Large‐scale mapping of anthropogenic relief features—legacies of past forest use in two historical charcoal production areas in Germany

With the increasing availability of high‐resolution digital elevation models (DEMs), large‐scale mapping of anthropogenic relief features has become feasible for more areas. However, the landscape‐scale distribution patterns of anthropogenic landforms and the quality of DEM‐based mapping can be highly heterogeneous. In this study, we mapped relict charcoal hearths (RCHs) in two study regions with differing environmental backgrounds, covering forest areas of more than 15,000 km2, analyzed the RCH distributions and evaluated possibilities for predictive modeling of RCH occurrence with respect to natural and cultural landscape structures. More than 45,000 RCHs were recorded in each region, with high site densities even in areas remote from charcoal‐consuming industries. Variations in the quality of DEM‐based mapping were related to small‐scale differences in the DEM quality and larger‐scale substrate heterogeneity. A clear association between RCHs and historical industrial sites was found in the Northern European Lowland; while the density of mapped RCHs was predominantly related to geology and morphology in the lower mountain ranges. The results show that variations in mapping quality across scales and the natural and cultural background of a region need to be considered so that the mapping of anthropogenic relief features can contribute to an improved understanding of land‐use history.

have often focused on detailed pedological or anthracological analyses of a few sites. During recent years, large numbers of RCHs have been described for more regions (Hardy & Dufey, 2012;Hazell, Crosby, Oakey, & Marshall, 2017;Johnson & Ouimet, 2014;Mastrolonardo, Francioso, & Certini, 2018;Rutkiewicz, Malik, Wistuba, & Osika, 2019;Schmidt, Mölder, Schönfelder, Engel, & Fortmann-Valtink, 2016), and the focus of many studies was on mapping sites and analyzing their spatial distribution over larger areas (Carter, 2019;Raab et al., 2019;Rutkiewicz et al., 2019). Automated mapping methods for improved or more timeefficient mapping have been developed and evaluated Trier, Zortea, & Tonning, 2015;Witharana, Ouimet, & Johnson, 2018). The growing database of RCH sites allows for the comparison of site densities and morphologies between regions, but a thorough understanding of the possibilities and limitations of DEM-based mapping is a prerequisite for meaningful interpretations of the distribution patterns of mapped features. This prerequisite applies to DEM-based mappings of RCHs but also to other anthropogenic relief features, such as burial mounds, agricultural terraces, and ridge and furrow systems.
Although LIDAR-based mapping often records large numbers of sites, ground-truthing studies indicate that a considerable proportion of sites cannot be captured from DEMs. Because of the effort required for ground-truthing and the limited number of field-based studies, whether DEM-based mapping results in biased databases of RCHs and other small relief features remains unclear (e.g., whether smaller or older sites within charcoal production areas are underrepresented in the mapping results, or whether the mapping quality differs between areas with different environmental background conditions).
Naturally, many mapping studies focus on regions where large numbers of sites can reasonably be expected, for example, in regions where RCHs are already described in archaeological records or where intensive charcoal production is to be assumed because of the presence of historical industry with high charcoal consumption (e.g., Ludemann, 2010;Raab et al., 2015;Rutkiewicz et al., 2019). A mapping of RCHs in such areas can provide valuable information for further studies on charcoal production practices, forest-use history, or industrial history and can contribute to the development of workflows and practices in archaeology and conservation strategies that record and consider RCHs. However, when transferring the results of small-to-intermediate scales to larger areas, the overall number and density of features might be overestimated because of the common focus on "hot spot areas". Furthermore, studies on RCH stratigraphy show clear differences in the practices of using hearth sites between regions, which limits the possibilities of drawing conclusions on the intensity or duration of charcoal production based on the study of site densities. However, the relationships between RCH density and natural or cultural background structures (e.g., geomorphology, soils, industrial history, etc.) are poorly understood.
The central aims of this study were to (a) record the numbers, spatial densities and distribution patterns of RCHs on a large scale, including forest areas close to and remote from historical industries; and (b) contribute to an improved understanding of the possibilities and limitations of DEM-based mapping by comparing RCH mapping results between regions with differing environmental background conditions. The specific objectives of the mapping were to 1. assess the spatial differences in mapping quality across scales, 2. evaluate the spatial heterogeneity in site densities with respect to natural background structures and site locations, and 3. derive a basic understanding of regional differences in charcoal production practices from RCH morphologies and site locations.
Therefore, we mapped RCHs in forest areas within two study regions that include forest areas of 10,000 and 5,000 km 2 and analyzed the RCH densities and their spatial distributions in relation to the environmental background conditions and basic structures of the industrial history of the regions.

| Study areas
The mapping areas of our study cover the state of Brandenburg in northeastern Germany and the historical iron production area of the Upper Palatinate/Northern Bavaria in southeastern Germany ( Figure 1). The study regions were delimited based on data availability and the distribution of historical industrial sites. RCHs were mapped in all the forest areas in both study regions.
The study region of Brandenburg is situated in the Northern European Lowland and is characterized by homogeneous geology and topography shaped by Quaternary glaciations. The morphology is predominantly flat, and the substrates are loose and mainly sandy.
Forests cover approximately 10,300 km 2 of the study region (the total area is approximately 29,500 km 2 ) and are dominated by pine in large parts. The potential natural vegetation (pnv) are beech forests in the northern and western parts and basswood-hornbeam, pine-oak and pine forests in the central and eastern parts (Hoffmann & Pommer, 2005). Most of the important charcoal-consuming industries were operating in modern times, with a small number of large iron-and other metalworking centers that were established in the 17th and 18th centuries in different regions. In addition, smaller iron hammer mills operated mainly in the southeastern part of the state.
Historical glassworks are concentrated in the northern part of the study region and were operating mainly from the 18th century, with many sites operating for only short time periods.
The study region of Northern Bavaria is situated in the lower mountain ranges of southern Germany. The geology, morphology, and soils in the region are heterogeneous, with a dominance of Mesozoic sedimentary rocks in the western part and metamorphic and igneous rocks of the Bohemian Massif in the eastern part of the study region.
Approximately 5,000 km 2 of the 10,700 km 2 study region is covered by forests, which are mainly dominated by spruce or pine. The pnv for the largest parts of the region is dominated by beech forests, and fir-beech forests are present in the eastern part (Bayerisches Landesamt für Umwelt, 2012). The study region extends around the mining and iron production area of the Upper Palatinate, which was of supraregional importance in the High Medieval and early modern period. Ironworks are concentrated along several small rivers, especially in the central part of the study region, whereas glassworks were mainly concentrated in the eastern part of the region.

| Large-scale RCH mapping
RCHs were mapped from LIDAR-based DEMs provided by the state topographic survey for forest areas in both study regions (Landesvermessung und Geobasisinformation Brandenburg, LGB; and Bayerische Vermessungsverwaltung, BVV). The denominated original ground point density of the LIDAR datasets was at least 4 points per m 2 for most areas in both study regions except for the northeastern part of the Bavarian study region, where it was at least 1 point per m 2 (Figure 1).
The mapping areas were confined to forest-covered areas based on state forest maps provided by Landeskompetenzzentrum Forst Eberswalde (LFE) for Brandenburg and CORINE Land Cover data (European Environment Agency, 2016) for Northern Bavaria. The DEMs show considerable surface disturbances by land in some parts of the study regions use (e.g., in mining or military training areas), which we delineated during the mapping based on the DEMs and topographic maps to be able to specifically consider such areas in further GIS analyses. Because of differences in data availability, we followed slightly different RCH mapping procedures in the two study regions.
For the Brandenburg study region, we mapped the location and diameter of specific RCHs in three consecutive steps. First, we examined the occurrences of RCHs in a 1 km 2 grid for all forest areas based on the shaded relief map (SRM) visualization of the 1 m DEM provided by the LGB via a Web Map Service (WMS). Cells of the 1 km 2 grid with low and high numbers of RCHs were identified from the WMS SRM with a scale of 1:4,000. In the second step, specific RCHs were mapped for areas with high numbers of prospected sites using a combination of automated mapping by a template matching algorithm  and manual postprocessing using a principal component analysis (PCA)-SRM visualization (Devereux, Amable, & Crow, 2008) at a scale of 1:3,000. DEM derivatives based on elevation data aggregated to a 2 m grid were used in both methods. In the third step, we manually mapped specific RCHs based on the 1 m WMS SRM at a scale of 1:3,000 for areas with low numbers of prospected sites. Finally, we revised the mapping for all areas based on the 1 m WMS SRM displayed at scales of 1:2,500 and 1:1,500. RCHs were mapped as circular polygons along the outline of the RCH platform, that is, the area within the surrounding ditch. Sites too small for a clear identification of their extent were assumed to have a diameter of 8 m, which corresponds to the minimum hearth site diameter mentioned in historical forest regulations for Prussia (cf. Raab et al., 2019). For the area of the Tauersche Forst in southeastern Brandenburg (Figure 1), we incorporated the RCH data set described by Raab et al. (2019).
For the Northern Bavaria study region, the mapping was based on the SRM visualization of the 1 m DEM provided by the BVV, which could be visualized in a web viewer but could not directly be integrated into ArcMap. Therefore, instead of mapping the location of specific sites, we recorded the number of RCHs in the cells of a 1 km 2 grid along with additional information on the RCH morphometry assigned as attributes to the grid cells. To allow for further analyses of the RCH distribution, we, in addition, generated a RCH point data set based on this information. Therefore, points representing approximative RCH locations were randomly generated within the forest areas in the grid cells according to the number of RCHs recorded from the SRM of the area.

| Mapping quality assessment
The small-scale quality variations of RCH mapping from LIDAR data available for the Brandenburg study region could be estimated based on the results of previous studies. Detailed ground-truthing studies Raab et al., 2019) show a mapping sensitivity of 40-60%, with a better mapping quality for areas with large and regularly distributed RCHs and pine forest cover, compared with areas with small and irregularly distributed sites and deciduous forest cover. We used ground survey data from the 0.4 km² area described by Bonhage et al. (2019) to assess the RCH mapping quality based on the DEM derivatives used for large-scale RCH mapping, that is, the PCA visualization of the aggregated 2 m DEM and the WMS SRM visualization of the 1 m DEM. The area is covered by oak and pine forest and dominated by sandy substrate and even topography.
For Northern Bavaria, the small-scale variations of mapping quality was assessed by comparing the DEM-mapped RCHs to RCHs mapped in a field survey described by Zenger (1972) for a 16 km 2 area. Field survey RCHs were therefore digitized from the overview map provided by Zenger (1972) after georeferencing the map. The survey area is mainly covered by pine forest with undergrowth of varying density and covers different geologic units (see Table 1). The original anticipated ground point density of the DEM for this area, as reported by the data provider, is at least one point per m 2 . To allow for a detailed, spatially distributed analysis of the results, we mapped the locations of specific RCHs for this area.
To perform a basic assessment of the larger-scale mapping quality, we validated the mapping results against field surveys in several forest areas and mainly focused on verifying or refuting mapped RCHs. No regular survey pattern was used but we recorded additional sites in the vicinity of previously mapped

| RCH distribution-GIS analyses
The spatial distribution of the RCH density was analyzed in relation We determined the altitude, slope and aspect values for this aggregated DEM. Subsequently, the terrain roughness was characterized by determining the standard variation of slope values within a 5 × 5 cell moving window following Frankel and Dolan (2007). With this method, terrain roughness was described independently from the overall slope (Berti, Corsini, & Daehne, 2013) within an area of 1 km 2 to reflect the effort of transporting charcoal from the hearth site related to relief variations. The morphometric parameters were then transferred from raster datasets to the mapped RCH datasets, and the distribution of mapped RCHs with respect to the morphometry was analyzed.
The locations of historical industries that represented potentially significant consumers of charcoal or competing consumers of firewood were integrated into the GIS database. For Brandenburg, we assembled information on metal works, glassworks, lime kilns and tar kilns based on the historical literature and gazetteers (e.g., Bratring, 1804;Cramer, 1872Cramer, -1809Büsching, 1775;Enders, 2013). Additional historical T A B L E 1 Mapping quality assessment for the ground-truthing areas in the Tauersche Forst, Brandenburg, and in the forest areas west of Weiherhammer, Northern Bavaria (see Figure 2) Brandenburgfield-mapped RCHs: n = 120    Götschmann (1986) and Hirschmann (1997). The databases and catalogs of the Bavarian State Offices for Monument Protection (BLfD) were used to localize additional iron, metal and glassworks and lime and tar kilns. The distances between the RCHs and industrial sites were determined using the resulting datasets for both study regions.

| RCH distributionstatistical analyses
To test for systematic distribution patterns of RCHs, we compared the mapped RCH datasets with randomly distributed point datasets.
Randomly distributed points were generated within a minimum distance of 25 m of each other within the forest areas. We compared the distances of RCHs and random points to industrial sites and waterbodies with Mann-Whitney U tests and the frequencies of mapped RCHs and random points for examined landscape units with chi-squared tests in R 3.5.1 (R Core Team, 2018).
Based on the results of exploratory data analyses and distribution pattern tests, we combined the environmental background variables in binomial logistic regression models, using the occurrence of RCHs in a cell as the dichotomous dependent variable of the models. The probability of RCH mapping for a 1 km 2 area was modeled as follows: Note: The RCH distribution among units significantly (p-value <.005) differs from a random distribution in both study regions. The superscripted symbols for the units show the petrography and soil class assignments for the logistic regression analysis and model (see Table 5). Abbreviation: RCHs, relict charcoal hearths. categorical variables (see Table 2). For the Northern Bavaria study region, we included all the variables into the model, whereas for Brandenburg, we did not consider the altitude and distance to waterbodies relevant variables. Logistic regression was carried out and evaluated in R 3.5.1 using the glm (generalized linear model) function and the ROCR package (Sing, Sander, Beerenwinkel, & Lengauer, 2005

| Large-scale RCH distribution
In the Brandenburg study region, we mapped more than 41,000 RCHs in addition to the almost 6,000 RCHs in the data set from Raab et al. Within the charcoal production fields, RCHs occur in a large variety of distribution patterns, from a nearly uniform distribution over large areas to frequent "microclusters" (see Raab et al., 2019) and larger isolated clusters of several RCHs. However, the mapping results do not reveal clear relationships between these spatial distribution patterns and the environmental background structures or RCH diameters.
In the Northern Bavaria study region, we recorded more than 47,000 RCHs, that is, a similar number of sites to that mapped for Brandenburg despite the considerably smaller extent of the study region and forest areas. The RCH densities are clearly highest in the westernmost part of the study region, south of Nuremberg, where we found more than 20,000 sites over an area of only 400 km 2 . RCH densities of more than 50 sites per km 2 also occur in other charcoal production fields, mainly in the vicinity of modern-age ironworks (e.g., Auerbach, Weiherhammer, Bodenwöhr, Figure 3). Lower RCH densities up to 25 sites per km 2 were mapped in other forest areas in the vicinity of historical industrial sites, that is, around the accumulation of medieval to early modern iron hammer mills in the southern part of the Vils River valley ( Figure 3) and in the eastern part of the study region (see Figure 1). Very few RCHs were recorded on the heavily disturbed military training areas in the western part of the study region.
We observed a large variety of RCH morphologies and sizes over the study region. In the areas with considerably high RCH densities, | 551 on slopes were found in high density in the area west of Auerbach and southwest of Weiherhammer ( Figure 3) and dominated in most other parts of the study region with lower RCH densities. Especially in the northern parts of the study region, we also observed RCHs shaped as platforms within a surrounding elevated ridge as described by Hardy and Dufey (2015) and Swieder (2018), and they were often located on flat hilltop positions or on level areas within sloped terrain. There is a high level of variability in the diameters of RCHs on flat terrain, and they range between a few meters and up to approximately 25 m. The slope platform RCHs show less variability in their diameters than the flat-terrain RCHs.

| RCH distribution with respect to the environmental background structures
The distributions of mapped RCHs with respect to soils and geology significantly differ from random distribution patterns for both study regions (p-values <.001, Table 2, Figure 4). In the forest areas in Brandenburg, the geology is clearly dominated by sand, and 77% of the randomly distributed points in the forest areas ("expected RCHs") are located on predominantly sandy substrate. However, 93% of the mapped RCH sites are located on sandy substrate while mapped RCHs are clearly underrepresented compared with the expected RCHs on the silty and clayey substrate and in peat areas. Although the soil quality is poor in most of the forest areas, a concentration of RCHs in areas with poorer soil quality is observable (Table 2, Figure 4a). There are clearly more mapped than expected RCHs in Podzol and Arenosol areas, and there are considerably fewer mapped than expected RCHs in Retisol and Luvisol areas.
The RCH distribution over petrographic units (

| Prediction models for the RCH mapping probability
The coefficients and quality measures of the logistic regression models for the RCH mapping probability are listed in Table 5.  Table 2. Abbreviation: AUC, area under the receiver operating characteristics curve.
studies that conducted a systematic ground-truthing of DEM-based RCH mappings (Hardy, 2017;Ludemann, 2012;Raab et al., 2019) and similar anthropogenic relief features (Trier et al., 2015). The relatively low mapping sensitivity found for the Northern Bavaria ground-truthing area is most likely related to the lower original ground point density in the DEM for this region (Figure 1). Smallscale spatial variations in the mapping quality were observed within both ground-truthing areas of our study, and a detailed visual analysis of the DEMs together with field surveys suggests that these variations are mainly related to differences in the vegetation cover characteristics, which presumably affect the LIDAR ground point density and, therefore, the DEM quality (Reutebuch, McGaughey, Andersen, & Carson, 2003;Spaete et al., 2011). Considering such effects, we observed only a slightly decreased mapping sensitivity in the RCH mapping from a 2 m aggregated DEM compared with the mapping from the original 1 m DEM. A bias in the mapping sensitivity related to feature size, that is, higher rates of missed features for smaller RCH diameters, was found in detailed analyses by Raab et al. (2019) and is also reflected in the validation against the ground survey data for our mapped RCHs. On the landscape scale, this phenomenon might also result in spatially varying mapping sensitivity, with lower proportions of RCHs captured in areas with characteristically smaller sites.
Field surveys for ground-truthing are mostly limited to small areas and mainly carried out for study areas with heterogeneous conditions and good feature preservation. Therefore, the mapping sensitivities found in such approaches cannot be directly transferred to DEM-based mapping at the landscape scale. Validation of DEMbased mapping against field surveys and archaeological records does not reflect a large-scale spatial heterogeneity of mapping quality, but validation against historical maps indicates that the proportions of RCHs that can be mapped from DEMs differ within our study region.
Generally, several additional and often interrelated effects on the mapping quality need to be considered in the interpretation of largerscale mapping results: -LIDAR-based DEMs for large areas are often composed of multiple datasets recorded in several individual flight campaigns over several years (Baltsavias, 1999), which can result in heterogeneous elevation datasets, even if the final processed DEM is provided at a continuous spatial resolution. We observed spatial differences in RCH visibility in DEMs related to documented differences in the ground point density in the Northern Bavaria study region. We also noted differences in the representation of small surface structures within the Brandenburg study region; however, we were not able to directly relate these differences to information on data acquisition or processing given by the WMS DEM provider (LGB, 2019).

| RCH distribution with respect to charcoal consumers
High densities of RCHs in the vicinity of major, especially modern, ironworks are noticeable for both regions (Figure 3). In the Brandenburg study region, this phenomenon is also reflected in the relatively small mean distances between RCHs and metal works; whereas a statistically significant association between RCH locations and historical industrial sites is not observable for Northern Bavaria.
The less clear associations for the Northern Bavaria study region presumably relate to the high spatial density of historical industrial sites and the more heterogeneous geology and geomorphology. The relationships between RCHs and industrial production centers might not be adequately captured in our analyses due to variations in charcoal consumption between different ironworks. The historical literature for the Brandenburg study region states that the metal works in Peitz, Zehdenick, Gottow and Eberswalde clearly dominate over other metal works in the region regarding their production and charcoal consumption (Bratring, 1804(Bratring, -1809. The production times and charcoal consumption of the older ironworks in Northern Bavaria can hardly be quantified despite the detailed analyses of the historical archives by Götschmann (1986).
In addition to the charcoal production fields related to iron-and glassworks, our results clearly show that high densities of RCHs occur in several areas without an apparent connection to historical industries in both study regions, most obviously in the forest areas north and east of Berlin and southeast of Nuremberg. This finding reflects that in addition to major industrial sites, clusters of small industries and households need to be considered important charcoal consumers and that long trade routes, presumably along waterways, occurred. Large RCH fields have mainly been related to major iron production areas in most previous studies (e.g. Groenewoudt, 2005;Raab et al., 2015;Rutkiewicz et al., 2019), whereas intensive charcoal production for smaller consumers has only been assumed for some areas, such as the Black Forest (Ludemann, Brandt, Kaiser, & Schick, 2017) or the Besançon area (Dupin et al., 2017). Furthermore, we recorded relatively low RCH densities in some forest areas for which intensive charcoal production is assumed, most obviously along the Vils River valley (Figure 3), but also around historical iron-and glassworks in the easternmost part of the Northern Bavaria study region.
With a mean distance of 15 km to industrial sites of more than 7 km, the RCH distribution in the Brandenburg study region clearly suggests charcoal transport over large distances, although previous analyses for several charcoal and iron production areas indicate that charcoal-consuming industries were preferably located in the immediate vicinity of charcoal production areas in forests (as summarized, e.g., by Schmidt et al., 2016). Information on specific charcoal production areas associated with industrial sites is available from the historical literature for some of the ironworks in Brandenburg and confirmed transport distances of more than 10 km (e.g., Bratring, 1804-1804, detailed discussion in Müller, 2017. In the Northern Bavaria study region, analyses of wood consumption for some individual ironworks also suggest charcoal transport over distances of up to 15 km during periods of wood scarcity (Hirschmann, 1997). The assumption that the industries' charcoal delivery areas were expanded, especially in periods of high charcoal demand and scarce supply, which was also made by Rutkiewicz et al. (2019) and Pasmore (1965), is confirmed by the widespread RCH fields mapped around several major and longexisting ironworks in our study regions.

| RCH distribution with respect to geology, soil, and morphology
The relationships between the RCH distribution and the substrate characteristics differed between the study regions. A correlation between high RCH densities and sandy parent material within the forests in Brandenburg is observable ( that this is mainly a consequence of regional differences in charcoal production practices, which were adapted to the environmental conditions. Higher RCH densities on the sand and sandstone weathering products are almost certainly related to the low effort of hearth construction and, therefore, a dominance of single-use charcoal burning sites. Lower densities in other areas reflect the repeated use of hearth platforms. A spatial concentration of RCHs in valley bottoms, settlement areas and traffic routes, which leads to higher proportions of disturbed sites, might further affect the lower mapped RCH densities in limestone and crystalline rock areas. Historical instructions for charcoal production support several aspects of this interpretation. Several such documents note that charcoal hearth locations should be chosen carefully with a view to minimizing the work involved in site preparation (Berg, 1860;Klein, 1830;Krünitz, 1773Krünitz, -1858. Furthermore, they note that homogeneously structured substrates are preferable for hearth operations while coarse and fractured substrates should be avoided; in addition, clays or clayey loams have considerably bad substrate conditions for charcoal burning (Berg, 1860;Krünitz, 1773Krünitz, -1858. It was also noted that pure sand provides unfavorable conditions that can be overcome by an admixture of topsoil material (Berg, 1860). Similarly, the observed underrepresentation of RCHs on Gleysols, Fluvisols and Histosols, and stagnic soils probably results from hearth location preferences because instructions for charcoal production consistently note that wet sites have unfavorable conditions for hearth operations (Berg, 1860;Klein, 1830), especially because of the higher groundwater tables during the last centuries (Kaiser et al., 2015). Among other forest areas, those with poorer soil quality (Podzols and Arenosols) clearly show higher RCH densities. Higher RCH densities in poorer soils were similarly observed by Rutkiewicz et al. (2019); however, the soil distribution over the total landscape was interpreted in this study and the pattern might largely reflect general site preferences for forest use. The relation between RCH concentration and poorer soils in Brandenburg might be related to the specific location and spatiotemporal contingency of forest areas in relation to soil properties. In fact, most of the large, historically old (i.e., continuously forested) state forest areas in the region are located in areas with poorer soils (Wulf, 2004). Forests with decent soil quality, for example, Retisols and Luvisols, are often fragmented and distributed over areas dominated by agricultural land use. High RCH densities in these areas might further be related to wood type distribution under the assumption that hardwood was preferably used for purposes other than charcoal production. This assumption is supported by the observation of a clear dominance of pine wood in RCHs on the poor soils of the Tauersche Forst .
Therefore, our results do not reflect higher densities of RCHs in areas with better wood growth. This finding is in contrast to the analyses by Schmidt et al. (2016) and the frequent observation of a preferred use of hardwood in hearths, which allowed for the production of higher-quality charcoal.
The mapped RCHs in both study regions are clearly concentrated on flat and even landscape positions, but the distribution patterns related to relief vary slightly between the regions, presumably indicating differences in hearth site location and operation. In Brandenburg, the RCHs are concentrated in areas with slopes <0.5°but they are only slightly overrepresented at slopes between 0.5 and 1°. This finding suggests that colliers preferred flat positions over positions on very slightly inclined surfaces, which might be related to the observation that wood for the hearths was commonly stacked directly on top of the undisturbed soil surface, as shown by stratigraphic analyses of RCH soils in the Tauersche Forst  prohibited since the 14th century (Walz, 1986). In addition to the low numbers of RCHs mapped in these forest areas, this ban on charcoal production might also contribute to the exceptionally high numbers of RCHs recorded in the forested areas south of Nuremberg. Lowland. Higher spatial variations in the variables affecting the RCH distribution might exist within the more heterogeneous mountainous charcoal production areas of Central Europe, where the variety of RCH morphometries and characteristic site locations suggest a high regional variety of charcoal production practices and traditions.

| Implications for interpreting RCH distribution patterns
The results of our study confirm that comparisons of the number of hearths in these sites. Furthermore, the remains of only upright charcoal hearths can usually be captured by DEMs, and charcoal production in pits rarely leaves structures that can be captured by elevation data or even field surveys. These phenomena need to be considered when interpreting the number of RCHs. Although the general transition from charcoal burning pits to upright hearths can be assumed to have occurred around the 10th to 11th century (Groenewoudt, 2005), the continued use of pit hearths at least until the High Medieval Period is described for several regions (Deforce, Vanmontfort, & Vandekerkhove, 2018;Groenewoudt, 2005;Risbøl et al., 2013). For Northern Bavaria, records of charcoal acquisition of several iron hammer mills report that up to one-third of the total consumed charcoal was pit charcoal, which indicates that pit and upright hearths were operated in parallel, even in later centuries (Götschmann, 1986).
Generally, the distribution of mapped RCHs is dependent not only on the actual original distribution patterns of hearth sites but also on the spatial patterns of feature preservation and visibility.
Both factors are difficult to separate; however, detailed information and a thorough understanding of the industrial and technical history and specific charcoal production practices of a region can provide a more detailed understanding of regionally characteristic preferences for hearth site locations and usage.

| CONCLUSIONS
The results of DEM-based mapping in our study confirm that RCHs are widespread legacies of past land use. Capturing small anthropogenic relief features, such as RCHs, from LIDAR-based DEMs is feasible, even for large areas with different environmental characteristics; however, the results of DEM-based mapping are clearly affected by small-and largescale variations in mapping quality. We suggest that this spatial heterogeneity in mapping quality along with general measures for mapping sensitivity and accuracy should be considered when comparing the results of DEM-based mapping between regions. Our mapping results confirm that the spatial density of RCHs is high in many forest areas, and they also show that high RCH densities are not necessarily associated with major historical iron production sites. GIS-based analyses of RCH distribution patterns in our study regions show a close relationship between charcoal production and historical industrial sites for regions with homogeneous environmental backgrounds but a stronger dependence of the RCH distribution on the relief and geology in more heterogeneous landscapes. These differences between study regions show that comprehensive analyses of database quality and the environmental and cultural background parameters of a region are prerequisites for meaningful interpretations of the spatial distribution patterns of anthropogenic relief features.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.