Geomorphology 115 (2010) 67–77 Contents lists available at ScienceDirect Geomorphology j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / g e o m o r p h Reconstructing the Roman topography and environmental features of the Sarno River Plain (Italy) before the AD 79 eruption of Somma–Vesuvius Sebastian Vogel a,⁎, Michael Märker b,c a b c German Archaeological Institute, Germany, Podbielskiallee 69-71, D-14195 Berlin, Germany Heidelberg Academy of Sciences and Humanities, University of Tübingen, Rümelinstraße 19-23, D-72070 Tübingen, Germany Department of Soil Science and Plant Nutrition, University of Florence, Italy a r t i c l e i n f o Article history: Received 10 June 2009 Received in revised form 9 September 2009 Accepted 18 September 2009 Available online 1 October 2009 Keywords: Sarno River plain Somma–Vesuvius Paleo-topography Paleo-environment Modeling Landscape reconstruction a b s t r a c t A methodology was developed to reconstruct the Roman topography and environmental features of the Sarno River plain, Italy, before the AD 79 eruption of the Somma–Vesuvius volcanic complex. We collected, localized and digitized more than 1800 core drilling data to gain a representative network of stratigraphical information covering the entire plain. Besides other stratigraphical data including the characteristics of the pre-AD 79 stratum, the depth to the pre-AD 79 surface was identified from the available drilling documentations. Instead of a simple interpolation method, we used a machine based learning approach based on classification and regression trees to reconstruct the pre-AD 79 topography. We hypothesize that the present-day topography reflects the ancient topography and related surface processes, because volcanic deposits from the AD 79 eruption coated the ancient landscape. Thus, ancient physiographic elements of the Sarno River plain are still recognizable in the present-day topography. Therefore, a high-resolution, presentday digital elevation model (DEM) was generated. A detailed terrain analysis yielded 15 different primary and secondary topographic indices. Subsequently, a classification and regression model was applied to predict the depth of the pre-AD 79 surface combining present-day topographic indices with other physiographic data. This model was calibrated with the measured depth of the pre-AD 79 surface. The resulting pre-AD 79 DEM was compared with the classified characteristic of the pre-AD 79 stratum, identified from the drilling documentations. This allowed the reconstruction of pre-AD 79 environmental features of the Sarno River plain such as the ancient coastline, the paleo-course of the Sarno River and its floodplain. To the knowledge of the authors, it is the first time that the pre-AD 79 topography of the Sarno River plain was systematically reconstructed using a detailed database and sophisticated data mining technologies. © 2009 Elsevier B.V. All rights reserved. 1. Introduction The Plinian eruption of the Somma–Vesuvius volcanic complex in AD 79, which completely buried the Roman settlements of Herculaneum, Oplontis, Stabiae and Pompeii belongs to the most well-known eruptions in history. During this eruption the pumice lapilli fallout was dispersed by stratospheric winds predominantly in southeastern directions up to a distance of more than 70 km from the vent (Sigurdsson and Carey, 2002). Thus, almost the entire Sarno River plain was covered by approximately 3.6 km3 of dense rock equivalent of volcanic deposits (Sigurdsson et al., 1985) showing a specific and therefore easily identifiable stratigraphy. Since these deposits are distinctive and laterally extensive, they can be considered as an ideal chronostratigraphic marker. According to the phase of the eruption and the chemical composition and density of the magma, the Plinian ⁎ Corresponding author. Tel.: +49 30 187 711 358; fax: +49 30 187 711 168. E-mail address: sv@dainst.de (S. Vogel). 0169-555X/$ – see front matter © 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.geomorph.2009.09.031 volcanic deposits of AD 79 consist of white phonolitic pumice and grey tephritic–phonolitic pumice fallout deposits, interrupted as well as overlain by six ash layers of pyroclastic surge and flow deposits (Lirer et al., 1973; Sigurdsson et al., 1985; Carey and Sigurdsson, 1987; Civetta et al., 1991; Cioni et al., 1992; Pescatore et al., 1999, 2001; Sigurdsson and Carey, 2002; Luongo et al., 2003; Pfeiffer et al., 2005). Fig. 1 illustrates the approximate dispersion of the volcanic material during the eruption AD 79 as modeled by Pfeiffer et al. (2005) presuming a northwestern wind profile. The AD 79 eruption covered a wide area of the Sarno River plain nearly isochronously (within 19 h time) with volcanic deposits of some meters. This not only caused a caesura in the existence of an entire landscape, but also contributed to the excellent preservation of the paleo-surface and the ancient paleo-environmental conditions before the eruption of AD 79 (Foss et al., 2002). Consequently, almost 2000 years later, this paleo-surface still remains in situ and is accessible for stratigraphical investigations. In the following the term pre-AD 79 refers to the conditions before the AD 79 eruption of Somma–Vesuvius. 68 S. Vogel, M. Märker / Geomorphology 115 (2010) 67–77 Fig. 1. Isopachs of the thickness [cm] of the white (A) and grey (B) pumice fallout, deposited during the eruption of Somma–Vesuvius AD 79. Modeled by Pfeiffer et al. (2005) (RUN V1). Preliminary geoarchaeological studies in the Sarno River plain were carried out in the 19th century (Ruggiero, 1879). However, it is only since the 1980s that more standardized methodologies of geoarchaeological investigation have developed. Since then there have been several attempts to reconstruct the paleo-landscape before AD 79, among them Cinque and Russo (1986), Livadie et al. (1990), Furnari (1994), Pescatore et al. (1999), Di Maio and Pagano (2003) and Stefani and Di Maio (2003). These studies mostly dealt with the delineation of the coastline before AD 79 and the paleo-course of the Sarno River and its estuary mouth using data from stratigraphical drilling and archaeological excavations. As Pescatore et al. (1999) state about previous coastline reconstructions, the results are sometimes rather different (Fig. 2). That is why the position of the paleo-coastline and the course of the paleoSarno cannot yet be clearly defined from today's state of research. Barra et al. (1989) and Livadie et al. (1990) used about 50 stratigraphical drillings between Torre Annunziata and the Sarno River and archaeological data from Roman and pre-Roman times to reconstruct the pre-AD 79 topography. According to them the ancient coastline was situated approximately 1 km inland, parallel to the modern coast with lagoons that were protected from the sea. There they also assume the location of the ancient marine harbour of Pompeii. Moreover, they predicted the ancient sea level to be approximately 4 m below the modern one. In the hinterland of the ancient coastline they discovered fossil littoral dune deposits of preRoman times (Messigno, 5600–4500 years BP and Bottaro/Pioppaino, 3600–2500 years BP; Cinque, 1991). Furthermore they related predominantly fluvio-palustrine deposits to the paleo-course of the Sarno River whose estuary they assumed southeast of the present-day mouth between Masseria and Resinaro (Barra et al., 1989; Livadie et al., 1990). Furnari (1994) carried out 51 stratigraphical drillings and used some data from archaeological excavations to reconstruct the pre-AD 79 topography. After identifying the pre-AD 79 stratum underneath the volcanic deposits of AD 79 in 39 of 51 drillings, he characterized it using micropaleontologic and granulometric analyses and categorized it according to its origin into four classes: (i) palustrine, (ii) alluvial, (iii) littoral and (iv) Roman paleosol. According to the location of the S. Vogel, M. Märker / Geomorphology 115 (2010) 67–77 69 Fig. 2. Past reconstructions of the pre-AD 79 coastline of the Sarno River plain, before (A) and after (B) 1990 (Iacono, 1913; Malandrino, 1988). drillings he could delineate the approximate position of the paleocoastline, the paleo-Sarno and terrestrial areas of the mainland. Pescatore et al. (1999, 2001) utilized data from 66 drillings and known archaeological sites for studying the pre-AD 79 environment around Pompeii. The stratigraphical data derived from own drillings, past construction projects, from Ruggiero (1879) and Furnari (1994). They state that the ancient shoreline may have run more or less parallel to the modern coast approximately 1 km southwest of the ancient city of Pompeii and 1 km landwards of the present-day coast. The mouth of the paleo-Sarno River they presume to be located south of the modern estuary (Pescatore et al., 1999, 2001). Di Maio and Pagano (2003) used 15 drillings, as well as five archaeological sections and data from archaeological excavations to illustrate a 2.5 km long section of the pre-AD 79 coastline near the ancient town of Stabia (Fig. 3) north of the Lattari Mountains. According to them the coastline before AD 79 followed the terrace of San Marco at a distance of 100 to 200 m, whereas it is possible that, at this location, the Lattari Mountains represented an active cliff. The interpretation of the drilling data showed strong variation in the thickness of the AD 79 eruptive material, especially between the slopes and the adjacent Sarno River plain. Hence, they argue that shortly after their deposition, the volcanic sediments must have been mobilized by heavy rainfall and slope water, moved downslope as a mudflow and were re-deposited as an alluvial fan delta within the plain. Thereby parts of the ancient coastline propagated to the NW. Stefani and Di Maio (2003) used a total of 78 drillings including new ones and those of Furnari (1994) as well as an extensive dataset of past archaeological excavations. They state that the eruption AD 79 coated the ancient topography with volcanic deposits up to a thickness of 3 to 6 m. Consequently the present-day topography more or less reflects the ancient topography. Despite the volcanic deposits, main physiographic elements of the Sarno River plain such as the volcanic hill of Pompeii, the prehistoric dune ridge of Bottaro/ Pioppaino and the terrace of San Marco are still recognizable. Thus, the present-day topography provides valuable clues in reconstructing the ancient topography. As these past studies mostly dealt with separate aspects of the ancient landscape pre-AD 79, there exists as yet no geoarchaeological study covering and investigating the paleo-geomorphology and paleo-environmental conditions of the entire Sarno River plain. Consequently this research project focusses on the reconstruction of the paleo-topography of the Sarno River plain before the eruption of Somma–Vesuvius in AD 79. Therefore we use a methodology that combines stratigraphical information from former and self conducted core drillings, present-day topographical data, and classification and regression methods. With this approach we expect to gain more detailed information concerning paleo-environmental features such as the position of the ancient coastline and the course of the paleoriver network before AD 79. 2. Study area The Sarno River plain situated south of the Somma–Vesuvius volcanic complex is a cultural landscape characterized by continuous anthropogenic activity since the Middle Bronze Age. It stretches across an area of 210 km2 and is drained by the Sarno River and its tributaries. In the west it opens to the Tyrrhenian Sea, whereas in the south and in the east, the plain is flanked by the Mesozoic calcareous rocks of the Lattari and Sarno Mountains that are part of the Southern Apennine chain (Fig. 3). The Sarno River plain belongs to the great graben structure of the Campanian Plain which was formed during Plio-Pleistocene extensional tectonic phases at the end of the Apennine orogenesis and is filled with marine, alluvial and volcanic deposits lying on a carbonate platform (Cinque et al., 1987; De Vita and Piscopo, 2002; Cella et al., 2007). During the last 25,000 years of Somma–Vesuvius' eruptive history, several Plinian eruptions have occurred, each one marking the beginning of a new eruptive cycle after a period of quiescence ranging from 1400 to 4000 years, such as Mercato/Ottaviano (7070–6770 BC), Avellino (1890–1630 BC) and Pompeii (AD 79) (Delibrias et al., 1979; Santacroce, 1987; Rolandi et al., 1993a,b; Andronico et al., 1995; Civetta et al., 1998; Sigurdsson, 2002; Cella et al., 2007). Among them 70 S. Vogel, M. Märker / Geomorphology 115 (2010) 67–77 Fig. 3. Present-day digital elevation model (DEM) and fluvial network of the Sarno River plain with the location of more than 1800 stratigraphical core drillings. Letters indicate sites of interest. The small scale map of Italy shows the Campanian plain, the Sarno River plain and the Apennine mountain ridge. the eruption of AD 79 is the most well-known thanks to the detailed written records of Pliny the Younger. 3. Materials and methods Since today the Sarno River plain is one of the most densely populated and urbanized areas in Italy, there are many constraints in comprehensively applying methods of stratigraphical research over the entire plain. Consequently it is very important to use a rather noninvasive or minimally-invasive scientific approach. Stratigraphical core drillings have become a well-established and standardized method of geoarchaeological prospection in this region since it can be carried out with comparatively minimal technical and bureaucratic effort and costs. To reconstruct the pre-AD 79 geomorphology and paleo-environmental features of the Sarno River plain, the documentation of more than 1800 drillings from construction works, as well as from past archaeological and geological studies, were collected to gain a representative network of stratigraphical information for the entire Sarno River plain. The drillings were localized and digitized using geographic information systems (GIS) (Fig. 3). Furthermore, 20 new core drillings were conducted south and northwest of Pompeii at Moreggine and Boscoreale (Fig. 3). The drillings were carried out mechanically to a maximum depth of 10 to 18 m to extract a core of approximately 15 cm in diameter. By means of the drilling cores the stratigraphy was determined, the volcanic deposits of AD 79 and the pre-AD 79 surface underneath were identified, and the pre-AD 79 stratum was characterized. For this project the most relevant data taken from the drilling documentation were: (i) (ii) (iii) (iv) x–y-coordinates of each drilling point, elevation of the present-day surface, thickness of the volcanic deposits related to AD 79 eruption, depth to the pre-AD 79 surface which is the difference between the present-day surface and pre-AD 79 surface, and (v) characteristics of the pre-AD 79 stratum underneath the volcanic deposits of the AD 79 eruption. The characteristics of the pre-AD 79 stratum allow the reconstruction of the paleo-environmental conditions. The pre-AD 79 stratum was characterized on the basis of lithofacies distinguishing five different classes: (i) (ii) (iii) (iv) (v) terrestrial deposits, fluvial deposits, palustrine deposits, littoral deposits, and marine deposits. S. Vogel, M. Märker / Geomorphology 115 (2010) 67–77 In case the pre-AD 79 stratum could not be clearly related to one of these five classes, mixed classes were proposed to comprise intermediate forms of deposits. Instead of using simple interpolation methods as in former studies, we reconstruct the pre-AD 79 topography with a sophisticated geostatistical methodology based on a high-resolution present-day digital elevation model (DEM). This methodology considers the statement by Stefani and Di Maio (2003) that the AD 79 eruption caused a coating of the ancient topography of the Sarno River plain which left ancient physiographic elements still recognizable in the present-day topography. Consequently the present-day topography provides valuable hints for reconstructing the ancient conditions. The following hypotheses can be outlined: (i) geomorphic transport processes are related to topographic terrain characteristics (slope, curvature, etc.), (ii) the pre-AD 79 topography controlled volcanic deposition, erosion and transport processes, (iii) For pre-AD 79 and today, similar relief-forming processes have been active, and (iv) tectonic activities and resulting landforms before AD 79 are taken into account in this modeling approach. Recent tectonic activities are considered via pre-AD 79 stratigraphic information (e.g. elevation of the ancient coastline). Since the recent topography will be utilized to deduce the pre-AD 79 topography, a hydrologically correct present-day DEM of the Sarno River plain was generated using the interpolation method ‘Topo to Raster’ by Hutchinson (1988, 1989) implemented in ArcGIS 9.2. The DEM is based on digitized topographic points and contour lines in vector format of the SIT 1:5000 (Sistema Informativo Territoriale) (Provincia di Napoli, by courtesy of S.I.A.V., Soprintendenza Archeologica di Pompei). The resulting present-day DEM was postprocessed to eliminate all evident man-made structures such as dams, motorways, quarries and embankments of the Sarno River which was regulated in the 1850s (Stefani and Di Maio, 2003). After the sink drainage route detection and sink removal of the present-day DEM, 15 different primary and secondary topographic indices were delineated using the terrain analysis module of SAGA GIS (Table 1). Based on the hypotheses stated above, a classification and regression tree approach was used to reconstruct the pre-AD 79 Table 1 Predictor variables used for the modeling process in TreeNet (Salford Systems). Predictor variables Method/reference Elevation Altitude above channel network Aspect Catchment area Channel network Channel network base level Convergence index Curvature Curvature classification Plan curvature Profile curvature LS-factor Slope Stream power Watershed subbasins Wetness index Geology Topo to Raster (ArcGIS 9.2)/SIT 1:5000 SAGA terrain analysis module (Olaya and Conrad, 2008) SAGA terrain analysis module (Zevenbergen and Thorne, 1987) SAGA terrain analysis module (Olaya and Conrad, 2008) SAGA terrain analysis module (Olaya and Conrad, 2008) SAGA terrain analysis module (Olaya and Conrad, 2008) SAGA terrain analysis module (Köthe and Lehmeier, 1993) SAGA terrain analysis module (Zevenbergen and Thorne, 1987) SAGA terrain analysis module (Dikau, 1988) SAGA terrain analysis module (Zevenbergen and Thorne, 1987) SAGA terrain analysis module (Zevenbergen and Thorne, 1987) SAGA terrain analysis module (Olaya and Conrad, 2008) SAGA terrain analysis module (Zevenbergen and Thorne, 1987) SAGA terrain analysis module (Olaya and Conrad, 2008) SAGA terrain analysis module (Olaya and Conrad, 2008) SAGA terrain analysis module (Olaya and Conrad, 2008) Digitized geological map 1:10,000 (Autorità di Bacino del Sarno, 2003) Land use CORINE land cover 2000 (Autorità di Bacino del Sarno) Mean annual ground Interpolated (kriging) water well data (Autorità di Bacino del Sarno) water table 71 topography. Classification and Regression Trees (CARTs) is an algorithm used for exploratory data analysis and predictive modeling to discover features and understand structural patterns in large databases by describing the correlation between predictor variables and a response variable (Breiman et al., 1984; Myles et al., 2004). This algorithm can handle nominally scaled (categorical) data as well as continuously scaled (metric) data by applying either classification or regression trees. CARTs are based on binary recursive partitioning, i.e. a decision tree is generated by partition of a dataset into two subsets (binary) where each of the subsets is the basis for a further partition (recursive). This process is carried out until a partition of a subset is no longer reasonable, i.e. the elements of a subset are homogenous with respect to the response variable or if the number of elements of a subset has become too small. Eventually CARTs generate a model taking into account the combination of predictor variables with the highest correlation to explain the response variable (Dannegger, 1997; Schillinger, 2002). The software that was applied in the present study using the CARTs algorithm is TreeNet by Salford Systems. We utilized 16 present-day topographic indices and other physiographic data as predictor variables (Table 1) and the depth to the pre-AD 79 surface as response variable to predict the pre-AD 79 topography. TreeNet is based on Friedman's stochastic gradient boosting (Friedman, 1999). Gradient boosting constructs additive regression models by sequentially fitting a simple parameterized function to current ‘pseudo’ residuals by least squares at each iteration. The pseudo residuals are the gradient of the loss function being minimized, with respect to the model values at each training data point, evaluated at the current step (Friedman, 1999). Practically, the method derives several hundreds to thousands of small trees each typically containing six nodes. Each tree is devoted to contributing a small portion of the overall model whereas the final model prediction is constructed by adding up each of the individual tree contributions. This methodology has a variety of advantages: (i) it is not sensitive to data errors in the input variables, (ii) it is resistant to overtraining, and (iii) it is very fast even with large sets of trees. For the modeling process we used a TreeNet regression model with the Huber-M loss function. The maximum number of trees to use is 3000. Internal model validation is performed by selecting a sample of cases at random from the model data. Here we selected a fraction of 0.2 of the entire dataset (N = 1811) to train the regression tree at each iteration. The model is subsequently used to regionalize the depth to the pre-AD 79 surface for the entire Sarno River plain. To get the absolute elevation above sea level of the pre-AD 79 surface, the predicted depth is thereafter subtracted from the present-day DEM. For additional external model validation we compare the resulting pre-AD 79 DEM with the stratigraphical point type information taken from the drilling data, which is not included in the model. To specify the paleogeomorphological and paleo-environmental characteristics of the study area and its dynamics, we perform a terrain analysis with SAGA GIS on the resulting pre-AD 79 DEM. The approximate location of the ancient coastline, the paleo-Sarno and its flood plain were already predefined by the littoral and the fluvial/palustrine deposits of the Roman stratum, respectively. We expect a more precise approximation of the ancient coastline and the paleo-Sarno from the adaptation of the existing drilling data to the modeled pre-AD 79 topography by deducing contour lines and depth contours from the pre-AD 79 DEM. 4. Results and discussion Table 2 shows the predictor variables used for the generation of the model arranged according to their particular importance to predict the pre-AD 79 surface. For the final model not all of the 72 S. Vogel, M. Märker / Geomorphology 115 (2010) 67–77 Table 2 Ranking of importance of predictor variables used for the model generation. Variable Importance [%] Elevation Watershed subbasins Aspect Channel base level Stream power Depth groundwater table Slope Altitude above channel network Elevation groundwater table Curvature classification Wetness index Curvature LS-factor Catchment area Convergence index Profile curvature Plan curvature Channel network 100.0 98.1 78.3 66.9 66.5 58.4 57.3 56.0 55.8 49.7 48.4 47.5 46.3 45.0 42.5 41.2 40.7 22.2 |||||||||||||||||||||||||||||||||||||||||| ||||||||||||||||||||||||||||||||||||||||| ||||||||||||||||||||||||||||||||| |||||||||||||||||||||||||||| |||||||||||||||||||||||||||| |||||||||||||||||||||||| |||||||||||||||||||||||| ||||||||||||||||||||||| ||||||||||||||||||||||| |||||||||||||||||||| |||||||||||||||||||| ||||||||||||||||||| ||||||||||||||||||| |||||||||||||||||| ||||||||||||||||| ||||||||||||||||| |||||||||||||||| ||||||||| available predictor variables were entered into TreeNet. The categorical variables geology and land use turned out to be of rather minor suitability for the modeling process since preliminary results showed that their areal distribution had a strong artificial impact on the regionalized pre-AD 79 surface. Fig. 4A illustrates the model performance with an attained minimum mean absolute error of the training dataset of 0.92 m and of the test dataset of 1.64 m. The smallest test mean absolute deviation is based on 2115 grown trees. The overall performance of the regression model on the given dataset is reported in Fig. 4B. The gain chart illustrates the percentage of population (x-axis) in relation to the percentage of the target variable (y-axis). The higher the percentage of the target variable with respective low population percentages, the better the model. In this case the model performance shows a fairly good correlation. Fig. 5 illustrates the modeled depth to the pre-AD 79 surface of the Sarno River plain which shows a distinct spatial distribution of the volcanic deposits since the AD 79 eruption. The modeled depth to the pre-AD 79 surface ranges from >0 to 15 m whereas the average depth is 5.7 m. In this study the Sarno River plain was only reconstructed northwards up to the modern town of S. Giuseppe Vesuviano (Fig. 3). From that location the drilling cores no longer contained AD 79 volcanic material which means that the pre-AD 79 surface could not be clearly identified anymore. The importance of the predictor variables illustrates that the spatial distribution of volcanic deposits since and including the AD 79 eruption is most notably controlled by the absolute elevation being the result of all relief-forming geomorphic processes. Of secondary importance are the aspect and hydrological variables such as watershed subbasins, channel base level and stream power. The aspect represents the direction in which the volcanic material was spread over the territory during the eruption. Originating from the Somma–Vesuvius volcanic complex, the pumice lapilli fallout was dispersed concentrically towards the southeast which fits well with the dispersion of the volcanic material during the AD 79 eruption as modeled by Sigurdsson and Carey (2002) and Pfeiffer et al. (2005) (Fig. 1). Consequently, on the southeastern flanks of Somma– Vesuvius, near the source of the eruption the volcanic deposits are thicker. The hydrological parameters represent the redistribution of the volcanic deposits after the eruption by erosion, transport and accumulation processes. Thus, greater thicknesses of the volcanic deposits along the Tyrrhenian coast most likely indicate re-deposited material that was mobilized by the Sarno River and its tributaries and transported towards the Tyrrhenian Sea. This corresponds with thinner deposits in the central part of the Sarno River plain. It is striking that along the southeastern foot slopes of Somma– Vesuvius the volcanic deposits are much thicker than in the river plain. Three main reasons may be responsible for the spatial distribution of the deposits: (i) As mentioned above the deposition of volcanic material is generally thicker close to the vent of Somma–Vesuvius being the source of the eruption. (ii) Apart from the white and grey pumice lapilli fallout the slopes of Somma–Vesuvius were also subject to pyroclastic surge and flow deposits of AD 79. They are concentrated at the foot slopes of Somma–Vesuvius since they did not reach far into the plain (Di Vito et al., 1998; Di Maio and Pagano, 2003). (iii) On the slopes of Somma–Vesuvius the volcanic deposits of AD 79 are overlain by pyroclastic material and lava flows from more recent eruptions that were less powerful to reach more distant areas. Thickest deposits most notably could be related to the eruptions of 1631, 1742–1761 and 1944 (see geological map 1:10,000, Autorità di Bacino del Sarno, 2003). An accumulation of volcanic deposits at the foot of the Lattari and Sarno Mountains may result from slope deposits being mobilized by mountain torrents or lahars after intense rainfall events. Debris flows or hyperconcentrated flows are known phenomena in the mountain areas adjacent to the Sarno River plain. The mountains are mantled by irregular layers of pyroclastic materials such as tephra and pumice fallout deposits which are intercalated with buried soils. Especially due to the presence of allophane, volcanic soils generally have a high water retention and low bulk density, which means that the weight of retained water can exceed the dry weight of the soil (Terribile et al., 2007). Due to this irregular cover of different strata which is characterized by a strong vertical variability of physical–mechanical properties, intense rainfall events can trigger slope instabilities, mass movements and thick submontane accumulations. These processes are still active today as the catastrophic landslide events of Sarno in 1998 have shown (Rao, 1995; Terribile et al., 2000; De Vita and Piscopo, 2002; Crosta and Dal Negro, 2003; Fiorillo and Wilson, 2004; Zanchetta et al., 2004; Terribile et al., 2007). Slightly thicker deposits in the southwestern area of Pompeii running NW–SE fit well with the ancient dune ridge of Bottaro– Fig. 4. Mean absolute error [m] of the depth to the pre-AD 79 surface versus number of trees of the training and test dataset (A) and gain chart of the training dataset (B) (TreeNet). The minimum mean absolute error of the training dataset of 0.92 m and of the test dataset of 1.64 m was attained at 2115 grown trees. S. Vogel, M. Märker / Geomorphology 115 (2010) 67–77 73 Fig. 5. Modeled depth to the pre-AD 79 surface of the Sarno River plain. Pioppaino. This may result from the fact that the dune ridge itself shows a slightly higher elevation than the surrounding terrain. Thus, the effects of fluvial and coastal erosion processes are limited. Moreover, the surface of the dune ridge is relatively flat, for instance in comparison to the Pompeiian hill, with low slope gradients and related erosion processes. Rather thin deposits on the Pompeiian hill on the other hand indicate that after deposition AD 79 volcanic material was eroded from the ridge and accumulated in the adjacent plain due to higher slope gradients. Fig. 6A and B compares the present-day topography with the predicted pre-AD 79 topography. Even though main physiographic elements appear on both DEMs, several differences can be identified. Most evident is that the pre-AD 79 surface is approximately 5.7 m deeper than the modern surface. Consequently the entire coastal area is lying below the recent sea level. Comparing the present-day Sarno River (Fig. 6A) with the modeled fluvial network before AD 79 (Fig. 6B) it is striking that the main elements seem to have preserved over time and are still visible in the present-day river network, i.e. the streams from the slopes of Somma– Vesuvius, the stream from Nocera and the intersections near the Sarno Mountains. However, in comparison to the canalized present-day Sarno River, the paleo-Sarno River, which follows the depth contours of the pre-AD 79 DEM, shows meander type fluvial patterns. In the southern section of the Tyrrhenian coast at the base of the Lattari Mountains a big fan delta can be clearly identified on both the present-day DEM and the pre-AD 79 DEM (Fig. 6). In the present-day geomorphological map of Cinque (1991) it is called ‘conoide Muscariello’. It is orientated towards the main discharge of this part of the Lattari Mountains, the ‘Fosso di Gragnano’. The fan delta on the modeled pre-AD 79 DEM indicates that this landform already existed in Roman times. This corresponds with Cinque (1991) and Cinque et al. (1987, 1997) who identified fan deltas of a first and second generation which they dated to the Middle and Late Pleistocene, respectively. Di Maio and Pagano (2003) also detected strong variations in the thickness of the AD 79 eruptive material between the slopes and the adjacent Sarno River plain which they related to the formation of fan deltas due to heavy rainfall after the AD 79 eruption. However, in their reconstruction of the paleo-coastline near the ancient town of Stabia, they state that the fan delta did not exist at the moment of the AD 79 eruption. Consequently they delineated the coastline between 100 and 200 m parallel to the terrace of San Marco (Fig. 2). The external validation of the pre-AD 79 DEM was performed using the stratigraphical information from the core drillings (see Materials and Methods). Furthermore by comparing the categorized character of the pre-AD 79 stratum with the pre-AD 79 DEM we reconstructed some paleo-environmental features before AD 79 (Fig. 7A and B). The results show that the littoral deposits before AD 79 are distributed nearly parallel to the present-day coastline at a horizontal distance of 1000 to 1500 m (Fig. 7A). In the north of the Tyrrhenian coast, most of the littoral deposits are located between the −3 and −2 m contour lines, and near the Sarno River and in the south of the bay, between the − 1 and 0 m contour lines. This matches with findings of Sigurdsson et al. (1985) and Livadie et al. (1990) who state that after AD 79 a northwards dipping subsidence of the Roman littoral deposits to approximately −4 m occurred. Near the mouth of the Sarno River the same deposits could be found at − 1 m. Pescatore et al. (2001) state that a more rapid subsidence appears near Somma– Vesuvius in comparison with more distant areas. This subsidence is related to the collapse of the graben structure of the Campanian plain of approximately 2 mm year− 1 during the entire Quaternary. In the Sarno River plain subsidence is compensated by the constant accumulation of volcanic deposits due to the activity of Somma– Vesuvius which finally resulted in a westward propagation of the coastline (Cinque et al., 1987; Livadie et al., 1990; Cinque, 1991). Converging contour lines at the foot of Somma–Vesuvius, the Lattari Mountain and the Pompeiian hill that are orientated towards 74 S. Vogel, M. Märker / Geomorphology 115 (2010) 67–77 Fig. 6. Present-day DEM (A) and modeled pre-AD 79 DEM (B) of the Sarno River plain with 5-m contour lines and fluvial network. the sea reveal escarpments that may have derived from abrasion action of the sea related to previous coastlines (Fig. 8). The protohistorical dune ridges of Messigno and Bottaro/Pioppaino could be delineated by confronting their location from the geological map (Autorità di Bacino del Sarno, 2003) with both the modeled preAD 79 topography and the distribution of marine/littoral deposits from the drilling data that were found inland from the ancient shoreline (Figs. 7 and 8). The distribution of fluvial/palustrine deposits taken from the drilling data correlates well with the modeled paleo-river network. Especially around the mouth of the paleo-Sarno River and near Longola–Poggiomarino, these deposits are characterized by the 1 to 2 m isolines of the ‘vertical distance to channel network’ index from the SAGA terrain analysis module. This area contains 71% of all fluvial/palustrine deposits found in the drilling data. Consequently, this area can be identified as the flood area in which the paleo-Sarno River most likely had its ancient riverbed. Near Pompeii the paleoSarno River bends southwards, approximately 1 km away from its present-day river bed to flow around the more elevated protohistorical dune ridges of Messigno and Bottaro/Pioppaino. This agrees S. Vogel, M. Märker / Geomorphology 115 (2010) 67–77 75 Fig. 7. Pre-AD 79 DEM with the drilling data on the pre-AD 79 stratum (A) and inferred reconstruction of pre-AD 79 environmental features (B). well with the paleo-Sarno River of Stefani and Di Maio (2003) and the geological map (Autorità di Bacino del Sarno, 2003). Due to the plain relief, this area will most likely be characterized by meanders, backwaters and an extensive flood area. The approximate mouth of the paleo-Sarno can be localized around 1500 m east of its presentday mouth which coincides with a noticeable inland propagation of the − 3 and − 2 m contour lines. This location corresponds well with the findings of Barra et al. (1989) and Livadie et al. (1990) stated above. Based on deeper littoral deposits in the northern coastal area, the pre-AD 79 DEM revealed an elevation discontinuity southwest of the excavation of Pompeii (Figs. 5 and 8). The discontinuity is more than 3 km long and oriented N–S. It includes several elevation levels from 1 to − 5 m a.s.l. and has a vertical offset from east to west of − 0.5 to 76 S. Vogel, M. Märker / Geomorphology 115 (2010) 67–77 Fig. 8. Pre-AD 79 environmental reconstruction of the Tyrrhenian coast of the Sarno River plain. −1.5 m. Thus, relative to the western part, the eastern part of this discontinuity is displaced towards the south indicating a horizontal transversal fault. Neotectonic deformations of the Apennines are predominantly oriented SW–NE or NW–SE. However, reactivated faults of an E–W or N–S orientation can sometimes be found (Nicotera and Civita, 1969a,b; Ortolani and Aprile, 1978; Aprile and Ortolani, 1979; Cinque et al., 1987). This discontinuity seems to be overprinted by recent deposition as it cannot be identified on the present-day DEM. This implies that it was formed before AD 79 possibly associated with the seismic activity preceding the AD 79 eruption, such as the big earthquakes of AD 62 and AD 64. Marturano (2008) and Marturano et al. (2009) proposed a ground uplift of Pompeii of around 2 m prior to the AD 79 eruption which fits well with the modeled discontinuity. The crustal deformations are presumably caused by the inflation of magma bodies underneath the Somma–Vesuvius volcanic complex. drillings. The pre-AD 79 DEM will be utilized to determine potential locations of Pompeii's marine or fluvial harbour, whose exact location is still debated (e.g. Stefani and Di Maio, 2003). Moreover, the combination of the pre-AD 79 DEM with archaeological findings enhances the paleo-environmental reconstruction of the Sarno River plain. In the next project phase the pre-AD 79 DEM will be used to determine the paleo-topographic setting of the known ‘villae rusticae’ of the Sarno River plain (Casale and Bianco, 1979; Kockel, 1985, 1986). In addition to paleo-pedological analyses, we will use the pre-AD 79 topography to reconstruct soil and land use characteristics. These data will help us to simulate the ancient cultural landscape of the Sarno River plain. In summary, the modeled pre-AD 79 DEM has a great potential for future geoarchaeological investigations. Acknowledgments 5. Conclusions The pre-AD 79 topography and paleo-environmental features of the entire Sarno River plain were reconstructed for the first time using sophisticated geostatistical methods. The estimated pre-AD 79 topography and the derived paleo-river network fit well with the stratigraphical data from the drilling documentation. In addition to the internal validation, the semantic validation also yielded good results. Nevertheless this reconstruction is only a model of the pre-AD 79 conditions based on the hypothesis stated above. Consequently the modeled paleo-river network is only approximate and cannot be considered as the exact paleo-course of the Sarno River before AD 79, since it is derived from the regionalized pre-AD 79 DEM. However, the deduced contour lines enclosing the fluvial/palustrine and littoral deposits from the drillings data more reliably indicate the approximate location of the paleo-Sarno and the paleo-coastline, respectively. In future studies the pre-AD 79 DEM will be used to identify zones of particular interest for carrying out additional stratigraphical This study is part of the geoarchaeological research project entitled ‘Reconstruction of the ancient cultural landscape of the Sarno River plain’ undertaken by the German Archaeological Institute in cooperation with the Heidelberg Academy of Sciences and Humanities. It was initialized by Florian Seiler and partly funded by the German Archaeological Institute, Berlin Head Office (Cluster 3) and the Deutsche Forschungsgemeinschaft (German Research Foundation). We would like to thank our local project partners and all their collaborators for their cooperation, particularly the Autorità di Bacino del Sarno, the Soprintendenza Speciale per i Beni Archaeologici di Napoli e Pompei, the Soprintendenza per i Beni Archaeologici di Salerno e Avellino. We also thank Giovanni Di Maio, Giovanni Patricelli and Gaetana Saccone for various supports and Andrew Kandel for reviewing the English manuscript. Finally, we would like to thank Carmen Rosskopf, Jean-Paul Bravard and Takashi Oguchi for reviewing and substantially improving the paper with constructive comments and suggestions. S. Vogel, M. Märker / Geomorphology 115 (2010) 67–77 References Andronico, D., Calderoni, G., Cioni, R., Sbrana, A., Sulpizio, R., Santacroce, R., 1995. Geological map of Somma–Vesuvius volcano. Period. Mineral. 64, 77–78. Aprile, F., Ortolani, F., 1979. Sulla struttura profonda della Piana Campana. Boll. Soc. Nat. Napoli 88. Autorità di Bacino del Sarno, 2003. Cartografia Geologica, scala 1:10 000. Progetto CARG, Regione Campania. Barra, D., Bonaduce, G., Brancaccio, L., Cinque, A., Ortolani, F., Pagliuca, S., Russo, F., 1989. Evoluzione geologica olocenica della piana costiera del fiume Sarno (Campania). Mem. Soc. Geol. Ital. 42, 255–267. Breiman, L., Friedman, J., Olshen, R., Stone, C., 1984. Classification and Regression Trees. Chapman and Hall, New York. Carey, S., Sigurdsson, H., 1987. Temporal variations in column height and magma discharge rate during the 79 A.D. eruption of Vesuvius. Geol. Soc. Amer. Bull. 99, 303–314. Casale, A., Bianco, A., 1979. Primo contributo alla topografia del suburbio Pompeiano. Antiqua 4, 27–56. Cella, F., Fedi, M., Florio, G., Grimaldi, M., Rapolla, A., 2007. Shallow structure of the Somma–Vesuvius volcano from 3D inversion of gravity data. J. Volcanol. Geotherm. Res. 161, 303–317. Cinque, A., 1991. La trasgressione versiliana nella piana del Sarno, Campania. Geogr. Fis. Din. Quat. 14, 63–71. Cinque, A., Russo, F., 1986. La linea di costa del 79 A.D. fra Oplonti e Stabiae nel quadro dell'evoluzione olocenica della Piana del Sarno (Campania). Boll. Soc. Geol. Ital. 105, 111–121. Cinque, A., Alinaghi, H.H., Laureti, L., Russo, F., 1987. Osservazioni preliminari sull'evoluzione geomorphologica della piana del Sarno Campania, Appennino Meridionale. Geogr. Fis. Din. Quat. 10, 161–174. Cinque, A., Aucelli, P.P.C., Brancaccio, L., Mele, R., Milia, A., Robustelli, G., Romano, P., Russo, F., Russo, M., Santangelo, N., Sgambati, D., 1997. Volcanism, tectonics and recent geomorphological change in the bay of Napoli. I.A.G. IV Int. Conf. on Geomorphology: Geogr. Fis. Dinam. Quat., vol. III (T2), pp. 123–141. Cioni, R., Marianelli, P., Sbrana, A., 1992. Dynamics of the A.D. 79 eruption: stratigraphic, sedimentological and geochemical data on the successions from the Somma– Vesuvius southern and eastern sectors. Acta Vulcanol. 2, 109–123. Civetta, L., Galati, R., Santacroce, R., 1991. Magma mixing and convective compositional layering within the Vesuvius magma chamber. Bull. Volcanol. 53, 287–300. Civetta, L., Di Vito, M.A., De Lucia, M., Isaia, R., 1998. Storia eruttiva del Somma–Vesuvio. In: Guzzo, P., Peroni, R. (Eds.), Archeol. Vulcanol. Campania, pp. 9–16. Naples. Crosta, G.B., Dal Negro, P., 2003. Observations and modelling of soil slip-debris flow initiation processes in pyroclastic deposits: the Sarno 1998 event. Nat. Hazards Earth Syst. Sci. 3, 53–69. Dannegger, F., 1997. Tree stability diagnostics and some remedies against instability. Institute of Statistics, Ludwig-Maximilians-University of Munich, Sonderforschungsbereich. 386. Delibrias, G., Di Paola, G.M., Rosi, M., Santacroce, R., 1979. La storia eruttiva del complesso vulcanico Somma–Vesuvio ricostruita dalle successioni piroclastiche del Monte Somma. Rend. Soc. Ital. Mineral. Petrol. 35, 411–438. De Vita, P., Piscopo, V., 2002. Influences of hydrological and hydrogeological conditions on debris flows in peri-vesuvian hillslopes. Nat. Hazards Earth Syst. Sci. 2, 27–35. Dikau, R., 1988. Entwurf einer geomorphographisch-analytischen Systematik von Reliefeinheiten. Heidelb. Geogr. Baust. 5, 1–45. Di Maio, G., Pagano, M., 2003. Considerazioni sulla linea di costa e sulle modalità di seppellimento dell'antica Stabia a seguito dell'eruzione vesuviana del 79 d.C. Riv. Studi Pompei. XIV, 197–245. Di Vito, M.A., Sulpizio, R., Zanchetta, G., Calderoni, G., 1998. The geology of the South Western Slopes of Somma–Vesuvius, Italy, as inferred by borehole stratigraphies and cores. Acta Vulcanol. 10, 383–393. Fiorillo, F., Wilson, R.C., 2004. Rainfall induced debris flows in pyroclastic deposits, Campania (southern Italy). Eng. Geol. 75, 263–289. Foss, J.E., Timpson, M.E., Ammons, J.T., Lee, S.Y., 2002. Paleosols of the Pompeii area. In: Jashemski, W.F. (Ed.), The Natural History of Pompeii. Cambridge University Press, Cambridge, pp. 65–79. Friedman, J.H., 1999. Stochastic gradient boosting. Technical Report. Department of Statistics, Stanford University, USA. http://www.salford-systems.com/treenet.html. Furnari, E., 1994. Nuovi contributi all'identificazione del littorale antico di Pompei. AA. VV., Neapolis Temi Progettuali, Roma, vol. II, pp. 219–291. Hutchinson, M.F., 1988. Calculation of hydrologically sound digital elevation models. Third International Symposium on Spatial Data Handling, Sydney. International Geographical Union, Columbus, Ohio. Hutchinson, M.F., 1989. A new procedure for gridding elevation and stream line data with automatic removal of spurious pits. J. Hydrol. 106, 211–232. Iacono, L., 1913. Note di archeologia marittima. In Neapolis I, f. III–IV, pp. 353–371. Kockel, V., 1985. Archäologische Funde und Forschungen in den Vesuvstädten I. Archäol. Anz. 495–571. Kockel, V., 1986. Archäologische Funde und Forschungen in den Vesuvstädten II. Archäol. Anz. 443–569. Köthe, R., Lehmeier, F., 1993. SARA — Ein Programmsystem zur Automatischen Reliefanalyse. Z. Angew. Geogr. 4, 11–21. 77 Lirer, L., Pescatore, T., Booth, B., Walker, G.P.L., 1973. Two Plinian pumice-fall deposits from Somma–Vesuvius, Italy. Geol. Soc. Amer. Bull. 84, 759–772. Livadie, C.A., Barra, D., Bonaduce, G., Brancaccio, L., Cinque, A., Ortolani, F., Pagliuca, S., Russo, F., 1990. Evoluzione geomorphologica, neotettonica e vulvanica della piana costiera del fiume Sarno (Campania) in relazione agli insediamenti anteriori all'eruzione del 79 d.C. Pact 25–13, 237–256. Luongo, G., Perrotta, A., Scarpati, C., 2003. Impact of the AD 79 explosive eruption on Pompeii, I. Relations amongst the depositional mechanisms of the pyroclastic products, the framework of the buildings and the associated destructive events. J. Volcanol. Geotherm. Res. 126, 201–223. Malandrino, C., 1988. Il “Pagus” Marittimo di Pompei. Note di topografia antica. In: D'Amelio (Ed.), Torre Annunziata, p. 40. Marturano, A., 2008. Sources of ground movements at Vesuvius before the AD 79 eruption: evidence from contemporary accounts and archaeological studies. J. Volcanol. Geotherm. Res 177, 959–970. Marturano, A., Aiello, G., Barra, D., Fedele, L., Grifa, C., Morra, V., Berg, R., Varone, A., 2009. Evidence for Holocenic uplift at Somma–Vesuvius. J. Volcanol. Geotherm. Res. doi:10.1016/j.jvolgeores.2009.05.020.. Myles, A., Feudale, R., Liu, Y., Woody, N., Brown, S., 2004. An introduction to decision tree modelling. J. Chemom. 18, 275–285. Nicotera, P., Civita, M., 1969a. Indagine idrogeologiche per la captazione delle sorgenti S. Marina di Lavorate Sarno). Mem. Note Ist. Geol. Appl. Univ. Napoli 11, 25–44. Nicotera, P., Civita, M., 1969b. Ricerche idrogeologiche per la realizzazione delle opere di presa delle sorgenti Mercato e Palazzo Sarno). Mem. Note Ist. Geol. Appl. Univ. Napoli 11, 1–63. Olaya, V., Conrad, O., 2008. Geomorphometry in SAGA. In: Hengl, T., Reuter, H.I. (Eds.), Geomorphometry: Concepts, Software, Applications. Elsevier, Amsterdam, pp. 293–308. Ortolani, F., Aprile, F., 1978. Nuovi dati sulla struttura profonda della Piana Campana a SE del Fiume Volturno. Boll. Soc. Geol. Ital. 97, 591–608. Pescatore, T., Senatore, M.R., Capretto, G., Lerro, G., Patricelli, G., 1999. Ricostruzione paleoambientale delle aree circostanti l'antica Città di Pompei Campania, Italia) al Tempo dell'eruzione del Vesuvio del 79 d.C. Boll. Soc. Geol. Ital. 118, 243–254. Pescatore, T., Senatore, M.R., Capretto, G., Lerro, G., 2001. Holocene coastal environments near Pompeii before the A.D. 79 eruption of Mount Vesuvius, Italy. Quat. Res. 55, 77–85. Pfeiffer, T., Costa, A., Macedonio, G., 2005. A model for the numerical simulation of tephra fall deposits. J. Volcanol. Geotherm. Res. 140, 273–294. Rao, S.M., 1995. Mechanistic approach to the shear strength behavior of allophanic soils. Eng. Geol. 40, 215–221. Rolandi, G., Maraffi, S., Petrosino, P., Lirer, L., 1993a. The Ottaviano eruption of Somma– Vesuvio 8000 y B.P. a magmatic alternating fall and flow-forming eruption. J. Volcanol. Geotherm. Res. 58, 43–65. Rolandi, G., Mastrolorenzo, G., Barrella, A.M., Borrelli, A., 1993b. The Avellino plinian eruption of Somma–Vesuvio 3760 y B.P— the progressive evolution from magmatic to hydromagmatic style. J. Volcanol. Geotherm. Res. 58, 67–88. Ruggiero, M., 1879. Del sito di Pompei e dell'antico lido del mare. Pompei e la regione sotterata dal Vesuvio nell'anno, vol. 79. Mem. e Note pubb. dall'ufficio tecn. degli scavi delle prov. meridionali, Napoli, pp. 5–14. Santacroce, R., 1987. Somma–Vesuvius. C.N.R., Quaderni de ‘La Ricerca Scientifica’ 114, Rome, 251 pp. Schillinger, M.P., 2002. Flächenhafte Schätzung mit Classification and Regression Trees und robuste Gütebestimmung ökologischer Parameter in einem kleinen Einzugsgebiet. Doctoral Thesis, Faculty of Biology, Chemistry and Geo Sciences, University of Bayreuth. Sigurdsson, H., 2002. Mount Vesuvius before the Disaster. In: Jashemski, W.F. (Ed.), The Natural History of Pompeii. Cambridge University Press, Cambridge, pp. 29–36. Sigurdsson, H., Carey, S., 2002. The eruption of Vesuvius in A.D. 79. In: Jashemski, W.F. (Ed.), The Natural History of Pompeii. Cambridge University Press, Campbridge, pp. 37–64. Sigurdsson, H., Carey, S., Cornell, W., Pescatore, T., 1985. The eruption of Vesuvius in A.D. 79. Natl. Geogr. Res. 1, 332–387. Stefani, G., Di Maio, G., 2003. Considerazioni sulla linea di costa del 79 d.C. e sul porto dell'antica Pompei. Riv. Studi Pompei. XIV, 142–195. Terribile, F., Basile, A., De Mascellis, R., di Gennaro, A., Mele, G., Vingiani, S., 2000. I suoli delle aree di crisi di Quindici e Sarno: proprietà e comportamenti in relazione ai fenomeni franosi del 1998. Quad. Geol. Appl. 7 (1), 59–95. Terribile, F., Basile, A., De Mascellis, R., Iamarino, M., Magliulo, P., Pepe, S., Vingiani, S., 2007. Landslide processes and andosols: the case study of the Campania region, Italy. In: Arnalf, Ó., Bartoli, F., Buurman, P., Óskarson, H., Stoops, G., García-Rodeja, E. (Eds.), Soils of Volcanic Regions in Europe. Springer-Verlag, Berlin, Heidelberg, pp. 545–563. Zanchetta, G., Sulpizio, R., Pareschi, M.T., Leoni, F.M., Santacroce, R., 2004. Characteristics of May 5–6, 1998 volcaniclastic debris flows in the Sarno area Campania, southern Italy. relationships to structural damage and hazard zonation. J. Volcanol. Geotherm. Res. 133, 377–393. Zevenbergen, L.W., Thorne, C.R., 1987. Quantitative analysis of land surface topography. Earth Surf. Process. Landf. 12, 47–56.