Introduction

The placement of Iron Age hillforts across the British landscape has been a subject of archaeological debate for over a century. Two dominant hypotheses have emerged: the defensive model, which argues that hillforts were positioned to maximize visibility and control over surrounding territory, and the subsistence model, which proposes that proximity to fertile, workable land was the primary determinant.

The Welsh Marches offers an ideal study area. The region contains over 120 documented hillfort sites across varied topography, from the Shropshire Hills to the Black Mountains.

Study Area and Data Sources

The study area encompasses approximately 4,800 km² of the Welsh Marches. Elevation ranges from 30m AOD in the river valleys to 886m at the summit of Plynlimon.

Dataset	Source	Resolution
Hillfort locations	Historic Environment Records	Point data
Elevation	Environment Agency LiDAR	1m DTM
Soils	Cranfield Soilscapes	1:250,000
Land cover reconstruction	Fyfe et al. (2013)	5km grid
Geology	British Geological Survey	1:50,000

We identified 127 confirmed hillfort sites from the Clwyd-Powys, Herefordshire, and Shropshire HERs.

Methodology

Viewshed Analysis

For each hillfort, we calculated a cumulative viewshed using the LiDAR DTM at 5m resolution. Observer height was set at 1.7m, following Lock and Harris (1996). The total visible area (TVA) within a 5km radius was computed for each site.

Soil Fertility Index

We constructed a composite soil fertility index (SFI) combining natural soil drainage, soil texture, and slope gradient. The SFI for each hillfort was calculated as the mean value within a 2km catchment radius.

Predictive Model

A MaxEnt model used hillfort locations as presence points with environmental layers including TVA, SFI, elevation, slope, distance to watercourse, distance to ridge line, and topographic prominence. Validated using 10-fold spatial cross-validation.

Results

Viewshed Analysis

The mean TVA for hillfort sites was 12.3 km² (σ = 4.8), compared to 8.1 km² (σ = 5.2) for random points (p < 0.001). The distribution was strongly bimodal: 34 hillforts in the Shropshire Hills showed TVA exceeding 18 km², while 28 lowland sites had values below 7 km².

Soil Fertility

Lowland hillforts showed significantly higher catchment SFI (mean 3.8) than upland sites (mean 2.4). The Herefordshire cluster sat atop some of the richest soils in the study area.

Predictive Model

The MaxEnt model achieved an AUC of 0.82.

Variable	Contribution (%)
Topographic prominence	28.4
Soil fertility index	22.1
Total visible area	18.7
Elevation	14.2
Distance to watercourse	9.3
Slope	4.8
Distance to ridge	2.5

Regional Variation

Upland (n=58): Viewshed (31.2%), Prominence (25.8%), Elevation (18.4%) Lowland (n=69): Soil fertility (34.7%), Distance to water (22.3%), Prominence (19.1%)

Discussion

Our findings reveal a landscape of strategic pluralism: upland hillforts optimized for territorial surveillance, lowland hillforts optimized for resource control. The prominence vari