From features to fingerprints: A general diagnostic framework for anthropogenic geomorphology

While many species alter their environments, humans are Earth's ultimate ecosystem engineers, engaging in a broader range of more potent environment-modifying behaviours than any other species (Smith, 2007; Smith and Zeder, 2013). By engineering environments using fire, tools of increasingly complex design and domesticated species, and by harnessing non-human energy to accomplish these modifications, human niche construction practices have radically enhanced the adaptive fitness of human individuals, social groups and societies, enabling human societies to increase in scale and to extend their reach across Earth's terrestrial surface (Ellis, 2015; Ellis et al., 2018). Most importantly, human capacities to engineer environments are not biological, but sociocultural, and these sociocultural capacities have evolved and accumulated over time together with human capacities for cooperation and social learning (Ellis, 2015; Ellis et al., 2018) (Figure 1). Human sociocultural niche construction has diversified, scaled up and utilized increasing amounts of energy over millennia, from the first use of fire to clear land (for greater success in hunting and foraging), to the construction and management of agricultural landscapes using animal labour, to urban settlements powered by fossil fuels, to the global networks of exchange infrastructure that have made contemporary human societies the most interdependent in history and have made humanity a global force of nature (Figure 1).

Human modification of landscapes has tended to increase in scale and complexity together with the scale and complexity of human societies (Ellis, 2015). By the Late Pleistocene, opportunistic use of naturally occurring fire was supplemented by technologies of fire-maintenance and fire-making (Sandgate and Berna, 2017), amplifying the ability of humans to transform ecosystems and the erosive and

hydrologic processes accompanying land clearing over extensive areas, either intentionally or unintentionally. The construction of settlements by sedentary hunter-gatherers and early farmers left more robust geomorphic evidence due to the systematic and deliberate reconfiguration of landscapes, often reflecting distinctive patterns in social behaviour and material culture (Fletcher, 2009). As societies scaled up over time, into the first agricultural, urbanized and then industrial societies, their ecosystem engineering also increased in scale, complexity, durability and in the sheer amounts of material moved (Figure 1).

Even more significant complexity in anthropogenic landforms has emerged over time, as the geomorphic features engineered in support of one society have come to be inherited by, overlaid by, overwritten and reconstructed by later societies, such that the anthropogenic features and landforms of regions with sustained human occupation represent not the directed efforts of a single society, but rather the complex sociocultural palimpsests of multiple diverse societies (Bailey, 2007). Moreover, given that natural geomorphic processes continue to act during and after societal processes, the landform palimpsests of anthropogenic landscapes represent the sustained interplay of sociocultural and biogeophysical processes over time (Johnson and Ouimet, 2018).

Individual anthropogenic features, as specific human modifications of land surfaces such as an individual graves, buildings or roads, show a wide variety of forms. In aggregate, however, the anthropogenic features produced by a given society may evince a degree of consistency within a given sociocultural milieu, as illustrated in Figure 1. Thus, as archaeologists have long recognized, the material cultures of different societies pattern the surface of Earth in specific ways that produce distinctive and recognizable sociocultural fingerprints (Butzer, 1974; 1982; David and Thomas, 2016).

Anthropogenic landforms have tended to increase in scale and complexity in parallel with the scale and complexity of the societies that produce them. Composed of individual land surface modifications, or “anthropogenic features”, these are the products of socially and culturally directed efforts by individuals and social groups, increasingly assisted by animal labour, fossil-fuelled machinery and other non-human energy, to remove, transfer, introduce and reshape elements of the physical environment to achieve the

goals of these individuals, social groups and entire societies. This social production of space generates a virtually endless variety of physical forms. We must therefore acknowledge at the outset the difficulties in dividing such myriad diversity of physical forms into neat categories according to their perceived function or meaning. Elements of material culture can, of course, be multifunctional, and polysemic, in ways that are arbitrary, unknowable and subject to variability over time. We argue, however, that there is practical and heuristic value in structuring the complexity of these physical forms into a generalized ontology of anthropogenic features, as in Figure 2: symbolic (e.g. graves and monuments), habitation (e.g. housing, villages, cities), transport and exchange (e.g. roads, canals), subsistence (e.g. plough furrows, irrigation networks), mining, refuse disposal and warfare (forts and battlements), among others.

Individual anthropogenic features are readily detectable on Earth’s surface (or below) through observations made on their geometric forms, and possibly through a range of other physical or chemical attributes. Smaller scale societies, such as those of the first farmers, tend to produce a relatively narrow range of anthropogenic features. However, even small-scale societies produce a diversity of distinctive anthropogenic features on the landscape within the framework of their material culture. For example, a horticultural society produces building and settlement structures, hand-tilled fields, waste middens and graves, while the material culture of more complex societies is characterized by large-scale urbanism, ploughed fields, irrigation systems, mines and quarries, roads, monuments, cemeteries and many other anthropogenic features. These individual features are the fundamental elements of material culture at the landscape scale and, considered in aggregate, it is possible to identify the distinctive spatial patterning of a society elaborated across time and space on the surface of Earth: the sociocultural fingerprints of societies and cultures. The

layering of these sociocultural fingerprints over landscapes over time, and their interactions with each other and the natural geomorphic processes that simultaneously shape them, combine to produce the landforms that cover most of Earth’s terrestrial surface today.

To understand the socioeconomic and sociocultural forces that shape landscapes, it is first

necessary to recognize that these operate at the scale of the societies engaging in them (Ellis, 2015). Human societies have long shaped terrain at local and regional scales. Now, human societies numbering in the billions are reshaping landforms globally through global-scale processes enabled by interconnected systems of economic, social, material and energy exchange, increasing levels of technological innovation and rapid socioeconomic development (Fox et al., 2017; Zalasiewicz et al., 2017). Yet, even while the anthropogenic features and fingerprints generated by societies have tended to extend spatially across their areas of influence, their distributions are usually highly heterogeneous within and across these areas. Societal demands for food, minerals, settlements, economic exchange, symbolic communications and other needs are generally expressed unevenly across societal areas of influence to access the distribution of opportunities afforded by natural patterns of resource-use potential, such as fertile soils, level terrain, access to water and other societally perceived patterns defining their suitability for different uses, and also based on pre-existing social structures, such as settlements, transport infrastructures and areas of symbolic significance that define their value and accessibility for societal use (Butzer, 1982b; Ellis, 2015).

The economic and social demands of global-scale societies are now met through processes that engage with social and environmental differences around the world at regional scales – where land, resources and labour are less expensive and more accessible. As a result, as socioeconomic demands and exchange processes have scaled up, some world regions are experiencing rapid increases in anthropogenic modification caused by deforestation and tillage to meet far-away food demands, road-building to interconnect global supply chains or dam construction to supply energy, while at the same time other regions are experiencing rapid decreases in such modifications and even complete abandonment by human populations (Meyfroidt et al., 2013). While uneven spatial patterns of anthropogenic landscape modification are just as common in small-scale societies, they tend to be produced by very different socioeconomic and sociocultural processes. In general, the forms, extents and distributions of anthropogenic landscape modifications must all be seen as variable across and within societies as a function of both pre-existing natural patterns, largely related to terrain and climate, and to pre-existing sociocultural patterns, largely related to the cultural and technological capacities and patterns of social structures in space, and the

varying interactions of different societies with these conditions (Butzer, 1982b; Ellis, 2015).

Anthropogenic landscape modifications are therefore a combination of cultural forms, technologies and natural environmental patterns, and can involve anthropogenic interventions that accelerate or delay natural geomorphological processes, producing hybrid anthropogenic/natural features, or engineer entirely distinct anthropogenic features with purely cultural functions. For this reason, anthropogenic landforms vary widely in form and function, in the degree to which they are distinct in their spatial patterning from natural landforms and in the degree to which they may be recognized as the product of a specific cultural milieu. Whereas natural mass flows generally move from high to low positions or from upstream to downstream under the influence of natural forces, and their capacity to move dissipates over distance, landforms engineered through direct human agency are shaped primarily by social and cultural imperatives that produce landforms distinctly different from natural physical processes. Anthropogenic landscape engineering often represents elaborate attempts to subvert natural processes in culturally specific ways – by damming floodplains and diverting sediment-laden floodwaters into walled field systems, for example – which can in turn feed back into natural systems and generate consequential changes within them. To understand human modifications of landscape evolution across multiple scales of time and space, a basic framework for describing and classifying anthropogenic landforms is needed, together with an understanding of the underlying generative principles that produce them, and the nature and degree of their ability to alter the functioning of landscapes.

The first symbolic features of human societies are likely burial sites (Belmonte, 2014). Entombments in ancient times have included caves and underground burials with different designs relating to the wealth and status of the deceased (Cuezva et al., 2016). As societies evolve, so have burial practices, from ground and cave burials to more elaborate constructions (Sidebotham, 2014; Tomczyk et al., 2011). In many cultural contexts these funerary structures represent massive investments in reconfiguring the natural landscape, and their sheer size makes them relatively resistant to natural weathering and degradation

(Guthrie, 2015). In some instances, for example with geoglyphs, it is clear that landscape-scale transformations of the natural environment have been undertaken in order to encode meaning in geometric patterns (Briones-M, 2006; Tapete et al., 2013, 2017). Geoglyphs can play a role in mortuary practices as well as aiding in the trapping of migratory animals, and serving as cleared areas for camps, houses and animal enclosures (Kennedy, 2011). They are formed of durable materials such as soil, stones, clastic rocks, live trees, gravel and earthworks (Sparavigna, 2010). Prehistoric agricultural societies, especially in parts of Europe, also produced large stone structures, or megaliths, including large single standing stones, as part of buildings, as portal dolmens, rotundas and passage graves, as well as henges, all of which served diverse but essential societal functions (Beck and Chrisomalis, 2008; Fleming, 1999, 2005; Holtorf, 1998; Midgley, 2010). Historical and contemporary societies also produce a wide variety of symbolic landscape features in parks and gardens, including stylized berms incorporated into artworks, and many shaped to mimic stylized natural landforms, such as those of central park in New York City (Portal, 2017).

Early human encampments relied on rock shelters formed by natural processes, such as caves (Farrand, 2001; Judson et al., 2005; Simms and Russell, 1997), whose natural and anthropogenic deposits offer details on human activities, natural environments inside such caves and their dynamic interactions over time (Courty and Vallverdu, 2001; Farrand, 2001). With the rise of sedentary societies, especially those sustained by agriculture, people began to build more permanent shelters and to concentrate these in ever larger settlements, from seasonal encampments, to villages, towns, cities and ultimately into interlinked complexes of urban and rural landscapes sustaining large-scale social formations, or states (Goudie and Viles, 2016).

As settlements grew in scale, these came to include not only housing but also specialized buildings for governance, crafts production, temples and other cultural needs, water infrastructure, road networks and boundary walls or fortifications (Beranek, 1988; Campbell, 2006; van der Spek, 2017). Many early cities prospered, dwindled and disappeared, such as the Mesopotamian city of Ur, in ruins by 450 BCE (Azara, 2015; Launderville, 2013). Other ancient cities have survived to the present, such as Jericho, one of the oldest inhabited cities in the world, well-known for its protective walls (Issar, 2008). As with Jericho, defensive enclosures are one of the most easily recognizable components of the archaeological record (Boyce et al., 2015; Cassel, 2012; Rojas, 2010). Other urban forms include the large-scale urban complexes and hydraulic systems that evolved in densely forested landscapes in Asia and Amazonia (Evans et al., 2013; Liu et al., 2017; Roberts et al., 2017). The production of built structures, from walls to housing and those for other specialized purposes, induces a variety of geomorphic processes ranging from concentrated excavation to the intensive transport, deposition and restructuring of materials, to the creation of new artificial landforms. All of these intentional structuring processes tend to further enhance other geomorphic processes, including water accumulation, the deposition of waste materials in excavated areas, destabilizing of new landforms and underground structures, intensified erosion and other environmental consequences (Courtney and Vallverdu, 2001; Farrand, 2001).

Urban development is also accompanied and facilitated by the development of transport networks composed of paths, unimproved and improved roads, railways and waterways connecting rural resources and people to distant markets and population centres (Bell and Iida,

1997; Kara and Verter, 2004; Morriss, 2003, 2005; Ruiz, 2016; Snead et al., 2009). Roads are ubiquitous features of landscapes and different shapes of road network represent the different evolution of civilization at different times (Strano et al., 2012). Road networks can reflect the central or decentralized organization of countries, and provide important pathways for mobility, migration and demographic change. They induce the rural to urban migrations as well as the massive emigration. In addition, since the Industrial Revolution, railroads have been an important form of transportation (Li, 2017; Morriss, 2003), shaping not only cities and economic growth (Hanedar, 2017; Nerlove, 1966), but also landscape processes (Bell and Iida, 1997; Blanton and Marcus, 2009, 2013; Kumar et al., 2014; Martinović et al., 2018). Railways, for example, can induce selective disclosure of the countryside (Antrop, 2004). Villages that received a station in some cases (but not always) developed rapidly into urban-like centres and their surrounding landscape changed accordingly. Developed road networks could also affect new urban development in tourist resort areas that once were remote rural regions with limited access (Petrov et al., 2009). The development of both networks (road and rail) is also accompanied by the use or even reshaping of the geomorphologic and topographic setting to create new infrastructure, tunnelling and bridges (Booth et al., 2011; Buchanan and Jones, 1980; Day, 1995; Pratesi et al., 2016; Watson et al., 2001). More recently, the development of underground tunnels for transport has created extensive subterranean networks in many large cities (Zalasiewicz et al., 2014).

Together with deposits of stone tools, hearths and other evidence of controlled fire for cooking, warmth and protection from predators are likely the earliest evidence of human transformation of geophysical environments (Roebroeks and Villa, 2011; Smith, 2007; Wrangham, 2009). The earliest intentionally built fires were little more than controlled burns of woody and other flammable organic materials, but these evolved into hearths constructed of stones and earth, and ultimately into ovens, kilns and other larger scale fire containing structures. A typical hearth pit of the past formed a bowl-shaped soil discoloration less than 1 m in diameter and less than 0.5 m deep, with evidence of soil or rock exposure to high temperatures (Wandsnider, 1997), with differences in hearth-pit size and shape indicating different functions (Crombé et al., 2015; Pearson and Pearson, 1999). Human engineering of environments for food production begins in the late Pleistocene with hunter-gatherer modification of landscapes to concentrate and trap game and the use of fire to clear and enhance vegetation growth, followed by the rise of agriculture in the early Holocene, involving the digging of holes for planting, soil tillage and the construction of livestock pens and fencing, activities continuing to the present day (Ellis, 2015; Ellis et al., 2013; Smith, 2013; Smith and Zeder, 2013). Soil tillage leaves especially clear and widespread geomorphic signatures across landscapes in terms of anthropogenic soils and erosive deposits (Certini and Scalenghe, 2011; Ellis, 2011; Tarolli et al. 2019). The need to exploit steep lands for agriculture also introduced the construction of terraces (Chase et al., 2014; Tarolli et al., 2014; Vogel, 1987). Geomorphic features produced by livestock management include animal pens with soils altered by trampling and manure deposits, their surrounding fencing, sometimes constructed using rocks, earth and other durable materials, and the extension of fencing and animal paths across landscapes, often associated with water and shelter, including induced the deposition of stones and earth near fences, soil compaction and erosion from paths and disturbance and vegetation removal caused by overgrazing (Amy and Robertson, 2001; Mwendera

et al., 1997; Tarolli et al., 2013; Wu et al., 2017; Yong-Zhong et al., 2005; Zhang and Zhao, 2015).

Although mining is as old as the manufacture of stone tools and leaves clear geomorphic signatures, it occupies a relatively small surface area worldwide (Tarolli and Sofia, 2016). Mining activity has been driven by a wide variety of objectives, including obtaining materials for making items such as tools, utensils, weapons, ornaments, decoration and currency (Timberlake, 2017). Mining of clay for bricks is also widespread and continues globally (Shakir and Mohammed, 2013). Further mining drivers have been the need for structures, and machinery, and obtaining resources for energy, electronics, nuclear fission (Hartman and Mutmanský, 2002; Herrington, 2013) and for low-carbon energy production (Vidal et al., 2013). As a consequence, various so-called “mining landscapes” have emerged all over the world since the 19th century. They are characterized by unique excavated, accumulated and planed landforms (Dávid, 2010) that bear the signature of the material sought. Such signatures can be broadly separated into two classes: surface and underground, although combinations of the two may occur in time and space (Mossa and James, 2013). The main characteristic is the persistence in time of the landform itself (Chen et al., 2015; Hooke and Martin-Duque, 2012; Xiang et al., 2018) and of the related effects on the environment (Kircher et al., 2003; Mossa and James, 2013; Rivas et al., 2006; Toy and Hadley, 1987).

Although several forces beyond water management combine to shape societies, the availability of water strongly influenced the trajectory of past societies (Scarborough, 2017). As humans began to settle as farmers during the Neolithic, water wells and water infrastructures began to rise (Angelakis and Andreas, 2012; Kollyropoulos et al., 2017; Tegel et al., 2012). These features leave a distinctive topographic signature that can still be identified on the landscape today, thousands of years later (Yevjevich, 1992). An example of this would be Qanat, which are gently sloping underground channels designed to transport water from an aquifer or water well to the surface for irrigation and drinking (Wessels, 2014), originating in the first century and distributed mainly in the Iranian plateau and east of Xinjiang Province in China (Goes et al., 2017; Harandi and de Vries, 2014; Luo et al., 2014). Another example is the famous hydraulic infrastructure of Pont du Gard in southern France (Dumas, 2011; Vrancic, 2010). Ditches provide further water infrastructure for reclamation purposes, irrigation and drainage alongside roadways or fields (Terry and Hughes, 1978). As well, canals provide water conveyance, navigation and water storage (Swamee and Chahar, 2015). Along water infrastructures, especially in floodplains, humans built embankments, levees and scarps. These infrastructures not only offer protection from floods, but they also change the flood frequency and create disconnectivity between the river landscape and its floodplain (Chin et al., 2013; Gregory, 2006; Overeem et al., 2013; Thoms et al., 2005).

Another example of water control measures is provided by dams and reservoirs, which have increased dramatically in number and scale globally since the 1950s (Chao, 1995), to the point that water impounded in artificial reservoirs since the 1950s is by far the most significant anthropogenic hydrological change in terms of water volume (Vörösmarty et al., 2003, 2004). These structures are built for irrigation purposes, water for human consumption, agricultural use, power generation, land drainage and reclamation, flood mitigation and recreational use (Castelletti et al., 2008; FAO, 2001; Gordon and Meentemeyer, 2006; Hall and Shelby, 2000; Le et al., 2007). These

functions are, however, accompanied by environmental changes, including subsequent impacts on lowland rivers and coastal systems, climate-changing greenhouse gases emissions, changes in the temperature regime, sedimentation, water pollution and destruction of ecosystems due to the reservoir effect on downstream environmental flows and sediment supply (Dević, 2015). Artificial wetlands have also become ubiquitous in recent decades, and are a distinctive form of water management system created for the purpose of treating municipal or industrial wastewater, greywater or stormwater, to act as a biofilter and remove pollutants from the water and for land reclamation after mining, refineries or other ecological disturbances (ElZein et al., 2016; Vymazal and Kröpfelová, 2008; Zhang et al., 2009). Some of the most dramatic anthropogenic transformations of Earth's surface occur at the interface between terrestrial and aquatic environments, which include massive and elaborate mechanisms for displacing and managing water. Recent advances in engineering technology have allowed for extensive land reclamation and the creation of artificial islands to provide new spaces for development and to achieve political and strategic objectives (Goudie and Viles, 2016; Martín-Antón et al., 2016).

Middens are among the first and most prominent examples of physical remnants of resource exploitation activities in the archaeological record, with potential consequences for a range of related physical and chemical characteristics, for example increased soil alkalinity, and in particular increases in levels of nitrogen, calcium, potassium and manganese (Cook-Patton et al., 2014). Middens offer an important archive of information about human dispersal and group diversification because they can be created either at the household level or the community level (Trebsche, 2009). They preserve records that are particularly valuable for the development of interdisciplinary approaches in investigating human–environment interaction, social relations and the role of resource exploitation in the developmental trajectory of human groups (Álvarez et al., 2011; Rick et al., 2005). They are also suitable for radiocarbon dating (Brady, 2016). Another widespread remnant of past resource exploitation survives in the form of tells, artificial features (mounds) formed from the accumulation of refuse over millennia of settlement activities (Portal, 2017). Their size and shape carry the fingerprint of diverse cultures and, in most cases, they are representative of the earliest settlement systems, beginning in the Neolithic period (Menze et al., 2006). In modern times, middens and tells evolved as landfills used for waste management and processing waste material in the most cost-efficient way (Stirling, 2015). The physical structure of landfills embodies diverse input characteristics, such as aspects of terrain, existing land use, aesthetic value, government regulations and public opinion about the sites themselves (Sharma, 2010).

Landform features have also been created for warfare or defence. These features include not only artificial structures, such as fortifications (De Matos-Machado et al., 2015; Ilyés, 2010; Moss and Erlandson, 1992), trenches (Baer and Ashworth, 1981; Houx, 2001; Power, 2009), improvised explosive device (IED) command wires (McDonald and Schumer, 2016), bunkers and missile silos, among others, but also existing underground terrain (Eastler, 2004) or rock defences (Moss and Erlandson, 1992). The most famous example of trench warfare is the Western Front World War I, which has become the classic example of stalemate, attrition and futility in modern conflict (Baer and Ashworth, 1981). However, this class of features also

includes direct geomorphological evidence, such as bomb and mine craters (Hupy and Schaetzl, 2006; Kiernan, 2015), uranium mining (Blustain, 2016) or landscape modification due to military structures (Yatsko, 2016). Clearly, conflicts leave a significant topographic signature on the landscapes, and they also result in long-term landscape modifications (Bothe, 2007), with impacts on remoteness and naturalness, or on physical phenomena such as geodiversity (geology, landforms, soils and the natural processes that give rise to them) (Kiernan, 2015).

Despite the limitation due to shadowing, angle of incidence or lack of topographic information under vegetation cover (Mostafa and Abdelhafiz, 2017; Quartulli and Olaizola, 2013; Sowmya and Trinder, 2000), archaeologists have made use of radar and satellite images to reveal differences in texture, roughness, moisture content, topography and geometry of features and surfaces related to human activities, for example (Evans et al., 2007; Fowler, 2002; Holcomb, 2001; Holcomb and Shingiray, 2006; Moore et al., 2006; Ricketson et al., 2003). More recently, images from satellites have been used not only in relation to anthropogenic features in the context of urban studies (see Ghanea et al., 2016, for a review), for cadastral boundary extraction (Crommelinck et al., 2016; Wassie et al., 2017) and oil spill detection (Brekke and Solberg, 2005), but also to identify target features in ongoing and post-humanitarian crisis scenarios (Witharana, 2012; Witharana and Civo, 2012). In analyses of this type, image analysis techniques, such as wavelets and fusion algorithms, or texture analyses, can enhance and localize patterns related to human activities, and they can be used to extract features of interest, such as not only buildings, but also shelters and war/flood/earthquake damaged structures. To illustrate this point we

show in Figure 3 a Pleiades satellite image. The study area is spatially stratified in environmentally similar land units (Figure 3(d)) using image analysis techniques (Figures 3(b) and (c)) and, thus, it is possible to identify those areas that are most likely to present anthropogenic landscapes (Figure 3(e)). Feature extraction from satellite images is, however, limited by shadows (Mostafa and Abdelhafiz, 2017) and other technical issues, such as angle of incidence (Quartulli and Olaizola, 2013; Sowmya and Trinder, 2000). Satellite images, furthermore, cannot be used to extract ground features under vegetation cover.

A further step in feature extraction is allowed by the availability of three-dimensional (3D) models (Digital Terrain Models – DEMs). When DEMs are available, the most straightforward approach to anthropogenic feature analysis is visualization, namely the use of hillshade or shaded-relief maps (Figure 4(a)), for example (Bennett et al., 2012; Chase et al., 2014; Deforce et al., 2013; Doneus et al., 2008; Golden et al., 2016; Harmon et al., 2006; Johnson and Ouimet, 2016, 2018; Menze et al., 2006; Risbøl et al., 2013; Tapete et al., 2017).

This approach simulates lights and shadows to enhance the presence of features captured by the input dataset. Slightly more advanced techniques, based on compound approaches, have been also developed, such as the sky view factor (Figure 4(b)) (Bennett et al., 2012; Devereux et al., 2008; Kennelly, 2008; Štular et al., 2012) or advanced analysis of hillshades (O’Neal, 2012). The advanced use of topographic parameters directly derived from the DEMs allows not only for the visualization, but also for the classification of anthropogenic features. The most direct and straightforward approach is offered by the so-called “detrending” techniques. Data are filtered using “low pass” filters, resulting in a smoothed surface to subtract from a LiDAR DEM (Figure 4(c)), for example (Cazorzi et al., 2013; Hesse, 2010; Howey et al., 2016; Luscombe et al., 2015; Schindling and Gibbs, 2014; Sofia et al., 2014c). As the $z$ values in the detrended data represent the height difference from a smoothed surface, peak values in either the positive or negative domain (depending on the approach) highlight the microtopographic element in the landscape, such as drainage features or pits, or walls. Further useful topographic parameters can be used, such as slope (Figure 4(d)), (e.g. McCoy et al., 2011; Riley, 2012), slope and also exposure under a water surface (Cappucci et al., 2017) and curvature (Figure

4(e)), (Drăguț and Blaschke, 2006; Sofia et al., 2014a; Tarolli et al., 2014). Recent literature has highlighted how image processing local filters can also be applied to a DEM to detect high-frequency variations (e.g. Laplacian – Figure 4(f), Sobel’s filters, SLLAC – Slope Local Length of Correlation) (Sofia et al., 2014b, 2016; Stal et al., 2010; Štular et al., 2012). These enhancing techniques can be the basis to segment and detect anthropogenic patterns and features using GEOBIA (GEOgraphic-Object-Based Image Analysis) (see Blaschke et al., 2014; Cerrillo-Cuenca, 2017; Diaz-Varela et al., 2014; Eckert et al., 2017; Hay and Castilla, 2008) or automatic thresholding (e.g. Figure 5).

Extractions carried out with thresholding or object-oriented approaches, although able to detect the two-dimensional (2D) location of anthropogenic elements, are not able to quantify whether or not they are continuous features in the landscape. Nevertheless, they can offer the basis for further evaluations of lengths, densities and volumes, for example (Sofia et al., 2014c; Sofia and Tarolli, 2017; Sofia et al. 2019), or the extent of 2D features. In addition, high-resolution DEMs can offer the basis to measure the shape and sizes of anthropogenic elements directly in the digital realm (Johnson and Ouimet, 2016; Sofia et al., 2016a). A further step in anthropogenic features characterization is offered by change detection techniques or time-series data analysis, which

has gained significant attention due to its capability of providing volumetric and extent measures in time (see Qin et al., 2016, for a review). These techniques rely on the availability of multitemporal datasets, real or simulated, to be used to compare anthropogenic features in time (Haas et al., 2016; Prosdociimi et al., 2015; Wróżyński et al., 2017; Xiang et al., 2018; Yucel and Turan, 2016). Figure 6 shows an example of geomorphic change detection evaluated considering the Wheaton et al. (2010) approach. For this area, multitemporal orthophotos (Figures 6(a) and (d)) and LiDAR DEMs (Figures 6(b) and (e)) are freely available (Institut Cartogràfic De Catalunya (ICC), 2005, 2008) for the years 2005 and 2008. From these maps, it is possible to compute a volumetric difference (Figure 6(c)), taking into account the errors that might be present in the DEMs themselves (see Lane et al., 2003, for specification on the method).

Improvements in spectral, spatial and temporal resolution are revealing more and more rich and detailed information to decipher landscape palimpsests. The synoptic view offered by compound analyses integrating satellite, airborne and photogrammetry enable integrated landscape investigations across scales in terms of resolution and coverage, from small features to entire regions, and comparisons across and within regions, and long time series allow

monitoring and diachronic analysis of changes in landforms. Recent availability of wide-area, high-resolution LiDAR is revolutionizing these efforts, owing to its unique capability to penetrate vegetation canopies and identify earthwork features at high spatial resolutions even under dense vegetation cover.

Challenges remain for using these data to detect anthropogenic features and their distributions across landscapes and regions. A better understanding of the representation of anthropogenic features within the data itself will be critical, combined with improvements in processing techniques that make analyses of vast quantities of data more manageable. Moreover, each study presents its own challenges. For example, traces of ancient ruins and infrastructures might lie below relatively modern roads, buildings and agricultural fields, or ancient structures might be made of the same material that underlies a whole study area. A further impediment to reliable feature identification is that long-term exposure to natural environmental processes, such as erosion and sediment deposition, can make anthropogenic features extremely difficult to distinguish from natural ones (Liu et al., 2017). A clear example of this is the effect of natural weathering processes on landfills or tells: ancient mounds nowadays might resemble natural hillslopes, and modern landfills will, over time, tend to smooth out to the same level as the surrounding land surface.

Developments in computer science and statistics offer a wide range of powerful tools to

augment the detection and interpretation of anthropogenic features, and potentially the sociocultural fingerprints of entire societies across large regions. These tools include advanced systems for geographic information systems (GISs), 3D modelling, predictive modelling, visualization, simulations and machine learning (e.g. Guyot et al., 2018). Ironically, however, there remains no standard ontological framework within which to classify and understand anthropogenic landscapes and the features within them, despite their coverage of the vast majority of Earth's terrestrial surface (Ellis et al., 2013). By developing standardized frameworks for feature and fingerprint detection and interpretation, and ultimately by automating these, rapidly expanding Earth coverage by high-resolution 3D remote sensing can be utilized to examine Earth's global geomorphic transformation by human societies, together with a wide range of other useful societal applications (Tarolli et al., 2017).

The first principle of anthropogenic geomorphology is that the sociocultural processes that form anthropogenic landforms are fundamentally different to natural processes like erosion or deposition. Even though natural and anthropogenic geomorphic features may share real similarities in form and even function (Figure 7), the processes forming them are fundamentally different. Unlike the forces of gravity, hydrology and geophysics, which have not changed since the formation of this planet, the forces of landscape change generated by human societies can and have changed dramatically over time, together with societies themselves (Figures 1 and 8).

The first step towards a process-based understanding of anthropogenic geomorphology is therefore the development of a timeline of known sociocultural changes across the landscape of interest, with the aim of understanding the historical layering of anthropogenic features, together with a timeline of major natural environmental changes and events, including major changes in precipitation, floods, tectonic and volcanic activity that have interacted with anthropogenic processes in forming the landscape under observation. This basic timeline should include a catalogue of the types of anthropogenic features created by different societies, including any geometric, symbolic or other cultural identifiers that can assist in reconstructing the layers of anthropogenic features produced across landscapes during specific periods of time. By combining social-geomorphologic timelines with detailed spatial data on landform structure, the spatial patterning of anthropogenic features and landforms may be understood in terms of the sociocultural and natural processes that have formed them.

While the full range of sociocultural patterns, processes and dynamics observed across human societies (Figure 1) is far larger than those occurring within any particular landscape, Figure 1 illustrates that anthropogenic geomorphic practices can be completely different in different societies, and that larger, more complex societies tend to engage in a larger number of different types of anthropogenic geomorphic practices and that the size of geomorphic features is also associated with societal scales (larger scale societies are generally capable of producing larger scale features). As the features of earlier societies can be added to by societies that come later, the total number of different types of anthropogenic features should be expected generally to increase over time, although the removal, destruction and overwriting of earlier features by later societies can erase or reduce the accumulation of

landform complexity, as can erosion and other natural processes.

A second principle of anthropogenic features and landforms is that their constituent materials may differ dramatically from those found naturally within a specific landscape or region (Zalasiewicz et al., 2017). The presence and composition of these materials, from stone, wood and other materials distinct from local materials and transported from other areas, to bricks, concrete, steel and other materials produced exclusively by human societies, and the presence and spatial organization of anthropogenic materials can be diagnostic of anthropogenic geomorphology – and detectable using remote sensing. The use of anthropogenic materials also enables the construction of anthropogenic features without natural analogues, from buildings to highway overpasses. As with feature types and sizes, the number of different types of materials utilized in feature construction also tends to increase with societal scale, with larger scale societies generally deploying a greater number of novel materials across anthropogenic landscapes.

A third principle of anthropogenic geomorphology is that within the regions influenced by a given society, the sociocultural functions that create anthropogenic features are not homogeneous, but rather differ across space in relation to both social and natural patterns. For example, within agrarian societies, there are often distinct clusters of housing, connected

by paths and roads, with agricultural fields and irrigation systems in the flatter lands in between, with less used areas on hillslopes. Larger scale urban societies show similar social patterns, but these may operate at much larger scales, and even globally at the present time. These patterns are aligned with the broader model of human sociocultural niche construction (Ellis, 2015), as modified and illustrated by the state equation: Anthropogenic landforms = $f(\backslash text\{biome,\; terrain,\; societies,\; centrality,\; time\})$.

This model describes the anthropogenic landforms produced in a given location in terms of its general biogeophysical conditions (biome, e.g. desert, savannah, rainforest), the specific terrain (terrain, e.g. floodplain, hillslope, mountain top), the societies that have operated at the location, the social position within the societies' operating spaces (social centrality, i.e. the travel distance from social centres, such as villages, towns and cities, the opposite of remoteness) and the amount of time over which the anthropogenic processes have operated. Based on this simple conceptual model, the types of features present within a specific location are contextualized both within the social space of the societies operating across a given site or region, with anthropogenic features produced by habitation expected to occur at high frequencies in areas of high social centrality, with transport features interconnecting these, while subsistence, mining, refuse, water infrastructure and other anthropogenic features are distributed in relation to suitable terrain and distance from social centres. In smaller scale societies, these functions and features tend to be tightly clustered in space, sometimes with the exception of symbolic and mining features, while larger scale societies tend to distribute these functions and features more widely, even globally. By combining these general principles of sociocultural landscape formation in relation to underlying biogeophysical conditions, the spatial patterning of anthropogenic features and landforms can be understood as the sociocultural fingerprints of the societies that have constructed them.

Although anthropogenic features can increasingly be detected through remote sensing and distinguished from natural features as a function

of their shape and material composition, a separate process of anthropogenic landscape interpretation is needed to understand their sociocultural and functional identity and their relation to the sociocultural processes that formed them – the basis both for feature classification and labelling and for the interpretation of features across landscapes and regions to identify the sociocultural fingerprints of societies. Automated feature classification is now increasingly capable of identifying classes of relatively homogeneous 3D entities (Figure 3(d)), but without further understanding of their sociocultural functions, such as habitation, symbolic culture or water infrastructure, these classifications remain purely geometric, and feature shape can at times be an unreliable indicator of function. Nevertheless, there is growing technological capacity for defining feature templates describing the geometry of specific anthropogenic feature classes, thereby enabling automated feature function identification and labelling using template matching (Schneider et al., 2015), object-oriented techniques (Blaschke et al., 2014) and/or machine learning approaches (Valentine and Kalnins, 2016). There is great promise in using these and other methods, including supervised feature classification leveraging spectral and geometric signatures developed by archaeologists and physical geographers with knowledge of the social-geomorphic timeline of a given landscape and its associated anthropogenic features, to train automated procedures to identify specific functional forms of anthropogenic features, such as buildings (Figure 3(e)). Caution is necessary, however, as natural landforms can have similar shapes to anthropogenic features: for example, a hearth-pit, a bomb crater and a karstic sinkhole might all have very similar geometry.

The confusion of natural and anthropogenic patterns across landscapes can present similar challenges to the automated analysis of sociocultural fingerprints. A wide variety of natural spatial features have been shown to be statistically self-similar over many scales, suggesting that fractal patterns are a signature of natural geomorphic patterns (Goodchild and Mark, 1987; Rodriguez-Iturbe and Rinaldo, 1997; Tarboton et al., 1988). For example, road networks can assume dendritic forms very similar to those of river networks (Figure 7), and agricultural terraces can resemble river terraces.

Yet, built-up environments may also possess similar structures at several different scales (Batty, 2008; Batty and Longley, 1994; Frankhauser, 2008; Thomas et al., 2008). Road networks, for example, are inherently fractal (Liu et al., 2014). Nevertheless, patterns of statistical self-similarity may yet prove capable of identifying differences between natural geomorphology and sociocultural fingerprints by developing rule-based systems for the classification of spatial-scale dependence of large-scale geomorphic patterns (Jiang and Brandt, 2016). The automated detection of larger scale patterns by aggregate assessment of feature composition and configuration across space through deep learning algorithms offers a clear way forward, analogous to progress with automated fingerprint detection in forensics (Schmidhuber, 2015).

The future of anthropogenic geomorphology will depend on a level of integration between theory and methodology that goes far beyond what we have presented here. Nevertheless, the way forward is clear, requiring a focus on the broader sociocultural, spatial and temporal contexts of anthropogenic landscape formation, rather than the mere presence or absence of specific anthropogenic features. Analysing the spatial structuring of features across landscapes in relation to the sociocultural and material systems that have shaped them can be essential to determine whether features have originated naturally, have anthropogenic origins or have emerged through the interplay of social and

natural processes. One example of this broader focus on sociocultural fingerprints is the differential patterning of urban landscapes around the world (Figure 8). Centuriation (Figure 8(a)) is typical of regions historically conquered by Romans (such as Italy (Figure 8(b)) and Spain (Figure 8(c))). In such cases, the city/agriculture follows a grid traced by extending the ancient Roman roads (Cardo Maximus and the Decumanus Maximus of the ancient cities) into the surrounding agricultural land. Parallel secondary roads were then traced on both sides of the initial axes, dividing the territory into square areas. Residential areas within modern cities present a different structure (Figure 8(d)), where similar patterns are produced by street geometry adapted to exclude traffic at the local street level and facilitate flow at the collector and arterial levels. In such systems, major internal roads run between communities, rather than through them, and they create grid squares where the road network uses cul-de-sac streets complemented by (for example) bike and footpaths that connect the entire sector and beyond.

Other examples of sociocultural fingerprints include the spatial patterning of waste disposal sites from informal middens to the regionally planned systems of larger scale societies (Sharma, 2010), the integration of green spaces into cities, the development of “agro-urban landscapes” integrating traditional agrarian landscape patterns within contemporary urban/industrial developments (Cavallo et al., 2016; Evans, 2016; Seto and Fragkias, 2005) and the spread of road networks (Corcoran et al., 2013; Strano et al., 2012). Even more detailed understanding of the functional roles of anthropogenic geomorphology can even enable the detection of specific cultivation practices. For example, different patterns of ditch networks (Figures 5(c) and (f)) support different cultivation practices (e.g. tobacco (Figure 5(a) and corn (Figure 5(d))), and identifying these patterns requires a deeper understanding of the sociocultural practices of the societies that formed and/ or operated them. Such an understanding, formed through careful reconstructions of the material cultures of societies across sites, is at the core of much archaeological research (Butzer, 1982b; David and Thomas, 2016). As the spatial and social scale of societies has increased, so have their capacities to reshape landforms locally, regionally and globally. Evidence already supports the hypothesis that anthropogenic transformations of land surface processes now move more of Earth's surface than any pre-existing natural process

(Wilkinson and McElroy, 2007; Zalasiewicz et al., 2017). Yet, these broad estimates of anthropogenic global change demand to be assessed with the greater precision and spatial context made possible by high-resolution reconstructions of anthropogenic geomorphic changes at local and regional scales. These measurement capabilities, for example the computation of volumetric changes caused by anthropogenic reshaping of landforms (e.g. Figure 6(c)), are only increasing as a result of advances in sensor systems, computational technology and large-scale data sharing, making the prospect of quantifying processes of anthropogenic geomorphic change globally over the long term a real possibility (Tarolli et al., 2017). The prospect of global assessment of anthropogenic geomorphology is illustrated in Figure 9, examining a suite of sites around the world in which anthropogenic geomorphology has been mapped using satellite and LiDAR data available to the public as free open-data or as samples. Given that different uses of land are expected to shape different landforms differently (e.g. Sofia et al., 2014b), stratifying these observations in relation to global patterns of land use, such as the anthropogenic biomes (Figure 9), may offer a general sampling framework for the global mapping and quantification of anthropogenic geomorphic change.

Anthropogenic geomorphology forms just one of many layers of the evolving geomorphic palimpsests that now cover most of Earth's dynamic terrestrial surface. Through a systematic assessment of anthropogenic geomorphology around the world and across time periods, it may yet be possible to develop a more general framework, perhaps even a predictive theory, on the geomorphological niche of societies, uniting the analysis of distinct geomorphic features to assess the evolving sociocultural fingerprints of societal change.