Lidar visualization techniques for the construction of geoarchaeological deposit models: An overview and evaluation in alluvial environments

Lidar has become an essential tool for the mapping and interpretation of natural and archaeological features within the landscape. It is also increasingly integrated and visualized within geoarchaeological deposit models, providing valuable topographic and stratigraphic control from the contemporary ground surface downwards. However, there is a wide range of methods available for the visualization of lidar elevation models and a review of existing research suggests that it remains unclear which are most appropriate for geoarchaeological applications. This paper addresses this issue by providing an overview and quantitative evaluation of these techniques with examples from archaeologically resource‐rich alluvial environments. Owing to the relatively low‐relief nature of the terrain within these temperate lowland flood plain environments, the results show that there is a small number of visualization methods that demonstrably improve the detection of geomorphological landforms that can be related to the variable distribution of archaeological resources. More specifically, a combination of Relative Elevation Models combined with Simple Local Relief Models offered an optimal approach that subsequently allows integration with deposit models. Whilst the presented examples are from a flood plain setting, deposit models are pertinent to a range of landscape contexts and the methodology applied here has wider applicability.

significantly in terms of their visual outputs from relatively simple twodimensional vertical cross-sections and horizontal surfaces to more computationally demanding deposit thickness maps and pseudo-threedimensional models. Some advanced deposit models also integrate a range of airborne and terrestrial remote sensing (Carey et al., 2019;Crabb et al., 2022;Schmidt et al., 2019) and geophysical survey data (Bates & Bates, 2016;Engel et al., 2022;Verhegge et al., 2016Verhegge et al., , 2021).
Yet, the coalescing of such disparate, albeit complementary, measurements of flood plain surface features together with records of subsurface sediment stratigraphy and architecture is not straightforward. In particular, lidar data have become a vital part of developing very accurate three-dimensional (3D) models, but it is questionable whether these data are used to their full potential within geoarchaeological investigations. Consequently, this paper aims to explore this potential, review how lidar has been previously applied and establish several key parameters for its future inclusion within geoarchaeological projects.

| Deposit modelling in archaeology
Although the specific term 'deposit model' is not always applied, in the last decade, there has been an increasing number of geoarchaeological investigations that produced land classifications to construct a 'model' of the subsurface. Such an approach is widely used across Europe and North America, either to inform researchdriven projects (e.g., Bini et al., 2015;Carlson & Baichtal, 2015;Castanet et al., 2022;van Dinter et al., 2017;Fontana et al., 2017;Gregory et al., 2021;Mozzi et al., 2018) or as part of the design of mitigation and heritage management strategies within commercial archaeology (Carey et al., 2019;Gearey et al., 2016;Stastney et al., 2021). Most of the models produced by such investigations are managed and generated using computational and GIS tools, but there is also a range of software available that has enabled more widespread construction of deposit models in recent years (e.g., Rockworks, Groundhog, Voxler, Strater, etc.). In tandem with this, several 'best practice' papers have recently been drafted to provide guidance for their wider application by practitioners, consultants and planning archaeologists (Carey et al., 2019;Historic England, 2020).
Whilst these documents allude to the incorporation of lidar and other remote sensing data sets for the definition of surface topography and landforms, there is limited discussion as to how best to achieve this.
Moreover, most applications of lidar data within geoarchaeological deposit modelling contexts utilize relatively simple or unaltered visualizations of lidar-derived Digital Elevation Models (DEM).
However, a better definition of features may be achieved through additional visual enhancements, but there has been little discussion/ research into which techniques are most appropriate.
A focused search of existing English-language literature containing the keywords 'Lidar' and 'Geoarchaeology' using Google Scholar and Scopus (Elsevier's abstract and citation database) in September 2022 returned 71 items dating from 2006. These principally comprise research articles as well as a small number of book sections and conference proceedings concerned with regions in the temperate zone ( Figure 1). Of these, 72% dealt with alluvial environments, and around half include an interpretation analogous to a deposit model (i.e., they map the distribution of buried deposits of geoarchaeological interest across a site or landscape). In each case, lidar is integrated into the analysis to some extent, but only 18% use additional data transformation techniques to enhance the visibility of resources. In addition, there is frequently little or no justification for the selection of these visualizations, although there are exceptions to this (Mayoral et al., 2017). For example, hillshading is very widely used, but there is rarely an explanation as to what advantage this has over other visualization techniques or unaltered versions of lidar DEM. While alternative visualizations may not always be appropriate or necessary for every project, they have the potential to define a wide range of features and landforms more accurately. As such, this paper aims to provide a detailed review and case study of the suitability and efficacy of visualization techniques applied for the identification of geoarchaeological resources within the valley floors of the rivers Lugg and Wye, Herefordshire (UK). It is hoped that this study will facilitate wider integration of appropriate visualizations of lidarderived DEMs into investigations of complex depositional environments not only within fluvial systems but also more broadly within the study of sediment systems.

| Using lidar to define the archaeological potential
Within many parts of the temperate zone, postglacial lowland flood plains are low-relief environments containing a complex assemblage of alluvial landforms that provide a record of the evolution of the river system (Brown, 1997;Howard et al., 2015). Topographic expressions of natural landforms such as palaeochannels, gravel islands, levees, bars and other bedforms may be present and can exert a significant influence on past settlement patterns, human exploitation and impact on the landscape . Mapping and understanding these landform assemblages can, therefore, offer significant insights into the distribution of archaeological remains (Carey et al., 2006;Challis & Howard, 2003Passmore & Waddington, 2009). However, the lateral and vertical accretion of fine-grained alluvial sediments within lowland flood plains, which are associated with human activity and land-use changes, can conceal and/or reduce the topographic expressions of these landforms (Brown, 2009;French, 2003;Howard & Macklin, 1999). Thus, while lidar visualizations aid in the interpretation of geomorphological processes and landscape evolution, they cannot be taken at face value and are most effective when integrated with other intrusive data sets (e.g., boreholes, test-pitting, etc.) or deeper methods of geophysical survey such as low-frequency ground penetrating radar (GPR), electrical resitivity tomography (ERT), or electromagnetic induction (EM) (Bates & Bates, 2016;Engel et al., 2022;Verhegge et al., 2016Verhegge et al., , 2021.
Over the last two decades, lidar has become a staple method used within both landscape archaeological and geoarchaeological projects, primarily within temperate regions of the northern CRABB ET AL. | 421 hemisphere ( Figure 1). This is partly due to the accessibility of national or near-national data coverage within these areas at a relatively high spatial resolution (0.5−2 m), which has led to the wide use of these data sets to study the topographic expression of archaeological remains and natural landforms that can contain significant archaeological and palaeoenvironmental archives (e.g., Crutchley & Crow, 2010;Höfle & Rutzinger, 2011;A. F. Jones et al., 2007;Notebaert et al., 2009;Opitz & Cowley, 2013). Beyond this, numerous geoarchaeological projects have also used lidar to map geomorphological landforms within alluvial environments to model the distribution of archaeological resources (Brunning & Farr-Cox, 2005;Carey et al., 2017;Challis, 2006;Challis, Carey, et al., 2011;Stein et al., 2017). However, alongside simply applying lidar topography to map and model landforms and archaeology, an increasingly diverse range of methods that provide visual enhancements of the topographic models have been developed, allowing for better identification of surface and subsurface features and sediments (Devereux et al., 2008;Hesse, 2016;. Lidar visual enhancement methods vary in complexity and applicability across the full range of landscape settings (e.g., from shallow superficial cover in uplands to deeper sequences found in lowland alluvial and intertidal areas). Within the context of geoarchaeological applications, lidar is commonly used for topographic modelling of landforms of variable archaeological or palaeoenvironmental potential (Corrò & Mozzi, 2017;Mozzi et al., 2018;Ninfo et al., 2011Ninfo et al., , 2016Passmore & Waddington, 2009) and given its relatively wide use for geoarchaeological deposit modelling, it is timely to provide a more thorough evaluation of visualization techniques that are most effective for this purpose.
Moreover, alluvial environments offer a unique and challenging setting, where consideration of these visual enhancement techniques is of specific importance for the enhancement of subtle geomorphological features and associated modelling of their archaeological potential.

| LIDAR FUNDAMENTALS
Lidar (Light Detection and Ranging) is an active form of remote sensing that uses pulses of laser light to measure distances to the earth's surface. Most lidar systems operate using very narrow beams of laser (NIR) light (typically 1064 nm) and are operated from airborne platforms, though there is an increasing array of UAV-mounted F I G U R E 1 Visual summary of existing geoarchaeological literature that integrates lidar as part of their methodology. The term deposit model is applied in its broadest sense (e.g., the distribution of buried deposits of geoarchaeological interest is mapped across a site or landscape) and additional data transformations refer to any visualization method applied, beyond the simple display of an unaltered DEM. The total number of publications identified was 71 and full details are available in the table provided as Supporting Information Material. DEM, Digital Elevation Models.
(drone) sensors and terrestrial laser scanners that have been used for archaeological research and a variety of Earth observation applications (Pádua et al., 2017;Telling et al., 2017). However, this paper is principally concerned with airborne laser scanning and its geoarchaeological application within temperate alluvial environments.
Lidar is typically acquired as discrete return or full-waveform data, which are then converted into a point cloud. Discrete return data record information only from targets that yield strong returns over a predefined threshold, whereas in full-waveform data, the whole waveform is practically digitized, regardless of intensity or strength, to provide a vertical profile over time .
Discrete return data sets are most common and are sufficient for most geoarchaeological applications but full-waveform data may offer some distinct advantages, particularly in densely vegetated areas, where it can increase the number of data points derived from the ground surface (Doneus et al., 2008;Lasaponara et al., 2011;Stott et al., 2015) while also informing understanding of vertical structure (Brolly et al., 2016).

| Data acquisition
Lidar data sets have been extensively collected by national governmental survey agencies and research institutions, many of which are freely available, providing an exceptional resource for geoarchaeological and archaeological projects. An extensive list of agencies or organizations that provide access to these lidar data sets can be found in  and Kokalj and Hesse (2017). In England, the Environment Agency's National Lidar Programme (NLP) provides high spatial resolution (0.5-2 m) elevation data for most of the country (Environment Agency, 2021). Surveys are principally undertaken during the winter months and are available as Open Government license data, which can be downloaded from the DEFRA Data Services Platform as point cloud and rasters (Department for Environment Food and Rural Affairs, 2021).
For projects that integrate airborne lidar collected by government agencies, the choice of instrumentation or data collection parameters is predetermined and, as a result, in some cases, it can be beneficial to commission a bespoke lidar survey from a private company, when specific data acquisition is required. Whilst this can be expensive, as with most technologies, this type of data is becoming more affordable (Bluesky International Ltd., 2022). The cost of lightweight lidar sensors that can be mounted on UAV systems (drones) is also coming down, but these range from surveygrade instruments to repurposed sensors originally designed for the automotive industry. As the characteristics of these instruments vary, numerous data acquisition parameters require consideration to ensure their effective use, including range accuracy, beam divergence (footprint size and shape), wavelength, number of returns recorded and accuracy of the GPS and IMU (Kellner et al., 2019). There are also numerous complexities surrounding mission planning and subsequent data processing to ensure the generation of accurate terrain models (Casana et al., 2021). However, archaeological applications of this technology are increasing and perform well in a variety of conditions (Casana et al., 2021;Risbøl & Gustavsen, 2018;van Valkenburgh et al., 2020). The main advantages are the increased flexibility, low flight altitudes, small laser footprint and the advantages of a farreaching field of view, which ultimately enable a higher point-cloud density (Risbøl & Gustavsen, 2018). Previous research has also shown that this higher spatial resolution enables an improved definition of the physical properties of topography (Resop et al., 2019). This may also provide improved capabilities within wooded environments, but since most sensors provide discrete return rather than full waveform data, this may be relatively limited, as only a small number of the last returns may relate to the ground.

| Standards and guidance
There is currently no standard approach to the integration of lidar in archaeological practice but there are several overviews of approaches to airborne data (Crutchley & Crow, 2010;Historic England, 2018;Kokalj & Hesse, 2017;Opitz & Cowley, 2013). The

Europae Archaeologiae Consilium (EAC) Remote Sensing Working
Group is, however, currently preparing guidelines for the use of lidar in heritage management across Europe, which is due to be published in 2024 (Europae Archaeologiae Consilium, 2022). In addition, a recent research article has provided a comprehensive review of archaeology-specific workflows for airborne lidar data acquisition, processing and interpretation . This highlights a series of common steps and sub-processes centred around pointcloud processing and derivation of products, followed by archaeological interpretation, dissemination and archiving. However, the specific considerations and workflow of each project will depend on the nature of the landscape in question and whether it involves commissioning the acquisition of new data or integration of existing data.

| Data products
Regardless of whether lidar data are captured from an airborne or UAV-based platform, the primary data set for lidar is a point cloud, which comprises information regarding the X, Y and Z coordinates of the returns and additional attribute information such as GPS time, intensity and scanning angle . These points can be classified into different object types such as ground, low, medium and high vegetation classes using an automated or manual process , although manual classifications are rarely done on a large scale. Ideally, point-cloud data will be provided in a.LAZ format, which must be decompressed into a.LAS format for use within GIS software. These files are not always available, can be very large and can be efficiently converted through the open-source LAZ converter provided by LAStools (rapidlasso GmbH, 2021). These data can then be used in the creation of different DEMs. Point-cloud data allow the data quality and effectiveness of the classification to be CRABB ET AL. | 423 interrogated/checked (Kokalj & Oštir, 2018;White, 2013). They also enable the gridding of Digital Surface Models (DSMs) and Digital Terrain Models (DTMs) by only considering the first and last returns, but preprepared DSMs and DTMs can also be downloaded in a standard ASCII format.
DSMs are created from the lidar pulses returned to the sensor relating to all ground surface objects, whereas DTMs are created from the last return classified as ground by filtering out surface objects (Environment Agency, 2021). For the DTM, manual filtering can also be undertaken to improve the automated classification routines and produce the most likely ground surface model. While surface objects such as modern buildings and vegetation are useful to orientate data in the landscape, they can be distracting for palaeolandscape reconstructions associated with geoarchaeological deposit models and DTMs are, therefore, most extensively used, as this provides a refined bare earth (ground) model. The intensity data for each area can also be used to provide a measure of the amount of laser light from each laser pulse reflecting from an object (Environment Agency, 2021). This reflectivity is a function of the near-infrared wavelength used and the material incident upon, but also the angle of incidence, flight altitude and the number of returns that can be used as a proxy to analyse the reflectivity of the surface (Historic England, 2018). Although previous applications of these data within an alluvial context have proved inconclusive, it can aid interpretation, particularly relating to the location of organic and waterlogged deposits ,

| Processing software and tools
Lidar data can be processed through a variety of proprietary or opensource GIS and remote sensing software (e.g., ArcGIS, ENVI, GRASS, SAGA and QGIS). These enable the generation of standard lidar products from the original point-cloud data and there are numerous tutorials and guides available (e.g., Davis, 2012). In addition, numerous visualization methods are also integrated or can be installed as an additional plugin but some methods may require bespoke model building or equations. Given that most geoarchaeologists will be familiar with GIS, this is perhaps the most convenient environment for processing lidar data. In addition, there are freely available toolboxes (e.g. RVT, LiVT and Whitebox tools) that are dedicated to the production of lidar visualizations and facilitate a wide range of advanced visualization procedures (Hesse, 2013;Kokalj, Zakšek, Oštir, Pehani, et al., 2013).

| CONTEMPORARY APPROACHES TO LIDAR VISUALIZATION
Lidar-derived DEMs can be difficult to interpret in their raw data form but visualization techniques can be used to improve the definition of features of interest or transform the data into other physical quantities such as degree of slope or aspect (Kokalj & Hesse, 2017). While the simple visualization of lidar data within standard GIS software converts the original data into new values by applying a histogram stretch that distorts but does not store the original data, other visualizations transform the data values, which can then be stored as a new independent raster data set. A growing number of these transformation techniques that vary significantly in complexity but can be broadly grouped into three categories that use similar approaches are provided as follows: 1. Illumination techniques.

Blending.
While there have been several empirical assessments of these techniques (e.g. Bennett et al., 2012;Challis et al., 2011b;Devereux et al., 2008;Štular et al., 2012;Thompson, 2020), it can be very difficult to establish the most appropriate technique for a particular site or landscape. There has also been a small number of more objective evaluations for a limited range of archaeological features (e.g., field systems and burial monuments) and landforms such as palaeochannels and landslide scars (Guyot et al., 2021;Mayoral et al., 2017;Notebaert et al., 2009) and when new visualization methods are proposed, they are normally assessed in relation to pre-existing approaches (e.g., Doneus, 2013;Hesse, 2010;Orengo & Petrie, 2018;Zakšek et al., 2011). However, the results of these studies have emphasized that there is no single visualization, or combination of visualizations, that consistently performs well in all situations. Despite this, it is often possible to identify techniques that are most suited to specific types of terrain. For example, Local Relief Models (LRMs) are commonly found to perform well in low-relief areas, whereas the Sky-View factor may be more appropriate in areas of steeper ground Mayoral et al., 2017).
Since alluvial landscapes contain rich archaeological records closely linked to a range of landform assemblages, it is challenging to select appropriate visualization methods that can enhance the visibility of the full range of resources that are potentially present. Many of the approaches to lidar visualization are designed to highlight small-scale archaeological features, but larger-scale natural landforms such as broad palaeochannels, levees and gravel islands may only exist as slight topographic variations that extend over hundreds of metres (Orengo & Petrie, 2018). As such, there is a need to establish which techniques are most appropriate for the identification of geoarchaeological features within alluvial environments, as well as consider how they might be integrated within a deposit modelling framework. The remainder of this paper provides an overview and quantitative evaluation of a range of commonly used lidar visualization techniques in archaeology and geomorphology and highlights those that are likely to be most applicable from a geoarchaeological perspective.

| What makes a good visualization?
Within geoarchaeological deposit modelling, the purpose of using lidar is to define the surface expressions of geomorphological landforms and relate these to more deeply buried deposits of geoarchaeological interest across a site or landscape, which can then be interpreted in terms of their archaeological and palaeoenvironmental potential. In essence, this is not dissimilar to archaeologists working in dryland environments to identify topographic features associated with past human activity or geomorphologists studying landforms and landscape evolution; rather, it is a combination of the two. However, as discussed, the physical shape and extent of archaeological features and geomorphological landforms are very different, ranging from small to large scale. Consequently, effective geoarchaeological visualizations need to account for this wider range of different feature types.
More generally, the visualization of lidar data should follow good practice principles required for any cartographic or raster data set (e.g., Borland & Taylor Ii, 2007). Grey or colour scales should be easily understandable and representative of the order of data (Kirk, 2019).
However, if the data values have a zero point, a dichromatic scale may be more appropriate to emphasize deviation from this (Campbell & Shin, 2011). It is also sometimes useful to use pseudo-realistic multipart colour schemes, but the number of contingent colours should not be excessive so that they cause confusion. Some colour scales, such as rainbow palettes, should be avoided as their nonsequential luminosity intensities can introduce false borders or artefacts in the data .
Beyond increasing image contrast for the human perception of archaeological features and natural landforms, lidar visualizations are also utilized to improve automated, or semiautomated, object detection (Davis, 2019;Verschoof-van der Vaart & Lambers, 2022).
In geomorphology, these procedures are increasingly critical for providing a quantification and recognition of landforms that possess unique morphometric characteristics (Evans, 2012;Jasiewicz & Stepinski, 2013;Lin et al., 2021;Wang et al., 2010). In archaeology, the focus of these procedures has also been more restricted to the detection of morphologically distinct features such as pits, linear features, mounds and other structures (e.g., Cerrillo-Cuenca, 2017; Freeland et al., 2016;Niculiță, 2020;Schneider et al., 2015;Trier & Pilø, 2012). However, algorithms for multiple and more complex feature types are increasingly being developed, particularly through the application of deep learning techniques such as convolutional neural networks (Bonhage et al., 2021;Bundzel et al., 2020;Meyer et al., 2019;Trier et al., 2019Trier et al., , 2021Verschoof-van der Vaart & Lambers, 2019;Verschoof-van der Vaart et al., 2020). As such, some visualizations are targeted toward ensuring that the broadest range of features is detectable by combining optimal aspects of various techniques (Guyot et al., 2021;. However, there is no agreement in terms of which visualization methods produce the best results for these approaches. Despite this, the principles of what makes a good visualization are largely the same, regardless of whether it is produced for an automated or manual interpretation. Thus, although this study does not attempt any automated procedures, the considerations presented here are important for any future research that may aim to utilize these methods within an alluvial setting. The default visualization of raster data sets within GIS software is a linear grey scale, with black denoting low values and white defining high values. A very simple method of improving the contrast of this is to constrain the image to a more appropriate data range, which can be achieved using a linear histogram stretch or manual saturation of extreme values (e.g., identifying the mean and range of elevation values contained within an area of interest such as a flood plain). 'Elevation' colour ramps are also widely used as they are very intuitive, where light blue relates to low-lying positions and darker red and white relate to higher areas. However, where the overall topography deviates from horizontal, it can be difficult to select the optimal elevation range for the full suite of landforms within a flood plain (Kokalj & Hesse, 2017).
One approach to overcoming the issue of identifying effective elevation ranges was developed by A. F. Jones et al. (2007). This comprised classifying multiple elevation ranges to enable the visualization of different landforms at different intervals (e.g., 4, 2, 1, 0.5 m, etc). This approach allowed the delineation of both large features, such as prominent terrace edges, and subtle features, such as shallow palaeochannels, which were mapped sequentially depending on the range of these intervals. Whilst this approach was successful, it also highlights the fundamental issue with such visualizations: that it is difficult to characterize the full range of features contained within a DTM of the flood plain using a single image. Moreover, it is also still subject to numerous biases of the landscape such as the inherited form and morphology of the flood plain profile and downstream slope.

| Illumination techniques
A wide range of image enhancement techniques has been developed for lidar, many of which explore the interaction between the landscape and a hypothetical light source (summarized in Figure 2).
The best known and well established of these is hillshading, where each cell is assigned a value (or shading) based on the illumination at a specified azimuth and elevation angle. Areas perpendicular to the light beam are most illuminated, while areas with an incidence angle equal to or greater than 90°are dark (Kokalj & Hesse, 2017). Under very low light source angles (below 10°), more subtle features are enhanced but the use of a single illumination can be problematic for any features that are aligned parallel with the light source (Davis, 2012). To overcome this, different illumination angles can be produced from multiple directions and displayed as an RGB composite image (Devereux et al., 2008). Then, Principal Components Analysis (PCA) can also be used to summarize the variance contained across these multiple (>3) hillshade directions (Crutchley & Crow, 2010). Within such images, high levels of the original data set variance are accounted for within the first three components (c. 99%), meaning that the first three-component bands presented as a false colour composite can be an extremely effective way of reducing data dimensionality.
Alternative illumination techniques such as Sky-view factor (SVF) and openness can also be used to overcome the directional problems CRABB ET AL. of hillshading (Kokalj & Hesse, 2017). Both techniques involve calculating the proportion of visible sky above (or below, as for negative openness) a certain observation point by computing the horizon angle in different directions to a specified radius (e.g., 10-20 m) (Kokalj & Hesse, 2017;Yokoyama et al., 2002;Zakšek et al., 2011). Openness can be provided in a positive (OPP) and   Slope models calculate the slope severity for each cell (calculated as the degree of slope) and darker shades typically represent more steeply sloping areas, with lighter areas corresponding to flatter terrain (Smith & Clark, 2005). Challis et al. (2011b) found slope gradient as the best visualization technique for the identification of most archaeological resources in a flood plain setting. However, a limitation of this method is that it is difficult to distinguish between convex and concave features, as slopes of the same gradient are presented with the same colour regardless of whether they are inclined or declined (Kokalj & Hesse, 2017).

Other topographic filters such as Local Dominance (LD) and Local
Relief Models (LRM) attempt to remove the influence of broad topographic trends to highlight more localized variations. Within LD images, the brightness of each pixel corresponds to the mean angle from which a virtual observer looks within a given radius, giving an impression of how a pixel dominates its local surroundings (Hesse, 2016). Simple Local Relief Models (SLRM) are very similar to Local Dominance but apply a low-pass filter to the original DEM, which provides a smoothed version only showing large-scale landscape forms. This is then subtracted from the original DEM to enhance small-scale features. Within a full LRM, a further processing step is applied where a purged DEM is created from the zero-metre contour lines in the SLRM, which is then subtracted from the original DEM to create a less biased elevation of small-scale features than the SLRM, as the elevations of features are relative to the surrounding landscape. However, the SLRM is faster to compute than LRMs and the visibility of the features of interest remains (Trier et al., 2021). In general, these techniques are well suited for very subtle positive relief features such as former upstanding field boundaries and slight depressions such as infilled ditches (Kokalj & Hesse, 2017). They have also been shown to be well suited to areas of relatively flat topography (Kokalj & Hesse, 2017;Mayoral et al., 2017)  An increasingly popular filtering method in geomorphological research is Relative Elevation Models (REMs), which were developed to facilitate examination to delineate fluvial processes and channel migration (Coe, 2022;Notebaert et al., 2009). They are sometimes referred to as Height Above River (HAR) models as they normalize the elevation of the active river channel by creating a detrended DEM and subtracting it from the original data set. REMs can be created using several different methods, including using bespoke cross-sections of the flood plain (J. L. Jones, 2006) or smoothing algorithms such as Kernel Density (Dilts et al., 2010) and Inverse Distance Weighting (IDW) (Olson et al., 2014). However, the principles behind the process are essentially the same, whereby the DEM is detrended by using the water surface of the present river channels to remove the influence of the downstream slope. In the IDW method, the analysis involves first extracting elevations along the channel to generate a detrended DEM using the IDW tool (Olson et al., 2014). Finally, the detrended DEM is subtracted from the original DEM, resulting in elevation values relative to the water surface of the channel. This produces positive values, except where there are low-lying zones (below the height of the river) within the flood plain. This has obvious merits for identifying upstanding alluvial landforms that may have a high archaeological potential (e.g., gravel islands or terraces) as any elevated parts of the flood plain will be more readily apparent.

| Blending
There have been several attempts to capitalize on the relative advantages of multiple techniques through image fusion and allow for the simultaneous display of distinct topographical features in a single (enhanced) image . This is potentially less onerous than a manual or automated interpretation of multiple images, but it can be difficult to identify exactly what topographic characteristics are displayed within the resulting visualizations.

Kokalj and Somrak (2019) developed a specific Visualization for
Archaeological Topography (VAT) using blending techniques to combine hillshading, slope, positive openness and sky-view factor.
It includes options for 'normal', complex and very flat terrain and can be very effective for small-scale features or local geomorphological characteristics, but larger-scale topographic variations that might relate to alluvial landforms may not be well represented (Guyot et al., 2021).
Another blending technique is Red Relief Image Maps (RRIMs), which were developed by Chiba et al. (2008) to overcome the limitations of openness by combining it with a slope gradient. To achieve this, positive openness (OPP) and negative openness (OPN) are combined using the following formula, which is sometimes referred to as the I-factor; The slope image is then presented in a red-colour scale and

| Visualization of archaeological resources in alluvial environments
Each of the above visualization techniques has advantages and disadvantages, and some are more appropriate than others for the study of geoarchaeological resources in alluvial environments. For example, some methods are known to be effective over low-relief terrain (e.g., SLRM) or are specifically designed for use within flood plain settings (REM). However, it is impractical and time-consuming to interpret multiple lidar visualizations, particularly when many of the techniques repeat the same information, highlighting features to a slightly different degree. While the use of blending techniques can be used to overcome this issue, the resultant images are often unintuitive for nonspecialists and difficult to relate to the topographic characteristics they were originally derived from. Consequently, it is useful to identify a smaller set of demonstrably effective imageenhancement techniques for use within geoarchaeological projects.
To achieve this, a quantitative evaluation has been undertaken for a series of landforms identified within the Lugg and Wye Valleys of Herefordshire, UK. This is combined with qualitative statements and a discussion of their applicability and potential integration with geoarchaeological deposit models.

| The Lugg and Wye Valleys, Herefordshire, UK
The Lugg and Wye Valleys were selected for this analysis as they represent typical lowland flood plain settings within the temperate zone, and have been a recent focus for investigating the application of hyper-and multi-spectral data sets within an alluvial geoarchaeology context (Crabb et al., 2022). Both river systems have complex depositional histories, with closely related human-environmental interactions demonstrated by a palimpsest of archaeological records, which have been the focus of previous geoarchaeological investigations (Brown et al., 2005;Carey et al., 2017;Dinn & Roseff, 1992;Hemingway & Dinn, 1996;Jackson & Miller, 2011;Pears et al., 2020). This has established secure chronostratigraphic knowledge of the landform assemblages, in turn aiding geoprospection at a reach scale, which may be tentatively applied more widely.

The specific study areas comprise sections of the Middle and Lower
Lugg and the Middle Wye Valleys (Figure 4), each containing a range of alluvial landforms to enable the assessment of lidar visualization.

| Lidar data
The lidar data used in this analysis were downloaded as 5 × 5 km tiles from the UK Environment Agency NLP (Department for Environment

| Calculation of lidar visualizations
In total, 16 different visualizations have been calculated for each of the study areas. These have been created using the Relief Visualization Toolbox (RVT) or ArcGIS and associated plugins or toolboxes (Qiusheng, 2022; Whitebox Geospatial Inc). The specific input parameters are provided (Table 1) and wherever possible, they have been optimized for low-relief variations . For many of these, a search radius of 20 pixels was used to transform the data as this helped to enhance many of the features but also preserved some finer details of the palaeochannels, ridge and swale, and levees adjacent to the river. Although a larger operating window (e.g., 50 pixels) may provide better results for extremely large features, this was found to give a 'washed out' appearance to the smaller-scale features.

| Evaluation of lidar visualizations in alluvial environments
There is no consensus on the best method to evaluate the capability of lidar visualizations and approaches have varied from qualitative assessments of their performance (e.g., the number of features detected; Bennett et al., 2012;Thompson, 2020) to more empirical judgements based on the success of automated interpretation procedures (e.g., Guyot et al., 2021;Mayoral et al., 2017). A small number of studies have evaluated signal-to-noise ratios as a proxy for image contrast (Štular et al., 2012) (Swain & Davis, 1978).
The main aim of undertaking these separability analyses is to identify optimal images or assess the effectiveness of training data sets for the automatic classification of different land surface types (Tso & Mather, 2009). Even though automated procedures are becoming increasingly commonplace for archaeological applications of lidar data sets (Bennett et al., 2014;Sevara et al., 2016), it is often not clear why certain visualizations are chosen over others. This is potentially because, until relatively recently, the number of available visualization techniques was limited, and therefore such analyses were not required. Moreover, most evaluations of the quality of lidar data are often more concerned with spatial accuracy (Bakuła et al., 2017), registration of features (Arnold et al., 2006), or classification of point clouds , as opposed to ensuring maximum feature contrast or enhancement. However, given the increasing diversity of lidar visualization techniques and the availability of open-source software tools, it is necessary to consider such issues more rigorously, and separability measurements provide one method of achieving this.
The M-statistic is a relatively simple separability metric that calculates the difference between the means μ ( ) of two ROIs, normalized by the sum of the standard deviation σ ( ) using the following formula: This provides a measurement of the separation between the histograms of two classes, with M > 1.0 indicating good separability and M < 1.0 indicating poor separability (Kaufman & Remer, 1994). As archaeological features and natural landforms are often very subtle, they will typically fall below this threshold, but the metric can still be used as an evaluative tool. Although it does not enable the separability of features within multiband images, their constituent parts (individual RGB bands) can be assessed. While other separability measurements (e.g., JM distance or transformed divergence) are better suited to multiband images, it is difficult to compare the results to the assessment of single-band images provided through the M-statistic (Crabb et al., 2022;Richards, 2013). However, in contrast to these and other common statistical analyses of variance and difference (ANOVA/t tests), the M-statistic does not account for sample size.
Although this can influence the results, it is of less concern for larger data sets, as the influence of any outliers is reduced. As such, the M-statistic was selected to assess the capability of the 16 visualization techniques undertaken as part of this analysis, but it was ensured that each ROI consisted of >1000 pixels.   Figure 5. Each ROI pair has been assigned a colour, with the surrounding background response highlighted in grey (Table 2).

| Quantitative assessment
The results of the M-statistic calculations are shown in Figures 9 and 10. The cumulative scores in Figure 9 illustrate which visualization methods work best overall, across each of the feature types covered by the study areas. In contrast, Figure 10 shows the variable effectiveness of each technique for these different feature types (ROI). Perhaps one of the clearest and most important aspects of this is that the original (unaltered) DTM provides some of the best separability results of all the visualizations, particularly for larger features. For example, both ROI 3a/3b and 5a/5b produce scores above the threshold (M > 1) and 1a/1b is also relatively high. These all represent larger-scale upstanding landforms (e.g., river terraces and gravel islands). In contrast to this, the smaller-scale features such as palaeochannels (ROI 2a/2b and 4a/4b) and ridge and swale (6a/6b) achieve lower separability scores, but the original DTM is by no means the worst-performing visualization analysed.
The best-performing illumination technique is SVF, which provides effective enhancement of the low-lying palaeochannel at ROI 2a/2b and the ridge and swale at 6a/6b. This is also the case for both the positive and negative openness images, with OPP providing a slightly higher score for 6a/6b. However, in general, all the illumination techniques 6a/6b). However, they also work relatively well for some larger-scale landforms, although these do not exceed those achieved by the unaltered DTM.
The slope gradient image does not provide a visual enhancement of most of the ROI pairs, and only the palaeochannel at ROI 2a/2b is improved. In fact, it is less effective than many of the illumination techniques, excluding hillshading.
The best-performing technique of the topographic filters and other visualization types is the REM images. REM produces high separability scores for most of the ROI pairs, with 1a/1b and 5a/5b exceeding the (M > 1) threshold. Both of these relate to gravel terraces with adjacent palaeochannels and are associated with large- While these can be very effective in a more diverse landscape, they are evidently less effective in a narrow, low-relief flood plain setting and may, therefore, offer limited benefits for the identification of geoarchaeological resources. However, in scenarios where the flood plain is less constrained, such as the Mississippi (Chamberlain et al., 2020) or  (Bennett et al., 2012;Štular et al., 2012). In addition, many of the topographic filters also attempt to reduce the effect of broader topographic trends to enhance localized relief variations (Guyot et al., 2018;Hesse, 2016Hesse, , 2010. Thus, while this is indeed very effective for small-scale landforms, it has a negative effect on the identification of larger-scale features such as gravel terraces or islands, which are of significant interest within a geoarchaeological context since they are often the focus of settlement and other human activities. It is challenging to select a single visualization technique that enhances the visibility of every resource that might be present within alluvial environments. This is consistent with previous empirical assessments (e.g., Bennett et al., 2012;Challis, Forlin, et al., 2011;Devereux et al., 2008;Štular et al., 2012;Thompson, 2020) and objective evaluations (Guyot et al., 2021;Mayoral et al., 2017), which have considered a range of terrain types. However, within specific lowlying flood plain settings, with minimal relief, it is possible to make more specific recommendations on which image enhancement techniques produce optimal definitions of geoarchaeological resources, which are summarized in Table 3. | 435 combination of visualization techniques. As REMs are tools specifically intended to enhance the investigation of geomorphological and hydrographic aspects of river systems (Notebaert et al., 2009;Olson et al., 2014) and SLRM has been previously found to perform well in low-relief areas, Mayoral et al., 2017) Examples of the most effective visualization methods are shown in Figure 11, which highlight the location of the ROI studied here.
These data can then be used to create a deposit model for the Lugg and Wye Valleys, where lower-lying (wetter) areas and palaeochannels are more likely to contain palaeoenvironmental resources (e.g. plant macrofossils, pollen, and other ecofactual material), whereas higher (drier) zones, relating to upstanding gravel terraces or islands, will unlikely contain such natural waterlogged remains, but were more attractive for a range of past human activities. These relatively simple statements can also guide the application of subsequent geoarchaeological procedures and/or archaeological mitigation strategies by defining areas where standard archaeological field techniques (e.g., field walking, shallow geophysical methods, and trial trenching) will be ineffective. Moreover, they may also be useful in determining techniques that have the greatest applicability at the prefieldwork, desk-based planning stage of any project such as establishing areas where the efficacy of aerial photographic analysis will be effective. Thus, the integration of lidar data can provide significant insights into the distribution of archaeological resources and can also provide a vital baseline data set from which to investigate the subsurface. However, in terms of integrating these lidar visualizations within the framework of deposit modelling, other factors must also be considered.
Geoarchaeological deposit models provide a visual representation of the spatial and stratigraphic relationships between subsurface sediments, archaeological features, and palaeoenvironmental remains (Carey et al., 2018 (Crabb et al., 2022) and geophysical surveys (Bates & Bates, 2016;Verhegge et al., 2016Verhegge et al., , 2021. To facilitate the effective integration of multiple, disparate data sets into deposit models, it can be useful to use consistent projections and datum values. As this can be achieved by simply using unaltered DTM lidar data sets, there is a strong case for incorporating more simplistic visualizations, particularly given that the unaltered DTM achieved very good M-statistic separability scores. However, it is possible that landforms can be missed if they are extremely subtle, as significant cross-and down-slope elevation changes may obscure their visualization. Moreover, for these reasons, unaltered DTMs can also be difficult to utilize over wider areas, and other techniques that account for the influence of general topographic trends (e.g., REM and SLRM) may be more reliable, particularly when studying larger alluvial flood plains.
While this study has focused on alluvial environments, it has implications for the application of lidar data where archaeological resources do not lie immediately below the modern ground surface but are buried beneath accumulated sediments (e.g., colluvial, aeolian, coastal, estuarine, and lacustrine deposits). Lidar has been used extensively in geoarchaeological research to study a wide range of different environments (e.g., Bowen et al., 2018;Carlson & Baichtal, 2015;Gregory et al., 2021;Lausanne et al., 2021) and many of these investigations would benefit from the integration of lidar within a deposit modelling framework to define surface evidence for any variations within these complex depositional zones. As such, it would be beneficial to expand this study to evaluate a larger number of landforms and a wider range of environments. However, the results of this research suggest that such an approach provides a robust baseline from which to investigate the subsurface and future research should, therefore, be directed toward ensuring its integration within a more diverse range of settings. However, currently, this is inconsistently implemented, and alternative visualizations beyond the simple display of DEM are rarely considered.
The focused literature search and review provided at the outset of this paper showed that for applications of lidar in geoarchaeological investigations of alluvial environments, approximately half included an interpretation analogous to a deposit model. In each of these cases, lidar data were integrated into the analysis to some extent, but only 18% use visualization techniques to enhance the visibility of resources. Although some visualizations may not always be appropriate or necessary, as this research has shown, they can help to better define a wide range of features and landforms.
Moreover, while there has been a barrier in terms of the technological expertise required to produce visualizations, there is now a wide range of open-source toolboxes and plugins available, which improve their accessibility (Hesse, 2013;Kokalj, Zakšek, Oštir, Pehani, et al., 2013;Qiusheng, 2022). Consequently, given the increasingly widespread and open-access nature of data sets, lidar data processed through the suggested methodologies presented here should form part of a standard approach to geoarchaeological deposit modelling, which has landscape evolution at its core. approach. However, the original, unaltered DTMs were also very effective and may enable better integration with deposit models as their values are more consistent with other data sets (e.g., intrusive methods, HER records, and geological mapping), especially when depths below ground level are required. However, lidar visualizations can help to better define the distribution of buried deposits of geoarchaeological interest across a site or landscape, which can then be interpreted and further investigated in terms of their archaeological and palaeoenvironmental potential.

| CONCLUSION
As alluvial environments offer the unique challenge of combining rich, well-preserved archaeological and palaeocological records with conditions where archaeological prospection methods are ineffective, appropriate visualization of lidar data has significant potential to aid their investigation. Consequently, it is argued that these techniques should become a standard part of geoarchaeological deposit modelling and future research should aim to apply these methods more widely within river flood plains and other complex depositional zones.

ACKNOWLEDGEMENTS
The authors are also grateful to Dr. Niall Burnside, Richard Higham (University of Brighton), and Robin Jackson (Worcestershire Archaeology) for their assistance with and discussion of aspects of this research. We are also grateful to the two anonymous reviewers for their comments whose comments helped to improve the quality of this paper. This work was supported by the UK's Engineering and