<p><strong>Abstract.</strong> In order to mitigate environmental risk in Thailand it is essential to understand where and when specific geographic areas will be exposed to individual and multiple natural hazards. However, existing national scale approaches to natural hazard risk assessment are poorly adapted to deal with multiple hazards where significant uncertainties are associated with input variables and prior knowledge of the spatiotemporal nature of hazards is limited. To overcome these limitations, a geospatial approach has been developed that integrates machine learning within a GIS environment. Four hazards were investigated by Naïve Bayes while multiple hazards and their causalities were analysed via a Bayesian Network. Geospatial and Earth observation data representing past hazard events and their trigger variables were analysed to derive the probability of a hazard. Results revealed that lowland areas covering 22,868 and 139,193 km<sup>2</sup>, or 5% and 29% of total lowland areas were at-risk at a 90% probability-level of floods in rainy-seasons and droughts in the summer. High mountains and the plateaus were exposed to landslides over 90% probability in rainy, and forest fires in summer with over 60% probability, covering 37,727 and 40,069 km<sup>2</sup>, respectively. Within the Bayesian Network four relations of multiple hazards were investigated. At a 90% significance level approximately 190,250 km2 was at risk from a combination of forest fires and droughts. At a 80% or greater probability, 161,450, 120,027, and 102,628 km<sup>2</sup> of land were at risk from a combination of 1) floods and landslides, 2) forest fires, floods, and landslides, and 3) all four hazards, respectively. The results were then used to produce the first fine-spatial scale multi-hazard assessment to support national policies on risk mitigation.</p>
Application and implementation of the hazard risk matrix to identify hospital workplaces at risk for violence