Portfolio Flood Modelling for Insurance Client

Project Overview

Insurance and reinsurance companies have traditionally relied on vendor catastrophe models to assess flood risk across their portfolios. These models — produced by companies such as JBA Risk Management, RMS (now Moody’s), and Verisk — use national-scale flood mapping and statistical methods to estimate losses at a portfolio level. They are valuable tools for aggregate risk management and capital modelling, but they have significant limitations when applied to individual high-value properties.

The fundamental problem is resolution. Catastrophe models operate at a grid resolution of 25-50 metres or more, which is too coarse to capture the specific flood risk to an individual building. A distribution centre sitting on a raised platform 1 metre above the surrounding ground may be shown as being within the flood extent on a catastrophe model, when in reality it is well above the design flood level. Conversely, a property in a topographic depression may appear to be outside the flood extent when it is actually at significant risk of ponding during intense rainfall events.

This project was commissioned to bridge that gap — providing site-specific flood modelling at a resolution that catastrophe models cannot achieve, while maintaining the consistency and comparability needed for portfolio-level analysis.

Methodology

Consistent Modelling Framework

To ensure comparability across the portfolio, we developed a standardised modelling framework that was applied to all 15 sites. The key elements were:

• Software: TUFLOW HPC (GPU-accelerated) for all 2D modelling.
• Grid resolution: 2-metre grid for all models, providing sufficient detail to resolve individual buildings, boundary walls, and local topographic features.
• Return periods: A consistent set of five return periods (1 in 20, 1 in 50, 1 in 100, 1 in 200, and 1 in 1,000 year) applied to all sites.
• Climate change: All return periods were modelled both with and without climate change allowances, using the EA’s current upper end allowances for the relevant river basin district.
• Outputs: Flood depth, extent, velocity, and hazard maps for each return period and scenario, delivered in GIS format (GeoTIFF and shapefile).

Flood Mechanism Identification

Each site was assessed individually to identify the relevant flood mechanisms. Of the 15 sites:

• 8 sites were primarily at risk from fluvial flooding (proximity to main rivers).
• 4 sites were primarily at risk from surface water flooding (no significant watercourse nearby, but located in topographic depressions or areas with poor drainage).
• 2 sites were at risk from combined fluvial and tidal flooding (coastal locations).
• 1 site was at risk from combined fluvial and surface water flooding (river nearby with additional overland flow paths).

For the fluvial sites, we obtained existing EA 1D model data where available and used it to generate boundary conditions for the 2D TUFLOW models. Where no EA model was available, we used the ReFH2 methodology to generate design hydrographs from catchment descriptors.

Model Build and Quality Assurance

The models were built by a team of three modellers working in parallel, with a senior modeller providing quality assurance reviews on each model before the design runs were executed. The QA process checked model setup, boundary conditions, terrain processing, roughness assignments, and results against the EA’s published flood mapping for plausibility.

This parallel workflow enabled us to deliver all 15 models within the 12-week programme. Each model took approximately one week to build, one week to run and post-process, and one week for QA and reporting, with three models progressing simultaneously through the pipeline.

Key Findings

Properties Where Risk Was Overestimated

Seven of the 15 properties showed significantly lower flood risk than the catastrophe models predicted. In most cases, this was because the site-specific modelling captured local topographic features — raised platforms, boundary walls, or embankments — that provided protection but were not resolved in the catastrophe model’s coarser grid.

For one distribution centre in the Midlands, the catastrophe model showed flood depths of 0.5-1.0 metres during a 1 in 100 year event. Our 2D model, which captured the 1.2-metre-high concrete boundary wall around the site perimeter, showed that the property remained dry in all events up to and including the 1 in 200 year event, with shallow flooding (less than 100mm) occurring only in the 1 in 1,000 year scenario.

Properties Where Risk Was Underestimated

Three properties showed higher flood risk than the catastrophe models suggested. Two of these were located in topographic depressions where surface water ponding during intense rainfall was the primary risk — a flood mechanism that catastrophe models, which focus primarily on fluvial risk, tend to underrepresent.

The third was a data centre located behind a river flood defence that was in fair-to-poor condition. The catastrophe models assumed the defence was fully effective, while our modelling included a breach scenario that showed the property could experience rapid, deep flooding in the event of defence failure.

Portfolio-Level Insights

The portfolio-level analysis ranked all 15 properties by their modelled annual average loss (AAL) and identified the properties that contributed most to the portfolio’s aggregate flood risk. Four properties accounted for over 70% of the total AAL, providing a clear focus for the syndicate’s risk management and reinsurance strategy.

Ongoing Engagement

Following the success of the initial commission, the syndicate instructed annual model updates for the five highest-risk properties. These updates incorporate any changes to flood defences, land use, or climate change allowances, ensuring that the underwriting team always has current flood risk intelligence for their most exposed assets.

The project demonstrated the value of site-specific flood modelling for insurance applications — providing the granularity that catastrophe models lack while maintaining the consistency needed for portfolio-level decision-making.