• Home
• Research
  • Archaeology/Historic Preservation
  • Digital Library and Data Archives
  • Geospatial Modeling and Visualization (GMV)
  • Environmental Studies
  • Geoinformatics (GIS/RS)
  • Geophysics
  • Spatial / Temporal Social Analysis
  • Urban Infrastructure Projects
  • Visualization/Animation
  • Publications
• Outreach
  • Communities
  • Government
  • International
  • Private Sector
  • University
• Education
  • Digital Institute for Archaeology
  • Field Programs
  • University
  • Continuing and Professional
  • K-12
  • Class Support and Information
  • How do I...
• About CAST
  • Contact
  • CAST 20 Years
  • Our Mission
  • Facilities
  • Computing & Research Equipment
  • Center of Excellence Sponsors
  • Collaborators
  • Staff Web Pages
  • Publications
  • CAST in the News
  • Our Community
  • Directions to CAST

• A - Z
• Support

Center for Advanced Spatial Technologies

University of Arkansas

Multi-Sensor Data Fusion for Historic Resource Identification

Home » Research » Geophysics » Multi-Sensor Data Fusion

New Approaches to the Use and Integration of Multi-Sensor Remote Sensing for Historic   Resource
Identification and Evaluation (SERDP SI-1263)

Kenneth Kvamme, Eileen Ernenwein, Michael Hargrave, Thomas Sever, Deborah Harmon, Fred Limp

The research focus of this project, funded by the Defense Department's Strategic Environmental Research and  Development Program (SERDP), was development of powerful new analytical approaches that demonstrate the effectiveness of non-invasive archaeological methods and the deployment of tools that offer an opportunity to recover a great deal of information about site content while reducing costs associated with traditional archaeological survey and excavation. Exploration and assessment of the benefits of combining a large suite of ground, aerial and space-based sensor data for the detection of subsurface archaeological features is central to this research which was completed in 2006.

Executive Summary

The Center for Advanced Spatial Technologies and the Department of Anthropology, University of Arkansas, ERDC CERL and NASA Marshall Center have conducted research in a project titled New Approaches to the Use and Integration of Multi-Sensor Remote Sensing for Historic Resource Identification and Evaluation (SERDP CS-1263). The focus of this research is the identification of specific combinations of remote sensors and data integration methods for the detection, identification, and interpretation of cultural resources in various environments and archaeological circumstances. Four sites were selected for inclusion in this project: Army City (Fort Riley, Kansas), Pueblo Escondido (Fort Bliss, New Mexico), Kasita Town (Fort Benning, Georgia), and Silver Bluff (located near the DOE Savannah River facility, Aiken County, South Carolina). An extensive suite of ground-based geophysical, aerial, and satellite technologies were employed for this task. The methods investigated include magnetometry, magnetic susceptibility, electrical resistivity, electromagnetic conductivity, ground penetrating radar, aerial thermal infrared and high resolution multispectral satellite imagery.

The goals of this research include:

• Application of a suite of established and newly fielded high-resolution space and aerial instrumentation to the detection of cultural resources.
• Examination of a suite of existing and newly developed ground based instruments applied to the detection and evaluation of subsurface deposits.
• Application of existing and the development of new computational approaches for the integration of an extensive suite of satellite, aerial, and ground-based remote sensing data sets.
• Assessment of the results of these approaches through an archaeological field validation program in a range of environments and archaeological site types.
• Evaluation of the effectiveness of various sensor combinations and fusion methods against site conditions, and development of guidance documents for their use in various settings.

 PROBLEMS ADDRESSED

Some of the nation's most significant historic and prehistoric cultural resources are contained within the 25 million acres of public land administered by the Department of Defense. Protecting these heritage resources is a fundamental part of the Department's primary mission. Heritage management issues central to that mission have focused on the economics of identifying and maintaining historic facilities, the impact of archaeological sites on construction and training programs, and the disposition and curation of artifacts. Cultural resources, defined in DoD Instruction 4715.3, include buildings, structures, sites, districts, and objects eligible for, or included in, the National Register of Historic Places regulation (36 CFR 60). Management of these resources, in compliance with existing laws and regulations such as Native American Graves Protection and Repatriation Act  (25 USC §3001), the Archaeological Resources Protection Act (16 USC §470 aa-ll), American Indian Religious Freedom Act (42 USC §1996), and the standards in the Curation of Federally-Owned and Administered Archaeological Collections (36 CFR 79), necessitates the development of innovative and cost-effective methods for archaeological site identification, evaluation, and protection.

The Strategic Environmental Research Development Program (SERDP) Statement of Need entitled "Cultural Resources Management Detection and Evaluation Technologies" (CSSON-02-01) requested research into a range of advanced archaeological methods. Specifically, it called for methods (1) that "effectively detect, locate and identify historic and pre-historic archeological resources on military and DoE lands and ranges," (2) that produce "improved models for predicting the location of resources," and (3) that identify "improved technologies for detecting surface and/or subsurface resources."  Non-invasive processes and procedures were "strongly encouraged in order to reduce the possible disturbance of human remains and associated artifacts."  Finally, validation of findings through excavation was deemed "necessary to demonstrate the feasibility of the proposed technologies and associated procedures."

This Statement was timely, because mainstream archaeological methods for the identification and evaluation of most historic resources remain little changed from those employed in the early twentieth century. Surface survey and excavation, the traditional field methods for discovery of artifacts, architectural elements, and other archaeological features, continue to predominate in spite of the fact that these techniques are extremely time consuming, expensive, and unreliable. Their shortcomings manifest themselves in many ways. Frequently there is no discernable surface evidence of buried archaeological features, making surveys ineffective. Excavation of small shovel test-pits can sometimes locate archaeological sites, but this method entails substantial additional survey costs. In National Register of Historic Places eligibility assessments, ever-increasing labor costs restrict the number of units that can be excavated, resulting in failures to locate features of significance. More importantly, traditional invasive methods regularly lead to the damage or destruction of the very resources they were designed to investigate and generate costs associated with collection curation. Aware of these limitations, archaeologists have adopted a conservative, preservationist approach, with the result that the Department of Defense protects many sites of marginal scientific importance. It is apparent that the SERDP SON implicitly recommends archaeological remote sensing as a primary vehicle for meeting current needs.

SCIENTIFIC QUESTIONS

A principal goal of this SERDP research project is a determination of remote sensing methods and techniques that work well individually, and that complement each other collectively when integrated, for the identification and evaluation of buried archaeological remains. Numerous methods for data integration were explored. With the exception of certain multi-band visualization techniques and overlays of vectors representing interpreted anomalies, most of the methods investigated have not been applied previously in archaeology. Several advanced computer graphic methods were explored, discrete methods that range from Boolean overlays to sums of categorized portrayals of sensor outputs and cluster analyses, continuous methods that include sums, products, ratios, principal components, regressions, and probability surfaces, intelligent knowledge-based systems like C5.0 and Cubist, and Expert Systems approaches. This research demonstrates that certain integrating methods yield more information about the subsurface than others, but what may be realized in each approach may depend on overarching purpose. Some fusion techniques yield visually pleasing results that appear to well-combine available information, for example, while others may seem less revealing but offer greater interpretive or predictive potential. In this process, the nature of similarities, differences, redundancies, and performance characteristics of results were examined. An important aspect of this research is an assessment of the added value of the fused product compared to traditional, individual-sensor based analysis.

RESULTS AND FUTURE APPLICATIONS                                       

These are some initial views of fusion of Army City data using relatively simple methods applied to the unprocessed data

The process of archaeological remote sensing as carried out in this project is a multi-step undertaking. The first stage, designed to meet the primary data integration goal, includes remote sensing data      collection, processing to clarify anomalies in individual sensors established within GIS databases, data fusion to integrate information from all sensors, definition of potentially "significant" cultural anomalies,   and classification of the anomalies into likely types of cultural features. This is normally the final product of most remote sensing projects, and the point at which archaeological fieldwork takes over. This project endeavored to go several steps farther with a second principal goal.

The second goal of this research project was very different and designed to meet an important criterion in the SERDP Statement of Need. Specifically, "ground-truth" testing was called for to demonstrate the feasibility of the data integration technologies and associated procedures. Three additional project tasks were therefore designed. They include development of a sampling design that allowed archaeological excavation of representative anomalies of each defined type to provide validation of remote sensing predictions about the subsurface. This validation phase often turned into a learning process, however, because the soils, geology and the archaeology in each site are unique, idiosyncratic, and confound predictability. In other words, remote sensing predictions cannot be perfect and a look into the ground through excavation offers additional insights that allow modifications to original predictions. Consequently, a final stage was defined that includes modification of original remote sensing predictions, based on excavation findings. Numerous scholarly presentations and publications were produced during the project. Preparation of a final report was, of course, the ultimate task.   These operations were undertaken at each of four prehistoric and historic archaeological sites distributed across time and space in a wide diversity of environmental settings from South Carolina to New Mexico. This provided a variety of contexts in which to assess the value of the methods investigated at very different archaeological sites with very different remains in very different environments. In so doing a better understanding of which methods consistently worked and offered useful results could be achieved, but this knowledge was also augmented by the considerable experience of project team members.

Integrating multiple geophysical data sets offers large potential for improved understanding of the subsurface. A single survey, for example, might reveal only part of a buried building. Integrated information from several surveys may illustrate the entire structure as well as interior components. Moreover, integrated data may simultaneously show relationships between conductive, resistant, magnetic, thermal, and metallic anomalies, potentially improving knowledge of features within a site, inter-sensor relationships, enhancing overall interpretations.

Graphical solutions for data integration are easy to implement and effectively combine information from disparate sources into interpretable displays. They allow complex visualizations of the subsurface, but their weakness rests in relatively low dimensionality-only 2-3 data sources may effectively be represented. Moreover, these methods are purely descriptive, yielding only images, not new data that may subsequently be analyzed. Discrete integrating methods, on the other hand, allow application of readily available Boolean operations to any number of geophysical data sets. A shortcoming is that the binary maps upon which these methods are based rely on arbitrary thresholds to define significant anomalies, while more subtle ones must be ignored. Continuous data integrations can yield insights beyond the capabilities of other methods. Robust and subtle anomalies may be simultaneously expressed, producing composite imagery with high information content. Interpretive data are also generated in the form of principal component scores, factor loadings, or regression weights that add to understanding of interrelationships and underlying dimensionality. Supervised and unsupervised classification methods are noteworthy because they introduce a predictive aspect to the integrating process. Patterns in these data fusions may point to anomalous conditions much less visible in any single data set that might otherwise be overlooked. They therefore offer a possible means to augment prospecting capabilities. Although the approaches to geophysical data integration examined here span a wide range of commonly available techniques, they are by no means exhaustive. A host of other supervised and unsupervised classification algorithms exists, as well as new context-based image segmentation, and intelligent knowledge-based methods.

If the foregoing results can be generalized, it is that robust anomalies exist in the data and tend to dominate any form of fusion, regardless of the method employed. The consequence is amazingly parallel results between widely different forms of integration. Consequently, they really should be considered as offering new information about subsurface variation.

The determination of which integrating methods are best may depend on purpose. Some yield visually pleasing results that appear to well-combine available information while others may seem less revealing but may offer interpretive or predictive potential. If a goal is to define discrete classes of anomalies that may be subsequently interpreted through comparison with primary data then categorical methods may be best. If a goal is merely a continuous-tone image that represents most of what is known about the subsurface then a composite color graphic or mathematical-statistical integration may be most suitable. Of course, continuous methods yield quantitative data that may subsequently be analyzed, plus regression weights, PCA scores, or factor loadings that give additional insights beyond graphical representations, important for improved understanding of the subsurface and its interaction with geophysical methods. In practice, a variety of different integrating methods may work best in practice, because each variation may give new insights about a different aspect of the subsurface.

The results of the integrated data sets clearly illustrate the very substantive subsurface site characteristics that are discoverable from the integrative methods used. Based on these results a dramatically clearer picture of the subsurface is realized, compared to traditional site evaluative methods. By more clearly imaging the totality of information about the subsurface from all sources, a better understanding of site content, structure, and organization may also be achieved. The amount of information provided from these methods dramatically improves the ability to assess the site properties consistent with eligibility evaluations.  The extensive amount of information yielded by the approach also will serve as important guidance should site mitigation be needed. Compared to the typical site evaluation results these methods provide orders of magnitude more information on the nature of the internal sites structures and its characteristics. This is particularly evident at sites such as Army City and Pueblo Escondido but to a lesser degree at all the sites.

It should be noted that while these methods are very effective in the horizontal delineation of site characteristics, such as the mapping of a house foundation, they are somewhat more limited in their capabilities in delineation of the site's vertical characteristics. However, as this project demonstrates, GPR does give good depth estimates and potentially allow 3D modeling and portrayal. Multiple depth slices showing structural changes with depth at Escondido and the other report examples of 3D models incorporating the vertical dimension illustrate these potentials.. 

The methods developed in this investigation will increasingly serve as critical steps in the evaluation of archeological properties as required by the National Historic Preservation Act. Use of these methods can increase the effectiveness and (often reduce the cost) of the evaluation efforts. Since excavation of entire sites or settlements, or even large areas of them, is impossible owing to funding limitations and ethical concerns, it may be only through integrated remote sensing that real understandings of the content, structure, and extent of archaeological sites may be achieved. It is anticipated that the methods pioneered here provide an important step in the direction of that goal.

The project final report is available here.