316 Layers of Perception – CAA 2007

Ignacio Fiz – Hèctor A. Orengo

Simulating Communication Routes in Mediterranean Alluvial Plains

Abstract: Societies exchange knowledge, ideas and merchandise throughout their territories. Topography plays a fundamental role in the trajectory of such movements whilst helping to explain the distribution of human constructions. Standard GIS functions have been employed widely to simulate communica- tion routes between settlements, but the straight application of published least cost route models proved inadequate for Mediterranean alluvial plain areas in which seasonal floods become an important factor to acknowledge. The objective of this study is the production of a new model, using topographic and hydro- logic factors as variables from which it would be possible to simulate a route, and test it against known Ro- man itineraries. The selected Roman stretches are – Coll de Pannisars and – Montblanc. The new model shows the need to consider each case individually but also stresses the hydrologic factor, expressed in seasonal floods, as being of prime importance in the creation and development of in Mediterranean alluvial plains.

Introduction accuracy in these environments. The selected mod- els were those developed by: v a n Le u s e n (2002), and The aims of this paper are twofold: to create a model De Si l v a / Pi z z i o l o (2001). They were chosen accord- that allows the simulation of communication routes ing to their different approach to cost simulation. that are adapted to the typical morphology of Medi- The first model includes variables as types of terrain terranean alluvial plains and to evaluate which and transport means while the second one is topog- factors contribute to the setting of Roman roads in raphy dependent. these areas. Those aims are intimately related since The Roman routes Coll de Panissars – Girona and the setting of Roman roads was highly influenced Tarragona – Lérida, documented in the Vicarello’s by the physiological character of the terrain. Techni- Goblets and the respectively as cal and cultural aspects also play an important role Summo Pyreneo – Gerunda and ab Asturica-Terra- and are, thus, taken into consideration. cone, were chosen to test the models. GIS has the potential to assist these studies, Numerous historical and archaeological studies through the generation of cost surfaces adapted were consulted to properly document these routes to environmental (topography being the most sig- but, ultimately, the most interesting sources were nificant) and cultural factors, suggesting least cost those from the Roman period, describing contempo- route (LCR) models. These can be updated as new rary itineraries: Vicarello’s Goblets, Peutinger Table relevant information is implemented on the cost (medieval copy of the Roman cursus publicus) and surface. GIS-related archaeological literature has the Antonine Itinerary. Old maps from the 1920s made wide use of such models (De Si l v a / Pi z z i o l o proved valuable for the identification of old cattle 2001; v a n Le u s e n 2002; Be ll / Wi l s o n / Wi c k h a m 2002; movement paths, known as camins rals, that have Ej s t r u d 2005). been commonly acknowledged as following the trace of ancient roads.

Methodology Different digital map sources were tested in an at- tempt to check their capability: a 100 m2/cell digital To tackle the objectives of this study various least terrain model (DTM) and a 30 m2/cell DTM, freely cost route models from the literature were applied distributed by the Cartographic Institute of Cata- to documented Roman roads in the Catalonian lonia (ICC), appeared to be inadequate since some alluvial plains in order to check their predictive important landmarks, such as mountain passes or Modelling Movement and Perception 317

sharp riverbanks, were smaller than the minimum sink identification and filling operations. Then, cell size these maps offer. In the end, a 5 m2/cell both flow direction and flow accumulation (FA) op- DTM was produced from digital topographic maps erations were performed to finally obtain a rivers of a scale of 1:5000 with spot heights, also distributed network. A layer named “river” was also created to by the ICC. This was performed using the “natural depict the existence of permanent water courses. neighbour” interpolation algorithm implemented in The effective friction (EF) was calculated employ- ESRI’s ArcGIS 9 software. ing De Silva and Pizziolo’s formulae (De Si l v a / The resulting DTMs were checked against or- Pi z z i o l o 2001). In order to make the rivers and the thorectified vertical aerial photographs of the study effective friction layers proportional an index was areas, taken in 1956 by the American army, in order developed: to confirm that these areas were not modified signif- icantly during the past fifty years as was the case for Index = Max(EF)/Max(river, FA) most of the Spanish territory. Old maps were also employed to this end. The hydrophobia factor (the tendency to avoid rivers The first implementation of the published models or water flows) was calculated as follows: suggested an over reliance on slope as a cost factor, producing long detours in the search of soft inclina- Hydrophobia = (FA + river) × index tions and also a tendency to follow riverbeds while constantly crossing from one bank to the other. Once the Hydrophobia factor was obtained it was These models have proved of great efficiency in sug- added to the effective friction values to calculate the gesting least cost routes in mountain environments final cost surface: but they failed to explain the setting of communica- tion routes in plain areas designed not only for the Cost surface = EF + Hydrophobia movement of people but also for goods. Coll de Pannisars – Gerunda route presented an additional problem: some of the areas crossed by Model Application the LCR model were marshlands during the Roman period. Although this is merely a consequence of in- Summo Pyreneo – Gerunda Route sufficient landscape modelling, it also suggests the problems associated with the tendency of the LCR The Hispanic stretch of the Roman road between models to follow flat areas. Gades (modern Cadiz) and Rome is known as Via Ch e v a l i e r (1997, 107) has pointed out the need to Augusta. This road was organised during the reign avoid marshlands, moor lands, swamps or flooding of on the prehistoric pathway known as areas in the location of Roman roads. These environ- Via Heraclea (No ll a / Ca s a s 1997, 142). ments are supposed to have a higher friction index The stretch to be analysed in this study goes from and, consequently, can lead to significant delays the mountain pass known as Coll de Pannisars1 during the journey. Seasonal floods are common (Summo Pyreneo) to the city of Girona (Roman in southern Mediterranean countries and not only Gerunda). Two itineraries have traditionally been do they render transportation difficult but they can proposed to describe this stretch: the east route, clos- also interrupt passage while their effect on routes er to the coast, and the west route (Fig. 1). The first creates a need for continuous repairs of river fords, simulations, applying the “Hydrophobia model” as bridges and roads. a cost surface, resulted in a LCR closely resembling The importance of water-related factors in route the suggested east route. However, when flow ac- modelling analysis became evident and led to the cumulation values are significantly incremented in creation of a rivers’ network for a GIS-based hydro- the cost surface map the resulting LCR varies signif- logical analysis (Je n s o n / Do m i n g u e 1988). In order icantly, adapting itself to the west route’s proposed to ensure the creation of a continuous vector-based course. rivers network, that is, without any gap in their In order to choose between those two different trail, the 5 m DTMs were subjected to small-scale LCRs (and the literature’s suggested routes that

1 This pass over the communicated the (Narbonense Roman province) with the (Tarraconense Roman province). It was monumentalised through the so called Pompeius’ Trophy. 318 Layers of Perception – CAA 2007

Fig. 1. Predicted routes using the hydrophobia model. The left route was developed increasing the flow accumulation values for this area. they so closely resemble) it was decided to calcu- the over the route, following the distanc- late the total distance between Summo Pyrenae es indicated in the goblets, produced interesting and Gerunda as reported in Vicarello’s Goblets results, since Cinniana, the only unlocated man- and compare it to the LCR total distance. sio, coincided with a series of toponyms closely Moreover, Vicarello’s Goblets document the resembling that of Cinniana: Boscos de Cinyana existence of five mansio (an official stopping (forests of Cinyana), Riera de Cinyanella (river place on a Roman road) in this stretch: Summo of Cinyanella), Camps de Cinyana (fields of Cin- Pyrenae (Coll de Pannisars), Deciana (La Jon- yana). The suggested spot for Cinniana mansio quera), Iuncaria (Figueras), Cinniana (unknown was also very close (about 400 meters) to a series location), Gerunda (Girona), including the dis- of important Roman archaeological sites among tance between each one of those mansio (Fig. 1). which a ceramic production centre is the most The sum of the distance between all mansio significant (Ma r t i n 1981). favours the predicted west route as the one de- Thus, in conclusion, the traditionally sug- scribed in the Vicarello’s Goblets. The location of gested west route is most probably the principal Modelling Movement and Perception 319

Fig. 2. Francolí River flooding in 1994. Fig. 3. Francolí River flooding model and the different predicted routes. Flooding area in dark blue, the normal route, coinciding with that described in Vicarel- volume of water in light blue. Roman sites represented as red dots. lo’s Goblets. Worth noting is also the importance of river level risings in the setting of communica- ates from the old pathway to follow a flatter area tion routes, since the increment of flow accumu- close to Francolí riverbed (Fig. 3, stretch B). The old lation values as a cost factor directed the LCR path was in a slightly steeper area away from the through the west route. riverbed, probably due to the Francolí seasonal floods. Evidence for the floods exists in photo- – Ilerda Route graphs taken in 1994 (Fig. 2), highlighting their importance and virulence/turbulence in this area. The Tarraco-IIerda route, quoted in the Antonine There is also written documentation of multiple re- Itinerary as Ab Asturica Terracone, connects the pair works of the old stone bridge over River Fran- coast with inner regions of the . colí, granting access to Tarragona city. A simulation The stretch of road of interest to this study spans of the River Francolí flooding area was produced from Tarraco (modern Tarragona) to Montblanc. (Fig. 3), which was then added as an extra cost factor Old maps identify this route as an old Camí Ral. to the already developed cost surface. The new LCR Ample archaeological evidence also supports this almost coincided with Camí Ral (Fig. 3, stretch A) hypothesis: a miliarium from the third century AD (max. difference of 250 meters) and passed through and an inscription from a roads supervisor (curator the Centcelles Villa; arguably the most important vil- viae) are both located on this pathway. Furthermore, la in the province. Furthermore, this flooding model this route connects some of the most important provided some explanation for the location of those Roman villas of the province. Roman sites near River Francolí, as they were all lo- The “hydrophobia model” cost surface was de- cated at the edge of the flooding area, indicating the veloped and applied to this area. The resulting pre- importance of seasonal river floods not only in the dicted LCR overall favours the old path connecting setting of roads but in all constructed archaeological Tarragona and Montblanc rather than the modern features, with the possible exception of some related road. Nonetheless, at some point the route devi- principally to cultivation. 320 Layers of Perception – CAA 2007

Conclusions References

The methodological approaches suggested here are Am i c h / Ca s a s 1998 a first attempt to incorporate and combine multiple N. Am i c h / J. Ca s a s , La Tabula Peutengeriana i la Xarxa parameters in Roman route modelling in Mediter- viària al nord-est de Catalunya. Una nova lectura. In: ranean alluvial plains. The exploratory analysis pre- Institut d’Estudis Ceretans, Comerç i vies de comu- sented here has the potential to contribute to better nicació (1000 aC–700 dC, XI Colloqui Internacional route models. Historical sources also appear to be d’arqueologia de Puigcerdà (Puigcerdà 1998) 175–181. an invaluable tool in the creation and evaluation of Be ll / Wi l s o n / Wi c k h a m 2002 these models. T. Be ll / A. Wi l s o n / A. Wi c k h a m , Tracking the Samnites: Although, the primary focus of the analysis was Landscape and Communications Routes in the Sangro to develop a cost surface model adapted to the study Valley (Italy). American Journal of Archaeology 106:2, areas’ local character, this study led to a deeper un- 2002, 169–186. derstanding of the numerous factors involved not Ch e v a l i e r 1997 only in the planning of Roman roads but also in the R. Ch e v a l i e r , Les voies romaines (Paris 1997). distribution of habitation sites in the landscape. The De Si l v a / Pi z z i o l o 2001 hydrological proxy, and in particular seasonal flood- M. De Si l v a / G. Pi z z i o l o , Setting up a “Human Cali- ing, proved to be an important factor in the choice brated” Anisotropic Cost Surface for Archaeological of communication routes in Mediterranean alluvial Landscape Investigation. In: Z. St a n č i č / T. Ve l j a n o vs k i plains, at least, during the Classical period. The de- (e d s .) Computing Archaeology for Understanding the velopment of this model was also useful in contrast- Past. Computer Applications and Quantitative Meth- ing historical sources such as Vicarello’s Goblets. ods in Archaeology 2000. BAR International Series 931 In accordance with these results, it becomes ap- (Oxford 2001) 279–286. parent that cost surface models and other types of Ej s t r u d 2005 GIS-based predictive analysis must always adapt B. Ej s t r u d , Cost Surface Analysis and Ancient Roads: to local particularities and seasonal variations, as a Comparison. In: J. Be r g e r (e d .), Temps et espaces de straightforward application of general models or l’homme en société. Analyses et modeles spatiaux en models created for different areas proved inad- archaéologie. XXVe rencontres internationales d’arché- equate. ologie et histoire d’Antibes (Antibes 2005) 135–140. Least Cost Route analysis must be employed as Er i c s o n / Go l d s t e i n 1980 an exploratory tool and never as an explanatory J. Er i c s o n / R. Go l d s t e i n , Work Space: a New Approach mechanism. Therefore, rather than offering a defini- to the Analysis of Energy Expenditure within Site tive answer the results presented here must be con- Catchments. In: F. Fi n d l o w / J. Er i c s o n (e d s .), Catchment sidered as a base to be further tested. Analysis: Essays on Prehistoric Resource Space (Los Angeles 1980) 21–30. Je n s o n / Do m i n g u e 1988 Acknowledgements S. Je n s o n / J. Do m i n g u e , Extracting Topographic Struc- ture from Digital Elevation Data for Geographic Infor- The authors would like to thank Ana Ejarque (Cata- mation System Analysis, Photogrammetric Engineering lan Institute of Classical Archaeology, Tarragona, and Remote Sensing 54:11, 1988, 1593–1600. Spain) and Dr. Mark Gillings (School of Archaeol- Ma r t i n 1981 ogy & Ancient History, University of Leicester) for M. Ma r t i n , El taller de ceràmiques emporitanes de Fell- their comments and support. ines. Estudi General: Revista de la facultat de Lletres de la Universitat de Girona 1, 1981, 37–50. No ll a / C a s a s 1997 J. No ll a / J. Ca s a s , Nouvelles donnés sur la via Augusta dans le secteur nord-est de la Catalogne. In: G. Ca s t e ll - v i / J. Co mp s / J. Ko t a r b a / A. Pe z i n (e d s .), Voies romaines du Rhône à l’Ebre: via Domitia et via Augusta. dAF 61 (Paris 1997) 142–148. Modelling Movement and Perception 321

v a n Le u s e n 2002 Ignacio Fiz M. v a n Le u s e n , Pattern to Process: Methodological In- vestigations into Deformation and Interpretation of Spa- Catalan Institute of Classical Archaeology tial Patterns in Archaeological Landscapes. PhD Thesis Universitat Rovira i Virgili (Leiden 2002). http://irs.ub.rug.nl/ppn/239009177 [31 Pl. Rovellat, s/n. Dec 2007]. 43003 Tarragona, Spain [email protected]

Hèctor A. Orengo

Catalan Institute of Classical Archaeology Universitat Rovira i Virgili Pl. Rovellat, s/n. 43003 Tarragona, Spain [email protected]