Final Project: Predictive Modeling of native sites in Dartmouth and Westport, MA

The study area for this project, located in southeastern MA, was home to native Wampanoags during the Archaic, Woodland, and Contact Periods. By 1800, their permanent and seasonal settlements along the shores of the Slocums, Little, Apponagansett, and Westport Rivers had disappeared.

Seasonal and permanent settlements had been reported along the shores of rivers from 1602 until 1800.  Trails and waterways were transportation routes used first by natives and later by colonial settlers.

By 1856, natives were no longer living in settlements but were dispersed throughout the community.

The Digital Elevation Model (DEM) is a starting point for creating a predictive model.

Elevation is divided into classes to show areas of similar values.  

A contour map is created so that slope and aspect maps can be derived from it.  

The slope map shows areas where the terrain is steep near the rivers.  These areas would not have been ideal for settlements.

Aspect illustrates places where the land faces N,S,E, or W.  Those areas facing south were often favored for settlements because of their warmer climate and protection from harsh winds.  Eastern and western facing slopes were the next most desirable.  North facing land was least desirable.

The resources found in rivers, streams, lakes, ponds, marshes and coastal areas were used by native people both seasonally and year-round.  The study area shows a far-reaching network of these environments.

Surface geology could have played a role in site selection.  However, more work needs to be done to identify characteristics that would have been favorable or not favorable, based on what is known about identified native sites.

To predict the location of unknown sites, environmental variables such as Elevation, Slope, and Aspect can be weighted and combined into a Weighted Overlay map which creates a pictures of  areas where sites are more or less likely to be found.

Several types of analysis can be done using known sites and random points to see if the predictive model is useful, and how the variables relate to one another and to the sites.  Elevation was a particularly useful variable in this model.

Biscayne Shipwrecks: Analysis

The sea-bed attributes most commonly associated with shipwrecks are used to create a map showing where wrecks are likely to be found.  The first step is to reclassify this data, giving more weight to these classes, and less to attributes were shipwrecks are absent.

The benthic map shows areas in red are most likely to have shipwrecks.  The bathymetric map is based on sea depth.

The two classified maps are combined to create a predictive map.  The benthic map has a greater influence (70%) on the overlay; the bathymetric map has less (30%), reflecting the greater importance of sea-bed type over sea depth in the location of existing shipwrecks.

Biscayne Shipwrecks Week 1

Shipwrecks are common in this area of Florida where the Gulf Stream hugs the coast.  Reefs and storms were two common perils that claimed ships over the last several centuries.  Some sites are known, but others remain uncharted.  By examining bathymetric maps of the sea floor, combined with historic and modern charts and data on characteristics of the bottom type (reef, pavement, etc.), patterns can be detected, and models created to predict where other sites may exist.  

Scythian Burial Mounds: Report

The process of modeling the spatial distribution of Scythian burial mounds in Tuekta, Siberia, focuses on the relationship of the mounds to their environment.  First, a DEM of the region was clipped to the study area, and three secondary surfaces were created from it – slope, aspect, and elevation.  Slope was reclassified to weigh more heavily in favor of flat or gently sloping terrain.  Aspect was reclassified to favor southern facing areas.  Elevation was reclassified to emphasize areas with a similar elevation range.  The Tuekta mounds sites were digitized into a point shapefile to be used in the OLS regression analysis, and then an additional 100 random points were created.  These were merged, creating a new shapefile for the dependent variable – the presence or absence of sites.  The data was edited to reflect the slope, aspect, and elevation of each point, and to populate each point with XY data.  The data could now be used in the OLS Regression.

The OLS Regression was done on the model to identify trends and to see relationships between the dependent and explanatory variables. The results showed that all three explanatory variables were contributing to the model (Aspect = 0.088971, Slope = 0.130735, Elevation = 0.581643).  These positive values are expected with models using reclassified values that weight the higher numbers to aspects of the landscape that are favorable for site location. The Adjusted R-squared value was .718372, indicating a high percentage of site presence or absence would be predicted using a model with these variables. The Spatial Autocorrelation showed the clustering of sites.  The p-value of 0.000 indicates a 100% confidence level that the patterning is completely non-random, and that there is a spatial variable influencing the data.  The z-score of 13.3590348071 indicates that there is a less than 1% likelihood that this clustered pattern could be the result of random chance.  There are areas where the model has under-predicted the presence or absence of sites.  These include areas on the edges of the valleys in areas of transition to higher elevation and increasing slope, and along the valley floor in the eastern part of the study area. Other variables could be included to further refine the model, such as geological features, soil type, and vegetation.  Proximity to rivers could also be included as a variable.  This model is limited because of its use of only three variables, so the addition of these others would make it more precise, but clearly it shows the significance of these variables.  A Geographically Weighted Regression model is another option, because it is suited to a regional scale and clustered data, while the OLS regression model is better for non-clustered data.  However, the dataset used here was not large enough for GWR Regression.  

Scythian Mounds: Analysis

The Tuekta mounds are located in a long valley surrounded by mountains, as can be seen in the contour map and the reclassified Elevation map.  The Slope map is weighted so that the greatest weight (4) goes to  flat or gently sloping terrain, found on the valley floor.  The Aspect map highlights areas in bright green that face in a southerly direction.  Analysis of these maps reveals a pattern of location characteristics favored by the builders of the mounds.  This information can be used to better understand the experience of those who built the mounds and for people for whom the mounds were part of the landscape.  It can also help predict where other mounds may be located.

Scythian Mounds: Analysis

The Tuekta mounds are located in a long valley surrounded by mountains, as can be seen in the contour map and the reclassified Elevation map.  The Slope map is weighted so that the greatest weight (4) goes to  flat or gently sloping terrain, found on the valley floor.  The Aspect map highlights areas in bright green that face in a southerly direction.  Analysis of these maps reveals a pattern of location characteristics favored by the builders of the mounds.  This information can be used to better understand the experience of those who built the mounds and for people for whom the mounds were part of the landscape.  It can also help predict where other mounds may be located.

Scythian Burial Mounds Part I

These Scythian burial mounds are located in a valley in Siberia.  A mosaic DEM illustrates the variation in elevation in this region.  By examining the characteristics of the landscape, it may be possible to understand the reasons why the burial mounds were located here.

The inset map shows a georeferenced image of the mounds, overlaid on a DEM of the study area.

Predictive Modeling

Predictive modeling can be a useful tool for archaeologists trying to narrow down likely areas where cultural resources may be found.  Environmental factors such as proximity to water, soil and vegetation types, elevation, slope, and aspect (which direction a slope faces) are understood to have some value in predicting where settlement and activity occurred in the past.  This type of information is becoming increasingly easy to acquire, and the resulting models can be used to narrow down areas where field survey is more likely to result in finding cultural remains.  The map above shows a weighted overlay map indicating areas where there is a high, medium and low probability of finding archaeological material.  By choosing specific variables, and by weighting them according to their relative importance, archaeologists can guide field survey, saving money and time.  However, this technique has been criticized for being too focused on environmental variables, and not taking into account cultural factors that would likely impact choices made by those who settled in a particular area.  Predictive modeling is currently used in CRM work most often.  When used in an academic situation, it is critical that substantial ground-truthing and field survey are conducted in addition to the use of the predictive model.

Finding Angkor's Hidden Sites

While the use of satellite images for identifying potential archaeological sites is successful in several parts of the world, it is less so in others.  In Cambodia, where many monumental stone structures are hidden in the dense tropical vegetation, and where land mines and unexploded bombs pose a threat to those conducting ground survey, it is possible to use training samples to classify images for the identification of previously unknown sites.  However, it is problematic.  This map shows a supervised classification of the area surrounding the core of Angkor's monumental architecture.  The classification does identify several areas where potential sites may be located, and indeed one area, Phnom Kulen, has recently been identified as a previously undiscovered urban landscape associated with early Angkorian settlement (Evans et al, 2013).  The classification is not good at distinguishing the stone monuments from other classes, such as dense forest and water.  However, patterning that points to hydraulic features and geometric lines associated with Angkorian architectures is visible.

It appears that the use of lidar in this situation is far superior to the results that can be achieved using Landsat imagery, as seen here.  The following website and article provide additional information.

http://angkorlidar.org/publications/

Evans, D. H., R. J. Fletcher, C. Pottier, J.-B. Chevance, D. Soutif, B. S. Tan, S. Im, D. Ea, T. Tin, S. Kim, C. Cromarty, S. De Greef, K. Hanus, P. Bâty, R. Kuszinger, I. Shimoda and G. Boornazian. 2013. “Uncovering archaeological landscapes at Angkor using lidar,” Proceedings of the National Academy of Sciences of the United States of America 110: 12595-12600

Band Combinations, Training Samples, and Supervised Classification

Different band combinations can be used to bring out various characteristics in the environment.  The top two maps here use two different combinations of Landsat satellite imagery.  The NVDI map uses a False Color image composite, which joins bands 2,3, and 4.  When the NVDI process is added, the negative ouputs show up as red.  Bare rocks, sand, and snow have an output close to zero, and those with a higher measure of "greenness" have higher values.  This allows dense vegetation, like tropical rainforest, to show up clearly.  The combination of bands 4,5, and 1 in the second map is used to differentiate vegetation that is stressed and sparse with healthy vegetation.  

The map at the bottom shows a map resulting from a supervised classification,  First a training signature file was created by drawing polygons around samples of each class.  Known Mayan pyramid sites were used to create this class's sample.  The classification shows the locations of possible new sites, and creates a tool to be used in survey and ground-truthing.

Maya Pyramids Part I

This series of maps shows several different band combinations that can be useful in locating potential sites.  Landsat imagery from the USGS can be viewed in ways which reveal certain characteristics depending on the combination of bands.  ArcMap Image Analysis and Processing tools are used to create combinations suitable for different purposes.

For example, the Landsat Band 8 is a high resolution panchomatic view, used here by itself to show the location of the Mirador pyramid.  It can also be used to create a sharper composite image, using the Pan-sharpening option. 

The Natural Color map shows a band combination that displays a color image to visualize data like a color photo using Bands 1,2 and 3, which show visible light.  Band 1 distinguishes soil from vegetation, Band 2 is useful for showing which plants are stressed, and which are more healthy, and Band 3 is also used to highlight vegetation.

The False Color map uses Bands 2,3,and 4, adding the Near Infared which emphasizes biomass content.  This band combination is most useful for the dense jungle where the Mirador pyramid is located because the red band (#3) indicates areas where chlorophyll is being absorbed, and the NIR band (#4) indicates areas of high refelectivity of plant materials.  The NDVI(Normalied Difference Vegitation Index) tool is used to show relative biomass in the image.

Mapping the History of New Bedford

Printed historic maps have been used in the past to answer questions about social, environmental, and political dynamics within a regional landscape.  

This study uses GIS analysis to begin to identify and understand changes in political, social, and economic life that took place between the post-War of 1812 economic boom tied to the whaling industry, and the peak of manufacturing in New Bedford, MA.   Several families, including the Kemptons, Rotches, Rodmans, Hathaways, Russells, Allens, and Morgans, were prominent during the early development of the city.  Their influences in terms of land and business ownership shaped the political, social, and economic landscape.  Free African-Americans and immigrants from the Azores and Cape Verde Islands working on the waterfront and in the whaling industry shaped New Bedford as well.  As the industrial base shifted from whaling to manufacturing, and as the city expanded in terms of population, ethnic diversity, and physical size, one would expect to see the development social and spatial boundaries between those who owned property and were successful in the initial establishment and later growth of the city into a whaling and commercial capital, and those who settled later as the city expanded to the north and south.  The GIS analysis presented here shows where and how these boundaries developed. 

By 1815 a social, political and industrial center has been established, dominated by those who owned valuable waterfront property and businesses related to whaling and shipping.  The 1850 map shows many of the Registered Historic Sites clustered north of the city center, several blocks away from the waterfront.  These were built during the heyday of whaling and shipping and illustrate the prosperity and growth the city was experiencing.  Many of these are modest one-family homes, not businesses or mansions built by those involved in whaling and commerce.  Those were being built, in smaller numbers, to the south of the center and close to County Street at the top of the hill overlooking the busy harbor.  Two mills had been built by 1850, but housing was not yet being built specifically to accommodate mill workers.  The environmental context here relates to the developing social inequality in New Bedford, as those with lower incomes settle in areas away from the center. 

Central business district showing the establishment of several prominent families as dominant property holders.

The first mill is built, whaling is in decline.

 The housing boom that can be seen in both the 1871 and 1891 maps reflects the increase in population between 1850 and 1920 as immigrants came to work in the mills.

Urban growth to the north and south of the city's commercial center, and to the west of County St.

The 1891 map in particular shows clusters of identical rows of housing built near the Wamsutta and Howland Mills complexes.  As foreign-born workers arrived in greater numbers, social differences increased.  By 1911 it is possible to see this clearly when analyzing the surnames of property owners in various neighborhoods. Properties closer to industrial areas, in particular mills and other factories, show an abundance of surnames that reflect the immigrant make-up of these parts of the city.   By 1900, over 40% of New Bedford’s population was foreign born, the majority being Portuguese, Azorian, and Cape Verdean.  Other immigrant groups included French-Canadians, Irish, and English.  Property along County Street, and still to some extent in the downtown center, is still very much dominated by surnames from the past whaling and commercial era.  This is an example of the political landscape “constantly being defined and redefined to further accentuate the social boundaries that underlie ideologies of political order” (Kosiba 2013). 

Growth in mill housing construction.

1911: Near-peak manufacturing production

1911: Pattern of land ownership for elite families

1911: Housing in the Wamsutta Mills area, inset showing highway construction in 1960's and 1970's.

The city of New Bedford has a rich and colorful history which should continue to be preserved and celebrated.  Several organizations are committed to this mission, including the New Bedford Whaling Museum and the Waterfront Historic Area League (WHALE).  GIS analysis of historic maps can be extremely valuable in supporting it.  Research that focuses on the diverse aspects of urban development spreads the focus of preservation efforts to include the history of all social groups, not just those who are most often credited with the city’s establishment and success in whaling and commerce.  Ultimately this will help ensure that the entire picture of New Bedford’s development is represented in preservation efforts. 

The findings of this study reveal patterns of settlement in New Bedford as it transformed from a small village to a center of manufacturing.  By investigating the demographic makeup of the people living here at different points in history, and considering the ways in which the political landscape shapes social differences, a more complete understanding of the complexities of urban development here can occur. 

Classified Images

These two images of the same area reflect two types of classification - the top one is Unsupervised, where ArcGIS classifies the land cover based on a selected number of land types, in this case 8.  It creates its own signature first, then does the classification.  As you can see, there are errors and overlaps in the land types, and it is not very accurate.

The second image shows a Supervised classification, where I created signatures for 5 different land cover types based on 30 points spread out to capture the land types most accurately, especially in the area of the Cahokia Monk's Mound.  I also used only 5 classes.  Clearly this was much more accurate and effective.  The Unsupervised classification is a good starting point, but it is important to further refine that information for accuracy.

3D Map Models

This week's lab gave us a taste of what is one of the coolest new ways of visualizing archaeological data, in my opinion - 3D models.

The first task was to create a 3D box representing the study area.  Then points representing shovel test pits were extruded upwards into "poles" with 3 different colored sections, each representing the depth of that layer in that particular test pit.  Then rasters were created from the data, and an interpolated surface resulted, showing each layer as a continuous surface with elevation based on the original point data.

A third task was to do a fly-through of the 3D scene showing the extruded test pits.  Here is a link to the video.  Clearly I am not a fantastic pilot - more practice is necessary.

https://www.youtube.com/watch?v=sFUhG8E3kM0


The last task involved taking a series of points representing the path of a proposed pipeline that would cut across the study area, and extruding them to show the levels of each layer at each point in the cross section/proposed path.  For some reason the shapefile with the elevations did not have z values, so I wasn't able to extrude the points and complete this part.

3D Map Models

This week's lab gave us a taste of what is one of the coolest new ways of visualizing archaeological data, in my opinion - 3D models.

The first task was to create a 3D box representing the study area.  Then points representing shovel test pits were extruded upwards into "poles" with 3 different colored sections, each representing the depth of that layer in that particular test pit.  Then rasters were created from the data, and an interpolated surface resulted, showing each layer as a continuous surface with elevation based on the original point data.

A third task was to do a fly-through of the 3D scene showing the extruded test pits.  Here is a link to the video.  Clearly I am not a fantastic pilot - more practice is necessary.

https://www.youtube.com/watch?v=EZ5vKog3rx4

The last task involved taking a series of points representing the path of a proposed pipeline that would cut across the study area, and extruding them to show the levels of each layer at each point in the cross section/proposed path.  For some reason the shapefile with the elevations did not have z values, so I wasn't able to extrude the points and complete this part.

Surface Interpolation

Above are several examples of ways in which surface data can be visualized to see patterns in settlement in a region.  In this case, points connected to data on artifact numbers and population estimates were used to show where communities were located in the past.  By looking at these patterns, one can infer the level of local and inter-community communication in past societies.  

Oaxaca Survey Grids - Georeferencing and Digitizing

  






































Paper maps and documents are often the primary sources of historical data and site reports.  This week we focused on how to take this type of data and incorporate it into ArcMap documents.  Being able to view and examine landscape data in this way is an important new method for analysis and interpretation in settlement pattern studies.

Maps and site data from a published report of a survey done in Oaxaca, Mexico were available as scanned files in .pdg format.  The maps were made into jpegs and then georeferenced to a topographic basemap.  Then they were digitized: grid squares, soil maps, and site plan maps.  By combining these map layers, archaeologists can make inferences about settlement patterns and how these relate to environmental conditions.

The last step involved joining data in an Excel spreadsheet from the site report with the newly created shapefiles showing the sites in each grid.  In this case, the resulting data shows the low population figures for the period IIIB sites in grid N9E8,

Looking at this map, it is difficult to make a guess about whether or not population size was a function of land type, since only 3 sites are involved.

Georectifying James Cook's 1785 Map of Macao

Historical maps can be fantastic sources of information, but they are rarely accurate enough to layer over current images without going through the process of georeferencing and rectifying.  This map shows a georectified map of Macao in 1785 layered over a recent image.

To do this, there must be at least 2 layers, and control points are added between the historic map and the modern one, using points that can be identified on both maps.  In this lab, we also added a topographic map to better identify specific locations.  As more points are added, the historic map shifts in location so that the 2 maps line up as much as possible.  In this case, 10 points were added before the map was rectified using a type of sampling called Cubic Convolution, which is used for data like images and photographs.

The purpose here is to highlight the changes in landscape and environment over time.  Other data can be examined using this method, such as population trends and settlement patterns. It can be used at a variety of scales, from neighborhoods and cities, to larger regions.

GIS Methods for Incorporating Historic Records and Documents

Historic records and documents can be incorporated into GIS maps, which is incredibly useful for examining and interpreting the past. The example above combines information from a variety of sources - a historic map from the 1890's-1920's, a user-created map of the Freedom Trail from ArcGIS Online, census data, an image, and informational text.  In addition, hyperlinks were created so that when the user clicks on Paul Revere's House using the Hyperlink tool, they can access the census and see the image.  An HTML pop-up was added to the map, so when you open the link it takes the user to the Google Maps Street View scene in front of the house.  It's easy to see how much information can be pulled together this way.

This is a phenomenal tool for organizing, visualizing, and interpreting a huge variety of historical data. Even more can be learned from using it in tandem with data in other fields.  Readings this week focused on two methods for using these methods to identify and document sites that were threatened, and to prioritize urgency and inform policy decisions.  In Peru, aerial photographs from the 1940's and Google Earth images spanning more recent years were used to document looting, and to identify additional cemeteries.  It will continue to be used to monitor sites and to help policy makers decide how to manage the resources to prevent more looting.  In Georgia, GIS methods were used to determine what areas were threatened by erosion, what sites existed there, and which ones were most at risk.  These are just two examples.  The potential for these methods of analysis in archaeology is limitless.

Mapping Jordanian Archaeological Sites

This map of Jordanian Archaeological Sites was made by creating a new file database, opening a new map document, adding a basemap, and creating points to show the location of sites.  The site of Petra was created by creating a new feature class and adding fields for lat/long and notes describing the site.  After the shapefile was added to the map, it was opened for editing so that the lat/long and description information could be manually added.  To do this, you can either click to add the point to the map, or you can enter the "absolute XY"  in decimal degrees.  This is fine if you only have a limited number of features. Another way, better for large numbers of features, is to add points by importing the coordinates from an Excel file. Location information found using the META website was added to the database, and then a new feature class was created from the XY table of the Excel spreadsheet. The output was a new feature class of archaeological sites, saved to my geodatabase and added to the map as a new layer.

Many of the sites shown here have been heavily looted and damaged.  To help reduce these incidents, a system was set up to provide information about the location and current status of each site.  The database, called the Middle Eastern Geodatabase for Antiquities(MEGA), is open-source, and uses Google Earth for imagery, so it is free and user-friendly.  There is even training available.  Users are granted varying degrees of access - Guests have limited query access, and at the top end, Administrators have the ability to add to and edit the information.  By allowing such open access, the greatest number of people can be involved in helping to preserve archaeological resources.  Logistically, it is much more readily updated than if information was gathered by field survey or using aerial photography, both of which are slower and also more expensive methods.

The benefits of having such open access must be weighed against the risks of having the data available to people who may misuse it.  Site locations are seen in the map above, and published material is often quite specific about not only location, but also the numbers and types of material remains which could become the object of looting.  

Decisions about granting public access to data is guided by ethical principles adopted by archaeological organizations including the SAA, AAA, AIA, RPA and SHA that state the responsibilities to the profession, the stakeholders, and the integrity and preservation of resources.   These ethical standards are helpful as archaeologists try to balance the different points of view of various stakeholders who may be working with a different set of ethical standards and goals.

Using Clip and Query to Examine and Interpret Historical Data

 

This lab focused on using the clip and query functions in ArcMap to answer questions about the Chicago Fire of 1871 - where it started, the extent of the fire's destruction, the direction the fire travelled and where the wind was coming from, and how Chicago grew after the fire in terms of geographic expansion, the building of new landmarks, and the restructuring of some of the wards, especially in the fire zone. 

 

Another useful skill we learned this week was how to take an Excel file with Lat./Long. information for all the landmarks in the Chicago area, and create a new shapefile showing their locations as points on the map. That file can then be used to select for various attributes, such as address or year built.

 

More data is available within these layers that would be useful to answer different questions - clearly these tools will be very useful in many situations.