A GeoAI Framework for Built-Up Valuation through Building Typology: A Replicable Open-Source Approach for Regional-Urban Systems

Abstract This paper presents an open‑source, replicable, and scalable GeoAI framework for estimating Built-Up valuation (fixed capital) using building typology as a proxy. The methodology addresses a critical gap in urban planning and risk analysis: the absence of up‑to‑date, spatially explicit information on the value of the built environment, particularly in the Global South where official cadastral and census data are often intermittent, incomplete or unavailable.The approach is developed and tested in the Central‑Pampean Region of Argentina (Córdoba and Santa Fe provinces, 87,640 km²), a territory characterized by diverse ecoregions, urban forms, and productive landscapes. A labeled dataset of 4,800 buildings was constructed from open data (OpenStreetMap, Google Open Buildings V3, Street View) and manual labeling, integrating 13 spatial and morphological features into a prediction data frame. A Random Forest classifier was optimized via randomized cross‑validation, achieving a weighted F1‑score of 0.825 for base building typology (seven classes) on a hold‑out test set. Spatial autocorrelation analysis (Global Moran’s I) showed no significant residual clustering, confirming model adequacy.The study makes three main contributions: (1) a fully transparent workflow implemented with open‑source software (Python, QGIS) and open data, made publicly available with the dataset and DOI; (2) the introduction of Spatial Units of Land Occupation (SULO) -a novel spatial unit derived from road networks- that enables multiscalar analysis from individual buildings to metropolitan and productive regions, integrating agglomerations with operational landscapes in a Regional-Urban System (Brenner & Katsikis, 2020); and (3) a demonstration of technological capability, showing that institutions in the Global South can generate high‑quality territorial knowledge without proprietary licenses or paid data, using a standard PC hardware.The results validate that building typology, when combined with spatial and morphological variables, is a reliable proxy for fixed capital. In this case we focus on the building fabric. The methodology is transferable and scalable to other regions where open building footprints, heights and material networks are available, offering a practical tool for risk analysis, urban planning and mass appraisal. All code and dataset are provided under an open license.Keywords: building typology; Built-Up; massive valuation; mass appraisal; GeoAI; fixed capital; random forest; urban planning; spatial modeling; risk analysis; open‑source; replicable methodology; technological sovereignty; open science