<!-- ARTICLE_BODY_START -->
A laser survey can make an ancient landscape suddenly legible. Banks become lines, lines become enclosures, and a slope that looked unremarkable acquires a geometry. The temptation is to treat the resulting image as history with the trees removed. Yet the image describes a present surface, assembled from measurements and processing decisions. The past enters through an argument about that surface.
A study published in Nature on July 29, 2026 makes the distinction especially important. Martti Pärssinen and colleagues reported an Amazonian airborne survey conducted in 2024, covering 4,497 square kilometres along 4,430 kilometres of flight. Its data documented 432 earthworks, 36 previously known. The much larger number in the paper's title, over 20,000, concerns the authors' regional estimate, combining survey evidence with extrapolation. It is not a count of 20,000 individually scanned monuments. The 2026 study
Both results can matter. The direct observations enlarge an archaeological inventory. The estimate proposes how extensive an ancient constructed landscape might have been. They require different tests, however. Rechecking an individual enclosure asks whether the feature survives alternative processing and field inspection. Rechecking the regional estimate asks whether sampled areas represent unsurveyed ones and whether visibility varies systematically. A dramatic map should invite those questions rather than settle them by appearance.
The useful story of archaeological lidar therefore begins before the coloured relief image and continues well beyond it. It includes a laser's path, a ground classification, a field sampling plan, a chronological model and an agreement about who may circulate the results. Each stage can strengthen the eventual interpretation. Each can also introduce a mistake that a beautiful final illustration conceals.
The instrument measures a reflection
Airborne lidar combines laser ranging with information about the aircraft's position, orientation and scanner geometry. Those components locate reflections in three dimensions, producing a point cloud. NOAA's National Geodetic Survey describes the integration of GPS, inertial measurements, scan angles and calibration data; a range alone does not establish a point's geographic position. NOAA's technical explanation
Consider a deliberately simplified instrument example. A measured round trip of two microseconds corresponds to approximately 300 metres of range when the speed of light is rounded to 300 million metres per second. The division by two matters because the pulse travels out and back. That figure is distance along the beam. Converting it into terrain elevation still requires the instrument's location and beam direction. It would be a category error to label every such range as height above the ground directly below the aircraft.
This distinction also explains why flight geometry belongs in an archaeological discussion. Imagine two strips crossing the same flat clearing. If their points form two slightly separated planes, a later surface model can inherit a seam. A reader seeing only the shaded raster might interpret that seam as a low bank. The appropriate first question is whether the supposed bank follows the landscape or the survey's organisation. This is an analytical example, not a reported defect in either study discussed here.
The ground is one possible reflecting surface among many. Leaves, branches and buildings may also contribute points. The US Geological Survey distinguishes this original cloud from a derived elevation model: a bare-earth product is created by selecting appropriate points and representing the terrain in a regular surface. A terrain raster is consequently a processed interpretation of the measurements, not the measurements themselves. USGS on point clouds and elevation models
An archaeological reader should preserve that separation in language. Saying that an enclosure appears in a terrain model makes the evidential object explicit. Saying that a laser found a ceremonial centre silently adds a cultural classification. The latter may eventually be well supported, but the instrument did not measure ceremony. Its returns become evidence for that interpretation through other observations and reasoning.
Forest removal is a computation, not a transparent canopy
Historic England's 2018 lidar guidance explains why selecting the last recorded return is not equivalent to selecting the ground. Vegetation may still produce the final return, and successful terrain representation depends on vegetation, acquisition conditions and classification. The guidance also warns that interpolation can conceal missing observations and that processing artefacts can resemble archaeological features. The Light Fantastic, technical guidance
A useful way to read a forested survey is to imagine three separate maps. One shows where the instrument recorded returns. Another shows which returns the processing classified as ground. The third displays the surface fitted through those selected points. The third is normally easiest to understand. The first two explain how much confidence its apparent detail deserves.
Here is an illustrative calculation. Suppose a survey averages twenty recorded returns per square metre, and five per cent are classified as ground. Its mean ground-return density is one per square metre. A quarter-metre raster divides that same square metre into sixteen cells. Creating those cells does not create sixteen independent ground observations. Some values will be estimated from neighbouring points, and some neighbourhoods may be much less supported than the overall average suggests.
The numbers are hypothetical, not a recommended survey specification. Returns are also not interchangeable with emitted pulses: a pulse can contribute more than one return. Their purpose is to expose a common advertising ambiguity. A large point-density number is uninformative until the reader knows what was counted and where those points occurred. For an earthwork under trees, a dense canopy cloud cannot compensate automatically for a sparse ground cloud.
OpenTopography's gridding guidance explicitly connects sensible raster size with the spacing of the points used to build it, while warning that there is no universal resolution rule. Its examples distinguish the full cloud from the lower density remaining after ground selection. OpenTopography on choosing grid resolution
The most informative comparison would keep a candidate feature visible beside its supporting points. If an apparent ditch crosses a broad gap with no ground returns, the honest conclusion is that its continuity is unresolved there. Drawing a dashed segment may communicate the evidence better than a seamless shaded surface. An analyst should be able to retain the hypothesis of an enclosure without pretending that every metre of its perimeter has equal observational support.
Which ground does the classifier preserve?
An earthwork creates an awkward classification problem. It rises above, or cuts into, a surrounding surface, yet that departure from ordinary terrain is precisely what an archaeologist wants retained. A procedure designed to remove above-ground objects can therefore remove part of the target. Conversely, a procedure that preserves every small rise can retain vegetation noise.
A concrete published example comes from the 2024 Antiquity study of environmental lidar in Campeche. Luke Auld-Thomas and colleagues found that the original terrain models truncated larger archaeological mounds. They reprocessed the point clouds with parameters suited to archaeological work, improving the representation of ancient architecture. The same paper explicitly reported that its analysis had not yet received field validation and cautioned that some rockpiles could be mistaken for buildings. The Campeche research article
The lesson is more demanding than simply choosing an archaeology-labelled setting. Suppose a hypothetical mound survives one classification but disappears in another. Neither outcome alone decides its status. The analyst can examine the original points, look for coherent profiles, compare adjacent terrain and ask which rule rejected the relevant returns. A meaningful sensitivity test changes an interpretable processing choice and records what happens to the candidate.
It is equally important to preserve failed interpretations. Imagine an inventory containing a hundred proposed features. Ten disappear when a strip-alignment problem is corrected. Deleting their identifiers and renumbering everything would make an old report difficult to reconcile with the revised map. Retaining those identifiers with a withdrawn classification explains the change. It also prevents a later analyst from rediscovering the same processing artefacts as apparently new archaeology.
For a research archive, the minimum useful chain is therefore relational: source acquisition, point classification, surface version, visualisation version, feature interpretation. A feature should point back through that chain. This is a proposed documentation design rather than a claim that all existing repositories implement it. Its practical advantage is that disagreement becomes localised. Two teams can discover that they agree about the points and disagree about the feature's meaning, instead of arguing vaguely about whose map is more convincing.
A further benefit is restraint. If only a supplied raster survives, it may be impossible to test the filtering decision. That does not render the image worthless. It narrows the claim to what can actually be checked and identifies the original cloud as a valuable future acquisition.
Illumination changes what the eye finds
Terrain visualisation is an analytical choice. The Relief Visualization Toolbox's September 2019 manual describes how single-direction hillshading can obscure features aligned with illumination, and how display stretches and neighbourhood-based calculations affect the image. It also records processing parameters and warns about edge effects. These are properties of a particular documented software version, not a statement about every current implementation. Relief Visualization Toolbox 2.2.1 manual
Think of a hypothetical square enclosure viewed under artificial light from the northwest. Two sides may acquire conspicuous shadows while the others nearly disappear. Rotate the illumination and a different pair becomes obvious. Someone tracing only the first image could record an incomplete structure and then invent an explanation for its missing sides. Comparing views tests whether that incompleteness belongs to the terrain or to the rendering.
A local-relief image presents another interpretive challenge. If a broad slope is removed to emphasise small departures, the result can clarify low banks. But a viewer who forgets that transformation may compare the apparent heights of features that do not share the same local reference. The safer reading moves between the enhanced view and an ordinary elevation profile. An enhancement helps locate a question; the underlying values help describe it.
An image caption can carry surprisingly useful information here. It can identify the surface version, grid spacing, visualisation method, scale and any vertical exaggeration. Those details let a reader distinguish a dramatic appearance from a dramatic landform. A low ridge exaggerated for visibility should not become a towering fortification in the prose simply because the illustration makes it look imposing.
The 2026 Historic England standards for aerial investigation and mapping require multiple lidar visualisations, including multidirectional hillshade, within their specified project framework. They also recommend assessing lidar with aerial photographs and explain why surface models can retain useful detail in open areas while terrain models help beneath vegetation. These are standards for that institutional context, not a universal legal requirement. Standards published June 30, 2026
For an independent reader, the principle is simple enough to apply without reproducing an entire professional workflow: ask to see a second view that could challenge the first. A convincing enclosure ought to survive more than a single carefully chosen picture. Where it does not, the instability is part of the result.
Fieldwork should test the empty spaces too
Ground validation is often described as a trip to confirm what the screen has already revealed. That framing wastes some of its value. A field campaign can assess false alarms, missed features, differences between landforms and the reliability of an interpretation under different vegetation conditions. Its strongest contribution may be to show where the initial classification was systematically wrong.
Consider a fictional survey with two hundred candidate mounds. A team visits fifty and finds evidence supporting an ancient constructed feature at forty. That is an eighty per cent confirmation rate among the visited candidates under the team's stated criteria. It is not automatically an eighty per cent rate for all two hundred. If the team selected the clearest shapes and most accessible locations, its sample contains a built-in advantage.
Now imagine that the team also examines twenty areas where the lidar interpretation found no mound, selecting them through a separate, explicit sampling design. Discovering low or damaged features in those areas would reveal a different failure: omission. A survey can have excellent confirmation of its visible candidates while still missing much of the archaeological landscape. Precision among detections and completeness across the landscape answer different questions.
The field record should preserve what was actually observable. A candidate may be inaccessible, obscured by recent debris or impossible to inspect safely. Marking such cases as rejected would turn lack of access into evidence of absence. Marking them as confirmed because the image looks persuasive would do the reverse. An unresolved category is not administrative indecision; it protects the meaning of both the positive and negative categories.
One possible sampling strategy is to divide candidates by size, terrain, vegetation and confidence, then ensure the fieldwork includes examples from each group. Another is to choose independent survey areas before consulting the candidate map, reducing the chance that expectations direct every observation. These are proposed designs for discussion with qualified project teams, not instructions to enter land, clear vegetation or excavate without authority.
Field evidence can itself require interpretation. A visible bank may be old, recently rebuilt or partly natural. Surface material can be displaced. A feature's absence at the expected location might reveal a positioning issue rather than an archaeological error. The label ground truth should not imply that every observation made on foot is final. The advantage of fieldwork is that it supplies different evidence and exposes the terrain model to tests it cannot perform on its own.
A classifier inherits its training history
Automated interpretation can help researchers triage enormous datasets, but it does not remove the need to define a target. In work presented at the 2019 IEEE Big Data conference and posted to arXiv in April 2020, Conrad Albrecht and colleagues combined domain-informed segmentation with neural-network classification of archaeological features. Their example used existing archaeological labels and knowledge from Angamuco. That is evidence for a particular research pipeline, not proof of universal automated discovery. The authors' paper
A hypothetical model trained mainly on intact rectangular platforms might struggle with rounded, eroded or differently built structures. More importantly, it might assign low scores to an entire class of cultural features absent from its training labels. A high average score on familiar examples cannot tell a project how well the model recognises a different architectural tradition.
The evaluation unit matters as much as the score. Suppose overlapping image tiles cover the same feature. If some tiles enter training and neighbouring tiles enter testing, the algorithm may encounter nearly identical terrain on both sides of the split. A geographically separated test asks a harder and more useful question: does the learned pattern transfer? This is an original methodological caution, not an allegation about the cited study's evaluation.
Thresholds also encode the intended use. A low threshold might send many uncertain candidates for expert inspection, accepting false alarms to reduce omissions. A high threshold might produce a cleaner public map while discarding ambiguous features. Neither threshold is inherently honest or dishonest. Trouble begins when the output is described without the selection rule or when a triage system's candidates are promoted into a definitive site inventory.
An archaeologist could make the system more interpretable by reviewing examples at several score levels, recording why particular predictions fail and retaining the unclassified areas. The result would be a map of knowledge and uncertainty rather than a colour-coded assertion of certainty. Especially valuable would be a disagreement set: locations where human interpreters, alternate models or different renderings diverge. Those locations can reveal which assumptions most need testing.
Automation is most useful when it makes scrutiny more selective without making the selection invisible. The scarce resource is not merely computing time. It is informed attention, field access and the capacity to turn a plausible shape into a defensible archaeological statement.
A map flattens several centuries into one surface
The most consequential missing dimension is often time. Two enclosures visible together may never have been used together. A route may cut across an older boundary. A platform can contain several construction phases while appearing as a single object in the current terrain. A spatially coherent map can therefore conceal a historically composite landscape.
The Aguada Fénix research illustrates the evidential division clearly. In their 2020 Nature paper, Takeshi Inomata and colleagues described both airborne lidar and excavation, dating construction to 1000–800 BC through Bayesian analysis of radiocarbon dates. The chronological claim depended on dating evidence and its interpretation, rather than on a laser return. The accessible abstract supports this bounded account; the full article was not available through the access route used here. The authors' abstract and publication record
Chronology also needs a precisely stated event. The age of organic material recovered from a construction deposit is not self-evidently the day that construction occurred. Historic England's 2022 radiocarbon guidance discusses sample selection, the old-wood effect, archaeological context and the need to specify model assumptions. Carbon in an inner tree ring can predate the tree's death, while material can move before entering the deposit eventually sampled. Radiocarbon dating and the Bayesian process
Imagine a fictional ditch with three observations: its cut crosses a buried surface, its earliest fill contains a short-lived plant sample, and a later bank overlies that fill. These observations suggest relations among events, but they do not all date the same event. A sample associated with filling may help constrain when the ditch was already open. It need not identify when digging began. The question is the relationship between the sample and the archaeological event, not simply whether a laboratory supplied a precise number.
A chronological diagram can be more honest than a single date printed beside a polygon. It can distinguish construction, use, modification and abandonment, marking which intervals have evidence and which remain conjectural. Where only a broad stylistic attribution is available, a colour representing a narrow century would give the map a precision the research does not possess.
This has immediate implications for claims of roads, districts and political organisation. A line connecting two centres today is suggestive. To argue that it connected functioning centres in the past requires evidence that the relevant route and settlements overlapped in time. Otherwise the map may join an early monument, a later settlement and a still later boundary into an institution that never existed.
Counting structures is not counting people
Population estimates are attractive because they translate terrain into a human scale. They are also a stack of assumptions. A mapped feature must first be interpreted as a building or residential space. The analyst must then consider how many such spaces were occupied at the same time, how many people used each and how much relevant settlement escaped detection.
A simple hypothetical example makes the leverage visible. Suppose an interpretation identifies one hundred plausible residential platforms. Assuming four residents per occupied platform and complete contemporaneity yields four hundred people. Keeping the same occupancy assumption but making only half the platforms contemporary yields two hundred. Neither answer changed because a new laser point arrived. The difference came from the historical model.
A second ambiguity concerns the denominator. One hundred platforms in a one-square-kilometre block and one hundred scattered over ten square kilometres describe very different spatial arrangements. But choosing the boundary around the densest patch can inflate an apparent density. Including a large unsurveyed margin can depress it. A density statement should make clear whether its denominator is surveyed land, interpretable land, habitable land or an administrative region.
The 2026 Amazonian paper itself acknowledges that some non-detections could reflect dense forest preventing an adequate terrain model rather than an absence of earthworks. Its regional totals consequently deserve to be read alongside its sampling and visibility analysis, rather than as a complete census. The authors also withheld exact locations of two features in contemporary Indigenous territories. Study discussion of detection limits and disclosure
That combination is instructive: the same research can offer a large historical hypothesis and retain concrete limits on what is observed or disclosed. A responsible reader need not choose between enthusiasm and scepticism. The useful response is to identify which further observation would most change the conclusion. Better chronology may matter more than finer pixels; a broader sample may matter more than another spectacular enclosure.
Volume estimates deserve similar care. In a fictional earthwork, an average reconstructed height of half a metre across two thousand square metres produces a thousand cubic metres of apparent fill. Change the assumed original ground surface and the volume changes. Turning that volume into labour requires further assumptions about material, tools, transport, work organisation and construction duration. The calculation can structure a historical question without answering all of it.
Reused surveys change the sample, not just the price
The Campeche study used lidar collected in 2013 for forest monitoring, allowing archaeological analysis outside a survey designed specifically around known sites. Its authors described this as a pseudorandom opportunity relative to archaeological questions. That wording matters: an environmental survey has its own selection logic even when it avoids some archaeological targeting biases. Environmental lidar study
Consider two fictional research programmes. One flies around ten celebrated monuments; the other reuses transects chosen to compare forest conditions. The first provides rich context around places already considered important. The second may expose ordinary or sparsely settled terrain that the first never samples. They answer different questions, and their findings should not be combined as though each square kilometre had the same probability of inclusion.
Reused data can be especially valuable when the original acquisition and processing survive separately. An archaeologist may have a reason to revisit ground classification that a forestry project did not share. Conversely, a dataset excellent for canopy structure may offer limited ground support in precisely the places relevant to archaeological interpretation. A new purpose does not retroactively change the acquisition conditions.
The USGS identifies its Lidar Base Specification as the requirements framework for collections under the 3D Elevation Program; its public index records the June 10, 2025 release of version 2025 revision A. This gives a reader a versioned acquisition context. It does not certify that every archaeological target in a compliant dataset will be detectable. USGS specification index
A practical reading question follows: what task was the data product designed to satisfy? A survey may be entirely successful for its original purpose and still be insufficient for another. Evaluating that mismatch is more informative than dismissing the source as poor quality. It also helps a project decide whether to commission new coverage, reprocess existing points or redirect fieldwork toward the places where uncertainty is greatest.
The archive can then become a sequence of investigations rather than a collection of orphaned images. Each new interpretation adds value if it preserves the original evidence, states its changed assumptions and explains how another analyst could revisit the result.
Authority belongs in the research design
The terrain may be newly mapped to a university team without being unknown to the people who live with it. A research announcement should therefore specify what is new: a recorded measurement, a published inventory, a particular interpretation or recognition within a disciplinary literature. Calling a place undiscovered can erase existing knowledge without adding any scientific precision.
The CARE Principles for Indigenous Data Governance were published by Stephanie Russo Carroll and colleagues in November 2020. They centre collective benefit, authority to control, responsibility and ethics, complementing data-management approaches concerned with findability and reuse. The framework explicitly includes data about Indigenous territories and environments, not only personal records. It is a governance framework, not a substitute for the applicable law or a universal consent form. The original CARE publication
For a hypothetical lidar project, those principles raise decisions that should occur before acquisition. Who helps define the research question? Who can review proposed interpretations? Which community institutions should receive usable data and training? Who decides whether a location can appear in an open repository? What happens if a proposed secondary use conflicts with the original agreement? Asking these questions after a public map has circulated leaves fewer meaningful choices.
The distinction between consultation and authority is substantive. A meeting at which concerns are heard does not establish who can approve a release. Nor does participation by one person automatically settle representation across a whole community. A project should describe its actual relationships and permissions rather than allowing a generic acknowledgements paragraph to imply a broader mandate.
There is also an intellectual issue. Community knowledge may change the classification of a feature, the relevant historical question or the boundary of what should be studied. Treating that knowledge solely as supporting colour for an interpretation already completed wastes its analytical contribution. Conversely, researchers should not publish knowledge simply because it makes the article more compelling. Its custodians may distinguish information appropriate for public explanation from information requiring restricted circulation.
A strong collaboration can produce several outputs for different purposes: an authorised scholarly argument, a community-controlled detailed dataset, accessible local teaching material and a restricted conservation record. Those are proposed possibilities, not a claim about the arrangements of the studies cited here. Their value is that public communication need not be the only measure of whether the work has returned something useful.
An open image can become an access route
Archaeological data can create new risks when it makes vulnerable places easier to find. The issue extends beyond a table of coordinates. A detailed terrain tile, a recognisable road junction or an annotated screenshot can reveal location indirectly. Removing a coordinate column while publishing everything needed to reconstruct it is a weak form of protection.
Arches National Park's disclosure policy, updated April 18, 2025, illustrates one institutional response. It differentiates classes of cultural sites according to vulnerability and protective arrangements, withholding location information for its most vulnerable class. That is a specific park policy operating within United States legal responsibilities, not a rule automatically applicable to archaeology elsewhere. Arches cultural-resource disclosure policy
A hypothetical research team preparing a public demonstration could use a synthetic terrain example, a location already authorised for interpretation or a deliberately generalised regional figure. Each choice serves a different purpose. Synthetic data can explain filtering without exposing a real place. Generalised mapping can show a regional pattern while withholding the detail needed for unauthorised visits. Neither should be represented as unmodified survey evidence if it has been altered.
Restricted access need not mean that an interpretation is beyond challenge. A qualified reviewer might inspect sensitive data through an agreed process while public readers receive the method, non-sensitive summaries and a clear account of what is withheld. The exact arrangement depends on rights, jurisdiction and custodial decisions. The important editorial obligation is to explain the limitation rather than suggesting that unrestricted downloads are the only form of scientific transparency.
Protection also changes the meaning of a successful announcement. If publication increases attention but leaves no resources for management, a discovery can burden the people responsible for the place. A useful research budget therefore considers the consequences of visibility: documentation, local capacity, monitoring and routes for reporting damage. These are practical implications of making a site easier to recognise, not a demand that every paper promise an unlimited conservation programme.
This article intentionally provides no sensitive undisclosed site coordinates, access instructions or detailed locational clues. Its examples concern how evidence is evaluated, which can be explained without turning the discussion into a guide to reaching vulnerable remains.
Revisiting a landscape requires comparable measurements
A second survey can support conservation as well as discovery. Historic England's 2026 mapping guidance notes that earlier coverage may preserve evidence of features subsequently levelled, while later coverage can improve visibility through acquisition conditions or processing. A difference between images can therefore have more than one explanation. Guidance on multi-temporal lidar
Imagine two fictional surveys of a low bank, acquired five years apart. The second terrain model places the bank's crest twenty centimetres lower. That observation could motivate an inspection, but the comparison first needs a common spatial reference and an assessment of stable surrounding terrain. If the whole second survey sits twenty centimetres lower, the apparent change at the bank offers no independent evidence of erosion. If neighbouring stable surfaces agree and only a short bank segment differs, the case for local change becomes stronger.
An uncertainty example shows why small differences deserve care. Suppose, purely for illustration, that each survey has an independent vertical error with a standard deviation of ten centimetres at the comparison location. The standard deviation of their difference would be approximately fourteen centimetres, obtained by taking the square root of the sum of the two squared errors. A twenty-centimetre difference is then only about 1.4 times that illustrative uncertainty. It should not be advertised as an unambiguous measured loss without further evidence. Real errors may be correlated or spatially structured, so the calculation is a teaching example rather than a universal detection threshold.
The pattern of change can be more informative than a single pixel. A coherent depression with adjacent displaced material suggests a different process from a uniform offset or a stripe along a tile edge. Yet pattern recognition still needs corroboration. A new ground-classification routine could alter the representation of an entire feature coherently. Comparing the original clouds and processing histories helps distinguish a changed place from a changed computational description.
A conservation report should also separate observation from attribution. Detecting fresh surface disturbance does not by itself establish who caused it, whether it was authorised or what material was removed. Those questions require other evidence and appropriate investigation. Publicly assigning blame from an elevation difference would exceed what the measurement supports.
The reporting cadence should follow the question. A project monitoring a slowly eroding bank might prioritise repeatable profiles and consistent reference points. A team evaluating a larger land-use change might need broader context. In either case, documenting an unchanged feature is useful when the observations were capable of detecting the change of concern. Repeated pictures alone provide weaker reassurance if each uses different processing and unknown sensitivity.
This shifts the purpose of repeat mapping from producing another impressive image toward maintaining an interpretable history of condition. The scientific value lies in knowing what could have changed, what the observations could detect and what actually remains unresolved.
The most useful final product may be a better question
The Society for American Archaeology's revised principles, adopted in 2024, connect stewardship with collaborative management and recognise both the archaeological record and research documentation as responsibilities extending beyond a project's immediate interpretation. They also address sensitive information in reporting. These are the society's professional principles, not evidence that every project or jurisdiction follows identical rules. SAA's 2024 principles
One way to make those responsibilities concrete is to imagine a future researcher receiving the project ten years later. Could that person distinguish measured points from interpolated terrain? Could they identify which candidates were inspected, which were rejected and which remain unresolved? Could they reconstruct the assumptions connecting a shape to an age or a population estimate? Would the access conditions still be intelligible after staff and software had changed?
A polished illustration cannot answer those questions by itself. The necessary handover may include linked observations, explicit uncertainties and permissions that survive the individual who negotiated them. It also needs enough interpretive explanation to make revisions possible. An archive containing every file but no account of why particular features were classified is large without necessarily being useful.
The next research question should emerge from that uncertainty. If competing interpretations differ mainly over age, another season of denser airborne measurements may have limited value. If the dispute concerns a supposed gap in settlement beneath exceptionally dense vegetation, targeted validation could matter more. If the principal obstacle is permission or community concern, a technical improvement cannot substitute for resolving that relationship.
This is where lidar's real power becomes clearer. It can move archaeology from isolated monuments toward relationships across a landscape, while allowing researchers to formulate more discriminating questions about what happened there. The gain is not that interpretation disappears. It is that the interpretation can be confronted with a much richer spatial record.
The strongest claim at the end of a survey may therefore sound quieter than its first image: these features are supported by these measurements; these relationships remain uncertain; these histories require further evidence; and these people retain a say in what happens next. That statement leaves room for discovery without pretending that a view of the ground is already a complete account of the past.