-Jones, 2015). However, outside of Rome and even within Rome after Frontinus, surviving records are patchy, particularly as the empire and its capital faced ongoing monumental challenges. For example, these included multiple outbreaks of disease such as the Antonine plague (A.D. 166), the third-century A.D. military anarchy or "Imperial Crisis," and repeated military incursions over the next several centuries across the northern and eastern frontiers that together caused the break-up of the Western Roman Empire in the fifth century A.D. (Heather, 2005). For the city of Rome, it is not until the late medieval period (ca. A.D. 1300) that documentation exists rivaling that of the high empire a millennium earlier, with flood records of the Tiber River. Therefore, much remains unknown about this period, during which there was gradual breakdown of the numerous aqueduct systems. This gap in historical information has proven particularly vexing for scholars attempting to draw inferences from ancient Rome to evaluate the possibilities of the future collapse of modern global society (Ferguson, 2010; Grinin et al., 2010; Yoffee and Cowgill, 1991).

An invaluable archive of information about human activity and climatic change during these poorly documented periods of history are the calcium carbonate (calcite, CaCO₃) limestone deposits that encrusted the interior floors, walls, and sometimes even ceilings of aqueduct channels that extended throughout the Roman Empire (Ashby, 1935; Hodge, 2002a, 2002b). Due to their striking earthen colors and stratigraphic successions of thin, dark–light laminations, these highly prized Roman aqueduct limestone deposits were later quarried and polished for use in church construction throughout Rome and other cities (Ashby, 1935; Grewe, 1991). The geoscience classification of similar CaCO₃ limestone deposits precipitated in natural terrestrial environmental settings includes the terms travertine (precipitates from high- to low-temperature springs, also called carbonate sinters), tufa (precipitates from low-temperature springs, lakes, and waterfalls), and speleothem (precipitates from waters in high- to low-temperature subterranean caves or fracture and fault systems) (Sanders and Friedman, 1967). However, previous archaeological studies have used a wide variety of terms for the limestone deposits that formed within Roman water supply infrastructure, including travertine, tufa, calcareous crusts, calx, calcium carbonate sinter, encrustation, lime, and lime scale (e.g., Aicher, 1995; Bobée et al., 2011; Brinker, 1986; Carlut, 2011; Carlut et al., 2009; Carrara and Persia, 2001; Coates-Stephens, 2003a, 2003b, 2003c; Dubar, 2006a, 2006b; Garbrecht and Manderscheid, 1992; Gilly et al., 1971; Gilly, 1971; Hodge, 1992; Lombardi, 2002; Schulz, 1986). In the present study, limestone deposited from the water that flowed within aqueducts will be referred to as travertine (Fouke, 2011; Pentecost, 1995a, 2003, 2005; Pentecost and Viles, 1994; Sanders and Friedman, 1967). Furthermore, the term travertine will be used here in a purely descriptive sense that is solely determined by basic rock properties (e.g., crystalline structure, mineralogy, chemical composition, and stratigraphy) and without a priori interpretation of the specific depositional or diagenetic environmental conditions in which it was deposited.

Roman aqueduct travertine deposits primarily formed as a result of mineral precipitation directly from the fast-flowing, chemically saturated waters being carried by the aqueducts. Only a minor component of the travertine was formed by downstream hydraulic transport of sedimentary particles and plant debris within the aqueducts. Frontinus (Frontinus, 2004) and others in ancient Rome mention aqueduct travertine accumulation as a maintenance problem, but they do not give many details of how it was handled (Fahlbusch, 1991; Frontinus, 2004; Leveau, 1991; Taylor, 2000). Textual and physical evidence suggests that travertine was regularly removed by aquarii slaves working along the aqueduct channels using hammers, chisels, spades, and perhaps even vinegar, to prevent flow restriction within the aqueduct channels (Bobée et al., 2011; Bruun, 1991; Coates-Stephens, 2003a, 2003b; Fahlbusch, 1991; Grewe and Blackman, 2001; Leveau, 1991; Passchier, 2015; Taylor, 2000). However, not all of the travertine was removed during cleaning, and it therefore continued to accumulate from the flowing water after maintenance ceased and the aqueducts fell into neglect. Travertine preserved within aqueduct ruins therefore provides an invaluable crystalline stratigraphic and geochemical record of the hydrology and chemistry of the water that flowed within ancient aqueducts throughout the Roman Empire (e.g., Ashby, 1935; Bobée et al., 2011; Garbrecht and Manderscheid, 1992; Hostetter et al., 2011; Lombardi et al., 2005; Passchier and Surmelihindi, 2019; Passchier et al., 2021; Passchier et al., 2016a; Passchier et al., 2016b; Passchier et al., 2020; Passchier and Sürmelihindi, 2010; Passchier et al., 2013; Passchier et al., 2011; Puliti et al., 1986; Sürmelihindi, 2013; Sürmelihindi, 2018; Sürmelihindi and Passchier, 2013; Sürmelihindi et al., 2013a; Sürmelihindi et al., 2019; Sürmelihindi et al., 2013b).

Over the past decade, a groundbreaking series of studies have reconstructed paleoenvironmental conditions and archaeological information from the crystal growth, stratigraphic layering, and geochemistry of aqueduct travertine deposited in the ancient Roman aqueducts of Italy, Turkey, France, and Jordan (e.g., Hostetter et al., 2011; Passchier and Surmelihindi, 2019; Passchier et al., 2021; Passchier et al., 2016a; Passchier et al., 2016b; Passchier et al., 2020; Passchier et al., 2013; Sürmelihindi, 2018; Sürmelihindi and Passchier, 2013; Sürmelihindi et al., 2013a; Sürmelihindi et al., 2019; Sürmelihindi et al., 2013b). In addition to containing stratigraphic sequences of dark–light laminae, these aqueduct travertine deposits exhibit systematic changes in their isotopic and trace element chemostratigraphy. Coupled with changes in travertine crystal growth morphology and evidence of human maintenance, this chemostratigraphic evidence has been utilized to reconstruct annual air temperature, rainfall, source water input, and Roman maintenance strategies (e.g., Berking et al., 2018; Hostetter et al., 2011; Passchier and Surmelihindi, 2019; Passchier et al., 2021; Passchier et al., 2016a; Passchier et al., 2016b; Passchier et al., 2020; Passchier et al., 2013; Sürmelihindi, 2018; Sürmelihindi and Passchier, 2013; Sürmelihindi et al., 2013a; Sürmelihindi et al., 2019; Sürmelihindi et al., 2013b). At the same time, extensive research on analogous travertine deposits in natural hot-spring, river, lake, and cave systems around the world has shown that travertine deposition results from complex, intertwined abiotic and biotic processes (Andrews and Brasier, 2005; Della Porta, 2015; Fouke, 2011; Frisia, 2015; Pentecost, 1995b). The similarity in crystalline structure and composition of these natural travertine deposits suggests that aqueduct travertine also results from complex and ever-changing physical, chemical, and biological processes that were active at the time of deposition. Later diagenetic physical, chemical, and biological processes then serve to strongly influence the extent and products of post-depositional alteration (Bathurst, 1975; Mcillreath and Morrow, 1990; Rodríguez-Berriguete, 2020; Tucker and Bathurst, 1990; Tucker and Wright, 1990). However, to date, these understandings of travertine deposition and diagenesis in natural systems have not been fully applied to determining the preservation and diagenetic alteration of aqueduct travertine.

Of importance to the study of ancient aqueduct travertine is the recognition that twenty-first–century scientific exploration is rapidly moving toward transdisciplinary approaches that are capable of simultaneously characterizing physical, chemical, and biological processes within complex natural and engineered systems (National Research Council, 2014). This scientific integration is made possible by the ongoing development of new technologies capable of simultaneously characterizing geological and biological components comprising complex natural systems across broad spatial and temporal scales (U.S. Department of Energy, 2009; Fouke, 2011; Wang et al., 2015). Called GeoBioMed when applied to human medicine, this approach has recently been applied to reconstruct the crystallization and diagenetic history of kidney stone formation (Saw et al., 2021; Sivaguru et al., 2020; Sivaguru et al., 2021a; Sivaguru et al., 2018a). For example, standard optical brightfield (BF; transmitted light), polarization (POL), and autofluorescence (AF) thin section microscopy has previously been conducted on aqueduct travertine using standard research-grade petrographic microscopes with a maximum spatial resolution of 3–5 μm (Passchier et al., 2021). However, microscopes developed for medical research have recently been used on kidney stones to permit super-resolution AF (SRAF) imaging at a spatial resolution of 140 nm (Saw et al., 2021; Sivaguru et al., 2020; Sivaguru et al., 2021a; Sivaguru et al., 2018a). This represents a more than ten-fold enhancement in thin section optical and laser imaging resolution, which is nearing resolutions attained with scanning electron microscope (SEM) imaging. Another example is three-dimensional (3-D) microcomputed tomography (microCT) X-ray analyses, which now permit external surface rendering and non-destructive internal virtual cross-sectioning of whole hand sample solid crystalline deposits from millimeters to tens of centimeters in diameter at a spatial resolution of 3 μm (Saw et al., 2021; Sivaguru et al., 2020; Sivaguru et al., 2021a; Sivaguru et al., 2018a). This represents an improvement of two orders of magnitude over standard X-ray image resolution and offers expansive new opportunities to understand the internal and external structure of solid crystalline deposits such as kidney stones and aqueduct travertine. Application of these next generation optical, laser, electron, and x-ray imaging capabilities to human kidney stones have demonstrated that they record multiple events of original crystal deposition that are strongly impacted by repeated events of diagenetic alteration (Saw et al., 2021; Sivaguru et al., 2020; Sivaguru et al., 2021a; Sivaguru et al., 2018a).

The present study was undertaken to apply these GeoBioMed imaging and conceptual approaches to reconstructing the depositional and diagenetic history of travertine deposited within the archaeologically significant and protected ruins of the Anio Novus aqueduct at Roma Vecchia (Fig. 1). Several previous studies have described, measured, and documented the structure, design, and architecture of the Anio Novus aqueduct itself (Ashby, 1935; Mancioli and Sartorio, 2001; Reina et al., 1917; Van Deman, 1934). Furthermore, the hydrology and conveyance of the Anio Novus has also been reconstructed from the 3-D distribution and total thickness of travertine deposited within the aqueduct channels (Keenan-Jones et al., 2015; Motta et al., 2017). Taken together, these extensive previous studies make the Anio Novus and its travertine deposits an exceptionally well-suited archaeological laboratory within which to apply GeoBioMed, high-resolution optical, laser, X-ray, and electron microscopy. This establishes the resulting depositional and diagenetic reconstructions of the Anio Novus aqueduct travertine as exemplars for studying travertine deposited in other ancient water transport and storage systems around the world.

The study of travertine deposited within aqueducts is confronted at the onset by challenges posed by scale, complexity, and time (Anderson and Lewit, 1992; Goldenfeld et al., 2006). To address these factors, a "Powers of Ten" conceptual framework for aqueduct travertine was adopted (Fouke, 2011). This approach recognizes that the primary components that make up ancient aqueducts span length scales of 10⁻⁷ m to more than 10² m. This nine-orders-of-magnitude dynamic range can be studied with a combination of standard, well-established and next-generation microscope instrumentation, which in the present study range from 140-nm-resolution SRAF microscopy to field-based observations and measurements of the upstream to downstream sample sites along a 140 m run of the Anio Novus aqueduct channel (Fig. 1). This Powers of Ten framework for the study of Anio Novus aqueduct travertine: (1) served to spatially and temporally frame the experimental design and sampling strategy, which is dictated by permit access, site conservation, and the number and size of samples that can be collected; (2) determined the choice and application of next-generation light, laser, electron, and X-ray microscopy techniques; (3) permitted direct comparisons of the aqueduct travertine with travertine deposited in hot springs, caves, and other natural environments; and (4) guided the interpretation and discrimination of aqueduct travertine depositional and diagenetic crystalline fabrics.

The largest and most significant expansion of the water supply system of ancient Rome commenced with construction of the Claudia and Anio Novus aqueducts, which was started by Emperor Caligula in A.D. 38 and finished by Emperor Claudius in A.D. 52 (Fig. 1) (Aicher, 1995; Ashby, 1935; Hodge, 2002a). The Anio Novus aqueduct, built directly on top of the Claudia aqueduct (Figs. 1C–1D and 2A–2B), was the highest and farthest-reaching of all 11 of Rome's aqueducts. The Anio Novus carried water over a distance of 87 km from its muddy and turbid Aniene (Latin: Anio) River source near Subiaco (Fig. 1A), where a settling tank was built by the Romans between the river and the aqueduct intake to help clean the water (Aicher, 1995; Ashby, 1935; Hodge, 2002a). Constructed underground for most of their length, the Claudia and Anio Novus aqueducts emerge on raised embankments and arcades at Roma Vecchia (Fig. 1C–1D), which is 9 km southeast of Rome (Fig. 1A) (Aicher, 1995; Ashby, 1935; Hodge, 2002a). Here, the Anio Novus and Claudia aqueduct ruins mark the beginning of an arcade, up to 32 m high, that brought the aqueducts into Rome's city center (Fig. 1A), where they are now a protected United Nations Educational, Scientific and Cultural Organization (UNESCO) World Heritage Site. The waters of the Anio Novus were then mixed with those of the Claudia aqueduct at the Vigna Belardi cistern near Porta Maggiore in Rome. Maintenance of the Anio Novus aqueduct likely ceased sometime between the fifth and eighth centuries A.D. (Coates-Stephens, 1998, 2004; Van Deman, 1934; Van Deman et al., 1991).

Measurements of the thickness and distribution of travertine deposits at multiple locations along the entire length of the Anio Novus aqueduct channel have been published (Keenan-Jones et al., 2014; Motta et al., 2017). The experimental design of the present study was to collect and compare age-equivalent upstream and downstream samples of travertine deposited on the floor of a continuous 140-m-long section of the Anio Novus aqueduct channel at Roma Vecchia (Figs. 1B–1C). Sample collection was completed in March 2010 under research permits granted by the Soprintendenza Speciale per il Colosseo, il Museo Nazionale Romano e l'Area Archeologica di Roma. These included an upstream 0 m site (sample RNRV3-2A) and a downstream 140 m site (sample RNRV1-2A; Figs. 1B–1C). Travertine samples collected from both the 0 m and 140 m sites (Figs. 1B–1C) were composed of: (1) a 1-cm-thick uppermost portion of the underlying Roman mortar lining the floor of the aqueduct; (2) the time-zero (t₀) surface at the contact between the underlying mortar and the overlying travertine deposits; and (3) the 8-cm-thick layer of aqueduct travertine deposited immediately above the t₀ surface. The 140 m site contains a 27-cm-thick section of travertine (Fig. 2C) (Keenan-Jones et al., 2015). However, only the lowermost 8 cm was sampled and studied at the 140 m site (Figs. 2C–2F) because only the lowermost 8 cm of the travertine section was preserved at the 0 m site. Age-equivalency of the 8-cm-thick travertine samples at both the 0 m and 140 m sites was established via correlation to the t₀ surface (Fig. 3).

At the 0 m site (Figs. 1B–1C), only the underlying support arcade remained intact. Most original building materials comprising the Claudia aqueduct itself, as well as both the walls and the vaulted ceiling of the Anio Novus aqueduct, have been removed. Only a fallen, tilted block composed of the floor of the Anio Novus aqueduct was preserved, and the 0 m site aqueduct travertine sample was collected from the uppermost surface of this block (Figs. 1C). Conversely, the Anio Novus aqueduct infrastructure at the 140 m site (Figs. 1B–1C) was better preserved (Figs. 2A–2C). Here, the supporting arcades and Claudia aqueduct are intact, as are the floor and lower portions of the east wall of the overlying Anio Novus aqueduct channel (Figs. 2A–2C). The travertine deposits at both the 0 m and 140 m sites were partially covered with soil and fully exposed to the elements (Figs. 1B–1C and 2C–2D).

Aqueduct travertine samples from the 0 m and 140 m sites (Figs. 1B–1C and 2C–2D) were carefully removed using a hammer and a small, clean, well-sharpened chisel composed of hardened steel. Each travertine sample was labeled (sample number, upstream-downstream context, and flow direction determined by contextual orientation of the sample within the aqueduct channel), bagged, and shipped in a padded container to the Microscopy and Imaging Core Facility of the Carl R. Woese Institute for Genomic Biology (IGB) at the University of Illinois Urbana-Champaign, Illinois, USA. After X-ray analysis and orientation (described below), samples were cut on a clean, water-cooled, diamond-embedded tile saw in an orientation parallel to the upstream-downstream flow direction of the channel. Samples were then thoroughly washed with deionized water, polished, and photographed with a Nikon SLR D7000 digital camera.

Two types of X-ray imaging were conducted on the aqueduct travertine samples. The first type was microcomputed tomography (microCT) X-ray imaging completed on a North Star Imaging X5000 system (Feinfocus 225 kV) at 63 μm resolution at the University of Texas High-Resolution X-Ray Computed Tomography Facility, Austin, Texas, USA. The 3-D microCT data sets were analyzed using either Imaris (Bitplane, Zurich, Switzerland) or Avizo (FEI, Thermofisher, USA) 3-D rendering software for both visualization and segmentation in the Microscopy and Imaging Core Facility of the IGB. Several thousand microCT images were compressed and converted into a maximum intensity 3-D volume projection as virtual 5-mm-thick slices. These virtual cross-section images were used to determine where to cut 5-mm-thick, vertical slices of the travertine hand samples parallel to the downstream flow direction on a diamond-embedded tile saw for thin sectioning. The second type of X-ray imaging was completed on a standard X-radiography system (Siemens Model #10092624, 70 kV) at 150 μm resolution in the Veterinary School of Medicine at Illinois. Image compression, averaging, gray-scale corrections, and line profile analyses were conducted using open-source NIH ImageJ software. Averaged microCT and X-radiography images were converted to TIFF files at both 8-bit and 16-bit gray scales for line transect analyses. Line profile analyses were performed after the line widths were adjusted to 200 pixels to retrieve intensity peaks and valleys for location-wise comparison of depositional stratigraphic patterns.

Five billets cut from the travertine sample collected at the 0 m site and five billets cut from the travertine sample collected from the 140 m site (Figs. 1B–1C) were prepared by Wagner Petrographic (Linden, Utah, USA) as Petropoxy impregnated, doubly polished, uncovered, 25-μm-thick sections mounted on standard-sized petrographic glass slides. Optical microscopy of these thin sections was completed on multiple microscopes, all of which are housed in the Illinois IGB Microscopy and Imaging Core Facility and have been previously described in detail (Sivaguru et al., 2014a; Sivaguru et al., 2019a; Sivaguru et al., 2014b; Sivaguru et al., 2019b; Sivaguru et al., 2020; Sivaguru et al., 2012; Sivaguru et al., 2018a; Sivaguru et al., 2021b; Sivaguru et al., 2018b). In brief, modalities included bright-field (BF), polarization (POL), phase-contrast (PC), cathodoluminescence (CL), and wide-field and confocal auto-fluorescence (AF) with merged pseudo-colored red, green, and blue (RGB) channels using DAPI, FITC, and Rhodamine filters. Analyses were done across a broad range of magnifications (10×: 0.3 numerical aperture (NA); 20×: 0.8 NA; 63×: 1.4 NA; and 100×: 1.46 NA) with Plan Neofluar (10×), Plan Apochromat (20–63×), and Alpha Plan Apochromat (1.46 NA) objectives.

BF, POL, PC, wide-field, and confocal (LSM 700, Carl Zeiss, Oberkochen, Germany) AF microscopy was conducted at a resolution of 250 nm on a ZEISS Axio Observer Z1 system with 20× (0.8 NA) Plan Apochromat and 50× (0.95 NA) Plan Neofluar POL objectives for both brightfield (BF) and polarization (POL) microscopy. In addition, CL microscopy was conducted on a custom-built Reliotron cathodoluminoscope stage (RELION Industries LLC, Bedford, Massachusetts, USA) operating at 11 kV and 550 μA and mounted on a ZEISS AxioZoom.V16. The same system has also been used to collect BF, POL, and AF images from thin sections using either a DL 450 LED white light source base or an X-Cite metal halide mercury fluorescent lamp (for AF images) with a ZEISS Axiocam 512 color camera for imaging (Carl Zeiss, Oberkochen, Germany). All microscopy images were processed using ZEISS ZEN Blue software. Red, green, and blue (RGB) curves for each image were adjusted and presented as linear or with a gamma adjustment of 0.4–0.5, min/max, best mode or manually adjusted in the display properties window in the ZEISS ZEN software for representative brightness, contrast, and clarity. Images obtained from the Avizo, Imaris, and ZEN programs were compiled in Adobe Photoshop and Adobe Illustrator after further intensity and size adjustments as required (Adobe Inc., San Jose, California, USA). These instruments and accompanying image analysis software workstations are housed in the Illinois IGB Microscopy Core Image Analysis Facility.

BF, POL, and AF microscopy was also completed on a ZEISS Axio Scan.Z1 whole slide scanning system (with 20× [0.8 NA] Plan Apochromat and 50× [0.95 NA] Plan Neofluor POL objective) and a ZEISS Axio Zoom.V16 microscope (with 1.0× [NA 0.25] Plan Apochromat objectives; Carl Zeiss Company, Oberkochen, Germany). Thin sections were illuminated with a DL 450 LED light source base and imaged with a ZEISS Axiocam 512 color camera (Carl Zeiss, Oberkochen, Germany). Travertine AF from thin sections was then further analyzed using a ZEISS LSM 880 Confocal Laser Scanning Microscope with Airyscan SRAF (Carl Zeiss, Oberkochen, Germany) as described previously (Sivaguru et al., 2019a; Sivaguru et al., 2014b; Sivaguru et al., 2020; Sivaguru et al., 2012; Sivaguru et al., 2018a; Sivaguru et al., 2021b; Sivaguru et al., 2018b). Excitation and emission wavelengths that were collected included: 405 nm excitation (emission collected between 410 nm and 460 nm), 488 nm excitation (emission collected between 500 nm and 550 nm), and 561 nm excitation (emission collected between 570 nm and 615 nm).

Small 1 cm³ cubes were cut from the Anio Novus aqueduct travertine. These were washed, air dried, and attached to a sample holder stub (Z1506P, SPI Supplies West Chester, Pennsylvania, USA) using double-sided carbon tape (cat. no. Z05073, SPI Supplies). These samples were then sputter-coated with an ~60-nm-thick layer of gold–palladium (Desk II TSC sputter coater, Denton Vacuum, Moorestown, New Jersey, USA). Each sample was imaged using an environmental scanning electron microscope (ESEM) in high-vacuum mode (Quanta 450 FEG ESEM, Thermo Fisher FEI, Hillsboro, Oregon, USA) at 5–10 kV and multiple magnifications, housed in the Microscopy Suite of the Imaging Technology Group in the Illinois Beckman Institute for Advanced Science and Technology, Illinois, USA.

Standard reflected light photography, combined with high-resolution optical, laser, electron, and X-ray microscopy, indicates that the Anio Novus aqueduct travertine deposited at both the upstream 0 m and downstream 140 m sites (Figs. 1B–1C) is composed of two depositional horizons herein called Units 1 and 2 (Figs. 3–5). Unit 1 (the lowermost 5 cm section) and Unit 2 (the uppermost 3 cm section), which were age-correlated by means of the t₀ surface, are strikingly similar in color, crystalline texture, and stratigraphic layering at both the 0 m and 140 m sites (Figs. 3–6). Unit 1, deposited directly on the t₀ surface, is composed of high-frequency interlayering of 0.1–1-mm-thick, dark brown laminae and light beige laminae (Figs. 3–6). Laminae within the basal 1 cm of Unit 1 at the 0 m site are nearly planar and horizontal (Figs. 3–6). Conversely, at the 140 m site, the stratigraphic layering within Unit 1 exhibits an angular unconformity (Boggs, 2012) where underlying horizontal beds are eroded and overlain by a sequence of inclined laminae (Figs. 4B and 5B). Unit 1 stratigraphy above the 1-cm-thick basal interval exhibits up-section increasingly larger and more pronounced travertine crystal growth ripple morphologies (i.e., stoss, crest, lee, and trough; Figs. 3–6) (Keenan-Jones et al., 2022). On bedding surfaces observed in the field and hand samples (Figs. 4C–4D), as well as virtual 3-D micro-CT horizontal sections (Figs. 5C), these bedforms have been identified as linguoid and sinuous ripples (Boggs, 2012). As these ripple sets vertically accumulated, they shifted from downstream progradation to upstream retrogradation, which created zig-zag stratigraphic geometries in vertical cross-section (Figs. 3–6). A bedding surface observed within Unit 1 at a location 9 m downstream from the 0 m sample site (Figs. 1B–1C) also exhibits hummocky ripples (Fig. 4D).

In contrast, both the aqueduct travertine crystal growth ripples and laminae stratigraphy change dramatically in Unit 2 (Figs. 3–6). While the thicknesses of the dark brown laminae remain relatively consistent between Units 1 and 2 (hundreds of µm thick to ~1 mm thick), the light beige laminae become significantly thicker in Unit 2 (~1–3 µm thick; Figs. 3–6). Bedforms observed in hand sample (Fig. 2F) and virtual horizontal microCT cross sections (Fig. 5D) indicate that Unit 2 is composed of sinuous and some hummocky ripples (Boggs, 2012). Like the ripples in Unit 1, the aggrading crests of the ripple sets in Unit 2 exhibit progradation and retrogradation, which create similar zig-zag stratigraphic patterns in vertical cross-section (Figs. 3–6). However, these stratal patterns are less obvious in Unit 2 because there are fewer lag deposits leeward of the ripple crests due to the longer ripple wavelengths (Figs. 3–6).

A common depositional feature of the Unit 1 and 2 linguoid and sinuous ripple-marked travertine is the accumulation of siliciclastic grains on the lee slope of ripple crests, herein called lee sands (Fig. 7) (Keenan-Jones et al., 2022). These lag deposits contain an assortment of fine to coarse and angular to rounded siliciclastic sands, which were densely packed on the lee slope of each linguoid ripple. These lee sands are well-documented in 3-D microCT imaging of the aqueduct hand samples prior to sawing and thin section preparation (Figs. 7D–7F). However, the process of cutting and sectioning plucked and removed the majority of these siliciclastic sand grains (Figs. 7A–7C). The remaining void spaces exhibit periodic bridging by continuous layers of euhedral calcite crystal cements, in which some coated grains became encrusted and suspended (Figs. 7A–7C). These age-equivalent, calcite-cemented depositional layers can be continuously traced downstream across the ripple crest, through the lee sands, and into the ripple trough (Fig. 7C). Furthermore, the lee sand concentrations geometrically track the vertical aggradation of the stratigraphic ripple sets (Figs. 7D–7F), which serve to accentuate the zigzag stratal geometries formed during downstream progradation and upstream retrogradation (Figs. 3 and 7). Lee sands are also found on ripple lee slopes in Unit 2 (Figs. 3 and 7). However, there are fewer lee sands deposited in Unit 2 than in Unit 1 (Figs. 3 and 7), which correlates with longer ripple wavelengths and results in the formation of fewer ripple crests and therefore fewer lee sand lag deposits.

The term "spar" has not been used in this study, because the term is often broadly linked with interpretations of specific environments of crystallization (Folk, 1959, 1962). Instead, the term "crystal" will be used in its place as a purely descriptive term (Dunham, 1962; Tucker and Wright, 1990). The Unit 1 and 2 aqueduct travertine deposits are composed of two types of crystalline growth morphologies, which are described here as “shrubs” (Fouke et al., 2000) and “radiaxial calcites” (Kendall, 1977; Kendall and Broughton, 1977; Tucker and Kendall, 1973). Both types of these crystalline deposits contain stratigraphic sequences of interlayered dark brown and light beige laminae (Figs. 3–7). The travertine shrubs are composed of 100-μm-tall, dendritically branching aggregates of small (1–3-μm-diameter) euhedral calcite crystals (Figs. 8A–8C). The travertine radiaxial calcites are composed of variably sized (hundreds of microns to tens of millimeters), upward-radiating and branching crystalline bundles that crosscut the alternating dark–light laminae stratigraphy (Figs. 3–7). The radiaxial calcites are composed of elongated crystals with irregular intercrystalline boundaries, subcrystals that diverge away from the substrate, a convergent length-fast structure, and twin lamellae (Figs. 3–7). These growth forms are consistent with radiaxial calcites previously reported from multiple types of natural limestones throughout the geological record (Kendall, 1985; Kendall and Broughton, 1978; Kendall and Dunn, 1977; Tucker and Kendall, 1973). Vertical and lateral contacts between the travertine shrubs and radiaxial calcite are irregular (Figs. 8D–8H) and composed of individual travertine shrub calcite crystals that are partially engulfed and entombed within large clear radiaxial calcite crystals (Figs. 8D–8H). These dark brown laminae extend undisturbed laterally through the travertine shrubs and radiaxial calcites (Figs. 8D–8H). Under BF microscopy, these dark brown laminae form thicker and more seemingly diffuse tens-to-hundreds-of-μm-thick, dark brown laminae within the travertine shrubs (Figs. 8D–8G). The laminae then laterally transition into multiple significantly finer (1–3-μm-thick) and more distinct brown laminae within the radiaxial calcite crystals (Figs. 8D–8G). These dark brown laminae in both the travertine shrubs and radiaxial calcites exhibit bright green AF and bright orange CL emissions, which are significantly brighter within the radiaxial calcites (Figs. 9 and 10).

High-resolution optical, laser, electron, and X-ray microscopy analyses were conducted in the present study to characterize the crystalline structure, bedform geometries, and laminae stratigraphy of the travertine deposited within the Anio Novus aqueduct. From these analyses, the depositional histories of the aqueduct travertine crystalline shrubs and their laminae are discussed within the context of the hydraulic setting of the Anio Novus, modern-day aqueous chemistry of the Aniene River source water, and comparisons are made to other analogous ancient aqueduct and natural systems. The diagenetic history of the Anio Novus aqueduct travertine is then evaluated using the same high-resolution microscopy analyses, which suggest that the abundant radiaxial calcite crystals are diagenetic replacement products of the original travertine. This historical framework of deposition and diagenesis is then used to evaluate the potential for future reconstructions of human history and paleoclimate from travertine deposits preserved within water conveyance and storage infrastructure built by ancient civilizations around the world.

While the absolute age of the 8-cm-thick travertine deposited on the t₀ channel floor surface at the 0 m and 140 m sites at Roma Vecchia (Figs. 1B–1C) remains uncertain, circumstantial evidence implies that the travertine may have been deposited relatively soon after construction of the Anio Novus aqueduct. Because the t₀ surface cannot be older than the A.D. 38–52 construction period of the Anio Novus itself (Ashby, 1935), possible depositional ages include: (1) the entire Roman Imperial period (A.D. 30–476) (Stevenson, 2009), when the Anio Novus was actively flowing, used, and maintained (Coates-Stephens, 2003a, 2003b), and (2) much later into early and middle medieval times (ca. A.D. 477–1000) (Stevenson, 2009), when the Anio Novus still functioned to transport water but was no longer maintained (Coates-Stephens, 2003a, 2003b). However, there is no evidence of rebuilding of the channel or its mortar lining after the Anio Novus aqueduct's initial construction at Roma Vecchia (Ashby, 1935). Furthermore, common Roman cleaning techniques included physical chiseling, scraping, and perhaps dissolution from the application of heated wine vinegar (Bruun, 1991; Fahlbusch, 1991; Grewe and Blackman, 2001; Leveau, 1991; Taylor, 2000). As a result, the lack of gouging or scraping along the t₀ surface implies that the Anio Novus had not been removed during maintenance prior to deposition of the basal, 8-cm-thick travertine deposits at the 0 m and 140 m sites. There is also no evidence that a waterproof mortar lining was placed over the encrusted aqueduct travertine, which has been commonly documented in other ancient Roman aqueducts (Passchier and Surmelihindi, 2019; Passchier et al., 2021; Porath, 2002; Sürmelihindi, 2013). This extra thick mortar was routinely added to further protect the vulnerable corner contacts of the floor and wall (Hodge, 2002b). Frontinus also reported that this additional aqueduct masonry served as structural reinforcement to support the immense weight of the travertine precipitated within some aqueducts. However, the absence of these cleaning and maintenance features, combined with the perfectly smooth and planar nature of the t₀ surface, implies that the lowermost 8 cm of Anio Novus travertine was deposited relatively soon after construction and not subjected to maintenance cleaning prior to deposition of the travertine analyzed in the present study.

The water temperature and chemistry of the upper Aniene River watershed, which is where the Subiaco water intake of the Anio Novus aqueduct was located, serves as a valuable modern analog for the ancient flowing water that was transported through the Anio Novus aqueduct (Bono and Boni, 1996; Bono et al., 2001; Bono and Percopo, 1996). The Aniene River basin drains the karsted Mesozoic carbonate lithologies of the Simbruini Mountains in the Lazio Region of the Appenines (Bono and Percopo, 1996). A gauging station at Subiaco has long provided measurements of modern-day temperature, flow discharge, and suspended solid load of the Aniene River under both normal and storm atmospheric conditions (Bono and Percopo, 1996). As a result, ancient Aniene River water, supersaturated with respect to calcite, is thought to have entered the Anio Novus intake at temperatures of 10–15 °C and a pH of 6.7–7.2, with aqueous chemical compositions typical of karsted limestone terrains (Bono and Boni, 1996). Furthermore, calcite precipitation rates, aqueous chemistry, and flow conditions have been quantified for the Tartare karst spring near Subiaco (Bono et al., 2001). These modern-day Aniene River watershed aqueous analyses, combined with ancient conveyance rates reconstructed from the thickness of travertine deposited on the floor, walls, and ceiling of the Anio Novus (Keenan-Jones et al., 2015; Motta et al., 2017), suggest that the ancient flowing Anio Novus aqueduct water at Roma Vecchia remained supersaturated with respect to calcite at 10–15 °C and a pH of 6.7–7.2 by the time it reached Roma Vecchia (Bono and Boni, 1996).

When the Anio Novus was in operation, travertine accumulations on the floor, walls, and ceiling of the aqueduct channel at Roma Vecchia would have reduced the cross-sectional area of the aqueduct channel (Keenan-Jones et al., 2014; Motta et al., 2017) Aqueduct water flow velocity under these restricted conditions is estimated to have varied by more than 1 m/s from place to place along the 80-km-length of the Anio Novus due to significant changes in the amount of travertine deposition and hillslope topography (Keenan-Jones et al., 2015; Motta et al., 2017). Average flow velocity estimates for the entire Anio Novus range from 0.7 m/s to 0.9 m/s (Bono and Boni, 1996; Fahlbusch, 1987). Additionally, it has recently been estimated that full gravity flow of the Anio Novus aqueduct was 1.2 ± 0.4 m³/s, which has been reconstructed from the wetted perimeter of the outermost surface of the travertine at Roma Vecchia (Keenan-Jones et al., 2014, 2022).

Hydraulic reconstructions have also been made from quantitative analyses of the wavelength, amplitude, and steepness of the Anio Novus travertine linguoid, sinuous, and hummocky crystal growth ripples at Roma Vecchia (Kennan-Jones et al., 2022). Taken together, these measurements suggest that the critical shear Reynolds number of channel flow at the time of travertine deposition was equivalent to being much higher than those calculated from bedforms measured in caves, ice, and fluvial and marine siliciclastic systems (Keenan-Jones et al., 2015, 2022; Motta et al., 2017). In addition, the high-velocity confined channel flow of the Anio Novus at Roma Vecchia occurred at virtually constant kinematic viscosity, where ripple wavelength decreased with increased shear velocity. These extremely high flow rates would have decreased the thickness of the boundary layer of the base of the flowing water on the upper surface of the growing aqueduct travertine. In addition, surface scouring (sand blasting) would have taken place on the upper surface of the travertine by the lee sands that were deposited and preserved on the lee side of travertine crystal growth ripples (Fig. 7) (Keenan-Jones et al., 2022).

The combined effects of these hydraulic factors would have acted to prevent dendritically branching crystalline structures, such as the upward radiating bundles of radiaxial calcites observed in hand samples and with standard X-radiography and microCT (Figs. 3–6), from growing on the channel floor. This is further indicated by the dense mounded and rippled surfaces seen in hand sample and outcrop, none of which exhibits the terminations of large crystalline shrubs or other crystal growth faces (Figs. 2F, 3C–3D, and 4C–4D). This high-velocity flow regime may also explain the extremely small size of the aqueduct's travertine crystalline shrubs (Fig. 7) relative to the size of crystalline shrubs observed in analogous lower velocity natural systems (Della Porta, 2015). Importantly, these hydraulic constraints would also have held true for other ancient aqueducts throughout the Roman Empire in France, Turkey, and Jordan, in which travertine crystal growth ripples are well-preserved (Passchier and Surmelihindi, 2019; Passchier et al., 2021; Passchier et al., 2016a; Passchier et al., 2016b; Passchier et al., 2013; Sürmelihindi, 2018; Sürmelihindi and Passchier, 2013; Sürmelihindi et al., 2013a; Sürmelihindi et al., 2019; Sürmelihindi et al., 2013b).

Correlation of the t₀ surface and strong similarities in travertine color, depositional texture, and stratigraphic layering patterns (Figs. 3–6) suggest that Units 1 and 2 deposited at the 0 m and 140 m sample sites are equivalent in depositional age. Outcrop bedding surfaces, dark–light laminae stratigraphy, and microCT imaging in the present study (Figs. 3–6) indicate that Unit 1 is composed of aggrading linguoid and sinuous travertine crystal growth ripples, which exhibit multiple episodes of downstream progradation and upstream retrogradation as they vertically aggraded. In addition, the striking angular unconformity observed near the bottom of Unit 1 in the 140 m sample (Fig. 4B) indicates that some type of erosion took place, either due to changing flow conditions or possibly Roman maintenance. In contrast, Unit 2 is composed of sinuous ripples that aggraded vertically and exhibit lesser degrees of downstream progradation and upstream retrogradation as well as hummocky ripples.

The Unit 1 travertine linguoid and sinuous crystal growth ripples also exhibit lee sand lag deposits composed of poorly sorted siliciclastic grains (Fig. 7). These lee sands, transported and deposited during the ongoing high-flow conditions of the Anio Novus waters, would have been rotated and rolled by Kármán spiral eddies (vortices) leeward of each linguoid ripple crest (Fig. 6). At the same time, small calcite cements encrusted the outer surfaces of these siliciclastic sands to form coated grains (Bathurst, 1975; Mcillreath and Morrow, 1990; Tucker and Bathurst, 1990; Tucker and Wright, 1990). Depositional laminae laterally extended downstream from the linguoid ripple stoss slopes, through the ripple crests, lee slopes, and troughs (Figs. 7 and 9). Ashby (Ashby, 1935) and Blanco and Sebastiani del Grande (Blanco and Sebastiani del Grande, 2016) were likely referring to these coated grains when describing “calcareous pebbles, completely round” that filled the settling tank in the Anio Novus at Villa Bertone (700 m upstream of Roma Vecchia). It is also possible that the high-velocity aqueduct waters were capable of transporting some of these siliciclastic sands from the Subiaco source area (Fig. 1). Conversely, although to our knowledge not previously suggested in the literature, it may also be possible (but unproven here) that these siliciclastic grains were periodically added by the Romans at access points upstream of the 0 m and 140 m sites in an attempt to “sand blast” and clear travertine deposits from within the channel of the Anio Novus.

The high-resolution microscopy analyses conducted in the present study indicate that the Anio Novus aqueduct travertine was originally composed of crystalline shrubs, which were made of dendritically branching aggregates of small (1–3-μm-diameter), euhedral calcite crystals (Figs. 6, 8, and 9). These syn-depositional shrubs grew with dark brown laminae that formed diffuse layers tens to hundreds of microns thick (Figs. 6, 8, and 9). The strong AF emissions of these laminae indicate that they contain high concentrations of organic matter (Figs. 7D–7G and 9). The bright AF of biofilm laminae in polished travertine hand samples that had not been impregnated with epoxy indicates that these bright AF emissions are from the travertine itself and not the epoxy (Fig. 7G). Furthermore, CL petrography has been used to characterize the types of individual events of both original carbonate sedimentation and secondary diagenetic alteration observed within the Anio Novus aqueduct travertine (Barker and Kopp, 1991; Fouke et al., 2005; Meyers and Lohmann, 1985). The diffuse, dark brown biofilm laminae within the travertine shrubs generally exhibit a dull to bright orange CL (Figs. 9C, 9F, and 10). CL emissions in these types of carbonates are primarily caused by Mn²⁺ and some trace elements, which have a high distribution coefficient (K_D) and are mobilized under reducing aqueous conditions (Barker and Kopp, 1991). Therefore, if the biofilm laminae CL represents an original depositional signal and not diagenetic alteration, this may indicate periods of low-oxygen (dysoxia) within the flowing aqueduct waters where Mn²⁺ and rare earth elements were mobilized and concentrated due to high water-to-travertine ratios during deposition (Banner and Hanson, 1990a; Fouke et al., 1996; Fouke et al., 2005). Similar original travertine shrubs have been observed in travertine deposited within cisterns and around lead pipes of the Baths of Caracalla in Rome and the Castellum Aquae at Porta Vesvuii in Pompeii, which exhibit bright AF and CL in otherwise minimally altered travertine (Hostetter et al., 2011).

The original crystalline shrubs comprising the Anio Novus aqueduct travertine are also comparable to travertine crystalline shrubs that grow within modern-day natural terrestrial springs, rivers, and caves around the world (Amundson and Kelly, 1987; Capezzuoli et al., 2014; Chafetz et al., 1991; Dong et al., 2019; Dreybrodt et al., 1992; Fouke, 2001, 2011; Fouke et al., 2000; Hauck et al., 2012; Pentecost, 2005; Sanders and Friedman, 1967). The primary difference is the small 1–3 μm size of the euhedral calcites and their dendritically branching aggregates within the Anio Novus aqueduct travertine at Roma Vecchia, which continue to grow even during biofilm laminae deposition (Fig. 7). Previous studies of travertine crystalline shrubs within natural settings have described the biofilm laminae as “micrite layers” (Passchier et al., 2021). However, SRAF 140-nm-resolution microscopy illustrates that the crystalline, dendritically branching travertine shrubs are syn-depositional with the biofilm laminae (Fig. 7). Therefore, the dark biofilm laminae are not simply micrite layers composed of randomly organized, ≤3-μm-diameter calcite crystals (Bathurst, 1975). The primary factors controlling precipitation of these types of travertine crystalline shrub fabrics in natural settings include (Ford and Pedley, 1996; Fouke, 2001, 2011; Fouke et al., 2000; Pentecost, 1995a, 2005): (1) water chemistry (e.g., major and minor elemental abundance, isotopes, pH, HCO₃, pCO₂, and resultant saturation state); (2) physical processes (e.g., degassing, temperature change, steaming, boiling, dilution, and evaporation); (3) hydrology (e.g., flow rates, flux, and surface area); and (4) biotic activity (e.g., microbial photosynthesis, respiration, and biochemical effects).

The distinctive high-frequency stratigraphy of alternating dark brown and light beige laminae, a hallmark of aqueduct travertine, was previously used to estimate aqueduct water chemistry, flow conditions, and maintenance (Aicher, 1995; Bobée et al., 2011; Brinker, 1986; Carlut, 2011; Carlut et al., 2009; Carrara and Persia, 2001; Coates-Stephens, 2003a, 2003c; Dubar, 2006a, 2006b; Garbrecht and Manderscheid, 1992; Gilly et al., 1971; Gilly, 1971; Hodge, 1992; Lombardi, 2002; Schulz, 1986; Wilson, 2004). These studies assume that the individual dark–light laminae couplets represent annual depositional events and therefore attempt to use these stratigraphic sequences to date the aqueduct travertine in a manner analogous to that of tree ring, speleothem lamination, and lake varve chronology (Baker et al., 2008; Faraji et al., 2021; Haneca et al., 2009; Ojala et al., 2012; Zolitschka et al., 2015). This approach requires that the dark brown laminae represent times when organic matter (plant material, soil particles, tannins, humic acids, other biomolecules, and Mn²⁺ adsorbed onto organic matter or as extremely fine MnO particles) was washed into the Anio Novus during periods of seasonal storm events (Barceloux, 1999; Lester and Birkett, 1999; Passchier et al., 2021). If correct, this would be consistent with the bright green AF and bright orange CL observed in most of the dark laminae within the Anio Novus travertine (Figs. 10 and 11). Further supporting evidence comes from Frontinus, who described periods of increased soil turbidity in the Anio Novus during storms despite the use of settling tanks at the aqueduct intake. This is also seen in modern-day analyses of the Aniene River water at Subiaco, where increased discharge during storms correlates with an increase in suspended solids and turbidity (Bono and Percopo, 1996).

However, this assumption that the dark biofilm laminae represent storm events is not without its uncertainties. Studies of travertine deposition in natural systems such as caves, lakes, springs, and marine settings around the world have documented multiple environmental conditions that can form dark, organic matter-rich laminae (Baker et al., 2008; Dabkowski et al., 2016; Spear and Corsetti, 2013). Dark biofilm laminae in speleothems are often yellow to brown in color and exhibit bright green AF, which has been interpreted as due to rainfall events that washed in soil organic matter rich in humic and fulvic acids (Baker et al., 2008; Gascoyne, 1978; White, 1981). However, for high-frequency speleothem laminae, as well as varved lake sediments, the stratigraphic positions of absolute age dates for seasonal climatic change do not consistently or precisely align with the dark–light laminae stratigraphy (Baker et al., 2008; Faraji et al., 2021; Haneca et al., 2009; Ojala et al., 2012; Zolitschka et al., 2015). Therefore, other physical, chemical, and biological processes may also influence the formation of the dark biofilm laminae, which indicates that multiple lines of evidence need to be assembled when attempting to use aqueduct travertine biofilm laminae stratigraphy for chronological correlations and paleoclimatic reconstructions. Furthermore, the majority of the dark biofilm laminae preserved within the Anio Novus travertine are finer (Figs. 5–7) than the 1 μm average diameter of microbial cells (Fouke, 2011). Therefore, by analogy, these extremely thin, dark biofilm laminae would contain cell debris and biomolecules, but not complete cells, as is observed in the biofilm laminae of calcium oxalate human kidney stones (Sivaguru et al., 2020; Sivaguru et al., 2018a). As a result, the mechanisms responsible for forming the dark biofilm laminae in the Anio Novus aqueduct travertine, and other similar types of carbonate deposits, remain to be systematically tested and determined.

The potential role of microbial influence on Anio Novus aqueduct travertine deposition can also be inferred via comparison with crystalline calcite shrubs that precipitate at temperatures of less than 25 °C and a pH of 8 in the Distal Slope Facies at Mammoth Hot Springs in Yellowstone National Park (Figs. 12A–12B) (Fouke, 2011). Here, the dendritically branching calcite crystalline shrubs are fully covered with microbial biofilms composed of cells and biomolecules that include extracellular polymeric substances (EPS; Figs. 12A–12B). Culture-independent 16S rRNA gene sequence surveys indicate that these biofilms are composed of a moderately diverse microbial assemblage. However, despite the fact that these hot springs are not impacted by seasonal variations in rainfall (Fouke, 2011), travertine deposited in the Distal Slope Facies still forms dark and light biofilm laminae (Fouke, 2011). In addition, field experimentation has demonstrated that microbial cells, biomolecules, and EPS (Figs. 12A–12B) cause dramatic increases in travertine precipitation rate via protein catalysis (Fouke, 2011; Kandianis et al., 2008). The abundance of entombed, dark brown, organic matter-rich biofilm laminae within the Anio Novus travertine implies that this type of microbial biofilm influence may also occur during formation of the rippled travertine surfaces within the Anio Novus aqueduct.

The radiaxial calcites observed within the Anio Novus aqueduct travertine are similar to crystals observed in marine limestone and cave speleothems (Kendall, 1985; Kendall and Broughton, 1978; Kendall and Tucker, 1971, 1973). However, the complex growth features of these radiaxial calcite crystals in marine and cave deposits have led to controversial and often contradictory interpretations of their formation. As a result, it is likely that radiaxial calcites can form as original depositional features, as well as be the product of diagenetic alteration, depending on the specific environmental conditions of deposition and diagenesis (Kendall, 1977; Kendall, 1985; Kendall and Broughton, 1978; Kendall et al., 1985; Tucker and Kendall, 1973). Furthermore, the Anio Novus aqueduct's travertine radiaxial calcite crystals are morphologically distinct from dendritic feather calcite crystals formed in terrestrial springs (Jones and Renaut, 2009; Turner and Jones, 2005) as well as from fracture-filling calcite cements precipitated under high temperature and pressure (Passchier and Trouw, 1998). In the present study, the hydraulic setting of the Anio Novus aqueduct is combined with high-resolution microscopy of the travertine to suggest that the radiaxial calcite crystals in this setting formed as a result of diagenetic replacement of the original travertine crystalline shrubs and dark biofilm laminae (Bathurst, 1975; Mcillreath and Morrow, 1990; Tucker and Bathurst, 1990; Tucker and Wright, 1990). These multiple lines of evidence are summarized and evaluated in the following discussion.

An important contextual consideration is that the large, upward-branching radiaxial calcite crystals within the travertine (Figs. 3–5 and 9) are inconsistent with their original precipitation within the high-velocity hydraulic environment of the Anio Novus aqueduct. The evolution of the aqueous boundary layer thickness along the wetted perimeter of the aqueduct would have induced shear forces onto the growing upper surface of the travertine (Niño et al., 2003). Combined with erosion during transport of the lee sands (Fig. 7), these hydraulic factors would have prevented the growth of upward-branching radiaxial calcite crystals on the uppermost surfaces of the ripple-marked travertine. Furthermore, the upward-growing, radiaxial calcite crystals are symmetrical, branching in both upstream and downstream directions (Fig. 11). As has been documented in the travertine Proximal Slope Facies of hot-spring drainage channels, travertine shrubs that extend above the overlying boundary layer grow into and toward the direction of flow from which dissolved ions are being delivered (Fouke, 2011). Therefore, the symmetrically branching growth structures of the radiaxial calcite are inconsistent with their growth in the high-velocity, unidirectional flow of the aqueduct water.

Another factor is that as the upward-branching radiaxial calcites aggrade within the travertine stratigraphic section, they track upstream to downstream shifts in the position of the ripple crests (Figs. 3–6). The hydraulic shear force of the flowing aqueduct water would have been strongest at the ripple crests and therefore could even more effectively prevent the growth of branching radiaxial crystals (Raudkivi, 1963). As a result, this suggests that the concentration of radiaxial calcites within the ripple crest travertine reflects preferential diagenetic alteration of the ripple crest travertine shrubs rather than original growth in this hydraulically vulnerable high-shear position (Figs. 3–6). The Anio Novus aqueduct's travertine radiaxial calcites also consistently exhibit upward-radiating, branching crystalline growth structures within the travertine deposits, which are relatively symmetrical in both the downstream and upstream directions (Fig. 11). Each upward-radiating syntaxial branch of radiaxial calcite crystals is itself discontinuous, being irregularly separated by partial to complete intervals of crystalline travertine shrubs. This discontinuous, syntaxial nature of individual radiaxial calcite branches is consistent with a diagenetic origin, whereas pauses in primary crystallization would not result in C-axis continuity after intervals of crystalline shrub growth.

Dark biofilm laminae (Figs. 11 and 12) are thicker and more diffuse within the travertine shrubs, while age-equivalent laminae within the RC are finer and more sharply defined. When considered within the stratigraphic context of each radiaxial calcite crystalline branch, the older, dark biofilm laminae at the bottom are smooth and flattened (Fig. 12). These dark biofilm laminae shapes within the radiaxial calcite are consistent with travertine shrub surfaces observed in cross-section and on ripple-marked bedding planes (Figs. 2F and 4D). Conversely, the uppermost younger terminations of branches of radiaxial calcites exhibit either: (1) euhedral, rhombohedra-shaped crystal growth terminations of tens to hundreds of microns in scale; or (2) irregular, porous, and jagged anhedral crystalline textures (Fig. 12). Therefore, neither of these crystal termination shapes and textures are consistent with the smooth and rounded dark biofilm laminae within the radiaxial calcite crystals (Figs. 11 and 12). If the radiaxial calcites were syn-depositional with the travertine shrubs, which are in situ crystalline growths and not the result of micritic sedimentation or cementation (Fig. 8), the dark brown biofilm laminae would have grown over and preserved the rhombohedral-shaped geometry and/or porous textures of the radiaxial calcite crystal terminations (Figs. 8 and 9). This strongly suggests that the biofilm laminae formed on the growing upper surface of small, rounded travertine shrubs (Figs. 2F and 4D) that subsequently underwent fabric-preserving (mimetic) diagenetic replacement (Figs. 8 and 9). This interpretation is consistent with previous observations within speleothem stalagmites, where dark biofilm laminae preserve rhombohedra-shaped crystal terminations within original, unaltered radiaxial columnar calcites (Kendall and Broughton, 1978; Frisia, 2015).

A diagenetic replacement origin for the radiaxial calcites within the aqueduct travertine is further suggested by the irregular lateral gradational contacts observed between crystalline shrubs and radiaxial calcites (Figs. 8D–8F). If the radiaxial calcites were syn-depositional, the travertine shrubs and their small, dendritically branching euhedral calcites would exhibit abrupt contacts abutting the lateral faces of the radiaxial calcite crystals. Instead, the individual 1–3-μm-diameter euhedral calcite crystals comprising the travertine shrubs exhibit a gradational lateral contact with some calcite crystals, which further implies a replacement process (Figs. 8D–8F). In addition, age-equivalent, individual dark biofilm laminae that are diffuse within the travertine shrubs become finer, more sharply defined, bright AF laminae within the large, clear radiaxial crystals (Figs. 11 and 12). The bright orange CL character of the dark biofilm laminae within the radiaxial crystals (Figs. 9C, 9F, and 10) may further imply that they are mimetically preserved remnants from the original travertine shrubs. In addition, the irregular and highly porous textures of many of the radiaxial crystals (Fig. 12) are consistent with the dissolution and diagenetic alteration commonly observed in carbonate deposits (Bathurst, 1975; Mcillreath and Morrow, 1990; Tucker and Bathurst, 1990; Tucker and Wright, 1990).

Comparison of 0-year-old and 100-year-old Proximal Slope Facies travertine deposits at Mammoth Hot Springs (Fig. 13; Fouke, 2011) reveals diagenetic radiaxial calcite replacement alteration fabrics similar to those observed in the Anio Novus aqueduct travertine. The Proximal Slope Facies travertine was deposited at higher temperatures than the Anio Novus aqueduct travertine and is composed of aragonite rather than calcite (Fouke, 2011). However, diagenetic alteration of the Proximal Slope Facies provides important insight into how crystalline travertine shrubs can be recrystallized into radiaxial calcites during meteoric diagenesis. The Proximal Slope Facies travertine at Mammoth Hot Springs is composed of 1–30-μm-long aragonite needles organized into shrub-like, dendritically branching, crystal aggregate growth structures on the spring outflow drainage channel floor (Figs. 13C and 13D; Fouke, 2001, 2011; Fouke et al., 2000). The 0-year-old, newly deposited travertine shrubs exhibit no CL emissions (Figs. 13C and 13D). Conversely, the original, non-CL travertine shrubs are diagenetically replaced by mauve to bright orange CL that radiates upward and branching, radiaxial calcite crystals (Figs. 13E and 13F). These CL emissions from radiaxial calcites within travertine preserved in a non-burial meteoric environment at Mammoth Hot Springs, a diagenetic environment analogous to that of the Anio Novus aqueduct travertine, suggest that this diagenetic replacement alteration was caused by high water-to-travertine reaction ratios with dysoxic diagenetic freshwaters in which Mn²⁺ and other trace elements were mobilized (Banner and Hanson, 1990b; Barker and Kopp, 1991; Brand and Veizer, 1980, 1981; Fouke et al., 1996; Fouke et al., 2005; Richter et al., 2003).

Several previous studies have attempted to reconstruct the timing, velocity, chemistry, and source area of the ancient aqueduct's flowing waters by analyzing the dark–light laminae stratigraphy that is the hallmark of Roman aqueduct travertine (Aicher, 1995; Brinker, 1986; Carlut, 2011; Carlut et al., 2009; Carrara and Persia, 2001; Coates-Stephens, 2003a, 2003b, 2003c; Dubar, 2006a, 2006b; Garbrecht and Manderscheid, 1992; Gilly et al., 1971; Gilly, 1971; Hodge, 1992; Lombardi, 2002; Schulz, 1986). Collectively, these studies have proposed that the dark–light laminae couplets reflect a variety of processes, including: (1) changes in paleoclimate, seasonal temperature, and solar cycles; (2) variations in water velocity and turbulence, mixing of waters, water chemistry, pressure, and depth; (3) chemical reaction with the surrounding aqueduct mortar and building stones; and (4) periods of disuse and other archaeological events. However, as described previously, uncertainties surrounding the determination of the underlying mechanisms that control formation of the dark laminae (Baker et al., 2008; Faraji et al., 2021; Haneca et al., 2009; Ojala et al., 2012; Zolitschka et al., 2015) call into question the accuracy of many of these interpretations.

Recently, a series of benchmark studies presented standard petrographic and high-resolution chemostratigraphic analyses of Roman aqueduct travertine deposits in France, Turkey, and Jordan (Passchier and Surmelihindi, 2019; Passchier et al., 2021; Passchier et al., 2016a; Passchier et al., 2016b; Passchier et al., 2013; Sürmelihindi, 2013; Sürmelihindi, 2018; Sürmelihindi et al., 2018; Sürmelihindi and Passchier, 2013; Sürmelihindi et al., 2013a; Sürmelihindi et al., 2019; Sürmelihindi et al., 2013b). As is the case for the Anio Novus aqueduct, many of the walls and ceilings in these other aqueducts distributed throughout the ancient Roman Empire are in ruins or completely absent. This has exposed the aqueduct travertine deposits to millennia of freshwater percolation that may have resulted in diagenetic alteration of their original crystalline fabrics and geochemical composition (Morse and McKenzie, 1990). Field and hand sample photography and standard petrographic analyses (BF and POL) of 30–35-μm-thick sections were used in these studies to suggest that the original depositional fabric of the aqueduct travertine was composed of: (1) dark laminae of fine-grained micrite, which appear opaque at the 3–5 μm resolution of the petrographic microscopes that were used and (2) light laminae of dense, coarse-grained, and transparent “sparite” calcite crystals elongated along their c-axes (Passchier and Surmelihindi, 2019; Passchier et al., 2021; Passchier et al., 2016a; Passchier et al., 2016b; Passchier et al., 2020; Passchier et al., 2013; Sürmelihindi, 2013; Sürmelihindi, 2018; Sürmelihindi et al., 2018; Sürmelihindi and Passchier, 2013; Sürmelihindi et al., 2013a; Sürmelihindi et al., 2019; Sürmelihindi et al., 2013b).

Although uncertain, the “dark micrite laminae” described in these previous studies may be similar to the travertine shrubs observed in the Anio Novus travertine, which are composed of micrite-sized, 1–3-μm-diameter, dendritically branching calcite crystals (Figs. 7A–7C). These fine crystalline shrub fabrics would appear as dark, opaque micrite when analyzed under BF and POL on standard petrographic microscopes within 30–35-μm-thick petrographic thin sections. In addition, the light-colored “sparite” laminae observed in other aqueduct travertine may be comparable to the radiaxial calcite mimetic replacement crystals observed in the Anio Novus aqueduct travertine (Figs. 3–5 and 9). Previous descriptions (e.g., Passchier et al., 2021) include: (1) heterogeneous crystal distributions along individual laminae, which are comparable to radiaxial calcite distributions observed in the present study (Fig. 8C); and (2) large crystals that cross-cut the dark–light laminae stratigraphy and commonly radiate vertically to connect through the entire thickness of the deposit, which are comparable to radiaxial calcite distributions observed in the Anio Novus aqueduct travertine (Fig. 9).

These previous studies of Roman aqueduct travertine distributed throughout the Roman Empire also presented high-resolution δ¹⁸O, δ¹³C, and trace element analyses conducted within “sparite” calcite crystals (Claes et al., 2017a; Claes et al., 2017b; Claes et al., 2015; Passchier and Surmelihindi, 2019; Passchier et al., 2021; Passchier et al., 2016a; Passchier et al., 2016b; Passchier et al., 2020; Passchier and Sürmelihindi, 2010; Passchier et al., 2013; Sürmelihindi, 2013; Sürmelihindi, 2018; Sürmelihindi and Passchier, 2013; Sürmelihindi et al., 2013a; Sürmelihindi et al., 2019; Sürmelihindi et al., 2013b). Chemostratigraphic data in these studies exhibit consistent trends in δ¹⁸O, δ¹³C, and trace element composition, which strongly suggest they record original seasonal paleoclimatic trends in water temperature, flow rate, rainfall, and aquifer recharge at the time of aqueduct travertine deposition (Passchier and Surmelihindi, 2019; Passchier et al., 2021; Passchier et al., 2016a; Passchier et al., 2016b; Passchier et al., 2020; Passchier et al., 2013; Sürmelihindi, 2013; Sürmelihindi, 2018; Sürmelihindi and Passchier, 2013; Sürmelihindi et al., 2013a; Sürmelihindi et al., 2019; Sürmelihindi et al., 2013b). Combined with standard petrography, this comprehensive data set was used to reconstruct a history of maintenance, floods, droughts, and earthquakes (Passchier et al., 2021). The dark laminae preserved within their radiaxial “sparite” calcite crystals (Passchier et al., 2021) may be similar to the crystalline replacement fabrics observed in the Anio Novus radiaxial calcites (Figs. 8 and 9). Preservation of original chronostratigraphic trends in mimetic replacement calcites has been identified in marine carbonates (Fouke et al., 1996; 2005) and would be consistent with the mimetic replacement radiaxial calcites observed within the Anio Novus aqueduct travertine.

Integration of the field, hand sample, and high-resolution microscopy analyses conducted for the present study, combined with field, petrographic, and chemostratigraphic approaches established in previous work (Bobée et al., 2011; Hostetter et al., 2011; Passchier and Surmelihindi, 2019; Passchier et al., 2021; Passchier et al., 2016a; Passchier et al., 2016b; Passchier et al., 2020; Passchier et al., 2013; Sürmelihindi, 2013; Sürmelihindi, 2018; Sürmelihindi et al., 2018; Sürmelihindi and Passchier, 2013; Sürmelihindi et al., 2013a; Sürmelihindi et al., 2019; Sürmelihindi et al., 2013b), will permit future studies to comprehensively and routinely evaluate the potential impact of diagenetic alteration on reconstructions of paleoclimate and archaeological information from travertine deposited in ancient water transport and storage systems around the world. This systematic approach would include: (1) establishment of a Powers of Ten 3-D contextualization for field, hand sample, and thin section analyses; (2) optical, laser, electron, and X-ray microscopy analyses; (3) quantitative measurement and correlation of the dark–light laminae stratigraphic layering sequences; (4) expansion of the aqueduct travertine chemostratigraphic data set to include carbonate-relevant δ¹⁸O, δ¹³C, and ⁸⁷Sr/⁸⁶Sr as well as Ca, Mg, Sr, Mn, and Fe elements with distribution coefficients (K_Ds) that are both greater than and less than one (Banner and Hanson, 1990; Brand and Veizer, 1980, 1981); (5) covariation modeling to evaluate and correct for potential mixing during sampling between the original travertine and diagenetic components (Langmuir et al., 1978; Sivaguru et al., 2019a), as well as cumulative water-rock molar ratio mass balance diagenetic water-rock interactions among freshwater and the aqueduct travertine (Banner and Hanson, 1990; Fouke et al., 2005); (6) evaluation of multiple other original biotic and abiotic influences during travertine deposition (Fouke, 2001, 2011; Fouke et al., 2000; Kandianis et al., 2008; Veysey et al., 2008); (7) addition of U-series isotope dating if the samples pass petrographic and geochemical screening for diagenetic alteration; and (8) controlled laboratory experimentation using microfluid testbeds such as the GeoBioCell (Fouke et al., 2022) to determine how and why the aqueduct travertine ripple marks form as a result of in situ crystal precipitation processes rather than downstream hydraulic transport of sedimentary grains (Keenan-Jones et al., 2022).

The depositional and diagenetic history of travertine deposited within the ruins of the Anio Novus aqueduct of ancient Rome was studied by combining hydraulic reconstructions with high-resolution optical, laser, electron, and X-ray microscopy analyses. Samples were collected at upstream to downstream 0 m and 140 m sites along a continuous run of the Anio Novus aqueduct channel at Roma Vecchia. The depositional history of the aqueduct travertine included precipitation of dendritically branching aggregates of 1–3-μm-diameter euhedral calcite crystalline shrubs with high-frequency, dark–light biofilm laminae, linguoid, sinuous and hummocky crystal growth ripples, and lee sand lag deposits. The diagenetic history of the Anio Novus aqueduct travertine included the precipitation of fabric destructive and mimetic fabric preserving upward-branching radiaxial calcite replacement crystals, which cross-cut the dark–light laminae stratigraphy and crystal growth ripple bedforms. Future studies aimed at reconstructing human activity and paleoclimate will be able to incorporate these approaches to create the type of depositional and diagenetic frameworks required for more accurate reconstructions from travertine deposited in other ancient water conveyance and storage systems around the world.

This research was completed in recognition of, and appreciation for, the innumerable personal and professional lifetime achievements of Walter Alvarez at the University of California, Berkeley. People around the world have had their curiosity ignited, and their fundamental approach to scientific inquiry redirected, by Walter's avant-garde research vision, engaged teaching, and inspirational writing. Walter infused his passion for Italian language, food, art, music, and history into all of us who were members of his Renaissance Geology research group. We gratefully acknowledge that permission to conduct this research was provided by the Soprintendenza per i Beni Archeologici del Lazio (especially Dott. Zaccaria Mari) and the Soprintendenza Speciale per i Beni Archeologici di Roma (especially Arch. Giacomo Restante). This research was supported by the Andrew W. Mellon Foundation through the Illinois Program for Research in the Humanities, the Italian Government, the late William and Janet Gale, Macquarie University, the British Academy and British School in Rome, the Ed and Barbara Weil Fund for Universal Biomineralization at the University of Illinois Urbana-Champaign, and the National Aeronautics and Space Administration (NASA) Astrobiology Institute (cooperative agreement NNA13AA91A) issued through the Science Mission Directorate. The support of the Chester and Helen Siess Professorship and the M.T. Geoffrey Yeh Chair in Civil and Environmental Engineering at the University of Illinois Urbana-Champaign is also gratefully acknowledged. We also thank Glenn Fried for modifying and enlarging the cathodoluminescence stage and assisting with cathodoluminescence petrography, Charlie Kerans and Jeff Trop for invaluable scientific discussions, and Julia Waldsmith and Megan Ward for assistance in the field and laboratory. Conclusions and interpretations presented in this study are those of the authors and do not necessarily reflect those of the funding agencies and permitting entities.