Tikal has been mapped, charted and re-mapped for a century, and yet on the morning of 14 March we watched a stone road that nobody knew existed surface from a 3.4-billion-point cloud. The sacbé runs for 11.4 kilometres south-east from a satellite plaza near Group H, threads between two minor centres on the Holmul side, and meets the Tikal causeway network at the Maler. It has been under jungle since at least the ninth century. It was hiding under one centimetre of leaf litter.

This is a field note on how we got it, written for anyone who runs vegetation-filtered LIDAR and has ever felt the lurch of "wait, is that real?" in the middle of a processing pipeline.

The campaign in numbers

We flew three passes between January and March with a Riegl VQ-1560 II mounted on a Cessna Caravan, gridded over an 84 km² polygon to the south-east of the central Tikal core. The flight specification was conservative: 600 m AGL, 50 % strip overlap, and a pulse rate that gave us a nominal 28 returns per square metre on the canopy. Once the canopy is filtered out, the ground point density on the densest passes settles around 6 returns/m² — enough to resolve a low limestone parapet to roughly ±15 cm.

The point cloud came back at 3.42 billion returns over three flights, which is large but not exotic. The first pass alone weighs 412 GB on the LAS files.

Filtering vegetation is the entire job

People underestimate how much of an airborne LIDAR campaign is just patient ground-classification. We used a two-stage pipeline: an initial Cloth Simulation Filter through PDAL with a 0.6 m cloth resolution, then a TIN-based refinement pass in LAStools with a step size of 4 m. The defaults in either tool will give you something credible. They will also flatten anything shorter than about 80 cm into "ground", which is exactly the height of a Maya sacbé curb.

The first time we ran the production filter against Pass II, the causeway was invisible. It was being eaten by the classifier as a low rise in the terrain. We re-ran with the cloth resolution dropped to 0.35 m and the step in the TIN refinement halved, and the road appeared — a 4-metre-wide ribbon of slightly raised stone running east-south-east, with two small platforms at intervals that looked exactly like the route markers documented on the Tikal–Uaxactún sacbé by the Carnegie team in the 1930s.

A vegetation filter is not a neutral instrument. Every parameter you pick is a hypothesis about what the ground beneath the trees probably looks like, and the road only exists in the dataset that admits the possibility of a road.

We re-ran with three more parameter sets to make sure we weren't conjuring a feature out of noise. All three retained the road. Two of them also surfaced a second, narrower linear feature about 900 m north — possibly a related approach, more likely an old logging trail. We are deliberately not making a claim on that one yet.

What changed when we showed it to people in Flores

The Instituto de Antropología e Historia in Flores has been collaborating on the survey since 2024. We brought the filtered DEM in on a laptop and projected it on the wall of the small office behind the Museo Santa Bárbara. Two of the Guatemalan archaeologists who have been working the Petén basin for over twenty years — and who have walked some of this ground on foot in the dry season — recognised the alignment immediately as fitting a regional pattern of secondary causeways radiating from the major centres. One of them had even mapped a low rubble line in 2017 that turns out to be the surface trace of the same road, three kilometres further south.

That is exactly the kind of triangulation a remote-sensing finding needs: another method, run by someone else, that landed on the same feature without knowing about ours.

What we are doing next

Four things, in order.

1. Ground-truth two transects. A short walking survey in October will put a tape across the road in two places and confirm the curb morphology against the LIDAR DEM. This is non-invasive — no excavation, no surface scraping.
2. Publish the filtered DEM. The cleaned 1 m DEM for the entire 84 km² polygon will go to the Open Topography repository in November under a CC-BY licence, along with the PDAL pipeline files and the LAStools parameter sets. This is non-negotiable; if you can't reproduce the filter, you can't argue with the finding.
3. Run photogrammetric documentation of the two platforms along the route using a DJI Mavic 3E and Agisoft Metashape, at <2 cm GSD. The mesh density is overkill, but it gives us a basis for future condition monitoring.
4. Sit on the press release. We are intentionally not naming the smaller plazas the road connects to. There is a real-world looting pressure on undocumented Maya sites in this part of the basin and we will not be the reason a new one is added to the list.

A note on what this is and isn't

This is not a "lost city". It is one road, 11 km long, between sites that are already in the archaeological record. The result matters because it adjusts the catchment maps that we use to reason about Classic Tikal's labour reach — a southern lobe that we previously bracketed at about 14 km has to be extended by at least another 7. That has knock-on effects for the carrying-capacity estimates that have been argued back and forth in Petén archaeology for two decades.

But it is also a small case study in how the NCALM-class instrumentation is now sensitive enough that the bottleneck in this kind of survey is no longer the hardware — it is the patience and care you apply to the vegetation filter. The road was always in the data. We just had to ask the right way.

I will be back at the Flores office in October. If you work on Petén LIDAR and want to swap pipeline files, write to [email protected].

— Elara