Optimal Spectral Domain Selection for Maximizing Archaeological Signatures: Italy Case Studies

Rosa Maria Cavalli,Simone Pascucci,Stefano Pignatti

doi:10.3390/s90301754

Abstract

Different landscape elements, including archaeological remains, can be automatically classified when their spectral characteristics are different, but major difficulties occur when extracting and classifying archaeological spectral features, as archaeological remains do not have unique shape or spectral characteristics. The spectral anomaly characteristics due to buried remains depend strongly on vegetation cover and/or soil types, which can make feature extraction more complicated. For crop areas, such as the test sites selected for this study, soil and moisture changes within near-surface archaeological deposits can influence surface vegetation patterns creating spectral anomalies of various kinds. In this context, this paper analyzes the usefulness of hyperspectral imagery, in the 0.4 to 12.8 μm spectral region, to identify the optimal spectral range for archaeological prospection as a function of the dominant land cover. MIVIS airborne hyperspectral imagery acquired in five different archaeological areas located in Italy has been used. Within these archaeological areas, 97 test sites with homogenous land cover and characterized by a statistically significant number of pixels related to the buried remains have been selected. The archaeological detection potential for all MIVIS bands has been assessed by applying a Separability Index on each spectral anomaly-background system of the test sites. A scatterplot analysis of the SI values vs. the dominant land cover fractional abundances, as retrieved by spectral mixture analysis, was performed to derive the optimal spectral ranges maximizing the archaeological detection. This work demonstrates that whenever we know the dominant land cover fractional abundances in archaeological sites, we can a priori select the optimal spectral range to improve the efficiency of archaeological observations performed by remote sensing data.

Highlights

As the planet’s exploding human population results in massive developments and changes to the landscape, there is a consequent need for efficient and cost-effective methods to locate, map, and acquire information from sites of our cultural heritage before they are forever lost [1]
As the spectral anomalies characteristics due to buried remains depend on vegetation cover and/or soil types, the Spectral Angle Mapper (SAM) algorithm was applied to Multispectral Infrared Visible Imaging Spectrometer (MIVIS) images to verify the land cover occurring on the anomaly-background system of the 97 known archaeological anomalies
A properly supervised classification of the five archaeological areas should imply the use of a higher thematic level, which will be site-specific and not useful for identifying the optimal spectral range for archaeological prospection at a broad-scale, such as the aim of this study

Summary

Introduction

As the planet’s exploding human population results in massive developments and changes to the landscape, there is a consequent need for efficient and cost-effective methods to locate, map, and acquire information from sites of our cultural heritage before they are forever lost [1]. The basic assumption of image-interpretation for the recognition of the buried structures is that they can alter the natural trend of the superficial soil and vegetation growth and such alterations can develop into permanent surface spectral features [1,2] These changes can mark out the pixel appearing with differences, with respect to the adjacent pixels, in color, texture, brightness or combination thereof [4]. The identification of these relevant anomalies, expected in presence of buried man-made structures, depends usually on the experience of the photointerpreter and his knowledge of the territory [5] Environmental factors such as the compaction of soil, moisture content and vegetation impact the effectiveness of the technique to detect subsurface remains [2,5]. Recent studies carried out by [9]

Methods

Results

Conclusion