1.

INTRODUCTION

Archaeological looting and trafficking, often exacerbated by conflicts, present significant challenges for many EU Member States. These challenges include protecting widely dispersed and remote heritage sites with limited resources in terms of equipment and specialized personnel, and funding. Recognized first at an international level in 1954 with The Hague Convention [1], the gravity of cultural property crimes has since been emphasized by UNESCO in 1970 [2] and 2001 [3], and the Council of Europe in 2017 [4]. Yet, these crimes persist, harming global heritage.

Satellite technology has shown great potential for analyzing archaeological looting through various academic studies worldwide [5, 6, 7]. To achieve more reliable and accurate results, semi-automated detection processes have been developed, focusing on multitemporal site monitoring, which traditionally required laborious and error-prone manual comparisons of satellite imagery. Despite these advancements, managing the increasing volume of information can be challenging for final users who may not fully leverage the benefits of the space segments’ operations.

In the framework of the ENIGMA Project [8], a remote sensing toolkit for endangered site monitoring is currently being developed using Earth Observation (EO) data, which will be able to provide early warnings to relevant authorities. It will exploit optical and radar satellite images using various processing techniques, such as interferometric SAR (InSAR), and Change Detection (CD).

The current study focuses on the description of the EO toolkit with a particular emphasis on the overall architecture and the preliminary results of the Change Detection processing optical data. The developed pipeline is presented together with the achieved results and current limitations.

2.

METHODOLOGY

Looting pits represent a typology of ground anomaly that is extremely difficult to detect using remote sensing techniques due to their size and variability in geometry and texture. In this context, the resolution of satellite images, particularly for optical satellite data, is crucial. Low-resolution data, with a magnitude of ≥ 5 meters, may not be indeed sufficient to detect illicit archaeological excavations. For this reason, the current study has utilized the Planetscope dataset [9], which offers a resolution of 3 meters per pixel (resampled) with an average temporal resolution on 1 day. The methodology employed for generating Change Detection maps is concisely outlined in Figure 1.

Figure 1.

EO toolbox architecture

The functionalities of the Earth Observation toolkit were developed using the Python programming language and leveraged the capabilities of GDAL libraries.

Calculate Normalized Difference Vegetation Index (NDVI) per single image and save it in grayscale

NDVI is used to identify looting because it might effectively highlight changes in vegetation [10]. The latter indeed can be indicative of ground disturbances such as those caused by archaeological looting. To save the NDVI index of a multiband image in grayscale, a series of steps is followed. First, per each clipped image, the NDVI is computed for each pixel in the image using the formula (NIR - Red) / (NIR + Red), which compares the near-infrared and red-band reflectance. Once the NDVI values are calculated, they are normalized to fit the grayscale color space by scaling them to a range of 0 to 255. Finally, the scaled NDVI values are converted into a grayscale image, where each pixel’s intensity accurately represents the corresponding NDVI value (Figure 2, left). This grayscale image visually illustrates the vegetation index, facilitating further analysis and interpretation.

Figure 2.

NDVI Map (left), and NDVI Multitemporal Change Detection Map (Right)

Compute PCA components

Principal Component Analysis (PCA) algorithm is a statistical procedure used in several domains and developed to transform a set of correlated variables into a new set of uncorrelated variables considering the principal directions in which the data are spread in space [11]. The process begins by stacking the image bands, combining the various spectral bands into a single multi-dimensional array. This comprehensive dataset is then standardized, normalizing the pixel values across all bands to have zero mean and unit variance. With the data prepared, PCA is performed, transforming the dataset into a set of orthogonal components, each representing a direction of maximum variance. Finally, these transformed data are extracted as PCA components, which can be saved and further analyzed to reveal underlying patterns and features that may not be immediately apparent in the original spectral bands (Figure 3, left).

Figure 3.

PCA Map (left), and PCA Multitemporal Change Detection Map (Right)

3.

CONCLUSIONS

The Earth Observation toolkit is designed to automate the identification of illegal archaeological excavations using remote sensing techniques. The workflow utilizes the Python programming language, GDAL libraries, image processing techniques, and Change Detection algorithms. These components work together to trigger an alarm system, enabling authorities to quickly intervene and mitigate the risks associated with illegal activities. The increasing revisit capabilities of various satellite providers enhance the system’s potential to improve responses to threats and protect archaeological sites. However, it is important to note that the resolution of the input data is critical for accurately identifying ground anomalies associated with illegal excavations and looting. Additionally, factors such as cloud coverage or water proximity can hinder the accurate interpretation of detected changes, potentially impacting the effectiveness of the system in real-world scenarios.