Governing Generative Landscape Design: Spatial AI and Indigenous Protocols for Heritage Reconstruction in Morocco's High Atlas

Advances in spatial artificial intelligence and generative modeling are transforming how landscapes are analyzed, designed, and reconstructed following environmental disruption. Yet these technologies are typically developed within technical frameworks that treat landscapes primarily as datasets, abstracted from the social institutions, governance systems, and cultural practices that shape how communities inhabit and steward place. This project examines how generative spatial modeling systems can be governed so that their data structures, training inputs, and design outputs reflect the knowledge systems and decision-making protocols of the communities whose landscapes they represent. Focusing on post-earthquake reconstruction in Morocco’s High Atlas Mountains, the research develops a spatial data pipeline that integrates semantic segmentation, generative modeling, and participatory metadata governance with Indigenous Amazigh institutional practices to produce community-guided digital reconstructions of heritage landscapes. The research centers on villages affected by the September 2023 earthquake, where the destruction of earthen architecture and communal infrastructure has prompted both national reconstruction initiatives and locally organized rebuilding efforts. Digital planning tools are increasingly used to guide reconstruction, but these systems often operate through top-down technical frameworks that risk sidelining vernacular architectural knowledge, collective governance systems, and culturally embedded spatial practices. This project explores an alternative approach in which spatial AI models are co-developed so that local governance protocols are embedded directly into the metadata structures, classification systems, and generative rules that guide landscape reconstruction. Methodologically, the project combines field-based spatial documentation with computational modeling techniques. High-resolution imagery and LiDAR scans are processed through semantic segmentation workflows that classify building typologies, circulation networks, agricultural terraces, and communal infrastructure. These segmented spatial datasets are integrated into generative modeling environments that simulate reconstruction scenarios informed by both historical building patterns and locally defined design priorities. To coordinate these computational processes across scales, space syntax analysis is introduced as a macro-level analytical framework linking spatial configuration, circulation patterns, and communal interaction. Because syntactic measures of connectivity, integration, and spatial hierarchy capture how spatial systems structure movement and access, they provide a shared analytical language through which communities, planners, and model designers can evaluate generative outputs. Within this framework, space syntax also functions as a governance interface that guides iterative editing of metadata taxonomies and model parameters. By linking spatial metrics to generative design rules, the project enables communities and planners to assess how modifications to training datasets, classification systems, and generative constraints influence reconstructed landscape configurations. Rather than treating generative models as neutral design tools, the project conceptualizes them as governance infrastructures that shape which forms of knowledge become legible within planning systems. By embedding Indigenous Amazigh governance protocols, including communal decision-making structures and vernacular design logics, into metadata standards and generative design pipelines, the research proposes a model for community-governed spatial AI systems. Beyond the Moroccan case study, the project contributes to broader debates about spatial AI governance, climate adaptation, and digital heritage systems in the Global South.