Modelling cultural ecosystem services of river landscapes in the Iberian Peninsula with deep learning and social media images

Cultural Ecosystem Services (CES) are essential for human well-being, particularly those provided by river landscapes. Yet, CES remains overlooked in river conservation strategies due to its intangible nature and the methodological challenges involved in their assessment. This study introduces a nov...

Descripción completa

Detalles Bibliográficos
Autores: Comalada i Pla, Francesc, Acuña, Vicenç, Garcia Acosta, Xavier
Tipo de recurso: artículo
Estado:Versión publicada
Fecha de publicación:2025
País:España
Institución:Varias* (Consorci de Biblioteques Universitáries de Catalunya, Centre de Serveis Científics i Acadèmics de Catalunya)
Repositorio:Recercat. Dipósit de la Recerca de Catalunya
OAI Identifier:oai:recercat.cat:10256/27684
Acceso en línea:http://hdl.handle.net/10256/27684
Access Level:acceso abierto
Palabra clave:Cursos d'aigua -- Conservació
Stream conservation
Ecologia fluvial
Stream ecology
Aprenentatge profund
Deep learning
Descripción
Sumario:Cultural Ecosystem Services (CES) are essential for human well-being, particularly those provided by river landscapes. Yet, CES remains overlooked in river conservation strategies due to its intangible nature and the methodological challenges involved in their assessment. This study introduces a novel AI-based framework that integrates deep learning for image recognition and machine learning for modelling to assess CES across river landscapes at regional scale. ResNet-152 convolutional neural network was fine-tuned to classify 6911 Flickr images into CES categories. The classified photos were then linked to biophysical variables using an XGBoost model, enabling interpretable predictions of biophysical CES drivers across heterogeneous landscapes. Residual analysis of population-based predictions revealed spatial clusters of “added CES value,” highlighting cultural benefits not explained by demographic factors alone. This integrated approach goes beyond previous CES assessments by combining automated image classification, large-scale spatial mapping of CES, and interpretable modelling of biophysical variables, allowing the cost-effective identification of under-recognized CES hotspots. Findings highlight the value of quotidian urban rivers and protected areas as key CES hotspots. The framework is transferable, reproducible, and openly available, thereby bridging AI methods and conservation planning