Towards a Portugal’s LADM-VM country profile: Conceptual integration of 3D variables and Deep Learning for rural property valuation

Resumen

In the context of Land Administration Systems (LAS) and with the aim of international standardization, the Land Administration Domain Model (LADM) was introduced in 2012. LADM emphasizes the importance of registering information about both public and private properties, encompassing not only surface-level aspects but also subsurface and above-surface components (3D aspects), while linking legal dimensions of ownership. To incorporate the property valuation component, the LADM Valuation Information Model (LADM-VM) was recently published (July 2025) as Part IV of the second edition of LADM. Although Portugal has not yet implemented a comprehensive 2D cadastre covering its entire territory, it has made continuous efforts to register rural properties. A significant milestone was reached in 2017 with the establishment of the Balcão Único do Prédio (BUPi), a simplified 2D cadastral system for rural land. However, this system was not designed to support real estate valuation, even for fiscal purposes, which in Portugal still relies on the tax asset value (Valor Patrimonial Tributário – VPT). The VPT faces notable limitations, particularly due to the consistently low taxation of rural land. In recent years, several research groups have proposed models to revise how property tax is determined. However, these approaches have not incorporated LADM-VM, nor have they considered the integration of volumetric and 3D spatial data, elements that could substantially enhance the robustness and accuracy of property valuations. This study proposes the initial development of a Portugal’s country profile based on LADM-VM, aiming to modernize cadastral systems and rural property valuation practices for fiscal purposes. The proposed valuation approach relies on the productive potential of the land as the basis for determining a revised VPT, ultimately contributing to more equitable and efficient land administration policies. The strategy seeks to integrate cadastral information, 3D spatial variables, and key property attributes. This approach is grounded in LADM-VM and leverages Deep Learning architectures, such as Convolutional Neural Networks for processing image data, alongside Multilayer Perceptrons for handling alphanumeric data. The methodology adopted in this study follows the three-phase framework proposed by Kalogianni et al. (2021), consisting of scope definition, profile creation, and testing. In addition, a review of relevant literature was conducted to support the methodological approach. Building on the results of this initial investigation, the first phase has established the foundation for developing the Portuguese Country Profile. The next steps include designing the UML diagram by combining standard LADM classes with newly proposed extensions.