This doctoral research project explores a multimodal artificial intelligence approach to rediscovering and interpreting lost cultural heritage in rural landscapes. The project integrates computer vision on deteriorated historical manuscripts with multispectral territorial analysis (drone imagery, satellite data, and soil sampling) to identify and geolocate forgotten heritage assets and historical landscape features. The proposed pipeline encompasses: deep learning-based image restoration and Handwritten Text Recognition (HTR) for Spanish historical documents; Natural Language Processing (NLP) for entity extraction; georeferencing of extracted knowledge within a Historical Geographic Information System (GIS); and a predictive model correlating archival document information with spatial data to predict locations of potential heritage features, historical land use patterns, and landscape transformations. Results will be validated primarily through non-invasive archaeological techniques. A Retrieval-Augmented Generation component will link extracted knowledge to assist experts in interpreting results and exploring the enriched heritage knowledge base. The present document delineates the motivation, state of the art, methodology, expected outcomes, and future work.