Predictive Modeling for Site Detection in Central New Mexico using Remotely Sensed Data on Phenology

Author(s): Scott Kirk; Amy Thompson; Christopher Lippitt

Year: 2016


The potential for remotely sensed metrics of phenology and a Multi-Layer Perceptron (MLP) neural network to accurately model potential archaeological sites in central New Mexico is high. Focusing on two different environments, the Galisteo Basin and the Sandia-Manzano Mountain range, this study attempts to distinguish between archaeological sites and their surroundings based on differential growth in vegetation. Using multi-spectral satellite data, a time series of Normalized Difference Vegetation Indices (NDVI) were created to characterize patterns of vegetation growth in the study areas. Through the use of a neural network, these patterns were analyzed to distinguish archaeological sites from their surroundings. By training the network using a series of known archaeological sites, the results of an output activation layer indicate possible locations of previously unknown sites according to similarities in phenology. Treated as a site suitability model, the output activation layer can then be validated using a Receiver Operating Characteristic (ROC) under the curve. The results of this analysis show promise for detecting new archaeological sites in large, open areas such as the Galisteo Basin. However, this technique needs refining in more heterogeneous environments such as the Sandia-Manzano Mountains.

Cite this Record

Predictive Modeling for Site Detection in Central New Mexico using Remotely Sensed Data on Phenology. Scott Kirk, Amy Thompson, Christopher Lippitt. Presented at The 81st Annual Meeting of the Society for American Archaeology, Orlando, Florida. 2016 ( tDAR id: 404824)

This Resource is Part of the Following Collections

Spatial Coverage

min long: -115.532; min lat: 30.676 ; max long: -102.349; max lat: 42.033 ;