Codes for: Assessing Anthropogenic Ground Deformation on Italy's North Adriatic Coasts Using Deep Learning

Anthropogenic ground deformation poses significant environmental risks in vulnerable coastal regions, where it combines with natural subsidence, sea-level rise and extreme weather events. This study focuses on Italy's North Adriatic coast, an area affected by hydrocarbon extraction and gas storage. We develop an integrated framework using GNSS data and deep learning (DL) to isolate human-induced deformation. Our approach constructs a comprehensive synthetic training dataset by simulating time series for 45 GNSS stations above active reservoirs and interpolating these into spatially continuous maps, thereby capturing seasonal operational cycles such as summer gas injection and winter extraction. We handle missing data using weighted principal component analysis and apply Kriging interpolation to transform time series into spatially continuous maps. These maps serve as input to two DL architectures (CNN-Unet and CNN-AE), which successfully extract anthropogenic deformation features and distinguish human-induced patterns from natural variability. Synthetic validation demonstrates that CNN-Unet achieves near-perfect structural fidelity (SSIM > 0.9) and low reconstruction errors (RMSE < 0.5 mm). Applied to 13 years of GNSS data (2010–2022), the CNN-Unet framework identifies cumulative subsidence reaching 56 mm near production zones and horizontal displacement up to 6 mm in storage areas, with temporal correlations confirming the model's ability to capture seasonal deformation cycles linked to gas operations. We contextualise these patterns within a multi-hazard assessment, demonstrating how subsidence acts synergistically with relative sea-level rise to increase flood vulnerability. This approach advances automated identification of human-induced ground motion and provides quantitative data for mitigating climate-exacerbated risks in resource-dependent coastal regions.