Building an analysis-ready geospatial dataset for Convolutional Neural Networks-based rural property valuation using LiDAR 3D point clouds

In Portugal, fiscal valuation of rural properties is currently based on the agricultural income obtained in the previous year, disregarding both the productive potential of the land and its market value. To address this limitation, we propose a framework in which rural property valuation is based on the productive potential of the land, following the best practices recommended by the Food and Agriculture Organization (FAO) and the agroforestry profitability model proposed by EU Directorate-General for Structural Reform Support (DG REFORM), while incorporating artificial intelligence techniques through convolutional neural networks (CNN). This paper presents the proposed methodology, which comprises four modules: data acquisition (including airborne LiDAR 3D point clouds, land use and land cover (LULC) data, Sentinel-2 imagery, and FAO datasets), geospatial data processing, data integration, and an AI-based rural property valuation model. The first three modules were successfully implemented, resulting in an integrated analysis-ready geospatial dataset composed of 16 normalized bands at a spatial resolution of 10 m in the Portuguese coordinate reference system PT-TM06/ETRS89 (EPSG:3763). In future work, this dataset will be integrated with additional territorial variables influencing rural property valuation, such as accessibility, legal restrictions, easements, environmental risks, and hydrological conditions, serving as the input for the proposed CNN-based valuation model.