Principles of Good Scientific Practice in Geoinformatics: From Data Management to Reproducible Code
Abstract
Principles of Good Scientific Practice in Geoinformatics: From Data Management to Reproducible Code Trial lecture (Probevorlesung, § 13 HabilO) for the Venia Legendi in Geovisual Analytics at the Faculty of Environmental Sciences, Technische Universität Dresden (October 15, 2026, Dresden, Germany). Reflecting the core principles discussed, this lecture and its underlying pipeline are openly archived as a version-controlled, persistent resource to demonstrate reproducible scientific practice in action. Abstract Spatial data science occupies a uniquely vulnerable position in the ongoing reproducibility crisis. Below high-level Python libraries lies a complex stack of compiled C/C++ system libraries (GDAL, GEOS, PROJ) where unannounced datum shifts or compiler updates silently alter coordinates without throwing errors. Concurrently, spatial planning practice frequently relies on GUI-based desktop GIS workflows and binary office documents (`.docx`), leaving zero audit trail for legal scrutiny or scientific verification. Grounded in the normative framework of the DFG Code of Conduct (Guidelines 7, 10, 11, and 12) and Mayer's principles of multimedia learning, this lecture outlines four foundational pillars to transition from fragile ad-hoc scripts to resilient, verifiable scholarship: Part I • Diagnose (The Reality Check): Identifying the root failure modes of spatial data science—from the "spatial curse" of compiled system dependencies and unrecorded point-and-click operations to the administrative friction of managing 22,000+ data cells across municipal Word documents. Part II • Structure (The Resilient Archive): Establishing core architectural principles: Separation of Concerns (SoC), Single Source of Truth (SSOT), and Knowledge Longevity. Demonstrating that tools (Typora, QGIS, Zotero) are merely disposable viewing lenses, whereas disciplined plain text (Markdown) and open formats (GeoPackage, GeoParquet) constitute the permanent archive of thought. Addressing the FAIR vs. ELSA dilemma through synthetic twins and probabilistic hashing. Part III • Construct (Freezing the Computation): Demonstrating parameterized, headless GIS batch pipelines. Using an authentic planning case study (1,500 wind turbine viewsheds calculated via headless GRASS GIS inside JupyterLab for regional planning), the lecture illustrates literate provenance, isolated Docker container layering (Carto-Lab), and automated GitLab CI/CD release pipelines. Part IV • Evaluate (The Generative Frontier): Governing the integration of non-deterministic Large Language Models (LLMs). Establishing Prompts as Code and conversational lineage archiving to maintain human-in-the-loop accountability, separating probabilistic model outputs from verifiable scientific scholarship. The lecture concludes with a pragmatic survival checklist for Master's theses and frames Good Scientific Practice through the lens of Calm Technology: disciplined conventions do not add administrative burden - they liberate researchers from technical friction to focus on intellectual synthesis. Artifacts in this Deposit `presentation.pdf`: Vector slide export rendered directly from the web presentation viewport. `presentation_source.zip`: Complete source repository containing Reveal.js Markdown files, assets, custom CSS, and interactive case study submodules. --- DFG Code of Conduct (Leitlinien zur Sicherung guter wissenschaftlicher Praxis): https://doi.org/10.5281/zenodo.14281892
