Aboriginal stone-walled intertidal fishtrap morphology, function and chronology investigated with high-resolution close-range Unmanned Aerial Vehicle photogrammetry

Stone-walled intertidal fishtraps surround the Australian coastline and are among the largest structures built by Indigenous Australians. Globally, fishtraps are considered important elements in food production, domestication, territoriality and ceremonial landscapes, yet the level of detail in documentation is highly varied and scholarly fishtrap knowledge sparse. Comparative analysis is currently restricted by a lack of detail and reproducibility in recording, hindering analysis of morphology, function and chronology. In this study we employ high-resolution close-range Unmanned Aerial Vehicle (UAV) photogrammetry and a suite of spatial information analytical techniques to investigate the Kaiadilt Aboriginal stone-walled intertidal fishtraps of Sweers Island, southern Gulf of Carpentaria, Australia. Tidal inundation modelling is undertaken to assess (1) fishtrap working range, (2) individual and simultaneous trap function, (3) seasonal functionality and (4) chronology based on function relative to sea-level history. Thirteen fishtraps were identified in the study area, ranging from 38m to 287m in length. Flow accumulation indicates that shape and placement of fishtraps reflects underlying topography. Inundation modelling shows that all fishtraps operate most efficiently at present mean-sea level (PMSL), indicating construction in the last 3500 years. Quantitative recording techniques, analytical procedures and terminology developed in this study provide an opportunity to improve approaches to recording large-scale stone features and standardise documentation of stone-walled intertidal fishtrap sites.


Introduction
Stone-walled intertidal fishtraps are some of the largest structures documented in the Australian archaeological record. Constructed with rock and/or organic matter, fishtraps are argued to be primarily designed to trap or control the movements of marine resources across tidal cycles in coastal or riverine contexts (Campbell, 1982;Dortch et al., 2006;Jeffery, 2013;Rowland and Ulm, 2011). For the purposes of this study, stone-walled intertidal fishtraps are defined as structures capable of controlling the movements of marine animals.
Due to the location of fishtrap structures in intertidal and riverine settings, access is often restricted and dependent on tidal movement, and in certain parts of the world the presence of marine predators can be hazardous to field researchers. Recording time and visibility is also controlled by tides, and can further be restricted by wind, causing swell and sediment to obscure structures. Such environmental factors, along with the impacts of recreational marine vessels, cause intertidal stone features to erode partially or completely, which underlines the urgency in recording remaining fishtrap structures (Elder et al., 2014;Memmott et al., 2008;Roberts et al., 2016;Rowland and Ulm, 2011;Rowland et al., 2014). Despite the urgency of documentation, and a global interest in fishtrap construction (e.g. Greene et al., 2015;Jeffery, 2013), most recordings consist of basic sketch maps of limited detail, with few quantitative data or photographic records (for exceptions see Coutts et al., 1978;Greene et al., 2015;Koivisto et al., 2018;Langouët and Daire, 2009;McNiven et al., 2012;O'Sullivan, 2004). Varied approaches to site recording has led to a proliferation of terms describing fishtrap attributes, which pose challenges for site management, comparison of sites, and the ability of fishtraps to be considered meaningfully in broader debates. This study focuses on intertidal stonewalled fishtraps and proposes a standardised high-resolution recording scheme for large-scale intertidal stone features, to improve knowledge of fishtrap construction, function, and age.

Background
In 2011, Rowland and Ulm published a comprehensive review of coastal and inland fishtraps and weirs in the state of Queensland, Australia. The review found that stone-walled fishtraps are generally situated on coastal points or estuaries sheltered from strong winds, and while limited, evidence indicated that organic traps and weirs were generally located inland. Multiple pens (or holding areas) are observed in the Torres Strait and Gulf of Carpentaria, while isolated single pen structures are found further south. Most coastal fishtrap structures across the state displayed an arc shape, and it was recognised that traps were constructed and utilised by both Indigenous and non-indigenous Australians (Rowland and Ulm, 2011). The authors concluded that the level of detail available in Queensland fishtrap recording was largely substandard and proposed standardised recording schemes with increased detail in documentation (Rowland and Ulm, 2011). Rowland and Ulm's (2011) findings apply to Australia more broadly, where site comparison is challenged by uncertainty in identification, a variety of recording techniques, and a wide range of associated terminology. Australian intertidal stone-walled fishtrap documentation is dominated by sketch maps to varied detail of fishtrap location, shape, and dimensions. More sophisticated documentation techniques and global navigation satellite systems (GNSS) have only been adopted in recent ground and aerial documentation strategies. While the recommendation by Rowland et al. (2014) to utilise Light Detection and Ranging (LiDAR) to capture large-scale coastal sites in detail has not yet been adopted in the Australian fishtrap context, various aerial recording techniques have been trialled. Low-level aerial photography was utilised in fishtrap site identification and analysis by Campbell (1982) at Hinchinbrook Island, Queensland, Dortch (1997 at Wilson inlet, Western Australia, and by Connah and Jones (1983) and Memmott et al. (2008) in the Gulf of Carpentaria. Photogrammetry, today a well-established technique in three-dimensional (3D) modelling (Sapirstein, 2016), has been sparsely applied in Australian fishtrap literature. The technique, allowing the generation of geometrically accurate photo-mosaics from which precise measurements can be retrieved, was used by Smith (1987) in a close range (< c.300 m) ground photo mosaic of a Bardi fishtrap, Dampier Peninsula, Western Australia, which provided a detailed map of the trap to scale. The anthropogenic inland stone-walled structures of Gunditjmara country, Lake Condah, southwest Victoria, have received the most detailed documentation to date. Van Waarden and Wilson (1994) used aerial photogrammetry (> c.300 m) to map the region, creating 1 m contoured topographic maps (1:5000), followed by Richards' (2013) detailed surface mapping using Real-Time Kinematic (RTK) Differential GNSS (DGNSS), and Builth's (2014) use of the same technology to create a 2 m × 2 m digital elevation model (DEM) for sites within the lava flow.
The various recording techniques applied in fishtrap studies have led to a wide range of terminology across the literature. While some studies describe stone-walled structures by physical composition (e.g. alignment by Dortch et al. (2006), continuous walls by Stockton (1975) or observed or perceived function e.g. barrier by Roberts et al. (2016)), the majority of studies focus on the morphology of structures. Aligning with Rowland and Ulm's (2011) findings, the arc (also described as U-shape or semi-circular), and circular terms are the most common shape characteristics applied across Australian fishtrap publications. Such morphological descriptors are problematic owing to their arbitrary and subjective nature, and risk neglecting or recording multiple traps as single features, and vice versa (e.g. one w-shaped trap or two v-shaped, or semicircular traps). Dimensions generally consist of a measure of the tallest and widest points of the trap, and an east-west and north-south measure of the enclosed area, but complete metrics are rarely presented for individual sites. Focusing on shape and size, fishtrap assessments generally neglect 3D aspects of structures, with the exception of Campbell (1982) who estimated holding capacity of the Scraggy Point fishtrap complex on Hinchinbrook Island. Although Campbell's (1982) early volume calculations assume homogenous wall height and a uniform substrate, it provides the only quantitative estimate of a fishtrap complex's holding capacity in the Australian literature.
The most significant challenges facing Australian fishtraps concern documentation, monitoring and management. The risk of structural degradation of fishtraps is an urgent practical implication of increasing coastal developmental pressures and climatic impacts (Memmott et al., 2008;Roberts et al., 2016;Rowland and Ulm, 2011). These threats cannot be appropriately managed without knowledge of current status of the intertidal stone-walled structures. To improve understandings of fishtraps, this study applies high-resolution Unmanned Aerial Vehicle (UAV) photogrammetry to document the stone-walled intertidal fishtraps of Sweers Island, southern Gulf of Carpentaria, Australia.

Case study area
The stone-walled intertidal fishtraps of Sweers Island are situated on the traditional lands of Kaiadilt people in the South Wellesley Islands, comprising an archipelago of 10 islands with Sweers Island the easternmost (c.13 km 2 ) ( Fig. 1). A local sea-level curve for the southern Gulf of Carpentaria demonstrates that rising post-glacial sea-levels separated the islands from the mainland c.8000 cal BP (Sloss et al., 2018). At 7700 cal BP sea-levels reached present mean sea level (PMSL), continuing to increase to +1.5 m-2 m above PMSL, with relatively stable sea-levels remaining until 4000 cal BP. Sea-levels rapidly regressed to 0.5 m ± PMSL between 4000 and 3500 cal BP (Sloss et al., 2018). The earliest documented occupation of the South Wellesley Islands occurs at 3483 cal BP on Bentinck Island and 3421 cal BP on Sweers Island, with a continuous occupation signal from around 2000 cal BP and strong evidence for permanent occupation in the last 1000 years (Memmott et al., 2016;Peck, 2016;Ulm et al., 2010). Archaeological and ethnographic evidence indicate that Bentinck Island was the focus of residence, with smaller surrounding islands, such as Sweers, visited for resource extraction and temporary occupation (Evans, 1995;Memmott et al., 2016;Tindale, 1962a;Ulm et al., 2010). The Kaiadilt population, believed to have reached a maximum of 123 individuals (Tindale, 1962b), were forcibly removed to a European mission on Mornington Island in 1948. Kaiadilt were the last coastal Aboriginal group to be institutionalised in Australia (Memmott, 1982).
The South Wellesley Islands generally experience a diurnal tidal range (one high and low tide each day) of approximately 3 m in amplitude, with an exception every fortnight where 'double' tides occur for 1-3 days, resulting in little water movement (Forbes and Church, 1983). Tidal fluctuations are most prominent during the wet season due to the strong northeast winds (Memmott, 1982). However, the southern part of the Gulf can experience varied tidal patterns when the combined effects of the shallow basin, strong winds, atmospheric pressure, and wet season run off coincide, and can halt and/or reverse typical clockwise currents, creating counter-clockwise currents and unpredictable tidal patterns (Forbes and Church, 1983). The extensive fishtrap structures (or ngurruwarra in Kayardild language) of the South Wellesley Islands have been described by several authors (Evans, 1995;Memmott et al., 2008Memmott et al., , 2016Robins, 1983;Saenger and Hopkins, 1975;Tindale, 1962aTindale, , 1962b. The first records of the Sweers Island fishtraps appear in Saenger and Hopkins (1975), who identified one fishtrap in Kabar Bay, one in Ngathald Bay and a third on the central eastern side of the island, later identified as the named Murarri trap (Robins, 1983). After an archaeological survey by Memmott and Robins in 1982, the number of reported fishtraps increased, later confirmed by low-level aerial photography as eight structures present on the island (Memmott et al., 2008), however the sites were not recorded in detail (Robins, 1983).

Photogrammetric UAV survey
In September and October 2015, fishtrap structures in Ngathald and Kabar Bays on the southwest Sweers coastline (Fig. 1) were subject to close-range photogrammetry utilising a Dynamic Spatial Solutions custom-built carbon frame octocopter. The UAV, equipped with 16" propellers, T-Motor U5 motors, a MikroKopter flight controller with GNSS, carried a 3 axes brushless gimble supporting a 24 megapixel DSLR Sony A6000 with a 20 mm prime lens (26.5 mm × 15.6 mm sensor). The vehicle was powered by 22Ah 6S or 16Ah 7S lithium polymer batteries, resulting in a total take-off weight of 8 kg. Five eight hectare (Ha) flight areas generated in ESRI ArcMap (10.2.1), utilising ESRI ArcGIS Online Gallery and Google Earth (v7.1.7) satellite imagery to determine survey boundaries, were exported to the mounted flight controller, providing a pre-programmed flight path, flown at 55 m elevation at a speed of 6 m/sec (Table 1) by Civil Aviation Safety Authority (CASA) certified pilots Michael Boland (Unmanned Aerial Vehicle Services) and Jason Scriffignano (Dynamic Spatial Solutions). Prior to recording, a permanent survey marker was established by placing a GNSS on an elevated beach ridge south of Kabar Bay (within 4 km of the survey area), to act as a base station, providing RTK geospatial corrections to a GNSS rover. To enable georeferencing of the photogrammetric model, ground controls points (GCPs), represented by numbered black and white linoleum tiles, were systematically placed at 50 m spacing in the intertidal zone and mapped with the RTK GNSS rover to retrieve high accuracy location and elevation measures (Fig. 2). Flights were conducted at 55 m elevation above the Australian Height Datum (AHD) at the lowest possible tide, daylight and weather permitted, within the timeframe of the field season, to capture the maximum extent of the stone-walled structures. Recording commenced at an outgoing tide, with tides ranging from a predicted low 1.17 m-2.62 m high above Lowest Astronomical Tide (LAT) ( Table 2).
While LiDAR has been suggested for data collection in coastal environments (Rowland et al., 2014), the dynamic conditions of the intertidal zone, with both terrestrial and submerged structures, are not suited as the infra-red pulse will be absorbed by water in inundated areas. Challenges of greater recording duration and resolution in bathymetric LiDAR, designed to record features at great depths, also proved unsuitable for high-resolution mapping of shallow submerged features (Doneus et al., 2013). Baltsavias (1999) additionally suggests that higher geometric resolution is achieved with aerial metric imagery captured from the same elevation as infra-red laser LiDAR.

Data processing
UAV data were prepared for processing in Folder2List, Microsoft Excel and ArcMap by aligning GNSS camera triggers with time-stamped photographs and re-projecting image coordinates. During recording, the flight controller sent triggers to the camera to capture images, while the internal GNSS simultaneously recorded geographic position, representative of where the camera was situated when the photograph was taken. The GNSS coordinates were recorded as the trigger was sent, causing a small time-lag between the recorded location and where the  Memmott's (1982) site map (which is reproduced in Robins, 1983), with orthography after Peck (2016). photograph was captured; potentially causing up to 2-3 m offset when travelling at a speed of 6 m/s, and a greater number of logged triggers than photographs. While the use of GCPs rectifies such inaccuracies during post-processing, camera triggers and photographs first require alignment to enable processing. To convert the flight controller's captured coordinates to a spatial reference suited for local-scale analysis, the aligned flight log was imported into ArcMap and re-projected into a local coordinate system, in this case Geocentric Datum of Australia 1994, Map Grid of Australia Zone 54. The projected photograph trigger coordinates were then exported as a text file and imported to Agisoft PhotoScan Professional (1.2.1) to create a 3D model of the mapped area.
To generate a point cloud, the re-projected coordinates and images from individual flights were imported into Agisoft PhotoScan Professional, software previously applied in archaeological 3D reconstructions (e.g. De Reu et al., 2013;Olson et al., 2013). A sparse point cloud was generated by aligning camera positions and applying a low-resolution mesh to create a continuous surface from which GCPs could be identified (Fig. 3). Image alignment, comprising 330-533 images per flight area, was made with high accuracy setting, generic pair selection, a key point limit of 40,000 and tie point of 1000. To georeference the model, GCPs were visually identified in individual photographs and assigned the RTK GNSS coordinates, providing a precise and accurate location and elevation value on the digital surface. The procedure was finalised by optimising the alignment, where camera calibrations were generated using the GCP locations, producing a 3D root mean square error (RMSE). The 3D RMSE refers to the overall error of the photogrammetric model and the actual location of individual GCPs.
To produce an orthographic image, and a DEM for further analysis of the fishtraps sites, the geo-referenced and error evaluated sparse point cloud was processed into a dense point cloud. Agisoft PhotoScan Professional generates dense point clouds through advanced image matching algorithms (see Verhoeven, 2011), computing coordinates of

Analytical procedure
The relationship between fishtrap placement and tidal movement was assessed in ArcMap. By importing the UAV-generated DEM to ArcMap, a raster representing slope percentage was generated (Spatial Analyst toolbox) to aid in identification of linear features by distinguishing low and steep gradients in the topography. Drawing on McCoy et al.'s (2011) geographical information system (GIS) methodology, four categories were established for fishtrap identification. By reclassifying slope into categories, excessive gradient was reduced, and a stronger visual representation of features was achieved. The output additionally assists in identifying the highest point of structures, as top stones are displayed as flat surfaces in contrast to the vertical walls (Fig. 4). In this study, fishtraps are defined by their ability to control movements of marine animals, and therefore ability to isolate a body of water. A contour layer was generated to identify boundaries of individual traps, based on the probability of pooling water within the structures. To establish appropriate precision in contour intervals a trial and error approach was adopted, which demonstrated the significance of fine intervals across the low gradient surface, resulting in a generation of 0.10 m contour intervals. To further assess the likelihood of pooling water, an output of flow accumulation was generated, indicating direction of water flow and likely areas of pooling.
Once the extent of fishtraps was established, individual structures were manually digitized based on the reclassified slope and contour outputs, with cross-reference to the orthographic image. Natural features incorporated in the trapping system and constructed walls were not separated in this analysis, as the purpose of this study was to assess tidal and trap correlation, rather than individual trap components. Drawing a polyline through the centre of a trap, length and elevation metrics were extracted through the 3D Analyst toolbox. As individual traps vary in height across their construction, mean elevation of the polyline was used to establish the top elevation of fishtrap walls. To calculate the height of fishtraps, the same procedure was executed by digitising a polyline at the base of the wall facing the shoreline (Fig. 5). Minimum elevation was extracted as topography is not homogenous and water is likely to pool at lowest elevations. By subtracting minimum elevation of substrate from mean elevation of the top wall, a mean fishtrap height was calculated (Fig. 6).
To assess the current and past ability of fishtraps to enclose water, numerous tidal scenarios were simulated to determine fishtrap working ranges. Working range is taken to be the time a trap is in effective use, represented by the elevation range where fish and other mobile marine animals can theoretically be contained within individual structures (i.e. working range is the elevation between no water and walls over-topped by water). The range is determined by wall height and topography, based on the mean elevation of top wall and minimum elevation of substrate. The possibility of marine animals being collected below working range, when traps are dry, is acknowledged. However, for the purpose of comparison, tidal range exceeding or not reaching the established working range is not considered to be in effect.
To align sea-levels and the photogrammetric model in a compatible format, tidal heights required recalibration to local terrestrial heights. Tidal range predictions, in the Australian context, are referenced to a height above LAT, the predicted lowest tide of the year (Department of Transport and Main Roads (DTMR) 2017), while terrestrial heights are described as a height above AHD, an Australian datum providing vertical control in surface mapping (Geoscience Australia, 2018). Zero AHD is referenced, but not necessarily equal, to PMSL across the nation. With local PMSL variations, major ports have established height datums to enable calibration of local sea-level fluctuation. For most Australian ports, calibrated Australian National Tide Tables (ANTTs, with electronic versions referred to as AHP11) can be obtained by the Bureau of Meteorology (BOM) and relevant State Governments agencies. However, presumably owing to the relatively small size and remote location of Sweers Island, the case study area is not considered a major, or standard, port and does not host a local datum. Tidal predictions for Sweers Island's port Inscription Point are therefore extrapolated from mainland standard port Karumba (135 km southeast of Sweers Island) which experiences a similar tidal regime (DTMR, 2017), demonstrated by a predicted 0.04m difference in predicted mean sea-level (MSL) for the two ports from 2015. The Karumba port datum has proved reliable with an observed MSL within 0.26 m of predicted levels in 2015 (estimated mean sea-level predicted to 2.11 m, and observed levels reaching 2.37 m) (BOM, 2018). While it is not possible to calibrate a precise tidal AHD offset without long-term documentation of local coastal processes, the Karumba port data are considered adequate for inundation modelling of Sweers Island.
Prior to modelling inundation events, ANTT tidal predictions were re-calculated to AHD heights to correlate with the UAV-captured dataset. First, the predicted PMSL of 2.06 m above LAT for Inscription Point was subtracted from the Karumba port datum, where 0 m AHD is represented by 2.184 m above LAT giving a value of 0.124. This value  was deducted from 0 m AHD to give a tidal height of −0.124 m AHD. The Raster Algebra tool (Spatial Analyst toolbox), or raster reclassification with the same principle, have previously been utilised to visualise scenarios of sea-level rise by displaying DEM values below or above a set threshold (e.g. Kuhn et al., 2011;Lichter and Felsenstein, 2012), and was adopted for this study. To model a PMSL of −0.124 m AHD for Sweers Island, the raster tool was set to display all pixels containing values equal to or below −0.124 m. The output is a raster file containing areas that do not satisfy the criteria (0) and areas that do (1). To display the generated tide with the orthographic image, the output was reclassified to disregard values of 0 and display 1, followed by a conversion to a vector file, to reduce file size and aid drawing in ArcMap.
Modelling of tidal inundation was initially undertaken using predicted tides of the day of UAV recording, to evaluate the procedure against known conditions, followed by enhanced PMSL scenarios, and mean annual magnitude of wet and dry season regimes. Mean seasonal high tide was based on the mean of predicted higher high water (MHHW) and mean lower high water (MLHW), and low tide drawn from the mean of mean higher low water (MHLW) and mean lower low water (MLLW) of 2015 (DTMR, 2013). Following this procedure, several characteristics of function were assessed: (1) fishtrap working range, (2) individual and simultaneous trap function, (3) seasonal functionality and (4) relative timeframes of fishtrap function in relation to previous sea-levels.

3D root mean square error (RMSE)
The accuracy of the created imagery and DEM is defined by the magnitude of 3D RMSE. The geometric error is based on the accuracy of the 84 RTK GNSS captured on-ground locations, post-processing positioning of GCPs in individual photographs, and photograph image quality. The resulting error estimate reflects the model's virtual position as opposed to physical location. This was also correlated by the use of interactively allocating GCPs as control or reference points during the model process, where reference GCPs displayed a comparable error in location to the 3D RMSE. Across the merged flight areas, the total 3D RMSE was less than 2.5 cm, representing a horizontal error of 0.9 cm along the x-axis, 1.21 cm across the y-axis, and 1.93 cm of vertical inaccuracy. The error was computed at a 68.27% confidence level (one standard deviation), which reflects geographical uncertainty of individual GCP. With a total 3D error below 2.5 cm, the accuracy of the model is equivalent to spatial precision provided by current RTK GNSS systems, where horizontal accuracy is estimated to 1 cm+2 ppm (part per million) and vertical differentiation up to 2 cm+2 ppm, when survey is situated within 10-40 km of associated base station (United States Geological Survey, 2017). The 3D RMSE demonstrates the Sweers Island fishtrap model to be close to identical to original RTK GNSS accuracy, despite its post-processing procedures. With a GNSS base station-rover distance within 4 km and a 2 cm × 2 cm pixel resolution, the maximum potential model error is equivalent to a single pixel. The 3D RMSE is therefore considered minor, and not predicted to impact spatial analysis results.

Fishtrap working range
Thirteen fishtraps were identified in the study area. The stonewalled structures are situated in the nearshore zone between 20 m and 120 m from the current high water mark, ranging from 38 m to 287 m in length, 0.15-0.97 m in height, and an average width of 0.50 m (Table 3, Fig. S1). The various working ranges of individual fishtraps, determined by topographic elevation and mean wall height, demonstrate the traps to enclose water effectively at various tides; therefore, individual traps are considered to be separate fishtrap sites. While the reasoning behind fishtrap construction cannot be confirmed, the flow accumulation strongly indicates shape and placement to reflect the underlying topography (Fig. 7). The lateritic substrate underlying most traps in the study area further provides a fixed foundation from which reliable elevation measures were calculated. The fishtraps of Ngathald and Kabar Bay are constructed between elevations of 0.35 m and −0.41 m AHD, with traps anchored to or near the modern shoreline. The majority of structures are constructed near AHD, where the physical expression of the arbitrary surface is PMSL. PMSL at Sweers Island is approximately −0.124 m AHD. By comparing PMSL and tidal scenarios to trap location, visual and numerical representations of fishtrap function can be calculated.

Functionality
Modelling of present sea-levels demonstrate fishtrap sites to be located within the range of local annual mean high and low tides, being of most effective use by enclosing maximum amount of water at mid-tide. Structures N4 and N5 are constructed on elevated topography above AHD, making the traps effective during the higher magnitudes of present mid-tidal conditions. As structures drop below working range with receding tides, the enclosed areas are left effectively dry, and N1, N2, N3, N7 and K4, K5, K6 located on lower elevations near or below AHD, come into effect (Fig. 8, Fig. S2). N7, with the greatest height of 0.98 m, is the only structure with sufficient height to extend its working range across mid and low magnitude tides. When assessing simultaneous inundation of fishtraps, it is evident that the greatest number of fishtraps are effective near PMSL. Trap use at mid-tide is further supported by ethnographic accounts of Kaiadilt paying daily visits to a fishtrap near camp at mid-tide (Tindale, 1962b). The documented fishtrap sites suggest two groups of stone-walled features operating in two intertidal zones, differentiated by approximately 0.5 m in elevation (Fig. S3), which may indicate a construction strategy that optimises the potential of enclosing marine resources during mid-tide across various tidal cycles.
With the variety of tidal patterns in the southern Gulf of Carpentaria, the island group is exposed to fluctuating tidal magnitude and frequency throughout the year (Forbes and Church, 1983). To assess seasonal functionality of the stone-walled structures, the observed mean high and low tide of two consecutive months representative of the wet (December-January) and dry (June-July) seasons were calculated. While tidal range was not predicted to significantly alter across seasons (0.02 m greater in the wet season), the shore inundation is less in the drier months. However, with the positioning and height of the documented fishtraps, working range is not impacted by seasonal tidal oscillations. Under average tidal conditions, the structures are most effective during high-to-mid-tide in the dry season, while mid-to-low tide provides optimal conditions during the wet (Fig. 9).

Past sea-level conditions
Geomorphological studies demonstrate peak sea-levels of +1.5 m-2 m between 7700 and 4000 cal BP before receding to current levels (Sloss et al., 2018). Past sea-levels were modelled at magnitudes up to +2 m greater than PMSL. An average tidal range relative to current regimes was assumed for modelling purposes and demonstrated that all structures would be submerged at mean sea-level. Only trap N4, anchored to the current high water mark, could have been of semi-effective use at the lowest tide (Fig. S4). With sea-levels receding to +1 m greater than PMSL, all fishtraps are situated within the tidal range, but only effective during low tides (Fig. S5). When comparing fishtrap function at current and enhanced sea-levels, the structural ability of controlling water movement is optimal, with the largest number of traps within working range at current, or close to current, sea-levels (Fig. 10). The individual and simultaneous function throughout PMSL tidal cycles indicates a late Holocene antiquity for the 13 fishtraps documented in this study area; aligning with the human occupation signal across the archipelago (Memmott et al., 2016).

Fishtrap operation and construction
The GIS-generated sea-level scenarios demonstrate that the stonewalled structures of Ngathald and Kabar Bays would not have had effective working ranges prior to c.3500 cal BP, which correlates with the earliest occupation dates of Sweers and Bentinck Islands dating to c.3500 cal BP (Memmott et al., 2016). With a continuous signal of occupation from c.2000 cal BP in the archipelago, and ethnographic records outlining the need for ongoing fishtrap maintenance for optimal function (e.g. Tindale, 1960), it can be concluded that the fishtraps were constructed within the last two millennia. With the current fishtrap placement and a consistent trap height below 1 m, the traps of Ngathald and Kabar Bays would have functioned with tides 0.5 m ± PMSL. With the relatively narrow fishtrap working range, sea-level shifts of ± 0.5 m would have a significant impact on trap function. As further refined high-resolution regional and local sea-level curves become available it will be possible to further refine construction chronologies. It is acknowledged that trap height may have decreased over time, however as such erosion cannot be demonstrated without longterm monitoring of fishtrap morphology, analysis is based on the properties of current structures. Based on the findings of fishtrap construction and interaction with past and current tidal regimes, we recommend high-resolution documentation of fishtrap metrics (length, height, width), particularly elevation in relation to a national height datum. While the method and majority of the analytical procedure can be applied to other archaeological site types, photogrammetric UAV recording proved particularly useful for documenting features in the intertidal zone. Previous fishtrap documentation has varied in detail, and cross-site comparison is limited due to ambiguous descriptions. GIS assessment of fishtrap physical features enabled quantitative analysis of geometric properties and formed a basis for robust comparative studies. By describing the structures through geometric variables, the risk of misidentification is minimised and the potential for site comparison and modelling maximized. Simple yet inclusive terms, such as fishtraps, representing structures (natural or constructed) enclosing bodies of water, described by metric properties is recommended to move towards a standardised objective industry practice. To further knowledge of the site type, geomorphological studies of environmental impacts on local ecosystems within the structures will assist in understanding local resource management, and potential erosion and sedimentation impacts on the wider intertidal zone. Such recent and historical information, of structural function and environmental impact, can contribute to development of site-specific cultural heritage management plans for stonewalled features in low energy intertidal environments. Evidence of local marine resource technologies, and their impacts, will further allow indicators of concepts such as domestication, cultural niche constructions, and potential aquaculture paradigms, to be evaluated and explored further.

Data acquisition and processing
Documenting large-scale remote archaeological structures through photogrammetric UAV survey proved to be a time effective recording procedure. The intertidal zone, with particular environmental constraints and wildlife hazards, make pedestrian recording difficult. Through UAV documentation, the entire area was captured in great detail without disturbing the site or archaeological features. The birdseye perspective proved vital to accurately assess the contextual landscape in which fishtraps are situated, including relationships between  features, which cannot be easily or accurately identified from a horizontal perspective. The method provided objective data collection, enabling post-collection analysis and interpretation not exclusively based on the initial judgement of the recorder. The highly accurate dataset further provides certainty in the results presented, and potential for future expansion of the analytical procedure presented in this research. Recording elevation in relation to AHD further aligns with Australian geographic survey standards (Intergovernmental Committee on Surveying and Mapping, 2014) and enables nation-wide site comparison if adopted in future site documentation. While the recording procedure required licensed pilots due to UAV weight class regulations under the CASA legislation (CASA, 2018), and data processing required some professional training, the method is relatively user-friendly, affordable, and accessible. Despite the compelling advantages of recording fishtraps utilising UAV and close-range photogrammetry, prevailing intertidal conditions and the high-resolution data acquisition presented some challenges.
While the approach provides a high accuracy output at a very highresolution of a large-scale siterequired for water pooling analysis of low-energy intertidal zonesweather, time, and hardware availability govern the application. Studies of intertidal site types will be subjected to weather and tidal restrictions, and despite the custom-made octocopter's technical capability of flying in winds up to 47 km/h, such conditions compromise flight time and pose a significant challenge for accessing intertidal sites, especially by boat. Tidal movement and daylight further restricted the time available for recording. The stonewalled structures were recorded at low tide, which set a limited timeframe, further constrained by low tides occurring in late hours of the day which reduced available daylight. Data acquisition during the lowest possible tide was necessary, as water causes the DEM to calculate elevation values of the water surface, or refracted through the surface, rather than directly off the underlying topography. Specular reflection and wave action can further result in null values, displayed as 'holes' in the raster surface. Very shallow areas of water were present within a few traps during recording of the Sweers Island sites. However, with such minor occlusion areas, the surface extrapolation during the DEM generation provide accurate surface values across site. While the demonstrated time restrictions are a result of the intertidal site type, rather than the recording method, distributing GCPs was the most timeconsuming aspect of the UAV methodology; a routine now eliminated by mounting a compact RTK GNSS system to the UAV. High-resolution low-level aerial imagery further presented some challenges in file size and processing time. Data processing was performed on a workstation, equipped with an 8 Core Intel CPU, dual nVidia 4 GB GTX980 graphic cards and 128 GB RAM, generating a 63.5 GB dataset of the recorded Sweers Island sites, computed over approximately 160 h. The analytical procedure was performed on a desktop of the same capabilities, where individual GIS operations required no more than 2 min to generate, with the exception of flow accumulation which computed over 1.5 h. However, analysing such a dataset with hardware of less graphic capacity will require a substantially longer timeframe.
Subjectivity in the manual digitisation of fishtraps was the final challenge of the post-processing procedure. To exclude subjectivity of the digitiser, object-based image analysis (OBIA) was considered, a well-established technique in remote sensing and geomorphological research (e.g. Blaschke, 2010;Conrad et al., 2015). While the technique initially showed promising results for identifying fishtraps in open source software System for Automated Geoscientific Analysis (v.2.2.7), the large file size hindered the system to perform image segmentation required to assess recognisable features across the assemblage. Instead, subjectivity was minimised by performing analysis in ArcMap by a single digitiser, applying heads-up digitisation (regularly using the zoom feature and multiple generated layers) to consider context as well as fine-scaled components affecting the capability of stone-walled structures to enclose water. Cross-referencing with the orthographic image proved crucial in digitisation of fishtraps, as the high-resolution image displayed discrepancies in low-gradient features that computed operations were unable to distinguish. For instance, walls near shore appear to be inundated by sediment, creating a flat appearance of the surface that could mislead length measurements; a natural process that may have altered through time, but unable to be detected by outputs such as contour and slope.
The analytical procedure and model validity presented in this study can further be improved by utilising a local port datum (when available), accurately reflecting the fluctuating tidal regime, particularly to provide information on out of the ordinary tidal events. Utilising the DEM output to its full potential can enable volume calculations (m 3 ) of fishtrap holding capacity across sites. Analysis of enclosed submerged surface area during various tidal heights will shine further light on fishtrap placement and test the importance of shape characteristics. Understanding fishtrap holding capacity will contribute to discussions of resource management. Finally, to reach standardised, quantitative analytical procedures of archaeological site types, OBIA and pattern recognition software should be further trialled.

Fishtrap terminology
The GIS analysis supported previous findings of challenges in inconsistent fishtrap terminology, particularly where focus is placed on shape features (Rowland and Ulm, 2011). The elevation range of the Ngathald and Kabar Bay fishtrap placement, in conjunction with the flow accumulation output, provide strong indication that structures are shaped to control the movement of marine resources with receding tides (Fig. S6, Fig. S7). Flow direction may lead structures to have a V or U appearance, however, results indicate function as a driver of construction, rather than stylistic choice. While this may not be a surprising find, it demonstrates the strengths of adopting metric-focused terminology, rather than morphological descriptions, and provides the basis for a less subjective interpretation of fishtrap construction and function. Accurate geometric classifications of fishtraps proved crucial in producing a replicable analytical procedure and perform a standardised analysis across structures.

Conclusion
The highly accurate dataset obtained through high-resolution UAV photogrammetric mapping of Ngathald and Kabar Bays, Sweers Island, identified 13 Kaiadilt fishtraps, where local topography was determined to be the dominant consideration in fishtrap placement and construction. The intertidal locations effectively enable operation of the fishtraps all year around, and elevations of the stone-walled constructions place them at optimal use during PMSL. During times of increased sealevels, the structures have capacity to be of effective use during low tide in magnitudes up to +1 m above PMSL, and combined with archaeological records, indicate a construction date in the last 2000 years. The UAV documentation technique and GIS analysis presented in this study provides the basis for further refinement of the chronology of construction when higher-resolution regional sea-level histories are available for the late Holocene. Considering fishtrap working range during past sea-levels is the most recent, viable option to date stone-walled fishtraps located in the present intertidal zone. The impact of future sealevel rise on fishtraps can also be modelled for mitigation and cultural heritage management applications. Aligning recording methods and terminology to a standardised quantitative technique will further benefit management plans for preservation of the site type and contribute to a broader cross-disciplinary understanding of stone-walled intertidal fishtrap. When modelling PMSLs at a 2 m greater magnitude, during a relative mean tidal range similar to current tidal conditions, a single fishtrap comes into effect at the end of the diurnal tidal cycle at elevations below 0.6 m AHD. (B) At PMSLs enhanced by 1 m, a maximum of 7 fishtraps are predicted to simultaneously hold water during low tide, and (C) assessment of the simultaneous function of stone-walled structures demonstrate individual fishtraps to enclose bodies of water most effectively at current, or close to, present sea-levels.