Carto-Lab Docker: An RDM Infrastructure for Transparent, Open, and Reproducible Spatial Data Science

Spatial data science, which encompasses subjects such as cartography, social geography, and urban planning, faces significant challenges in managing research data (RDM). These include setting up complex computational environments comprising numerous spatial libraries that often conflict with each other (e.g. GDAL dependencies), ensuring computational reproducibility across different operating systems and machines over time, and facilitating collaboration and the publication of transparent research workflows. This paper introduces Carto-Lab Docker (CLD), a practical RDM infrastructure consisting of components designed to address these challenges and implement RDM principles. At the core of CLD is a versioned Docker container stack running JupyterLab, pre-configured with a curated Python environment (worker_env) containing essential, compatible spatial data science packages (e.g., GeoPandas, PySAL, Rasterio, HoloViz suite). This eliminates environment setup barriers for researchers. Crucially, CLD is versioned using semantic versioning, with each version available via a GitLab Registry. Researchers can pull specific container versions, ensuring long-term reproducibility by allowing code execution within the exact environment it was developed in, aligning with FAIR principles (Findable, Accessible, Interoperable, Reusable) [3]. In addition to environment management, CLD integrates a publication workflow. With the help of web templates and Continuous Integration (CI) / Continuous Deployment (CD) pipelines (GitLab), Jupyter notebooks can be automated to produce and deliver versioned, static HTML websites. These websites often include supplementary data or interactive cartographic visualizations that are essential for timely peer review and public outreach. This promotes transparency and facilitates archiving the entire research process, from code to narrative and results. CLD supports collaboration by enabling multiple users to work simultaneously within the same JupyterLab notebook session. Live collaboration fosters knowledge exchange between researchers with different areas of expertise, such as coding and domain-specific knowledge. We provide workflows for extending CLD by adding R environments, Mapnik (the OpenStreetMap renderer), and GRASS GIS, as well as for deploying secure, isolated instances for teams and individuals. These instances offer benefits beyond the cloud, such as zero startup time and user-controlled versioning, providing an alternative to institutional JupyterHubs. For broader audiences, Carto-Lab Docker can be deployed as a spatial JupyterHub container image, which has been demonstrated in Kubernetes in the Microsoft Azure Cloud. CLD was developed since 2018 and has been successfully used in university lectures at TU Dresden, as well as in conference workshops, such as the EuroCarto. It has also been used in training materials, such as the NFDI4Biodiversity Jupyter Book (training.fdz.ioer.info). CLD has supported the production of reproducible supplementary materials for multiple peer-reviewed publications (see resources). By abstracting technical complexities and integrating environment management, versioning, collaboration, and publication, CLD allows spatial researchers to focus on their analytical work while adhering to robust research data management (RDM) practices. In this way, CLD exemplifies "RDM in Action." We consider CLD and its associated CI/CD processes to be integral components of the RDM workflow, establishing a foundation for the sharing and promotion of transparent, open, and reproducible spatial science. CLD references the FORCE11 data and software citation principles [1, 2].