Identifying Landscape Modification using Open Data and Tools: The Charcoal Hearths of the Blue Mountain, Pennsylvania

In this technical brief I demonstrate two methodological points broadly relevant to historical archaeologists. While light detection and ranging (LiDAR), also known as airborne laser scanning (ALS), has been widely used to identify prehistoric archaeological sites, its use in historical archaeology could be expanded. LiDAR data are particularly valuable because they are frequently open access. By coupling open LiDAR data with open source software one can quickly, easily, and inexpensively identify historical landscape modification. I present an illustrative example, the identification of charcoal hearths in Pennsylvania, along with tools and techniques used to carry out the research. This method has allowed us to identify 758 charcoal hearths within a 74 km2 research area along the Blue Mountain of northeastern Pennsylvania.


Introduction
Analysis and interpretation of the physical landscape is a significant component of historical archaeology. In this technical brief I demonstrate, through the case of charcoal hearths from the Blue Mountain in Pennsylvania, the value of open light detection and ranging (LiDAR) data and of using open source software to process the data. The project has facilitated the identification of charcoal production sites (hearths) along the slopes of the Blue Mountain, revealing an extensive industrial landscape within the "wilds" of State Game Lands No. 217. Although not presented here, this newly revealed historical landscape creates an opportunity to better understand a poorly documented segment of 19thcentury society-the people, known as colliers, who produced the charcoal. In order to provide these richer interpretations, however, we need evidence upon which they can be built. Below, I discuss four important aspects of this project. First, I provide historical context for the project. Second, I discuss the ethical and practical advantages of open data and open source software along with the specific data and software used herein. Third, I present the analysis, which, though specific to this study, promotes both the reproducibility of this research and the use of these tools for other research projects in historical archaeology. Fourth, I present a very brief discussion of the results, though not the interpretation, of the methods discussed herein.

Charcoal Production
During the 19th century charcoal was widely used as a fuel for smelting iron (Birkinbine 1879;Kemper 1941;Straka 2014). Charcoal was ideal because it introduced few contaminants into the iron smelting process (McVarish 2008:268). In the Americas, vast tracts of "unoccupied" forest made charcoal an even more attractive fuel. Even after methods were devised to use coal for iron smelting, replacement of charcoal by coal was a slow process that took approximately 100 years (Schallenberg 1981). Between 1828 and 1878 two charcoal furnaces (Lehigh Furnace and East Penn Furnace) and one charcoal forge (East Penn Forge, a.k.a. Pennsville) were constructed in northern Lehigh County and southern Carbon County, Pennsylvania, along the northern edge of the Lehigh Valley ( Fig. 1) (Mathews and Hungerford 1884;Swank 1892;Brenckman 1913). At the time the only large tracts of standing forest remaining in the Lehigh Valley were along the Blue Mountain, a long (approximately 150 miles) ridge that acts as the boundary between Lehigh and Carbon/ Schuylkill counties. Bernard Fernow, who managed the tract from 1879 to 1887 and would become the third chief of forestry of the United States Department of Agriculture, labeled the mountainous tract "waste land" because it had little use beyond charcoal production (Fernow 1883). Between 6,000 and 15,000 acres along the Blue Mountain were owned and harvested by the owners of these furnaces (Straka and Ramer 2010).
The making of charcoal involves steps that ensure the careful and efficient conversion of wood to charcoal. Wood was cut into 4-ft. lengths by wood cutters, who were frequently farmers or colliers, working in the winter "off season." To begin, a flat, level, compact circle of earth between 8 m and 14 m in diameter was constructed. On slopes, this involved the removal of upslope earth that was placed downslope to create a circular or oval terrace. Wood was then piled vertically in three tiers around a central "chimney" placed in the center of the circle. Gaps left by stacking larger "billets" (4-7 in. in diameter) were filled with "lap-wood" (1-4 in. in diameter). The entire pile was covered with a layer of leaves and other detritus, which was then coated with a layer of earth and charcoal dust (Fig. 2). Colliers then ignited the wood through the chimney, which they then sealed. During the first 24 hours or so, the collier carefully monitored and controlled air vents in order to ensure the spread of the charring while avoiding burning the wood. Once the collier was sure that the smoldering had spread throughout the hearth, they did not need to remain at the hearth, instead returning periodically for monitoring and managing throughout the next seven days. After completion, the hot hearth was raked out to separate the charcoal from the soil (Fig. 3); the charcoal was then transported to its destination via carts along forest paths and roads (Kemper 1941;Walker 1966;Wigginton 1979;Zeier 1987;Wettstaed, 2003, Hart et al. 2008Strachan et al. 2013;Johnson and Ouimet 2014;Straka 2014).
We know little about how the forests were managed or about the lives of the colliers who, through skill and sleepless nights, produced the fuel to power the furnaces. This paper describes the LiDAR-based methodology employed that will allow us to begin shedding light on the lives of the colliers.

LiDAR
Light detection and ranging (LiDAR), also known as airborne laser scanning (ALS), has revolutionized archaeology in forested regions. In recent years, archaeologists have used LiDAR to dramatically rewrite our understanding of settlement patterns and farming practices in the Maya area, Hawaii, Mexico, and Cambodia (Chase, Chase, Weishampel et al. 2011;McCoy et al. 2011;Chase, Chase, Fisher et al. 2012;Evans et al. 2013;Fisher et al. 2017). Historical landscapes have also been investigated using LiDAR (Devereux et al. 2005;Harmon et al. 2006;Opitz et al. 2015;Johnson and Ouimet 2018). Charcoal hearths and their associated industry have been investigated in the northeastern United States and in Europe (A. Raab et al. 2015;Schneider et al. 2015;Hirsch et al. 2017;T. Raab et al. 2017).
In essence, LiDAR is a relatively simple technology (for detailed description see Fernandez-Diaz et al. [2014]). A laser pulse is shot at a surface, bounces off the surface, and the reflected signal that returns to the source, or "return," is detected. The time traveled is recorded and the distance is calculated (at the speed of light through a medium, typically air), thereby measuring the relative location of the surface compared to the laser. Highly sensitive equipment and powerful computing are required as thousands of pulses are shot and recorded per second from an airplane. The geolocation of the airborne laser is measured through the use of high precision global navigation satellite system (GNSS [GPS or global positioning system in the United States]) units and inertial measurement units (IMU). The first (GNSS) measures absolute location and the second (IMU) records relative movement. Between the two, airplane location can be measured with great accuracy and precision. From the airplane, lasers are scanned across the surface in a swath covering the study area.
After postprocessing, the resultant data are a point cloud, a series of points recorded with longitude, latitude, and altitude coordinates. Points are computationally classified (Table 1). Archaeologists are primarily concerned with ground points, which are relatively easy to identify because they are the "last return," that is the pulses that took the longest to return to the detector. Since laser pulses cannot penetrate the ground, "last returns" represent the surface. LiDAR data for much of the United States are available through aggregators. 1 For the ground return (class 2) data considered herein, an average of 6.25 LiDAR points were collected per 100 ft. 2 (or approximately 1 point/1.5 m 2 ).
Although the point cloud can be used as is, it is typically converted into a digital elevation model (DEM) (Werbrouck et al. 2011;Kokalj and Hesse 2017). A DEM is a raster that stores elevation in each grid unit, much as digital photographs store color in a grid of pixels. The difference is that DEM grids are located on the landscape. To convert the point cloud to a DEM, a triangular irregular network (TIN), one of the more common methods, is used to convert the point cloud into a solid surface by making contiguous 1 Aggregators include Open Topography (<http://www. opentopography.org/>, accessed 2 February 2019) and United States Interagency Elevation Inventory (<https://coast.noaa.gov/inventory/>, accessed 2 February 2019). triangles with three of the closest points as vertices (Fernandez-Diaz et al. 2014). This surface of triangles (Fig. 4b) is then converted into the regularly gridded surface of a DEM. Once prepared the DEM can be visualized using hillshade, ruggedness, slope, and slope aspect (Challis et al. 2011). For this analysis, each cell of the DEM is 1.3 × 1.3 m (1.69 m 2 ).

Why Open?
Only open data and open source software (or FOSS) were utilized for this analysis for practical and ethical reasons (Kansa et al. 2011;Ducke 2012;Wilson and Edwards 2015).
The FOSS tools described herein save time. It is often believed that open source software can have a high adoption threshold and a lack of support (for example via a help line), thereby costing precious time. When true, open source software can be difficult to use. The tools described herein are included because they are well-supported, stable, cross platform, and flexible and, therefore, have relatively low adoption thresholds.
The FOSS tools described herein save money. Because they are freely available on the internet, costs are limited to a computer and access to the internet. Low cost reduces (though cannot eliminate) barriers for underfunded archaeologists. Open source software can be seen to grease the wheels of entrance into professional archaeology. This is particularly significant for populations poorly represented in archaeology.  The use of open data and open software eases archaeologists' ability to abide by our organizational ethics, including those of the Society for Historical Archaeology (2015), the Society for American Archaeology (1996), and the Register of Professional Archaeologists (2018)  Open formats encourage responsible archiving and curation because they allow out-of-date formats to be easily read. Second, the ease of sharing also promotes "clean" data-data clearly understood by people other than those who collected it. Third, if we can share data easily then we must be more mindful of how it is collected in the first place. That is, openness promotes careful attention to both the data and the method of data collection, yielding benefits for the original collectors/users and unintended users. Fourth, the code for open source software is available for inspection. While few archaeologists will delve into the code, this availability means that they, or a knowledgeable expert, can check that code. This means that there are no more proprietary black boxes. Fifth, future uses of data are never fully predictable; openness encourages experimentation and reuse. Sixth, using open source software with students promotes skills (easily transferred to commercial software) and flexibility and frees them from oppressive subscription fees when they graduate.

Open Data
This project was initiated because of open data. Curiosity about odd features of the local landscape along the Appalachian Trail led me to the streaming, LiDAR-derived hillshade on the Pennsylvania Imagery Navigator website. This data allowed me to quickly and easily identify many odd features on the landscape that turned out to be the remnants of charcoal hearths.
The data used in this research are provided openly by PAMAP, a collaborative project of local, state, and federal agencies, and the Pennsylvania Spatial Data Access program and website as well as via the user-friendly Pennsylvania Imagery Navigator. LiDAR data were collected by PAMAP between 2006 and 2008 (Pennsylvania Department of Conservation and Natural Resources 2018a). Accuracy of the postprocessed LiDAR points is much better than the standards (18.5 cm vertical and 5 ft. horizontal accuracy) with approximately 1 point per 14 m 2 (Pennsylvania Department of Conservation and Natural Resources 2018b). Data are provided as a hillshade derivative of a DEM via the Pennsylvania Imagery Navigator website via streaming data for geographical information systesm (GIS) software or data download. However, the DEM (and therefore the hillshade) is based upon a subsample of the original data and, therefore, has a much lower resolution than the original data. Fortunately, the original (though processed) LiDAR point cloud data can also be downloaded in the open LAS format (American Society for Photogrammetry and Remote Sensing 2013).

Open Tools
Two FOSS applications were used for this analysis, LASTools 3 and QGIS 4 (for additional FOSS GIS see Orengo [2015]). LASTools is a lightweight collection of tools for viewing, modifying, and managing LiDAR data in a graphical user interface that can also be run via command line. The licensing, and therefore the openness, of the tools within the suite varies. Some, such as las2las (see below), are FOSS. Others, such as lasview are free, but not open source. Others, such as lastile and blast2dem, are free for educational uses but are not free to others nor are they open source. While ideally all of the tools used herein would be open source, LASTools are the most efficient and widely used tools for working with LiDAR. LASTools can be used as standalone applications or within the QGIS (or ArcGIS) environment. The output of LASTools is typically a DEM for viewing and analysis within GIS software.
QGIS is a cross-platform (Windows, Mac, and Linux) open source GIS application. There is no need to repeat the many comparisons of QGIS with commercial GIS applications, especially since software packages and subscriptions change constantly (for example, Orengo [2015]). Generally speaking, the differences tend to be at the edges of GIS use-for most purposes, QGIS is equivalent to commercial packages. "Bleeding edge" uses are more variable. QGIS contains a plethora of native analysis tools that can be expanded with plugins. QGIS support is extensive and includes tutorials, videos, and a very active user community. Commercial support for QGIS can be purchased relatively inexpensively.

Method
In order to understand the landscape of charcoal production within the research area along the Blue Mountain (Carter 2018a), a wide variety of maps and other resources was consulted. Initially, using the online hillshade model provided by Pennsylvania Imagery Navigator, I was able to identify charcoal hearths because they are visible along the flanks of the mountain as flat, level, and round areas distinct from the slope (Fig. 4). To record the charcoal hearths, the hillshade derived from the LiDAR was brought into QGIS via streaming web mapping service (WMS). 5 WMS makes it easy to quickly view large datasets over the internet, but you can also download the large tiles from Pennsylvania Spatial Data Access (PASDA). Using only this data, we were able to identify 298 potential charcoal hearths within the study area (Fig. 4b).
However, the hillshade provided by PASDA was derived from a subsample of the original point cloud. Only class 8 points (see Table 1) were used, reducing the resolution of the data and limiting the visibility of many charcoal hearths (Pennsylvania Department of Conservation and Natural Resources 2018b) (Fig. 4b).
Happily, the Pennsylvania Imagery Navigator website provides the original LiDAR data openly via download.
Using LASTools, I converted the downloaded Li-DAR data into a DEM. I used las2las to separate the ground points, including class 2 and 8 points (see Table 1), from the other points. Using lastile, these larger files were broken into smaller 1 × 1 km tiles, which makes the work more efficient and gets around some of the licensing restrictions (without breaking the license). This is converted to a DEM using blast2dem then stitched back together using the merge function in QGIS (Carter 2018b).  DEMs are difficult to use, however. One of the most common methods of visualizing DEMs is a hillshade (Carter 2018c), which is an algorithmic construction of the landscape based upon illuminating the DEM with a "sun" in a particular position (Kokalj and Hesse 2017). This makes it much easier to understand the topography ( Fig. 4b and 4c). Visibility of features is dependent upon the direction of the sun, however. Therefore, a charcoal hearth may be more or less visible in a hillshade depending upon its location (Fig. 4c). An alternative was needed. Because many of the charcoal hearths are located on the slopes of the mountain, we can use the differing slope of the mountain versus the level charcoal hearth to better visualize the hearths (Kokalj and Hesse 2017). Using the slope analysis (Carter 2018e) we were able to identify 758 charcoal hearths within the research area (Fig. 5) (Carter 2018d). Note that this includes only hearths visible in the slope analysis. Flat hearths on flat land would not be visible in the slope analysis or the hillshade. We are currently working on methods to identify these hearths.

Results
The increased ability to identify charcoal hearths can be seen in Fig. 4. Figure 4a shows an aerial photo of a portion of the slope of the Blue Mountain. Figure 4b shows the streaming hillshade (based on class 8 points) provided by PASDA; only two charcoal hearths are visible. Figure 4c shows the hillshade constructed from the LiDAR data selected using the tools above (class 8 and 2). Figure 4d shows the slope analysis of the class 2 and 8 data. Note that, with each additional step, the visibility of the charcoal hearths increases. In the original streaming hillshade, we were able to identify 298 charcoal hearths within the research area. The transformation of the Li-DAR data described herein increased this number by 460 (154%) for a total of 758 charcoal hearths (Fig. 5). To date, 82 of these hearths have been ground truthed: two false positives and one false negative have been recognized. Ground truthing involved visiting the location and examining the feature for characteristic elements including a flat, level area 10-15 m in diameter, a distinct rim, dense charcoal near the surface at the edges, and, on slopes, a dramatically steeper slope uphill and downhill resulting from the removal of dirt uphill and placing it downhill. The reanalysis expanded our ability to discern charcoal hearths mainly in areas of lower slopes and more variable topography. Slopes of greater than 20% contain very few charcoal hearths, but these are highly visible. Charcoal hearths located on slopes between 15% and 25% are extremely common and are now easy to identify. Charcoal hearths located on slopes less than 15% are more difficult to identify, but are presumably more numerous since it would have been easier to harvest and move timber on flatter land. We are currently working on pedestrian surveys to help increase our ability to recognize charcoal hearths in relatively flat (i.e., <10% slope) areas including the top of the mountain and the lower slopes.

Conclusion
The work discussed herein is a presentation of a particular research project, but it is intended to stress two important components of this research that are broadly applicable. First, the essential importance of LiDAR/ ALS in historical archaeology, especially in the recognition of historical landscapes. Second, the benefits of using open data and open tools. In this case, the former is only possible because of the latter. Open LiDAR data are particularly important for historical archaeologists because they facilitate the recognition of activities that changed the landscape. Of particular note are mining and industrial activities, especially those that have been long abandoned and, therefore, are buried in forests. However, the recognition of activities with less significant impacts, such as historical trails and paths or agriculture, is also possible with LiDAR.