SACRED MOUNTAINS AND GLACIAL ARCHEOLOGY IN THE ANDES
ADINA E. RACOVITEANU B.A., Middlebury College, 2000
A thesis submitted to the Faculty of the Graduate School of the University of Colorado in partial fulfillment of the requirement for Masters Degree Department of Geography 2004 ADINA E. RACOVITEANU (M.A., GEOGRAPHY) SACRED MOUNTAINS AND GLACIAL ARCHEOLOGY IN THE ANDES
Thesis directed by Mark W. Williams, Associate Professor, Department of Geography, University of Colorado, Boulder
Archeological sites from the Inca period (15th-16th century) have been found on Andean Mountains at altitudes up to 6,700 m, including frozen mummies of Inca children. High-altitude archeologists believe that these sites were constructed for ritual purposes, to appease the Andean Mountain Gods, held responsible in local beliefs for droughts, volcanic eruptions and glacier-related hazards. Currently, these archeological sites are threatened by looting, tourism and changes in glacier cover that may expose the artifacts to decomposition. There is a need to develop new tools to identify archeologically sensitive areas to guide archeological fieldwork at high altitudes.
This thesis proposes to identify recurring spatial patterns of archeological sites using the existing survey data, multivariate statistical analysis and the latest geo-visualization tools. Here I argue that the location of the ritual sites is related to environmental and social factors such as topography, climate and accessibility. First, I review existing hypotheses about archeological site location and I construct a spatial inventory of the archeological sites found to the present day. I then present the construction of the digital elevation model (DEM), and the development of the key environmental and social factors associated with high altitude archeological sites. Possible Inca access routes are reconstructed using least cost path algorithms and topographic data from the DEM, and validated with climbing routes surveyed during fieldwork. I then develop an archeological predictive model using a combination of univariate statistics, logistic regression, and the spatial analysis capabilities of GIS.
The model is validated on Nevados Coropuna (6,426 m) and Pichu Pichu (5,650 m), two sacred mountains in the Peruvian Andes. I surveyed these mountains in two field expeditions in 2003, and located Inca ruins, tombs and a mummy. Using these data, the model predictions are compared against with actual locations of the sites recorded in the field, with the goal of refining the model. "*-.6 +$#&,$-32
The research was supported by the NSF/IGERT- sponsored graduate training program entitled Carbon, Climate and Society Initiative (CCSI) at University of Colorado. I am grateful to my thesis committee members: Mark Williams, William Manley, James Dixon,
Jeremy Mennis and John M. Malville for providing advice. I thank the Global Land and Ice
Measurements from Space (GLIMS) team for facilitating free access to ASTER data through the NASA ESE Pathfinder project; the USGS EROS Data Center for providing SRTM and
ASTER elevation datasets; the Glaciology Unit at Intituto Nacional de Recursos Naturales
(INRENA) in Huaraz, Peru, for providing logistical assistance, and the UNAVCO facility in
Boulder for assisting with GPS corrections. I am grateful to all the research teams who offered their collaboration during fieldwork: l’Institut de Recherche pour le Développement,
France (GREAT ICE project), for supporting my participation in the ice-core drilling expedition on Coropuna in June 2003; Jean-Claude Thouret, Jean-Claude Thouret,
Laboratoire Magmas et Volcans, Université Blaise Pascal, France for collaboration; archeologists Mariusz Ziólkowski from University of Warsovia (Proyecto arqueológico
Condensuyo) and Jose Antonio Chávez Chávez from Universidad Católica Santa María de
Arequipa, for providing archeological guidance, and Arcadio Mamani, mountain guide,
Arequipa, for assisting with fieldwork.
iii ".-3$-32
CHAPTER ONE...... 1 INTRODUCTION...... 1 I.1 Sacred Mountains ...... 1 I.2 Mountain rituals at risk...... 2 I.3 Archeological predictive modeling overview...... 5 I.4 Research design and objectives ...... 9 I.5 Organization of the thesis ...... 11
CHAPTER TWO...... 12 HIGH-ALTITUDE ARCHEOLOGY: THE CONTEXT ...... 12 II.1 Introduction ...... 12 II.2 The cultural context...... 13 II.3 Existing hypotheses...... 22 II.4 Conclusions ...... 30
CHAPTER THREE ...... 31 EVALUATING DIGITAL ELEVATION MODELS FOR GLACIOLOGIC APPLICATIONS: AN EXAMPLE FROM NEVADO COROPUNA, PERUVIAN ANDES ...... 31 III.1 Introduction...... 31 III.2 Study area...... 34 III.3 MethodS ...... 35 III.4 Results and discussion...... 38 III.5 Conclusions and further applications ...... 55
CHAPTER FOUR ...... 57 CONSTRUCTION OF A SPATIAL DATABASE FOR HIGH-ALTITUDE ARCHEOLOGY IN THE ANDES...... 57 IV.1 Introduction...... 57 IV.2 Construction of the spatial database ...... 59 IV.3 Results and discussion ...... 67 IV.4 Conclusions...... 76
CHAPTER FIVE...... 78 ENVIRONMENTAL AND SOCIAL FACTORS ASSOCIATED WITH HIGH ALTITUDE ARCHEOLOGICAL SITES IN THE ANDES ...... 78 V.1 Introduction...... 78 V.2 Methods...... 80 V.3 Results and discussion...... 85 V.4 Conclusions ...... 99
ii CHAPTER SIX...... 101 RECONSTRUCTING INCA CLIMBING ROUTES USING LEAST COST PATH ANALYSIS ...... 101 VI.1 Introduction...... 101 VI.2 Methods ...... 105 VI.3 Results...... 112 VI.4 Discussion...... 123 VI.5 Conclusions and further applications...... 124
CHAPTER SEVEN ...... 125 PREDICTIVE MODELING FOR HIGH-ALTITUDE ARCHEOLOGICAL SITES IN THE ANDES ...... 125 VII.1 Introduction ...... 125 VII.2 Methods ...... 126 VII.3 Results ...... 128 VII.4 Conclusions ...... 135
References...... 136
Appendix 1 Human sacrifices discovered on Andean peaks to the present day...... 146
Appendix 2 Mountains surveyed prior to 1985 and reported by Beorchia Nigris (1985). 148
Appendix 3 Mountains surveyed in the last decade and reported by Ceruti (1999a; 1999b) and Reinhard and Ceruti (2000)...... 154
List of tables...... iv
List of figures...... vi
iii +(23.%3 !+$2
Table 3.1 Evaluation of different interpolation methods used to construct DEM’s from the topographic maps ...... 39 Table 3.2 Statistics summary of map differences for the glaciated areas vs. non-glaciated areas after trend removal...... 52 Table 4.1 Summary of the archeological survey data compiled from the available publications...... 61 Table 4.2 Metadata component storing information about projection systems, datums and accuracy of the data...... 63 Table 4.3 Attribute information recorded from archeological surveys...... 64 Table 4.4 Classification of archeological sites based on their function, the location on the mountain and the typical artifacts associated with these. Table translated and reproduced from Ceruti (1997)...... 65 Table 4.5 Distribution of surveyed archeological sites by country...... 68 Table 4.6 Frequency of site types based on their function. The site type is defined by a combination of site altitude, the type of artifacts and their location on the mountain. ....68 Table 4.7 Frequency of artifacts found at the surveyed archeological sites...... 69 Table 4.8 Correlations between archeological attributes and their location...... 70 Table 4.9 Similarity matrix for associations of artifacts, where each artifact is stored as binary. The matrix is derived using the Jaccard (“similarity ratio”) method. The highest match ratios are highlighted in shades of grey with the darkest shades for the highest scores...... 73 Table 5.1 Organization of the GIS data layers, the resulting variables and non-parametric tests for contrasts between sites, non-sites and site types...... 87 Table 5.2 General landscape and climatic characteristics of archeological site locations...... 91 Table 5.3 Percentage of archeological sites that occurs on glaciers, volcanoes and permafrost, where “0” indicates number of sites absent and “1” indicates number of sites present...92 Table 5.4 Climatic characteristics for archeological sites vs. non-sites ...... 92 Table 5.5 Vegetation zones and types occuring at archeological sites...... 93 Table 5.6 Topographic characteristics for archeological sites vs. non-sites...... 94 Table 5.7 Proximity factors for archeological sites vs. non-sites...... 98 Table 6.1 Terrain coefficients used to predict energy expenditure of load carrying in various types of terrain. Compiled from Soule and Goldman (1992) and Pandolf et al. (1977; 1976)...... 110 Table 6.2 Hiking speed for walking uphill, downhill and flat slopes for various categories of hikers. Reproduced from the Search and Rescue Society of British Columbia (SARBC 1992)...... 111 Table 6.3 Comparison of predicted and actual travel time for selected modeled routes...... 117
iv Table 7.1 Classification table for probabilities of archeological site occurrence, where “1” = sites and “0” = nonsites. The cut-off value is 0.5...... 130 Table 7.2 Classification table for probabilities of archeological site occurrence, where “1” = ceremonial sites, “2” = logistical sites and “3” = base camp sites...... 131
v +(23.%%(&41$2
Figure 1.1 Generalized flow chart illustrating the steps used in the model-building process. a) development of the archeological, environmental and social data layers; b) development and testing of the predictive model. Diagram adapted from Warren and Asch (2000). ..10 Figure 2.1 Map of the study area showing the Inca Empire. The box indicates the area containing high altitude archeological sites (Southern Peru to Chile)...... 14 Figure 2.2 Typical high-altitude archeological complex from Volcano Licancabur, Chile. Reproduced from Reinhard (1985a)...... 17 Figure 2.3 Ceremonial complex on Nevado Pichu Pichu: a) ceremonial platform (double Inca wall) indicated by the solid line. The dashed arrow indicates the lookout point; b) the base camp ruins with the summit in the background; c) a close-up view of the habitation constructions at the base...... 18 Figure 2.4 The ceremonial complex of Mauca Llacta of Pampacolca, at the base of Nevado Coropuna (6,426 m) in the Peruvian Andes...... 19 Figure 2.5 Paleoclimatic reconstructions in the Andean area. a) Paulsen’s (1976) series of rise and collapse of Andean Empires. b) Ice core record from Quelcaya Ice cap, Peru, reproduced from Thompson et al. (1984). Lower accumulation of ice and more negative δ18O values suggest decreased precipitation in the first half of the 15th century, overlapping with the rise of the Inca Empire...... 23 Figure 2.6 System of straight lines radiating from Cuzco, the Inca capital, to natural shrines (huacas) in the Sacred Valley of the Incas. Lines digitized from Farrington (1982) are shown on a shaded relief map of a digital elevation model...... 26 Figure 3.1 Location map of the study area. The ASTER Level 1A image from July 2001 is draped over a shaded relief map of the topographic DEM. Also shown are GPS transects obtained in the field...... 34 Figure 3.2 The effect of interpolation methods on representation of terrain topography at a subsection of the study area. a) original contour lines, with 25 m interval; b) shaded relief map of the DEM created with the IDW method; c) shaded relief map of the TIN data structure...... 40 Figure 3.3 Histograms of elevation values for the three DEM's analyzed. a) DEM created with TOPOGRID algorithm (TOPO DEM), b) SRTM DEM and c) ASTER DEM. Spiked on elevation histograms represent elevation values of the contour lines used in the interpolation (‘terracing’ effect)...... 42 Figure 3.4 Plots of height differences between the DEM’s from satellite data and GPS elevation along GPS transects. a) SRTM elevations minus GPS elevations; b) ASTER elevations minus GPS elevations...... 44 Figure 3.5 Color maps of height differences between the SRTM and topographic elevations draped over the topographic DEM. a) before trend removal, the SE-NW spatial trend of elevation differences is visible. b) after trend removal. Also shown is the glacier extent obtained by classification of the October 2000 L1B ASTER scene...... 47 Figure 3.6 Color maps of elevation differences between ASTER DEM and the TOPO DEM’s. a) 2001 ASTER DEM (before trend removal) minus 1955 TOPO DEM, with the NNW-
vi SSW spatial trend visible; b) 2001 ASTER DEM (after trend removal) minus the 1955 TOPO DEM...... 48 Figure 3.7 Frequency histograms of elevation differences between the DEM’s on non- glaciated vs. glaciated areas, after the trend removal. a) SRTM DEM minus TOPO DEM; b) ASTER DEM minus TOPO DEM...... 49 Figure 3.8 Correlation of vertical differences between the DEM’s with slope a) SRTM DEM minus TOPO DEM; b) ASTER DEM minus TOPO DEM. Largest vertical differences occur on steepest slopes...... 50 Figure 3.9 Radar charts of vertical differences between the DEM’s with respect to aspect. a) SRTM DEM minus TOPO DEM; b) ASTER DEM minus TOPO DEM...... 51 Figure 3.10 Correlation between vertical differences between the DEM’s and altitude on the glaciated area, after trend removal. a) SRTM DEM minus TOPO DEM; b) ASTER DEM minus TOPO DEM. Vertical differences increase with elevation on the glaciated areas.53 Figure 4.1 Distribution of surveyed archeological sites compiled from published surveys. The sites are georeferenced and mapped on a shaded relief map of the SRTM digital elevation model...... 59 Figure 4.2 An example of the archeological sketch used to document the location of the archeological ruins on Volcano Chilques in Chile. Reproduced from Beorchia Nigris (1985)...... 60 Figure 4.3 Frequency distribution of differences between SRTM- derived elevations and elevations reported by surveys...... 74 Figure 4.4 Horizontal displacement of archeological sites mapped using published coordinate information. The sites are mapped on a shaded relief of the SRTM digital elevation model...... 75 Figure 4.5 Comparison of vertical differences between SRTM DEM with survey elevations before and after the location corrections. The RMSEz represents the combined error between GPS measurements, topographic measurements and SRTM...... 76 Figure 5.1 Terrain roughness index illustrated for a subset of the study area (Nevado Coropuna, Peruvian Andes). Darker shades around ridges and summits represent higher variation of elevations (more rugged terrain)...... 82 Figure 5.2 Shelter measure derived from cylinder volumes, reproduced from Kvamme (1988b)...... 83 Figure 5.3 Primary and secondary GIS data layers...... 86 Figure 5.4 Distribution of archeological sites in the four quarters of the Inca Empire. Also shown is the Inca Roads system...... 89 Figure 5.5 Frequency histogram of site elevations with respect to relative elevation (percent elevation from base of the mountain in bins of 10 %)...... 90 Figure 5.6 Frequency distribution of archeological sites with respect to slope...... 94 Figure 5.7 Charts of frequency distributions of sites and non-sites with respect to aspect. a) archeological sites; b) random background sites. Frequencies are plotted in bins of 22.5 degrees...... 95
vii Figure 5.8 Frequency distribution of archeological sites with respect to terrain roughness index. Roughness index values are calculated based on variance of elevations around a site. High variances indicated rougher terrain...... 96 Figure 5.9 Frequency distribution of archeological sites with respect to shelter index. High shelter values suggest more exposure...... 97 Figure 5.10 Frequency histograms of proximity factors for archeological sites a) linear distance to Inca roads; b) linear distance to wood sources...... 98 Figure 6.1 Map of the study area showing the three mountains surveyed during field work in 2003. The dots represent GPS points taken along the climbing routes and around the base of the mountains. The GPS points and glacier boundaries are draped over a shaded relief map of the SRTM digital elevation model...... 106 Figure 6.2 Nevado Coropuna (6,426 m), Peruvian Andes. a) view of the NW and SW summits seen from the access road from Chuquibamba to Lake Pallacocha. b) offerings made by local people to worship Coropuna. Sketch reproduced from Guamán Poma de Ayala (1980 [1615])...... 107 Figure 6.3 Penitents forming on the glacier surface of Nevado Coropuna...... 107 Figure 6.4 The Inca camp of Ajocancha, with habitation complexes and water canal...... 113 Figure 6.5 Archeological sites along the climbing route on Nevado Pichu Pichu. Sketch reproduced from Linares Malaga (1966)...... 113 Figure 6.6 Speed of traveling expressed as a non-linear function of slope, based on the velocity equation developed by Gorenflo and Gale (1990)...... 114 Figure 6.7 Nonlinear relationship between hiking time and slope. Travel time is computed as distance / velocity (time necessary to cross 1 meter of horizontal distance at a given slope)...... 115 Figure 6.8 Friction surface for Nevado Coropuna in the Peruvian Andes draped over the hillshade of the SRTM DEM. White areas are cells with NODATA in the DEM. Cost is expressed as time it takes to cross a cell (90m by 90m) at a given slope...... 116 Figure 6.9 Least cost climbing routes on Nevado Coropuna, Peruvian Andes draped on a shaded relief map of the 30 m resolution DEM. Possible climbing routes from two starting points to the two main summits are created using the time-dependent model based on the Gorenflo and Gale’s (1990) equation. Also shown are the GPS points of the actual routes climbed during fieldwork in 2003...... 116 Figure 6.10 Nonlinear relationship between actual energy expenditure and slope. Energy is expressed in Watts expended to cross 1 meter of horizontal distance at a certain slope, assuming standard speeds of walking uphill (0.43 m/s) and downhill (0.86 m/s) for an expert hiker weighing 70 kg and carrying a 30 kg-load...... 118 Figure 6.11 The effect of terrain type on energy expenditure of load carrying while hiking. The graph is created using the equation provided by Pandolf et al (1977) and the terrain coefficients provided by Soule and Goldman (1972) and Soule et al (1978)...... 119 Figure 6.12 The effect of terrain type on climbing routes. The least east cost paths are derived from the 30-m resolution DEM. The yellow line is the path model with the time- dependent model and the red line using the energy-dependent model with parameters for 35-cm deep snow. The red line estimates better the present-day climbing route (not shown)...... 121
viii Figure 6.13 The effect of DEM resolution on least cost paths. The smoother solid line represents the path derived from the time-equivalent model from the 30 m DEM. The coarser, straighter dotted line represents the path derived from the 90 m resolution DEM...... 122 Figure 7.1 Probability image draped on a shaded relief map. The effect of topography (shelter and slope) is visible in this model. Areas with high shelter are assigned a low archeological potential...... 133 Figure 7.2 Probability of site occurrence. a) Ceremonial sites; b) intermediate sites; c) base camp sites...... 134
ix "' /3$1.-$
(-31.#4"3(.-
I.1 Sacred Mountains
Mountains throughout the world have been worshipped for their inspiring beauty, their sacred significance and their importance as water sources. Different cultures have regarded mountains as sacred centers (“axis mundi”), places of revelation, or places of the divine
(Eliade, 1963; 1987). The idea of the mountain as a spiritual center is embedded in traditional worship practices such as pilgrimages around the mountains (circumambulation), the construction of Buddhist hemispheric buildings (stupa) or the Hindu temples built on concentric circles (mandalas). Ritual practices in the Andes and the Himalayas are centered on the concept of water as purifying medium and source of fertility. For Hindus, Dudh Kunda
(Milk Lake) at the base of the sacred mountain Shorung Yul-lha (Numbur) in the Solu region of Nepal is a site of a yearly pilgrimage and purifying ritual bathing. Similarly, the sacred lakes Huaringas in Cordillera del Wamani (Northern Peru) are believed to have medicinal powers, and are used by the shamans for healing purposes. Glacier ice from the Sinaqara
Glacier in Peru is also believed to have medicinal powers (Gow, 1974). To this day, Andean people climb to the top of the mountains to perform worship rituals to the mountain Gods
(Reinhard, 1985a). Human sacrifices have even observed in 1942 and 1945 in Peru (Beorchia
Nigris, 1985a) and even as recently as 1986 in the Puno province of Peru (Tierney, 1989).
1 I.2 Mountain rituals at risk
These worship practices are affected today by changes in climate, glacier retreat and tourist development. Since the second half of the 19th century, there has been rapid melting of
Andean glaciers (Kaser, 1999b). For example, some glaciers of Cordillera Blanca of Peru have shown alarming rates of retreat compared to the Little Ice Age maximum extent (Kaser,
1999b). The increased melting observed in tropical glaciers is synchronized with similar trends detected during 1961 - 1997 in mid- and high-latitude glaciers around the world
(Dyurgerov, 2002). These climatic changes may cause less accumulation and therefore induce insufficient replenishing of glaciers during the wet season, changing the water supply balance.
In the Andes, rituals have been mainly directed to glaciated mountains because they were considered to be source of meteorological phenomena such as rain, snow and hail
(Reinhard, 1985a). Indeed, glacial melt in the Andes constitutes the major water supply during the dry season. Recent changes in glacial cover are affecting the way in which water and fertility rituals are performed in the Andes. For generations, the Qoyllur Rit'i pilgrimage in the Peruvian Andes has involved the practice of carving medicinal ice from the Sinaqara glacier (4,700 m) and carrying it to villages to ensure fertility of the crops and welfare of the community (Gow, 1974). In the past, Catholic priests have unsuccessfully tried to stop this
Andean ritual practice arguing that harvesting of glacier ice is increasing the rate of glacial retreat. However, due to concern about climate change-induced glacier retreat, the guardians of the ceremony (“Pablitos”) refused to participate in the ritual in 2003 (BBC, 2003), based on statements of local people.
In addition to causing changes in local traditions and worship practices, current mass wastage of glaciers worldwide also exposes already existing artifacts to natural decomposition processes and looting. Prehistoric human remains preserved by natural freeze-
2 drying processes in glacial environments are presently revealed by glacier retreat. For example, in 1991, shrinkage of a small glacier in the Hauslabjoch area in the Tyrolean Alps revealed the “Iceman”, a 5000 year old mummy (Baroni and Orombelli, 1996). The body was buried at the end of an era of minimum ice extent (9000 - 5000 yrs. BP), remained covered by glacial ice during “Neoglaciation”- a period of maximum glacier expansion that started around 5300 – 5050 cal. yr. BP - and was exposed at the end of the ablation season in 1991
(Baroni and Orombelli, 1996). Glacial melt due to volcanic eruptions can also destroy archeological remains from local glacial retreat. An example is the frozen mummy of an Inca girl found in 1992 on Nevado Ampato in Southern Peru, after the eruption of the nearby volcano Sabancaya. A series of recent eruptions of Sabancaya starting in 1990 covered the ice cap on the Nevado Ampato with hot ash, changing the energy balance of the ice cap and inducing melting (Thouret et al., 2001a). This exposed the Inca route and tombs and triggered the fast decomposition of the mummy.
When exposed by climate change-induced glacier changes, these artifacts may also be subjected to destruction by looters, tourists and climbers. Looters searching for Inca treasures have destroyed the archeological artifacts on a number of Andean peaks. For example, the Inca sites on Nevado Chachani in Southern Peru and Nevado Quehuar in
Argentina have been dynamited by looters (Ceruti, 1999a; Chávez Chávez, personal communication). In addition, the re-utilization of Inca structures along the climbing routes and the removal of artifacts as “souvenirs” by climbers adds to increased deterioration of the archeological record (Ceruti, 1999a). Such events pose an urgency to locate and preserve these archeological sites.
Tourism and the climbing industry also affect traditional worship practices and views of the sacred mountains in the Andes and the Himalayas, conflicting with local traditions. For example, many Himalayan summits are viewed as abode of the Gods and are therefore off- limits to climbers (such as Macchapucchre in Nepal, Kangchendzonga in Sikkim, Nanda Devi
3 in India and Kailas in Tibet). The sanctity of Kailas, the most sacred mountain of the world, worshipped by four religions (Hinduism, Buddhism, Jainism and Bon-Po), was threatened in the last decade by several attempts of climbers to reach the summit (Hamilton, 2001a;
2001b).
Considering both natural and anthropogenic disturbances of the sacred sites, new methods are needed to help the management of these cultural resources. Bernbaum (1990;
1997) suggested that the spiritual importance of the mountains can be used as a vehicle for conservation of the sacred sites. An example is the tree-planting ceremony at Badrinath, a pilgrimage site in the Indian Himalaya, where local communities’ beliefs of sacred grooves was used to restore the forests in the area (Bernbaum, 1997). Such approaches are also used by local communities in the Andes in the ceremony of cleaning the canals (“acequias”) that bring water from the mountains (Reinhard, 1985a).
However, cultural resource management in the Andes is often limited by poor knowledge about the archeological locations. Various artifacts were discovered on top of
Andean peaks at high altitudes up to 6,700 m (Beorchia Nigris, 1985a). These include ceramics, sacrificed animals and in some cases frozen human beings. Frozen human remains are believed to be offerings made by the Incas in the 15th century to worship the Andean
Mountain Gods (“Apus”). Although mountain worship is encountered in different cultures, as shown by Bernbaum (1990) and Eliade (1963; 1987), the presence of human sacrifices at such high altitudes is unique to pre-Columbian societies (Schobinger, 1967a). Advances in
Inca technology made possible the occupation at high-altitudes in remote areas for ceremonial purposes Spanish chronicles such as Cobo (1990[1653]) and Guamán Poma de
Ayala (1980[1615]) mention Inca customs and religion related to mountain worship. Most of these historical sources were used in the Spanish campaigns to located and extirpate all traces of Andean religion (Ceruti, 1999a). While a large number of lower elevation shrines were destroyed during the Spanish occupation, high-altitude sites could not be reached by the
4 Catholic campaigns due to their location in remote mountain terrain. Consequently, they may still contain a valuable archeological record. New tools such as spatial technologies are needed to locate archeological sites threatened by glacial retreat, looting and tourism in rugged terrain in the Andes Mountains. Predictive modeling has been used in the past to locate archeological sites (Kvamme, 1988b; 1992; Duncan, 2000, Warren and Asch, 2000), but these approaches have not been yet applied in Andean high altitude archeology.
I.3 Archeological predictive modeling overview
Recently, predictive modeling has been identified as a valuable method in cultural resource management applications, including archeological site location studies (Van Leusen,
2002). New approaches in archeology include spatial modeling with the aim of explaining past human behavior with respect to environmental and social variables (Judge and Sebastian,
1988). The main assumption underlying these approaches is that human behavior is patterned, and that patterns between existing archeological sites and environmental or terrain variables can be used to derive a location model for unknown sites (Rose and Altschul, 1988). This constitutes a change from traditional archeology where the focus was on the attributes of artifacts and less on their spatial distribution and their relationship to larger spatial scales
(Wheatley and Gillings, 2002).
Recent advances in Geographic Information Systems (GIS) since the 1980’s have facilitated the implementation of spatial techniques in archaeology (Kvamme, 1999). The analytical capabilities of GIS offer powerful tools to investigate the spatial patterns of artifacts and their relationship with environmental factors. The main advantages of GIS lie in its ability to combine multiple data layers such as topographic maps, remote sensing information, aerial photography and field data (Kvamme and Kohler, 1988). With recent development of digital elevation models (DEM’s) from satellite imagery with better coverage and resolution, it is possible to derive terrain morphology data including elevation, aspect and
5 slope when these measurements are missing from the archeological record (Brandt et al.,
1992). In addition, accurate field data acquired with the use of Global Positioning System
(GPS) can be easily integrated in archeological applications.
Location applications in archeology typically address the problem of identifying patches that are suitable for archeological settlement. Geographical locations which contain archeological remains are referred to as “sites”, as opposed to “non-sites”, which are areas that do not contain archeological evidence. These naming conventions will be followed throughout this thesis. Two types of models have been commonly used in archeological site location studies: inductive (data-driven) models and deductive (theory-driven) models.
Inductive models rely on patterns derived from existing archeological observations (e.g. correlations between known archeological sites and characteristics of the physical landscape).
Deductive models are based on assumptions or a-priori knowledge about past human behavior, and are used when previous archeological data are not available (Wheatley and
Gillings, 2002).
Many modeling approaches developed so far are based on methodology proposed by previous studies, such as Altschul (1988), (Kvamme and Kohler, 1988) and Kvamme (1988b;
1990a; 1988).The model-building process typically involves the following stages: (1) archeological data collection and manipulation (2) reviewing the existing hypotheses about site location, (3) compiling GIS data and developing the key variables initially believed to influence site location, (4) exploring the relationship between site locations and environmental/social factors, and 5) development and testing the predictive model using field data (Altschul, 1988).
There are two main approaches used to construct a predictive model: (a) rule-based approaches and (b) regression-based approaches (Warren and Asch, 2000). Rule-based approaches rely on a decision rule that specifies the criteria for site occurrence based on a set of predictor variables (Wheatley and Gillings, 2002). A weighted map layer approach was
6 used by Brandt (1992) to develop an archeological site location model on flat terrain in the
Netherlands using GIS techniques. Another example of this approach is Wheatley (2002), who used a weighted map layer approach for archeological prediction in the coastal area of
Upper Chesapeake Bay. A GIS predictive model based on weighted map layer approach for mapping the archeological potential of ice and snow in the Alaskan glaciers and snowfields has been developed and tested by Dixon and Manley (2001; 2002).
The weighted map layer approach is adequate when the terrain lacks relief and when the layers used are mostly thematic, e.g. soil classes, geology or vegetation. However, this approach is not suitable in highly variable terrain since these terrain factors should be treated as continuous variables. In addition, in the weighted map layer approach, the goodness of fit of the model and the predictive power of each variable cannot be quantified (Wheatley and
Gillings, 2002).
More powerful tools for predictive models are based on statistical analysis (Rose and
Altschul, 1988). These tools include descriptive statistics and multivariate analysis such as linear multiple regression and multiple logistic regression (Kvamme, 1988b; Rose and
Altschul, 1988; Wheatley and Gillings, 2002). Regression analysis yields the weight and statistical significance for each predictor, allowing for an objective selection of variables.
Multiple linear regression is rarely used for archeological data because the normality assumption is often violated. Besides, this approach is not adequate because the dependent variable - site presence/absence or site probability- is not measured on an ordinal scale
(Wheatley and Gillings, 2002). An alternative statistical approach is multiple logistic regression, which calculates the probability of site occurrence based on supposed predictors for site location. Since logistical regression uses less assumptions than linear regression, it is the statistical approach most often used in archeological predictive models (Warren and Asch,
2000).
7 Warren (2000) used logistic regression and GIS techniques to predict archeological site location in an upland prairie region of Central Illinois. Similar techniques were used by
Duncan (2000) to predict the occurrence of archeological sites in Pennsylvania and Virginia.
They relied on GIS techniques to derive additional variables such as slope, aspect and a simple measure of insolation. Another example of the use of logistic regression in archeology is the optimal foraging model developed for the Northeastern Continental Shelf to locate sites that were submerged 18,000 BP and were recently exposed (Judge and Sebastian, 1988).
While the above mentioned works provide the foundation for building a GIS archeological predictive model, a few questions arise. For example, are these approaches applicable in other landscapes such as complex glaciated mountainous terrain? Which of the various types of models is appropriate to use and when? Furthermore, most archeological predictive models developed so far are for simple terrain (plains or high plateaus) in the
Netherlands and North America. Only a few predictive models have been developed for mountainous terrain, such as Kvamme (1990a; 1992). It is unclear whether these approaches can be successfully used to model archeological sensitivity in complex mountain terrain.
With recent advances in the use of quantitative techniques for archeology, it is hoped that these modeling approaches can be adapted to successfully map archeologically sensitive areas in the Andes Mountains. The model-building process employed here is based on methodology developed by Kvamme and Kohler (1988), Kvamme (1988a; 1988b) and
Wheatley and Gillings (2002), and involved two major steps. The first step involved the hypothesis development, the creation of the spatial database and the initial development of the key variables (Figure 1.1a). The second step was the development and testing of the predictive model using logistic regression (Figure 1.1b). The model takes an inductive approach based on the existing archeological record and on the statistical relationships between environmental variables and archeological site distribution.
8 I.4 Research design and objectives
This research is conducted with the long-term goal of helping cultural resource preservation in the Andes by identifying archeologically sensitive areas for selected mountains in the Andes. In order to achieve this goal, a set of intermediate objectives has been established to be achieved in this stage of the project. These objectives are: (1) to construct a spatial inventory of existing high-altitude archeological data in the Andes by integrating and evaluating various data sources; (2) to review existing knowledge and initial hypotheses about high-altitude archeology in the Andes; (3) to illustrate the methodology used to develop the key model components and (4) to develop and test a predictive archeological site location model using the revised archeological record and spatial techniques (logistic regression and GIS). Here, it is helpful to mention that the model should not be regarded as means to predict the exact locations of undiscovered sites. Rather, it is used as a tool to extrapolate from known locations and to map unsurveyed locations that display similar patterns to those discovered.
There are several justifications for this study. First, there are problems with the existing archeological record in the Andes. There is limited accessibility to previous survey data. Although a few previous surveys have been published internationally, such as Beorchia
Nigris (Beorchia Nigris, 1985a) and Ceruti (1999a), some are available only in South
America. Data relevant to elevation and geographical location are scarce and not well documented. A further concern is the lack of consistency in the data sources. Various instruments and survey methods were used to record these data, leading to inconsistencies in data formats. Hence, there is a need to inventory the archeological site attributes in a common format.
9 Figure 1.1 Generalized flow chart illustrating the steps used in the model-building process. a)
development of the archeological, environmental and social data layers; b) development
and testing of the predictive model. Diagram adapted from Warren and Asch (2000).
10 Secondly, traditional archeological surveys in the Andes are limited by extreme environmental conditions. High altitude archeological research is limited by difficulty of fieldwork at high altitudes, on glaciated terrain and in extreme climatic conditions. The retrieval of artifacts in such conditions is difficult, and local expertise prepared to deal with these extreme conditions is not readily available. Furthermore, the logistical support for this kind of work is limited, and obtaining archeological permits is a lengthy process. There is a need to identify archeologically sensitive areas to advise archeological survey team about where to spend their limited resources.
I.5 Organization of the thesis
Chapter 2 provides an overview of the cultural, climatic and landscape context of the high-altitude archeological sites found in the Andes Mountains. It also presents existing hypotheses about site location, as a basis for collection of ancillary data needed to develop an archeological predictive model. Chapter 3 presents the construction of a spatial database for inventorying archeological sites found at high altitudes. Chapter 4 focuses on the initial choice and development of environmental and social factors believed to influence archeological site location. Chapter 5 presents a reconstruction of possible Inca climbing routes using least cost path analysis approaches. Chapter 6 presents the development and testing of an archeological sensitivity model for mountainous areas in the Andes.
11 "' /3$136 .
'(&' +3(34#$ 1"'$.+.&83'$".-3$73
II.1 Introduction
Prior to developing an archeological predictive model, it is necessary to develop an understanding of the archeological record. The archeological record consists of artifacts deposited in a past cultural environment, which are no longer used by the present society
(Schiffer, 1987). It is further necessary to reconstruct the spatial and temporal context in which the artifacts were deposited, the paleo-climatic conditions, and the post-depositional processes that might affect these artifacts. Recent geoarcheology approaches help reconstruct the context of artifacts by applying stratigraphy, geochronology and absolute dating techniques (Waters, 1992).
The discovery of artifacts and mummies at high elevations on Andean peaks
(Reinhard, 1985a) raises various questions. First, what culture constructed these sites?
Second, where did these mummified individuals come from? Third, why were they located at high altitude sites? This chapter provides a literature review pertaining to high-altitude archeological sites, to help answer these questions. Here I review the existing hypotheses about site location on two scales: a macro scale (all the mountains containing archeological sites within the Inca Empire extent) and a micro scale (inter-site level, pertaining to multiple archeological sites on a mountain) following the methodology proposed by Ceruti (1997).
The macro scale addresses the question of why these sites were constructed; the micro scale addresses the question of how their location on a specific mountain was chosen.
These hypotheses are presented in light of the cultural, paleo-climatic and paleo- landscape context of the high-altitude archeological sites found in the Andes. The cultural context includes clues about the ritual practices of the Incas during the 15th – 16th century
12 inferred from ethno-historical sources. The cultural context includes a spatial dimension representing the geographical provenance, or origin of the artifacts and a temporal dimension representing their age. The climatic context is reconstructed from temperature and precipitation proxies such as ice cores and dendrochronology. The landscape context includes the geomorphic and geographical setting, and the natural disasters associated with these settings. An emphasis is placed on the development of hypotheses about site location based on these cultural and environmental settings. These hypotheses constitute the basis for the collection of ancillary environmental and social data that is further needed to develop an archeological predictive model.
II.2 The cultural context
II.2.1 Mountain worship in the Inca culture
The Inca Empire extended from present Colombia to Northern Chile over 4,300 km
(Bauer, 1992) (Figure 2.1). The empire was called “Tawantinsuyu” and was organized in four regions, or “suyus”: Condensuyu, Collasuyu, Andesuyu and Chinchaysuyu (Guamán Poma de
Ayala, 1980[1615]). In a short period (ca.1475 to 1532), the Incas developed a complex system of administrative centers, roads, canals and architecture. These technological advances made possible the use of high altitudes for ritual purposes (Schobinger, 1969). In
Inca religion, a special importance was accorded to certain mountains, considered Gods and designated by the terms “Apus”, “Wamanis”, “Achachilas”, “Awkillas”, “Apusuyus” or
“Apuosentos” (Martinez, 1983). The mountaintops of Coropuna, Pariacaca, Pitusiray, and
Vilcanota were considered guardians of each of the four quarters of the Inca Empire
13 Figure 2.1 Map of the study area showing the Inca Empire. The box indicates the area
containing high altitude archeological sites (Southern Peru to Chile).
(Guamán Poma de Ayala, 1980[1615]). Other volcanic peaks such as Ampato and Sara-Sara in Southern Peru were also regarded as “Gods” by the Incas, and to this day they are still considered so by the local communities (Reinhard, 1985). According to ethno-historical sources such as Cobo (1990 [1653]) and Guamán Poma de Ayala (1980[1615]), the Incas climbed these peaks to perform worship ceremonies during special occasions. Remains of these worship sanctuaries are found throughout the extent of the Inca Empire in the Central
Andes of Chile, Argentina, Peru and Bolivia at altitudes ranging from 900 m (Esmeralda, in
Chile) to 6,739 m (Volano Llullaillaco, Argentina)(Ceruti, 1997).
Although mountains are worshipped in various religion traditions around the world
(Eliade, 1987), the practice of human sacrifice at high altitudes is exclusively encountered in pre-Colombian societies, especially Andean cultures (Schobinger, 1967a). Spanish chronicles
14 (Guamán Poma de Ayala, 1980 [1615]) documenting Inca religion and customs refer to a ritual called Capac Cocha, as the most important ceremony in the Inca religion. This ritual involved sacrificing a chosen child on special occasions: to designate a new emperor
(Quevedo and Durán, 1992), to worship the Sun deity (Cobo, 1990 [1653]), to appease the volcanic mountain gods (Thouret et al., 2001a) or to pray for water and fertility (Reinhard,
1985a). These hypotheses are reviewed in the second part of this chapter.
II.2.2 High-altitude archeological investigations
Artifacts at high altitudes were reported by local Andean people, explorers and climbers as early as the end of the 19th century (Schobinger, 1967b). The first ruins were discovered on top of Chuculai (5,420 m) in Chile by the geographer Francisco San Román
(Vitry, 1997). During the first half of the 20th century, local people and climbers reported ruins on Socompa (6,031m), Morrado (5,200m), and volcano Gallán (6,000m) in Chile. Since the 1960’s, archeological surveys have been performed by a small group of researchers from
Peru, Argentina and Chile, mainly from two institutions: the Center for Archeological
Investigations of High Mountains (CIADAM) from Argentina and the Catholic University of
Santa Maria from Arequipa, Peru. Antonio Beorchia Nigris - director of CIADAM and Juan
Schobinger from the National University of Cuyo, Argentina conducted extensive archeological surveys on more than 100 peaks in Northeastern Argentina between 1963 and
1970 (Ceruti, 1999a). The results of these archeological surveys are reported in Spanish in the publications of CIADAM (Beorchia, 1985; Ceruti, 1999b). Most recently, investigations were led by Constanza Ceruti from the National University of Salta, Argentina (Ceruti, 1997;
1999a) and Johan Reinhard from the Mountain Institute (Reinhard, 1985a; 1996; 1999;
Reinhard and Ceruti, 2000; Reinhard and Chávez, 2001). In Peru, José Antonio Chávez
Chávez from the Catholic University of Santa María, Arequipa, has been directing archeological surveys under the project Museum of Andean Sanctuaries and conducted
15 surveys on a number of peaks in Southern Peru, such as Pichu Pichu (5,634 m), Chachani
(6,056 m), Misti (5,821m) and Ampato (6,270 m)(Chávez Chávez, personal communication)
In addition to these local efforts, archeologists from University of Varsovia initiated the
Condensuyos project (Ziólkowski and Belan Franco, 2001) and conducted extensive surveys in Southern Peru from 1996 to the present. Their findings are reported in Ziólkowski and
Belan Franco (2001) and Sobczyk (2000).
High altitude archeological research has been limited to such a small group of researchers because of the difficulty of conducting fieldwork at high altitudes, in extreme climatic conditions. Local expertise prepared to deal with these extreme conditions is not readily available (Ceruti, 1999a). It is worth mentioning that due to the difficulties of collecting field data, local archeologists have not always been willing to collaborate or share archeological data. This leads to inconsistencies in the archeological record due to multiple measurements and differences in survey methods.
II.2.3 The archeological record
The archeological record in the Andes consists of ecofacts and artifacts that provide evidence of ritual activity at high altitudes. Although generally these sites are referred to as
“high altitude sites”, in reality not all of them were found at extreme altitudes. Beorchia
(1985a) defines “high altitude sites” as sites situated between 5,000 m and 6,700 m,
“medium” sites between 3,000 and 5,000 m and “low altitude” sites less than 3,000 m.
These archeological sites are distributed on the mountains along climbing routes similar to present-day climbing strategies (Ceruti, 1999a). Each site had one or several functions based on their inferred use during the ritual activity. These functions can be utilitarian (techno-functions) - when the artifacts are used to extract and process resources or symbolic - when the artifacts are used in social functions or ideologic - when the artifacts serve to convey cultural beliefs (Schiffer, 1987). In this case, sites typically include a base
16 camp (used for temporary habitation purposes), a few intermediary logistical sites (situated along the climbing route), the main ceremonial site and a sometimes a lookout point. An example of a typical system from volcano Licancabur in Chile is shown in Figure 2.2.
Figure 2.2 Typical high-altitude archeological complex from Volcano Licancabur, Chile.
Reproduced from Reinhard (1985a).
Ceremonial sites were used for ritual purposes to make sacrifices and offerings to the mountain Gods (Ceruti, 1997). They typically include a circular/rectangular platform (Figure
2.3a), and a number of ritual objects such as offerings of coca leaves, textiles, gold and silver statues, Spondyllus1 shells and burned wood (Beorchia Nigris, 1985a). Ritual sacrifices (both human and zoomorphic) were also found at ceremonial sites, for example Aconcagua,
Llullaillaco and Ampato (Beorchia, 1985). These sites tend to be located in the highest part of
1 Spondyllus is a shell of bright red color encountered only in warm tropical waters in Ecuador, México, Central America and Caribbean. Since the Humboldt current maintains a colder ocean temperature on the coasts of Peru, the shells encountered on top of the mountains from Peru to Chile were probably brought by the Incas from Ecuador (Beorchia Nigris, 1985a). Spondyllus was used as
17 the mountain, when the topography of the summit is flat and ample (Ceruti, 1999a). Sites along the climbing route which display a large visual perspective were used as lookout points
(“miradores”) (Figure 2.3a).
Fotos: A. Racoviteanu
Figure 2.3 Ceremonial complex on Nevado Pichu Pichu: a) ceremonial platform (double Inca
wall) indicated by the solid line. The dashed arrow indicates the lookout point; b) the
base camp ruins with the summit in the background; c) a close-up view of the habitation
constructions at the base.
It is useful to note that ceremonial sites are not always located on the summits, and that mountain worship sites are also encountered at the base of the mountain. For instance, the most important ceremonial site on Nevado Coropuna, the ruins of Mauca Llacta de
symbol of water and fertility; it might signify that the rituals were performed in a dry period to appease the mountain Gods to bring rain.
18 Pampacolca are located at the base of the mountain (Figure 2.4) (Sobczyk, 2000; Ziólkowski and Belan Franco, 2001).
Figure 2.4 The ceremonial complex of Mauca Llacta of Pampacolca, at the base of Nevado
Coropuna (6,426 m) in the Peruvian Andes.
Logistical sites served as temporary shelters during the ascent and were situated along the climbing routes. Base camps consist of low constructions and various structures tend to display large habitation complexes with a main square, and large quantities of ceramics used for domestic purposes. These sites are usually located at lower elevations in close proximity to access routes (Ceruti, 1999a)(Figure 2.3b and 2.3c). Both base camp and logistical sites display artifacts with techno-utilitarian functions used for biological needs, such as processing of resources and construction of shelters (Schiffer, 1987). Examples include habitation constructions, icchu grass, wood used for heating, and enclosures used for burning wood (fogones) (Beorchia Nigris, 1985a; Ceruti, 1997). Such sites were found on
Ampato, Ascotan de Ramaditas, Licancabur, Llullaillaco and Pichu Pichu (Ceruti, 1997).
Rock cairns arranged in a pyramidal cone (‘apachetas’) found along access routes are used
19 by local people in the present day to make offerings to Mother Earth (Pachamama) and ask absolution from mistakes (Gow, 1974).
II.2.4 The temporal and spatial context of the high-altitude archeological record
The excellent preservation of some of the frozen mummies found on Andean peaks allows reconstructing the temporal context (age) and the spatial context (provenance of the artifacts and the paleo-environmental conditions). To the present day, a total of 25 frozen bodies were reported on Andean peaks (Appendix 1). Most of these bodies pertained to children of less than 10 years of age with a male/female ratio of 11/7 (several not identified), found in a fetal position (Beorchia Nigris, 1985a). Some bodies were destroyed by looters, such as the mummy on Nevado Chachani in Peru (Chávez Chávez, personal communication) and Nevado Quehuar (Ceruti, 1997). Only three bodies have been used for radiocarbon dating, provenance studies, and isotopic analyses (Fernández et al., 1999). Results of these studies are reported in detail in Fernández et al.(1999), Wingenroth (2001), Reinhard and
Ceruti (2000) and Schobinger (1964; 1966; 1969; 2001a).
Radiocarbon dating (14C) revealed that the ritual sacrifices dated from the Inca period
(15th –16th century) (Schobinger, 2001a; 2001b). However, there is no consensus about the
Inca period (Schobinger, 2001a). For example, the 14C dating from the Geochron
Laboratories2 initially placed the mummy of Aconcagua around cal 1580 ± 70 years, while the Beta-Analytic Laboratory3 placed it around cal 1470 ± 40 (Schobinger, 2001b). Other sites - Negro Ovrero, cal 1380 ± 80 and Mercendario, cal. 1540 ± 80, most probably around
1480 – were placed earlier in the first Inca period (Beorchia Nigris, 1985a)4. The 14C age of the Ampato mummy “Juanita” was cal. 1290-1450 (Thouret et al., 2001a), so it’s uncertain to
2 AMS Dating method, using a bone fragment and pre-treatment methods (filter, wash, boil in slight acid) and computerized correction (CALIB 3.03) 3 AMD Dating method using hair samples and calibration
20 which Inca period it belongs. Some of these results are difficult to interpret due to unknown transformations of organic materials exposed to cosmic rays at high altitudes (Ceruti, 1999a).
When 14C dates are doubtful, the architecture of the archeological constructions may help identify the temporal context of the artifacts.
Samples extracted from the Inca mummies were used to infer their provenance (the geographical region of origin) based on isotopic analysis. This method has been commonly used to reconstruct paleo-environmental conditions and more recently to infer the diets of prehistoric individuals (Fernández et al., 1999). Two studies (Fernández et al., 1999) and
(Wingenroth, 2001) used stable isotopes to determine the geographic origin of the mummy of
Aconcagua. Carbon, nitrogen and sulfur isotope ratios from bone collagen and hair samples were compared to ratios measured in locals from both high-altitudes and coastal locations
(Fernández et al., 1999). The analysis determined that the sacrificed child had a terrestrial diet, suggesting that it came from the mountainous region near to Aconcagua, rather than from a coastal location (Fernández et al., 1999). Similarly, Wingenroth (2001) found that pollen grains found in textiles, exposed hair, fecal matter and vomit come from plants currently growing on the slopes of Aconcagua. Thus, vegetation patterns at the time of the sacrifices may have been similar to present day patterns.
In basis of these analyses, and the offerings found at these sites, it was concluded that the frozen mummies found on Andean peaks pertain to the Inca period (Beorchia Nigris,
1985b; Schobinger, 1991).
4 Date communicated by Juan Schobinger at VI Congreso Nacional de Arqueologia Argentina, San Luis, Nov.1982). The dating was done on wood found in the ceremonial platform at the summit. It is mentioned that the wood could be old or dry thus explaining the early date suggested by the analysis.
21 II.3 Existing hypotheses
II.3.1 Macro scale: Why were the sites constructed?
Various hypotheses about why the Incas performed human sacrifices at high
altitudes on Andean peaks were previously formulated. The most common hypotheses
include the sun worship hypothesis (Mostny, 1957; Rebitsch, 1966), the water and fertility
hypothesis (Reinhard, 1985a), the natural disasters hypothesis (Chávez Chávez, 1993;
Thouret et al., 2001a), the astronomical hypothesis (de Molina, 1989 [1575]) and the
expansion hypothesis (Schobinger, 2001a).
a) Sun worship hypothesis
Some of the first theories (Mostny, 1957; Rebitsch, 1966) suggested that the high-
altitude Inca ruins were constructed for the purpose of Sun worship. Sun (Inti) was the
main deity worshipped by the Incas (Beorchia Nigris, 1985a). Alignments of archeological
constructions with the position of the sun during solstices were common in Inca astronomy
(Aveni, 1990; 1996; Bauer and Dearborn, 1995; Urton, 1978; Zuidema, 1977). Such
examples include the shrines on top of Huanacauri in Cuzco (Bauer and Dearborn, 1995)
and Macchu Pichhu (Reinhard, 1991). Since orientation was important in Inca astronomy,
the rituals may have been related to the movement of the sun in the sky on the East-West
axis (Aveni, 1996). Many high altitude archeological sites have openings to the East, the
cardinal direction of the rising sun (Beorchia Nigris, 1985). However, it is unclear whether
all high altitude archeological sites were exclusively used for sun worship and other
astronomical purposes (Reinhard, 1985a).
b) Water and fertility hypothesis
Mountain Gods were held responsible for water supply in pre-Columbian beliefs
(Reinhard, 1985a). To the present day, Andean rituals are centered on the concept of water
22 and fertility (Reinhard, 1985a). Present day pilgrimages to high sanctuaries occur during the dry season (June to October), when local people to pray for fertility of the fields during the harvest season (Sallnow, 1974). It was suggested that water shortages motivated the Incas to perform human sacrifices to appease the angry Gods (Chávez Chávez, personal communication).
Based on abandonment patterns in Santa Elena peninsula of Ecuador, Paulsen (1976) argued that climatic anomalies determined the sequence of rise and collapse of different
Andean Empires, including the Incas. Her chronology shows that a dry period began during the Late Horizon (1476 to 1532), when the Inca Empire was established (Figure 2.5a). a. b.
Figure 2.5 Paleoclimatic reconstructions in the Andean area. a) Paulsen’s (1976) series of rise
and collapse of Andean Empires. b) Ice core record from Quelcaya Ice cap, Peru,
reproduced from Thompson et al. (1984). Lower accumulation of ice and more negative
δ18O values suggest decreased precipitation in the first half of the 15th century,
overlapping with the rise of the Inca Empire.
Proxy records of annual precipitation (measured ice accumulation) and temperature
(oxygen isotopes) extracted from ice cores provide climatic reconstructions for the past 1500
23 years. The best records are from Huascaran, Quelcaya, Sajama and Illimani (Henderson et al.,
1999; Shimada et al., 1991; Thompson et al., 1994; Thompson et al., 2000; Thompson et al.,
1984; Thompson et al., 1995). The ice accumulation record from Quelcaya provides evidence of a sustained period of aridity from 1100 to 1500, overlapping with the rise of the Inca
Empire (Figure 2.5b) (Thompson et al., 1988; Thompson et al., 1984). The ice core records are consistent with periods of reduced precipitation in other regions and even globally
(Chepstow-Lusty et al., 2003). For instance, a pollen record from the Cuzco region supports the evidence of reduced precipitation starting with AD 1100 (Chepstow-Lusty et al., 2003).
Given this climatic evidence, the question is: Were Inca rituals performed during the drier period at the end of 15th century? Pollen analysis from textiles buried with the mummy from Nevado Ampato showed that the ritual happened at the end of the dry period (Chávez
Chávez, 1993). The presence of Spondyllus shells in the tombs seems to indicate that the rituals were linked to water shortages. However, additional 14C dates place other rituals during 1480 – 1530, at the end of the dry period and the beginning of the humid period
(Corte, 2001). Hence, it appears that some rituals were performed during the prolonged drought period (first part of the Inca period) and others after the drought (second part of the
Inca period), to prevent further water shortages.
c) Natural disasters hypothesis
The climate hypothesis is complemented by another hypothesis that links human sacrifices with volcanic activity with during the Inca period (Thouret et al., 2001a). Tectonic activity in the Andes has continually generated natural disasters such as earthquakes, ice avalanches and other glacier-related hazards. For example, the city of Huaraz at the base of
Cordillera Blanca in Peru has been partially destroyed by an outburst flood from a glacier lake in 1702, then again by an ice avalanche in 1725 (Morales-Arnao, 1999). More recently, it was completely destroyed during a major earthquake in May 1970. The earthquakes caused
24 debris avalanches to originate from the north peak of Nevado Huascarán in 1962 and 1970
(Ericksen et al., 1970). The 1970 avalanche debris traveled at an average velocity of 280 km/h(Ericksen et al., 1970), and completely destroyed the towns of Yungay and Ranrahirca.
The casualty resulting from the 1962 and 1970 events combined was more than 25,000 people (Morales-Arnao, 1999).
Such catastrophic events also occurred in the Pre-Columbian era. Ancient debris provides evidence of a much bigger pre-Columbian avalanche from the north peak of
Huascaran (Ericksen et al., 1970). In Southern Peru, volcanic eruptions of Misti and
Sabancaya occurred around 1466, destroying the city of Arequipa (Thouret et al., 2001b). A series of continual eruptions of the volcano Sabancaya occurred from 1990 to 1998 (Thouret et al., 2002). This induced the melting of the ice cap on near-by Nevado Ampato and revealed the mummy of a young Inca girl sacrificed in a ritual on top of the mountain. The tephra layers found at the mummy site on Ampato were similar to those of an ancient eruption of the
Sabancaya-Ampato massif, suggesting that a similar series of eruptions might have happened at the time of the ritual sacrifice (Thouret et al., 2001a; Thouret et al., 2002). It has been thus hypothesized that some of the human sacrifices were performed to appease the volcanic Gods
(Chávez Chávez, 1993; Thouret et al., 2001a).
d) Ceques hypothesis
Worship at high-elevations in the Peruvian Andes may also be linked with lines called „ceques”, that radiate out from a center (i.e. village or temple) to natural shrines (hills, mountaintops, rivers). According to (Zuidema, 1977) and Aveni (1996), the ceques were designed as straight lines and the sacred sites were chosen because they fell on the lines. For example, it was hypothesized that straight lines connect Cuzco with hilltops and mountaintops in the Sacred Valley of the Incas (Farrington, 1992) (Figure 2.6).
25 Figure 2.6 System of straight lines radiating from Cuzco, the Inca capital, to natural shrines
(huacas) in the Sacred Valley of the Incas. Lines digitized from Farrington (1982) are
shown on a shaded relief map of a digital elevation model.
Straight lines up to 10 km, similar to the ceques, were observed in the Sajama region in Bolivia (Morrison, 1978) and in the village of Socaire in Chile (Reinhard, 1985a). These lines connect the villages with mountaintops that are worshipped in the present day
(Reinhard, 1985a). However, there is no consensus about the pathways used by the Incas to travel to the ritual sites. It is possible that the Incas traveled in a straight line to smaller hilltops to perform the Capac Cocha ritual of human sacrifice, as is suggested by some ethno- historical sources (de Molina, 1989 [1575]). However, when the ritual was performed on high mountains, zigzag lines may have been used, as suggested by Rowe (1979,1981), Bauer
(1998) and Niles (1987).
26 e) Expansion hypothesis
Another theory suggests that Inca mountain worship may be related to the expansion of the Inca Empire in search of natural resources. The Incas extended the road system from present day Ecuador to Chile (Hyslop, 1984). Trans-mountain roads built over mountain passes at altitudes up to 5,000 m were especially important for communication
(Hyslop, 1984). Therefore, it was hypothesized that when a trans-mountain segment was constructed, the mountaintop near it was chosen by the Incas as a ceremonial site
(Schobinger, 2001a). For example, the Inca sanctuary of Aconcagua, containing the mummified body of an Inca child, was located next to an important trans-mountain road
(“Camino Real”) which today links Argentina and Chile. The human sacrifice of Aconcagua may have been offered to the mountain to mark the expansion of the Empire (Schobinger,
2001a). There are eight additional examples of human sacrifice sites located near segments of
Inca Roads: Nevado Chachani, Nevado Pichu Pichu, Licancabur, Azufre, Toro, Mercendario and Tortolas (Ceruti, 1999b). The expansion hypothesis is supported by radiocarbon dating
(14C) from wood extracted from Cerro Mercendario, which placed the construction of the site on Mercendario around 1480, at the beginning of the Inca expansion (Ceruti, 1999b).
Thus, it has been also hypothesized that the construction of some high-altitude archeological sites was related to mining (Ceruti, 1999b). Certain mountains containing minerals such as gold, silver and copper were worshipped by the Incas (Cobo, 1990 [1653]).
Examples include Volcano Antofalla, Esmeralda, Potosi and Puntiudos for silver mines; Las
Cuevas for gold and Nevado Acay for copper (Beorchia Nigris, 1985a).
II.3.2 Micro scale: How were site locations chosen on a mountain?
Terrain characteristics are believed to influence site location in most archeological studies (Kvamme, 1988b). In addition, certain environmental factors might have determined the location of Inca archeological complexes on a mountain (Ceruti, 1997). These factors are:
27 water availability, topography (altitude, slope, local relief, exposure, roughness), geomorphology (landscape, mountain type), climate (precipitation and temperature), cryonival processes (snow, ice and permafrost), vegetation cover, resource availability, accessibility and visibility (Ceruti, 1997). Hypotheses about how each of these factors may influence the choice of a site on a particular mountain are summarized below.
a) Water resources
Water resources constitute an important consideration in archeological site location (Kvamme, 1988b). It has been observed in various archeological investigations that distance from archeological settlements to water sources such as lakes and streams tends to be minimized (Kvamme, 1988b). In the Inca high-altitude archeology, given the large contribution of glacial melt in water supply in the dry season, mountains were recognized as a source of water. The hypothesis here is that glaciated peaks might have been chosen for worship.
b) Topography
Topography determines to a certain extent where archeological sites are located on a mountain (Kvamme, 1988b). The steepness of the terrain is commonly considered in location studies since most archeological sites tend to occur on leveled surfaces (Kvamme,
1988b). The choice of a flat area is believed to have been motivated by the need to accommodate large groups of people (Ceruti, 1997). In addition, local relief determines the accessibility of a site - areas with less relief are more accessible. Aspect provides a measure of shelter from the sun. In the Southern Hemisphere, South-facing aspects provide more shelter, and are therefore more suitable for preservation of artifacts in the cryosphere. The exposure of a site to natural elements such as wind is also believed to be important in choosing a site (Kvamme, 1988b). For instance, Inca logistical sites used as acclimatization stations or stops en route to the summit, might have benefited from locations sheltered from the wind (Ceruti, 1999a).
28 c) Geomorphology
The Andes Mountains have a complex morphology, due to the last folding events during the Pliocene-Quaternary (Narshkikh, 1996b). Today, the Andean mountain range is continually shaped by erosional processes, notably glacial, fluvio-glacial and wind (aeolian and arid) morphosculpture (Narshkikh, 1996a). The location of high altitude archeological sites may vary with each mountain depending on morphology. For instance, ceremonial platforms tend to be located in the highest part of the mountain (summit or pre-summit).
However, when high elevations on mountain were rugged with a narrow and abrupt summit, these sites were not suitable and the ceremonial platforms were build in pre-summit locations instead (Ceruti, 1997).
d) Climate and cryonival processes
Climatic factors contribute to determining optimal conditions for preservation of human sacrifices, as inferred by Corte (2001) and Reinhard and Ceruti (2000). For example, cold temperatures and the presence of glacier ice favor the preservation of sacrificed bodies by mean of natural freeze-dry processes (liofilization). It has been hypothesized that the Incas possessed the knowledge about these processes and that the sacrifices were performed on summits that were permanently covered by ice
e) Vegetation cover
Vegetation cover was important for the Incas since it determined the availability of wood and icchu grass. Large quantities of wood were found on top of some Andean peaks
(up to 4,000 kg on Licancabur, 200 kg on Chilques and 200-300 kg on Tambillos)(Beorchia
Nigris, 1985a). Wood was used in combustion for heating at the logistical sites, or to burn offerings at the ceremonial sites. Wood sticks were also used as tools to dig the tombs
(Beorchia Nigris, 1985a). The icchu grass was used at intermediate sites for roofs and insulation. Since transporting these large quantities of wood up the mountain requires
29 considerable effort, effort might have been minimized by choosing locations closer to wood sources (Ceruti, 1997).
f) Accessibility
Site accessibility is determined by two factors: proximity to access roads
(described in the previous section) and local topography. Elements from the morphology of the mountain (shape, type of rock, landscape) may influence the placement of a site. In addition, snow cover may influence the accessibility to a site and the suitability of a site for construction of archeological complexes. For instance, travel over a snow-covered area requires significant effort to transport the materials and construct the sites. Therefore, snow- free climbing routes might have been preferred to minimize effort (Ceruti, 1997). This consideration is important assuming that the Incas lacked technical equipment to tackle complex mountain terrain (Ceruti, 1999a).
II.4 Conclusions
These hypotheses provide guidance to the data collection for the creation of a database of environmental and social factors associated with high altitude archeology. Only hypotheses at the micro-scale (such as topography and climate), and social hypotheses such as distance to roads and mines are tested in this thesis (Chapter 5). It is worth mentioning that some of the present day climatic variables may not be adequate for assessing environmental conditions at these sites during the Inca period.
30 "' /3$13'1$$
$5 +4 3(-&#(&(3 +$+$5 3(.-,.#$+2%.1&+ "(.+.&(" //+(" 3(.-2 - $7 ,/+$%1.,-$5 #.".1./4- /$145( - -#$2
Adina E. Racoviteanu1, 2 ∗, William F. Manley2, Yves Arnaud3 and Mark W.Williams1,2
1 Department of Geography, University of Colorado, CB 260, Boulder CO 80309 2Institute of Arctic and Alpine Research, University of Colorado, CB 450, Boulder CO 80309 3 Institut de Recherche pour le Développement, LGGE, BP 96, 38402, Saint Martin d’Hères, France
III.1 Introduction
Digital elevation models (DEM's) are beginning to see wide use in glaciological applications. Some studies have used DEM’s to extract components of glacier topography
(slope and aspect), which were then combined with satellite images to map glacier areas
(Duncan et al., 1998; Kääb et al., 2002; Klein and Isacks, 1996; Paul et al., 2002; Sidjak,
1999). In addition, DEM’s have been used as tools to derive hypsometry maps at different time steps and to quantify vertical surface changes on glaciated areas in remote areas, as indirect measurements of mass balance (Khalsa et al., 2003).
Several studies have explored ways to assess glacier mass balance and volumetric change by using a time series of digital elevation data. For example, Etzelmüller (2000),
Etzelmüller and Björnsson (2000) and Etzelmüller et al. (1993) discussed GIS techniques to quantify changes in elevation, terrain roughness, glacier hypsometry and flow patterns using grid-based DEM’s. Rentsch et al. (1990), Vignon et al. (2003) and Rivera and Casassa (1999)
∗ Corresponding author. Email: [email protected], Fax. 1-303-492-6388
31 estimated changes in glacier volume and mass balance based on reference elevations from topographic data.
The availability of new remote sensing platforms with high resolution, global coverage and low costs provide the potential to calculate glacial mass balances in remote areas with little existing glacial information, such as the Andes of South America. Two of the new sensors are the Advanced Spaceborne Thermal Emission and Reflection Radiometer
(ASTER) sensor and the Shuttle Radar Topography Mission (SRTM). These sensors acquire simultaneous stereo images from different directions, suitable for generation of DEM’s.
Glaciologists would like to evaluate changes in glacial mass balance over time by comparing changes in DEM properties acquired at different times (Etzelmüller, 2000). Efforts are being undertaken to provide accuracy assessments of the new DEM’s from SRTM and ASTER imagery. For instance, ASTER – derived DEM’s (30 m resolution) have been validated at several sites (e.g. Hirano et al., 2003; Kääb, 2002; Lang and Welch, 1999; Welch et al.,
1998). The recently released SRTM-3 (90 m resolution) datasets have been evaluated only at a few study sites, on non-glaciated terrain (e.g. Falorni et al., 2003b; Falorni et al., 2003a;
Rabus et al., 2003). Therefore, it is unclear if the data available to glaciologists from these new remote sensing instruments provide sufficient spatial and temporal resolution to detect a glacial signal without extensive calibration.
Combining these new satellite-derived DEM’s with DEM’s constructed from topographic maps has the potential to extend the time series of glacial change over many decades. This ability to construct glacial mass balances over many decades is particularly well-suited to remote areas such as the Andes where there is little historical research in glaciology. In such areas, digitized elevation contours from old topographic maps still constitute a ready source of historical data on glacier elevation and area. However, there is no established interpolation method especially suitable for creating continuous elevation data from these topographic maps for accurate representation of glacier terrain. The accuracy of
32 various techniques to construct DEM’s from digitized contour data has been addressed in GIS literature, e.g. Burrough and McDonnell (1998); Wood and Fisher (1993), but the glaciological community has yet to agree on a suitable interpolation method. For instance,
Etzelmüller and Björnsson (2000) used an Inverse Distance Weighted (IDW) interpolator to create a continuous surface from radar profile lines on a glacier. Other authors (e.g. Mennis and Fountain, 2001) chose a spline interpolation for surface representation of glacier and sub- glacier topography from digitized contour lines. Alternatively, Gratton et al. (1990) chose a
Triangular Irregular Network (TIN) derived from digitized contours to represent rugged glacier topography at the Columbia Icefield. The preference of one interpolation over the other depends on terrain topography and the type of data analysis needed. So far, only a few glaciological studies (Cogley and Jung-Rothenhausler, 2004) provided a careful quantitative evaluation of interpolation accuracy over glaciated area using ground data. At present we do not know how sensitive glacial mass balance calculations are to the type of interpolation method used to create glacier elevation surfaces.
Here we assess the suitability of readily available SRTM and ASTER datasets for mass-balance studies at a remote mountain area in the Peruvian Andes. The SRTM DEM’s released by USGS and the ASTER DEM’s generated at the EROS Data Center constitute information in the public domain available to all glaciologists. Our objectives are: 1) to evaluate the suitability of various interpolation techniques to construct DEM’s for glaciological studies; 2) to assess elevation differences between DEM’s from satellite data and DEM from topographic data, 3) to identify the spatial distribution of these errors with respect to topographic characteristics (elevation, slope and aspect) and 4) to ultimately distinguish a glacier signal from multi-temporal DEM’s.
33 III.2 Study area
Our study area is Nevado Coropuna (Figure 3.1), situated in the volcanic Cordillera
Ampato in Southern Peru (15°24' -15°51'S latitude and 71°51' - 73°00'W longitude). The
Ampato range consists of 93 glaciers, with an estimated average glacier thickness of ~35 m and a total glaciated area of 146.73 km2 based on 1962 aerial photography (Ames et al.,
1989). Nevado Coropuna is the highest peak in Cordillera Ampato, and it extends over 25 km with elevation ranging from ~ 4,600 m at the base (Lake Pallacocha) to over 6,400 m at the main summit, with gentle sloping lava flows and glaciated terrain. There have been no comprehensive field measurements of glacial properties on Nevado Coropuna until recently.
The only glacier mapping was carried out by Ames et al.(1989) who reported a glaciated area
Figure 3.1 Location map of the study area. The ASTER Level 1A image from July 2001 is
draped over a shaded relief map of the topographic DEM. Also shown are GPS transects
obtained in the field.
34 of 82.6 km2 for Nevado Coropuna based on planimetric analysis of 1962 aerial photography.
This manuscript fills this gap and complements two ice-core-drilling expeditions conducted on Coropuna in 2003 by l’Institut de Recherche pour le Développement, France (GREAT
ICE project) and by the Ice Core Paleoclimatology Research Group at the Byrd Polar
Research Center, Ohio State University. Results from these paleoclimatic studies will provide an isotopic record for the region, which can help understand the climatic variability in the region and assess present-day glacier fluctuations in the area.
III.3 MethodS
III.3.1. Field data collection
In June and August 2003, GPS points were obtained using a handheld Garmin Etrex
GPS unit on both rock and glaciated areas (Figure 3.1). The light and portable Garmin was preferred over the bigger Trimble Pathfinder unit. However, a disadvantage of the Garmin unit is that its accuracy is lower because it cannot be differentially corrected. The accuracy of the Garmin unit was tested at Green Lakes Valley Long-Term Ecological Research (LTER) study site in Colorado (Ackerman et al., 2001). Horizontal errors of measurements taken with the Garmin unit were within 3.9 m of the differentially corrected data obtained with the
Trimble Pathfinder (Ackerman et al., 2001). As a rule of thumb, we consider the vertical accuracy of the Garmin unit to be about 1.5 times bigger than horizontal errors (< 10 m). The
GPS elevations referenced to the WGS84 ellipsoid were converted to orthometric heights
(heights above the EGM96 geoid) by subtracting geoid heights calculated based on latitude and longitude at the UNAVCO facility in Boulder, Colorado (Rapp, 1996).
35 III.3.2. Construction of the DEM from topographic data
Two 1:50,000 topographic maps, constructed from 1955 aerial photography by
Instituto Geográfico Nacional (IGN) of Peru were needed to cover the study site. The maps used Provisional South American Datum of 1956 for Peru, and elevations referenced to mean sea level. The maps were scanned and georeferenced based on UTM grids with a positional accuracy (calculated as root mean square error in the X and Y coordinates) of 3.9 m. Contour lines with 25 m spacing were digitized on screen, and assigned the corresponding elevation values read from the topographic map. Additional GIS layers digitized from the topographic maps included lakes, streams, spot heights and the 1955 snowline.
We examined common interpolation routines to create continuous data from the digitized contours using the Arc Info 8.2 software package. These included: Inverse Distance
Weighted (IDW), Splines (TOPOGRID) and Triangulated Irregular Network (TINs). The
IDW method estimates the Z value of an unknown point based on a distance-weighted average of elevation points within a neighborhood (Burrough and McDonnell, 1998). Spline techniques use a piece-wise function to fit a curve through all the data points. The
TOPOGRID algorithm available in Arc Info is a more sophisticated spline technique (thin plate spline) that fits a smoothing surface through the data points to minimize artifacts
(excessively high or low spurious values) (Burrough and McDonnell, 1998). TOPOGRID interpolates directly from the contour lines by determining areas of steepest slope and generating terrain morphology. Ancillary hydrologic data (streams and lakes) are used to define drainage based on the ANUDEM algorithm for hydrologic modeling described by
Hutchinson (1988). Triangulated Irregular Network (TIN) data structures are terrain models represented by continuous triangular facets that store elevation at irregularly spaced nodes
(Burrough and McDonnell, 1998).
36 III.3.3. SRTM and ASTER datasets
The Shuttle Radar Topography Mission (SRTM) acquired data in February 2000, from which digital elevation models are created (Rabus et al., 2003). Preliminary elevation datasets with 90 m resolution (‘SRTM-3’) were recently released for South America. An elevation dataset (1-degree latitude by 1-degree longitude) was obtained for the study area, referenced to UTM projection and resampled to 30 m resolution. Elevations are in meters referenced to the EGS84 EGM96 geoid (USGS, 2003).
Two ASTER scenes acquired by along-track stereo channel (3), with nadir (3n) and aft-viewing (3b) orientations (Kääb, 2002) were obtained from the Land Processes DAAC at
EROS Data Center: one Level 1 B scene from October 2000 and one Level 1A scene from
July 2001. The cloud-free 2001 ASTER scene, shown in Figure 3.1 was used to extract a
DEM using automated stereo auto-correlation procedures with PCI Geomatica software package at the USGS EROS Data Center. Ground control points (GCP’s) were required to obtain an “absolute” ASTER DEM (where locations are fitted to UTM coordinate system and elevations referenced to mean sea level) (Hirano et al., 2003). Eight GCP’s were digitized from the topographic maps at river crossings, spot elevations and road intersections and identified on the 3n and 3b bands of the Aster image, following the protocol of Hirano et al.(2003) and Khalsa et al.(2003). The ASTER-derived DEM has 30 m post spacing.
Various ASTER scenes were evaluated to find the one that provided the best glacial extent, based on image contrast and minimal snow coverage. We used the ASTER L1B scene obtained in October 2000 (end of the dry season) to delimitate the glacier outline. An unsupervised ISODATA clustering (Aniya et al., 1996; Paul, 2001; Tou and Gonzales, 1974) was performed using ASTER VNIR channels (1, 2 and 3) to delimitate the ice extent. The resulting raster image was converted to polygon coverage and visually checked to ensure correspondence with glaciated areas on the color composite image.
37 III.3.4. DEM validation and comparison
We focused on evaluating errors in the vertical coordinate (Z), estimated as root mean square errors (RMSEz). The Z coordinate is the only unconstrained value, since X and
Y coordinates were used to locate corresponding grid cells in all DEM's. Moreover, elevation is the coordinate of interest in glaciological applications because changes in surface elevation over time can be an indicator of mass balance changes (Etzelmüller, 2000). The RMSEz of the various interpolated methods was calculated with respect to evenly distributed spot elevations digitized from topographic maps. The RMSEz of the SRTM and ASTER DEM’s was calculated with respect to GPS points from non-glaciated areas. Visualization techniques
(shaded relief maps, elevation contours and slope maps) were used to examine representation of topography in each DEM.
Difference maps were constructed by subtracting the DEM from topographic data from both the ASTER and SRTM-derived DEM's on a cell-by-cell basis. We examined correlations between vertical differences and topographic characteristics (elevation, slope and aspect). Errors on the non-glaciated areas (bias) were quantified by performing trend surface analyses on the difference maps. After removing the bias, we examined the remaining elevation differences on glaciated areas to distinguish a glacier signal, using histograms, summary statistics and color maps of the height differences.
III.4 Results and discussion
III.4.1. Topographic interpolation results
An examination of the RMSEz values DEM’s derived from topographic data (Table
1) shows that no interpolation method performed perfectly. The vertical accuracy of the DEM created with the TOPOGRID algorithm is 14.7 m based on 61 spot elevations. The other interpolation methods yielded RMSEz values that ranged from 21 to 24 m, which is 30 – 40
38 % greater than the TOPOGRID algorithm (Table 3.1). A one-way analysis of variance
(ANOVA) test showed that there was a significant difference in the DEM’s created with various interpolation methods at the 0.1 significance level (p-value = 0.07). The RMSEz of
14.7 m using the TOPOGRID algorithm is only slightly bigger than half of the contour interval (25 m), which is considered an acceptable vertical accuracy for DEM’s derived from topographic maps (Cogley and Jung-Rothenhausler, 2004).
Table 3.1 Evaluation of different interpolation methods used to construct DEM’s from the
topographic maps
Interpolation Other RMSEz (m) Terracing method artifacts
TOPOGRID 14.7 Light Cones
IDW 24.2 Severe No
TIN 22.0 No Triangulation
SPLINE 21.1 Moderate No
All DEM’s constructed from contours lines display ‘terracing’ effects due to denser sampling along the contour lines, because points closer to the contour lines are interpolated using the same elevation values (Figure 3.3) (Burrough and McDonnell, 1998).
39 Figure 3.2 The effect of interpolation methods on representation of terrain topography at a
subsection of the study area. a) original contour lines, with 25 m interval; b) shaded relief
map of the DEM created with the IDW method; c) shaded relief map of the TIN data
structure.
The terracing effect is most visible on flat surfaces where contours are spread apart, and is most severe when using local interpolator methods such as IDW (Figure 3.2b).
Etzelmüller and Björnsson (2000) used the IDW interpolation method to create a continuous surface of glacier thickness. This systematic ‘terracing’ artifact was also reported in glacier studies that used tension splines for surface representation (e.g. Mennis and Fountain, 2001).
Terracing is known to affect subsequent calculations of topographic characteristics (slope, aspect and profile curvature) (Kamp et al., 2003; Wilson and Gallant, 2000) which are of interest glaciological applications. For our study area, the DEM created with the TOPOGRID algorithm yielded the smoothest surface, suitable for the gentle sloping terrain. Minimal terracing is detected in this DEM, and appears as spikes on the histogram of elevation values
(Figure 3.3a).
40 Other glaciological studies (e.g. Gratton et al., 1990) preferred TIN’s because of their advantage of capturing complex terrain variations, accurately representing ridges and streams, and reducing data redundancy on flat terrain (Burrough and McDonnell, 1998). For
Coropuna, the TIN structure introduced noticeable triangular discretization on the gentle- sloping lava flows and smooth glacier surface (Figure 3.2c), which we considered unacceptable. Our results show that there are large differences in the glacier surface representation by DEM’s as a function of interpolation algorithm used. Based on minimizing both RMSEz and artifacts (terracing and triangulation), we chose the DEM created with the
TOPOGRID algorithm (denoted as ‘TOPO DEM’) as the 1955 reference elevation dataset.
III.4.2. Accuracy assessment for the SRTM and ASTER - derived DEM's
SRTM elevations and ASTER elevations were checked against 64 GPS points from non-glaciated terrain. We focused on non-glaciated terrain to validate the DEM’s because elevation changes might have occurred on the glacier between 1955 and 2000/2001. We present the elevation differences of the DEM’s as RMSEz relative to the GPS points, and not the absolute vertical accuracy. The RMSEz of the SRTM DEM relative to the GPS points was 23.4 m. Since the vertical accuracy of GPS points is less than 10 m (cf. Section 3.1), this gives an absolute vertical accuracy of 23.4 m ± 10m. The specified SRTM accuracy standard is ± 16 m for global coverage (Rabus et al., 2003). The frequency histogram of SRTM- derived elevations (Figure 3.3b) indicates a normal distribution, with a few anomalously high values (spikes). Water bodies are not well defined and appear "noisy" or rough due to low radar backscatter (USGS, 2003). Height differences between SRTM elevations and GPS elevations tend to be randomly distributed (Figure 3.4a), with SRTM elevations being both lower and higher than the GPS elevations.
41 Figure 3.3 Histograms of elevation values for the three DEM's analyzed. a) DEM created
with TOPOGRID algorithm (TOPO DEM), b) SRTM DEM and c) ASTER DEM.
Spiked on elevation histograms represent elevation values of the contour lines used in the
interpolation (‘terracing’ effect).
42 The comparison of elevations from the ASTER DEM with GPS points shows both a large RMSEz and a vertical bias. The RMSEz of the ASTER DEM relative to the GPS points is 61.2 m. This corresponds to an absolute vertical accuracy of 61.2 m ± 10m, which is bigger than the specified accuracy of 7 – 50 m for absolute ASTER DEM’s (Lang and Welch, 1999).
ASTER-derived elevations are consistently higher than the GPS points (Figure 3.4b). The magnitude of the vertical differences between ASTER and GPS increases with elevation along the GPS transects, pointing towards some vertical bias, with the ASTER DEM being too high in the upper terrain. Such vertical errors were reported in other studies. Kääb (2002a) found an overall accuracy of ± 60 m RMSEz when an absolute ASTER DEM was compared to a reference DEM on complex mountain terrain. However, better accuracy (±18 m RMSEz) was found at a section of moderate topography in the same study (Kääb, 2002a), suggesting that errors in the ASTER DEM’s tend to increase in rugged mountain terrain.
Some errors come from some noise due to ‘banding’ in the ASTER L1A scene, visible in the
DEM as spikes on the elevation histogram (Figure 3.3c). Comparison of contours derived from the ASTER DEM with contours from topographic data revealed positional offsets as much as 300 m in X and 200 m in Y. These offsets were not consistent throughout the study area, pointing towards a distortion in the ASTER DEM in the X and Y coordinates. Such horizontal offsets have been observed at other study areas with high relief (Dwyer, LP DAAC
Project Scientist, USGS EROS Data Center, personal communication). To correct the offsets, the ASTER DEM was fitted to the georeferenced ASTER L1A scene using a second order polynomial transformation based on 15 control points identified at lakes, stream crossings, and noticeable terrain features such as ridges.
Additional validations of the SRTM and ASTER-derived DEM’s were performed by comparing their elevations with the reference 1955 DEM from topographic data on a cell-by- cell basis. Subtracting the reference DEM from the SRTM DEM yielded:
43 Figure 3.4 Plots of height differences between the DEM’s from satellite data and GPS
elevation along GPS transects. a) SRTM elevations minus GPS elevations; b) ASTER
elevations minus GPS elevations.
44 Mean difference = –1.8 m
Standard deviation = 15.7 m
The range of differences was –113 m/ +121 m, with the largest differences occurring on non- glaciates areas, at valley bottoms and sharp moraine ridges, as well as on a few flat areas where interpolation from contour lines produced erroneous values (either spikes or sinks).
Subtracting the reference DEM from the ASTER DEM yielded:
Mean difference = 80.5 m
Standard deviation = 28.1 m.
The range of differences was –86 m/ +500 m. The large positive differences of +500 m come from ‘spikes’ of erroneous elevation values that occur on glaciated summits and sharp ridges. Such spikes of up to 500 m in ASTER DEM’s were noted at other areas on sharp peaks (Kääb et al., in press). Elevation "waves" with about 200-300 m amplitude on low contrast glacier areas were reported by the USGS EROS Data Center (Wessels, U.S.
Geological Survey Alaska Science Center, personal communication). These elevation errors are due to either steep northern slopes which are missed by the back-looking band 3b (Kääb et al., in press) or low contrast in the ASTER scenes over snow and ice, causing failure in the image-matching process (Toutin, 2002a). While distortions in the ASTER DEM may be due to lack of adequate ground control points, it was shown by other studies (Kääb et al., in press) that introducing more GCP’s does not significantly remove this effect.
Height differences between the SRTM/ASTER DEM’s and the reference DEM shown on color maps in Fig. 5a and 6a, indicate a systematic bias in both difference maps, with positive residuals at non-glaciated areas in the north and negative residuals in lower valleys in the south. Similar biases in residuals were noted in other comparisons of ASTER DEM’s with
IGN topographic data (e.g. Vignon et al., 2003) as well as comparison of ASTER DEM’s with photogrammetric-derived DEM’s (Kääb et al., in press). We modeled the variation of residuals over the non-glaciated area for the two datasets by fitting various polynomial
45 surfaces through the residuals. The best fit in terms of R-squared was obtained by a first order polynomial, suggesting that the magnitude of the residuals increases linearly with location.
The polynomial is derived by multiple regression on X and Y coordinates and is an inclined surface of the form:
f {(x,y)} = a0 + a1 X + a2 Y, where a0 is the intercept and a1 and a2 are the slopes (Burrough and McDonnell, 1998).
The bias of elevation differences between the SRTM DEM and the reference DEM is a tilted surface, oriented towards the NNE (5.42 degrees), which dips at a rate of 1.9 m vertical per 1 km Northing, and has a range of -21 m to 20 m across the DEM. For the ASTER minus
TOPO DEM, the bias is a tilted surface oriented towards the NNW (349.30 degrees), which dips at a rate of 2 m vertical per 1 km northing, and has a range of 47 m to 96 m across the
DEM. Once the trend was removed, the mean statistics for the elevation differences on non- glaciated areas yielded:
SRTM DEM minus TOPO DEM (Fig. 8a): mean = 0, std. deviation = 9.5 m
ASTER DEM minus TOPO DEM’s (Fig. 8b): mean = 0, std. deviation = 20.6 m
46 Figure 3.5 Color maps of height differences between the SRTM and topographic elevations
draped over the topographic DEM. a) before trend removal, the SE-NW spatial trend of
elevation differences is visible. b) after trend removal. Also shown is the glacier extent
obtained by classification of the October 2000 L1B ASTER scene.
47 Figure 3.6 Color maps of elevation differences between ASTER DEM and the TOPO DEM’s.
a) 2001 ASTER DEM (before trend removal) minus 1955 TOPO DEM, with the NNW-
SSW spatial trend visible; b) 2001 ASTER DEM (after trend removal) minus the 1955
TOPO DEM.
48 The histograms of elevation differences on non-glaciated areas are close to normally distributed (Figure 3.7 a-b). Large standard deviations on non-glaciated areas point to artifacts in the DEM’s (high or low values) and they do not affect subsequent analysis of the glaciated areas.
Figure 3.7 Frequency histograms of elevation differences between the DEM’s on non-
glaciated vs. glaciated areas, after the trend removal. a) SRTM DEM minus TOPO
DEM; b) ASTER DEM minus TOPO DEM.
49 After the trend removal, we examined the impact of slope and aspect on the vertical differences between the DEM’s. Correlations with slope yielded a coefficient (Pearson’s r) of
0.54 for SRTM minus reference DEM and 0.69 for ASTER minus the reference DEM. On slopes less than 50 degrees, there is almost no difference in SRTM-derived elevations and the topographic DEM, but ASTER elevations are consistently higher than the reference DEM.
The plots of vertical differences with respect to slope (Figure 3.8) show that elevation errors in the SRTM and ASTER DEM’s tend to increase with slope. For the SRTM DEM, elevation errors of up to –25/+50 m occur on ~ 60 degree slopes. For the ASTER DEM, elevation errors greater than 100 m occur on steep slopes (60 – 77 degrees) and correspond to the
‘spike’ artifacts in the ASTER DEM. These results are consistent with trends noted in
ASTER-derived DEM’s. For instance, Kääb (2002) and Kääb et al. (in press) found large vertical differences in the ASTER DEM’s compared to reference DEM’s from topographic data at steep slopes.
Figure 3.8 Correlation of vertical differences between the DEM’s with slope a) SRTM DEM
minus TOPO DEM; b) ASTER DEM minus TOPO DEM. Largest vertical differences
occur on steepest slopes.
50 An examination of the mean vertical differences (SRTM – reference DEM) plotted in aspect classes (Figure 3.9a) shows that the SRTM elevations do not depend on aspect. The mean elevation differences between SRTM and reference DEM range from –5 m on S-facing slopes to 1 m on NNE-facing slopes. The ASTER DEM displays bigger mean vertical differences ranging from –3m on W-facing slopes to +19 m on E-facing aspects (Figure
3.9b). We expected bigger elevation errors on N-facing slopes, which are normally missed by the back-looking band 3b (Kääb et al., in press). For instance, decreased vertical accuracy of
ASTER DEM’s on northern slopes was reported by Kääb (2002a) and Kääb et al. (in press).
Our results suggest that mean errors tend to occur on aspects between 0 and 180 degrees, not only on N-facing slopes. This suggests that the large vertical errors in the ASTER DEM cannot be entirely explained by the back-looking ASTER sensor.
Figure 3.9 Radar charts of vertical differences between the DEM’s with respect to aspect. a)
SRTM DEM minus TOPO DEM; b) ASTER DEM minus TOPO DEM.
51 III.4.3. Glacier signal from the SRTM DEM
We checked elevations from the SRTM and ASTER DEM’s against GPS points on glaciated areas. The difference between SRTM elevations and 56 GPS elevations acquired on the glaciated area yielded a RMSEz of 27 m, which is ~ 4 m bigger than on non-glaciated areas. SRTM elevations are both higher and lower than the GPS points on the glacier (Figure
3.4a). The RMSEz of ASTER elevations with respect to GPS points on the glacier was 98 m, and residuals increase with altitude (Figure 3.4b). This large vertical bias implies that the
ASTER DEM, created with a limited number of GCP’s, is not suitable for glaciological interpretation.
To quantify the glacier signal from the SRTM DEM, we examined the mean elevation differences (SRTM minus topographic DEM) on glaciated areas after removing the
N-S spatial trend. Once the trend was removed, the elevation differences on the glaciated area are negatively skewed (Figure 3.7a), with a mean of - 5 m and a standard deviation of 15.8 m
(Table 3.2). We consider the remaining mean difference of –5 m ± 15.8 m as a signal of glacier thinning (95% confidence interval).
Table 3.2 Statistics summary of map differences for the glaciated areas vs. non-glaciated
areas after trend removal.
SRTM - TOPO DEM (m) ASTER - TOPO DEM (m) Statistics Glaciated Non-glaciated Glaciated Non-glaciated
Mean - 5 0.0 28.5 0.0 Std. 15.8 9.5 26 20.5 Deviation
Cell-by-cell comparison of elevations from SRTM data with topographic DEM within the glaciated area (Figure 3.5b) show ablation at the toes of the glaciers (- 25 m to –75 m surface lowering) along with an apparent thickening at the summits (25 – 50 m). Average
52 height differences between the SRTM and topographic DEM on the glaciated area increase with altitude, with a correlation coefficient (Pearson’s r) of 0.62 (Figure 3.10). Similar comparisons of ASTER data to topographic data in Cordillera Blanca (Peru) revealed a loss of altitude of as much as –23 m at the glacier toes (Vignon et al., 2003). Ablation in the lower parts of the glaciers (via ice melting and sublimation) was also observed from field measurements in other tropical glaciers (Kaser, 1999a; Kaser et al., 1990).
Figure 3.10 Correlation between vertical differences between the DEM’s and altitude on the
glaciated area, after trend removal. a) SRTM DEM minus TOPO DEM; b) ASTER DEM
minus TOPO DEM. Vertical differences increase with elevation on the glaciated areas.
Thickening in the accumulation zone of the glaciers is a less common trend and was observed in some mountain glaciers around the world during the 1961 – 1997 time period
(Dyurgerov and Meier, 2000). However, in the climatic context of Coropuna, an average thickening of 25 – 50 m firn in 50 years, or 0.5 m / year would represent a mean increase in precipitation of 250 – 500 mm water equivalent. At the col, our results agree with field data
53 from the ice core drilling of June 2003, which also point to an accumulation of 0.5 - 1 m firn
/year in the col (Ginot, Laboratoire de Glaciologie et Geophysique de l’Environnement,
Grenoble, personal communication). However, at the summits, the increase of 0.5 - 1m firn / year from the DEM comparison is 2 - 4 times bigger than ice core results (.26 m firn/ year)
(Ginot, Laboratoire de Glaciologie et Geophysique de l’Environnement, personal communication).
In the upper part of the glaciated area the noise is too high to be able to infer a positive change of 25 – 50 m in altitude. The elevation differences between ASTER elevations and reference DEM on glaciated area are positively skewed (Figure 8b), with a mean of 28.5 m and a standard deviation of 26 m. Comparison of GPS points with corresponding ASTER elevations on glaciated areas (Figure 3.4b) shows that the ASTER
DEM is too high on upper glaciated terrain, with a RMSEz error of 98.3 m with respect to
GPS points. However, an examination of cell-by-cell differences between the ASTER DEM and the reference DEM (Figure 3.6b) shows negative residuals (–50 to –25 m) in the ablation areas of the southern glaciers. The surface lowering is consistent with results from SRTM
DEM. We could not quantify the glacier signal from the ASTER DEM due to the altitudinal bias and the large elevation ‘spikes’ on the glacier surface, which are affecting the mean statistics.
III.4.4. Changes in glacier extent and volume
`For 1962, Ames et al.(1989) reported a glaciated area of 82.6 km2 on Nevado
Coropuna based on planimetric analysis of 1962 aerial photography. Based on the ASTER
L1B scene from October 2000, we obtained a glacier area of 60.8 km2, which represents a loss of 26% in glacier area from 1962 to 2000. Our results are consistent with glacier retreat observed in Cordillera Ampato during the last few decades. Ames et al.(1989) reported a total glaciated area of 146.7 km2 based on 1962 aerial photography. The total glaciated area in the
54 Ampato range was estimated to be 105 km2 based on more recent Landsat TM imagery
(Morales-Arnao, 1999). This corresponds to a retreat of 27 % in Cordillera Ampato from
1962 to the end of the 20th century. Glacial retreat in Peru has also been observed in other areas, especially in Cordillera Blanca (Georges, 2004; Hasternath and Ames, 1995; Kaser et al., 1990).
III.5 Conclusions and further applications
Using DEM's derived from topographic and satellite data at different steps in time holds potential for glacier analysis. DEM's constructed from old topographic data still constitute a valid elevation dataset for comparison with more recent DEM’s for glaciology purposes. Here we created a DEM from 1:50,000 topographic data for Nevado Coropuna and tested different interpolation techniques. Based on RMSEz and visual analysis, the TOPOGRID algorithm was found to be superior to the other techniques examined, with the smallest RMSEz error and least interpolation artifacts.
Error analyses were performed on all DEM’s to characterize the bias present in the various DEM's. We removed the bias on non-glaciated areas to distinguish a glacier signal.
We found that the SRTM dataset with a RMSEz of 23.4 m ± 10 m was suitable for glaciological applications after some calibration. In areas of rugged terrain, the SRTM resolution (90 m) is not sufficient to represent the topography. Comparison of the 2000
SRTM DEM with the DEM from 1955 topographic data points to an average thinning of ~ 5 m on the glacier surface, with a significant lowering of the glacier surface at the glacier toes and an apparent accumulation on the summits. We attribute the changes at the summits to large errors in the data at higher elevations with steeper slopes. While lowering of glacier surface at the toes was visible in the ASTER DEM, large elevation errors and altitudinal bias did not allow quantifying a glacier signal from the ASTER data.
55 The analysis of multi-temporal DEM’s to quantification of glacier changes is extremely sensitive to the quality and spatial resolution of the DEM’s. For studies of glacier change using DEM's, we found that several steps were necessary: referencing all elevation data to the same vertical datum; evaluation of DEM differences in non-glaciated areas; testing the DEM's against field GPS survey points; visualization techniques such as shaded relief, slope angle and comparison of contours; removing the biases in the elevation datasets.
Future steps to minimize large error differences occurring in DEM’s derived from satellite data include filtering and smoothing of the DEM's (Hirano et al., 2003; Toutin, 2001;
2002a). These techniques may help to better distinguish and quantify glacier surface changes.
56 "' /3$1%.41
".-2314"3(.-.% 2/ 3( +# 3 ! 2$%.1'(&' +3(34#$ 1"'$.+.&8(-3'$ -#$2
IV.1 Introduction
Archeological databases are mostly focused on inferring the spatial patterning of human behavior from the artifacts present at a site (Waters, 1992) and their chemical properties (e.g. shape, size, provenance and weight). These patterns constitute the systemic context of a site (Schiffer, 1987). In traditional archeology, there was little emphasis on the geographical location of the artifacts (Wheatley and Gillings, 2002), on the natural site formation processes and on paleo-environmental conditions. Recent approaches in geoarcheology aim at reconstructing the spatio-temporal context of the sites using geologic approaches (Waters, 1992). These techniques require accurate spatial information, which is often missing from the archeological record (Wheatley and Gillings, 2002). In the last two decades, with advances in spatial techniques such as Geographic Information Systems (GIS) and Global Positioning System (GPS), it has become easier to construct archeological databases with a spatial component (X, Y and Z coordinates) (Wheatley and Gillings, 2002).
Such spatial databases have not been developed yet for archeological sites in remote areas of the Andes Mountains. Artifacts dating from the Inca period (14th-15th century)
(Schobinger, 1967a) were investigated in the last few decades using traditional methods (i.e. grid systems). These data were not designed for digital databases, thus limiting the abilities for mining their wealth of information. The archeological record is currently maintained by two local institutions: the Center for High Altitude Archeology (CIADAM) in Argentina and the Catholic University of Santa Maria (UCSM) in Peru.
57 There are some problems with the existing archeological record in the Andes. First, there is a concern with data quality. The first explorations were made by local communities or climbers with little archeological expertise and little means to record exact locations
(geographic coordinates). Archeological sketches were not drawn to scale, making it difficult to georeference the sites. Secondly, there are internal inconsistencies within the same archeological survey record due to the use of various survey instruments. For instance, the
GPS measurements reported by Ceruti (1999) are consistently higher than elevations reported from topographic maps or altimeters. Third, the horizontal and vertical accuracy of these instruments, and the projection and datum of the measurements are not specified.
Furthermore, multiple investigations of the same site by different survey teams yield contradictory information. For instance, elevations of archeological sites reported in one survey often differ from other surveys. Positional errors are expected because of the difficulty of working at high altitudes in extreme climatic conditions, and the lack of logistical support
(Ceruti, 1999a). Such positional inaccuracies are a common problem in archeological surveys
(Kvamme, 1988a). Lastly, access to these data is limited, and only a limited number of publications in Spanish are available locally. Access to topographic maps that were used in the original surveys is limited. Verifying the location of the archeological sites in the field poses a challenge due to logistical constraints, the large extent of the study area (2,312,467 km2) and the remoteness of the sites.
This chapter presents the evaluation of the archeological survey data from available publications and the design of a spatial database. The purpose of constructing this database is to: (a) collate existing information on high-altitude archeological attributes into one database that can be accessed, queried and displayed (b) convert the various formats used to report artifact data into one common format; (d) locate the archeological sites on a digital elevation model (DEM) so that they can be assigned environmental attributes. This effort complements
58 the descriptive archeological record that resulted from decades of explorations conducted by
Andean archeologists.
IV.2 Construction of the spatial database
IV.2.1 Digital elevation model
The study area covers 2,312,467 km2 of the Inca Empire, from Southern Peru to
Chile (Figure 4.1). Elevation data with 3-arc seconds (~ 90 m) resolution acquired by the
Shuttle Radar Topography Mission (SRTM) in 2000 (Rabus et al., 2003) was obtained from the USGS Data Center. These data were mosaicked into a digital elevation model (DEM) covering the study area.
Figure 4.1 Distribution of surveyed archeological sites compiled from published surveys. The
sites are georeferenced and mapped on a shaded relief map of the SRTM digital
elevation model.
59 IV.2.2 Archeological data sources
Archeological data were acquired from four publications in Spanish, in form of descriptions of the archeological findings found on each mountain surveyed or reported by local people and climbers. The sites surveyed and their location attributes were stored as shown in Table 4.1. Beorchia Nigris (1985a) compiled information on 117 mountains surveyed prior to 1985 (Appendix 2). Latitude and longitude lists of 108 summits were available; however, there was no geographical coordinate information for multiple sites on each of the 117 mountains. Site elevations, measured with altimeters or read from topographic maps were also reported. In addition, archeological sketches were also provided for a limited number of peaks surveyed. An example of the sketches used for volcano
Chilques shown in Figure 4.2.
Figure 4.2 An example of the archeological sketch used to document the location of the
archeological ruins on Volcano Chilques in Chile. Reproduced from Beorchia Nigris
(1985).
60 Table 4.1 Summary of the archeological survey data compiled from the available publications.
Ceruti (1999) Beorchia Nigris (1985) and Reinhard and Ceruti Estimated Total (2000) sites Sites number source number source added deleted Mountains surveyed With summit coordinates and Topographic Topographic 108 29 34 elevations maps maps Missing summit coordinates 9 0 0 Total 117 29 34 9 171 Archeological sites With lat/long coordinates and Topographic GPS 59 47 54 26 elevations maps altimeters 134 Missing lat/long coordinates 80 9 62 Total 139 56 54 9 196
61 Ceruti (1999a; 1999b), and Reinhard and Ceruti (2000) provided information on 47 additional mountains from Northwestern Argentina (Salta, Jujuy, Catamarca and Tucuman provinces) (Appendix 3). These two publications contained latitude, longitude and elevation of the summits from topographic maps, as well as GPS measurements of multiple sites found on each mountain. Orientation measured using the grid method was also reported. The accuracy of the GPS units and the precision used to store the coordinates were not quoted. In addition with coordinate lists, detailed descriptions of the artifact type, frequency, dimension, form, weight and cultural affiliation were available for each site from these publications.
Descriptive labels about the topography at each site (such as slope and morphology) were also available.
IV.2.3 Database design
The database consists of three components: (a) the spatial component, (b) the archeological (non-spatial) attribute component and (c) the metadata component. The spatial component stores the geographical location of each archeological site (X, Y and Z coordinates). The archeological attribute stores information about the physio-chemical properties of the artifacts such as the shape, size, weight, date and provenance in addition to descriptive labels about topography. The archeological attribute component is dynamic, and is continually updated to incorporate new or revised information. The artifact attribute and the spatial component are developed as separate databases but linked by an unique identifier assigned to each site. The identifier is composed of the initials of the investigator, the year of the publication and the number of the record (e.g. BN1985_001). The metadata component stores information about the projection systems and datums used for each site, the source of the data (topographic maps or GPS measurements), and the RMS errors encountered in each dataset.
62 IV.2.4 Georeferencing the archeological sites
The archeological data were manipulated using the ArcGIS software package and the ArcInfo command line. The latitude and longitude coordinates were converted from deg.min.sec to decimal degrees and stored in digital format with precision to the nearest meter. The coordinate lists were used as input to generate point layers of the archeological sites, for each of the two archeological datasets. Data sources were treated separately due to different datums and projection systems used in the two sets of surveys. Topographic maps in
South America use Transverse Mercator projections with either Provisional South American
Datum of 1956 (PSAD 1956) or South American Datum of 1969 (SAD 1859). Maps produced in Argentina use the Gauss-Kruger projection system, with Campo Inchauspe (CAI) datum. GPS measurements use the WGS 1984 Datum. The sites were initially georeferenced using the geographic coordinate system, and then projected to UTM WGS 1984 to match the elevation dataset. This information was stored in the metadata component of the database
(Table 4.2). The two archeological datasets were then merged and stored in both vector format (point coverage) and raster format (grid with a cell size of 90 m).
Table 4.2 Metadata component storing information about projection systems, datums and
accuracy of the data.
Vertical accuracy Data source Original coordinate system Datum (m)
SRTM DEM UTM WGS84 16*
Beorchia Nigris (1985) Argentina: Geographic CAI 978.11** Other countries: Geographic PSAD1956 Ceruti (1999a;1999b) Reinhard and Ceruti Geographic CAI 557.68** (2000) Resulting dataset UTM WGS84 88.63**
* absolute vertical accuracy on a global scale (Rabus et al.., 2003) ** RMSEz error with respect to SRTM DEM elevations
63 IV.2.5 Archeological attributes
A number of attribute data were stored in a database, separately from the spatial layer (Table 4.3). The location information included country and province where the site was found, the mountain range and the name of the site. Ten artifact types (ruins, enclosures for burning wood, stone piles, ceramics/pottery, wood, mummies, statues, Inca trail fragments, animal bones and icchu grass) were stored in the database as binary variables
(presence/absence). Seven additional attributes pertained to the physical properties of the constructions (length, width, height and radius, shape (i.e. rectangular, circular), frequency
(number of constructions), area covered, topography (flat or rugged) and visibility ( “low” to
“high”). Frequency analysis (number of occurrences of a type of artifact on a particular site type), correlations and associations (co-occurrence of artifacts) were perform for the binary artifacts.
Table 4.3 Attribute information recorded from archeological surveys.
Attribute Data storage Location information Mountain String Site name String Latitude, longitude Numerical (1 m precision) Elevation of the mountain Numerical Base of the mountain Numerical Country String Province String Archeological information Site type Categorical Ruins: Form/function Descriptive Ruins: shape Descriptive Ruin size (L,l,h,Ø) Numeric Ruin frequency Numerical Artifact type Binary (1 = present, 0 = absent) Area Numeric Topography Descriptive Visibility Descriptive Orientation String
64 The sites were categorized into four types: base camp, logistic, ceremonial or lookout points based on labels assigned in published archeological surveys. When these were not available, the site categories were defined by a combination of the types of artifact present, the location on the mountain, the elevation of the site and the type of inferred activity following the approach of Ceruti (1997) (Table 4.4). For instance, the presence of human sacrifices, offerings and ceremonial platforms indicated a ceremonial site. Tambos (cabins, or places of rest), found along with shelters and a large number of domestic ceramics indicated a logistical site.
Table 4.4 Classification of archeological sites based on their function, the location on the
mountain and the typical artifacts associated with these. Table translated and reproduced
from Ceruti (1997).
Site type Relative Location Constructions Function Tambo Base, slopes Plaza, enclosures Logistic Ceremonial (?) Cemetery High and low constructions, Ceremonial Base, slopes, pre-summit platforms, walls Shelter Slopes, pre-summit Enclosures, walls Logistic
Habitation Enclosures, walls, low Base, slopes, pre-summit Base camp, logistic complex constructions Lookout Slopes, pre-summit Platform, walls Ceremonial point Platforms, terraces, ramps, Sanctuary Pre-summit, summit high and low constructions, Ceremonial stone piles, circular constructions and plazas
65 Elevation values were normalized by calculating the percentage from the base of the mountain where a site was located, according to the formula:
SITE _ ELEV − BASE REL _ ELEV = ⋅100 SUMMIT _ ELEV − BASE
The base of the mountain was taken from the archeological survey data and was often defined as the vertical drop between the summits and the altiplano (high plains, 4000 – 4500 m above sea level found in Argentina, Bolivia, Chile and Peru). When the base elevation was not specified, mountains in the same geographic region were assigned the same base elevation. The site type was encoded as a categorical variable ranging from 1 to 4, where 1 =
“ceremonial”, 2 = “logistical”, 3 = “base camp” and 4 = “lookout point”. A “nodata” value was assigned to sites that did not contain any artifacts.
IV.2.6 Locational data accuracy
Root mean square errors of the elevations (RMSEz) were calculated to evaluate the accuracy of the elevation measurements provided in the archeological survey data. The accuracy of elevations from topographic maps was evaluated separately from GPS measurements. At each site, elevations corresponding to each site were extracted from the
SRTM DEM using GRID overlay functions. The SRTM elevations were then compared to site elevations reported in the archeological surveys using the root mean square error formula:
n 2 ƒ(Z observed − Z predcted ) RMSE = i=1 , z n
where the observed values (Zobserved) represent site elevations reported in the archeological surveys, the predicted values (Zpredicted) represent corresponding site elevations extracted from the DEM, and n = 108 elevations from topographic maps or n = 47 GPS elevations.
66 Incorrect locations were adjusted based on ancillary data. These data included site elevation, information about where on the mountain the site was located (e.g. its orientation, location on the saddles, on a ridge etc.). I mapped the archeological sites on the DEM and visually estimated the positional errors. I compared the elevations reported in the archeological surveys with the elevations of the DEM at the same point; I then used archeological sketches of the summits (e.g. Figure 4.2) along with descriptions from the surveys to estimate a “correct” location. Each point feature in the ArcInfo coverage was then shifted to the estimated location using the on-screen digitizing tool. I also identified sites that originally did not have coordinates, and digitized them as points. I then evaluated the relative accuracy of the estimates by calculating the RMSEz between the estimated elevations and the elevations extracted from the SRTM DEM.
IV.3 Results and discussion
IV.3.1 Archeological attributes
From the data sources, I compiled information on 146 mountains surveyed up to present. Some of the mountains had multiple archeological sites, resulting in a total number of 205 sites surveyed, distributed as shown in Table 4.5. Of these 205 sites, 17 sites lacked significant artifact evidence. Since these were considered present-day sanctuaries, and their use as Inca worship sites was doubtful (Beorchia Nigris, 1985a), they were classified as
“non-sites”. In addition, two sites were located outside of the area covered by the DEM, and were eliminated. As a result, the archeological database consisted of 196 archeological sites with artifacts.
67 Table 4.5 Distribution of surveyed archeological sites by country.
Country Frequency Percent
Argentina 105 51.2 Chile 64 31.2 Peru 30 14.6 Bolivia 5 2.4 Ecuador 1 .5
Total 205 100 %
Based on an altitudinal criteria specified by Beorchia Nigris (1985), 50 % of the surveyed mountains included in the archeological database were classified as “high” mountains (5,000 – 6,700 m), 13.9 % were “medium” mountains (3,000 – 5,000 m) and 4.7
% were “low” mountains (less than 3,000 m). The remaining percentage represents sites that lacked elevation measurements. More than half of the surveyed sites (52.7 %) were classified as “ceremonial” in basis of their location, elevation, and the type of constructions (Table 4.6).
Table 4.6 Frequency of site types based on their function. The site type is defined by a
combination of site altitude, the type of artifacts and their location on the mountain.
Site type Frequency Percent Ceremonial 108 52.7 Logistical 53 25.9 Base camp 25 12.2 Lookout point 5 2.4 NODATA 14 6.8 Total 205 100.0
Logistical sites occurred in a percentage of 25.9%, and base camp sites were less frequent (12.2%); only 5 sites had a lookout point. Ruins were the most common artifacts,
68 encountered at 83.4 % of the sites (Table 4.7). Wood was also frequently encountered (48.8
% of the sites had large quantities of wood). It was estimated that up to 4,000 kg of wood were deposited on volcano Licancabur (5,921 m) in Chile, and quantities of 200 – 300 kg were estimated on other peaks such as Chilques (5,778 m) in Chile and Tambillos (5,747 m) in Argentina (Beorchia Nigris, 1985a).
Table 4.7 Frequency of artifacts found at the surveyed archeological sites.
Type of artifact absent present
Ruins 16.6% 83.4% Fogon 92.7% 7.3% Apacheta 86.1% 13.9% Inca trail 90.0% 10.0% Statues 93.7% 6.3% Ceramics 84.6% 15.4% Mummy 93.0% 7.0% Bones 91.5% 8.5% Icchu 96.6% 3.4% Wood 51.2% 48.8%
IV.3.2 Correlation analysis
The frequency of constructions and site types was correlated with attributes pertaining to their geographical location (latitude) and location on the mountain (site elevation and the vertical drop). Results of this correlation analysis indicate a negative correlation between latitude and the elevation of the sites, with site elevation decreasing as latitude increases (Table 4.8). The correlation was significant at the 0.05 level (95% confidence interval). This suggests that at higher latitudes south, archeological sites may be encountered at lower elevations. The ruin frequency is also negatively correlated with site
69 elevation and relative elevation (the percent up the mountain where a site occurs). This suggests that as elevation increases, the number of archeological construction decreases, i.e. the highest frequency of constructions, is found at lower elevations. This correlation was not significant at the 0.05 level (Table 4.8). Site type was negatively correlated with relative elevation, with a significant coefficient (.-452) at the 0.01 level (99% confidence interval).
This suggests that the higher up on the mountain, the more ceremonial sites (site type = “1”) are encountered.
Table 4.8 Correlations between archeological attributes and their location.
Pearsons’ r Attributes correlated 2-tailed Site elevation and -.201* Latitude Ruin frequency and -.032 site elevation Ruin frequency with -.058 relative elevation Site type with relative -.452** elevation * Correlation is significant at the 0.05 level (2-tailed). ** Correlation is significant at the 0.01 level (2-tailed).
IV.3.3 Association analysis
A distance similarity matrix was created to examine which artifacts tend to occur together. I chose the Jaccard method, which flags positive associations or co-occurrence of artifacts. In the Jaccard method, conjoint absence (0,0) is ignored and conjoint presence (1,1) is more important. The values in Table 4.9 represent the ratio of positive matches (1,1) to the sum of positive matches and mismatches (1,0) and (0,1). Higher values suggest more positive
70 matches; lower values suggest less positive matches. Wood and ruins have the highest score of conjoint presence (score = .475) (Table 4.9).
Wood was found in large quantities on the summits with ceremonial sites, as documented by Beorchia Nigris (1985). Mummies tend to be found together with ceramics
(score = .250) and statues (score = .174). Ceramics seem to occur in conjunction with animal bones, icchu, wood and ruins. These results confirm survey data where these last five types of artifacts were generally associated with logistical sites. The icchu grass and wood were used for heating (Beorchia Nigris, 1985a). The presence of llama bones suggesta the use of animals to transport artifacts to the logistic sites (Reinhard and Ceruti, 2000). This analysis is useful to identify artifacts occurring together and defining types of sites when a-priori knowledge is not available.
IV.3.4 Accuracy assessment of location information
Of the 117 mountains surveyed prior to 1985 and reported by Beorchia Nigris
(1985), 108 had summit coordinates reported from topographic maps. Only 55 sites were located at the summit. In addition, 29 mountains were surveyed in the last decade and reported by Ceruti (1999a; 1999b) and Reinhard and Ceruti (2000). These had GPS and topographic measurements for the summits as well as for multiple sites on the same mountain.
First, the relative elevation accuracy of the summit sites was calculated as root mean square error of reported elevations with respect to DEM elevations. The RMSEz for the topographic measurements was 944.46 m. The RMSEz for the GPS measurements was
557.68 m (Table 4.2). The frequency histogram of elevation differences shows vertical differences of up to 3,750 m between SRTM elevations and topographic measurements and up to 2,442 m between SRTM and GPS measurements (Figure 4.3).
71
Table 4.9 Similarity matrix for associations of artifacts, where each artifact is stored as binary. The matrix is derived using the Jaccard (“similarity
ratio”) method. The highest match ratios are highlighted in shades of grey with the darkest shades for the highest scores.
RUIN FOGON APACHETA INCA_RD STATUE CERAMIC MUMMY BONES ICCHU WOOD RUIN 1.000 FOGON 0.082 1.000 APACHETA 0.139 0.075 1.000 INCA_RD 0.099 0.061 0.021 1.000 STATUES 0.077 0.120 0.051 0.031 1.000 CERAMICS 0.156 0.070 0.054 0.020 0.189 1.000 MUMMY 0.076 0.036 0.024 0.000 0.174 0.250 1.000 BONES 0.094 0.067 0.023 0.000 0.000 0.116 0.033 1.000 ICCHU 0.041 0.100 0.029 0.038 0.053 0.118 0.050 0.091 1.000 WOOD 0.475 0.130 0.156 0.054 0.088 0.152 0.057 0.106 0.061 1.000
73 Figure 4.3 Frequency distribution of differences between SRTM- derived elevations and
elevations reported by surveys.
These large elevation differences with respect to the reported GPS elevations were caused by positional shifts. Visually, I detected inaccurately located sites by mapping them on a shaded relief map of the SRTM DEM using the coordinates reported by the surveys. For example, the site located on the summit of Carachipampa, according to the archeological survey data, was placed in the high plains at 3,009 m (on the DEM) instead of on the mountain (4,500 m reported in the survey) (Figure 4.4).
.
74 Figure 4.4 Horizontal displacement of archeological sites mapped using published coordinate
information. The sites are mapped on a shaded relief of the SRTM digital elevation
model.
There were large vertical differences between GPS measurements reported and SRTM elevations (Figure 4.5). The RMSEz of the GPS measurements was 666.33 m. The location of these archeological sites was \adjusted by at least a few pixels as described in the methods section. A total of 26 sites large positional errors and were eliminated. In addition, I estimated the locations for 54 additional sites described in the archeological survey, which were missing location coordinates initially. This resulted in 134 georeferenced archeological sites.
Once the locations were corrected, the RMSEz of the new (adjusted) locations was 83.68 m.
Up to 27 m of this error is attributed to differences between the SRTM DEM and GPS measurements, based on the validation presented in Chapter 3. The remaining error of up to
55.68 m is attributed to errors inherent in the source data (topographic maps), transcription errors that might have occurred when transferring coordinates from paper to the digital database, and uncertainties introduced during the re-location of the archeological sites on the
75 DEM. The frequency distribution of the vertical differences in elevations between the new locations and the DEM elevations (Figure 4.5) shows differences closer to 0.
Figure 4.5 Comparison of vertical differences between SRTM DEM with survey elevations
before and after the location corrections. The RMSEz represents the combined error
between GPS measurements, topographic measurements and SRTM.
IV.4 Conclusions
The existing archeological data provides a wealth of information about high altitude archeological sites located in the Andes Mountains surveyed to the present day. Compiling this dataset into a spatial database has been a challenging process due to: 1) inconsistencies in the data sources; 2) the descriptive nature of these sources and 3) the lack of access to original survey data. Positional corrections performed using ancillary data (a digital elevation model, archeological sketches and descriptions provided in the surveys) resulted in a smaller but improved dataset. This database does not consist of the full archeological record found at
76 high altitudes in the Andes, and can be improved for greater detail and accuracy with new and revised archeological data.
Developing and querying a spatial database that includes both archeological sites and environmental attributes helps testing hypotheses about site location. By constructing a spatial database, I can then locate the archeological sites on a digital elevation model (DEM) and extract topographic variables characteristic to each site. Furthermore, querying the database allows for quantitative ways to categorize various types of archeological sites.
Quantitative techniques include clusters analysis, and associations. The full potential of this spatial database has not been exhausted.
77 "' /3$1%(5$
$-5(1.-,$-3 + -#2."( +% "3.12 22."( 3$#6 (3''(&' +3(34#$ 1"'$.+.&(" +2(3$2(-3'$ -#$2
V.1 Introduction
New approaches in archeology rely on the use of spatial technology such as
Geographic Information Systems (GIS) to locate archeological sites in unsurveyed areas. A common approach is to map locations displaying environmental patterns similar to surveyed archeological sites. This inductive approach relies on two assumptions: (1) that the choice of a site location was determined by a set of suitable environmental conditions (Rose and
Altschul, 1988) and (2) that present day environmental conditions are used as proxies for past conditions at the site location.
In remote areas of the Andes Mountains, archeological surveys have been limited to a small group of researchers due to remote locations, extreme climate and lack of logistical support (Ceruti, 1999a). Predictive models can assist field work by providing maps of archeologically sensitive areas, and thus narrowing the area to be surveyed on each mountain.
Obtaining the necessary environmental and social data necessary to build such predictive models posed a challenge in the past because data relevant to terrain characteristics are not available from archeological surveys (Altschul and Nagle, 1988). Socio-cultural factors are equally difficult to obtain due to the lack of a writing system in pre-Columbian cultures, including the Incas.
Another problem encountered in previous archeological surveys is that most location data reported (i.e. geographic position, altitude and orientation) only pertain to “sites”.
Location information about sites surveyed but where no artifacts were found (“non-sites”) is not provided. This poses a problem for statistical methods commonly used in predictive
78 modeling such as logistic regression, which require data about both “sites” and “non-sites”
(Kvamme, 1988a). To overcome this problem, it was suggested that the “non-sites” dataset can be obtained using GIS by extracting random samples from the study area (referred to as the “background environment”) and assigning them environmental variables (Kvamme,
1988b; Kvamme and Kohler, 1988).
GIS methods have been successfully used in a number of archeological predictive models to collect environmental data and to derive a dataset for “non-site” locations
(Kvamme and Kohler, 1988). For example, Warren and Asch (2000) used GIS to derive a set of secondary data layers such as distance to streams, topographic relief and soil characteristics, as well as a set of “non-sites” extracted from the background environment.
Dixon and Manley (2001; 2002) used GIS to map variables such as distance from mammal ranges, mineral licks, prehistoric trails, glaciers and persistent snow cover to identify archeologically sensitive areas in Alaska’s Wrangell St. Elias National Park. Other studies such as Brandt et al. (1992), Duncan and Beckman (2000) and Kuiper and Westcott (1999) also exemplify the use of GIS for obtaining environmental data layers used in predictive modeling.
This chapter focuses on the initial choice and development of key factors believed to influence site location. At a local scale, it has been hypothesized that certain environmental factors determined the location of Inca archeological complexes on a mountain (Ceruti,
1997). These factors are: water availability, topography, geomorphology, climate, cryonival processes, accessibility, vegetation cover, resource availability and visibility (Ceruti, 1997).
These hypotheses constitute a framework to guide the data collection process presented in this chapter.
Three types of data were compiled and integrated into GIS: archeological data, environmental data (topography, climate, hydrology, vegetation and landscape) and socio- cultural data (Inca roads, empire quarters and mining sites). These data were used to develop
79 topographic factors and proximity measures. The contrasts between site and non-site locations with respect to these environmental and social factors were examined using descriptive statistics. Furthermore, this allowed to were used to explore the relationship between and, and to test the hypotheses about site location, described in Chapter 2.
V.2 Methods
V.2.1 Archeological data layers
The process of compiling and evaluating a spatial archeological database for high altitude archeology was discussed in detail in the previous chapter. The archeological dataset consisted of 151 sites with geographical location and archeological attributes. Grid cells that contained artifacts were denoted as a “site”, whereas the cells where no artifacts were present were denoted as a “non-site”. To obtain the non-site data, background sites were extracted randomly from the whole extent of the study area using a stratified random sampling method suggested by Kvamme (1988b) and Kvamme and Kohler (1988). There is no established method regarding how many non-sites vs. sites should be used. Some studies (Kvamme,
1988b) used the same number of samples for sites and non-sites, while others (Warren and
Asch, 2000) used larger number of nonsite cells (5,208) to contrast against a set of 265 site cells. Kvamme (1990b) noted that the random sampling might not capture the true range of environmental variation in the study area; however, it has the advantage of reducing spatial auto-correlation in non-site samples and it has been used with good results in other studies
(Kvamme, 1988b). I chose a larger number of cells for this study to cover the large extent of the study area. I also checked that none of the non-sites overlapped with the actual archeological sites. I used the SAMPLE command in the GRID module, and obtained a set of
456 non-sites cells, which were used subsequently as a control group in statistic analysis.
80 V.2.2 Environmental data layers
Primary GIS data layers obtained from various sources were used to create secondary layers of terrain topography and proximity factors in ArcInfo version 8.2 software package.
The resulting variables were assigned to both the 151 archeological “sites” compiled from pervious surveys, and to the 456 “non-sites” extracted randomly from the study area. These attributes were stored as independent variables needed in the subsequent stage of the project
(logistical regression). Terrain factors extracted data extracted from raster datasets were stored as continuous variables; other environmental factors extracted from vector datasets were stored as categorical variables. Some variables were stored as binary (presence or absence).
a. Topographic data
Terrain factors, especially slope and aspect, are commonly used in location studies and are most frequently derived as continuous surfaces from a digital elevation model
(DEM) (Kvamme, 1988b). Elevation data with 3 arc-second (90 m resolution) derived with interferommetric techniques from the Shuttle Radar Topography Mission (SRTM) were obtained from the USGS EROS Data Center. The elevation grids were imported into ArcInfo, georeferenced and mosaiced to create a DEM covering the study area.
Slope and aspect were calculated from the elevation dataset. The slope surface was further incorporated in a cost-surface analysis to derive the least cost paths to all the archeological sites (Chapter 6). Aspect values were transformed using the cosine function since 0° and 360° both express the same direction (Kvamme, 1988b), resulting in values ranging from –1 to 1.
Terrain roughness index
A cell-by-cell measure of terrain roughness was derived by calculating the variance of elevation values in a 4 x 4 window (270 m by 270 m) with a focal function, following
81 methods used by Kvamme (1988b). High variances reflect more rugged terrain, while low variances reflect smoother, leveled terrain (Kvamme, 1988b). The algorithm was tested for
Nevado Coropuna in Peruvian Andes as shown in Figure 5.1 below.
Figure 5.1 Terrain roughness index illustrated for a subset of the study area (Nevado
Coropuna, Peruvian Andes). Darker shades around ridges and summits represent higher
variation of elevations (more rugged terrain).
Shelter index
A simple measure of exposure introduced by Jochim (1976) and described by
Kvamme (Kvamme, 1988b) was used to compute a shelter index based on the shape of surrounding terrain. A cylinder with a radius r and height h is placed on a site, and the volume of air it contains gives an indication of the exposure (Figure 5.2). This concept was implemented in ArcInfo using a focalsum function and the equation provided by Kvamme
(Kvamme, 1988b):
82 2 ≈ 9 ’ π r ∆ ÷ Volume = ∆12 h + 8 ⋅ E 0 − ƒ E i ÷ 12 « i =1 ◊
where Eo is the site elevation and Ei are the nine surrounding elevations at a specified radius.
I used a height of 100 m above the site and a radius of 180 m. A high volume indicates low shelter (i.e.ridges). A low volume indicates high shelter (i.e. valleys or depressions).
Figure 5.2 Shelter measure derived from cylinder volumes, reproduced from Kvamme
(1988b).
b. Present-day climate and hydrologic data
Gridded annual temperature and precipitation datasets in ASCII format were obtained from the R-Hydronet project (Legates and Willmott, 1990a; 1990b). The matrices of values were imported into ArcInfo to generate grids at 1,000 m resolution. Glacier boundaries in vector format were obtained from the National Snow and Ice Data Center (NSIDC), Boulder.
These boundaries were compiled from the Digital Chart of the World and the World Glacier
Monitoring Service's World Glacier Inventory (Raup et al., 2000). Snow cover, permafrost
83 line and solar radiation were obtained from ArcAtlas in vector format (ESRI, 1999).
Permafrost was stored as a binary layer (absence/presence). Solar radiation values were obtained from observations collected at the World Center of Radiation Data in St. Petersburg,
Russia (Morozova and Berlyant, 1996).
c. Vegetation and landscape data
Vegetation cover and present landscape data were obtained from ArcAtlas.
Vegetation classification was derived at the Russian geobotanic school (Safronova and
Khramtsov, 1996). The vegetation layer was reclassified based on vegetation zone codes to obtain possible wood sources: (1) dry, partly sclerophyllous forest and open woodland; (2) mountain vegetation of dry forest and open woodland and (3) evergreen and partly deciduous forests, open woodland and shrubland. Polygons contained in these zones were reclassified as wood sources.
d. Social data
The social data collected for the model development included Inca Roads, Empire quarters (suyus) and mining sites. A paper map of major Inca road segments was obtained from the Inca Road Project conducted by the Institute of Andean Research (Hyslop, 1984).
Inca Road segments were manually digitized as vector data layers and were subsequently used to determine the accessibility of each archeological site based on the cost-distance from the roads (Chapter 6). The Inca Empire quarters were digitized and georeferenced using published paper maps. A dataset containing modern mining sites in South America was obtained from the ArcAtlas database and stored as point coverage.
V.2.3 Proximity factors
Two factors commonly used in location analysis are accessibility (e.g., distance to nearest road) and availability of natural resources such as fuel, wood or lithic resources
84 (Kvamme, 1988b).Proximity to Inca Roads, mining sites, wood sources, glaciers and volcanoes was calculated on a cell-by-cell basis using linear (Euclidean) distance. Cost distance from Inca Roads was calculated based on slope, terrain characteristics and weight carried. These procedures are described in detail in the following chapter.
V.2.4 Statistical tests
Descriptive and univariate statistics were used to examine contrasts between sites and non-sites with respect to environmental and social variables. Contrasts between the four archeological site types were also examined. The Man-Whitney U test was used for categorical data (e.g. landscape and vegetation classes). The Kolmogorov-Smirnov test was used for continuous variables derived from the DEM.
V.3 Results and discussion
The compilation of various data sources resulted in a total of 33 primary and secondary layers stored in GIS with the same extent and spatial reference (Figure 5.3). These variables were assigned to both archeological sites and non-sites. Of these variables, 21 represented environmental layers, 10 social layers and 2 archeological layers (sites and non- sites). Topographic factors were stored as continuous variables, and the rest as categorical
(Table 5.1).
85 Figure 5.3 Primary and secondary GIS data layers.
86 Table 5.1 Organization of the GIS data layers, the resulting variables and non-parametric tests
for contrasts between sites, non-sites and site types.
Non- Description Source Organization Code Variable type parametric test Primary GIS layers Kolmogorov- Elevation SRTM Grid ELEV Continuous Smirnov** (1,3)(2,3)** R-HYDRO Kolmogorov- Temperature Grid TEMP Continuous NET Smirnov** Kolmogorov- R-HYDRO Precipitation Grid PRECIP Continuous Smirnov** NET (1,3)* Digital Glaciers Chart of the Polygon GLCR Binary - world Mining sites ArcAtlas Point MINES - -
Volcanoes ArcAtlas Point VOLC Binary -
Lakes ArcAtlas Polygon LAKES Categorical - Mann-Whitney Vegetation ArcAtlas Polygon VEG Categorical U* Mann-Whitney Landscape ArcAtlas Polygon LAND Categorical U Mann-Whitney Snow cover ArcAtlas Polygon SNOW Categorical U** Mann-Whitney Solar radiation ArcAtlas Polygon RAD Categorical U** Permafrost ArcAtlas Polygon PERMFRST Binary Mann-Whitney Morphology ArcAtlas Polygon MORPH Categorical U* Morpho- Mann-Whitney ArcAtlas Polygon MSCULPT Categorical sculpture U**
Countries ArcAtlas Polygon COUNTRY - -
Empire quarters Digitized Polygon SUYUS - -
Empire extent Calculated Polygon EXTENT - -
Inca Roads Digitized Line INCA_RD - - Archeological CIADAM Grid ARCHEO Categorical - sites Non- sites Sampled Grid NONSITES Categorical -
87 Non- Description Source Organization Code Variable type parametric test Secondary GIS layers Kolmogorov- Slope Elevation data Grid SLOPE Continuous Smirnov** (1,3)** Kolmogorov- Aspect Elevation data Grid ASPECT Continuous Smirnov** Kolmogorov- Roughness Elevation data Grid VARIANCE Continuous Smirnov** Kolmogorov- Shelter Elevation data Grid SHELTER Continuous Smirnov** (1,2,3)** Linear distance Distance DIST_RDS_ Kolmogorov- Grid Continuous to Inca Roads buffer LIN Smirnov** Linear distance COST_DIS Kolmogorov- DEM Grid Continuous to Inca Roads T Smirnov** Kolmogorov- Hiking time DEM Grid TIME Continuous Smirnov** Energy Kolmogorov- DEM Grid ENERGY Continuous expenditure Smirnov** Distance to Kolmogorov- Distance DIST_GLC Grid Continuous Smirnov** glaciers buffer RS (1,2)(1,3)* Distance to Distance DIST_VOL Kolmogorov- Grid Continuous volcanoes buffer C Smirnov** Distance to Distance DIST_MIN Kolmogorov- Grid Continuous mines buffer ES Smirnov**
Distance to Distance DIST_WOO Kolmogorov- Grid Continuous woodlands buffer D Smirnov*
Notes for the non-parametric tests:
Asteriscs mark significant statistical differences between sites and non-sites with respect to the environmental and social variables.
* p<0.05
** p<0.001
88 V.3.1 Archeological site distribution
Archeological sites in the Andes Mountains occur at latitudes from –9 degrees to –33 degrees south, with the highest frequency of archeological sites is observed in the southernmost part of the Inca Empire (Collasuyu) (Figure 5.4).
Figure 5.4 Distribution of archeological sites in the four quarters of the Inca Empire. Also
shown is the Inca Roads system.
89 Sites occur at a wide range of elevations from 905 m to 6,712 m with a mean of 5178 m.
Their frequency increases with elevation gain from the base of the mountain, as shown in
Figure 5.5. Elevations of the sites were statistically different than elevations of the non-sites based on the Kolmogorov-Smirnov test (p-value = 0.000). This supports the hypothesis that higher elevations were chosen for worship, either because these sites were closer to the Sun
(Mostny, 1957) or because higher mountains were more sacred.
Figure 5.5 Frequency histogram of site elevations with respect to relative elevation (percent
elevation from base of the mountain in bins of 10 %).
Archeological sites tend to be located in high mountain environments and plateaus, in the tropical belt (Table 5.2). Only one site occurs in the subequatorial climatic belt (in
Ecuador). More than 75% of sites occur above the limit of permafrost (Figure 5.6 and Table
5.3), which supports the hypothesis that the Incas chose climatic conditions suitable for preservation of human remains (Corte, 2001; Corte and Zhijui, 1985). In the Andes, permafrost occurs in large areas in the Central Andes, at high elevations in the tropical Andes
90 and at lower elevations in the Patagonian Andes (Kadomtseva, 1996b). Cold temperatures are an ideal medium for preservation of human remains by lyophilization5 in glacial environments.
Table 5.2 General landscape and climatic characteristics of archeological site locations.
Relief Count Percent High mountains 119 75.32 Middle mountains 2 1.27 Low mountains 4 2.53 Plateaus and tablelands 33 20.89
Climatic belt Subequatorial 1 0.63 Tropical 133 84.18 Subtropical 19 12.03 Intrazonal 5 3.16
Of the surveyed sites, 13.3 % were located on glaciers (Table 5.3). One example is the mummy from Cerro El Plomo, found under a 30 cm-thick ice cap (Corte, 2001).
Descriptive statistics of climate variables (Table 5.4) suggest that archeological sites tend to occur in a somewhat smaller climatic range than non-sites. The mean annual temperature at site locations is 14.61 degrees C, and the mean annual precipitation of 270.7 mm is about 3 times lower than mean precipitation at non-sites (737.4 mm) (Table 5.4). The warm mean temperature and low precipitation at site locations support the hypothesis that preservation of
5 Lyophilization refers to the process of freeze-drying that can occur naturally under extreme conditions of low temperature and pressure. Water in the frozen body is lost via sublimation of ice while maintaining the physio-chemical properties of the body tissue.
91 human sacrifices was also possible by desiccation in snow-free environments under arid conditions (Corte 2001).
Table 5.3 Percentage of archeological sites that occurs on glaciers, volcanoes and permafrost,
where “0” indicates number of sites absent and “1” indicates number of sites present.
Location 0 1 Missing
Permafrost 22.2 77.6 1.3 Glaciated 84.2 13.13 2.5 Volcanic 77.2 19.6 3.2
Table 5.4 Climatic characteristics for archeological sites vs. non-sites
Std. Climate factor Min Max Mean Dev. Temperature(°C) Sites 7.8 18.8 14.6 2.2 Nonsites 6.00 26.8 16.0 4.7 Precipitation (mm) Sites .1 836.7 270.7 261.3 Nonsites .1 4303.1 737.4 758.03 Solar radiation (J/km2/day) Sites 16 24 21.5 2.0 Nonsites 14 24 20.3 2.4 Snow cover duration Sites 0 365 26 69 Nonsites 0 365 13 45
Site locations receive more solar radiation and have shorter snow duration than non- sites. Radiation and snow values were statistically different for sites and non-sites at the
0.001 and 0.01 level respectively (p-values: 0.000 and 0.002). However, the climate variables only indicate general patterns since data were not available on a pixel-by-pixel basis. Other
92 studies calculated solar radiation based on elevation, slope and aspect (Duncan and Beckman,
2000). More complex algorithms (TOPORAD) allow to calculate radiation from a digital elevation model in snow covered mountainous terrain (Dozier, 1980). Due to the large extent of the study area, high-resolution snow cover was not available. Therefore, climate variables are interpreted cautiously here and are not expected to influence the location model due to the low resolution of the data.
Five types of vegetation specific to highland arid areas are encountered at archeological site locations (Table 5.5). High-mountain grassland (puna) and shrublands
(subparamo) are most commonly encountered. These vegetation patterns suggest that archeological sites tend to occur in dry highland areas and agree with elevation and climatic data.
Table 5.5 Vegetation zones and types occuring at archeological sites.
Vegetation zone Vegetation type Percent
Dry, partly Deciduous forest and sclerophyllous forest 1.3 sclerophyllous shrubland and open woodland Mountain vegetation of Mountain and high dry forest and open mountain xerophytic 37.3 woodland shrubland (“subparamo”) Desert (subequatorial) Succulent shrubs 1.3
Mountain vegetation of High mountain grassland 54.4 desert and semi-desert (“puna”)
Halophytic vegetation Halophytic vegetation 1.3
Desert (subtropical) Shrubs 4.4 Analysis results indicate significant contrasts between site locations and non- sites with respect to all topographic variables examined (elevation, slope, aspect, shelter and terrain roughness) (p-values<0.001, 99% confidence interval) (Table 5.1). Both archeological
93 sites and non-sites tend to occur on gentle slopes (around 10 degrees) (Table 5.6). However, the frequency histogram of sites (Figure 5.6) shows a positively skewed distribution, with the number of sites decreasing exponentially on slopes steeper than 15 degrees.
Table 5.6 Topographic characteristics for archeological sites vs. non-sites.
Terrain factor Min Max Mean Std. Dev. Elevation (m) Sites 905 6712 5178.2 871.8 Nonsites 23 5328 2416 1644.9 Slope (degrees) Sites .32 38.0 9.9 7.6 Nonsites .11 42.9 10.3 10.2 Aspect (degrees) Sites .00 353.0 129.8 98.3 Nonsites .00 359.5 174.7 106.0 Roughness index Sites .44 3281.1 406.7 545 Nonsites 0 4696 419 700.1 Shelter index Sites 23 229 177.0 27.1 Nonsites 0 179 81.0 56
Figure 5.6 Frequency distribution of archeological sites with respect to slope.
94
Site locations display a strong tendency towards north and north-east facing slopes
(Figure 5.7a). This tendency in the data reflects a significant pattern since non-sites display a more even distribution in all aspect classes, with a predilection for south-east facing slopes
(Figure 5.7b). Many of these high altitude archeological sites have openings to the East, the cardinal direction of the rising sun (Beorchia Nigris, 1985). The orientation of the sites to the east is consistent with alignments of archeological constructions with the position of the sun during solstices, which was common in Inca astronomy (Aveni, 1990; 1996; Bauer and
Dearborn, 1995; Urton, 1978; Zuidema, 1977). These results suggest that the human sacrifice rituals may have been related to the movement of the sun in the sky on the East-West axis. a. b.
Figure 5.7 Charts of frequency distributions of sites and non-sites with respect to aspect. a)
archeological sites; b) random background sites. Frequencies are plotted in bins of 22.5
degrees.
95 There was no strong preference for terrain roughness in archeological site, with similar mean roughness for sites and non-sites (406.7 and 418.9 respectively) (Table 5.6).
However, the differences in means between the two datasets are statistically signifiacant based on Kolmogorov-Smirnov test (p-value < 0.001). The frequency of sites decreases as terrain roughness increases (Figure 5.8) indicating a preference for smoother terrain.
Figure 5.8 Frequency distribution of archeological sites with respect to terrain roughness
index. Roughness index values are calculated based on variance of elevations around a
site. High variances indicated rougher terrain.
Archeological sites tend to occur on more exposed terrain such as ridges and edges of plateau tops (shelter index of 171.02) while non-sites tend to be more sheltered (81.06 for random sites) (Table 5.6). Site frequency increases as shelter index increases (Figure 5.9).
These results suggest that exposed areas with high visibility might have been preferred instead of sheltered sites of low visibility. Shelter can be calculated more accurately on a pixel by pixel basis from a DEM using methods developed by Winstral et al (2002) if the
96 wind direction is available. Due to the extent of the large extent of the study area, localized wind patterns were not available for this method.
Figure 5.9 Frequency distribution of archeological sites with respect to shelter index. High
shelter values suggest more exposure.
V.3.2 Proximity factors
Archeological sites tend to be located closer to Inca roads, mines, woodlands and volcanoes than random background sites, as indicated by smaller mean distance values (Table
4.7). The frequency of archeological sites decreases as distance from Inca Roads increases
(Figure 5.10a), indicating a preference for more accessible locations. The frequency of sites also decreases with distance from wood sources (Figure 5.10b), suggesting that locations closer to wood sources might have been preferred. However, distance to wood was not significantly different for sites and non-sites at the 0.001 level. All the other proximity measures were significant (p<0.001), indicating distinctive pattern in distance to roads, mines and volcanoes resources between sites and non-sites. Preliminarly, the expansion hypothesis can be accepted based on this analysis. Archeological sites tend to be located further from glaciers than non-sites, which does not support the hypothesis that proximity to water sources is important.
97 Table 5.7 Proximity factors for archeological sites vs. non-sites
Proximity factor Min Max Mean Std. Dev. Distance to Inca Roads (km) Sites .20 96.1 33.2 25 Nonsites N/a N/a N/a N/a Distance to volcanoes (km) Sites .0 818.1 116.2 113 Nonsites 6.5 985.0 335.1 235.7 Distance to mines (km) Sites 1.6 119.5 55.6 28.3 Nonsites 2.8 415.7 80.1 63 Distance to glaciers (km) Sites .0 611 268.5 194.8 Nonsites .0 722.7 196 159.5 Distance to woodlands (km) Sites .0 129.5 32.5 38.7 Nonsites .0 697.0 58.4 121 a. b.
Figure 5.10 Frequency histograms of proximity factors for archeological sites a) linear
distance to Inca roads; b) linear distance to wood sources.
98 V.3.3 Univariate statistics for multiple site types
Non-parametric tests for continuous variables (Kolmogorov Smirnov) show that base
camp sites are significantly different than ceremonial and logistical sites in terms of
elevation (p-value < 0.001). This suggests that logistical sites may be also found at higher
elevations. Site types 1 and 3 were also different with respect to slope (p<0.05), shelter
index (p < 0.001), precipitation (p<0.05) and distance to glaciers (p<0.008). Sites 2 and 3
were different with respect to shelter index (p<0.001). Site types 1 and 2 differed with
respect to terrain roughness at the 90% confidence interval (p<0.1). There were no
significant contrast between lookout points and any of the other types. It can be inferred
that ceremonial sites occur on different slopes, elevation, shelter index, precipitation and
distance to glacier than base camp sites. Furthermore, logistical sites and ceremonial site
tend to occur on different topography.
From non-parametric test Man-Whitney, only vegetation type and zone were
statistically significant for contrasting site types 1 and 2 (p<0.05). The other sites type
categories did not differ significantly with respect to landscape and vegetation patterns (p
>0.05). Based on these results, I identified two possible binary classifications: (1) sites vs.
non-sites and (2) ceremonial sites vs. logistic and base camp sites.
V.4 Conclusions
Descriptive statistics show that archeological sites tend to be located at higher elevations, on exposed ridges or plateaus with gentle slopes, oriented preferentially towards north and north-east. Many of the sites are found above the limit of permafrost. Although there are such characteristics at sacred mountains at lower latitudes in northern Peru and
Ecuador, no ruins or mummies have been found on high glaciated peaks in the northern parts of the Inca Empire (Beorchia Nigris, 1985a; Reinhard, 1985a). One explanation of the
99 predominance of sites found in Argentina is that the province of Salta in Argentina has been most intensely surveyed by members of the Center for High Altitude Archeology (CIADAM) in the last decade, thus leading to a larger archeological record (Ceruti, 1999a).
This chapter discussed the techniques used to integrate various data sources and to
develop variables initially believed to influence the location of high altitude archeological
sites. Exploratory data analysis using descriptive summaries and univariate statistics
identified contrasts between sites and non-sites as well as clusters of sites based on their
location. I concluded that archeological sites at high altitudes in the Andes occur on
distinctive topographic and precipitation patterns than the whole study area, but on similar
landcover and vegetation type. However, the lack of contrasts in these last variables may be
due to the coarse resolution of these data layers.
A major assumption here is that terrain characteristics derived from present-day digital
elevation model are representative of the landscape in the Inca times. However, as Church et
al. (2000) point out, present day conditions may not reflect paleo-environmental conditions
should be recognized in order to provide a robust model. Post-depositional processes (i.e.
cryosphere changes) were not taken into account here because glacier extents during the
Inca period were not available. Since the exact timing of the rituals is unknown and there are
seasonal variations in glacier cover, additional data would be needed in order to reconstruct
glacier cover.
100 "' /3$12(7
1$".-2314"3(-&(-" "+(,!(-&1.43$242(-&+$ 23".23/ 3' - +82(2
VI.1 Introduction
The effect of topography on time and energy expenditure in mountaineering has been explored in physiological studies (e.g. Minetti, 1995 and Minetti, 2002) and also applied to model accessibility to pre-historic sites in a few studies (Bell et al., 1988; Gorenflo and
Gale, 1990). The applicability of these approaches to modeling least cost access routes that may have been used by past cultures to reach sites located in mountain terrain has only been evaluated sparingly (e.g. Bell and Lock, 2000). Algorithms proposed by previous studies have been tested on individuals carrying loads and traveling on different types of terrain, but it’s unclear whether they are appropriate in inferring mountain paths used by ancient cultures (Minetti, 1995).
Due to the lack of written records on the movement of past societies through landscape, there often no information on routes used from archeological settlements to natural resources (Llobera, 2000). New approaches in archeology such as GIS technologies and digital elevation models allow a reconstruction of possible access routes when these ethno-historical data are not available, and can help understand the behavior of past societies and their movement through landscape (Llobera, 2000). The underlying assumption of these approaches is that distance to natural resources should be minimized (Kvamme, 1988b).
In this chapter I evaluate various approaches used to model cost-distance in mountain terrain and apply them to high-altitude archeology in the Andes Mountains. The objectives of this application are: (1) to find the optimum climbing routes that the Incas would have used to reach ritual sites on sacred mountains; (2) to assess accessibility to high altitude
101 archeological sites based on cost distance from Inca Roads and (3) to test the hypothesis that high-altitude Inca sites were preferentially located in proximity to mountain roads.
The discovery of archeological sites at high altitudes in the Andes Mountains raised a number of questions (Reinhard, 1985a): What routes up the mountain did the Incas use?
How did topographic factors influence the climbing strategies used to transport ecofacts and artifacts to these sites? Recently documented Inca roads (Hyslop, 1984) contain data on fragments of trails observed at lower elevations, but many mountain paths still remain unknown.
Second, was the proximity to cultural features (e.g. Inca roads and settlements) important in determining from which side a mountain was approached? It was hypothesized that some mountains were chosen by the Incas as ritual sites because they were situated in proximity to trans-mountain roads (Schobinger, 1967b). However, was proximity to these features important for all the worship sites, or only for certain ones? This chapter seeks to address these questions by reconstructing possible Inca climbing routes and accessibility from Inca roads.
Here I combine a raster-based GIS with archeological survey data and social data (Inca
Roads). The climbing routes are tested at a subset of the study area where climbing routes have been surveyed with a GPS. The proposed methodology relies on two major assumptions. The first assumption is that the Incas did not have the means or the necessary equipment to climb technical routes up the mountain, as hypothesized by Ceruti (1997).
From the clothing discovered with the mummified Inca children found on top of Andean peaks (Reinhard, 1996), it was found that Incas climbed in sandals (“ushutas”), in often cases in glacierized terrain (Reinhard and Ceruti, 2000).
The second underlying assumption is that site accessibility (understood as both distance to roads and topographic characteristics), was an important factor in determining the route used to transport the ecofacts (wood, icchu grass) and the artifacts (offerings) to the
102 ritual sites (Ceruti, 1997). Ritual activities were performed at extreme altitudes (more than
5,000 m), in severe conditions of low temperatures, difficult terrain and hypoxia (lack of oxygen). While high altitude native people have developed a set of particular adaptations to cope with lack of oxygen at high altitudes, such as increased blood pressure and larger lung volume to maximize oxygen uptake (Minetti, 1995), physiological constraints due to heavy load carrying may have constituted a limiting factor on Inca travel strategies. Large loads were transported to the ritual sites (Beorchia Nigris, 1985a; Ceruti, 1997). This is inferred from the large quantities of wood, Inca constructions and human sacrifices found on top of
Andean peaks at altitudes up to 6,700 m. For example, up to 4,000 kg of wood were found on top of volcano Licancabur (5,921 m) in Chile, and quantities of 200 – 300 kg were common on other peaks such as Chilques (5,778 m) in Chile and Tambillos (5,747 m) in
Argentina (Beorchia Nigris, 1985a). Therefore, the need to transport the construction materials and artifacts motivated the choice of an easier route up the mountain. In addition, the large size of some of the ritual constructions suggest an intensive labor process that would have involved numerous transports of the materials involving a large number of people (Beorchia Nigris, 1985a).
VI.1.1 Cost distance analysis: m odeling approaches
Accessibility to natural resources or transportation networks has been recognized as an important factor for location studies (Jochim, 1976). The main assumptions is that distance to economic resources or to access roads tends to be minimized (Kvamme, 1988b).
Accessibility is most commonly quantified in archeology using two fundamental concepts – distance and proximity (Wheatley and Gillings, 2002). Theoretical approaches to modeling movement of pre-histories societies rely mainly on gravity models originating from economic geography (Wheatley and Gillings, 2002). These models viewed the natural or cultural resources as central places, and the frequency of archeological sites decreasing with
103 distance from the resources (Kvamme, 1988b). This approach has been widely incorporated
in archeological predictive models developed so far to calculate distance to natural resources
such as roads, fuel, lithic sources or specific biotic zones (Kvamme, 1988b). However, using
linear distance as a measure of accessibility is rarely adequate when movement involves
travel through complex landscapes, where straight lines are not likely to be followed. A
more realistic approach is to use a different measure such as travel time or energy expended
rather than linear distance to these resources (Kvamme, 1988b). Various approaches to infer
the cost of travel are reviewed here to determine the best approach that can be applied in the
context of high altitude Andean archeology.
VI.1.2 Topographic cost and bioenergetic approaches
Cost-weighted distance functions model the difficulty of crossing one unit of landscape, referred to as “cost”. In mountaineering, cost is most commonly defined as hiking time from a starting point to a destination. Travel time has been used as a measure of proximity in other studies (Bell and Lock, 2000; Bell et al., 1988; Gorenflo and Gale, 1990).
Gorenflo and Gale (1990) inferred settlement patterns in the southern Basin of Mexico based on travel time between settlements. Other approaches define cost as the effort of traveling as a function of slope. A measure of relative energy cost required to ascend various slopes was proposed by Bell and Lock (2000) and used to model pre-historic movement patterns in
Ridgeway, England. Both of the above approaches assume that the cost of travel, expressed as either time or energy, relies heavily on topographic slope in a non-linear manner.
However, they do not take into account other topographic characteristics (surface roughness, snow cover).
The metabolic energy expenditure per unit distance can be determined using bioenergetic approaches, which incorporate the effect of slope, altitude, terrain type and load
104 carrying. This has been addressed in physiology studies (Minetti, 1995; Minetti and Ardigò,
2001; Minetti et al., 2002).
VI.2 Methods
VI.2.1 Study area
Two Andean mountains were chosen for validation of various least-cost path algorithms: Nevados Coropuna (6,377 m) and Pichu Pichu (5,634 m) in Southern Peru
(Figure 6.1), both sacred mountains of the Incas (Beorchia Nigris, 1985a; Reinhard, 1999).
Nevado Coropuna (6,426 m) (Figure 6.2) is situated in the glaciated Cordillera Ampato in
Southern Peru (15° 24' and 15° 51' S and 71° 51' and 73° 00' W), and is covered by a permanent ice cap with an area of approximately 82.6 km2 (Williams Jr. and Ferrigno, 1999).
Late in the dry season (June to August), large penitents form on the surface of the glacier
(Figure 6.3). Coropuna was the main mountain deity in Southern Peru, and it received ritual offerings of animals and human beings, as illustrated by of Guamán Poma de Aylla (1980
[1615]) (Figure 6.2). Nevado Pichu Pichu pertains to Cordillera Volcánica, situated between
16°07' and 16°33' latitude south and 71°12' and 71°33' longitude west. Cordillera Volcánica consists of small groups of ice-covered peaks with a total glacierized area of 15 km2. The highest peak is Nevado Chachani (6,100 m) (Williams Jr. and Ferrigno, 1999). Nevado Pichu
Pichu is not permanently covered by ice.
105 Figure 6.1 Map of the study area showing the three mountains surveyed during field work in
2003. The dots represent GPS points taken along the climbing routes and around the base
of the mountains. The GPS points and glacier boundaries are draped over a shaded relief
map of the SRTM digital elevation model.
106 Figure 6.2 Nevado Coropuna (6,426 m), Peruvian Andes. a) view of the NW and SW
summits seen from the access road from Chuquibamba to Lake Pallacocha. b) offerings
made by local people to worship Coropuna. Sketch reproduced from Guamán Poma de
Ayala (1980 [1615]).
Photos: A. Racoviteanu
Figure 6.3 Penitents forming on the glacier surface of Nevado Coropuna.
107 VI.2.2 Field data collection
GPS points of present-day climbing routes on Nevados Coropuna and Pichu Pichu were obtained during fieldwork in 2003, using a handheld Garmin Etrex GPS unit (Figure
6.1). The light and portable Garmin was preferred over the bigger Trimble Pathfinder unit.
GPS ellipsoidal elevations were converted to orthometric heights referenced to the WGS84
EGM96 global geoid to match the topographic map, using a conversion program provided by the UNAVCO facility in Boulder. The GPS points were used to define the starting points and destination in the least-cost path algorithms, and to validate the results of the analysis.
VI.2.3 Data sources
Two elevation datasets with 3-arc seconds (~ 90 m) resolution, created from radar data acquired by Shuttle Radar Topography Mission (SRTM) in February 2000 were obtained from USGS. In addition, for Nevado Coropuna, a DEM with 30 m resolution was constructed from 1:50,000 topographic maps obtained from Instituto Geografico Nacional (IGN) of Peru using the TOPOGRID algorithm in ArcInfo. The DEM from topographic data and the DEM's from satellite data were validated using GPS points for Nevado Coropuna.
Glacier boundaries in vector format were obtained from the National Snow and Ice
Data Center (NSIDC), Boulder. These boundaries were compiled from the Digital Chart of the World and the World Glacier Monitoring Service's World Glacier Inventory (Raup et al.,
2000). Inca roads were digitized from paper maps of road fragments surveyed and documented by the Inca Road Project (Hyslop, 1984).
VI.2.4 Time-equivalent model
The time-equivalent model assumes that the topographic cost is based entirely on
slope, without considering load. The speed of travel was calculated based on slope,
independently of the load carried, using the equation from Gorenflo and Gale (1990):
v = 6e −3.5 slope + 0.05 (Equation 6.1)
108 where v is walking speed (in km/h), s is slope calculated as vertical change divided by
horizontal change and e is the base for natural logarithms (Gorenflo and Gale, 1990). The
function was derived from empirical data from soldiers traveling in various types of terrain
(cf. Imhof, 1950). This equation has not been tested in mountain terrain.
VI.2.5 Energy-dependent model
A formula for predicting energy expenditure accounting for the above factors has been presented by Pandolf et al. (1976; 1977):
2 ≈ l ’ M = 1.5w + 2(w + l)∆ ÷ + η (w + l )(1.5v 2 + 0.35v ⋅ s) « w ◊ (Equation 6.2) where: M = metabolic expenditure (Watts)
w = weight of the individual (kg)
l = load carried (kg)
η = terrain factor
v = speed of walking (m/s)
s = slope (%)
The formula predicts the total metabolic rate for walking at slow speeds while carrying loads on various slopes and various types of terrain. The terrain factor (η) in equation 6.2 reflects the type of terrain (e.g. dirt road, snow or sand). Coefficients for various type of terrain were obtained Soule and Goldman (1972) and Pandolf et al. (1976; 1977). They provide coefficient for travel at a constant speed with loads between 10 and 40 kg (Table
6.1). The energy expenditure formula was validated in the mentioned studies using empirical data from individuals carrying different loads, walking at different speeds over the entire range of slopes (Pandolf et al., 1977). However, the metabolic cost of climbing on a slope, at speeds less than 0.7 m/s, still needs validation (Pandolf et al., 1977).
109 Table 6.1 Terrain coefficients used to predict energy expenditure of load carrying in various
types of terrain. Compiled from Soule and Goldman (1992) and Pandolf et al. (1977;
1976).
Terrain type Coefficient (η) Blacktop surface 1.0 Dirt road 1.1 Light brush 1.2 Hard packed snow 1.3 Heavy brush 1.5 Swampy bog 1.8 Loose sand 2.2 Soft snow (15 cm) 2.5 Soft snow (25 cm) 3.3 Soft snow (35 cm) 4.1
Equation 6.2 was used to calculate energy using various parameters for slope, speed, load carried and type of terrain. I developed two energy-equivalent models. The first model is independent of the speed of walking and dependent only on slope gradient and load carried expenditure for hiking with a backpack on different slopes while maintaining constant speeds was calculated. I used constant speeds for hiking with a backpack obtained from the hiker’s speed table developed by the Search and Rescue Society of British Columbia based on empirical data (SARBC, 1992) (Table 6.2). The following parameters were specified for Inca individual: a weight of 70 kg, load carried of 30 kg, and an expert category of hiking. The weight was chosen on the basis of physiological studies, where the average weight of the individuals is 70 kg. I chose a load of 30 kg based on hypotheses about the weight carried by the Incas up the mountain (20 – 30 kg) and also based on load carried in the present day by
Andean porters (Ceruti, 1999).
110 The second energy-equivalent model is dependent on the speed on walking, calculated from Gorenflo and Gale’s (1990) equation. I then evaluated the effect of slope and load by varying each of them at a time.
Table 6.2 Hiking speed for walking uphill, downhill and flat slopes for various categories of
hikers. Reproduced from the Search and Rescue Society of British Columbia (SARBC
1992).
VI.2.6 Least cost path analysis
Topographic cost was calculated on a cell-by-cell basis, for each DEM, using the
GRID module available at the ArcInfo workstation. The unit of landscape is represented by one grid cell with 90 m resolution for the SRTM DEM and 30 m resolution for the DEM from topographic maps. Each grid cell was assigned an accessibility index, or a cost, which represents the difficulty of crossing that particular cell.
The time and energy layers that resulted from Equations 6.1 and 6.2 were used as inputs in the COSTDISTANCE function in ArcInfo GRID module to create layers of accumulated cost and direction of traveling to each cell from the starting point. These cost- distance layers were used to find the least cost paths from the starting point (base of the mountain) to the destination (summit) using the COSTPATH function. I did two runs for each
111 model: one run starting at present-day base camps, to validate the model, and a second run starting at Inca tambos (archeological sites) to reconstruct Inca climbing routes. I then compared these with GPS points of climbing routes surveyed without a-priori established routes. I then evaluated each algorithm and applied it to assess accessibility to all the archeological sites in the database, by calculating cost-weighted distance to the closest Inca road.
VI.3 Results
VI.3.1 Field work results
In June and August 2003, I climbed the SW and NW summits of Nevado Coropuna, and circled the mountain searching for archeological evidence at the base of the mountain. I gathered a total of 400 GPS points. 24 points were taken along the climbing route from the village of Mauca Llacta to the NW summit (6,377 m), and 59 points were along the climbing route from Lake Pallacocha (3,760 m) to the SW summit (6.411 m above sea level recorded by the GPS). An ice-core drilling expedition was conducted by l’Institut de Recherche pour le
Développement (IRD France), GREAT ICE project in June 2003, with the goal of obtaining a paleo-climatic record. During the expedition, I observed the climbing strategies used by local porters to transport large loads up to the drilling site (6,100 m), surveyed the route with a
GPS and recorded the travel time.
The present-day climbing route to the SW summit of Coropuna starts at Lake
Pallacocha at 4,760 m, a possible Inca camp (Ziólkowski and Belan Franco, 2001). At 5,026 m is found the ancient Inca camp Ajocancha, which displays a big plaza with rectangular and circular ruins, and a water canal (Figure 6.4) (Reinhard, 1999), This site dates from the late
Inca period (Ziólkowski and Belan Franco, 2001), and it might have been used as a temporary settlement for small groups to gain access to the mountain for ritual purposes (Reinhard,
1999). The NW summit of Coropuna is rarely climbed due to difficulty of access and lack of an established route to the summit. A possible starting point is the village of Mauca Llacta,
112 situated at 4,400 m, dating from the Intermediate-Late Inca period, and possibly used as a temporary settlement for administrative and ceremonial purposes (Ziólkowski and Belan
Franco, 2001).
Figure 6.4 The Inca camp of Ajocancha, with habitation complexes and water canal.
Nevado Pichu Pichu (5,634 m) contains archeological ruins dating from the Inca time
(Linares Malaga, 1966). Location of a double wall platform that contained a mummy, ruins along the climbing route and at the base of the mountain are shown in Figure 6.5.
Figure 6.5 Archeological sites along the climbing route on Nevado Pichu Pichu. Sketch
reproduced from Linares Malaga (1966).
113 VI.3.2 Time-equivalent model
The non-linear relationship between speed of travel and slope is shown in graphical
form in Figure 6.6. The graph is symmetric, both steep uphill and downhill slopes yielding
low walking velocities. According to this formula, the highest speed of walking in mountain
terrain (5.95 km/h) is attained on slightly downsloping terrain (-3° slopes), rather than on flat
terrain (0° slopes), where the maximum velocity is 5.04 km/h.
Figure 6.6 Speed of traveling expressed as a non-linear function of slope, based on the
velocity equation developed by Gorenflo and Gale (1990).
The hiking time as a function of slope is also non-linear, and increases exponentially with slope, as shown in Figure 6.7. For example, on a flat slope (0°) it takes 0.01 minutes to cross 1 meter of horizontal distance, while on a 45° slope it takes 40 times as much (0.4 minutes), to travel the same horizontal distance. This formula produces a very steep curve for travel time on slopes greater than 45°, which become terrain barriers.
114 Figure 6.7 Nonlinear relationship between hiking time and slope. Travel time is computed as
distance / velocity (time necessary to cross 1 meter of horizontal distance at a given
slope).
The resulting time surface calculated for Nevado Coropuna using the SRTM elevation dataset is shown in Figure 6.8. Steeper slopes require more time to cross.
Figure 6.8 Friction surface for Nevado Coropuna in the Peruvian Andes draped over the
hillshade of the SRTM DEM. White areas are cells with NODATA in the DEM. Cost is
expressed as time it takes to cross a cell (90m by 90m) at a given slope.
115 This model was validated on Nevados Coropuna and Pichu Pichu using the starting points and summits obtained from GPS measurements. Starting from Pallacocha Lake, it is possible to access both the NW and the SW summits (Figure 6.9). Similarly, both summits are accessible from Armas (Mauca Llacta site). The two routes intersect at the col (6,100m).
The modeled routes to the SW summit correlate very well the GPS points taken in the field, showing that this route (which is not the standard climbing route) is the least cost-effective.
The NW modeled route deviates slightly from the GPS points, showing that the route I took without a-priori established route was not the least cost possible.
Figure 6.9 Least cost climbing routes on Nevado Coropuna, Peruvian Andes draped on a
shaded relief map of the 30 m resolution DEM. Possible climbing routes from two
starting points to the two main summits are created using the time-dependent model
based on the Gorenflo and Gale’s (1990) equation. Also shown are the GPS points of the
actual routes climbed during fieldwork in 2003.
116 I compared the actual hiking time documented in the field with the accumulated time calculated from the starting points to the summits. The estimated hiking time calculated based on this model matched the actual hiking time from my field measurements very closely
(Table 6.3). For instance, starting at Ajocancha camp, the hiking time is between 5 – 6 hours depending on speed on hiking and acclimatization. This route took me 5.30 hours; it took the local porters 4 hours carrying loads, and the algorithm estimates 5.05 hours. The algorithm also calculates the length of the route traveled. The modeled least cost paths are longer than the straight line distance from the same starting point and destination (Table 6.3).
Table 6.3 Comparison of predicted and actual travel time for selected modeled routes.
Modeled time Modeled time Actual time Route (hours) SRTM (hours) TOPO (hours) DEM DEM Lake Pallacocha- 26.19 8.46 ~ 7 – 8 summit Ajocancha Base 20.2 5.05 ~ 5 – 6 camp-col Armas- NW 25.5 8.75 13 summit Base camp Pichu 15 N/a 9 Pichu
VI.3.3 Energy equivalent model
Equation 6.2 yielded the relative energy expenditure of hiking with a backpack on
different slopes while maintaining constant speeds yields. The energy cost is represented by
a non-linear curve (Figure 6.10). Absolute values of slope were used to account for positive
metabolic expenditure at downhill slopes, following suggestions from Van Leusen (2002)
and Llobera (2000).
117 Figure 6.10 Nonlinear relationship between actual energy expenditure and slope. Energy is
expressed in Watts expended to cross 1 meter of horizontal distance at a certain slope,
assuming standard speeds of walking uphill (0.43 m/s) and downhill (0.86 m/s) for an
expert hiker weighing 70 kg and carrying a 30 kg-load.
The effect of terrain type on energy expenditure
Energy expenditure is affected by the type of terrain that is being crossed (Pandolf et al., 1977). To estimate the effect of terrain type on climbing, velocity values were kept constant at the 0.43 m/s for uphill travel. For snow, the coefficient varies depending on the hardness and depth of snow (e.g. hard-packed snow vs. fresh deep powder) (Table 6.1).
Higher energy costs are required to carry a load in the snow (Soule and Goldman, 1972). The cost of hiking on snow increases exponentially at a higher rate than while travelling on hard packed snow (Figure 6.11). There is little difference between energy expended while traveling on glacier and the energy expended while traveling on dirt/non-glaciated terrain.
118 Figure 6.11 The effect of terrain type on energy expenditure of load carrying while hiking.
The graph is created using the equation provided by Pandolf et al (1977) and the terrain
coefficients provided by Soule and Goldman (1972) and Soule et al (1978).
Visually, the effect of adding snow onto the terrain can be observed in Figure 6.12.
Here I compare the least cost travel path derived by the energy model for an individual hiking on 35-cm deep snow with the least cost path derived from the velocity equation. The resulting least cost path approaches the SW ridge to the summit instead of passing through the col.
This approach seems to approximate better the normal route of approach to the SW summit. I calculated that the route derived from the energy model is also shorter – 7.738 km from the lake than the path derived with the time model. When the terrain coefficient increases (e.g. more snow), the energy model finds the best route while minimizing the length of the path.
119 Figure 6.12 The effect of terrain type on climbing routes. The least east cost paths are derived from the 30-m resolution DEM. The yellow line is
the path model with the time-dependent model and the red line using the energy-dependent model with parameters for 35-cm deep snow. The
red line estimates better the present-day climbing route (not shown).
120 VI.3.4 Effect of DEM resolution and quality on least cost paths
When the time-dependent model was run using the SRTM DEM (90 m resolution), the least cost path obtained was coarser (Figure 6.13). The travel times derived from the
SRTM DEM are as much as 4 times higher than both the ones modeled using the 30 m DEM or the actual hiking times measured in the field (Table 6.3). This is caused by the voids in the
SRTM DEM at surfaces with low backscatter (water bodies and glacier cover) (Figure 6.13).
When calculating the time surface in the GRID module, these cells are avoided and act as travel barriers.
Figure 6.13 The effect of DEM resolution on least cost paths. The smoother solid line
represents the path derived from the time-equivalent model from the 30 m DEM. The
coarser, straighter dotted line represents the path derived from the 90 m resolution DEM.
121 VI.4 Discussion
Minetti (1995) compared experimental results on walking at various speeds and gradients and found that the most efficient gradient of mountain paths is ~25% for uphill and
–20% for downhill walking. The optimal ascensional speeds of walking at these gradients are
+0.16m/s and –0.36 m/s, respectively (Minetti, 1995). The effect of load carrying and terrain type was included in a few studies such as Cymerman et al. (1981), Pandolf et al. (1977) and
Soule et al. (1978). The effect of load carrying and terrain type was included in a few studies such as Cymerman et al. (1981), Pandolf et al. (1977) and Soule et al. (1978). The results presented in this chapter are consistent with these studies, which stress the effect of slope on climbing strategies. The time dependent approach modeled a gentle route, which avoided steep slopes. In contrast, the energy model yields shorter ridge routes, commonly used in present day mountaineering. While both approaches constitute possible climbing routes, the choice of the algorithm to reconstruct Inca climbing routes depends on other factors such as altitude, topography and cultural barriers. These factors might have influenced the route chosen during the Inca times. Cultural beliefs and local legends might impose that some areas should not be crossed (Wheatley and Gillings, 2002). For instance, abandoned settlements in the Andes Mountains are typically avoided.
While load carrying does not seem to influence the hiking time of the porters in the
Andes, terrain features such as penitents have not been accounted in the model. The complex mountain terrain poses physical travel barriers, such as cliffs, rock outcrops and water bodies that may have been impassable for past cultures (Wheatley and Gillings, 2002). Glacier features such as crevasses, seracs (ice towers formed at the toes of the glaciers) and penitents reduce hiking speed of travel and may even impede travel. Penitents (spikes in the snow resulting from ablation caused by sun or warm wind) are prevalent at high altitudes in the tropics, for instance in the Andes Mountains, and are extremely arduous to cross. While
122 penitents increase the hiking time considerable, they are difficult to incorporate in the model because they change very rapidly.
The effect of altitude on the speed of walking was also investigated in physiological studies (Minetti, 1995; Péronnet et al., 1991; Pulfrey and Jones, 1996) reported the effect of altitude on the speed of walking. The lack of oxygen at high altitudes induces hyperventilation, increasing energy cost and generally reducing the maximum aerobic power
(Minetti, 1995; Pulfrey and Jones, 1996). Minetti (1995) suggested that at extreme altitudes
(5,000 – 6,000 m), lower gradients (15-18 %) should be used to minimize energy expenditure. Altitude constraints were not included in the present analysis because it was assumed that Andean people have the physiological adaptations for high altitude (Minetti,
1995).
VI.5 Conclusions and further applications
The least cost path analysis has proved to be useful here in determining possible climbing routes that might have been used to access archeological sites on the summit of the mountains. A few limitations remain. The quality of the DEM was essential in estimating a correct hiking time. Void values in the DEM provide unrealistically long hiking time.
Furthermore, these approaches could benefit from further validation in the field. Another limitation is that altitude constraints were not included in the model. Accounting for the altitude effect and the glacier changes would constitute further improvements to this model.
A final remark is that the approaches used are based on the assumption that the topography remained unchanged since the Inca times. In glacierized terrain, however, important changed might have occurred during the last 500 years.
123 "' /3$12$5$-
/1$#("3(5$,.#$+(-&%.1'(&' +3(34#$ 1"'$.+.&(" +2(3$2(-3'$ -#$2
VII.1 Introduction
High altitude archeological surveys in the Andes are constrained by the remoteness of the sites, the difficulty of obtaining archeological permits and the limited logistical support.
Given these constraints, the efficiency of the fieldwork could be improved by providing maps of archeologically sensitive areas. Such maps can be constructed with predictive modeling approaches. These approaches rely on the assumption that patterns of human behavior can be inferred from environmental factors observed at known archeological sites, and that these factors can be used to derive location models for unknown sites (Rose and Altschul, 1988).
Predictive models are tools to extrapolate from known locations to unsurveyed locations that display similar patterns, and provide an indication of potential archeological sites to be surveyed in the field.
Here I propose the development and testing of a predictive model for high altitude Inca archeological sites in the Andes using the existing archeological record and spatial techniques
(logistic regression and GIS). The model-building process draws from methodology developed by Kvamme and Kohler (1988), Kvamme (1988a; 1988b), Wheatley and Gillings
(2002) and Warren and Asch (2000). The model takes an inductive approach, which relies on statistical relationships between environmental variables and archeological site distributions, extracted from the existing archeological record. The GIS methodology draws upon approaches from Mapping the Archeological Potential of Ice and Snow (MAPIS) model developed for the Alaska snowfields and glaciers by Dixon and Manley (2001; 2002).
The initial stages of the model –hypothesis development, construction of the spatial database and initial development of the key variables - were presented in the previous chapters. Here I focus on the development of the predictive model and preliminary testing
124 using field data. The final stage, model testing and refinement, is an on-going process where refinements are incorporated to construct a robust model that can be applied at other similar sites. The assumptions used are: 1) the sites used by the Incas to construct archeological sites were not randomly chosen; 2) topographic factors determined to a certain extent the choice of site locations and 3) these patterns can be inferred using quantitative techniques and can be used to identify suitable locations in the field.
VII.2 Methods
VII.2.1 Archeological survey data
The study area is located in the Andes Mountains within the extent of Inca Empire, from Southern Peru to Chile. A calibration dataset was developed using archeological surveys from this area, published mainly in South America (Beorghia Nigris, 1985; Ceruti, 1999a;
1999b; Reinhard and Ceruti, 2000). The validation area is Nevado Coropuna (6.426 m), located in Cordillera Ampato in Southern Peru. During the fieldwork season 2003, I identified and recorded locations of the Inca sites located on the mountain using a Garmin
Etrex GPS unit.
VII.2.2 Logistic regression
Logistic regression was chosen for constructing the model for a number of reasons.
First, logistic regression calculates the probability of site occurrence based on hypothesized predictors for site location. Second, logistic regression is used when the response is binary
(i.e. is has only two levels, presence or absence) (binary logistic regression). It is a suitable technique for archeological survey data since the dependent variable - site presence/absence or site probability- is not measured on an ordinal scale (Wheatley and Gillings, 2002).
Logistic regression can also be used to model individual site types within a region, when
125 there are multiple categories of sites (multinomial logistic regression). Third, logistic regression is a nonparametric technique, it does not assume a linear relationship between the dependent and independent variable and it does not require normally distributed independent variables. Since logistic regression uses less assumptions than linear regression, it is the statistical approach most often used in archeological predictive models (Warren and Asch,
2000). Logistic regression determines the strength of each independent variable and creates a formula of the form:
L = α + β1 xa + β 2 x2 + ... + β n xn
where βi are the regression weights.
A logistic transformation (Kvamme, 1988b) is applied to calculate the probability of site presence:
e Li 1 P = = 1+ e Li 1+ e −Li
The dependent variable
The dependent variable is site probability. Two forms of logistic regression (binary and multiple) were used to construct models of site suitability for archeological sites in the
Andes. Binary logistic regression was used to predict the probability of sites vs. non-sites occurring at a specified location. Multinomial logistic regression for modeling multiple site types occuring within the study region (ceremonial, logistic or base camp sites).
The independent variables
Environmental and social factors initially considered to influence the location of archeological sites were entered as independent variables in the model. Topographic variables were stored as raster datasets with 90 m resolution. Other environmental variables (such as
126 landscape and vegetation) were stored as categorical variables. These variables were assigned to both the 151 archeological sites and a set of 456 “non-site” cells extracted using a stratified sampling technique from the extent of the study area. The non-site cells were assumed to contain no archeological information. Univariate tests (Kolmogorov – Smirnov and Mann-
Whitney U test) were used to examine contrasts between site locations and the background environment as well as contrasts among multiple site types. An initial choice of variables was made in basis of their significance for indicating contrasts between site types. In basis of the results from statistical tests, various clusters of site types were chosen as dependent variables for the regression model. Statistical methods were performed with the SPSS software.
VII.2.3 Model validation
An internal assessment of the model performance was done initially by applying the model to the same calibration dataset used to generate the model, and examining classification tables, which indicate the percent sites and non-sites classified correctly.
External validation was performed by applying the model to an independent study site,
Nevado Coropuna in Peruvian Andes. Model performance was assessed in terms of the gain statistic measure introduced by Kvamme (1988b), defined as:
Gain = 1 – (% of total area covered by the model / % of total sites within the model area)
The model is useful if there is the area likely to contain archeologicl evidence is small relative to the total study area (i.e. if the gain approaches 1).
VII.3 Results
VII.3.1 Initial data interpretation
Non-parametric tests (Kolmogorov-Smirnov and Mann-Whitney U tests) indicate that there are significant contrasts between environmental characteristics of site locations as opposed to non-site locations (p< 0.001 for all tests) (Chapter 5). In addition, there were
127 contrasts between different types of sites on the same mountain (ceremonial, logistic and base camp). Ceremonial sites occurred on different slopes, elevation, shelter index, precipitation and distance to glacier than base camp sites. Furthermore, ceremonial sites tend to occur on different morphologic classes than intermediate and base camp sites. These results suggest that archeological sites tend to occur on different locations than locations without archeogological sites (“non-sites”) in terms of environmental characteristics.
VII.3.2 Binary logistic results
A binary logistic regression was fitted to the 151 site-present data and 456 background sites using the 33 environmental and social variables described in Chapter 5. The differences between sites and non-sites were significant for six environmental variables: slope, shelter index, temperature, distance to glaciers (p-value < 0.001) and aspect (p-value <
0.05). The following logistic regression equation was obtained:
L = -21.42 – (.063·SLOPE) – (.004·ASPECT) + (.001·ELEV) + (.064·SHELTER) +
(.004·DIST_GLCRS) + (.490·TEMPERATURE)
Positive coefficients of the variable indicate site presence, and negative coefficients are associated with site absence. From the regression coefficients above, a couple of observations can be made. First, site probability decreases as slope increases, suggesting that flat areas have a higher site potential value. Second, site probability decreases as aspect increases, suggesting that slopes with lower aspect values (i.e. north-facing) have a higher site potential.
Third, site potential increases as one moves to higher elevations, which supports the hypothesis that sites higher altitudes were chosen as archeological sites. Fourth, the site potential increases as one moves further away from glaciers and at higher temperatures. This
128 result does not support the hypothesis that glaciated sites with lower temperatures were chosen for preservation of human remains. Rather, it suggests that accessibility (i.e. snow- free areas) were preferentially chosen for archeological sites.
The binary logistic regression model is significant at the .001 level according to the
Hosmer and Lemeshow test, and the Nagelkerke R Square is .813. The performance of the model was evaluated by deriving a contingency table (Table 7.1). Locations were assigned to either “sites” or “non-sites” based on a cutoff value of p = 0.5. This means that any location with a probability greater than 50% is considered a “site” as opposed to “non-site”. At the 0.5 cut-off value, 94.1 % of sites were classified correctly and 83.6% of the non-sites were classified correctly.
Table 7.1 Classification table for probabilities of archeological site occurrence, where “1” =
sites and “0” = nonsites. The cut-off value is 0.5.
Observed Predicted Percentage BINARY Correct .00 1.00 BINARY .00 440 14 96.9 1.00 21 107 83.6 Overall Percentage 94.0
VII.3.3 Multinomial logistic results
The multinomial logistic model predicts the probability of a location of being one of the multiple site types specified (ceremonial, logistic or base camp). These sites represent different functional site types, derived in basis of archeological survey data (Chapter 3). In this case, the probability values for each of the site types sum to 1.0 for any given location
(i.e. grid cell on the digital elevation model), thus:
p(s1) + p(s2) + p(s3) = 1, where s1, s2 and s3 are the three types of sites.
129 The model is significant at the 0.001 level, and the Nagelkerke R square is 0.8. This model predicted 95.3% ceremonial sites correctly, 68.8 % logistical sites correctly, and 88.2% base camp sites correctly (Table 7.2).
Table 7.2 Classification table for probabilities of archeological site occurrence, where “1” =
ceremonial sites, “2” = logistical sites and “3” = base camp sites.
Predicted Observed Percent 1 2 3 Correct 1 81 2 2 95.3% 2 3 11 2 68.8% 3 2 0 15 88.2% Overall 72.9% 11.0% 16.1% 90.7% Percentage
The classification tables presented above provide an estimation of percentage of rach site type classified correctly. It can be noted that a percentage of sites are still incorrectly classified by each model. However, for field validation purposes, the model cut-off value can be adjusted to increase the prediction accuracy for a certain site type while decreasing the accuracy for another site type predicted (Kvamme, 1988b). The binary logistic model predicts
96.9 % of the sites correctly, which is considered sufficient, so no lower cutoff was specified.
The multinomial logistic regression predicted 95.3% of the ceremonial sites correctly but only 68.8% of the intermediate sites were correctly assigned the site category. The model cut- off point could be adjusted to bias the model and predict a larger percentage of intermediate sites correctly while lowering the accuracy of the ceremonial sites predicted. These model specifications depend on the practical application and the needs of the real-world survey.
Here, I using the default cutoff value of 0.5 yielded a percentage of sites correctly predicted larger than 90%, which is considered acceptable (Kvamme, 1988b). Furethemore, the second model is biased towards predicting ceremonial sites more accurately than intermediate sites
130 or base camp sites. This is acceptable since most archeological survey focus on ceremonial sites, which contain the most valuable archeological record.
VII.3.4 Field testing the regressio n models
The internal model validation performed using classification tables might yield an inflated view of model performance (Kvamme, 1988b). In practice, archeologists would like to be able to asses the model performance by applying it to an independent dataset, at an unsurveyed area. I chose Nevado Coropuna for the external validation of the model. A few antecedent explorations confirm the archeological potential of Coropuna. Testimonies of local people mentioned Inca constructions at the base of the mountain. Fragments of Inca pottery, and two mummies were retrieved from Coropuna, but their location was not mentioned in the published studies (Linares Malaga, 1978). Recent archeological sites were documented by survey teams from Proyecto Condensuyo (Ziólkowski and Belan Franco,
2001). I used these findings along with the field data acquired with a GPS unit in 2003 to assess the performance of the two models presented above.
Fieldwork results consist of 240 GPS points that contained archeological evidence on
Nevado Coropuna. At 90 m grid cell resolution, this represents a total area of 1.94 km2 (0.22
% of the study area). When the probability results are mapped onto a digital elevation model, certain pixels are more likely to contain archeological ruins than others. For the binary logistic regression model, the probability layer derived on a cell by cell basis for the study area is shown in Figure 7.1. Of the total area, 0.02 % is classified with a probability bigger than 99% of being an archeological site. Based on the percent area covered by archeological sites and percent area predicted with a probability > 99%, the gain statistic for this model is
0.91, which implies that the model has predictive utility. It can be noted that pixels with flat slopes and higher exposures display a higher archeological probability.
131 Figure 7.1 Probability image draped on a shaded relief map. The effect of topography (shelter and slope) is visible in this model. Areas with high shelter are assigned a low archeological potential.
For multinomial logistic regression, the maps in Figure 7.2 (a, b and c) shows the probabilities for each site type occuring. Each pixel is assigned to one of the three types of sites possible (ceremonial, logistic or base camp). The maps suggest that the site potential for the three site types is largely dependent on elevation. An improvement to this model would be to include an alternative of no-site as one of the site categories.
132 a.
b.
c.
Figure 7.2 Probability of site occurrence. a) Ceremonial sites; b) intermediate sites; c) base
camp sites.
133 VII.4 Conclusions
Logistic regression was used to calculate the potential of archeological sites occurring at a location given a set of topographic factors. While the classification tables indicate percentages of correct estimates above 90%, results required evaluation with independent field data for an evaluation for real-worls applications. The analysis presented above suggest that 96.9 % of the archeological sites may be predicted correctly by the regression model.
Applying the model to an independent study site on nevado Coropuna yielded probabilities in the range of 0.03 to 0.77 for the actual archeological sites. While the gain statistic (0.91), suggests that the model is robust predictive since it predicts a small percentage of potential archeological site locations, further calibration is needed in order to predict high site- sensitivity areas at unsurveyed locations. Furher model improvements include adjusting the cutoff values, using different classification schemes, including new variables that may offer increased predictive power, and field testing at other areas.
This phase of the modeling process is a continual process of refining and testing. Using the field data and drawing from validation approaches used in other predictive models, it is hoped that the model will be refined to accurately map archeologically sensitive areas in the
Andes.
134 1$%$1$-"$2
Ackerman, T., Erickson, T. and Williams, M. W., 2001. Combining GIS and GPS to improve our understanding of the spatial distribution of snow water equivalence (SWE). Proceedings of ESRI USERS Conference, San Diego, CA. Altschul, J. H., 1988. Models and the modeling process. In: Judge, W. J. and Sebastian, L., Eds., Quantifying the present and predicting the past : theory, method, and application of archeological predictive modeling. US Department of the Interior, Bureau of Land Managament Service Center, Denver. Altschul, J. H. and Nagle, C. L., 1988. Collecting new data for the purpose of model development. In: Judge, W. J. and Sebastian, L., Eds., Quantifying the present and predicting the past : theory, method, and application of archeological predictive modeling. US Department of the Interior, Bureau of Land Managament Service Center, Denver. Ames, A., Muñoz, G., Verástegui, J., Zamora, M. and Zapata, M., 1989. Glacier Inventory of Peru. Part I. Hidrandina S.A. Unit of Glaciology and Hydrology. Huaraz - Perú. Aniya, M., Sato, H., Naruse, R. and Skvarca, P. C., G., 1996. The use of satellite and airborne imagery to inventory outlet glaciers of the Southern Patagonia Icefielf, South America. Photogrammetric Engineering and Remote Sensing 62, 1361-1369. Aveni, A., Ed. 1990. The Lines of Nazca. The American Philosphical Society, Philadelphia. _____, 1996. Astronomy and the Ceques System. Journal of the Steward Anthropological Society 24(1-2), 157-172. Baroni, C. and Orombelli, 1996. The Alpine "Iceman" and Holocene Climatic Change. Quaternary Research 46, 78-83. Bauer, B., 1992. Ritual pathways of the Incas: an analysis of the Collasuyu ceques in Cuzco. Latin American Antiquity 3(3), _____, 1998. Sacred landscape of the Inca:the Cuzco ceque system. Unversity of Texas Press,Austin. Bauer, B. S. and Dearborn, D. S. P., 1995. Astronomy and empire in the ancient Andes: the cultural origins of Inca sky watching. University of Texas Press,Austin. BBC, 2003. Melting glaciers threaten Peru. BBC News. Online: http://news.bbc.co.uk/1/hi/world/americas/3172572.stm. Bell, T. and Lock, G., 2000. Topographic and cultural influences on walking the Ridgeway in later prehistoric times. In: Lock, G., Ed. Beyond the Map: archeology and spatial technologies. NATO Science Series A: Life Sciences, IOS Press, Amsterdam. Bell, T. L., Church, R. L. and Gorenflo, L. J., 1988. Late Horizon regional efficiency in Northeastern Basin of Mexico: A location-allocation perspective. Journal of Anthropological Archaeology 7, 163-202.
135 Beorchia, A., 1985. El enigma de los santuarios indigenas de alta montaña. Revista del CIADAM 5. Bernbaum, E., 1990. Sacred mountains of the world. Sierra Club Books,San Francisco. _____, 1997. The spiritual and cultural significance of mountains. In: Ives, J. and Messerli, B., Eds., Mountains of the world:a global priority. Parthenon Publishing Group, New York;London. 39-60. Brandt, R., Groenewoudt, B. J. and Kvamme, K. K., 1992. An experiment in archeological site location: modeling in the Netherlands using GIS techniques. World Archaeology 24(2), 269-282. Burrough, P. A. and McDonnell, R. A., 1998. Principles of Geographic Information Systems. Oxford University Press,Oxford. Ceruti, M. C., 1997. Arqueología de alta montaña,Salta, Argentina. _____, 1999a. Cumbres sagradas del noroeste argentino : avances en arqueología de alta montaña y etnoarqueología de santuarios de altura andinos. Editorial Universitaria de Buenos Aires. EUDEBA,Buenos Aires. _____, 1999b. Recientes prospecciones arqueologicas. Revista del C.I.A.D.A.M.Corp Author(s): Centro de Investigaciones Arqueológicas de Alta Montaña 6, Chávez Chávez, J. A., 1993. La erupción del volcán Misti. Pasado, Presente, Futuro. Imprenta zenit, Arequipa. _____, Director, Museo Santuarios Andinos, Universidad Católica Santa María, Arequipa, personal communication. Chepstow-Lusty, A., Frogley, M. R., Bauer, B. S., Bush, M. B. and Herreras, A. T., 2003. A late Holocene record of arid events from the Cuzco region, Peru. Journal of Quaternary Science 18(6), 491-502. Church, T., Brandon, J. R. and Burgett, G. R., 2000. GIS applications in archeology: Method in search of theory. In: Westcott, K. L. and Brandon, J. R., Eds., Practical applications of GIS for archeologists: a predictive modeling kit. Taylor & Francis, London. Cobo, B., 1990 [1653]. Inca religion and customs. University of Texas Press, Austin. _____, 1990[1653]. Inca religion and customs. University of Texas Press, Austin. Cogley, J. G. and Jung-Rothenhausler, 2004. Uncertainty in digital elevation models of Axel Heiberg Island, Arctic Canada. Arctic, Antarctic and Alpine Research 36(2), 249 - 260. Corte, A. E., 2001. Las emplazamientos rituales Incas en el permafrost de los Andes (Observaciones Paleoclimaticas). In: Schobinger, J., Ed. El santuario Incaico del cerro Aconcagua. EDIUNC, Mendoza. Corte, A. E. and Zhijui, C., 1985. Comparative study of Geocryogenic features, conditions and processes in the Andes and Himalayas. Acta Cryogenica 3, Cymerman, A., Pandolf, K. B., Young, A. J. and Maher, J. T., 1981. Energy expenditure during load carriage at high altitude. J Appl Physiol 51(1), 14-18.
136 de Molina, C., 1989 [1575]. Relación de la fábulas y ritos de los Incas. In: H.Urbano and P.Duviols, Eds., Fábulas y mitos de los incas. Historia 16, Crónicas de America 48. Ediciones Atlas, Madrid. Dixon, J. E. and Manley, W. F., 2001. Archeology of the cryosphere. Proceedings of XIV International Congress of Prehistoric and Protohistoric Sciences, Pre-prints, Liege, Belgium. _____, 2002. Archeology of Alaska glaciers and snow fields. Proceedings of Poster presentation, 32nd Annual Arctic Workshop Conference, Boulder. Dozier, J., 1980. A Clear-Sky Spectral Solar-Radiation Model for Snow-Covered Mountainous Terrain. Water Resources Research 16(4), 709-718. Duncan, C. C., Klein, A. J., Masek, J. G. and Isacks, B. L., 1998. Comparison of late Pleistocene and modern glacier extents in central Nepal based on digital elevation data and satellite imagery. Quaternary Research 49, 241-254. Duncan, R. B. and Beckman, K. A., 2000. The application of GIS predictive site location models within Pennsylvania and West Virginia. In: Westcott, K. L. and Brandon, J. R., Eds., Practical applications of GIS for archeologists: A predictive modeling kit. Taylor and Francis, London and New York. Dwyer, J., LP DAAC Project Scientist, USGS EROS Data Center, personal communication Dyurgerov, M., 2002. Glacier Mass Balance and Regime: Data of Measurements and Analysis. Dyurgerov, M. B. and Meier, M. F., 2000. Twentieth century climate change: Evidence from small glaciers. PNAS 97(4), 1406-1411. Eliade, M., 1963. Patterns in comparative religion. World, Cleveland. _____, Ed. 1987. Mountains. The encyclopedia of religion. v.10, MacMillian, NY. Ericksen, G. E., Plafker, G. and Fernández Concha, J., 1970. Preliminary report on the geologic events associated with the May 31, 1970, Peru earthquake,Washington U.S. Geological Survey. ESRI, 1999. ArcAtlas: Our Earth. CD-ROM. Etzelmüller, B., 2000. On the Quantification of Surface Changes using Grid-based Digital Elevation Models (DEMs). Transactions in GIS 4(2), 129-143. Etzelmüller, B. and Björnsson, H., 2000. Map analysis techniques for glaciological applications. International Journal of Geographical Information Science 14(6), 567- 581. Etzelmüller, B., Vatne, G., Odegard, R. S. and Sollid, J. L., 1993. Mass-Balance and Changes of Surface Slope, Crevasse and Flow Pattern of Erikbreen, Northern Spitsbergen - an Application of a Geographical Information-System (GIS). Polar Research 12(2), 131- 146. Falorni, G., Istanbulluoglu, E. and Bras, R. L., 2003b. A Method for Improving SRTM DEMs in High-Relief Terrain. EOS Trans. AGU 84(46). Falorni, G., Vivoni, E. R., Teles, E. R. and Bras, R. L., 2003a. Evaluating SRTM elevation products for hydrogeomorphological applications. Proceedings of EGS XXV General Assembly, Nice, France.
137 Farrington, I. S., 1992. Ritual Geography, Settlement-Patterns and the Characterization of the Provinces of the Inka Heartland. World Archaeology 23(3), 368-385. Fernández, J., Panarello, H. O. and Schobinger, J., 1999. The Inka Mummy from Mount Aconcagua: Decoding the geographic origin of the "Messenger of the Deities" by means of stable carbon,nitrogen, and sulphur isotope analysis. Geoarcheology:An international Journal 14(1), 27-46. Georges, C., 2004. The 20th century glacier fluctuations in the Cordillera Blanca (Perú). Arctic, Antarctic, and Alpine Research 36(2). Gorenflo, L. J. and Gale, N., 1990. Mapping regional settlement in information space. Journal of Anthropological Archaeology 9, 240-274. Gow, D., 1974. Taytacha Qoyllur Rit'i. Allpanchis Phuturinqa 7. Gratton, D. J., Howarth, P. J. and Marceau, D. J., 1990. Combining DEM parameters with Landsat MSS and TM imagery in a GIS for mountain glacier characterization. IEEE Transactions on Geoscience and Remote Sensing 28(4), 766-769. Guamán Poma de Ayala, F., 1980 [1615]. Nueva crónica y buen gobierno. Siglo Ventiuno, Mexico City. _____, 1980[1615]. Nueva crónica y buen gobierno. Siglo Ventiuno, Mexico City. Hamilton, L. S., 2001a. Mountain Protected areas update No.30, June 2001. _____, 2001b. Mountain Protected areas update No.31, September 2001. Hasternath, S. and Ames, A., 1995. Recession of Yanamaery glacier in Cordillera Blanca, Perú during the 20th century. Journal of Glaciology 41(137), 191 - 196. Henderson, K. A., Thompson, L. G. and Lin, P. N., 1999. Recording of El Nino in ice core delta O-18 records from Nevado Huascaran, Peru. Journal of Geophysical Research- Atmospheres 104(D24), 31053-31065. Hirano, A., Welch, R. and Lang, H., 2003. Mapping from ASTER stereo image data: DEM validation and accuracy assessment. ISPRS Journal of Photogrammetry and Remote Sensing 57(5-6), 356-370. Hutchinson, M. F., 1988. Calculation of hydrologically sound digital elevation models. Proceedings of Third International Symposium on Spatial Data Handling, Sydney, Hutchinson, M. F. and Gallant, J. C., 2000. Digital Elevation Models and representation of terrain shape. In: Gallant, J. C., Ed. Terrain analysis - principles and applications. Wiley, New York. Hyslop, J., 1984. The Inka Road System. Academic Press ,Orlando. Imhof, E., 1950. Gelaende und Karte. Rentsch, Zurich. Jochim, M. A., 1976. Hunter-gatherer subsistence and settlement: A predictive model. Academic Press, New York. Judge, W. J. and Sebastian, L., 1988. Predicting the past: correlation, explanation, and the use of archeological models. In: Judge, W. J. and Sebastian, L., Eds., Quantifying the present and predicting the past : theory, method, and application of archeological predictive modeling. US Department of the Interior, Bureau of Land Managament Service Center, Denver.
138 Kääb, A., 2002. Monitoring high-mountain terrain deformation from repeated air- and spaceborne optical data: examples using digital aerial imagery and ASTER data. ISPRS Journal of Photogrammetry and Remote Sensing 57(1-2), 39-52. Kääb, A., Paul, F., Maisch, M., Hoelzle, M. and Haeberli, W., 2002. The new remote- sensing-derived Swiss glacier inventory: II. First results. Annals of Glaciology 34, 362 - 366. Kääb, A., Huggel, C., Paul, F., Wessels, R., Raup, B., Kieffer, H. and Kargel, J., in press. Glacier Monitoring from ASTER Imagery: Accuracy and Applications. Proceedings of EARSeL LIS-SIG Workshop, Berne, March 11-13, 2002, Kadomtseva, T. G., 1996b. Permafrost. In: Liouty, A. A., Ed. ArcAtlas: Our Earth. ESRI. Kamp, U., Bolch, T. and Olsenholler, J., 2003. DEM Generation from ASTER Satellite Data for Geomorphometric Analysis of Cerro Sillajhuay, Chile/Bolivia. Proceedings of Annual Meeting Imaging and Geospatial Information Society (ASPRS), Anchorage, Alaska, May 5 - 9. Kaser, G., 1999a. A review of modern fluctuations of tropical glaciers. Global and Planetary Change 22, 93 - 103. _____, 1999b. A review of the modern fluctuations of tropical glaciers. Global and Planetary Change 22(1-4), 93-103. Kaser, G., Ames, A. and Zapata, M., 1990. Glacier fluctuations and climate in the Cordillera Blanca, Perú. Annals of Glaciology 14, 136 - 140. Khalsa, S. J. S., Dyurgerov, M., Khomova, T., B., R. and Barry, R. G., in press. Space-based mapping of glacier changes using ASTER and GIS tools. Proceedings of International Geoscience and Remote Sensing Symposium, Toulouse, France, July 21 - 25, 2003. Klein, A. G. and Isacks, B. L., 1996. Glaciers: tracking change in the central Andes Mountains. GIS World 9, 48-52. Kuiper, J. A. and Westcott, K. L., 1999. A GIS approach for predicting prehistoric site locations. Proceedings of Presented at the nineteenth Annual ESRI User Conference, San Diego, California, July 26-30, 1999. Kvamme, K. L., 1988a. Using existing archeological survey data for model building. In: Judge, W. J. and Sebastian, L., Eds., Quantifying the present and predicting the past: theory, method, and application of archeological predictive modeling. US Department of the Interior, Bureau of Land Managament Service Center, Denver. _____, 1988b. Development and testing of quantitative models. In: Judge, W. J. and Sebastian, L., Eds., Quantifying the present and predicting the past : theory, method, and application of archeological predictive modeling. US Department of the Interior, Bureau of Land Managament Service Center, Denver. _____, 1990a. The fundamental principles and practice of predictive archeological modeling. In: Voorrips, A., Ed. Mathematics and Information Science in Archeology: a flexible framework. Studies in Modern Archeology. Holos-Verlag, Bonn. 257-95. _____, 1990b. One-sample tests in regional analysis: new possibilities through computer technology. American Antiquity 55(2), 367-381. _____, 1992. A Predictive Site Location Model on the High-Plains - an Example with an Independent Test. Plains Anthropologist 37(138), 19-40.
139 _____, 1999. Recent directions and developments in Geographic Information Systems. Journal of Archaeological Research 7(2), 153-201. Kvamme, K. L. and Kohler, T. A., 1988. Geographic Information Systems:Technical aids for data collection, analysis and display. In: Judge, W. J. and Sebastian, L., Eds., Quantifying the present and predicting the past : theory, method, and application of archeological predictive modeling. US Department of the Interior, Bureau of Land Managament Service Center, Denver. Lang, H. R. and Welch, R., 1999. Algorithm Theoretical Basis Document for ASTER Digital Elevation Models, Version 3.0. Jet Propulsion Laboratory, Pasadena, CA, 69 pp. Legates, D. R. and Willmott, C. J., 1990a. Mean Seasonal and Spatial Variability in Gauge- Corrected, Global Precipitation. International Journal of Climatolgy 10, 111-127. _____, 1990b. Mean Seasonal and Spatial Variability in Global Surface Air Temperature. Theoretical and Applied Climatology 41, 11-21. Linares Malaga, E., 1966. Restos arqueologicos en el Nevado Pichu Pichu (Arequipa, Perú). Anales de arqueologia y etnologia XXI, 7-47. Llobera, M., 2000. Understanding movement: a pilot model towards the sociology of movement. In: Lock, G., Ed. Beyond the Map: archeology and spatial technologies. IOS Press, Amsterdam. Martinez, G., 1983. Los dioses de los cerros en los Andes. Journal de la Societé des Americanistes XXIX, 85 - 116. Mennis, J. L. and Fountain, A. G., 2001. A spatio-temporal GIS Database for monitoring alpine glacier change. Photogrammetric Engineering and Remote Sensing 67(8), 967- 975. Minetti, A. E., 1995. Optimum Gradient of Mountain Paths. Journal of Applied Physiology 79(5), 1698-1703. Minetti, A. E. and Ardigò, L. P., 2001. The transmission efficiency of backward walking at different gradients. European Journal of Physiology 442. Minetti, A. E., Moia, C., Roi, G. S., Susta, D. and Ferretti, G., 2002. Energy cost of walking and running at extreme uphill and downhill slopes. Journal of Applied Physiology 93(3), 1039-1046. Morales-Arnao, B., 1999. Glaciers of Peru. In: Ferrigno, J. G., Ed. Satellite image atlas of glaciers of the world. United States Geological Survey Professional Paper 1386-I. Morozova, I. V. and Berlyant, T. G., 1996. Solar Radiation. In: Liouty, A. A., Ed. ArcAtlas: Our Earth. ESRI. Morrison, T., 1978. Pathways to the gods:the mystery of the Andes lines. Harper&Row,New York. Mostny, G., 1957. La momia del cerro El Plomo. Bol.Mus.Nac.Hist.Nat.Chile 27(1). Narshkikh, R. S., 1996a. Morphosculpture - Introduction. In: Liouty, A. A., Finko, E. A. and Zueva, E. V., Eds., ArcAtlas: Our Earth. ESRI. _____, 1996b. Morphologic structure - Introduction. In: Liouty, A. A., Ed. ArcAtlas: Our Earth. ESRI.
140 Pandolf, K. B., Haisman, M. F. and Goldman, R. F., 1976. Metabolic Energy-Expenditure and Terrain Coefficients for Walking on Snow. Ergonomics 19(6), 683-690. Pandolf, K. B., Givoni, B. and Goldman, R. F., 1977. Predicting energy expenditure with loads while standing or walking very slowly. J Appl Physiol 43(4), 577-581. Paul, F., 2001. Evaluation of different methods for glacier mapping using Landsat TM. Proceedings of EARSeL Workshop on Remote Sensing of Land Ice and Snow, Dresden, June 16 -17, 2000, 1, 239-245. Paul, F., Kääb, A., Maisch, M., Kellenberger, T. and Haeberli, W., 2002. The new remote- sensing-derived Swiss glacier inventory: I. Methods. Annals of Glaciology 34, 355 - 361. Paulsen, A. C., 1976. Environment and empire:climatic factors in prehistoric Andean Culture change. World Archaeology 8(2), 121-132. Péronnet, F., Thibault, G. and Cousineau, D.-L., 1991. A theoretical analysis of the effect of altitude on running performance. Journal of Applied Physiology 67, Pulfrey, S. M. and Jones, P. J. H., 1996. Energy expenditure and requirement while climbing above 6,000 m. Journal of Applied Physiology 81(3), 1306-1311. Quevedo, S. and Durán, E., 1992. Ofrendas a los dioses en las montañas: Santuarios de altura en la cultura Inka. Bol.Mus.Nac.Hist.Nat.Chile 43(43), 193-206. Rabus, B., Eineder, M., Roth, A. and Bamler, R., 2003. The shuttle radar topography mission--a new class of digital elevation models acquired by spaceborne radar. ISPRS Journal of Photogrammetry and Remote Sensing 57(4), 241-262. Rapp, R. H., 1996. Use of potential coefficient models for geoid undulation determination using a spherical harmonic representation of the height anomaly/geoid undulation difference. Journal of Geodesy 71(5), 282 - 289. Raup, B. H., Kieffer, H. H., Hare, T. M. and Kargel, J. S., 2000. Generation of Data Acquisition Requests for the ASTER Satellite Instrument for Monitoring a Globally Distributed Target: Glaciers. IEEE Transactions on Geoscience and Remote Sensing 38(2), 1105-1112. Rebitsch, M., 1966. Santuarios indigenas en altas cumbres de la puna Atacama. Anales de arqueologia y etnologia 21. Reinhard, J., 1985. Sacred Mountains - an Ethno-Archaeological Study of High Andean Ruins. Mountain Research and Development 5(4), 299-317. _____, 1985a. Sacred Mountains - an Ethno-Archaeological Study of High Andean Ruins. Mountain Research and Development 5(4), 299-317. _____, 1991. Macchu Picchu, the sacred center. Nuevas imagenes,Lima. _____, 1996. Peru’s Ice Maidens. National Geographic Magazine 189(6), 62-81. _____, 1999. Coropuna- Lost temple of the Incas. South American Explorers 58. Reinhard, J. and Ceruti, M. C., 2000. Investigaciones arqueológicas en el Volcán Llullaillaco : complejo ceremonial incaico de alta montaña. Ediciones Universidad Católica de Salta,Salta, Argentina. Reinhard, J. and Chávez, J. A., 2001. The Inca Temple of Coropuna. Tawantinsuyu (in press),
141 Rivera, A. and Casassa, G., 1999. Volume changes on Pio XI glacier, Patagonia: 1975-1995. Global and Planetary Change 22(1-4), 233 - 244. Rose, M. R. and Altschul, J. H., 1988. An overview of statistical method and theory for quantitative model building. In: Judge, W. J. and Sebastian, L., Eds., Quantifying the present and predicting the past : theory, method, and application of archeological predictive modeling. US Department of the Interior, Bureau of Land Managament Service Center, Denver. Safronova, I. N. and Khramtsov, V. N., 1996. Vegetation. In: Kotova, T. V., Liouty, A. A. and Smirnova, E. V., Eds., ArcAtlas: Our Earth. ESRI. Sallnow, M., 1974. La peregrinación andina. Allpanchis, n°spécial: "La fiesta en los Andes" 7, SARBC, 1992. Hiker speed table. Available at: http://www.sarbc.org/sarbc/speed.html (accessed February, 2003). Schiffer, M. B., 1987. Formation processes of the archaeological record. University of New Mexico Press,Albuquerque, NM. Schobinger, J., 1964. Descubrimiento de una momia del período incaico en la cumbre del cerro El Toro (6300 m. Prov. San Juan), Mendoza. _____, 1966. La "Momia" del cerro El Toro : investigaciones arqueológicas en la cordillera de la Provincia de San Juan (República Argentina) : suplemento al tomo XXIo. de los Anales de arqueología y etnología. [Instituto de Arqueología y Etnología Universidad Nacional de Cuyo], Mendoza. _____, 1967a. Breve historia de la arqueologia de alta montaña en los Andes Meridionales. Sociedad arqueologica de Santiago 4, 23-34. _____, 1967b. Breve historia de la arqueología de alta montaña en los Andes Meridionales. Boletín de la Sociedad Arqueológica de Santiago 4, 23-24. _____, 1969. La "Momia" del Cerro del Toro : y sus relaciones con otros sitios arqueológicos de la Cordillera de los Andes. El Autor, Mendoza. _____, 1991. Sacrifices of the high Andes. Natural History 4, 63-70. _____, 2001a. EL santuario incaico del cerro Aconcagua. EDIUNC-Editorial de la Universidad Nacional de Cuyo,Mendoza, Argentina. _____, 2001b. Datación radiocarbónica de la momia del Aconcagua por el Laboratorio Geochron. In: Schobinger, J., Ed. El santuario incaico del cerro Aconcagua. Mendoza. _____, 2001a. Datación radiocarbónica de la momia del Aconcagua por el Laboratorio Geochron. In: Schobinger, J., Ed. El santuario incaico del cerro Aconcagua. Mendoza. _____, Ed. 2001b. EL santuario incaico del cerro Aconcagua. EDIUNC-Editorial de la Universidad Nacional de Cuyo, Mendoza, Argentina. Shimada, I., Schaaf, C. B., Thompson, L. G. and Mosleythompson, E., 1991. Cultural Impacts of Severe Droughts in the Prehistoric Andes - Application of a 1,500-Year Ice Core Precipitation Record. World Archaeology 22(3), 247-270.
142 Sidjak, R., 1999. Glacier mapping of the Illecillewaet icefield, British Columbia, Canada, using Landsat TM and digital elevation data. International Journal of Remote Sensing 20(2), 273-284. Sobczyk, M., 2000. Arquitectura funeraria prehispánica en la región del Nevado Coropuna, Perú. Boletín de la Misión Arqueológica Andina 2(4), Soule, R. G. and Goldman, R. F., 1972. Terrain coefficients for energy cost prediction. J Appl Physiol 32(5), 706-708. Soule, R. G., Pandolf, K. B. and Goldman, R. F., 1978. Energy-Expenditure of Heavy Load Carriage. Ergonomics 21(5), 373-381. Stanish, C., 2001. Regional research on the Inca. Journal of Anthropological Research 9(3), Thompson, L. G., Davi, M. E. and Mosley-Thompson, E., 1994. Glacial records of global climate:A 1500-Year tropical ice core record of climate. Human Ecology 22(1), 83- 95. Thompson, L. G., Mosley-Thompson, E. and Henderson, K. A., 2000. Ice-core palaeoclimate records in tropical South America since the Last Glacial Maximum. Journal of Quaternary Science 15(4), 377-394. Thompson, L. G., Davis, M. E., Mosleythompson, E. and Liu, K. B., 1988. Pre-Incan Agricultural Activity Recorded in Dust Layers in 2 Tropical Ice Cores. Nature 336(6201), 763-765. Thompson, L. G., Mosleythompson, E., Grootes, P. M., Pourchet, M. and Hastenrath, S., 1984. Tropical Glaciers - Potential for Ice Core Paleoclimatic Reconstructions. Journal of Geophysical Research-Atmospheres 89(ND3), 4638-4646. Thompson, L. G., Mosleythompson, E., Davis, M. E., Lin, P. N., Henderson, K. A., Coledai, J., Bolzan, J. F. and Liu, K. B., 1995. Late-Glacial Stage and Holocene Tropical Ice Core Records from Huascaran, Peru. Science 269(5220), 46-50. Thouret, J. C., Juvigne, E., Loutsch, I. and Chávez, J. A., 2001a. Activité volcanique et sacrifices humains chez les Incas au Pérou méridional. In: Juvigne, E. and Raynal, J.- P., Eds., Tephras: chronologie, archeologie. Les dossiers de l'Archéo-Logis. CDCRAD, Goudet. Thouret, J. C., Finizola, A., Fornari, M., Legeley-Padovani, A., Suni, J. and Frechen, M., 2001b. Geology of El Misti volcano near the city of Arequipa, Peru. Geological Society of America Bulletin 113(12), 1593-1610. Thouret, J. C., Juvigne, E., Mariño, J., Moscol, M., Legeley-Padovani, A., Loutsch, I., Davila, J., Lamadon, S. and Rivera, M., 2002. Late Pleistocene and Holocene Tephro- Stratigraphy and chronology in Southern Peru. Tephrocronologia. Tierney, P., 1989. The highest altar : the story of human sacrifice. Viking,New York. Tou, J. T. and Gonzales, R. C., 1974. Pattern recognition principles. Addison - Wesley Publishing Co. Toutin, T., 2001. Elevation modeling from satellite visible and infrared (VIR) data. International Journal of Remote sensing 22(6), 1097-1125. _____, 2002a. Three-dimensional mapping with ASTER stereo data in rugged topography. IEEE Transactions on Geoscience and Remote Sensing 40(10), 2241 - 2247.
143 Tuan, Y.-F., 1977. Space and place : the perspective of experience. University of Minnesota Press, Minneapolis. Urton, G., 1978. Orientation in Quechua and Incaic Astronomy. Ethnology 17, 157-167. USGS, 2003. SRTM documentation. Available at: ftp://edcsgs9.cr.usgs.gov/pub/data/srtm/Documentation/ (accessed May 16, 2003). Van Leusen, P. M. 2002. Pattern to Process: Methodological investigations into the formation and interpretation of spatial patterns in archeological landscapes. Doctoral Thesis, Rijksuniversiteit Groningen, 356 p. Vignon, F., Arnaud, Y. and Kaser, G., 2003. Quantification of glacier volume change using topograhic and ASTER DEMs. Proceedings of International Geoscience and Remote Sensing Symposium, Toulouse, France, July 21 – 25, 2003. Vitry, C., 1997. Arqueologia de alta montaña. YACHAYRUNA, Universidad Nacional de Salta, Argentina. Año 1, No.1. Warren, R. E., Ed. 1990. Predictive Modeling in archaeology: a primer. New Methods, Old Problems: Geographic Information Systems in Modern Archaeological Research. Warren, R. E. and Asch, D. L., 2000. A predictive model of archeological site location in the Eastern Prairie Peninsula. In: Wheatley, D. and Brandon, J. R., Eds., Practical applications of GIS for archeologists: a predictive modeling kit. Taylor & Francis, London. Waters, M. R., 1992. Principles of geoarchaeology : a North American perspective. University of Arizona Press, Tucson. Welch, R., Jordan, T., Lang, H. and Murakami, H., 1998. ASTER as a source for topographic data in the late 1990's. IEEE Transactions on Geoscience and Remote Sensing 36(4), 1282-1289. Wessels, R., U.S. Geological Survey Alaska Science Center, personal communication Wheatley, D. and Gillings, M., Eds., 2002. Spatial Technology and Archaeology: The archaeological applications of GIS. Taylor and Francis, London and New York. Williams Jr., R. S. and Ferrigno, J. G., Eds., 1999. Satellite image atlas of glaciers of the world - Glaciers of South America. USGS professional paper 1386-I. USGS, Wilson, J. P. and Gallant, J. C., 2000. Digital terrain analysis. In: Gallant, J. C., Ed. Terrain analysis - principles and applications. Wiley, New York. Wingenroth, M. C., 2001. Estudio palinológico de la Momia de Aconcagua. In: Schobinger, J., Ed. El santuario Incaico del cerro Aconcagua. Mendoza. Winstral, A., Elder, K. and Davis, R. E., 2002. Spatial modeling of wind-redistributed snow using terrain-based parameters. Journal of Hydrometeorology 3, Wood, J. D. and Fisher, P. F., 1993. Assessing Interpolation Accuracy in Elevation Models. IEEE Computer Graphics and Applications 13(2), 48-56. Ziólkowski, M. S. and Belan Franco, L. A., 2001. Proyecto Arqueologico Condensuyos. Boletín de la Misión Arqueológica Andina 1(3), 9-23. Zuidema, T. R., 1977. The Inca calendar. In: Aveni, F., Ed. Native American Astronomy. University of Texas Press, Austin. 219-259.
144
//$-#(7 Human sacrifices discovered on Andean peaks to the present day. Elevation Lat Long # of Age of Year Mountain Sex Origin Reference (°S) (°W) Summit Site bodies victim discovered (m) (m) 1 M 10 Inca 1985 Schobinger (2001) Aconcagua 32.48 70.03 6,959 5,250
Ampato 15.78 71.85 6,270 5,900 4 F(1), M(2) 14 Inca 1995 Thouret et al. (2001a)
Cajón 24.30 66.69 5,468 5,400 1 M 25 ? 1979 Beorchia (1985)
Coropuna 15.32 73.45 6,415 5,000 2 M adult Inca 1966 Beorchia (1985)
Chañi 22.83 67.85 6,060 6,060 1 F 6 Inca 1905 Beorchia (1985)
Chuscha 24.31 66.73 5,468 N/a 1 F N/a N/a 1922 Beorchia (1985)
Chachani 16.2 71.53 6,056 6,056 1 F 15 Inca 1896/1898 Beorchia (1985)
Esmeralda 20.22 70.12 905 905 2 M 9, 18 Inca 1977 Beorchia (1985) Reinhard and Ceruti Llullaillacu 24.71 68.53 6739 6712 3 F(1), M(2) N/a Inca 1999 (2000) Misti 15.52 71.68 5,821 N/a 6 N/A N/A Inca 1998 Thouret et al.(2001a)
Pichu Pichu 16.43 71.23 5,664 5,630 1 F 18 Inca 1963 Beorchia (1985)
Plomo 15.52 72.66 5,430 5,430 1 M 9 Other 1954 Beorchia (1985)
Quehuar 24.40 66.15 6,130 6,130 1 M 14 N/a 1974 Beorchia (1985)
Sara Sara 24.24 67.04 5,505 5,475 1 F N/a Inca 1996 Reinhard (1996) Schobinger Toro 24.36 65.66 6,160 6,130 1 M 20 Other 1964 (1966;1969)
146
//$-#(7 Mountains surveyed prior to 1985 and reported by Beorchia Nigris (1985).
SITE SITE METAKEY MOUNTAIN SITE COUNTRY PROVINCE LAT LONG ELEV ELEV(m) TYPE BN1985_001 Acay Acay3 Argentina Salta - - 5716 5658 1 BN1985_002 Aconcagua Aconcagua Argentina Mendoza - - 6959 5250 1 BN1985_003 Acotango Acotango Chile Atacama -18.42 -63.03 6050 6050 1 BN1985_004 Aguas Calientes Aguas Calientes Argentina Catamarca -27.23 -68.30 5517 5517 1 BN1985_005 Alma Negra Alma Negra Argentina San Juan - - 6120 6000 1 BN1985_006 Ampato Ampato1 Peru Arequipa - - 6270 6220 1 BN1985_007 Ampato Ampato2 Peru Arequipa - - 6270 5852 2 BN1985_008 Ampato Ampato3 Peru Arequipa - - 6270 5900 1 BN1985_009 Antofalla Antofalla Argentina Salta - - 6100 6100 1 BN1985_010 Aracar Aracar1 Argentina Salta -24.28 -67.78 6086 6086 1 BN1985_011 Aracar Aracar2 Argentina Salta - - 6086 3900 3 Ascotan de Ascotan de -21.78 -68.08 5505 5505 1 BN1985_012 Bolivia Sud Lipez Ramaditas Ramaditas1 Ascotan de Ascotan de - - 5505 4400 3 BN1985_013 Bolivia Sud Lipez Ramaditas Ramaditas2 BN1985_014 Aucanquilcha Aucanquilcha Chile Zona_Norte - - 6100 5400 2 BN1985_015 Belen Belen1 Chile Atacama -18.50 -69.08 5260 5260 1 BN1985_016 Belen Belen2 Chile Atacama - - 5260 4200 3 BN1985_017 Bismark Bismark1 Chile Santiago - - 4670 4200 2 BN1985_018 Bismark Bismark2 Chile Santiago -33.27 -70.22 4670 4670 1 BN1985_019 CerroBlanco CerroBlanco Peru Nazca -14.87 -74.83 2076 2076 1 BN1985_020 Bonete Bonete Bolivia Sud Lipez -21.75 -66.50 5656 5656 1
148 SITE SITE METAKEY MOUNTAIN SITE COUNTRY PROVINCE LAT LONG ELEV ELEV(m) TYPE BN1985_021 Bonete Grande Bonete Grande Argentina La Rioja 6412 5700 2 BN1985_022 Calcha Calcha Peru Arequipa -15.93 -71.43 5257 5257 1 BN1985_023 Carachipampa Carachipampa Argentina Catamarca -26.52 -67.48 4500 4500 1 BN1985_024 Cariquima Cariquima Chile Iquique -19.05 -68.70 5355 5355 1 BN1985_025 Cawaray Cawaray Chile Iquique -19.13 -68.72 5860 5860 1 BN1985_026 Chachani Chachani Peru Arequipa -16.20 -71.53 6056 6056 1 BN1985_027 Chani Chani3 Argentina Salta - - 6060 5790 4 BN1985_028 Chani Chani4 Argentina Salta - - 6060 5450 2 BN1985_029 Chani Chani5 Argentina Salta - - 6060 4910 3 BN1985_030 Chani Sur Chani Sur Argentina Salta - - 6000 4418 1 BN1985_031 Chaquicocha Chaquicocha Peru Huaraz -8.87 -77.77 4750 4750 1 BN1985_032 Chilques Chilques3 Chile Atacama -23.57 -67.70 5778 5778 1 BN1985_033 Chilques Chilques2 Chile Atacama - - 5778 5727 2 BN1985_034 Chilques Chilques1 Chile Atacama - - 5778 4500 3 BN1985_035 Chimberi Chimberi Argentina Tucuman -27.17 -66.08 5300 5300 1 BN1985_036 Chuculai Chuculai1 Argentina Salta - - 5420 4683 3 BN1985_037 Chuculai Chuculai2 Argentina Salta - - 5420 4713 3 BN1985_038 Chuculai Chuculai3 Argentina Salta -24.60 -68.53 5420 5420 1 Colorado Colorado Chile San Pedro de - - 5742 5723 1 BN1985_039 Atacama Colorado de Colorado de Chile Santiago - - 4500 4500 1 BN1985_040 Farellones Farellones BN1985_041 Cora Cora Cora Cora Bolivia Sur Lipez -19.55 -67.68 4800 4800 1 BN1985_042 Coropuna Coropuna Peru Arequipa - - 6415 5000 1 Curiquinca Curiquinca2 Chile San Pedro de -22.67 -67.85 5769 5769 1 BN1985_043 Atacama Curiquinca Curiquinca1 Chile San Pedro de - - 5769 4800 2 BN1985_044 Atacama BN1985_045 Dona Ana Dona Ana Chile Coquimbo -29.77 -70.12 5690 5690 1 BN1985_046 Dona Ines Dona Ines Chile Chanaral - - 5040 5040 1
149 SITE SITE METAKEY MOUNTAIN SITE COUNTRY PROVINCE LAT LONG ELEV ELEV(m) TYPE BN1985_047 Esmeralda Esmeralda Chile Iquique -20.22 -70.13 905 905 1 BN1985_048 Flechas Flechas1 Argentina San Juan -28.78 -69.67 5350 5350 1 BN1985_049 Flechas Flechas2 Argentina San Juan - - 5350 3900 3 General 6250 6000 2 BN1985_050 General Belgrano Argentina La Rioja Belgrano BN1985_051 Guana Guane Guana Guane1 Chile Atacama -18.15 -69.27 5050 5050 1 BN1985_052 Guana Guane Guana Guane2 Chile Atacama -18.15 -69.27 5050 5050 1 BN1985_053 Hualca Hualca Hualca Hualca Peru Arequipa - - 6025 5850 2 BN1985_054 Huanac Pacha Huanac Pacha Peru Arequipa - - 5920 5861 1 BN1985_055 Huaracante Huaracante Peru Arequipa -15.73 -71.52 5360 5360 1 BN1985_056 Huaychao Huaychao Peru Chavin -9.60 -77.22 4526 4526 1 BN1985_057 Illakata Illakata Peru Nazca -14.63 -74.33 4327 4327 1 BN1985_058 Iman Iman Argentina San Juan -29.23 -69.42 5413 5413 1 BN1985_059 Incahuasi Incahuasi Argentina Catamarca -27.03 -68.30 6620 6620 1 BN1985_060 Isluga Isluga1 Chile Atacama - - 5530 5200 2 BN1985_061 Isluga Isluga2 Chile Atacama -19.17 -68.83 5530 5530 1 BN1985_062 Jatamala Jatamala1 Chile Iquique -19.58 -69.13 4700 4700 1 BN1985_063 Jatamala Jatamala2 Chile Iquique -19.58 -69.13 4700 4700 1 San Pedro de -22.83 -67.83 5662 5662 1 BN1985_064 Juriques Juriques Chile Atacama BN1985_065 Lejia Lejia1 Chile Atacama -23.55 -67.77 5793 5793 1 BN1985_066 Lejia Lejia2 Chile Atacama - - 5793 5650 2 BN1985_067 Leon Leon Chile Atacama -22.15 -68.12 5760 5755 1 BN1985_068 Licancabur Licancabur1 Chile Atacama -22.83 -67.83 5921 5920 2 BN1985_069 Licancabur Licancabur2 Chile Atacama - - 5921 5830 2 BN1985_070 Licancabur Licancabur3 Chile Atacama - - 5921 5600 2 BN1985_071 Licancabur Licancabur4 Chile Atacama - - 5921 5200 2 BN1985_072 Licancabur Licancabur5 Chile Atacama - - 5921 5100 3 BN1985_073 Licancabur Licancabur6 Chile Atacama - - 5921 4600 3
150 SITE SITE METAKEY MOUNTAIN SITE COUNTRY PROVINCE LAT LONG ELEV ELEV(m) TYPE BN1985_075 Mercendario Mercendario1 Argentina San Juan 6770 6700 1 BN1985_076 Mercendario Mercendario2 Argentina San Juan - - 6770 6500 1 BN1985_077 Mercendario Mercendario3 Argentina San Juan - - 6770 6200 1 BN1985_078 Mercendario Mercendario4 Argentina San Juan - - 6770 5150 1 BN1985_079 Mercendario Mercendario5 Argentina San Juan - - 6770 5000 2 BN1985_080 Miniques Miniques Chile Atacama -23.82 -67.77 5916 5916 1 BN1985_081 Mino Mino1 Chile Atacama - - 5800 4300 3 BN1985_082 Mino Mino2 Chile Atacama -21.18 -68.62 5800 5800 2 BN1985_083 Mino Mino3 Chile Atacama - - 5800 5400 2 BN1985_084 Miscanti Miscanti Chile El Loa -23.67 -67.70 5622 5622 1 BN1985_085 Mismi Mismi Peru Arequipa -15.52 -71.68 5596 5596 1 BN1985_086 Misti Misti1 Peru Arequipa - - 5821 4750 2 BN1985_087 Misti Misti2 Peru Arequipa - - 5821 5720 1 BN1985_088 Mogotes Mogotes1 Argentina San Juan - - 5380 3900 3 BN1985_089 Mogotes Mogotes2 Argentina San Juan -28.63 -69.57 5380 5380 1 BN1985_090 Negro Overo Negro Overo1 Argentina La Rioja -28.93 -67.85 6050 6050 1 BN1985_091 Negro Overo Negro Overo2 Argentina La Rioja - - 6050 3900 3 BN1985_092 Negro Overo Negro Overo3 Argentina La Rioja - - 6050 5700 2 Pabellon del Pabellon del Inca Chile Atacama - - 5110 5110 1 BN1985_093 Inca BN1985_094 Pachatsusan Pachatsusan1 Peru Cuzco - - 4842 4400 2 BN1985_095 Pachatsusan Pachatsusan2 Peru Cuzco - - 4842 4832 1 BN1985_096 Palas Palas1 Chile Huasco - - 4993 4500 2 BN1985_097 Palas Palas2 Chile Huasco -28.95 -69.97 4993 4993 1 BN1985_098 Palpana Palpana Chile Atacama - - 6050 6006 5 BN1985_099 Paniri Paniri Chile Atacama - - 5946 5937 1 San Pedro de - - 6360 5946 1 BN1985_100 Parinacota Parinacota Chile Atacama BN1985_101 Pastos Grandes Pastos Grandes Argentina Salta - - 5810 5810 1 BN1985_102 Patos Patos Argentina Catamarca -27.28 -68.85 6250 6250 1
151 SITE SITE METAKEY MOUNTAIN SITE COUNTRY PROVINCE LAT LONG ELEV ELEV(m) TYPE BN1985_104 Peladeros Peladeros Chile Santiago -33.58 -70.22 3310 3310 1 BN1985_105 Penitentes Penitentes Argentina Mendoza -32.87 -69.88 4258 4258 1 BN1985_106 Pichu Pichu Pichu Pichu1 Peru Arequipa - - 5650 5634 1 BN1985_107 Pichu Pichu Pichu Pichu2 Peru Arequipa -16.43 -71.23 5650 5650 4 BN1985_108 Pichu Pichu Pichu Pichu3 Peru Arequipa - - 5650 5350 2 BN1985_109 Pichu Pichu Pichu Pichu4 Peru Arequipa - - 5650 4810 3 BN1985_110 Pichu Pichu Pichu Pichu5 Peru Arequipa - - 5650 4820 3 San Pedro de -23.28 -67.63 6060 6060 1 BN1985_111 Pili Pili Chile Atacama BN1985_112 Plomo Plomo1 Chile Santiago - - 5425 5400 1 BN1985_113 Plomo Plomo2 Chile Santiago - - 5425 5200 2 BN1985_114 Plomo Plomo3 Chile Santiago - - 5425 3400 3 BN1985_115 Potosi Potosi Bolivia Potosi -19.60 -65.67 4985 4985 1 Pular Pular Chile San Pedro de -24.20 -68.08 6225 6225 2 BN1985_116 Atacama BN1985_117 Puntiudos Puntiudos Chile La serena - - 2000 2000 1 San Pedro de -23.12 -68.67 4300 4300 1 BN1985_118 Quimal Quimal Chile Atacama BN1985_119 Sairecabur1 Sairecabur1 Chile Atacama -22.70 -67.87 5970 5970 1 BN1985_120 Sairecabur2 Sairecabur2 Chile Atacama - - 5970 4800 2 BN1985_121 Sara Sara Sara Sara Peru Arequipa - - 5505 5475 1 BN1985_122 Saye Saye Chile Atacama -22.63 -68.02 4884 4883 4 BN1985_123 Socompa Socompa3 Argentina Salta - - 6050 5000 2 BN1985_124 Socompa Socompa4 Argentina Salta - - 6050 4300 3 BN1985_125 Taapaca Taapaca Chile Atacama - - 5615 5415 1 Quebrada - - 1300 1300 1 BN1985_126 Talapata Talapata Chile Camarones BN1985_127 Tambillos Tambillos Argentina San Juan - - 5747 5427 1 BN1985_128 Tata Jachura Tata Jachura Chile Iquique -19.50 -69.10 5252 5252 1
152 SITE SITE METAKEY MOUNTAIN SITE COUNTRY PROVINCE LAT LONG ELEV ELEV(m) TYPE Tolar del Tolar del Carmen Chile Atacama -21.97 -70.10 2220 2220 1 BN1985_130 Carmen BN1985_131 Toro Toro1 Argentina San Juan -29.15 -69.80 6380 6380 1 BN1985_132 Toro Toro2 Argentina San Juan - - 6380 6300 1 BN1985_133 Toro Toro3 Argentina San Juan - - 6380 3500 3 BN1985_134 Tortolas Tortolas1 Argentina San Juan -29.93 -69.92 6332 6332 1 BN1985_135 Tortolas Tortolas Argentina San Juan - - 6332 5200 2 BN1985_136 Tortolas Tortolas Argentina San Juan - - 6332 4000 3 NO BN1985_137 Tulapalca Tulapalca Chile Atacama - - 4780 4726 DATA BN1985_138 Viracochan Viracochan Peru Cuzco - - 3670 3668 2 BN1985_139 Yanagaga Yanagaga Peru Ancash - - 4764 4729 2
153 //$-#(7 Mountains surveyed in the last decade and reported by Ceruti (1999a; 1999b) and Reinhard and Ceruti (2000).
SITE SITE METAKEY MOUNTAIN SITE_NAME LAT LONG COUNTRY PROVINCE ELEV BASE ELEV(M) TYPE CE1999_001 Acay Acay1 -24.38 -66.17 Argentina Salta 5716 4000 5716 1 CE1999_002 Acay Acay2 -24.42 -66.14 Argentina Salta 5716 4000 5320 3 CE1999_003 Aconquija Abra_Colorada -27.14 -66.06 Argentina Salta 5550 1550 4790 2 CE1999_004 Aconquija Becovel -27.11 -66.07 Argentina Tucuman 5550 1550 3850 3 CE1999_005 Aconquija Campo_Colorado -27.16 -66.05 Argentina Tucuman 5550 1550 4670 2 CE1999_006 Aconquija Cuevas1 -27.18 -66.02 Argentina Tucuman 4900 1550 4395 3 CE1999_007 Aconquija Cuevas2 -27.19 -66.01 Argentina Tucuman 4900 1550 4295 3 CE1999_008 Aconquija Cuevas3 -27.17 -66.03 Argentina Tucuman 5550 1550 4765 4 CE1999_009 Aconquija Cuevas4 -27.18 -66.03 Argentina Tucuman 5550 1550 5005 1 CE1999_010 Arizaro Arizaro -24.42 -67.99 Argentina Salta 5874 3500 5874 2 CE1999_011 Azufre Azufre1 -24.41 -66.79 Argentina Salta 5840 3800 5840 2 CE1999_012 Bayo Bayo1 -26.05 -66.10 Argentina Salta 4612 1512 4380 1 CE1999_013 Bertrand Bertrand -26.75 -68.12 Argentina Catamarca 5207 3800 5207 2 CE1999_014 Cachi Cachi1 -25.03 -66.34 Argentina Salta 6380 3800 5352 2 CE1999_015 Cachi Cachi2 -25.01 -66.35 Argentina Salta 6380 3800 5636 2 CE1999_016 Cachi Cachi3 -25.01 -66.34 Argentina Salta 6380 3800 5588 2 CE1999_017 Cajon Chuscha - - Argentina Salta 5468 1512 5400 1 CE1999_018 Cajon Chuscha1 - - Argentina Salta 5468 1512 5165 2 CE1999_019 Cajon Chuscha2 - - Argentina Salta 5468 1512 5185 2 CE1999_020 Castillo Castillo2 -24.38 -65.66 Argentina Salta 5565 1896 5620 1 CE1999_021 Cerro del Cerro_del_Medio -24.24 -67.04 Argentina Salta 4980 3780 4929 1 Medio CE1999_022 Chani Chani -24.05 -65.73 Argentina Salta 6060 1896 6000 1 CE1999_023 Collaguaima Collaguaima1 -22.80 -66.60 Argentina Jujuy 5635 4100 5635 1 CE1999_024 Collaguaima Collaguaima2 -22.84 -66.61 Argentina Jujuy 5635 4100 5264 2 CE1999_025 Granada Granada1 -22.56 -66.55 Argentina Jujuy 5697 4100 5697 2 CE1999_026 Granada Granada2 -22.56 -66.55 Argentina Jujuy 5697 4100 5670 2
154 SITE SITE METAKEY MOUNTAIN SITE_NAME LAT LONG COUNTRY PROVINCE ELEV BASE ELEV(M) TYPE CR2000_002 Lullaillacu Lullaillacu2 -24.68 -68.50 Argentina Salta 6739 4000 4960 3 CR2000_003 Lullaillacu Lullaillacu3 -24.70 -68.51 Argentina Salta 6739 4000 5200 3 CR2000_004 Lullaillacu Lullaillacu4 -24.70 -68.50 Argentina Salta 6739 4000 5615 2 CR2000_005 Lullaillacu Lullaillacu5 - - Argentina Salta 6739 4000 6260 2 CR2000_006 Lullaillacu Lullaillacu6 -24.72 -68.52 Argentina Salta 6739 4000 6555 2 CR2000_007 Lullaillacu Lullaillacu7 -24.72 -68.54 Argentina Salta 6739 4000 6712 1 CE1999_027 Macon Macon1 -24.48 -67.26 Argentina Salta 5611 3900 5611 1 CE1999_028 Malcante Malcante1 -25.08 -65.85 Argentina Salta 5226 4000 5202 1 CE1999_029 Morado Morado_de_Iruya -22.92 -65.18 Argentina Salta 5008 3500 5008 1 CE1999_030 Pano Pano1 -24.24 -65.71 Argentina Salta 5530 1896 5530 1 CE1999_031 Pocitos Pocitos1 -24.26 -55.99 Argentina Salta 5060 3850 5060 1 CE1999_032 Quehuar Quehuar1 -24.33 -67.07 Argentina Salta 6102 3800 6102 1 CE1999_033 Quehuar Quehuar2 - - Argentina Salta 6102 3800 6100 1 CE1999_034 Quehuar Quehuar3 - - Argentina Salta 6102 3800 5000 2 CE1999_035 San Miguel de San_Miguel_de_ 2 -24.54 -66.14 Argentina Salta 5750 3500 5750 la Poma la_Poma_1 CE1999_036 Sisilera Sisilera -23.52 -65.27 Argentina Jujuy 4742 3700 4742 2 CE1999_037 Socompa Socompa1 -24.41 -67.25 Argentina Salta 6050 3850 6050 1 CE1999_038 Socompa Socompa2 -24.42 -66.27 Argentina Salta 6050 3850 4675 1 CE1999_039 Tipillas Tipillas1 -27.17 -66.07 Argentina Tucuman 5450 1550 5396 1 CE1999_040 Tipillas Tipillas2 -27.38 -66.27 Argentina Tucuman 5450 1550 5220 2 CE1999_041 Tultul Tultul1 -24.20 -67.11 Argentina Salta 5352 3850 5352 NO DATA CE1999_042 Tuzgle Tuzgle1 -24.06 -66.49 Argentina Salta 5595 3850 5595 1 CE1999_043 Tuzgle Tuzgle2 -24.06 -66.49 Argentina Salta 5595 3850 5500 2 CE1999_044 Vilama Vilama1 -22.76 -66.98 Argentina Jujuy 5678 4100 5415 NO DATA CE1999_045 Volcan Blanco Volcan_Blanco1 -24.53 -67.86 Argentina Salta 4200 3500 4190 2 CE1999_046 Zucho1 Zucho1 -23.62 -65.28 Argentina Jujuy 5009 2995 4844 1 CE1999_047 Zucho2 Zucho2 -23.63 -65.27 Argentina Jujuy 5009 2995 5009 2
155 156