Specialized Session 8: Graspa Session – Environmental Issues and Sustainability

Metabarcoding has revolutionized species occurrence studies by enabling rapid, efficient identification. However, its unique data structure poses novel challenges. This study investigates the presence-absence of macroinvertebrate haplotypes across river monitoring sites in north-west Italy, focusing on the effect of altitude. We employ a latent Gaussian model with hierarchical random effects to account for station-level and species-specific variability. To improve variance estimation, we use the Hierarchical Decomposition (HD) prior framework, offering a more intuitive alternative to traditional priors. Our approach enhances the flexibility and transparency of prior choices in species distribution models, making Bayesian inference more accessible for ecological applications.

In Italy, the ecological status of water bodies is assessed by means of LIMeco index obtained by summarizing four key variables: ammoniacal nitrogen, nitrate nitrogen, total phosphorus, and dissolved oxygen in percentage of saturation. This study analyzes the impact of different chemicals on each LIMeco component, focusing on the river network of Piedmont. To capture the influence of these covariates on a target variable while accounting for the complex spatial heterogeneity of river systems, we introduce a spatially varying coefficient linear regression model on the network that leverages on a nonparametric estimation method. The results of a case study focused on the relationship between nitrogen nitrate and total pesticide in the water show the effectiveness of our proposal.

Landslides are a major geological hazard, causing widespread disruptions and posing increasing risks. Understanding the relationship between terrain conditions and landslide occurrences is crucial for identifying the factors that trigger instability and for developing effective monitoring strategies. In this study, we examine landslides that occurred in Italy and recorded in publicly available sources. To estimate spatial landslide susceptibility, we employ a zero-inflated Poisson (ZIP) model, which addresses the challenge of excess zero-event observations in the data. This probabilistic approach not only enhances susceptibility estimation but also helps pinpoint areas where landslide events may be underreported.

This study focuses on the application of Fixed Rank Kriging (FRK) to model daily nitrogen dioxide (NO2) levels across Italy over a 10-year period. We employed low-rank approximation techniques to efficiently handle the large-scale dataset. Additionally, the “sliding window approach” is investigated to further reduce computational costs. The study incorporates data fusion by considering two different sources of NO2: point data on an irregular grid, i.e., ground-based monitoring stations, and areal data on regular lattice, i.e., Chemical Transport Models (CTMs). Several external factors, including meteorological conditions, land use, and emissions, are utilized to better understand and explain the variations in NO2 concentrations. Finally, the FRK model addresses the challenges of modelling a non-stationary, spatially correlated process, which is essential for accurately capturing NO2 dynamics across the entire country.