Final Project - Chaco Canyon: A Site Prediction Model

For my final project, I chose to create a predictive model for Chaco Canyon, located in the San Juan Basin of northwestern New Mexico. This region is characterized by its extensive Puebloan ruins and large Chaco Great Houses. The purpose of this study was to create a predictive model for potential site locations based off generated secondary surfaces from a Digital Elevation Model (DEM) of the study area. These secondary surfaces would then be used as input towards a Weighted Overlay model and an Ordinary Least Squares (OLS) linear regression analysis. These methods are intended to aid in identifying trends, relationships between variables, and possible missing variables for the current predictive model and perhaps future ones.

All analysis was conducted using Esri’s ArcMap program. Data used for the study was gathered from various USGS data download websites and consisted of a DEM, an aerial imagery raster, and a hydrography shapefile. A polygon boundary for the study area and point shapefiles for known site locations were created in ArcCatalog to be used in the study. The DEM, aerial raster, and hydrography data were all clipped to the study area boundary to reduce tool process time and produce results relevant to the study.

Slope was calculated first from the DEM using the Slope (Spatial Analyst) tool. In the Symbology tab of the slope raster’s Properties, the color scheme was set to display the Slope scheme for better visual understanding. Next, the slope raster’s categories were transformed into more meaningful divisions through the Classify button in the Symbology tab. The Classification was set to the Method of Defined Interval, with an Interval Size of 12. After symbolizing the slope raster appropriately, it was then reclassified using the Reclassify (Spatial Analyst) tool. In the dialog box that appears for the tool, the Old Values reflect the category changes I made previously for the raster and are left as is, since they represent the range of slope breaks for the data. Meanwhile, the New Values were changed to give a higher weight to the most desirable slope for habitation.The values were inversely weighted as follows: 3 = 0-12⁰, 2 = 13-24⁰, 1 = 25-36⁰, and 1 = 37-48⁰.

 The next secondary surface to be calculated from the DEM was aspect, which was created using the Aspect (Spatial Analyst) tool. The output aspect raster was also re-symbolized by setting its color scheme to the Aspect scheme in the Symbology tab of its Properties. Next, the Reclassify (Spatial Analyst) tool was used to reclassify the aspect raster into appropriate categories . The Old Values column for this raster display the numbers 0-360 divided into 11-12 categories, which represent the numeric directional facing of the compass. In the New Values column, this was simplified by breaking down the directional facing into three categories, with a higher weight assigned to the most desirable directional facing. The values were inversely weighted as follows: 1 = north, 2 = east/west, and 3 = south.

This area is relatively consistent in terms of elevation, there only being a maximum distance of 300 meters between the highest (~2100m) and lowest elevations (~1800m). In the Classification dialog box of the DEM, the Classification Method was set to Manual, and the Break Values were set at 100 meters apart, creating three categories. The DEM was then ready to be reclassified with the Reclassify (Spatial Analyst) tool . The New Values were then inversely weighted, with the most desirable elevation assigned with a higher weight. The values are as follows: 1 = 2000-2100 meters, 2 = 1900-2000 meters, 3 = 1800-1900 meters.

The last secondary surface to be calculated was hydrography. A 200-meter buffer was created around the streams of the clipped hydrography shapefile in preparation to be converted into a secondary surface. The previously created stream buffer of 200-meters was then converted into a raster using the Feature to Raster tool. The hydrography raster was then reclassified with the Reclassify (Spatial Analyst) tool in order to give the no data value (leftover area with no hydrography data) a new value that would be meaningful for the study . Those values are as follows: 3 = streams, 1 = no data.

Finally, all four of these secondary surfaces were ran through the Weighted Overlay tool in order to create a basic predictive model for potential site locations. In the weighted overlay tool dialog box, the weight of each variable was set to be calculated at: Aspect (40%), Elevation (30%), Hydrography (20%), and Slope (10%) . 

With all known sites accounted for, non-sites were needed in order to conduct the additional predictive model. Non-site data was generated using the Create Random Points tool. A sum of 150 points, with a spacing of 30 meters between all points, was created to compare the sites against. The random points were then merged with known points with the Merge tool and labeled OLS points .

Finally, the Ordinary Least Squares (OLS) (Spatial Statistics) tool was conducted on the OLS points . In the tool dialog box, the Unique ID field was set to the previously calculated ID field. The dependent variables were set to the secondary surfaces slope, elevation, aspect, and hydrography. 

Lastly, the Hotspot Analysis (Getis-OrdGi) (Spatial Statistics) tool was ran to determine if there were any cold (under-predicted) or hot spots (over-predicted) in the data . 

Biscayne Shipwrecks - Analyze Week

   In this week's lab, analysis was conducted on the benthic and bathymetric data from last week. Buffers and clipping was used to determine what type of benthic features could be found within 300 meters of the Heritage Trail shipwreck sites. The benthic and bathymetric data was further analyzed through reclassification in order to determine areas that could be potentially dangerous for ships to travel through. Finally, this reclassified data was ran through a weighted overlay analysis, combining both inputs together to create an output that displays potential areas for shipwreck locations.

300 meter buffer displaying benthic features around each site

Reclassified data

Weighted Overlay Model output

Modeling Biscayne Shipwrecks - Prepare Week

   This week, we gathered data on shipwrecks in the Biscayne National Park. This data will be used to generate a weighted overlay model. Data gathered includes a historical nautical chart from 1892, a current ENC, and bathymetric data for the Biscayne Bay. The historical chart was downloaded from the NOAA's Historical Map and Chart Collection and georeferenced to the location. The bathymetric data was downloaded from the NOAA's National Geophysical Data Center and symbolized to reflect depth in meters (shallow in red, deep in blue).

Scythian Landscapes - Analyze Week

In this lab, Elevation, Slope, and Aspect are taken into consideration and reclassified in ArcMap to aid in the interpretation and analysis of the Tuekta Mounds study area. They were each simplified for further analysis by condensing their data into smaller groupings, as reflected in their legends. Contour lines in meters of elevation were made for the Tuekta area as well. Also, a shapefile was made including point locations of up to 50 mound sites in the georeferenced Tuekta Mounds image.

Scythian Landscapes - Data Prepare Week

This week, we explored USGS Earth Explorer for ASTER DEM data for our target study area. Because imagery in this region is difficult to come by, we also georeferenced an image of a known Scythian burial mound site in Tuekta to our DEM dataset. We further narrowed our study area by clipping the mosaic raster to a smaller polygon boundary. The goal of the study is to determine if the Scythians had a placement pattern of the mounds themselves.

Predictive Modeling

In this lab, predictive modeling was used to determine potential site locations in the Tangle Lakes Archaeological District in Alaska. Predictive modeling is a useful tool for estimating the amount of time and money needed to devote to any given area, as well as the amount of field survey effort. It is used to suggest broad trends in settlement patterns and resource utilization. However, it should be used as a supplementary tool for guiding and informing field surveys. It can not justify the development, avoidance, or destruction of an area without conducting field survey beforehand. Being only one tool, it generates only one possible interpretation and should not be taken as a definitive explanation.

The weighted overlay model I created, though a simplistic version, is significant in that it provides a model for determining archaeological site locations in a site dense area. The Tangle Lakes Archaeological District contains the densest known concentration of archaeological sites in the American subarctic, with over 600 sites identified. A model like this could help identify potential site locations as well as help park officials designate visitor friendly trails to prevent disruption and negative impact of sites. My model is seen below:

SWIR Supervised Classification - Angkor Wat

For my training sample of Angkor Wat I used a SWIR (short-wave infrared) composite image. From the USGS Earth Explorer search image options, I choose the Landsat 7 image taken on January 10, 2002. I narrowed down this image to be the clearest of cloud cover. I added nine additional classification features in addition to the temple feature. These are Water (grouped all large bodies of water into this), Urban, Dense Forest, Cloud, Cloud Shadow, Crop (which was hard to separate from ground, urban, and rivers), River, Grassland, and Forest. I created and merged additional samples for Crop, River, and Urban, as these samples created the most issues while classifying. The training sample I took of the temple resulted in several large clusters of potential archaeological sites.

Identifying Maya Pyramids: Data Analysis

The main focus of this week's lab was to conduct an Interactive Supervised Classification of a training sample created based on features in the Composite Band 4, 5, 1 image. First, a NDVI image was created in order to determine if it was suitable for analysis and detection of Maya pyramids. NDVI measures biomass, giving us a view of plant growth and stress. As the pyramid had not been discovered at this point in the image, or cleared of vegetation, it is not suitable to use for the detection of this pyramid. The 4, 5, 1 image predominately shows vegetation as reds and bare ground as greens. Band 5 is included to aid in the detection of plant growth or stress. This image was used for the training sample for the Interactive Supervised Classification. In order to get the classification image, features are selected on the 4, 5, 1 image to be represented as colors on the classification image. This way, if a pyramid feature is represented as red, other features identified in the process as pyramids show up as red. This helps predict future archaeological sites.

Below is my final map output, with additional information included:

Identifying Mayan Pyramids: Data Preparation

In this week's lab, our main focus was classifying and applying raster imagery to a real world scenario. We will analyze this imagery to identify Mayan pyramids in Mexico.

Below is my map output from this week's lab assignment.

The Landsat raster data was downloaded from USGS Earth Explorer and covers a portion of the Rio Azul National Park in Mexico. The downloaded data came with 9 different (raster) bands, for this lab we worked with Bands 1, 2, 3, 4, and 8.

Band 1 = Blue, Band 2 = Green, Band 3 = Red, Band 4 = Infrared, Band 8 = Panchromatic

Band 8 is panchromatic and has the highest resolution (at 15 m). It appears in black, white, and grey tones. Natural Color is the combination of the Bands 1, 2, and 3. They were combined using the Composite Bands tool in the Image Analysis window of ArcMap. The result is a natural looking color image. False Color is the combination of the Bands 2, 3, and 4, also using the Composite Bands tool. In false color, vegetation appears red because it readily reflects infrared energy.