Page 684 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

(continued)

Item Details
The use case was implemented as a pilot deployment at the Saruq Al-Hadid site in
Dubai, UAE. This site served as the testbed for evaluating the efficiency and accu-
racy of the integrated remote sensing, geospatial analysis, and machine learning
framework. Field validation and comparisons with historical excavation data were
used to assess the performance of the approach, demonstrating its potential to
guide targeted archaeological investigations in challenging arid environments.
These results validate the integrated AI–remote sensing framework in a real-world
Testbeds or desert context. In addition, supporting geophysical surveys—including Ground
Pilot Deploy- Penetrating Radar (GPR) and magnetic survey—conducted in parallel for validation
ments revealed further insights into the site’s subsurface anomalies. Key findings from
the AI–remote sensing and machine learning approach include the successful
identification of known excavation areas and the detection of new locations with
archaeological potential. Meanwhile, geophysical surveys at SAR.53 identified
a total of 462 anomalies, including five major subsurface structures (A–E), while
SAR.7 yielded 2,094 anomalies distributed across six sub-zones with archaeo-
logical potential (A–F).
Links: [1]

2 Use Case Description

2�1 Description

Archaeological research in arid regions such as the Saruq Al-Hadid site in Dubai, UAE, is
challenged by extensive sand cover, shifting dunes, and minimal vegetation, which limit the
effectiveness of traditional field surveys in detecting buried features. This use case addresses
these limitations through an innovative, integrated approach that combines high-resolution
remote sensing technologies—specifically multispectral imagery and SAR—with advanced
image processing, machine learning, and geospatial analysis. The objective is to automate the
detection and prediction of archaeological sites, reducing reliance on invasive, labor-intensive
surveys while guiding excavations more efficiently and accurately. The technical approach
includes PCA, spectral band transformations, and semantic segmentation using convolutional
neural networks (U-Net), alongside clustering techniques such as K-means++ and Kernel PCA
with Gaussian RBF kernels. These tools extract meaningful patterns from complex datasets.
Geospatial tools, including digital elevation models and slope maps, help contextualize findings
within the broader landscape. The framework was further validated through comparisons with
historical excavation data and supported by geophysical surveys—namely GPR and magnetic
survey—offering deeper insights into subsurface features. The solution has demonstrated high
detection accuracy, successfully identified known excavation areas and predicted new zones
with archaeological potential. It enhances decision-making and resource allocation by helping
archaeologists focus efforts on the most promising areas. The methodology is also scalable and
adaptable for use in other arid or environmentally sensitive regions. Compared to traditional
survey methods, it offers a major advancement by enabling automated, non-invasive, and data-
driven archaeological investigations at both surface and subsurface levels.

The outputs of this approach contributed to the creation of a site-specific geospatial knowledge
base for the Saruq Al-Hadid archaeological site. This includes digital elevation and slope maps,
hydrographic network data, classified geomorphological assemblages (Geo Assemblage I–III),
and AI-generated outputs such as unsupervised classifications, clustering results, and predicted

648