Model evaluation for km-scale simulations within the Climate Adaptation Digital Twin: the AQUA approach

High-resolution climate simulations at the km-scale present significant challenges for data processing and analysis due to the massive data flow, memory bottlenecks, and scaling limitations. These demands exceed the capabilities of traditional data processing pipelines, requiring the development of innovative approaches to handle and analyze the data efficiently.

The Climate Adaptation Digital Twin, part of the European Commission’s Destination Earth initiative, addresses these challenges with a unified end-to-end workflow. This workflow integrates the full chain, from global km-scale climate simulations to real-world applications in sectors most affected by climate change (such as energy, hydrometeorology, wildfire management etc.)

Embedded within the Climate-DT framework, AQUA (Application for QUality Assessment) is a Python-based tool designed for the scientific evaluation of these simulations, featuring a core engine optimized for efficient access to high-resolution data and modular diagnostics suite that ensure consistent and scalable analysis. It runs within a containerized environment on high-performance computing (HPC) systems, ensuring portability and compatibility across different machines and computational infrastructures.

AQUA evaluates climate model outputs through key diagnostics (such as model biases, timeseries, top-of-the-atmosphere energy balance, ocean circulation metrics etc.) while also providing innovative diagnostics to analyze tropical rainfall, cyclone structures, and other km-scale processes traditionally challenging to study.

Within the Climate-DT workflow, AQUA enables real-time monitoring, systematic comparisons, and quality control of ongoing simulations, building confidence in model results for climate adaptation decisions. The pipeline automatically generates intake catalog entries for new data and creates a low-resolution archive derived from high-resolution data, allowing for more efficient execution of default diagnostics. As each new simulation year is generated, analyses are automatically performed and the results are published on a dedicated website and dashboard, offering visualizations and actionable insights based on real-time data.