Object-Based Image Analysis of Ground-Penetrating Radar Data for Archaic Hearths

Object-based image analysis (OBIA) has been increasingly used to identify terrain features of archaeological sites, but only recently to extract subsurface archaeological features from geophysical data. In this study, we use a semi-automated OBIA to identify Archaic (8000–1000 BC) hearths from Ground-Penetrating Radar (GPR) data collected at David Crockett Birthplace State Park in eastern Tennessee in the southeastern United States. The data were preprocessed using GPR-SLICE, Surfer, and Archaeofusion software, and amplitude depth slices were selected that contained anomalies ranging from 0.80 to 1.20 m below surface (BS). Next, the data were segmented within ESRI ArcMap GIS software using a global threshold and, after vectorization, classified using four attributes: area, perimeter, length-to-width ratio, and Circularity Index. The user-defined parameters were based on an excavated Archaic circular hearth found at a depth greater than one meter, which consisted of fire-cracked rock and had a diameter greater than one meter. These observations were in agreement with previous excavations of hearths at the site. Features that had a high probability of being Archaic hearths were further delineated by human interpretation from radargrams and then ground-truthed by auger testing. The semi-automated OBIA successfully predicted 15 probable Archaic hearths at depths ranging from 0.85 to 1.20 m BS. Observable spatial clustering of hearths may indicate episodes of seasonal occupation by small mobile groups during the Archaic Period.


Introduction
Geophysical techniques are a noninvasive way to map and identify buried archaeological features. Their development has both paralleled and been enhanced by the evolution of geospatial and computer technologies, including GIS and Global Navigation Satellite Systems (GNSS) [1]. In 1999, Joseph Puyol-Gruart [2] wrote that 'Artificial intelligence is especially useful for experience-based knowledge', successfully predicting the future importance of digitizing multimedia information in the field of archaeology. Puyol-Gruart [2] further discussed extracting information from databases containing preprocessed data and computer models to identify patterns, while emphasizing the need for validation from a human expert. Semi-automated image analysis has become a multidisciplinary technique that began with aerial photography in the 1960s and satellite imagery in the 1970s. It has been employed in such fields as environmental science, microbiology, and medical imagery using pixel-based and object-based classification algorithms [3][4][5]. Pixel-based image analysis (PBIA) groups pixels based on spectral data values and was first implemented in the 1970s, predating object-based image analysis (OBIA) [3][4][5]. The OBIA approach utilizes both spectral and spatial data, using a two-step process to segment data based on the spectral values of pixels and then to classify objects based on user-defined spatial attributes [3]. This can be accomplished within geospatial software using computer-implemented algorithms [3]. Case studies comparing the two imagery analysis methods have shown significant

Archaic Period in the Southeastern United States
Indigenous populations have been in the southeastern US for at least 15,000 years, and archaeologists have divided this time into specific periods based on technological advances and cultural changes as follows: Paleoindian (before 8000 BC), Archaic (8000-1000 BC), Woodland (1000 BC to AD 1000), Mississippian (AD 1000-1570), and Protohistoric (AD 1570-1700) [45][46][47]. The Archaic Period in the southeastern US is characterized by cultural adaptations to both climate and landscape changes during the early and middle Holocene epoch [48]. It is further delineated into Early (8000-6000 BC), Middle (6000-3000 BC), and Late (3000-1000 BC) Archaic Periods [46,47]. Expanding deciduous forests created an ecosystem rich with nut-bearing trees, seed-bearing plants, and small game, while riverine and marine environments provided an abundance of fish and shellfish [46,47]. The archaeological record has shown that populations increased and that groups were highly mobile across the landscape with a focus on seasonal resource procurement [47]. Major technological changes included smaller projectile points, more expedient stone tools, steatite carved vessels, and domestic plant cultivation [46,47]. Certain cultural traits of the southeastern Archaic Period seem to be regionally localized such as earthen mounds, coastal shell middens, fiber-tempered pottery, and mortuary sites [48][49][50][51]. Seven additional Archaic sites have been recorded within a one-mile radius of the study area [52,53].

Ground-Penetrating Radar for Archaeology
GPR is an active geophysical technique that propagates electromagnetic waves into the subsurface and records reflections from buried interfaces and objects [54,55]. The GPR unit measures the reflection of returning waves from buried features and anomalies that can be metallic or nonmetallic [54]. Reflection magnitude is measured in decibels (db), while velocity is measured in

Archaic Period in the Southeastern United States
Indigenous populations have been in the southeastern US for at least 15,000 years, and archaeologists have divided this time into specific periods based on technological advances and cultural changes as follows: Paleoindian (before 8000 BC), Archaic (8000-1000 BC), Woodland (1000 BC to AD 1000), Mississippian (AD 1000-1570), and Protohistoric (AD 1570-1700) [45][46][47]. The Archaic Period in the southeastern US is characterized by cultural adaptations to both climate and landscape changes during the early and middle Holocene epoch [48]. It is further delineated into Early (8000-6000 BC), Middle (6000-3000 BC), and Late (3000-1000 BC) Archaic Periods [46,47]. Expanding deciduous forests created an ecosystem rich with nut-bearing trees, seed-bearing plants, and small game, while riverine and marine environments provided an abundance of fish and shellfish [46,47]. The archaeological record has shown that populations increased and that groups were highly mobile across the landscape with a focus on seasonal resource procurement [47]. Major technological changes included smaller projectile points, more expedient stone tools, steatite carved vessels, and domestic plant cultivation [46,47]. Certain cultural traits of the southeastern Archaic Period seem to be regionally localized such as earthen mounds, coastal shell middens, fiber-tempered pottery, and mortuary sites [48][49][50][51]. Seven additional Archaic sites have been recorded within a one-mile radius of the study area [52,53].

Ground-Penetrating Radar for Archaeology
GPR is an active geophysical technique that propagates electromagnetic waves into the subsurface and records reflections from buried interfaces and objects [54,55]. The GPR unit measures the reflection Remote Sens. 2020, 12, 2539 4 of 20 of returning waves from buried features and anomalies that can be metallic or non-metallic [54]. Reflection magnitude is measured in decibels (db), while velocity is measured in nanoseconds per meter (ns/m) [56]. GPR can also detect changes in soil and sediment properties including areas that have been disturbed by human occupation, making it highly suitable for archaeology [54,55]. Archaeologists utilize GPR to detect the remnants of architectural features, hearths, middens, graves, and even subtle features such as compaction that might not be detected during an excavation [54]. Depth range is potentially increased with lower antenna frequency, which can range from 12.5 to 2600 MHz [57,58]. Ground moisture and sediment type also affect signal penetration [54,55]. Dryer sediments and low conductivity sediment types such as sand allow the signal to penetrate up to 30 m below surface (BS), while wetter sediments and materials with a higher conductivity can decrease signal depth to less than one meter [57][58][59]. Most archaeological GPR studies use antennas ranging from 100 to 900 MHz, typically penetrating up to 5 m below the ground surface. The quality of results varies as ground moisture can change daily, while sediment characteristics are based on geological morphology that may drastically change across a small area [54].
GPR and other geophysical techniques have become essential and often primary methods for mapping, investigating, and understanding archaeological sites [60][61][62]). L.T. Dolphin and colleagues were the first to use GPR in the field of archeology in 1977 to explore underground caves in New Mexico [63]. In 1982, C. J. Vaughn identified 16th century features at a Basque whaling site in Canada using GPR [56]. Software advances in the 1990s and 2000s led to the development of processing tools that allow for horizontal depth slicing and 3D imaging of GPR data [64]. In the southeastern United States, research-driven geoarchaeological investigations employing GPR (along with other remote sensing techniques) have been focused mainly on historic (e.g., [65][66][67][68][69]), Mississippian and/or Protohistoric (e.g., [70][71][72][73][74][75][76]), and Woodland (e.g., [70,72,77,78]) sites. Geophysical investigations in the southeast have rarely been focused solely on Archaic sites ( [79] is a notable exception). This is partially because robust features such as structures, villages, and palisades are rarely present at Archaic (and Paleoindian) sites. Fortunately, many of these locations, including David Crockett Birthplace State Park, contain evidence of thousands of years of indigenous occupation (including Archaic) due to the suitability of the landscape for human habitation (e.g., [70,80]). One major advantage of using GPR at these sites is that the GPR signal can penetrate deep enough to reach older layers that may be out of reach for other geophysical methods. Even if these layers are not targeted for excavation, they are still recorded in the geophysical data. Another advantage of geophysical techniques in general is that known sites can be revisited and explored without breaking ground. When used in tandem, archaeological excavations and geophysical investigations enhance both the archaeological record and knowledge of geophysical practice, allowing investigators to not only pinpoint features, but answer research questions and develop new ones [60][61][62]77,[81][82][83].

Geophysical Survey
A geophysical survey of the park was completed in an area encompassing 2 hectares, which was being reconstructed into a historically accurate late 18th-century homestead. This was performed using a GSSI SIR-4000 unit with a 400 MHz antenna. The parameters were as follows: meters per mark = 1, ns time window = 50, sample/scan = 512, and scans/m = 100. The magnetometry survey was conducted with a Bartington Grad 601-2 Magnetic Gradiometer System containing two Grad-01-1000L sensors and a DL601 data logger. A grid network was created using a Real-Time Kinematic (RTK) GNSS, a Spectra Precision SP80 survey kit with a positional accuracy level of ± 0.05 m. Grids measured 30 × 30 m where possible and were surveyed north to south in a zig-zag pattern, starting in the southwest corner, and with survey lines spaced 0.5 m apart. Survey grid coordinates were recorded with a handheld data collector using SurveyPro software. A high-resolution digital orthomosaic created from Unmanned Remote Sens. 2020, 12, 2539 5 of 20 Aerial Vehicle (UAV) photos within Agisoft PhotoScan was used as an additional basemap. A workflow chart of the GPR data processing and OBIA steps is presented in Figure 2.

Ground-Penetrating Radar Data Processing
GPR-SLICE 7.0 was used to process and display the GPR data. Transects were displayed as radargrams showing profiles of depth levels starting at ground surface and ending at less than 2 m. Background removal and bandpass frequency filters were applied to all radargrams and the first break was used to set time zero. Slices and radargrams were examined at varying depth levels for reflections (see Figure 3) that had the potential to be archaeological features. A few of these (see Figure  4) were selected for ground truthing through both augering and excavation test units. Feature depth measurements made during excavation were used to calculate an average velocity of 0.78 m/ns for the archaeological layers, and this was used to export 22 horizontal amplitude depth slices with a thickness of 2.05 ns. Slices were gridded in Surfer software using the nearest neighbor method to avoid interpolation errors. Archaeofusion software was used to mosaic the surfer ASCII grid slices by matching adjacent grids based on regional statistical means. Slice mosaics were georeferenced in Archaeofusion using the geodetic coordinates of the local site datum and applying an azimuth rotation. Slice mosaics were further processed in Archaeofusion to remove spikes by using a standard deviation threshold moving window filter and a mean profile filter. Lastly, these mosaics were resampled to a pixel size of 0.125 m × 0.125 m and standardized with a mean of zero and a standard deviation of one.

Ground-Penetrating Radar Data Processing
GPR-SLICE 7.0 was used to process and display the GPR data. Transects were displayed as radargrams showing profiles of depth levels starting at ground surface and ending at less than 2 m. Background removal and bandpass frequency filters were applied to all radargrams and the first break was used to set time zero. Slices and radargrams were examined at varying depth levels for reflections (see Figure 3) that had the potential to be archaeological features. A few of these (see Figure 4) were selected for ground truthing through both augering and excavation test units. Feature depth measurements made during excavation were used to calculate an average velocity of 0.78 m/ns for the archaeological layers, and this was used to export 22 horizontal amplitude depth slices with a thickness of 2.05 ns. Slices were gridded in Surfer software using the nearest neighbor method to avoid interpolation errors. Archaeofusion software was used to mosaic the surfer ASCII grid slices by matching adjacent grids based on regional statistical means. Slice mosaics were georeferenced in Archaeofusion using the geodetic coordinates of the local site datum and applying an azimuth rotation. Slice mosaics were further processed in Archaeofusion to remove spikes by using a standard deviation threshold moving window filter and a mean profile filter. Lastly, these mosaics were resampled to a pixel size of 0.125 m × 0.125 m and standardized with a mean of zero and a standard deviation of one.

Archaeological Excavation
The deep anomaly discovered in Radargram 33 of Grid C13 was pinpointed for archaeological testing, and a 1 × 1 m unit was placed above the feature. At 1.02 m BS, a large circular hearth ( Figure 5) consisting of fire-cracked rock (FCR) was uncovered with a diameter greater than one meter. Chert and quartzite lithic debris were recovered through wet screening of sediment located within the hearth layer using a 1/16th inch mesh screen. Ash and wood charcoal were also recovered, and one wood charcoal sample from this level was analyzed by Direct AMS laboratories. The results of the Accelerator Mass Spectrometry (AMS) radiocarbon dating gives an uncalibrated mean (1-sigma) of > 8083 ± 40 BP (D-AMS 033192) with a 2-sigma calibrated age range of 7019-6832 cal BC. The calibration range was generated using OxCal 4.3 online software [84] with the IntCal 13 curve for the Northern Hemisphere [85]. Several cultural levels were examined within the geophysical data and identified during the archaeological investigations, including Woodland features and artifacts encountered at depth levels ranging from 0.36 to 0.55 m BS. The remnants of a partially burned Mississippian/Protohistoric structure were uncovered at 0.30 m BS at the base of the plowzone. Diagnostic artifacts from the house floor included broken vessel fragments and three European glass trade beads. AMS radiocarbon dates (Table 1) were obtained from organic material at all cultural levels.

Archaeological Excavation
The deep anomaly discovered in Radargram 33 of Grid C13 was pinpointed for archaeological testing, and a 1 × 1 m unit was placed above the feature. At 1.02 m BS, a large circular hearth ( Figure  5) consisting of fire-cracked rock (FCR) was uncovered with a diameter greater than one meter. Chert during the archaeological investigations, including Woodland features and artifacts encountered at depth levels ranging from 0.36 to 0.55 m BS. The remnants of a partially burned Mississippian/Protohistoric structure were uncovered at 0.30 m BS at the base of the plowzone. Diagnostic artifacts from the house floor included broken vessel fragments and three European glass trade beads. AMS radiocarbon dates (Table 1) were obtained from organic material at all cultural levels.

Object-Based Image Analysis
ESRI ArcMap software (version 10.6.1) was used to conduct the OBIA on the final processed GPR slice mosaics. Three slice mosaics (Slices 14, 15, and 16) were chosen with a depth range of approximately 0.20 m above and below the level of the excavated hearth. The raster datasets were imported in their original plane coordinates in order to retain exact pixel values that can be slightly altered due to automatic resampling associated with georeferencing. Each slice mosaic was

Object-Based Image Analysis
ESRI ArcMap software (version 10.6.1) was used to conduct the OBIA on the final processed GPR slice mosaics. Three slice mosaics (Slices 14, 15, and 16) were chosen with a depth range of approximately 0.20 m above and below the level of the excavated hearth. The raster datasets were imported in their original plane coordinates in order to retain exact pixel values that can be slightly altered due to automatic resampling associated with georeferencing. Each slice mosaic was reclassified into a binary image using a global threshold of two standard deviations. Next, a raster to vector conversion was performed to create polygons of the anomalies that had a reflectance value of 2σ or greater. Geometric attributes of the Archaic hearth excavated by the East Tennessee State University (ETSU) team, and hearths previously excavated and recorded by Benthall [40] were used to define probable hearth metrics and extract similar polygons. The OBIA used the following attributes: area, perimeter, L2W ratio, and circularity index. Perimeter and area values were automatically generated through the vector conversion tool. All documented Archaic hearths had diameters ranging from one to two meters, which corresponds to areas between 0.8 and 3.5 m 2 (see Equation (1)).
Area Selection Formula: (Area > = 0.8) AND (Area < = 3.5) The L2W ratio was created using the major and minor axis of the polygons with a minimum of 1.0 representing a perfect circle. Polygons were selected with an L2W ratio of less than 2.5 (Equation (2)) to allow room for potential hearths that may be more elliptical in shape, as was seen in Woodland layers during the ETSU excavation and Archaic hearths that were recorded by Benthall [40].
Length-to-Width Ratio: (L/W) < = 2.5 Several Circularity Index formulas have been developed to calculate a mathematical value indicating deviation from a perfect circle [16,18]. The circularity formula used in this study was obtained from a GEOBIA sinkhole study conducted by Doctor and Young [15]. In this formula, Po represents the observed or known perimeter of the shape, while Pe represents the expected perimeter value if the shape was a perfect circle (Equation (3) All polygons were evaluated by human interpretation to eliminate false positives and clutter from modern features such as utility lines, roads, gardens, and tree roots. Next, they were matched to their location in radargrams, and probable Archaic hearths were identified using expert judgment. Some features had ambiguous GPR reflections indicating possible prehistoric burials and modern utility lines, so these were excluded at this stage.

GIS Model
Incremental model results are presented in Table 2. A combined total of 8344 objects were created from Slices 14, 15, and 16 after segmentation by the global threshold. The OBIA steps radically reduced clutter from modern features by removing polygons outside the prescribed geometric ranges. The area metric reduced the number of polygons by 95%, eliminating many unwanted features such as small reflections created from the recently tilled gardens. The L2W ratio eliminated linear polygons, many of which were reflections from utility lines, ditch lines, and roads. The Circularity Index extracted features that were closer to the shape of a circle, eliminating irregular features such as tree roots. Overall, the automated OBIA steps reduced the total number of polygons by 98%. Further manual elimination of 31 objects was necessary to remove remaining clutter. The remaining 106 objects were matched to corresponding radargrams and expert judgment was employed to identify 18 polygons as probable Archaic hearths containing FCR. Two of the polygons in Slices 15 and 16 were identified as the hearth excavated by the ETSU team, while two more anomalies had overlapping polygons in Slices 14 and 15. Combining these reduced the number of probable hearths to 14. Expert judgment of radargrams also identified four additional reflection patterns as probable hearths, even though they had previously been eliminated. Two were too small, one too large, and the fourth was located at 1.40 m BS-deeper than allowed in the model. The final number of probable Archaic hearths was therefore 17, which is inclusive of the slightly under-and over-sized candidates but excluding the deep one because that GPR slice (Slice 18) was not used in the model. Figures 6-8 show examples of probable Archaic hearths at varying depth levels of Slices 14, 15, and 16 as they appear in (a) radargrams and (b) amplitude depth slices, along with (c) maps of objects extracted by global threshold segmentation, and (d) after OBIA metrics and manual elimination.

Auger Testing
The 17 probable Archaic hearths were tested with a four-inch (0.1-m) bucket auger. FCR was reached at the approximate predicted depths in 16 of the tests. Radargrams for probable hearths are shown in Figure 9, and corresponding auger test results in Table 3. Probable Hearth 8 did not contain FCR but was interpreted as a prehistoric pit based on the presence of ash, charcoal, and a lithic flake from 0.90 to 1.30 m BS. The pit feature had been previously excluded from the model based on area values greater than the metrics allowed, but it was added back during expert interpretation of radargrams. When obstructions were reached in three of the auger tests, they were moved 0.30 m grid north. Control tests were performed to depths of 1.40 m, at a range of 0.50 to 4 m outside of five of the test feature locations. There was no evidence of FCR in four of the control tests. FCR was reached at 0.99 m BS in one control test placed 0.5 m outside a probable hearth found at 0.92 m BS, which was likely a continuation of the same hearth. The automated OBIA model successfully predicted 15 out of the 17 probable hearths identified by geometry attributes. Probable hearths 4 and 12 had been added based on expert judgment and tested positive for FCR-both were slightly smaller than allowed by the area metric. Figure 10 uses Slice 14 to illustrate how the semi-automated OBIA approach selected thousands of objects and reduced them to 10 probable hearths for that depth interval. Figure 11 shows the location of all Archaic hearths confirmed by the presence of FCR, showing that they occur in a clustered arrangement across the survey area.

Auger Testing
The 17 probable Archaic hearths were tested with a four-inch (0.1-m) bucket auger. FCR was reached at the approximate predicted depths in 16 of the tests. Radargrams for probable hearths are shown in Figure 9, and corresponding auger test results in Table 3. Probable Hearth 8 did not contain FCR but was interpreted as a prehistoric pit based on the presence of ash, charcoal, and a lithic flake from 0.90 to 1.30 m BS. The pit feature had been previously excluded from the model based on area values greater than the metrics allowed, but it was added back during expert interpretation of radargrams. When obstructions were reached in three of the auger tests, they were moved 0.30 m grid north. Control tests were performed to depths of 1.40 m, at a range of 0.50 to 4 m outside of five of the test feature locations. There was no evidence of FCR in four of the control tests. FCR was reached at 0.99 m BS in one control test placed 0.5 m outside a probable hearth found at 0.92 m BS, which was likely a continuation of the same hearth. The automated OBIA model successfully predicted 15 out of the 17 probable hearths identified by geometry attributes. Probable hearths 4 and 12 had been added based on expert judgment and tested positive for FCR-both were slightly smaller than allowed by the area metric. Figure 10 uses Slice 14 to illustrate how the semi-automated OBIA approach selected thousands of objects and reduced them to 10 probable hearths for that depth interval. Figure 11 shows the location of all Archaic hearths confirmed by the presence of FCR, showing that they occur in a clustered arrangement across the survey area.  Table 3.

Discussion
The results of geophysical surveys vary due to site conditions and feature characteristics. In this case study, deep features were identified within the GPR data that were not found with magnetometry. This was more than likely due to soil conditions and interference from modern noise within the survey area. A micaceous sandy loam was found at all levels of auger testing that reached

Discussion
The results of geophysical surveys vary due to site conditions and feature characteristics. In this case study, deep features were identified within the GPR data that were not found with magnetometry. This was more than likely due to soil conditions and interference from modern noise within the survey area. A micaceous sandy loam was found at all levels of auger testing that reached a depth of 1.40 m. The low conductivity of sandy sediments likely allowed for greater GPR depth penetration, and this seems to have played a role in the ability of the GPR survey to receive strong reflections from features found at levels below one meter. Modern features affected both the magnetometry and GPR data, especially at levels closer to the ground surface. The survey area contained metallic debris, electrical utility boxes, and a large metal pylon supporting high-powered electrical lines, all of which masked magnetometry readings from more subtle archaeological features. Still, in limited regions with little to no modern magnetic interference, hearths identified by GPR were not detected by magnetometry. This suggests that the hearths do not have magnetic fields strong enough to be detected at these depths.
The OBIA model radically reduced clutter in the GPR data and successfully extracted probable archaeological features. Human interpretation and expert judgment were imperative to delineate all possible Archaic hearths from the GPR data, while excluding features such as utility lines and possible prehistoric burials. Manual examination of radargrams and auger testing also proved that the OBIA excluded some probable hearths; however, this could be corrected by expanding the metrics of the model. Further examination of the radargrams and amplitude depth slices also showed that one probable hearth (Probable Hearth 12 in Figure 9) not included by the OBIA had been separated into smaller polygons due to gaps in the reflectance of the anomalies, reinforcing the need for human interpretation. This problem could also be solved by adding a buffer or proximity analysis to the OBIA steps. All features that were selected by the OBIA and expert judgment tested positive for evidence of human occupation, with many containing charcoal and ash. The success of the OBIA was in part due to the size, shape, and depth of the Archaic features, along with knowledge of the attributes of previously excavated hearths. It is important to note that this type of analysis may not be suitable for all data. As stated by Verdonck et al. [3], 'Where the archaeological features belong to one class with a simple shape (e.g., circular structures), relatively simple algorithms can be used. ' The map of the auger-tested probable hearths ( Figure 11) shows noticeable patterns of clustering by depth and proximity, which may indicate seasonal occupation of small groups. Resources such as nearby springs were more than likely utilized, along with a local abundance of small and large game and freshwater species from the adjacent creek and river [40]. The quartzite and chert debris found during the excavation of the Archaic hearth suggests stone tool production and possible local raw material procurement, an idea previously proposed about this site by Benthall [40]. Additional AMS radiocarbon dating of charcoal samples may help determine a more robust occupation range. However, it is impossible to know the true nature of the occupation without a more extensive archaeological investigation. The FCR of the ETSU-excavated Archaic hearth was stacked at a height of greater than 0.10 m and was mounded in the center. Overall, the tested features were reached at depth levels ranging from 0.85 to 1.20 m BS, and exact occupation levels cannot be determined from auger testing alone. The survey area of the park is relatively flat due to decades of plowing and grading associated with modern construction. The slope and elevation of the prehistoric landscape and river terrace may have been dramatically different during the Archaic Period. Landscape changes and feature disturbance are particularly likely given the proximity to the meandering Nolichucky River and associated flooding.

Conclusions
Semi-automated OBIA can be utilized to quickly delineate homogenous subsurface archaeological features from preprocessed geophysical data. This method was successfully applied to extract Archaic hearths from a large GPR dataset using parameters of known Archaic hearths. A logical sequence was constructed employing computer-generated algorithms within GIS software to eliminate clutter. The result was a time-saving approach that reduced the number of anomalies by over 98% and eliminated some of the subjectivity and inconsistency associated with manual interpretation. Expert judgment, excavation, and auger tests were used to validate the model, resulting in the identification of 17 probable Archaic hearths, of which 15 were successfully predicted by OBIA. This method could be beneficial in processing large datasets where homogeneous features are expected to be found within geophysical data. OBIA could also be enhanced by future software development that would allow the technique to be implemented on anomalies found within GPR radargrams and 3D geophysical data. OBIA was implemented to target deep Archaic features that are sometimes overlooked, ignored, or not identified with geophysical techniques having more limited depth capabilities. The OBIA of the preprocessed data was completed for one mosaic dataset in under 20 min, and while additional time was needed for human interpretation, the selected anomalies were quickly pinpointed within GPR radargrams. Auger testing was a fast and suitable method, as FCR could be felt and heard (and sometimes retrieved) without the need for a full excavation, while still collecting sediment samples and cultural material. Possible occupation patterns were seen through GIS mapping of the probable Archaic hearth locations, adding to the archaeological record of the park and region. By combining GPR, GIS, OBIA, expert judgment, limited excavation, and auger testing, a cost-effective and labor-efficient method was developed that could be utilized to discover similar cultural components at local and regional prehistoric sites and adapted for other feature types and sites worldwide.