Combining Remote Sensing Approaches for Detecting Marks of Archaeological and Demolished Constructions in Cahokia’s Grand Plaza, Southwestern Illinois

Abstract

Remote sensing data are increasingly being used in digital archaeology for the potential non-invasive detection of archaeological remains. The purpose of this research is to evaluate the capability of standalone (LiDAR and aerial photogrammetry) and integration/fusion remote sensing approaches in improving the prospecting and interpretation of archaeological remains in Cahokia’s Grand Plaza. Cahokia Mounds is an ancient area; it was the largest settlement of the Mississippian culture located in southwestern Illinois, USA. There are a limited number of studies combining LiDAR and aerial photogrammetry to extract archaeological features. This article, therefore, combines LiDAR with photogrammetric data to create new datasets and investigate whether the new data can enhance the detection of archaeological/ demolished structures in comparison to the standalone approaches. The investigations are implemented based on the hillshade, gradient, and sky view factor visual analysis techniques, which have various merits in revealing topographic features. The outcomes of this research illustrate that combining data derived from different sources can not only confirm the detection of remains but can also reveal more remains than standalone approaches. This study demonstrates that the use of combination remote sensing approaches provides archaeologists with another powerful tool for site analysis.

1. Introduction

The application of non-invasive/non-destructive digital archaeology methods is vital to the ongoing effort to discover and identify potential archaeological remains. It is very difficult to reconstruct the archaeological record if an archaeological area is changed or destroyed by anthropogenic and destructive activities, especially when there is a lack of geospatial documentary data [1]. Although excavation is commonly used to identify remains, the process inherently removes evidence and can damage materials [1,2]. As a result, archaeologists often leave part of a site unexcavated to preserve it for future research using advanced techniques.

In the past 30 years, there has been a gradual shift from analogue to digital approaches in archaeology [3,4]. This paper makes a contribution to the ongoing development of digital approaches by investigating the effectiveness of methods that combine data from different digital sources.

While invasive methods continue to play a role in archaeological investigations, archaeologists are increasingly using geophysical and remote sensing methods [5,6,7]. Recently (since 2018), artificial intelligence (AI) approaches have also been used in archaeology based on the analysis of remote sensing data [8,9,10]. A very wide range of studies has been carried out in digital archaeology based on various remote sensing platforms [11,12,13,14,15,16,17,18,19,20]. The use of remote sensing data allows archaeologists to identify archaeological earthworks [11,12,13,14,15,21] and enhances the understanding of a known site/feature through better documentation. Such digital methods, e.g., photogrammetry and light detection and ranging (LiDAR), provide a non-destructive means of securing the “digital preservation” of archaeological areas as part of the process of preserving information about sites that are at risk of damage from various drivers (e.g., climate change, war, or development) [5].

In terms of quality and efficiency, the development of photogrammetry and LiDAR has led some studies to recommend applying one technique over another [14,15]. More recent studies, e.g., Liang et al., Filzwieser et al., and Luhmann [16,17,18], have found that combining both techniques is likely to boost the benefits of the acquired data to generate consistent and, to some extent, complete results. Hence, these methods could complement each other to enhance 2.5D and 3D models of archaeological areas [19,20]. The enhancement of 2.5D/3D models is conducted by combining various datasets through different approaches, such as data integration from the same sensor [22,23] and data integration from different sensors [18,24].

Data integration from the same sensor is an approach that has been applied in several archaeological studies such as [22,23,25]. The aim of this approach is to combine two or more different raster layers derived from the same source (photogrammetry or LiDAR). This integration is based on visual analysis techniques (VATs). VATs are techniques, such as hillshade, gradient, aspect, and the sky view factor (SVF), used to generate a realistic 2.5D raster from digital surface models (DSMs)/digital terrain models (DTMs). Each raster highlights traits of topographic surfaces. VATs are used to detect and interpret topographic features and to enhance the understanding of archaeological sites. Therefore, these techniques can be used towards the successful detection of archaeological features. The limitations of these approaches are mainly associated with illumination, raster distortion, and filtering; more information about these techniques is demonstrated in several archaeological studies, such as [26,27].

Previous studies such as [25,28,29,30] found that the edges of the archaeological features of the Barwhill in Scotland [30] and Beaufort County, South Carolina [28], were emphasised through red relief image maps (RRIMs). An RRIM is a VAT based on multiple layered topographic data, i.e., gradient and differential topographic data [24]. This raster is normally applied to archaeological sites since it provides a clearer and less-distorted view of topographic changes than the standalone VATs. Kokalj et al. [23] corroborated the conclusions of previous studies, suggesting the enhancement of the existing individual VATs to improve archaeological prospection and avoid missing possible remains. To enhance the visibility of detected remains, they created open-access relief visualisation tools (RVTs) to integrate various raster images derived from various fine LiDAR data (50 cm/pix and 25 cm/pix) from different archaeological areas. The study by Inomata et al. [25] was in line with previous studies such as Davis et al. [28] as they compared various VATs (e.g., hillshade, gradient, RVT, and RRIM) derived from LiDAR data (1 m/pix spatial resolution) under various conditions. The results of this study showed that the edge of marks is relatively more emphasized in RRIMs than in other applied VATs.

The second approach involves combining data from different sensors; this approach has been implemented in previous archaeological research, e.g., [18,24,31,32,33,34]. The concept of this approach is to combine two different remote sensing datasets, which are derived from different sources, after processing each one separately [35,36]. This includes the integration of 3D-to-3D dense cloud models [18,37,38] and 2D-to-2D raster images (photogrammetry and laser data) [18,24,31] to generate a new, enhanced, integrated dataset. Gunieri et al. [39] used photogrammetric and terrestrial laser scanning (TLS) data independently to create 3D models. After geo-referencing, both models were combined to generate a 3D representation of the area of interest (AOI).

A wealth of archaeological work has emphasised the importance of data combination in digital archaeology [23,37,39]. While a significant number of researchers have adopted the combination (integration/fusion) of TLS and photogrammetric point clouds for digital preservation [18,37,39,40,41,42], studies that combine LiDAR and photogrammetric data to detect archaeological remains are remarkably lacking [31]. Kokalj and Somrak [23] also emphasised the need to improve the detection of archaeological features by applying integrated approaches. This article, therefore, combines airborne LiDAR data with photogrammetric datasets to generate a new dataset in order to investigate whether the new data can enhance and improve the detection of archaeological remains at the Cahokia Grand Plaza study site in comparison to a standalone approach. We present results that were initially obtained from standalone approaches in order to critically investigate the enhanced potential of the new approach—the integration and fusion of LiDAR and photogrammetric data—in revealing a relatively greater level of detail (LOD) of the same archaeological area. Free, open-access LiDAR data are used for the AOI in this study as they are commonly used in archaeological studies, e.g., [2,28,43,44]. Drone photogrammetry is also used due to its cost-effectiveness compared to airborne photogrammetry. Ultimately, the aim of this research is to examine the capability of standalone and combination remote sensing approaches in enhancing the prospecting and interpretation of archaeological data. The approaches presented in this paper address a significant outstanding question related to the application of remote sensing in digital archaeology: which remote sensing combination approaches are most likely to improve the prospecting and interpretation of archaeological data? This question is divided into three sub-questions:

• How can standalone approaches be successfully combined to develop a practical, supported procedure for the capture and representation of archaeological data?
• What are the merits and limitations of individual sensors in observing and detecting archaeological sites?
• Which remote sensing combination approaches provide the most LOD in comparison to standalone approaches?

2. Materials and Methods

This section comprises an introduction to the study area and the rationale for its selection, as well as a review of previous studies that conducted excavation works at the AOI. Demonstrating previous discoveries at the AOI is important to distinguish between the previous findings and those originally reported in this paper, which were produced using combined remote sensing approaches. This section also includes a discussion of data acquisition (LiDAR and aerial images) and the standalone and combination approaches applied for detecting archaeological remains. The flowchart below (Figure 1) displays the main steps accomplished in this study.

2.1. Study Area: Cahokia’s Grand Plaza

Cahokia (Figure 2) takes its name from the Indigenous tribe who inhabited Cahokia city when the first French voyagers arrived in the late 17th century [45]. An ancient area, Cahokia was constructed by the Mississippians and is located in southwestern Illinois, USA (38.6551°N, 90.0618°W) [45,46]. Cahokia was the largest settlement of the Mississippian culture, who lived in the southeast of the USA from the Mississippi River to the shores of the Atlantic [47]. The city was developed for its time and densely populated (between 10,000 and 20,000 people). There was a reduction in the population two centuries later, which archaeologists theorise may have been due to drought, climate change, disease, or war [48]. The city contains the largest pre-Columbian earthen mound in North and South America. Up to 120 mounds have been recorded; however, some mounds were demolished or ruined by construction and agricultural purposes, leaving 80 mounds among them. These include Monks Mound, the largest mound on the North American continent (Figure 3). Monks Mound measures 30.5 m in height and spans an area of approximately 56,655.99 m² [2]. Nonetheless, there are no written records from the site, and its formation is still somewhat of a mystery [31,48].

The Grand Plaza (approximately 50.097 ha—measured in QGIS (version. 3.16), based on the structure from motion (SfM) mosaic imagery) is located between Monks Mound and the Twin Mound markets, gatherings, and ceremonies. Excavations were conducted several times at the study site. For instance, several remains were discovered in 1971, such as incised sandstone tablets on the east side of the Monks Mound and several watchtowers/fortifications surrounding the mounds. These fortifications were probably built by inhabitants in response to external threats (such as regional warfare) that had not previously existed [49]. Furthermore, archaeologists discovered potential constructions in front of Monks Mound and others located on the Plaza edges in 1997 [50]. In the early 21st century (from 2002 to 2010), surveys determined the exact location of copper tools that were originally discovered in 1950, traces of residential areas/ temples with wooden poles on top of the mounds, and flint and stones [51]. Therefore, many archaeological remains were already revealed in the AOI through the application of invasive methods. The reasons for selecting this study site are as follows:

• This case study is an extension of the existing study led by researchers Vilbig et al. (2020) [2] at Saint Louis University. They compared the analysis of standalone approaches (LiDAR and photogrammetry) at Cahokia Mounds and found that the digital models derived from photogrammetry provided comparable archaeological detail to LiDAR data. Testing the proposed method (standalone and integration/fusion approaches) of this study and comparing the findings with previous studies is vital to obtaining more archaeological data from the study area;
• The ancient city of Cahokia was abandoned in the 1400s and the reasons are still ambiguous as there are no contemporaneous records from this area. All the information received for this particular site is based on archaeologists’ hypotheses. Additionally, the outcomes of employing various remote sensing approaches are likely to suggest insights into appropriate applications for revealing new archaeological information.

2.2. Remote Sensing Data Acquisition

Two remote sensing datasets were employed in this research: (i) airborne LiDAR data and (ii) aerial-image-based photogrammetry. These datasets were used to investigate the potential of the integrated and fused datasets in enhancing the detection of archaeological remains.

2.2.1. LiDAR Datasets

The raw, airborne topographic LiDAR data for Cahokia’s Grand Plaza were captured in 2012 by a geospatial firm, Merrick & Company (https://www.merrick.com/about-us/) (accessed on 15 September 2021). The data were captured using a Leica ALS50II LiDAR Sensor for the Prairie Research Institute at the University of Illinois. The point density was 39.5 points/m² and the horizontal and vertical accuracy of the raw data was 1 m, calibrated by Merrick & Company [2]. The LiDAR epochs (the return pulses) contain information about the topographic features, constructions, vegetation, and water, which was grouped into 12 different classes that correspond to a variety of terrain surfaces. The data were obtained for the Illinois State Geological Survey (ISGS) (https://isgs.illinois.edu/about) (accessed on 15 September 2021), and the LAS (LASer) tiles for St. Clair County were downloaded from (https://clearinghouse.isgs.illinois.edu/data/elevation/illinois-height-modernization-ilhmp accessed on 15 September 2021). The LAS files were then processed into a digital model.

The workflow to create a bare-earth digital terrain model (DTM) from LAS files was implemented in ArcGIS Pro 2.6.0 (https://pro.arcgis.com/ accessed on 16 September 2021). A LAS file is a file format mainly used to store three-dimensional (3D) LiDAR point clouds; hence, the workflow comprised generating LAS datasets following recommendations from the previous studies [2,52]. Hence, producing these datasets was important to examining the LAS files and selecting appropriate classification codes. In this study, the ground class was the only code selected to represent the terrain features and to create a DTM of Cahokia’s Grand Plaza with a 0.50 m spatial resolution. LiDAR covers a much wider footprint when compared to aerial photogrammetry. In this study, we applied both techniques to reveal archaeological information about Cahokia’s Grand Plaza. For this reason, LiDAR data forthe AOI were extracted to be within the same scale of the study site as the photogrammetric DTM (approximately 50.097 ha). Both digital models were projected and geotagged at the same projection and coordinate system (NAD83-NSRS2007/Illinois West ‘ftUS’—EPSG:3531).

2.2.2. Photogrammetric Datasets

The final raw, unpiloted aerial vehicle (UAV) data were obtained from the research by Vilbig et al. [2]. Additional details are contained therein. Images were captured using a DJI Phantom 4 Pro drone equipped with an fc6310 digital camera (5472 × 3078 pixels). The study site was surveyed with a programmed flight using the open-source Pix4D Capture app (https://www.pix4d.com/product/pix4dcapture accessed on 3 October 2021). This app was used as a dynamic control supplement to plan flight missions. The flight height was 80 m over the study site, with a minimum overlap of 70% to generate a fine spatial resolution of 0.23 m. The drone was flown over the study site for 30 min for individual missions. Three flights were conducted to capture 1201 aerial images for the entire AOI.

Following the data capture of the UAV images, the structure from motion and multi-view stereoscopic (SfM–MVS) processing phase was implemented in this study using a high-performance computer (128 GB, AMD Ryzen 9 3900XT 12-Core Processor 3.79 GHz) on Windows 10. Several computer programs can run the SfM–MVS method (e.g., Agisoft Metashape, and Recap). In this research, Agisoft Metashape Professional software (v.1.6) (https://www.agisoft.com/ accessed on 20 July 2020) was used for its efficiency and effectiveness in generating accurate dense point clouds from aerial images [3,53,54,55]. The SfM–MVS processing was implemented in this research following previous studies such as [18,25,56,57]. The scale-invariant feature transform (SIFT) algorithm was used to reveal the tie points of the images, match common points, and triangulate point clouds. More than 97% of the original UAV images were aligned and applied in the 3D reconstruction. The remaining UAV images did not align due to an inadequate recognition of tie points in the common image features and were not used in the modelling process. Following point cloud triangulation, the bundle block adjustment (BBA) algorithm was executed to enhance the camera positions and improve the 3D reconstructions [58,59,60]. Subsequently, a 3D mesh was created based on the sparse point clouds of the scene. The fc6310 camera has a rolling shutter, so rolling corrections were used through camera calibrations. We used the following lens coefficients for self-camera calibration to optimise the accuracy of the geometric reconstruction: aspect ratio and skew parameters (b1, b2), principal point (cx, cy), focal length (f), radial distortion (k1, k2), and tangential distortion coefficients (p1, p2). Following the production of sparse points, dense point clouds were produced by applying the MVS method based on the aligned images and camera positions [61]. Approximately 3 million (3,138,269) points, with altitude errors of 2.59 m and a spatial resolution of 0.23, were created by processing the UAV images of the AOI. The outputs of the SfM–MVS processing were textured meshes, orthomosaics, and digital models (46.3 points/m² point density). These outputs were exported for further analysis to reveal archaeological remains.

2.3. Standalone Detection Approaches

The digital models derived from the LiDAR and photogrammetric data were used to generate various visualisation raster images in a free and open-source platform, QGIS (version. 3.16) (https://qgis.org/en/site/forusers/download.html, accessed on 20 July 2020). Examples of the visualization techniques applied in this study are hillshade (shaded relief map), slope image, and sky view factor (SVF), which could contribute to the detection of topographic information [53,62]. These VATs were used in previous archaeological studies, as is illustrated in Section 1. In this paper, the purpose of using these well-known VATs was to demonstrate how the new, original combination (integration and fusion) approaches applied in this study can overcome the limitations of standalone raster images in revealing marks of demolished structures in the same archaeological area.

The hillshade technique was produced in this study using the following illumination parameters, which are suitable for multiple visualisation analyses: azimuth (315°), altitude (45°), and Z factor (1) [27,53,55]. These parameters were applied to the LiDAR and the photogrammetric digital models to create shaded-relief raster images of the AOI. The purpose of generating various raster images was to verify the visibility of the topographic detail of the AOI. Another VAT created was the slope technique. This technique was produced using a 3 × 3 neighbourhood, as it is executed in most GIS packages [27,29]. The slope raster was generated in degree units.

The third raster successfully created in this study was the SVF. The SVF was generated by applying specific parameters, i.e., the number of search directions (Dn) and the maximum search radius (Rm). These are two vital parameters in determining the SVF raster image. The Dn is generally set based on either the sector method or the multi-scale method. The sector method was applied here since the multi-scale method generates pyramid layers, which creates less accurate raster images than the sector method [54]. The Rm was set to 15 and 25 pixels, following the recommendation from the study by Somrak et al. [55], as the topographic detail in the SVF raster is enhanced when the Rm value is between 10–30 pixels and the Dn value is more than the Rm value. However, the computation time increases significantly when the Rm value is above 50 pixels [56,57].

2.4. Combination Detection Approaches

This section is divided into two sections: data integration from the same sensor (Section 2.4.1) and data integration and fusion from different sensors (Section 2.4.2). This section addresses the first part of the research question and focuses on combination approaches. The RRIM technique (using the parameters applied in Section 2.4.1) was already tested in previous studies, e.g., [22,28,29,59]. The purpose of applying this technique was to investigate how new combination approaches (Section 2.4.2) could improve the detection of archaeological remains in comparison to well-known techniques (standalone and RRIM) in the same archaeological area.

2.4.1. Data Integration from the Same Sensor

Multiple visualisation raster images derived from the individual sensors were combined in this research to increase the visibility of archaeological remains and improve the interpretability of the remote sensing data for Cahokia’s Grand Plaza. Several blending modes (e.g., multiply) can be implemented to integrate multiple raster images. The multiply mode was applied in this study, as it treats every raster separately, integrating the top raster image (X) with the bottom raster (Y) [26]. Integrating any raster with a relatively darker layer normally creates a dark raster; this means that the outcome raster images from the multiply mode are relatively darker than the standalone raster images.

Based on the multiply mode, two RRIMs were created in this study: one from the LiDAR data and one from the photogrammetric data. Both RRIMs were generated by integrating the gradient raster with the topographic layers (positive and negative openness) derived from the DTM, as demonstrated in previous studies, e.g., [31,59]. The differential openness (D) was created by applying certain parameters: the radial limits were 10 m, and the number of sectors was set at 8 [60]. We applied the sector method, as it is less likely to generate pyramid layers and can provide more accurate openness parameters compared to the multi-scale method [60,63]. To execute the RRIMs, a topographic openness raster (D) was used to determine the positive and negative openness (P_o and N_o) [59]. When D is the differential openness raster, P_o represents the positive openness, which highlights topographical concavity, and N_o signifies the negative openness, which highlights topographical convexity [28,63]. Hence, the RRIM was created by merging the D in a white-to-black raster with the red gradient layer [31,60].

2.4.2. Data Integration and Fusion from Different Sensors

Two new pipelines for combining two different datasets, namely the LiDAR and photogrammetric data for Cahokia’s Grand Plaza, was applied herein. This study refers to the integration method as a combination of two different raster images (2.5D raster) derived from the range-based LiDAR and the image-based methods after processing the individual datasets separately. In comparison, the term fusion approach denotes the combination of 2D images together to create a fused model. Both techniques match common features/points between aerial images and laser data to achieve a successful combination.

Due to the limitations of the standalone approaches [18,61], we argue that combining raster images generated from the LiDAR and photogrammetric datasets creates a new, single, integrated raster that includes all the distinguished archaeological features highlighted (or shadowed) on single images. The new concept of this approach is to multiply two raster layers created from different data sources, i.e., to integrate individual VATs derived from the LiDAR with the corresponding VAT derived from photogrammetry to generate new, integrated VATs. An example of this combination is the integration of the SVF from the LiDAR data with the SVF from the photogrammetric datasets. The new, integrated raster images were then compared with each other regarding the topographic detail and improvements in the detection of archaeological remains. The purpose of this approach was to boost the existing raster layers and thereby enhance the detection of archaeological remains. However, we avoided the integration of more than two raster images (e.g., RRIM derived from the LiDAR data with the corresponding raster from the SfM method), as this is not an efficient process and features are likely to be misdetected [23].

Following the registration of the individual rasters, six new raster images were created in this study. These new raster images are the integrated hillshade, integrated gradient, integrated SVF, and SVShade (I) (SfM SVF with LiDAR hillshade), SVShade (II) (LiDAR SVF with the SfM hillshade), and fused raster (SfM-derived mosaic and LiDAR DTM). Creating multiple raster images to reveal archaeological remains is an effective way to validate the results as well as make sure that the findings from individual approaches are truthful and agree with each other. This integration was achieved for the LiDAR and photogrammetric datasets after multiple attempts to combine a variety of raster layers to generate single, enhanced raster images. Similar procedures to those applied in Section 2.4.1 were employed here, but with multiple sources of data. Specifically, the parameters for integrating the hillshade raster derived from the LiDAR were: stretch to min–max, blending mode (multiply), transparency (75%), and brightness (default). The same parameters were applied to the hillshade generated from the SfM–MVS method except for brightness, which was set at 20. In addition, the settings for the SVF raster images from both sources were stretched to min–max, blending mode (multiply), transparency (90%), and brightness (default). Lastly, the gradient raster layers were configured at stretch to min–max, blending mode (multiply), and brightness (50%), while the transparency for the layers derived from the photogrammetry and LiDAR were 65% and 90, respectively. Selecting appropriate integration settings is important for accomplishing the desired result. For this reason, the applied parameters of the integration approaches were selected based on their performance in highlighting distinct topographic details of the AOI in a single raster. All the settings applied to generate new integrated raster images are summarised in

Table 1. The settings applied in this study to integrate the raster images (hillshade, gradient, and SVF) created from LiDAR and photogrammetry.

Raster	Integrated Hillshade	Integrated Gradient	Integrated SVF	SVShade (I)	SVShade (II)
Contrast enhancement (Both data sources)	Min–max	Min–max	Min–max	Min–max	Min–max
Blending mode (Both data sources)	Multiply	Multiply	Multiply	Multiply	Multiply
Transparency (%)	LiDAR: 75	LiDAR: 70%	LiDAR: 70	LiDAR: 70	LiDAR: 60
	SfM: 75	SfM: 65%	SfM: 70	SfM: 70	SfM: 60
Brightness (%)	LiDAR: default	LiDAR: 50	LiDAR: Default	LiDAR hillshade: 10	LiDAR SVF: 20
	SfM: 20	SfM: 50	SfM: Default	SfM SVF: 20	SfM hillshade: 20

.

The fusion of DTMs with mosaics (F) was originally performed and investigated by Al-Najjar et al. [62] in their study of land cover classifications using a standalone approach (aerial photogrammetry). This fusion was applied in this study to enhance the detection of archaeology based on multiple sensor datasets. More specifically, we used an open-source computer vision (OpenCV) Python module using the ‘addWeighted’ function, which is clarified in Equation (1) [63]. The image arrays (i.e., IA1 and IA2) were imported together, and various weights (e.g., ∝: 0.2 and β: 0.8, ∝: 0.3 and β: 0.7 and 0.5) were used for each input in order to apply the most optimal pixel weights to create a new fused raster that revealed a relatively greater LOD of the AOI. In this way, ∝ and β values of 0.3 and 0.7 were selected for the mosaic and the DTM, respectively. Gamma, γ, was the constant value (weight) also added to all the image pixels.

$$F=IA\mathbf{1}.\propto +IA\mathbf{2}.\mathsf{\beta }+\mathsf{\gamma }$$

(1)

3. Results

This section includes the visual and statistical results of the detected remains obtained following the methodology implemented. The integrated raster images produced by combining LiDAR and photogrammetric data (Section 3.2) were originally and newly generated and evaluated with the standalone data (Section 3.1).

3.1. Standalone Approach Results

The findings of the standalone approaches that are reported and discussed in this section and Section 4 are crucial aspects to answering the second part of the research question, which focuses on the merits and weaknesses of single datasets in revealing archaeological features. In the previous study led by Kadhim and Abed [29], standalone approaches (LiDAR and photogrammetry) were applied separately to Chun Castle, England. In that study [29], it was recommended that fusion approaches be implemented in future work since it is possible to receive a relatively greater LOD of the archaeological sites. For this reason, we applied both standalone and combination approaches in this case study (Cahokia Grand’s Plaza). The results that were obtained from the standalone approaches are vital to critically assessing the potential of the new approaches—the integration and fusion of the LiDAR and SfM–MVS data—in revealing archaeological remains. Therefore, an investigation was implemented to examine the potential of the LiDAR- and photogrammetry-derived products to provide detailed data for detecting remains in the AOI. In this investigation, point clouds derived from the LiDAR data and the drone-based SfM–MVS photogrammetry created a 2.5D model representation of Cahokia’s Grand Plaza. The VATs were categorised into three main single raster images: hillshade, gradient, and SVF. The assessment was implemented for each raster in terms of the number of the detected remains (e.g., paths/roads, mounds, and marks of demolished buildings), their measured areas in square meters, and the variations in their measurements obtained from different products. We found that both the LiDAR- and the SfM–MVS photogrammetry-derived raster images successfully revealed archaeological features.

The detected marks at Cahokia’s Grand Plaza. which have rectangular shapes (divided into strips of land), were interpreted in this study as patterns of archaeological remains. The detected features are labelled (named) in this study with Latin numerals. Some archaeological marks (mounds), as well as the remains of modern archaeological features, were revealed from all the VATs, i.e., the ‘III’, ‘XX’, ‘XXIX’, and ‘XXX’ (

Table A1. The areas of the archaeological remains, e.g., mounds (Md) and marks of known modern houses that were demolished in the 1970s in Cahokia’s Grand Plaza were revealed in this study using the SfM–MVS and LiDAR data based on the standalone and combination approaches. The areas are auto measured in QGIS (version 3.16) in m², through various VATs, so these values do not represent the exact areas due to the absence of ground control points. The table also shows that the SfM-derived hillshade and RRIM as well as the integrated hillshade provided the highest Level Of Detail (LOD) compared to the other (VATs). The unrevealed archaeological remains have been identified in this table as not applicable (n/a) data. The locations of these remains with the labels of their names are presented in Figure 5.

	Standalone Raster (m2)						RRIM (m2)		Fused Data (m2)	Integrated Hillshade (m2)	Integrated Gradient (m2)	Integrated SVF (m2)	SVShad (I) (m2)	SVShad (II) (m2)
	Hillshade		Gradient		SVF
ID	SfM	LiDAR	SfM	LiDAR	SfM	LiDAR	SfM	LiDAR
Md 48	11,755.7	11,756.4	11,761.6	11,758.8	11,753.4	11,760.8	11,759.83	11,756.39	11,755.4	11,757.8	11,759.6	11,753.5	11,751.56	11,757.11
Md 49	1335.94	1330.09	1329.58	1327.42	1329.39	1325.82	1315.53	1331.53	1335.78	1335.75	1331.06	1338.9	1339.69	1335.11
Md 50	1325.85	1319.14	1328.67	1314.38	1320	1319.96	13,485.75	1312.83	1321.10	1327.64	1324.96	1323.9	1327.95	1330.68
Md 51	1265.56	1256.5	1260.1	1263.45	1261.55	1257.93	1258.39	1265.72	1262.08	1261.37	1261.92	1265.9	1264.96	1265.67
Md 54	653.91	655.85	657.74	660.38	656.3	665.66	649.73	653.83	655.79	649.58	657.2	653.5	651.87	657.52
Md 55	1989.63	1977.56	1983.56	1985.21	1987.22	1987.8	1983.47	1984.29	1983.92	1985.39	1980.73	1980.7	1988.34	1987.25
Md 56	2621.85	2630.23	2626.56	2624.97	2630.88	2627.99	2637.94	2632.75	2635.33	2639.97	2633.07	2637.7	2635.34	2639.53
Md 57	953.85	955.63	951.45	955.88	950.57	950.71	950.39	955.02	953.73	952.47	959.29	957.9	954.78	955.19
Md 59	5921.46	5923.49	5926.38	5930.82	5935.58	5935.21	5936.37	5939.83	5939.65	5938.27	5936.97	5933.5	5935.94	5935.11
Md 60	7137.74	7139.11	7140.56	7147.69	7150.85	7143.11	7152.49	7159.37	7149.29	7150.97	7153.62	7153.1	7153.28	7152.7
Md 61	1475.32	1467.68	1477.48	1477.83	1478.79	1475.24	1479.39	1472.79	1475.06	1472.36	1475.82	1471.2	1472.85	1470.68
Building	2275.37	2289.79	2276.72	2273.82	2271.83	n/a	2271.31	2276.48	2270.78	2273.93	2271.01	2272.1	2272.59	2270.75
Museum	4218.34	4203.77	4215.12	4211.63	4212.55	4210.32	4207.35	4215.58	4212.67	4211.5	4211.74	4215.4	4213.46	4217.39
I	1814.81	n/a	n/a	n/a	n/a	n/a	1811.28	n/a	n/a	1811.46	n/a	n/a	1821.20	1819.58
II	1832.03	1814.59	n/a	n/a	1831.3	n/a	1827.53	1825.6	1829.79	1823.82	1836.63	n/a	1822.19	1822.92
III	2059.66	2059.26	2060.14	2061.25	2063.85	2068.75	2069.03	2059.26	2061.03	2063.82	2057.39	2065.8	2063.39	2061.82
IV	1985.94	1976.14	n/a	n/a	1984.18	n/a	1989.72	n/a	1963.62	1988.97	1986.97	1980.8	1988.05	1985.47
V	1971.27	n/a	n/a	n/a	1932.46	n/a	1949.84	1949.56	1962.86	1975.55	1978.75	1953.6	1975.32	1977.94
VI	1936.01	1941.85	n/a	n/a	n/a	n/a	1947.23	1946.42	1938.27	1946.39	1940.7	1940.50	1941.98	1943.11
VII	1720.62	1748.25	n/a	n/a	n/a	n/a	1722.95	n/a	n/a	1719.36	n/a	n/a	1719.27	1723.07
VIII	2157.7	2127.64	n/a	n/a	n/a	n/a	2160.32	n/a	n/a	2157.43	n/a	2158.6	2159.83	2159.16
IX	1676.13	1699.96	1669.96	n/a	n/a	n/a	1673.39	n/a	1659.57	1668.94	1680.02	1672.33	1673.84	1675.38
X	2536.1	2534.94	2538.12	n/a	n/a	n/a	2548.67	n/a	n/a	2522.8	n/a	2547.98	2530.10	2525.79
XI	n/a	n/a	n/a	n/a	n/a	n/a	1727.88	n/a	n/a	1726.31	n/a	n/a	1726.96	1727.26
XII	2041.02	2016.46	n/a	n/a	n/a	n/a	2048.69	n/a	n/a	n/a	n/a	n/a	2055.50	2052.68
XIII	2127.6	2195.82	n/a	n/a	n/a	n/a	2132.14	2123.24	n/a	n/a	n/a	2124.37	2124.42	2122.83
XIV	1782.86	1775.3	n/a	1789.29	1788.37	n/a	1786.56	n/a	1785.56	1788.32	n/a	n/a	1789.57	1789.20
XV	2073.61	2068.5	n/a	n/a	2074.99	n/a	2075.84	2073.13	n/a	2072.1	n/a	2065.3	2067.92	2069.49
XVI	1933.42	1932.31	n/a	n/a	1942.19	n/a	1942.65	n/a	n/a	1929.79	n/a	n/a	1939.52	1935.23
XVII	2177.45	2162.31	n/a	n/a	2177.26	2174.89	2175.87	n/a	n/a	2163.15	n/a	2170.2	2169.38	2168.53
XVIII	1643.08	1662.08	n/a	n/a	1645.89	n/a	1644.87	n/a	n/a	1635.14	2000.62	1643.3	1649.29	1650.69
XIX	2569.58	2589.17	2569.09	n/a	2570.25	n/a	2571	n/a	n/a	2548.29		2569.0	2551.82	2549.23
XX	3897.08	3987.22	3909.86	3893.91	3902.63	3889.03	3903.71	3889.25	3970.69	3976.8	3979.78	3975.1	3985.71	3980.5
XXI	n/a	n/a	n/a	n/a	n/a	n/a	2217.78	n/a	n/a	n/a	n/a	n/a	n/a	n/a
XXII	4309.49	n/a	n/a	4318.24	n/a	n/a	4316.92	n/a	n/a	4327.83	n/a	n/a	n/a	4311.62
XXIII	2077.23	n/a	n/a	n/a	n/a	n/a	2086.45	n/a	n/a	n/a	n/a	n/a	2076.11	2075.83
XXIV	4310.1	n/a	n/a	n/a	4307.02	n/a	4328.57	4338.37	n/a	4349.57	n/a	4332.9	4319.25	4312.57
XXV	n/a	n/a	n/a	n/a	n/a	n/a	1266.55	1272.83	n/a	n/a	n/a	n/a	n/a	n/a
XXVI	n/a	n/a	n/a	n/a	n/a	n/a	2050.58	n/a	n/a	2074.33	n/a	n/a	n/a	n/a
XXVII	1895.02	n/a	n/a	n/a	n/a	n/a	1897.49	1896.07	n/a	1892.54	n/a	n/a	n/a	1897.76
XXVIII	2506.7	2498.62	n/a	2499.49	2475.38	2493.92	2548.21	2546.82	n/a	2539.03	n/a	2462.9	2505.19	2500.93
XXIX	2097.86	2097.86	2078.82	2078.82	2080.48	2076.78	2093.38	2088.06	2087.15	2092.99	2097.15	2088.4	2089.99	2090.29
XXX	5646.28	5657.59	5642.6	5644.2	5666.71	5644.32	5658.63	5636.36	5657.36	5651.56	5649.53	5652.3	5650.23	5647.01
XXXI	n/a	n/a	n/a	n/a	n/a	n/a	n/a	2069.36	n/a	2075.07	n/a	2065.7	2071.18	2071.95
XXXII	n/a	n/a	n/a	n/a	n/a	n/a	n/a	n/a	n/a	2332.05	2327.72	n/a	2332.38	2330.46
XXXIII	n/a	n/a	n/a	n/a	n/a	n/a	n/a	n/a	n/a	1621.17	n/a	n/a	n/a	n/a
XXXIV	n/a	n/a	n/a	n/a	n/a	n/a	n/a	n/a	n/a	1227.35	1233.76	n/a	1226.82	1229.91
XXXV	n/a	n/a	n/a	n/a	n/a	n/a	n/a	n/a	n/a	n/a	790.53	n/a	n/a	n/a

and Figure 4 and Figure 5). Examples of these remains derived from the SfM–MVS are ‘X’ and ‘IX’ in both hillshade and gradient, the ‘XIX’ in all the VATs, and the ‘XVII’ and ‘XVIII’ in both hillshade and SVF (Figure 4 and Figure 5). In addition, these structures obtained from the LiDAR data are ‘XVII’ in the hillshade and SVF, ‘XIV’ in the hillshade and gradient, and the ‘XXVIII’ feature in all the VATs (LiDAR VATs). The reasons for these differences in archaeological detection are discussed in Section 4. The SfM–MVS-derived hillshade uncovered relatively more marks of the study site when compared to other raster images. Specifically, the visual interpretation of the hillshade from the SfM and the LiDAR have detected 26 and 20 features (modern demolished structures), respectively, 14 known structures (e.g., mounds, museums, buildings), and well as linear features, i.e., footpaths/roads (Figure 4). However, more than 50% of the polygon features detected in the hillshade raster images were also revealed in the other VATs as linear features, not polygons.

3.2. Combination Approach Results

The findings of the integration and fusion approaches that are presented and discussed in this section and Section 4 are vital aspects to addressing the third part of the research question, which concentrates on assessing the LOD of the integration and fusion raster images with the standalone datasets. A strong agreement was found between the marks revealed in the standalone and combination approaches (integration from the same and different sensors). The standalone and combination approaches to Cahokias’s Grand Plaza detected 50% of common marks of modern features (e.g., walking paths/ roads) that have been demolished (Table A1, Figure 6, Figure 7 and Figure 8). These marks and the labels of their names are presented in Figure 5. Relatively more archaeological marks were revealed when two raster images were combined. In particular, the RRIM derived from the SfM photogrammetry (Figure 6) and the integrated hillshade (Figure 7) allowed for the identification of 30 rectangular structures, respectively (in addition to the mounds), as well as linear features (roads/ footpaths). They considerably enhanced the detection of archaeological data when compared with the single raster images. However, the LiDAR RRIM (Figure 6) provided less data than the SfM-derived RRIM; there were 14 rectangular features (marks of demolished constructions) revealed in the LiDAR RRIM (Figure 6). Figure 7 shows the outcomes of the new, integrated data applied using different sensors; it is evidently displayed that the integrated raster provided a comparatively better distinction of the remains in the AOI than the single raster.

In addition to the integrated data, some features were detected in the fused raster. The fusion approach created enhanced data when compared to the mosaic and the LiDAR DTM. Several features were revealed, such as the known features (mounds and museum) and the remains of modern, demolished structures (‘II’, ‘III’, ‘IV’, ‘V’, ‘XIV’, ‘XXIX’, and ‘XXX’) as well as linear features (Table A1 and Figure 9). The edges of the mounds were more easily revealed in the fused raster than in the integrated data (Figure 6, Figure 7, Figure 8 and Figure 9). However, these marks, such as ’VI’, ‘VII’, and ‘VIII’, were less clearly recognised in the fused data compared to the integrated data. The presented results suggest that although the fusion approaches improved the detection of the single data, the integration approaches in this specific case study provided a relatively more detailed detection.

The marks detected with quadrilateral shapes were auto-calculated in QGIS to assess the variations in area between the standalone and the combined data. The auto-calculation was executed using the remote sensing data for the AOI. The size of the detected marks varied in individual rasters, and these variations are relatively correlated to the spatial resolution and the traits of the individual VATs, as discussed in Section 4. The variations of the two measurements acquired from the standalone and the integrated data are depicted in boxplots (Figure 10 and Figure 11) to demonstrate whether the results from both approaches are significantly different from each other. The variations of the standalone data include the areas, in m², of the archaeological and modern remains detected from the six individual VATs (i.e., hillshade, gradient, and SVF) obtained from the Li-DAR and photogrammetric data. In parallel, the differences in the integrated results comprise the areas, in m², of the remains revealed from the six integrated VATs (i.e., integrated hill-shade, integrated gradient, integrated SVF, SVShade (I), SVShade (II), and the fused data).

The variations in the measured areas of the study area were determined separately due to the number of detected features. The boxplots in Figure 10 represent the variations in area (m²) of the traces of known features (mounds). These boxplots illustrate that, although there were differences between the two approaches in the computed areas (m²) delivered for the same known features, no significant differences were perceived. The maximum variations were found in Mound (Md) 50 and Md 60—the lower whiskers for the standalone and the combined (integrated and fused) data for the Md 50 were 1340 m² and 1320 m², respectively, while the lower whiskers of the Md 60 measurements from the standalone and combined data were 7138 m² and 7150 m², respectively (Figure 10). In contrast, the smallest variations were identified in Md 48 and Md 55: the differences between the upper and lower whiskers for the standalone and combined measurements did not exceed 2 m² (Figure 10). In addition, the upper whiskers for both the standalone and the combined data of Md 55 were 1990 m², while the lower whiskers were 1284 m² and 1282 m², correspondingly. The variations in the area of the marks of modern, demolished constructions were greater than those for the mounds. The greatest variations between the measurements of the two approaches were found in the feature ‘XX’: the upper whiskers for the standalone and the combined data were 3900 m² and 3980 m², respectively, while the lower whiskers were 3880 m² and 3875 m², respectively. In contrast, the smallest variations of the two measurements were identified in feature ‘III’: the upper whiskers were 2064 m² and 2068 m² for the standalone and the combined data, respectively, while the lower whiskers were 2060 m² and 2058 m², correspondingly (Figure 11).

4. Discussion

The integrated data (e.g., the SVShade and integrated SVF) appear to be at a satisfactory level in providing more detailed raster images when compared to data from most standalone and fusion approaches. Therefore, this section contributes to a better understanding of the reasons why various outcomes may be obtained by using different remote sensing approaches and accomplishes the overall aim of this study to identify the most appropriate approach for specific archaeological applications. The features detected by the single and combined VATs of the AOI are presented in Figure 12.

4.1. Standalone Data Outcomes

The purpose of detecting archaeological and modern remains in this study was to provide valuable LOD regarding the design of the study site in the past. This information enhances and adds significant detail to the existing/known archaeological data. The outcomes of this research will remain accessible to the public for educational purposes and further interpretation and exploration and conservation.

The marks of archaeological and modern remains that were detected in all VATs are less likely to be created from raster distortions. Examples of this include the remains of features from modern demolished structures, i.e., ‘III’, ‘XX’, ‘XXIX’, and ‘XXX’. These features were revealed in all standalone VATs (Figure 4). Bennett et al. [26] also found that applying more than one VAT (derived from LiDAR data) was an effective way to reveal and confirm the existence of detected remains. The investigations in this research, which used various VATs derived from multiple remote sensing data, are likely to considerably confirm the real existence of the common detections extracted from individual rasters. On the other hand, features that were revealed in only one VAT are considered potential errors, such as XXXV from the integrated gradient (Figure 7).

Despite these common detections, outlet points of the measured areas were also found. Specifically, the area variations in the ‘III’ feature of Cahokia’s Grand Plaza (Figure 10 and Figure 11) did not exceed 11 m². These variations indicate that the applied VATs have different specifications in visualising features. However, other unexpected variations emerged in this investigation. Particularly, the feature ‘XXII’ (Figure 4b,c and Figure 5) was revealed using the LiDAR gradient and SfM hillshade the (difference between them was 8.75 m²); however, those revealed by the integrated hillshade and SVShade (II) were comparatively significantly smaller (Table A1, Figure 4, Figure 7 and Figure 8). This is because not all of the entire quadrilateral feature was detected. This unanticipated result suggests that some edges of the ’XXII’ feature were not entirely revealed in the SfM SVF, so a smaller quadrilateral shape was revealed compared to those revealed by the LiDAR gradient and SfM hillshade.

Various archaeological features were obtained by applying multiple approaches due to the different conditions (e.g., sensors, spatial resolution, and the settings for data collection) of individual datasets. The outcomes for Cahokia’s Grand Plaza were rather similar to those obtained in [29]. In other words, the VATs derived from the SfM photogrammetry also revealed more remains than the LiDAR data, although the same processing and analysis methods were implemented in both datasets. This is an indication that the SfM–MVS method holds significant advantages over the LiDAR data, although the same processing and analysis methods were implemented in both datasets. This is likely to be due to the differences in spatial resolution between the two datasets. The spatial resolution of the SfM–MVS data was relatively higher (0.23 m) than that of the LiDAR data (0.50 m) for the AOI.

In this research, the ground sampling distances (GSDs) of the LiDAR and photogrammetric data were 0.5 m and 0.23 m/pix, respectively. This result also agrees with Vilbig et al. [2], who compared both datasets from the same AOI (Cahokia’s Grand Plaza) in terms of their qualities and efficiency. They stated that photogrammetric data were rather more feasible and practicable than LiDAR data, specifically in this archaeological area, which contains slight vegetation cover. Regardless of the differences in the aims of both studies, the conclusion of this part of our research meets [2] the conclusion that the UAV photogrammetry can generate comparable results to LiDAR data with less cost and greater ease.

Further, feature recognition relies not only on the dataset but also on the settings used to create the VATs. The settings for generating individual VATs depend on the topo-graphic type of the study area [23,27]. The latter study also applied various modes to create VATs to improve the recognitions of topographic features. Producing VATs by applying default settings, e.g., gradient setting: method (planer), Z factor (1), hillshade setting: horizontal angle (315°), vertical angle (40°), and Z factor (1), performed successfully in most archaeological areas, including the AOIs in our research and previous studies such as [25,27,29,55]. Nonetheless, the context of the SVF and the topographic openness layers (which were created in Section 2.3) should, in some cases, be modified from their default settings. The default settings are: radial limit (10,000 m), method (sectors), multi-scale factor (3.00), and the number of sectors (8). These settings were applied in this study and created a distortion of the raster image. Hence, the default settings were changed (Section 2.3). This suggestion was also reported in the study by Kokalj et al. [23], who recommended not applying the default settings for the SVF and the topographic openness as the default settings might not be effective for revealing some topographic features (specifically, relatively plain areas). Kokalj et al. [23] successfully identified an archaeological topography for areas that have relatively flat terrain in the SVF by applying the following parameters: directions (16° and 32°, respectively) and search radius (10 m). These configurations were unable to identify features of the AOI in this research as they did in previous studies [64,65]. In this study, a five-meter radius was used with the sector method; these were the most appropriate parameters, as the archaeological marks, which were only subtly indicated in this case study, became relatively clearer and more detectable. In accordance with these parameters, Daxer [54] demonstrated that applying a relatively smaller radial limit (e.g., two meters) is likely to provide a relatively more detailed topographic information than a higher radial limit (e.g., 50 m and/or 1000 m).

In addition to the settings, each VAT had different specifications. The hillshade data derived from the SfM photogrammetry and LiDAR data revealed relatively more marks compared to other VATs, particularly in Cahokia’s Grand Plaza. The hillshade raster images of the AOIs derived from both sensors were generated by applying the same parameters (vertical angle of 40°, azimuth of 315, and a Z factor of 1). These parameters were also demonstrated by other research [25,27] to be optimal settings for identifying archaeological remains and marks of ruined constructions. The outcomes of the hillshade in this case study (Cahokia’s Grand Plaza) would suggest considering the hillshade to be a more plausible detection technique than other standalone VATs. However, Challis et al. [66] found that an interpretation based merely on the hillshade tends to lead to the oversight of some remains in the study site. There are two studies [23,28] in agreement with the study by Challis et al. [66]: they stated that the illumination drawbacks of this technique could create topographical distortion, especially in archaeological detection.

The above claims indicate that although more features were revealed by the hillshade in this study, there is a possibility that some of them were generated from the distortion of the raster imagery. The outcomes of this study, to some extent, contradict these claims. Specifically, some of the features identified in the hillshade were also revealed as linear features in other VATs. Specifically, the feature ‘V’ was detected in the SfM hillshade and SfM SVF raster images of Cahokia’s Grand Plaza (Figure 4). These detections suggest that those features were not produced from the illuminations of the hillshade, as they were also revealed in the SVF. These outcomes agree with the findings of López et al. [67]. These previous studies normally prefer SVF over hillshade since one of the main merits of the SVF tends to overcome the hillshade’s distortions.

4.2. Combination Data Outcomes

Holata et al. [31] claimed that integrating airborne LiDAR and SfM–MVS data is not a common approach to revealing remains in archaeological areas. Kokalj and Somrak [23] also argued that there is a need to enhance the existing VATs and improve the detection of archaeological features. Various VATs derived from LiDAR data were combined in [23]. Quantifying and assessing archaeological features acquired from the combination approaches are vital for applications in prospecting archaeology, object detection, and assessing the most appropriate approach to explore new archaeological features/areas.

In this study, various VATs were applied. The combined raster images (e.g., SVShade and integrated gradient) that were derived from multiple sensors—LiDAR and photogrammetric data—provided comparatively more detailed and clearer raster images than the standalone VATs. Consequently, more archaeological marks were revealed.

The most important finding regarding the combined data is the new raster image, SVShade (I), which was originally created in this study (Figure 8). The SVShade (I) delivered the best object recognition compared to the other integrated data. It was originally generated by integrating the SfM SVF with the LiDAR hillshade. This integration appears to have overcome the limitations of the individual raster. Previous archaeological studies such as [24,27] integrated different raster images derived from the same source (e.g., LiDAR) and found that the visibilities of topographic features were enhanced. Inomata et al. [25] compared various VATs under various conditions. They found that the RRIM technique brought a relatively great visualization advantage to the end user when compared to other methods. In other words, the RRIMs provided, to some degree, better archaeological detection than other integrated techniques, e.g., the RVT in Jaber and Abed’s study [24]. Kokalj and Somrak [23] responded to [24], as they concluded that blending different data derived from the same sensor based on the (RVT) for archaeological detection is a viable alternative to the standalone approaches. Based on the results of this study, we would argue that the integration of two datasets derived from different sensors, specifically the SVShade (I), can significantly improve and enhance the recognition of archaeological remains and marks of modern demolished structures when compared to the RRIMs and standalone data (Figure 6, Figure 7, Figure 8 and Figure 9). This argument does not imply disagreement with the previous studies but instead highlights a significant advance in the understanding of new approaches that promise to provide a comparatively greater LOD for archaeological areas. It seems clear that, in this study, the integration approaches enhanced the edges of remains and thus improved their recognition. The SVShde was the most highly recommended raster to be applied for archaeological detection as it provided a detailed raster for Cahokia’s Grand Plaza.

Consequently, enhanced raster images were obtained to mitigate the weaknesses of the standalone data. As is illustrated in Section 2.2, the LiDAR and photogrammetric datasets were different in their GSDs in this study. Kokalj et al. [68] found that coarse spatial resolution data (e.g., 1 m/pix) possibly led to a reduction in detection capabilities. Integrating both datasets likely avoids the possibility of the occurrence of pseudo-detection and confirms the features identified from a single raster image. This is consistent with the study by Doneus et al. [69], which stated that LiDAR is likely to penetrate dense vegetation and improve the interpretation of the bare ground due to the higher point density, meaning that it has the capability to deliver relatively more detailed depth data compared to SfM photogrammetry. Consequently, LiDAR has provided improvement to object detection in this case. The combination of photogrammetry and LiDAR is, however, not commonly applied in digital archaeology. In this context, this study shows that the inadequacies of these products are mitigated when both datasets are combined, such as by combining SfM SVF with the LiDAR SVF and the SfM hillshade with the LiDAR hillshade. However, the possible feature ‘XXXV’ was only revealed in the integrated gradient of the AOI. This feature was considered a pseudo-detection since it was only revealed in one raster. This particular outcome of the integrated gradient in this study contradicts the results of other integrated datasets generated in this study since the weaknesses of the single gradient data were not overcome when both raster images were integrated (SfM gradient with the LiDAR gradient). Apart from the possible error detection in the integrated gradient, the integrated datasets, i.e., integrated hillshade, integrated SVF, SVShade (I), and SVShade (II), considerably refined the topographic detail in the raster and thus sharpened the distinction of archaeological and modern remains in the AOI.

Importantly, the outcomes of this research suggest that the integration of both hill-shade and SVF can mitigate the single limitations of each raster and further enhance the detection, providing crucial data for archaeological exploration (Figure 8). A similar result was achieved by Kokalj et al. [68], who integrated hillshade and SVF data derived from the same sensor (LiDAR). The results received in this study indicate that the integration of the SVF and hillshade derived from different sources is more effective in feature detection. The VATs derived from these sources also have various strengths and weaknesses. Specifically, both SfM SVF and LiDAR hillshade are different in terms of visual analysis. Kokalj et al. [68] found that the SVF was an effective analysis technique for improving the identification of archaeological features, particularly from high-resolution data. The standalone result of this study ties well with previous studies [68], wherein the SVF from both sources detected several archaeological and modern remains, especially from the high-resolution raster, i.e., the SfM SVF. The SVF was commonly used in several studies, such as [67], as an alternative VAT to hillshade due to its potential to overcome the limitations of the existing VATs, such as the diffusing illumination of hillshade. However, the findings from the integration approach in this study are slightly different, as the area variations of the detected features were less than those detected in the standalone approaches. This integration provides insight into digital preservation and object detection and demonstrates that combining various datasets can overcome some of the limitations of standalone data.

In addition to the outcomes of the integration approaches, another combined raster image was created from the fusion approach. This study implemented the fusion of the LiDAR DTM with the RGB (red, green, and blue) orthomosaic derived from photogrammetry to refine the final product and emphasise the mark edges identified by the standalone and integration outcomes. The fused data, as the integrated data, enhanced the detection of known, modern features (Figure 9). Examples of these features are ‘II’, ‘III’, ‘IV’, ‘V’, ‘XIV’, ‘XXIX’, and ‘XXX’ (Figure 5 and Figure 9). Additionally, the edges of the mounds were more easily identified in the fused raster (as well as in the aerial images and mosaic) than in the standalone and integrated data for this study (Figure 6, Figure 7, Figure 8 and Figure 9). The detections of these edges were enhanced as the forms of the features are extremely similar, especially in the absence of colour data, which means that the model (DTM/ VATs) might not be able to distinguish and detect them. In this case, the imagery had RGB colour, which contributed to the correct detection of archaeological features through the fused data and provided more realistic representations. Al-Najjar et al. [62] also applied this type of fusion (LiDAR DTM with SfM orthomosaic) for land-cover classification and found that the fused data improved the classifications among vegetation. However, it was anticipated in this research that connecting the mosaic with the height data (DTM) would generate clearer identifications of archaeological features than the integration approaches. Several archaeological features (e.g., ’VI’, ‘VII’, and ‘VIII’) that were detected through the integration approaches were not recognised in the fused raster. The absence of some marks of constructions in the fused raster of the study site led to the conclusion that the integration approaches in this study delivered a higher LOD about the study site than the fusion approach, although the detection of some existing features (mounds) was improved through the fusion approaches.

4.3. Limitations

Despite the importance of the applied approaches and the obtained results, there were some limitations to the applied methods tested in this study. One of the limitations is associated with the area variations of some of the detected archaeological features. The measurements of the detected features, particularly the areas of the quadrilateral marks in a certain VAT, were often >10 m² for corresponding features in another raster technique (Figure 10 and Figure 11). These variations might be caused by digitising errors compounded by limitations in the VATs. Due to the different specifications of individual VATs, detected features were visualised in slightly different ways. Hence, some features were exposed in slightly different dimensions (Figure 4, Figure 5, Figure 6, Figure 7 and Figure 8) and therefore do not represent the exact areas due to the absence of ground control points (GCPs). These differences led to the digitisation of, to some degree, imprecise polygons/polylines of the detected features. Importantly, these variations were nonetheless significantly reduced in the new, combined datasets.

The absence of GCPs is another limitation identified. A number of related archaeological studies, such as Orengo et al. [21], Moussa [37], and Agudo et al. [70], used GCPs as they are arguably important steps to georeference aerial images and to establish object scale. Azmi, Ahmad, and Ahmad [71] argued that control points are recommended to avoid possible distortions in image overlapping. Regardless of these arguments, GCPs were not used in this study since the traditional element of photogrammetric processing was changed. The RTK drones applied here produce high-resolution mosaics and raster imagery through the SfM–MVS method. Additionally, the absolute positions yielded from the RTK drone, in this study, generated scaled models and determined the accuracy of the reconstructed datasets. From above, if the UAVs/digital cameras are RTK-enabled, then establishing GCPs is not essential in SfM–MVS processing [72]. In terms of data combination, several related studies, e.g., Holata et al. [31] and Moussa [37], established GCPs to georeference the laser and photogrammetric data since both datasets had different geographic systems. Herein, the LiDAR and aerial images obtained for Cahokia’s Grand Plaza were geotagged at the same projection and coordinate system; thus, GCPs were not used in this specific case (obtaining georeferenced datasets). However, the absence of GCPs is considered one of the limitations of this study, as the exact coordinates of the detected features (i.e., the centimetre accuracy of their locations) were not conclusively identified.

5. Conclusions

The novelty of this research is presented in a two-step approach to detect and digitally preserve features in archaeology: (i) the application of remote sensing standalone approaches and (ii) combination approaches based on VATs. The original contribution of this study was to combine different datasets acquired from LiDAR and photogrammetry and generate new, combined datasets in order to examine whether the new data can enhance and improve archaeological information in comparison to the standalone method. In this study, the combination of the LiDAR and photogrammetric datasets was investigated to detect archaeological and modern remains in Cahokia’s Grand Plaza site. The results demonstrate that both standalone and combination approaches produced successful detections of several remains in the AOI. The results illustrate that the standalone VATs generated from the SfM–MVS method delivered a greater LOD of the AOI than those created from the LiDAR data. However, the integrated VATs boosted the edges of archaeological features (mounds, paths/roads, and traces of demolished constructions) and improved their identification to a relatively greater extent than the standalone VATs. Particularly, the SVShade raster images were the most highly recommended raster to be applied in archaeological detection. Therefore, the integrated raster images allowed us to further detect and confirm remains that were detected using the standalone approaches. For future work, deep learning object detection algorithms can be applied based on standalone and combined remote sensing datasets to automatically detect archaeological remains. We argue that the integrated raster images generated in this study from two different sensors are valuable data sources that can be applied successfully to detect new archaeological features. In addition to deep learning algorithms, they may contribute to a paradigm shift in digital archaeology.