PhD in AI‐Driven Flood Scenarios and Insurability Maps; Pillar

Are you our new colleague?

We are looking for a PhD candidate to develop AI methods that extend current sets of flood scenarios derived from physical and numerical models, incorporate climate change effects, and then use these enriched datasets to assess the future insurability of flood risks. In addition, you will develop methods that use AI to produce rapid damage estimates for financial institutions in the event of an impending flood.

About FIRM

Flood Insurance and Risk Management (FIRM) is a five‑year research programme (2026–2030), funded by TKI Deltatechnologie, Nationale‑Nederlanden and contributions from more than 10 partners from academia, the financial sector, engineering consultancies and government. The central aim of FIRM is to estimate flood and water nuisance risk information as realistically as possible and to make it suitable for use in the financial sector.

FIRM consists of four substantive research pillars and a work package on knowledge transfer. This vacancy concerns a PhD position within Pillar 3. You will collaborate closely with the PhD candidates in the other pillars, be supervised by researchers at the University of Amsterdam, and work together with experts from the field.

In many countries, a set of flood scenarios for the current climate is developed using hydraulic models. These scenarios describe the consequences of a flood in terms of flood depth, flood extend, economic damage and loss of life. These scenarios are constructed for a given load (a water level on sea or rivers, or river discharge, with an associated probability of occurrence).

Using these probabilities and the consequences, flood risk can be described. The current datasets, developed with physical and numerical models, are the result of decades of hydraulic modelling. These models are used to answer water management questions; applications in the financial sector raise new questions and requires different and additional scenarios. There are gaps in the probability domain between normative events for water systems and stress tests events, and from an ESG and stress‑testing perspective there is a need to understand different scenarios of climate change (based on IPCC scenarios) in a specific year — these scenarios often do not yet exist.

This PhD project develops AI methods to close these gaps so that flood impacts are better represented in climate analyses. The research consists of:

• Developing hybrid AI methods to extend and improve existing sets of flood scenarios with additional events that fill gaps in the existing dataset of scenarios, incorporate non‑linear effects such as failure of regional flood defences, and account for uncertainty in breach growth. In this way, flood impacts can also be described probabilistically and later added to the model of Pillar 1;
• Translating the current scenario sets into future climate states (using climate scenario as KNMI’23 and new insights of IPCC) and climate adaptation pathways for arbitrary horizon years, enabling the financial sector to quantify climate effects on portfolios;
• Building AI methods to integrate the flood scenarios to
  1) early warning: a rapid operational damage assessment using flood and weather forecasts to support emergency warnings by insurers and
  2)
• Future risk assessments by long‑term insurability maps for both data‑rich and data‑scarce contexts to identify potential future tippings points for insurance;
• Applying and validating the methods in case studies in the Netherlands and internationally, in collaboration with project partners.

Main duties and responsibilities:

• Developing hybrid AI methods to extend existing sets of flood scenarios for dike breaches and extreme precipitation, and to model the impacts probabilistically;
• Translating these scenario sets into future climate states for specific horizon years and emission pathways (KNMI’23);
• Building AI methods for rapid operational damage estimation and for strategic long‑term classification of insurability;
• Validating methods in both data‑rich areas (such as the Netherlands) and data‑scarce regions;
• Integrating results with the risk models of Pillars 1 and 2;
• Presentin…