The ISIMIP Repository

The Inter-sectoral Impact Model Intercomparison Project (ISIMIP) [1] brings together worldwide climate impact modelling communities by providing high quality climate related and direct human forcing data, consistent simulations across different sectors (e.g. agriculture, water), and the curation and publication of model output data. These data provide the link between climate change and society and are therefore relevant not only to the scientific community, but also for various social and political stakeholders. All data are freely available from the ISIMIP Repository [2]. The repository contains more than 180Tb of data (as of 2025). Over the last few years, we have developed our own workflow for data management, quality control, and publication. The whole process is governed by a machine-actionable simulation protocol, which consists of a human-readable web page informing about the different variables and experiments, and machine-readable files, which are used by other components of the workflow. The protocol is edited using a combination of markdown and YAML files and is automatically built using GitHub Actions. Modelling groups use this protocol, along with the specified input data, to perform climate impact simulations using their models. The output is then transferred to the ISIMIP data space at the German Climate Computing Centre (DKRZ). There, the ISIMIP data team uses the ISIMIP quality control tool to perform various format checks, which ensure compliance with the different specifications of the protocol (file naming, NetCDF file layout). The tool can be easily installed by the modellers to check the data even before uploading. In order to be always up to date, it fetches the machine-readable part of the protocol via the Internet. In addition to these formal quality checks, we have also implemented a semi-automatic system to assess the data. This is done by automatically generating plots of model ensembles and then manually searching for irregularities or outliers. In the future, we would like to use artificial intelligence methods for this task, to seamlessly integrate automatized outlier detection in the publication process. After the check and an optional embargo period, the data is published on the repository using the ISIMIP Publisher. This tool copies the data onto a dedicated machine, extracts metadata, and inserts a record into a relational database. The repository itself consists of this database, the interactive website, which can be used to search and filter the datasets, and the file server, from which the files are downloaded. Whenever data needs to be replaced, because of discovered problems or because better data is available, the old datasets are archived on tape. A comprehensive "Issues & Notes" system informs about these changes or any other caveats with the data. We register DOI (with DataCite) for each sector or input data category and create subsequent versions of this DOI whenever new data is published or redacted. To ensure user-friendly data processing, a dedicated File API [3] can be used via the "Configure Download" button to perform operations on the server before download, e.g. a cut-out of a specific region.