Lithic artifact assemblage transport and microwear modification in a fluvial setting: A radio frequency identification tag experiment

River processes are widely assumed to have impacted the integrity of lithic assemblages when artifacts are found in fluvial sediments, but the specifics of these influences remain largely unknown. We conducted a real‐world experiment to determine how the initial stages of fluvial entrainment affected lithic artifact assemblages. We inserted replica artifacts with radio frequency identification tags into a gravel‐bedded river in Wales (UK) for seven months and related their transport distances to their morphology and the recorded streamflow. In addition, nine artifacts were recovered at the end of the experiment and analyzed for microwear traces. In sum, our results show that in a gravel‐bedded river with a mean discharge of 5.1 m3/s, artifact length and width were the main variables influencing artifact transport distances. The experiment also resulted in characteristic microwear traces developing on the artifacts over distances of 485 m or less. These results emphasize the multifaceted nature of alluvial site formation processes in a repeatable experiment and highlight new ways to identify the transport of replica Paleolithic material.


| INTRODUCTION
Lithic artifacts are a main source of information for reconstructing the movements, technological behaviors, and diets of ancient hominins.
Their morphology, location, and spatial associations provide archaeologists with key data sets to test hypotheses. Paleolithic artifacts are frequently found embedded within Pleistocene fluvial deposits, reflecting both anthropogenic behavior and landscape taphonomy, thus making the latter crucial repositories for information on past hominins (van den Biggelaar, Balen, Kluiving, Verpoorte, & Alink, 2017;Bridgland & White, 2014;Bridgland et al., 2006;Chauhan et al., 2017;de la Torre, Benito-Calvo, & Proffitt, 2018;Westaway, Bridgland, Sinha, & Demir, 2009). At the same time, experimental archaeology is a valuable instrument in the researcher's investigatory toolkit (Eren et al., 2016;Lin, Rezek, & Dibble, 2018) and has been used periodically over the last half-century to determine questions such as whether lithic artifacts in a river behave as normal clasts or whether their unique shape and anthropogenic insertion points substantially modify their entrainment, movement, deposition, and abrasion.
Field experiments were first initiated by Isaac (1967) and Schick (1986), to understand the impact of hydrological sorting of lithic assemblages in ephemeral rivers in East Africa. Their findings, that smaller artifacts were selectively transported downstream, were elaborated upon by Petraglia and Nash (1987) who showed that a number of factors In combination, these previous studies have suggested that a number of main points remain unresolved and/or require further study: 1. How do artifact metrics affect their transport, entrainment, and deposition in the variety of river types known from Pleistocene archives (e.g., meandering, braided; cold-climate, and temperate)?
What is the best predictor of transport distance (if any)? Here, we present an experiment using radio frequency identification (RFID) tagged replica lithic artifacts inserted into a gravel-bedded, meandering river in a temperate, and mid-latitude environment. RFID tagging improves artifact recovery allowing for enhanced postentrainment analyses (Houbrechts et al., 2015;Lamarre, 2005). The method is derived from geomorphological studies and supports high recovery rates throughout the project area (Hassan & Bradley, 2017). The approach also allows for artifact surfaces to be largely unmodified (e.g., it avoids the use of artifact marking with paint to enable recovery and identification), permitting artifact modifications such as edge damage and microwear to be meaningfully studied. It also generates accurate artifact transport distances that can be linked to artifact size measurements, streamflow characteristics, and morphological modifications. Combined, it allows artifact positional changes to be monitored over time, and for potential distinctions between use-wear/retouch and river modifications to be explored.

| Background
Field experiments were carried out on a 1.14 km section of the  (Foulds, Griffiths, Macklin, & Brewer, 2014). The surrounding catchment terrain has a maximum height of 612 m above ordnance datum and is primarily comprised of grassland mainly used for forestry and sheep husbandry. During the experiment, the river generated a mean annual river discharge of 5.1 m 3 /s and a maximum daily mean discharge of 72.6 m 3 /s (measured at the Pont Llolwyn gauging station; 52.374642°N, −4.072693°W).
For the past 200 years, the section of the river has been characterized by aggradation related to historical mining. Predominantly aggrading regimes are likely to have been common during the transitional periods of Pleistocene climate cycles (after Bridgland, 2000), when sediment-supply rates were increased due to cold/cool climates with reduced vegetation cover (Lewin, Bradley, & Macklin, 1983). As a result of the recent aggradations, the current valley floor is filled with Holocene alluvium, dominated by sandy gravels predominantly derived from local impermeable Silurian shales and gritstones, with a high proportion of disc-and blade-shaped clasts with a mean flatness of 0.44 (c/b; . The coarse-grained river sediments have a median diameter (D 50 ) of 40 ± 10 mm (min 17 mm and max 53 mm) and a 95th percentile grain diameter of (D 95 ) 96 ± 27 mm (min 43 and max 2. The catchment's topography makes the Yswyth a flashy river; prone to rapid rises and falls in level depending on rainfall, thus promoting artifact entrainment and deposition and facilitating fieldwork. 3. The river has been used in previous artifact transport and geomorphological studies, providing comparative and contextual data (Brewer, Johnstone, & Macklin, 2009;Brewer, Maas, & Macklin, 2001;Harding et al., 1987;Hosfield & Chambers, 2004a, 2004bHosfield et al., 2000).

| MATERIALS AND METHODS
Replica artifacts were produced by experienced knappers, covering different Paleolithic forms including handaxes, blades, flakes, and cores (Table 1; Figure 2) from a range of European raw materials (partly from Lengyel and Chu, 2016). In addition, the handaxes were previously used in a well-documented experiment to butcher fallow deer (Dama dama; Machin, Hosfield, & Mithen, 2007). Artifacts were individually wrapped in aeroplast after production to preserve the integrity of their surfaces and edges. A 15 × 4 mm niche was then removed from the center of the artifacts with a water-jet cutter that incises material with a focused jet of water mixed with abrasive grit. An HPT12 Passive Integrative Transponder tag (Biomark; 9 × 2 mm; 134.2 kHz) with a unique serial code was inserted into the niche and fixed with epoxy resin. The maximum length, maximum width, and maximum thickness of the replica artifacts were recorded to the nearest 1 mm (Andrefsky, 2005). Weight was recorded to the nearest 0.1 g. Artifacts were then photographed on their dorsal and ventral sides.
Field experiments were conducted between March 13-15, 2017, August 3-6, 2017, and again in September 15-18, 2017 During the first field visit, a single "scatter" of 454 artifacts of differing lithologies was emplaced in a regular grid on a riverbank, where artifacts were positioned 25 cm apart from each other with a randomly oriented long-axis. During the second visit, another 114 artifacts were emplaced in a single pile (25 cm radius) simulating a knapping scatter.
Artifact locations were recorded with a Biomark HPR Plus Reader using a BP Plus portable antenna that has an integrated global positioning system (GPS) unit (horizontal accuracy of ±3 m) and is able to detect tags underwater and/or beneath gravels from a distance of up to 45 cm (Cassel, Piégay, & Lavé, 2017). During subsequent monitoring visits, the surface of the project area ( Figure 1) was scanned with the RFID antenna in 2 m strips traversing the river channel and floodplain to ensure the entire riverbed and banks were appropriately covered (Chapuis, Bright, Hufnagel, & MacVicar, 2014). When found, artifact locations, the date, time, and GPS coordinates were automatically recorded and if possible, orientation was recorded with a transit compass. Where visible, artifacts were recovered at the end of the experiment.
Artifact locations, measurements, and orientation values were later compiled and imported into QGIS (2.18). Equivalent artifact location points (pre-and posttransport) were matched. These were then converted to distances by creating a vector file of artifact travel distances using the R package "riverdist" which simulates the most parsimonious artifact travel path through a river. All subsequent statistical analyses were performed with SPSS 22.
Nine transported artifacts were further selected for microwear study based on their morphology and distance traveled (between 0 and 485 m). They were cleaned with 10% HCL solution for 20 min, rinsed with water, and immersed in a 10% KOH solution for 20 min.
For an extensive description of the microwear methodology, see van Gijn (1989van Gijn ( , 2010 Abbreviation: SD, standard deviation. and transport distances were tested with bootstrapped single linear regressions. The data used in these analyses were artifact measurements and the total transport distances for all artifacts between March and September 2017. The experimental artifacts provided a robust data set to explore the effects of artifact measurements on transport distances during the initial stages of fluvial reworking since they were transported up to 485 m, were exposed to the same maximum hydrological flows and were inserted along the same gravel bar. Figure 3 displays scatter plots reporting the results of bootstrapped linear regressions comparing transport distance with maximum length, maximum width, maximum thickness, weight, elongation, and refinement and

| A recovery bias in transport distance analysis?
A total of 102 (28%) of the artifacts were unrecovered after their insertion at the experimental site. To assess if artifact recovery was biased, artifact measurement means were compared with a bootstrapped  Table 5 shows the measurements of recovered and unrecovered artifacts. Unrecovered artifacts were statistically significantly smaller by length (μ = 9 mm), width (μ = 7 mm), thickness (μ = 2 mm), and weight (μ = 21 g).
Differences in measurements of transported artifacts were statistically insignificant for artifact elongation and refinement. These results suggest that the recovery of artifacts was biased towards longer, wider, thicker, and heavier artifacts and may indicate that smaller artifacts were more readily transported out of the project area.

| Where do artifacts typically become deposited in fluvial environments?
Artifacts were regularly redeposited within the center of the channel at the first river bend downstream from the point of insertion, while artifacts that were transported longer distances were typically found isolated in small (c. 50 cm in maximum dimension) scours at the margins or downstream ends of gravel bars where water velocities suddenly dropped ( Figure 1). They were commonly buried within fine -grained sands and silts or found resting directly on, but embedded within, the gravel bar surfaces. This pattern was consistent during both subsequent monitoring periods.
To test potential clustering of artifacts, hot and cold spots were created using optimized hot spot analysis (OHSA; ArcMap 10.5.1).
OHSA aggregates presence/absence point data and identifies statistically significant spatial clusters of high values (hot spots) and low values (cold spots), using the Getis-Ord Gi* statistic. The analysis showed that there were a greater number of significant "hot" (high value) spatial clusters in September than August (Figure 5a,b). In addition, when performed between the different shape categories (blade and oblate; Figure 5c,d) and size categories (very coarse pebble and small cobble; Figure 5e,f), the spatial distributions of highvalue clusters are broadly comparable, suggesting that they did not behave any differently. It was not possible to run the analysis on the other shape categories (equant and prolate) and size categories (coarse pebble and large cobble) due to sample sizes < 60.
Artifact dispersal increased c. ten-fold from March to August (based on all clasts; Table 6 and Figure 6). However, the degree of spatial dispersal varied between clast size groups: from c. ×20 for the smallest clasts (coarse pebbles) to c. ×2 (large cobbles). While the samples sizes for these extreme groups were small, the data suggest a relationship between increasing artifact weight and a reduction in spatial dispersal during fluvial transport. Trends were less clear among the clast shape groups, although the prolate group showed the greatest dispersal. Since this clast shape is elongated and spherical (axis ratios: b/a < 0.67 and c/b > 0.67), it is possible that this "rugby ball" prolate shape was preferentially vulnerable to greater dispersal through a rolling motion, although this conclusion is tentative given the small sample size. The similar dispersal distances of the flattened (c/b < 0.67) oblate and bladed clasts suggest that degree of elongation (oblate: b/a > 0.67; bladed: b/a < 0.67) was not a significant factor.
T A B L E 2 Descriptive statistics of recovered artifacts by clast size and clast shape Udden-Wentworth (Wentworth, 1922)

| Under what fluvial conditions is artifact orientation altered?
Artifacts were primarily buried within the river channel during the course of the experiment. Attempts to recover them required excavation, which was a time-consuming endeavor (c. 1 hr per artifact).
Due to time constraints, this was seldom performed except for some of the farthest transported artifacts. When performed, the excavation process rendered orientation data unreliable due to the low visibility through the water and due to the artifacts' movement during excavation. Therefore, the number of artifacts with recognizable F I G U R E 4 Clast transport data, (a) subdivided by clast size categories; (b) subdivided by clast shape categories. Clast sizes after the Udden-Wentworth (Wentworth, 1922) fig. 3.4). Artifacts defined according to largest axis. Clast shapes after Zingg (1935). Artifacts defined according to b/a and c/b ratios (Jones et al., 1999, fig. 3 fig. 3.4). Artifacts defined according to the largest axis.

| How does transport affect artifact damage?
All the analyzed pieces were affected by their stay in the river. Although polish. This directionality varied between the different spots ( Figure 7b; Figure 9a). The polish was not observed directly on the edge but was found spread both along and away from the edge. This polish was disordered, varied in texture, and lacked indicative characteristics. The polish was generally better developed on protruding parts, for example, the ridges (arêtes). All of these characteristics clearly distinguished these traces from anthropogenic use-wear traces. Where residual use-wear traces from the previous butchery experiments were still visible, these were partially obscured by the PDSMs, but were still clearly distinguishable (Figure 9). On many of the pieces, rounding from the current experiments was also visible.
Generally, it was only lightly developed along the edges, but more strongly developed on the ridges and protruding parts of the artifacts ( Figure 9). The results of the bootstrapped linear regressions showed that transport distance was significantly predicted by maximum length (R 2 = 0.010; Figure 3; p < .05) and maximum width (R 2 = 0.020; Figure 3; p < .05) though the R 2 values are low suggesting that other factors are involved in artifact transport. Weight, thickness, elongation, and refinement were statistically insignificant predictors of artifact transport (Figure 3; p > .05). These results suggest that transport distances of lithic artifacts in gravel-bedded rivers are partially, albeit weakly, dependent on overall length, and width of artifacts. Shorter, narrower artifacts tend to be transported farther than longer and wider artifacts. The results also indicate that weight and thickness are not significant predictors of artifact transport, suggesting that artifact transport is better predicted by overall dimensions than weight (Byers et al., 2015;Hosfield & Chambers, 2004a). Given that unrecovered artifacts were consistently smaller than recovered artifacts, and that this may in part be due to them being transported out of the experimental area, the effect of size on transport distance may be under-reported. Elongation and thickness were also insignificant predictors of artifact transport distances indicating that "dimensionless" artifact shapes (e.g., short squat flakes or long thin flakes) did not play a statistically significant role in transport. The results of the Kruskal-Wallis and ANOVAs showed that transport distance were not significantly different when grouped by artifacts size and shape classes regardless of monitoring periods which had different maximum discharges.
The results of the OHSA showed a greater number of significant "hot" (high value) spatial clusters in September than August (Figure 5a,b).
This is likely because more of the August scatters were concentrated in the same location (close to the insertion point) and thus relatively few hot spots. By contrast, as the material became more dispersed in September, the hot spots (still close to the insertion point) were more apparent.
intimately related to each other, the results of this study confirm that length and width are better predictors of transport distance than weight (Wilcock, 1997). However, while the relationships are statistically significant, the R 2 values for length and width are low (0.010 and 0.020, respectively) indicating that artifact size accounts for < 2% of variation in transport distance.
The differences between this study's results and those of Hosfield and Chambers (2005) are probably in part due to their lower F I G U R E 6 Degree of artifact concentration/dispersal, by observation period (March > August > September) and groupings (clast shape and clast size). Clast sizes after the Udden-Wentworth (Wentworth, 1922) grain size scheme: coarse pebble (16-32 mm); very coarse pebble (32-64 mm); small cobble (64-128 mm); large cobble (128-256 mm; Jones et al., 1999, fig. 3.4); Clast shapes after Zingg (1935). Artifacts defined according to b/a and c/b ratios (Jones et al., 1999, fig. 3.6). Dispersal area data calculated from Standard Distance statistics, generated using Standard Distance analysis in ArcMap 10.5.1; standard distance = circle radius (circle size set at 1 standard deviation; i.e. c. 63% of data points). For sample sizes see Table 6. One data point was not included in this analysis for March due to imprecise GPS coordinates. GPS, global positioning system T A B L E 7 Degree of spatial separation/overlap between artifacts, by observation period (August and September) and groupings (clast shape and clast size) Note: Clast sizes after the Udden-Wentworth (Wentworth, 1922) grain size scheme: coarse pebble (16-32 mm); very coarse pebble (32-64 mm); small cobble (64-128 mm); large cobble (128-256 mm; Jones et al., 1999, fig. 3.4); Clast shapes after Zingg (1935). Artifacts defined according to b/a and c/b ratios (Jones et al., 1999, fig. 3.6).
recovery rates but also to the longer timescale of their study that may have provided more opportunities for the materials to become mobile, including larger artifacts Ferguson, Bloomer, Hoey, & Werritty, 2002). This study used handaxes, flakes, and blades but the output regimen of the Ystwyth was never above 30 m 3 /s, while between 2000 and 2003 (Hosfield & Chambers 2004a) river discharge reached a maximum of 75.62 m 3 /s.

Hosfield and Chambers' (2004a) finding of an insignificant relation-
ship between artifact length, width, and thickness and horizontal displacement may, therefore, be the result of an output regimen too extreme to discriminate between small ranges of artifact sizes.
Many previous studies provide data allowing for direct comparison of horizontal displacement of artifacts and the results of this experiment are within these reported ranges. Among artifacts deposited on the riverbanks (Stations E and H) Petraglia & Nash (1987) reported that scatters were buried "in situ." Schick (1986) also reported seven experiments (14)(15)25,28,and 34) where artifacts emplaced on the banks of rivers were "minimally disturbed" (i.e., artifacts stayed mostly in place with a maximum transport of 7 m). However, Schick (1986) described other bankside experiments (Sites 13 and 20-22) as more heavily modified, reporting that transport values increased to as far as 19 m in one case. These differences were attributed to scatters being located lower and closer to the active channel than other scatters (Schick, 1987).
Experiments such as the current one where artifacts were placed within an active channel (either directly or on a bar) showed greater transport distances. Schick (1986) found that such sites were scoured, truncated, or experienced "major disturbance," resulting in transport of up to 90 m (Sites 1c,19,23,24,(26)(27)and 36). Petraglia and Nash (1987) reported similar results (Stations C and D) with in-channel artifacts dispersed up to an average of 33.1 m. Hosfield and Chambers (2004a) reported a maximum transport distance of 84.95 m while Harding et al. (1987) Hosfield and Chambers (2004a).
Comparing discharge values of the River Ystwyth to those of other rivers used in earlier experiments indicates that daily mean flow and average flows were low compared with those previously reported, though the high flows for this study of 30 m 3 /s were comparable with those of Petraglia and Nash (1987) and Chu (2016).  Potts, 1994;Schick, 1986;de la Torre et al., 2018). The absence of smaller artifacts after fluvial disturbance that has been recorded in some experiments of this kind has generally been interpreted as a possible indication of downstream fining and a potential means of detecting archaeological assemblage reworking (Isaac, 1967;Malinsky-Buller, Hovers, & Marder, 2011;Schick, 1987;de la Torre et al., 2018). Other experiments, however, have found no such association, suggesting that artifact dispersal is a highly complicated process influenced by the river's output regimen (Petraglia & Nash, 1987), the discard location of the artifacts (Dennell, 2004;Harding et al., 1987, p. 250), and the nature of the bedload/rugosity (Hosfield & Chambers, 2004a). This experiment has broadly, but not conclusively, supported the former view, highlighting

| Limitations and future work
The application of the Ystwyth experiment to the Paleolithic record has some limitations due to the scope of the experiment. However, some of these may be overcome through further work: First, the short duration of the tracer studies gives insufficient information about the long-term/long-range effects of artifact transport, although this may be partially overcome by longer artifact transport studies combined with improved theoretical models (Hassan, Church, & Ashworth, 1992;Klösch & Habersack, 2018;Milan, 2013).
Second, these experiments only examined a single river type and their application to Pleistocene river types is therefore limited to comparable settings.
Third, artifact transport and modification patterns are based on a maximum transport distance of 485 m, and thus may be principally relevant to locally disturbed sites. Therefore, these results should not be generalized to more heavily transported data sets as longer transport periods may generate different results.
Fourth, there are undoubtedly more variables involved in artifact fluvial transport that require exploration, notably channel morphology, Fifth, microwear still remains a largely subjective appraisal however future advances in quantitative microscopy and tribology may help to accurately assess taphonomic signatures (Stemp, 2018).
Finally, active RFID and tracer systems may clarify the ultimate locations of artifacts outside search areas and be able to identify the exact timing and distance of their movements with reference to flow regimes (Cassel, Dépret, & Piégay, 2017).

| CONCLUSION
Particle kinematics in gravel-bed rivers is a complex process,