Regional-Scale Archaeological Remote Sensing in the Age of Big Data

Abstract

With the ever expanding quantity of high-resolution aerial and satellite imagery available to archaeologists, numerous researchers have sought to address this "big data" challenge by developing automated methods to aid in the discovery and mapping of archaeological sites and features. This paper reviews several notable efforts to create automated discovery tools, including both spectral and object-based approaches, and highlights the difficulties these projects have encountered. Arguing instead for the critically important role of a human analyst in archaeological discovery, I illustrate interim results of an ongoing project that utilizes CORONA satellite imagery to document previously unknown sites in a 300,000 km 2 study area in the northern Fertile Crescent. The project is based on what I term "brute force" methods, relying on systematic exploration of imagery by trained analysts, and has now successfully created a database of more than 14,000 sites, some 10,000 of which are previously undocumented. Results of the project highlight the need for human intervention to make any archaeological discovery meaningful, suggesting that imagery analysis, like any act of archaeological investigation, requires an engaged, thoughtful and creative scholar.Desde la última década, la cantidad de imágenes aéreas y satelitales de alta resolución disponible a los arqueólogos ha crecido exponencialmente, y estos recursos ofrecen posibilidades enormes para el descubrimiento de elementos y sitios arqueológicos. La gran cantidad de datos aéreos y de satélite ya disponible a los arqueólogos puede ser abrumador, y esto ha causado que unos de nosotros busquemos herramientas automatizadas para poder manejar nuestra propia versión de "datos grandes." Yo argumento que el análisis de imágenes aéreas y satelitales para encontrar evidencia de actividades culturales pasadas es tanto un arte hábil como ciencia. Es un proceso que requiere un arqueólogo empeñado, con un entendimiento de la historia del asentamiento local y las practicas de uso de terreno locales, y que pueda explorar imágenes creativamente para encontrar e interpretar elementos de posible importancia. Este es un trabajo que no puede ser automatizado, ni debería de ser, como sería la construcción de robots autónomos de excavación que hicieran nuestras propias excavaciones para nosotros. Ilustro este punto con un estudio de caso, utilizando imágenes satelitales CORONA en un esfuerzo para documentar sitios previamente no conocidos en un área de estudio de 300,000 km cuadrados en el norte de la Creciente Fértil. La base de datos resultante contiene 14,000+ sitios.

From Above and on the Ground: Geospatial Methods for Recording Endangered Archaeology in the Middle East and North Africa

“…Remote sensing is not always the most appropriate method for identifying archaeological sites. Even the application of labour intensive remote-sensing visual analysis or 'brute force' methods as Casana [17] (p. 231) has termed them is in some regions simply not the best option for large-scale site detection and monitoring. For example, trials carried out by the EAMENA team using their methodology in Kuwait revealed that large numbers of 'known' archaeological sites could not be identified via imagery, even when given precise locations for these sites.…”

“…For Middle Eastern landscapes in particular, the declassification in the 1990s of Cold War satellite photography collected in the 1960s-1970s revolutionised this sub-field. This facilitates the mapping of features, especially as many sites have been damaged or destroyed during phases of agricultural and urban expansion in the last 40 years [15][16][17][18][19]. In North Africa, projects focusing on Libya initially made use of the Landsat sensors which have been collecting data since the 1970s (e.g., the Libyan Valleys Project, [20]).…”

“…The rapidity of recording that machine learning allows for is also significant, and given the accelerating loss of archaeology that EAMENA is identifying, is a factor to be taken into consideration. There are, however, also some good reasons for skepticism [17].…”

“…We also make use of declassified satellite imagery (KH7, KH4B) collected for surveillance purposes during the Cold War, which is a useful resource for mapping archaeological features in the condition they were in during the 1960s-1970s [40][41][42][43]. With accessibility and open access again in mind, the online Corona Atlas and Referencing System is an important resource [17]. Additional images can also be purchased at a relatively low cost from United States Geological Survey (USGS).…”

“…As Casana [17] (pp. 226, 228, 230-231) highlights, large training samples, weeks of training and, where possible, a first-hand understanding of local settlement histories, archaeologies and environments are all important tools that an image interpreter will be required to use within their work.…”