FreshCES-Net: A scalable deep learning approach to map freshwater cultural ecosystem services using social media data

• Maps freshwater CES using geotagged social media images. • Combines deep learning and spatial modelling in one framework. • Enables large-scale analysis with reproducible open-source tools. • Identifies CES hotspots across large scales. • Supports planning and conservation with high-resolution outputs. FreshCES-Net is a modular, scalable framework for mapping freshwater Cultural Ecosystem Services (CES) using geotagged social media images. It integrates automated photo retrieval, deep learning-based image classification, and spatial modelling in a fully reproducible pipeline. The classification module employs a fine-tuned ResNet-152 Convolutional Neural Network trained on 6911 Flickr images, achieving 0.92 accuracy and 0.91 recall across five CES categories. Spatial modelling is conducted using an XGBoost model trained on biophysical covariates such as population density, river order, naturalness, accessibility, protection status, and others. Model outputs include the weight of the biophysical variables over CES presence and maps that reveal areas with unexpected CES intensity not explained by demographic or environmental variables. The framework was applied across over 150 river basins in the Iberian Peninsula, enabling large-scale CES assessments with high spatial resolution. FreshCES-Net facilitates new research questions about how freshwater landscapes influence CES distribution at large scale, while also improving the reproducibility and scalability of existing methods. The software is designed for practical use by researchers, planners, and environmental managers, requiring only basic Python experience. It uses relative paths, modular notebooks, and intermediate outputs in CSV or Excel formats. Though not commercialized, the tool is actively used in applied research and is publicly available. FreshCES-Net offers a high-performance, accessible solution for integrating CES into freshwater planning, conservation strategies, and environmental decision-making at regional to continental scales.