Modelling Human Influence at Koh Ker Using Multiple Proxies

understanding the urban trajectory through sedimentary archives

W. Christopher Carleton | 2026-09-30

Introduction

Placeholder for Koh Ker landscape photograph

Prasat Prang, Koh Ker

Report on Carleton et al. (2026).

Overview

01 · The challenge

Reconstructing Koh Ker.

02 · The LHIF

Viewing proxies in sediments as latent human influence.

03 · The Pilot and Results

Ceramics and magnetic susceptibility from the pilot trenches (and new data).

04 · Recent Developments

Next phase: core sampling and more proxies.

THE URBAN SAMPLING PROBLEM

Space and Time

Lidar and survey

  • Extensive spatial coverage.
  • Limited chronological depth.

Angkor Wat Lidar

Excavation

  • Temporal depth.
  • Limited spatial coverage.

Catalhöyük

Source: https://www.flickr.com/photos/catalhoyuk/

How can we reconstruct an urban trajectory between these scales?

Different Proxies, Different Measurements

  • Proxy records differ in resolution, distribution, and measurement structure.
  • Joint inference needs a model linking observations to a shared process.

Proxy Process

Research Aim

Estimate past human influence at Koh Ker by combining heterogeneous sediment proxies to understand urban trajectories.

Proxy Fusion

THE LATENT HUMAN INFLUENCE FIELD

What Is a Latent Field?

  • Latent: Not directly measurable.
  • Field: intensity or properties can vary throughout space and or time.
  • e.g., temperature

Temperature Field

Latent Human Influence Field (LHIF)

  • Latent: human influence is inferred from its traces.
  • Field: influence can vary across the sediment archive (translated into time).
  • many proxies are noisy observations of a shared LHIF.
  • Proxy records differ in resolution, distribution, and measurement structure.
  • Joint inference needs a model linking observations to a shared process.

Theoretical LHIF

KOH KER PILOT AND RESULTS

Site and Context

Cannonically, Koh Ker was established by Jayavarman IV in 921 CE and became the sole capital from 928–944 CE. The site represents a brief but highly distinctive phase of Khmer urbanism, political history, monumental architecture, and sculpture. But it contains a much longer sedimentary record of human activity, which can be reconstructed through the LHIF framework.

Koh Ker in Regional Context

Pilot Study

  • Eight test trenches were dug in 2024 and we have just published the first set of results.
  • Paired magnetic susceptibility and ceramic abundance.
  • Rapid, low-impact excavation as a potential survey strategy.

Koh Ker Trench Locations

Trench Example

Report on Carleton et al. (2026).

Two Observation Channels

Ceramics

Ceramics Data

Magnetic Susceptibility

Magnetic Susceptibility

Model Structure

  • Bayesian hierarchical latent-variable model.
  • Proxy channels linked within a causal framework.
  • Depth-wise reconstructions for the pilot trenches.

Model Diagram

RESULTS

Reconstructing Human Influence

Per Trench Data

Ceramics Channel

Ceramics channel across all 8 trenches. The model captures the depth-wise variation in ceramic abundance, which serves as a proxy for human activity. The Latent influence field is inferred from the ceramics data, showing peaks corresponding to periods of increased human activity.

Ceramics Comparison

Magnetic Susceptibility Channel

Magnetic susceptibility channel across all 8 trenches. The model captures the depth-wise variation in magnetic susceptibility, which serves as a proxy for human activity. The Latent influence field is inferred from the magnetic susceptibility data, showing peaks corresponding to periods of increased human activity.

Magnetic Susceptibility Comparison

pXRF Channel

pXRF channel across all 8 trenches. We finished the planned pXRF analysis. Here we show multidimensional differences in each trench sample between the sterile bottom layers and upper sediments. As with magnetic susceptibility, we observe depth-wise variation in elemental composition, another proxy for human activity. The Latent influence field model will be updated to include this new channel along with FTIR data (which we are currently processing) and isotope data.

pXRF Comparison

Next Pilot Steps

Time

Co-registration of trenches and dating of sediments.

Sampling

Increased spatial coverage and additional proxies (eg FTIR, pXRF).

Coregistration

NEXT RESEARCH PHASE

Pinhole Sampling

At the moment, we have a “pinhole” view of the latent human influence field. The pilot trenches provide depth-wise reconstructions, but spatial coverage is limited. Future work could expand trenching and sampling to better understand the spatial distribution of human influence across the site.

Pinhole Sampling

2026 fieldwork update

We completed a trial terrestrial core sampling survey in March 2026, with dozens of cores collected across the site.

Coring in the field, step 1

Coring in the field, step 2

Coring in the field, step 3

References

Carleton, W. Christopher, Thon Tho, David Brotherson, et al. 2026. “Bayesian Modelling of the Latent Human Influence Field in Archaeological Urban Sediments at Koh Ker, Cambodia.” Journal of Archaeological Science 193: 106663. https://doi.org/10.1016/j.jas.2026.106663.

QR code linking to the paper DOI: 10.1016/j.jas.2026.106663