advertisement

Archaeological Predictive Model for the High Plains of Southwestern Kansas

43 %
57 %
advertisement
Information about Archaeological Predictive Model for the High Plains of Southwestern Kansas

Published on March 15, 2009

Author: jscampbell1

Source: slideshare.net

Description

Summarizes the results of my MA thesis on archaeological predictive modeling. In addition to a summary of the southwest Kansas model, this presentation also contains reference to previous work at Fort Hood, Texas and possible model enhancements.
advertisement

Archaeological Predictive Model for the High Plains of Southwestern Kansas Plains Conference 2006 Topeka, Kansas Joshua S. Campbell Department of Geography University of Kansas

Project Outline Use existing data sources to develop an empirical predictive model of open-air archaeological sites on the High Plains of southwestern Kansas The binary logistic regression model relates the presence or absence of archaeological material to geographic variables extracted from modern map data Final output is a probability surface in which each raster cell contains a probability score describing its environmental similarity to known site locations

Use existing data sources to develop an empirical predictive model of open-air archaeological sites on the High Plains of southwestern Kansas

The binary logistic regression model relates the presence or absence of archaeological material to geographic variables extracted from modern map data

Final output is a probability surface in which each raster cell contains a probability score describing its environmental similarity to known site locations

 

 

Point of Rocks Middle Spring

 

 

Archaeological Sample Selection Training Sample Site (1) = 151 sites (7,917 cells) Nonsite (0) = 12,303 random cells Testing Sample Site (1) = 75 sites (3,344 cells) Nonsite (0) = 3,142 random cells Over 90% of sites are less than 2,000 yrs (Brown, 1978)

Training Sample

Site (1) = 151 sites (7,917 cells)

Nonsite (0) = 12,303 random cells

Testing Sample

Site (1) = 75 sites (3,344 cells)

Nonsite (0) = 3,142 random cells

Over 90% of sites are less than 2,000 yrs (Brown, 1978)

Apriori Probabilities Site-present {S} event class : Pr(S) = 11,261/20,440,315 = 000550 ( 0.05% of all 30m 2 land parcels contain a site) Morton County = 0.3%   Site-absent class {S’} as: Pr(S’) = 20,429,054/20,440,315 =0.999449 ( 99.95% of all land parcels do not contain a site)

Site-present {S} event class :

Pr(S) = 11,261/20,440,315 = 000550

( 0.05% of all 30m 2 land parcels contain a site)

Morton County = 0.3%

 

Site-absent class {S’} as:

Pr(S’) = 20,429,054/20,440,315 =0.999449

( 99.95% of all land parcels do not contain a site)

Geographic Variables Slope Relief (3): 150m, 300m, and 600m radius Shelter Index Distance to Intermittent Water Distance to Perennial Water Distance to Playa Lake Landforms

Slope

Relief (3): 150m, 300m, and 600m radius

Shelter Index

Distance to Intermittent Water

Distance to Perennial Water

Distance to Playa Lake

Landforms

Landform Variable Generated by reclassifying the 229 soil series in the study area into 6 landform categories Upland Slopes Floodplain Sand Dunes Semi-Sand Playas (buffered to 90m to represent activity zone)

Generated by reclassifying the 229 soil series in the study area into 6 landform categories

Upland

Slopes

Floodplain

Sand Dunes

Semi-Sand

Playas (buffered to 90m to represent activity zone)

 

Landform – Site Distribution Sites are differentially located with respect to landform, while the Non-Site sample reflects a random sampling of the overall landscape   Entire Landscape   All Sites   Non-Sites Floodplain 6.0%   9.1%   5.9% Upland 55.8%   14.7%   56.2% Semi-Sand 13.0%   9.6%   12.5% Slopes 7.5%   39.1%   6.7% Sand 15.5%   27.0%   16.7% Playa 2.2%   0.5%   2.0%

Sites are differentially located with respect to landform, while the Non-Site sample reflects a random sampling of the overall landscape

Predictive Model The ‘Z’ equation in then defined as: Z = -2.701459 + (Dist. to Inter. * -0.000328) + (Dist. to Perr. * -0.000053) + (Floodplain * 0.005184) + (Upland * -0.969463) + (Semi-Sand * 0.509768) + (Slopes * 1.535854) + (Sand * 0.867705) + (Relief150 * -0.009843) + (Relief600 * -0.012883) + (Shelter Index * 0.001617) + (Slope * -0.025225) This equation was entered into the GIS and then further modified by the probability equation, Probability (event) = 1 / (1 + e -Z )

The ‘Z’ equation in then defined as:

Z = -2.701459 + (Dist. to Inter. * -0.000328) + (Dist. to Perr. * -0.000053) + (Floodplain * 0.005184) + (Upland * -0.969463) + (Semi-Sand * 0.509768) + (Slopes * 1.535854) + (Sand * 0.867705) + (Relief150 * -0.009843) + (Relief600 * -0.012883) + (Shelter Index * 0.001617) + (Slope * -0.025225)

This equation was entered into the GIS and then further modified by the probability equation, Probability (event) = 1 / (1 + e -Z )

 

 

 

Model Evaluation Determine the number of sample locations accurately predicted in the Site-Present and Site-Absent event classes Testing uses an set of sites and non-sites withheld from model development A ‘cut-point’ is established at the value in which 85% of sites are correctly classified

Determine the number of sample locations accurately predicted in the Site-Present and Site-Absent event classes

Testing uses an set of sites and non-sites withheld from model development

A ‘cut-point’ is established at the value in which 85% of sites are correctly classified

 

 

 

Cut-Point 85% of the Site-Present testing sample were accurately predicted at the probability value 0.11 At the 0.11 cut-point, 41% of the landscape is classified as in the Site-Present class Optimally this value would be 33% or less but the current results are inline with other published results

85% of the Site-Present testing sample were accurately predicted at the probability value 0.11

At the 0.11 cut-point, 41% of the landscape is classified as in the Site-Present class

Optimally this value would be 33% or less but the current results are inline with other published results

 

Conclusions Probability of finding a site in a cell classified as Site-Present is 2.15x more likely than random chance alone Probability of finding a site in the Site-Absent class is .25x as likely as random chance Model performs equally well in Morton County, even with the higher base-rate probability Considering there are over 20 million cells in the study area, this represents a significant increase over random

Probability of finding a site in a cell classified as Site-Present is 2.15x more likely than random chance alone

Probability of finding a site in the Site-Absent class is .25x as likely as random chance

Model performs equally well in Morton County, even with the higher base-rate probability

Considering there are over 20 million cells in the study area, this represents a significant increase over random

Model Enhancements Qualitative / ‘non-universe’ data integration Model results could be negatively impacted by including a dataset in which the complete universe is not known, or that has a very low occurrence Springs, quarries, trails, rockshelters, rock art, etc. A synthetic approach is proposed that integrates a statistical model as a ‘baseline’ with the addition of specific features

Qualitative / ‘non-universe’ data integration

Model results could be negatively impacted by including a dataset in which the complete universe is not known, or that has a very low occurrence

Springs, quarries, trails, rockshelters, rock art, etc.

A synthetic approach is proposed that integrates a statistical model as a ‘baseline’ with the addition of specific features

 

Model Enhancements Landscape change / Subsurface modeling Predict the age and cultural significance of a subsurface location (voxel) Accurate paleo-landscape reconstructions (Fort Riley – CHILD model) Multi-temporal archaeological models based on the paleo-landscape reconstructions A synthetic approach combining a geoarchaeological and statistical surface model has been developed for Fort Hood, Texas

Landscape change / Subsurface modeling

Predict the age and cultural significance of a subsurface location (voxel)

Accurate paleo-landscape reconstructions (Fort Riley – CHILD model)

Multi-temporal archaeological models based on the paleo-landscape reconstructions

A synthetic approach combining a geoarchaeological and statistical surface model has been developed for Fort Hood, Texas

Figure 8

 

 

Local Collectors (What about the living rooms?) Significant collections are in the hands of an aging group of collectors Archaeological resources are out there, it is up to the research community and the State to put together an integrated approach to address the issue It is hoped this model will help in that process

Significant collections are in the hands of an aging group of collectors

Archaeological resources are out there, it is up to the research community and the State to put together an integrated approach to address the issue

It is hoped this model will help in that process

Joshua S. Campbell Email: jsc1@ku.edu Department of Geography University of Kansas

Add a comment

Related pages

ARCHAEOLOGICAL PREDICTIVE MODEL OF SOUTHWESTERN KANSAS BY ...

ARCHAEOLOGICAL PREDICTIVE MODEL OF ... for the High Plains region of southwestern Kansas. ... archaeological predictive model for the High Plains ...
Read more

Archaeological Predictive Model for the High Plains of ...

1. Archaeological Predictive Model for the High Plains of Southwestern Kansas Plains Conference 2006Topeka, Kansas Joshua S. Campbell Department of ...
Read more

Curriculum Vitae Joshua Stewart Campbell - KARS Home

Curriculum Vitae Joshua Stewart Campbell ... Predictive Model of Southwestern Kansas ... Archaeological Predictive Model for the High ...
Read more

Mn/Model Final Report Chapter 3: Minnesota's Environment ...

The Mn/Model archaeological ... 1992 A Predictive Site Location Model on the High Plains: ... The Mn/Model process and the predictive ...
Read more

Joshua Campbell, PhD | LinkedIn

... entitled "Archaeological Predictive Model of Southwestern Kansas" ... Archaeological Predictive Model of ... on the western High Plains. ...
Read more

Topographic Prominence in an Archaeolgical Landscape

Using ARC/GRID to Calculate Topographic Prominence in an Archaeological ... in Southwestern Kansas. Plains ... Model on the High Plains: ...
Read more

Mn/Model Final Report Chapter 5: The Archaeological Database

The foundation of a predictive model of archaeological site ... in those areas of high site potential ... Mn/Model archaeological ...
Read more

Team Members | Metcalf Archaeology

Find out more about the team here. ... development of a predictive model of Knife River flint distribution ... western High Plains, and southwestern ...
Read more