Understanding and Predicting Roman Landscapes

The Roman Landscape Characterisation and Prediction Project (RoLCAP) is a privately funded research programme at the School of Archaeology in the University of Oxford that uses archaeological evidence to better understand settlement and farming in the Roman countryside, and develops innovative modelling to predict where new sites might await discovery.

RoLCAP arose out of the findings of previous research projects that looked at the agricultural landscape of Roman Britain, in particular the  English Landscapes and Identities  (EngLaId ) , the  Rural Settlement of Roman Britain  (RSRB), and  Fields of Britannia  projects. Of these projects, EngLaId took very much a big picture approach to the archaeology of England from the  Bronze Age to Domesday Book , collating a massive database of archaeological monuments and events from which broad scale analyses could be produced. RSRB focused purely on rural England and Wales in the Roman period, producing and analysing a database of detailed excavated information, but focused largely on sites excavated through open area excavation. RoLCAP has been attempting to strike a balance between these two approaches, by being more focused on detail than EngLaId but by also taking into account less well-excavated sites than RSRB. What could this type of approach add to the picture of rural life in Roman Britain?

Clearly, with a small team we would not be able to conduct a nationwide survey, so we focused on a series of case studies in areas known to have seen large amounts of modern archaeological intervention (primarily associated with development). These included the towns of Didcot (Oxfordshire), Swindon (Wiltshire), Wokingham and Bracknell (Berkshire), and Andover and Basingstoke (Hampshire). We also conducted case studies on the Berkshire Downs due to good preservation of ancient field systems in that area, in the Vale of White Horse (Oxfordshire), near the village of Chalton (Hampshire) due to a previous landscape scale survey, and around the small towns of Ampthill and Shefford (Central Bedfordshire).

For these case studies, we gathered together data from local Historic Environment Record (HER) offices and enhanced it with RSRB data for sites they had also analysed and with material from recent " grey literature " reports. We initially started with the Didcot, Swindon, and Wokingham/Bracknell case study areas, but later added more to fill in the picture as we explored the data and our questions.

Having gathered all of this data, the main question that then arises is whether we can begin to understand the character of the Roman farming landscape? We approached this question by testing densities of farms and fields against various potentially explanatory factors.

Once we have characterised the nature of the Roman farming landscape across our study areas, we can then use our data and quantifications to make predictions for new areas or to predict where missing sites might be in the current study areas.

This first predictive model is an attempt to suggest where any missing farms might once have been located on the Berkshire Downs, with redder colours meaning a higher probability of a farm having been missed or lost in those zones. The criteria for this model were very simply: slope <5°; greater than 1km from a known farm; and less than 1km from a known field. This is a very straightforward example of how gathering as much data as possible helps identify potential gaps, to guide future survey and investigation

The second model was attempting to model the predicted number of farms based on RoLCAP data for the  Oxford to Cambridge Arc , an area targeted by regional development agencies for growth and development.

We took the data from the initial set of case study areas (excluding the Vale of White Horse, and Avebury World Heritage Site), and used it to model predicted farm numbers by soil type. The model was collated by hexagons to make visualisation easier (as then all analytical units cover the same amount of space) and to convey a sense of uncertainty, as the predictive model is very simple.

The very high numbers of predicted farms at the Cambridge end of the Arc suggests that more study areas would be needed here to improve the predictive power of the model.

This last model is our theoretical reconstruction of Roman landuse. It shows areas of arable (defined using field system records), and areas of heath, meadow, pasture, and woodland, defined based upon soil, geology, and the character of the land surface (e.g. slope). The bar charts show a similar, but not identical, breakdown of landscape types seen in pollen analyses for the Roman period. Although clear differences exist, the bar charts do roughly correlate with their surrounding landscape as defined by us, so perhaps this model is more robust than its simple character might suggest. It is clearly more reliable in the areas around our case studies and definitely underpredicts arable (insofar as arable is only modelled where we have data for field systems), but it is a reasonable first attempt at understanding landuse in the Roman period in this part of Britain.

Having demonstrated that using HER and other readily available data, patterns of Roman farming and settlement can be identified on a finer scale than RSRB’s Central and Southern Zones and additionally that Natural Character Areas can be further sub-divided into ‘Character Zones’, one of RoLCAPs main aims has been achieved. The second theme: to test whether the available archaeological and topographic data is sufficient for reliable modelling (prediction) of Roman farming patterns, remains in it’s exploratory phase and may only progress slowly until the potential of multi-factoral analyses can be undertaken on a larger database – a task potentially best achieved by AI applications.

In the short-term, a series of ‘Characterisation Summaries’ based on relevant National Character Areas and Geological and Soil Type Summaries (farm and field values by geology and soil type) will be prepared. Various publication outlets are being explored to disseminate the results more widely and a new phase of RoLCAP is being contemplated. RoLCAP 2 might explore the characterisation and modelling/prediction process in other locations across central and northern England, including the Cheshire-Staffordshire Plain where, despite fewer large-scale development-related archaeological results, the impact of military occupation on patterns of settlement and land-use might be identifiable.