Multiscale anthropogenic feature detection in the Argentinian Andes: satellite machine learning, UAV, and pedestrian survey

It may state two separate claims as one.

At Casas Quemadas, previously undetected channels were identified that connect the main irrigation canal to the southern area of the site, where a check-dam system was documented in 2024.

“At the site of Casas Quemadas, previously undetected channels were identified that connect the main irrigation canal to the southern area of the site, where a check-dam system was documented in 2024 (Fig. 8).”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

The UAV orthophotograph had a resolution of 3.5 cm.

“The orthophotograph had a resolution of 3.5 cm (Fig. 6), the DEM a resolution of 8 cm (Fig. 7).”

It may state two separate claims as one.

The final cleaned dataset consists of 552 enclosure features and represents an ML-assisted map of enclosure features across the wider study area.

“The final cleaned dataset consists of 552 enclosure features (Fig. 5), and represents an ML-assisted map of enclosure features across the wider study area.”

It may state two separate claims as one.

In 2024, a new terrace system was detected in the Huayatayoc region and subsequently named El Sauce.

“In 2024, in the region where the Huayatayoc site is located (Fig. 9), a new terrace system was detected, subsequently named El Sauce.”

It may state two separate claims as one.

The ML run provided an initial set of predictions, with the final total of 552 features reflecting the manually validated and corrected dataset.

“The ML run provided an initial set of predictions, with the final total of 552 features reflecting the manually validated and corrected dataset.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

After processing, the UAV rasters covered an area of 8 square kilometres.

“After processing, the resulting rasters covered an area of 8 square kilometres.”

It may state two separate claims as one.

During 2023 and 2024, new agricultural structures were identified at Casas Quemadas and Huayatayoc, some of which were located within the UAV imagery area.

“During 2023 and 2024, new agricultural structures were identified at both sites, some of which were located within the UAV imagery area.”

It may state two separate claims as one.

Several ceremonial platforms, approximately 30 rock accumulations resulting from field clearing, and previously undocumented terrace systems were identified.

“Several ceremonial platforms, along with approximately 30 rock accumulations resulting from field clearing and previously undocumented terrace systems, were also identified.”

It may state two separate claims as one.

Imagery resolution varied: Bing and Maxar provided scenes at around 60 cm resolution, whereas Google imagery lacked sufficient detail for object detection tasks.

“Imagery resolution varied: Bing and Maxar provided scenes at around 60 cm resolution, whereas Google imagery lacked sufficient detail for object detection tasks (Fig. 4).”

It may state two separate claims as one.

The 0.90 threshold favours recall, whereas the 0.95 threshold reduces false positives but increases false negatives, particularly on the test dataset.

“The confusion matrices provide the relevant basis for threshold selection: the 0.90 threshold favours recall, whereas the 0.95 threshold reduces false positives but increases false negatives, particularly on the test dataset.”

It may state two separate claims as one.

At the selected confidence threshold (0.90), the model produced 572 candidate detections; 182 were removed as false positives during manual review, leaving 390 detections confirmed as true positives.

“At the selected confidence threshold (0.90), the model produced 572 candidate detections; 182 were removed as false positives during manual review, leaving 390 detections confirmed as true positives.”

• AssumptionRests on an assumption: The manual review and human interpretation correctly identify false positives and true positives in the machine learning detections.

It may state two separate claims as one.

Using the trained Mask R-CNN Object Detection Model, the automated detection phase with a confidence threshold of 0.90 yielded 572 detections, and with a confidence threshold of 0.95 it identified 471 potential enclosures across the study area.

“Using the trained Mask R-CNN Object Detection Model, the automated detection phase with a confidence threshold of 0.90 yielded 572 detections, and with a confidence threshold of 0.95 it identified 471 potential enclosures across the study area.”

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

The other Rockshelter Chullpa is part of the El Sauce site, which features a wide variety of agricultural structures.

“In the latter two cases, one was reused as a Herding Post with associated enclosures, while the other is part of the El Sauce site, which features a wide variety of agricultural structures.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

The confusion matrices summarise the trade-off between detection sensitivity and false positive rates.

“The resulting confusion matrices summarise the trade-off between detection sensitivity and false positive rates (Table 1).”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

The trained model was run across the 1750 square kilometres area of interest over the course of a day.

“The trained model was run across the 1750 square kilometres area of interest over the course of a day.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

The UAV DEM had a resolution of 8 cm.

“The orthophotograph had a resolution of 3.5 cm (Fig. 6), the DEM a resolution of 8 cm (Fig. 7).”

It may state two separate claims as one.

UAV imagery provided a detailed and up-to-date visual reference, with substantially higher spatial resolution than satellite imagery, and an aerial perspective that helped bridge ground-level observations and broader landscape patterns.

“These data provided a detailed and up-to-date visual reference, offering substantially higher spatial resolution than satellite imagery and an aerial perspective that helped bridge ground-level observations and broader landscape patterns.”

It may state two separate claims as one.

In this study, walled features were more readily identifiable in satellite imagery and therefore formed the basis of the training dataset, whereas deteriorated terraces and retaining walls were often too subtle to detect.

“In this study, walled features were more readily identifiable in satellite imagery and therefore formed the basis of the training dataset, whereas deteriorated terraces and retaining walls were often too subtle to detect.”

It may state two separate claims as one.

In this study, the UAV orthophotographs enhanced the dataset by revealing subtle features such as degraded channels, while the pedestrian survey added further nuance by identifying retaining walls and terraces, providing a complementary perspective.

“In this study, the UAV orthophotographs enhanced the dataset by revealing subtle features such as degraded channels, while the pedestrian survey added further nuance by identifying retaining walls and terraces, providing a complementary perspective.”

Probably not a substantial claim. The paper may give no reason why it holds.

Architectural features constitute one of the most reliable proxies for pastoral infrastructure in the region, where relative chronologies are often blurred by long-term reuse.

“Furthermore, architectural features constitute one of the most reliable proxies for pastoral infrastructure in the region, where relative chronologies are often blurred by long-term reuse.”

• AssumptionRests on an assumption: Architectural features constitute one of the most reliable proxies for pastoral infrastructure in the region.

Probably not a substantial claim. It may be too vague to check against evidence.

These results highlight the value of combining automated detection and UAV interpretation with pedestrian survey to build a more comprehensive understanding of the cultural landscape.

“These results highlight the value of combining automated detection and UAV interpretation with pedestrian survey to build a more comprehensive understanding of the cultural landscape.”

It may state two separate claims as one.

Although satellite data allows study of unprecedentedly large areas, its available resolution is often not sufficient for in-depth assessment and identification of individual features.

“Although the application of satellite data now allows the study of unprecedentedly large areas, the available resolution of the data is often not sufficient for in-depth assessment and identification of individual features20,21.”

• cites [20] Schuetter, J. et al. Autodetection of ancient Arabian tombs in high-resolution satellite imagery. Int. J. Remote Sens. 34, 6611–6635 (2013). sourceNot checked — no full text of the source
• cites [21] Rayne, L. et al. Detecting Change at Archaeological Sites in North Africa Using Open-Source Satellite Imagery. Remote Sens. 12, 3694 (2020). sourceCheck pending

Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.

Most of the 19 newly identified features have been in use for many generations.

“Most of these have been in use for many generations.”

It may state two separate claims as one.

Pedestrian survey documented and validated features within accessible areas, UAV imagery provided high-resolution visual information in selected zones, and ML detection extended enclosure identification across the wider landscape.

“Pedestrian survey documented and validated features within accessible areas, UAV imagery provided high-resolution visual information in selected zones, and ML detection extended enclosure identification across the wider landscape.”

Probably not a substantial claim.

In this context, the ML-assisted detections function as a practical tool for highlighting potential anthropogenic features, such as enclosures, that can guide targeted field verification.

“In this context, the ML-assisted detections function as a practical tool for highlighting potential anthropogenic features, such as enclosures, that can guide targeted field verification.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

Previous work in the Cusi Cusi region has identified more than 220 archaeological sites with diverse chronologies.

“Previous work carried out in the Cusi Cusi region has identified more than 220 archaeological sites with diverse chronologies10.”

• cites [10] Vaquer, J. M., Pey, L. & Gerola, I. The multitemporality of landscapes: an example from the Andean area (Cusi Cusi, Puna of Jujuy, Argentina) [La multitemporalidad de los paisajes: un ejemplo del área andina (Cusi Cusi, Puna de Jujuy, Argentina)] (in Spanish). Relaciones 48, e070 (2023). sourceCheck pending

Probably not a substantial claim. The paper’s conclusions may not depend on it.

The low precision and F1 scores reported here should be interpreted in relation to the rarity of enclosure features and the visually variable background class.

“The low precision and F1 scores reported here should therefore be interpreted in relation to the rarity of enclosure features and the visually variable background class.”

Probably not a substantial claim.

The machine learning results were subsequently reviewed and corrected through human interpretation.

“subsequently reviewed and corrected through human interpretation.”

It may state two separate claims as one.

Numerous sites showed evidence of reuse and modification over time, reflecting long-term continuity in land use practices.

“Numerous sites showed evidence of reuse and modification over time, reflecting long-term continuity in land use practices.”

It may state two separate claims as one.

Limited vegetation cover in the study area generally affords high visibility of the surface archaeological record, facilitating both pedestrian survey and remote sensing.

“This limited vegetation cover generally affords high visibility of the surface archaeological record, facilitating both pedestrian survey and remote sensing.”

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

The survey boundary represents a methodological and operational constraint defined by temporal and logistical resources, and potential spatial bias is recognised as a limitation of the study.

“Rather, this boundary represents a methodological and operational constraint defined by the temporal and logistical resources available for fieldwork, and potential spatial bias is therefore recognised as a limitation of the study.”

Probably not a substantial claim.

The interpretative model presented reveals the social processes and land use based on the sites studied.

“The interpretative model we present reveals the social processes and land use based on the sites we have studied.”

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

The data augmentations simulate differences in acquisition geometry, lighting conditions, and spatial scale commonly encountered in high-resolution satellite imagery and help reduce overfitting.

“Together, these augmentations simulate differences in acquisition geometry, lighting conditions, and spatial scale commonly encountered in high-resolution satellite imagery and help reduce overfitting.”

It may state two separate claims as one.

The pedestrian survey findings confirm the presence of a dispersed but persistent occupation across the survey area, with a strong emphasis on pastoral and agropastoral strategies.

“These findings confirm the presence of a dispersed but persistent occupation across the survey area, with a strong emphasis on pastoral and agropastoral strategies (Fig. 12).”

It may state two separate claims as one.

All reported confusion matrices and performance metrics are derived from the raw model detections; no False Positives were manually removed and no True Positives were manually added prior to quantitative evaluation.

“All reported confusion matrices and performance metrics are derived from the raw model detections; no False Positives were manually removed and no True Positives were manually added prior to quantitative evaluation.”

Probably not a substantial claim.

To ensure unbiased evaluation, the dataset was partitioned into 60% training, 20% validation, and 20% test data.

“To ensure an unbiased evaluation, the dataset was partitioned into 60% training, 20% validation, and 20% test data.”

It may state two separate claims as one.

The detections were initially suggested by the model but subsequently reviewed and corrected through human interpretation.

“The detections were initially suggested by the model but subsequently reviewed and corrected through human interpretation.”

• AssumptionRests on an assumption: The manual review and human interpretation correctly identify false positives and true positives in the machine learning detections.

It may state two separate claims as one.

These data were geographically non-overlapping, and no image tiles or annotations from validation or test areas were used during model training.

“These data were geographically non-overlapping, and no image tiles or annotations from the validation or test areas were used during model training.”

It may state two separate claims as one.

Image chips were sampled across the full spatial extent of the study area and include a wide range of background conditions, including rocky outcrops, riverbeds, erosion features, and anthropogenic landscape elements.

“Image chips were sampled across the full spatial extent of the study area and include a wide range of background conditions, including rocky outcrops, riverbeds, erosion features, and anthropogenic landscape elements.”

It may state two separate claims as one.

The area is part of the Puna phytogeographic province, where xerophytic vegetation is well adapted to lack of water, low temperatures, and grazing pressure.

“The area is part of the Puna phytogeographic province, where xerophytic vegetation is well adapted to the lack of water, low temperatures, and grazing pressure25.”

• cites [25] Oyarzabal, M. et al. Vegetation units of Argentina [Unidades de vegetación de la Argentina]. Ecol. Austral 28, 40–63 (2018). in Spanish. sourceCheck pending

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

Mask R-CNN has demonstrated effectiveness in similar archaeological contexts, including detection of burial clusters in the Mongolian steppe, burials and charcoal hearth sites in Germany, and terraces in Samoa.

“Mask R-CNN has demonstrated its effectiveness in similar archaeological contexts34, including the detection of burial clusters across the Mongolian steppe35, burials36 and charcoal hearth sites37 in Germany, as well as terraces in Samoa38.”

• cites [34] Bellat, M. et al. Machine learning applications in archaeological practices: a review, (2025). sourceCheck pending
• cites [35] Fisher, M. T. et al. Multidisciplinary digital methodologies for documentation and preservation of immovable archaeological heritage in the Khovd River Valley, Western Mongolia. F1000Research. 11, 1250 (2022). sourceCheck pending
• cites [36] Meyer-Heß, M. F., Pfeffer, I. & Juergens, C. Application of convolutional neural networks on digital terrain models for analyzing spatial relations in archaeology. Remote Sens 14, 2535 (2022). sourceCheck pending
• cites [37] Bonhage, A. et al. A modified Mask region-based convolutional neural network approach for the automated detection of archaeological sites on high-resolution light detection and ranging-derived digital elevation models in the North German Lowland. Archaeol. Prospect. 28, 395–410 (2021). sourceNot checked — no full text of the source
• cites [38] Quintus, S., Davis, D. S. & Cochrane, E. E. Evaluating Mask R-CNN models to extract terracing across Oceanic high islands: an example from Sāmoa. Archaeol. Prospect. 30, 477–492 (2023). sourceNot checked — no full text of the source

Probably not a substantial claim. The paper may give no reason why it holds.

Enclosure features leave distinct and recognisable anthropogenic traces in the landscape, visible at the available resolution.

“The ML component of this study focused specifically on identifying enclosure features, since these leave distinct and recognisable anthropogenic traces in the landscape, visible at the available resolution.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

Due to field conditions and time constraints, it was often not possible to wait for optimal lighting, such as the uniform illumination provided by overcast skies during daylight hours.

“Due to field conditions and time constraints, it was often not possible to wait for optimal lighting, such as the uniform illumination provided by overcast skies during daylight hours.”

It may state two separate claims as one.

Architectural elements were prioritised because they are clearly identifiable in satellite imagery and exhibit high visibility in the field, allowing consistent detection and spatial validation across remote-sensing and ground-based datasets.

“Architectural elements were prioritised because they are clearly identifiable in the satellite imagery employed in this study and exhibit high visibility in the field, allowing for consistent detection and spatial validation across both remote-sensing and ground-based datasets.”

Probably not a substantial claim. It may be too vague to check against evidence. The paper may give no reason why it holds. It may state two separate claims as one.

Developing and expanding new remote sensing-assisted survey approaches will allow documentation of heritage, which can help counter extractive pressures and support local communities' demands for material and symbolic resources.

“Developing and expanding new remote sensing-assisted survey approaches will allow us to document and highlight this heritage, which in turn can help counter extractive pressures and support local communities’ demands for material and symbolic resources.”

• AssumptionRests on an assumption: Large-scale lithium mining poses a significant threat to the environmental stability of the region and the preservation of its cultural heritage.

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

These landscape types are implicitly represented in the training data as background rather than as explicitly labelled negative classes, which is common practice in object-detection workflows.

“These landscape types are implicitly represented in the training data as background rather than as explicitly labelled negative classes, which is common practice in object-detection workflows.”

Probably not a substantial claim. It may state two separate claims as one.

The 0.90 threshold was adopted as the main operating point because the primary aim is to indicate the broader spatial extent of anthropogenic activity and to support pedestrian surveys.

“Because the primary aim of this study is to indicate the broader spatial extent of anthropogenic activity across the landscape and to support pedestrian surveys, we adopted the 0.90 threshold as the main operating point.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

In this study, the type of features identified by the ML workflow reflects the training dataset, as expected in supervised learning.

“In the case of this study, the type of features identified by the ML workflow reflects the training dataset, as expected in supervised learning.”

• AssumptionRests on an assumption: As with all supervised ML approaches, the detected feature types correspond to those represented in the training dataset.

It may state two separate claims as one.

The UAV photogrammetry process enabled validation of the automated detection results and offered a more detailed and up-to-date view of the current situation on the ground, complementing older satellite imagery and allowing assessment of recent changes.

“This process enabled validation of the automated detection results and offered a more detailed and up-to-date view of the current situation on the ground, complementing the older satellite imagery and allowing assessment of any recent changes.”

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

Vegetation cover is low during the dry season, which is advantageous for mapping purposes.

“Conveniently, vegetation cover is also at a low during the dry season, which is advantageous for mapping purposes.”

It may state two separate claims as one.

By overlaying the UAV orthophotograph and satellite imagery, it was possible to assess the potential structures identified through the ML process, and to gain a high-resolution, current view of the layout of the sites.

“By overlaying the UAV orthophotograph and satellite imagery, it was possible to assess the potential structures identified through the ML process, and to gain a high-resolution, current view of the layout of the sites.”

It may state two separate claims as one.

While not every feature can be ground-truthed, the ML-assisted results improve survey efficiency and help identify promising locations for future archaeological investigation that may clarify the chronology and long-term use of these sites.

“While not every feature can be ground-truthed, the ML-assisted results improve survey efficiency and help identify promising locations for future archaeological investigation that may clarify the chronology and long-term use of these sites.”

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

Rainfall in the region is highly seasonal, with at least 80% falling between December and March, ranging from 200 to 400 mm per year.

“Rainfall is highly seasonal, with at least 80% of rain falling between December and March, ranging from 200 to 400 mm per year23,24.”

• cites [23] Saravia, T. S. Geography of the Province of Jujuy [Geografía de la Provincia de Jujuy]. (in Spanish). Instituto Geográfico Militar, (1960).Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [24] Buitrago, L. The climate of the province of Jujuy [El clima de la provincia de Jujuy] (in Spanish). Facultad de Ciencias Agrarias, Univ. Nac. de Jujuy (1999).Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

It may state two separate claims as one.

Around the 13th century AD, populations from the Doncellas-Casabindo region settled in Cusi Cusi and created an agricultural landscape on a scale not seen before in the region.

“Around the 13th century AD, populations from the Doncellas-Casabindo region settled in the area, creating an agricultural landscape on a scale not seen before in this region10.”

• cites [10] Vaquer, J. M., Pey, L. & Gerola, I. The multitemporality of landscapes: an example from the Andean area (Cusi Cusi, Puna of Jujuy, Argentina) [La multitemporalidad de los paisajes: un ejemplo del área andina (Cusi Cusi, Puna de Jujuy, Argentina)] (in Spanish). Relaciones 48, e070 (2023). sourceCheck pending

Probably not a substantial claim.

The training data were based on previously identified and validated enclosures within the study area.

“These training data were based on previously identified and validated enclosures within the study area.”

It may state two separate claims as one.

The ML methodology showed considerable potential, as traces of human influence were visible throughout the study area, including in highly remote locations.

“In this regard, the ML methodology showed considerable potential, as traces of human influence were visible throughout the study area, including in highly remote locations.”

Probably not a substantial claim.

Prioritising architectural elements was an operational compromise to maximise analytical consistency and comparability within the temporal and logistical constraints of the project.

“We acknowledge that this approach does not fully capture finer-scale evidence such as low-density artefact scatters or small domestic enclosures; however, prioritising such architectural elements was considered an operational compromise to maximise analytical consistency and comparability within the temporal and logistical constraints of the project.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

The spatial delimitation of the surveys does not necessarily reflect the broader geographical extent of pastoral mobility systems, which may operate at much larger scales.

“We acknowledge that this spatial delimitation does not necessarily reflect the broader geographical extent of pastoral mobility systems, which may operate at much larger scales.”

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

The high-altitude environment resulted in lower air pressure, which reduced lift capacity and made it more difficult for UAVs to maintain stable flight.

“In addition, the high-altitude environment resulted in lower air pressure, which reduced lift capacity and made it more difficult for UAVs to maintain stable flight.”

It may state two separate claims as one.

The additional pedestrian survey carried out to validate the ML procedure confirmed several results of the ML and UAV work, and added a range of additional features to the ML and UAV-based dataset, including retaining slope walls and check-dams.

“The additional pedestrian survey carried out to validate the results of the ML procedure confirmed several results of the ML and UAV work, and added a range of additional features to the ML and UAV- based dataset, including retaining slope walls and check-dams.”

It may state two separate claims as one.

The test data were fully held out and only used for final performance assessment after model training and threshold selection, ensuring independence between training and evaluation data.

“The test data were fully held out and only used for final performance assessment after model training and threshold selection, ensuring independence between training and evaluation data.”

Probably not a substantial claim. It may be too vague to check against evidence.

Favouring recall over precision is preferable in this context because missed detections reduce the value of the results for survey planning and landscape interpretation.

“In this context, favouring recall over precision is preferable, as missed detections reduce the value of the results for survey planning and landscape interpretation.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

As a result, it was not possible to rely solely on the onboard sensors to determine when the UAVs should return to the take-off point for battery replacement after depletion.

“As a result, it was not possible to rely solely on the onboard sensors to determine when the UAVs should return to the take-off point for battery replacement after depletion.”

Probably not a substantial claim.

The resulting potential increase in false positives is acceptable because detections are treated as candidate locations for prioritised field verification rather than as definitive feature identifications.

“The resulting potential increase in false positives is acceptable, as these detections are treated as candidate locations for prioritised field verification rather than as definitive feature identifications.”

Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds. It may state two separate claims as one.

UAVs generally have limited flight time and require targeted deployment, which assumes some prior knowledge of the location of features or areas of interest.

“Unfortunately, UAVs generally have a rather limited flight time and require a targeted deployment, which assumes some prior knowledge of the location of features or areas of interest.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

The experience of the operators was crucial in determining safe flight durations and distances.

“Instead, the experience of the operators was crucial in determining safe flight durations and distances.”

It may state two separate claims as one.

High-resolution historical aerial images collected by airplane can supplement satellite data, provide wider views of a research area, and contribute to deeper analysis of landscape change.

“High-resolution historical aerial images collected by airplane can be used to supplement this data while providing wider views of a research area and contributing to deeper analysis of landscape change22.”

• cites [22] Peters, M. J. H. & Stek, T. D. Photogrammetry and GIS to investigate modern landscape change in an early Roman colonial territory in Molise (Italy). Meas. Cult. Herit. Acta IMEKO 11, 1–8 (2022).Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

It may state two separate claims as one.

Pedestrian survey combined with community interaction provides the highest-resolution data but significantly limits the spatial range of investigation, particularly in challenging terrain at high altitudes.

“Pedestrian survey, combined with community interaction, arguably provides the highest-resolution data, but it significantly limits the spatial range of investigation, particularly in challenging terrain at high altitudes10.”

• cites [10] Vaquer, J. M., Pey, L. & Gerola, I. The multitemporality of landscapes: an example from the Andean area (Cusi Cusi, Puna of Jujuy, Argentina) [La multitemporalidad de los paisajes: un ejemplo del área andina (Cusi Cusi, Puna de Jujuy, Argentina)] (in Spanish). Relaciones 48, e070 (2023). sourceCheck pending

Probably not a substantial claim. The paper’s conclusions may not depend on it.

The SfM process relies on consistent lighting across overlapping images for effective image matching.

“This limitation complicated the SfM process, which relies on consistent lighting across overlapping images for effective image matching.”

Probably not a substantial claim.

Based on extensive field experience in the study area, the authors were able to identify how target enclosure features manifest in satellite imagery.

“Based on extensive field experience in the study area, we were able to identify how the target enclosure features manifest in satellite imagery.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

The ResNet backbone is a convolutional neural network architecture based on stacked residual blocks that has become a standard in Deep Learning for image analysis.

“The ResNet (Residual Network) backbone is a convolutional neural network architecture based on stacked residual blocks that has become a standard in Deep Learning for image analysis39.”

• cites [39] He, K., Zhang, X., Ren, S. & Sun, J. Deep residual learning for image recognition. In Proc. IEEE Conf. Comput. Vis. Pattern Recognit. 770–778 (2016). sourceNot checked — no full text of the source

It may state two separate claims as one.

This agricultural-pastoral system continued until the Spanish invasion in the 16th century AD, when agricultural sites were abandoned and populations returned to pastoralism as their main means of subsistence.

“This continued until the 16th century AD with the Spanish invasion, when agricultural sites were abandoned, and populations returned to pastoralism as their main means of subsistence11–13.”

• cites [11] Gerola, I. An internodal approach to the Cusi Cusi microregion [Paisajes dinámicos de la Cuenca Superior del Río Grande de San Juan (Jujuy, Argentina). Una aproximación internodal a la microrregión de Cusi Cusi] (in Spanish). Rev. Mundo de. Antes 16, 269–303 (2022). sourceCheck pending
• cites [12] Pey, L., Sclafani, M. & Braun, E. Textile chakra: a genealogy of productive practices and “ways of doing” in Cusi Cusi (Puna of Jujuy, Argentina) [Chakra textil: una genealogía de las prácticas productivas y “maneras de hacer” en Cusi Cusi (Puna de Jujuy, Argentina)] (in Spanish). Relaciones 48, e069 (2023). sourceCheck pending
• cites [13] Vaquer, J. M. Logics of landscape and territory in Cusi Cusi (Jujuy, Argentina): establishing new dialogues between the past, present, and future from archaeology. Estud. Atacameños. 67, 1–22 (2021). sourceCheck pending

Probably not a substantial claim. It may be too vague to check against evidence. It may state two separate claims as one.

The survey results constitute a crucial reference dataset for validating remotely sensed features and for interpreting broader landscape use patterns over time.

“The survey results constitute a crucial reference dataset for validating remotely sensed features and for interpreting broader landscape use patterns over time.”

Probably not a substantial claim.

Recent work on machine learning in archaeological remote sensing has shown that limited labelled training data, class imbalance and the accuracy paradox are common problems when rare archaeological targets are detected within large and complex datasets.

“Recent work on machine learning in archaeological remote sensing has shown that limited labelled training data, class imbalance and the accuracy paradox are common problems when rare archaeological targets are detected within large and complex datasets40–42.”

• cites [40] Verschoof-van der Vaart, W. B., Lambers, K., Kowalczyk, W. & Bourgeois, Q. P. J. Combining Deep Learning and Location-Based Ranking for Large-Scale Archaeological Prospection of LiDAR Data from The Netherlands. ISPRS Int. J. Geo-Inf. 9, 293 (2020). sourceNot checked — no full text of the source
• cites [41] Argyrou, A., Agapiou, A., Papakonstantinou, A. & Alexakis, D. D. Comparison of Machine Learning Pixel-Based Classifiers for Detecting Archaeological Ceramics. Drones 7, 578 (2023). sourceCheck pending
• cites [42] Argyrou, A. et al. Learning from the rare: overcoming class imbalance in archaeological object detection with boosting methods. Heritage 9, 99 (2026). sourceCheck pending

Probably not a substantial claim.

The final combined result constitutes a mapping product that serves as a baseline for further spatial and contextual analysis.

“The final combined result constitutes a mapping product that serves as a baseline for further spatial and contextual analysis.”

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

The Puna of Jujuy is situated in the 'Lithium Triangle', which holds 58% of the lithium in the world.

“The Puna of Jujuy is situated in the ‘Lithium Triangle’; the wider region within Argentina, Bolivia, and Chile that holds 58% of the lithium in the world15.”

• cites [15] U.S. Geological Survey. Mineral commodity summaries 2021. U.S. Department of the Interior (2021).Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim. The paper’s conclusions may not depend on it.

Imagery availability in the high Andes is strongly biased toward the dry season because wet season cloud cover reduces the number of cloud-free scenes suitable for mapping.

“Imagery availability in the high Andes is strongly biased toward the dry season, because the wet season is associated with increased cloud cover, which reduces the number of cloud-free scenes suitable for mapping.”

Probably not a substantial claim. It may be too vague to check against evidence. It may state two separate claims as one.

Although this knowledge often exists within local communities, the study offers structured and tangible data to inform heritage management and policy.

“Although this knowledge often exists within local communities, we now offer structured and tangible data to inform heritage management and policy.”

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

The rugged terrain caused unpredictable wind conditions, which could easily disrupt the flight paths of the relatively lightweight UAVs used in this study, occasionally making control difficult.

“The rugged terrain also caused unpredictable wind conditions, which could easily disrupt the flight paths of the relatively lightweight UAVs used in this study, occasionally making control difficult.”

It may state two separate claims as one.

Flights were conducted over a central area, where the UAV-derived raster data confirmed the presence and geometry of the identified enclosure-type features, including wall alignments, enclosed spaces, and associated terrace structures, where preserved.

“Flights were conducted over a central area, where the UAV-derived raster data confirmed the presence and geometry of the identified enclosure-type features, including wall alignments, enclosed spaces, and associated terrace structures, where preserved.”

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

The available SRTM DEM, used for terrain following during programmed flights, has a 30 m resolution, which was not always sufficient in mountainous areas with steep slopes.

“Finally, the available SRTM DEM, used for the terrain following during programmed flights, has a 30 m resolution, which was not always sufficient in mountainous areas with steep slopes.”

Probably not a substantial claim. It may state two separate claims as one.

Many of the features documented are located in areas of limited present-day circulation, making this work a significant contribution to the identification of sensitive heritage zones.

“Many of the features documented are located in areas of limited present-day circulation, making this work a significant contribution to the identification of sensitive heritage zones.”

Probably not a substantial claim. It may be too vague to check against evidence. The paper’s conclusions may not depend on it.

These findings may contribute to Indigenous land tenure processes under Argentinian constitutional rights and territorial ancestral land rights (Argentine law No. 26.160).

“Beyond confirming patterns previously proposed in the literature, these findings support and add to previous studies done with different methods10,29,31, reinforce previously collected ethnographic data about Indigenous agropastoral practices14, and may contribute to Indigenous land tenure processes under Argentinian constitutional rights and territorial ancestral land rights (Argentine law No. 26”

• cites Argentine law No. 26.160Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim. The paper’s conclusions may not depend on it.

A targeted pedestrian ground-truthing survey was required to provide potential chronological context for the identified features.

“As with most remote sensing approaches, a targeted pedestrian ground-truthing survey was required to provide potential chronological context for the identified features.”

It may state two separate claims as one.

Although the satellite-based machine learning procedure performed well in detecting certain features, current UAV data was essential for improving spatial resolution and enabling a more accurate assessment of present-day conditions.

“Although the satellite-based machine learning procedure performed well in detecting certain features, current UAV data was essential for improving spatial resolution and enabling a more accurate assessment of present-day conditions.”

Probably not a substantial claim. It may be too vague to check against evidence.

The dataset offers communities an evidence-based resource to support informed decision-making processes aimed at the protection and long-term stewardship of their cultural heritage.

“In this way, the dataset offers communities an evidence-based resource to support informed decision-making processes aimed at the protection and long-term stewardship of their cultural heritage.”

It may state two separate claims as one.

While satellite-based data alone are not always sufficient to determine the precise nature of a feature, the combination of UAV validation and pedestrian survey has shown that when the algorithm detects evidence of human activity, it is generally reliable.

“While satellite-based data alone are not always sufficient to determine the precise nature of a feature, the combination of UAV validation and pedestrian survey has shown that when the algorithm detects evidence of human activity, it is generally reliable.”

It may state two separate claims as one.

The integration of ML with UAV imagery and pedestrian survey is essential for detecting these features, and significantly improves the resolution of current knowledge of water management networks.

“Therefore, although time-consuming, the integration of ML with UAV imagery and pedestrian survey is essential for detecting these features, and significantly improves the resolution of current knowledge of water management networks.”

It may state two separate claims as one.

The combined multiscale results suggest a long-term dispersed occupation in the Cusi Cusi region, characterised by both agricultural and pastoral activities.

“The combined multiscale results suggest a long-term dispersed occupation in the Cusi Cusi region, characterised by both agricultural and pastoral activities.”

Probably not a substantial claim. It may be too vague to check against evidence. The paper’s conclusions may not depend on it.

This case study contributes to the growing global body of work demonstrating the value of cost- and time-effective remote sensing methods for the documentation, preservation, and management of cultural heritage.

“Moreover, this case study contributes to the growing global body of work demonstrating the value of cost- and time-effective remote sensing methods for the documentation, preservation, and management of cultural heritage6–8.”

• cites [6] Prümers, H., Jaimes Betancourt, C., Iriarte, J., Robinson, M. & Schaich, M. Lidar reveals pre-Hispanic low-density urbanism in the Bolivian Amazon. Nature 606, 325–328 (2022). sourceCheck pending
• cites [7] Frachetti, M. D. et al. Large-scale medieval urbanism traced by UAV–lidar in highland Central Asia. Nature 634, 1118–1124 (2024). sourceCheck pending
• cites [8] Lozić, E., Eichert, S. & Štular, B. Interpolation of airborne LiDAR data for archaeology. J. Archaeol. Sci. Rep. 48, 103840 (2023).Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim. It may be too vague to check against evidence. The paper’s conclusions may not depend on it.

In the current context of the resurgence of extractivist activities in Argentina, the scale of the data presented in this study provides a valuable basis for the development of a community-based heritage database.

“In the current context of the resurgence of extractivist activities in Argentina, the scale of the data presented in this study provides a valuable basis for the development of a community-based heritage database.”

• AssumptionRests on an assumption: Argentina is currently experiencing a resurgence of extractivist activities.

It may state two separate claims as one.

Although not every ML-assisted detection could be validated on foot due to the physical constraints of the high-Andean environment, the correspondence observed in the surveyed zones demonstrates that the UAV results were adequate for verifying and expanding the satellite-based detections within the feasible fieldwork limits.

“Although not every ML-assisted detection could be validated on foot due to the physical constraints of the high-Andean environment, the correspondence observed in the surveyed zones demonstrates that the UAV results were adequate for verifying and expanding the satellite-based detections within the feasible fieldwork limits.”