Akram Elghouat

High mountain protected areas (PAs) are increasingly recognized not only for their role in conserving biodiversity but also for their contribution to the provision of cultural ecosystem services (CES). Despite their relevance, CES remain underrepresented in conservation planning, particularly due to challenges in quantifying their spatial distribution. This study combines geolocated social media data and ecological niche models (ENMs) to assess the spatial patterns and key drivers of CES supply across eight mountain PAs spanning distinct biogeographical regions in Spain and Portugal. Using deep learning techniques to classify more than 200,000 photographs into ten CES categories, we evaluated model performance under two modeling approaches and identified the most influential environmental and social predictors. Most CES categories exhibited good model performance (Boyce index > 0.5), though variation existed across services and regions. Nature & Landscape and Gastronomy CES showed strong associations with park boundaries and human settlements, respectively, while Religious and Cultural CES were spatially linked to culturally significant landmarks.. Our findings demonstrate the potential of combining social media data with ENMs to map CES distributions and reveal both universal and context-dependent drivers. This approach offers valuable insights for integrating CES into PA management and spatial planning, supporting more holistic and culturally inclusive conservation strategies.

This systematic review explores the use of social media data to spatially model Cultural Ecosystem Services (CES), which are non-material benefits provided by ecosystems that support human well-being. Based on a search of 510 scientific articles in Web of Science and Scopus, we carefully selected and analyzed those that focused on spatial modeling of CES using social media data. We aimed to (a) identify the diversity of CES assessed, (b) analyse the social media platforms used as data sources, (c) evaluate the modelling frameworks employed, and (d) summarise the predictor variables included in these models. We found that the most studied CES were those related to physical and psychological experiences (62%; especially recreation) and the main predictor variables were the presence of natural elements (52%), land use and land cover maps, and topographic variables, often weighted by applying distance-based metrics (24%). Twenty-four social media data sources were identified but Flickr was by far the most widely used one (40%). MaxEnt (37%) and Random Forest (16%) were the most commonly used modeling tools. The most commonly used metrics to assess model performance were AUC-ROC, AIC, and R2 values. While the use of social media offers an opportunity to study CES and provide cost-effective and scalable insights, this article discusses some limitations and considerations raised from the literature review to be taken into account when using this type of data. These include the quality and representativeness of social media data, the importance of a clear definition of CES, a proper labeling of social media data, and an appropriate selection of spatial modeling techniques. Future research should address these limitations and considerations by integrating different data sources and refining methodologies to improve the accuracy and applicability of CES models. This review provides a comprehensive overview of current practices and highlights areas for further investigation in the spatial modeling of CES using social media data.

Flash floods are highly destructive disasters, posing severe threats to lives and infrastructure. In this study, we conducted a comparative analysis of bivariate and multivariate statistical models and machine learning to predict flash flood susceptibility in the flood-prone Rheraya watershed. Six models were utilized, including frequency ratio (FR), logistic regression (LR), random forest (RF), extreme gradient boosting (XGBoost), K-nearest neighbors (KNN), and naïve Bayes (NB). We considered 12 flash flood conditioning variables, such as slope, elevation, distance to the river, and others, as independent variables and 246 flash flood inventory points recorded over the past 40 years as dependent variables in the modeling process. The area under the curve (AUC) of the receiver operating characteristic was used to validate and compare the performance of the models. The results indicated that distance to the river was the most contributing factor to flash floods in the study area. Moreover, the RF outperformed all the other models, achieving an AUC of 0.86, followed by XGBoost (AUC = 0.85), LR (AUC = 0.83), NB (AUC = 0.76), KNN (AUC = 0.75), and FR (AUC = 0.72). The RF model effectively pinpoints highly susceptible zones, which is critical for establishing precise flash flood mitigation strategies within the region.

Key message: Crown area, sapling height, and biovolume extracted from UAV-acquired RGB images provided accurate estimates of aboveground biomass and carbon stocks in a 5-year-old holm oak (Quercus ilex L.) plantation. Our models regressing UAV-derived sapling variables against ground-based measurements exhibited high R² values (0.78–0.89), thereby reflecting that RGB data can be used as an effective tool for measuring young individuals. Context: The monitoring of tree sapling performance from the early stages of reforestation is of particular importance in the context of the global efforts to restore forests. Yet, most models to estimate carbon sequestration are developed for adult trees. Thus, the few models specifically developed for young trees rely on ground-based field sampling of tree growth parameters, which is time-consuming and difficult to implement at large spatial scales. Aims: Our objectives were as follows: (1) to study the potential of UAV-based RGB imagery to detect and extract sapling variables (e.g., crown area, height, and biovolume) by comparing ground-based sapling measurements with UAV-derived data and (2) to compare the accuracy of the data estimated from RGB imagery with existing traditional field-based allometric equations. Methods: We used a 5-year-old holm oak (Quercus ilex L. subsp. ballota (Desf.) Samp.) plantation (N = 617 plants), and their crown area, height, and biovolume were estimated from RGB imagery. Subsequently, the plants were harvested and the UAV-derived data were compared with field-measured sapling height and aboveground biomass values. Carbon content in leaves and stems was measured in a subsample of the saplings to estimate carbon stocks. Results: The models fitted with UAV-derived variables displayed high performance, with R² values from 0.78 to 0.89 for height, leaf and stem biomass, total aboveground biomass, and carbon stocks. Moreover, aboveground biomass outputs calculated with field height and UAV-derived height using allometric equations exhibited R² values from 0.65 to 0.68. Conclusions: Given the affordable cost of RGB cameras and the versatility of drones, we suggest that UAV-based models may be a cost-effective method to estimate the biomass and carbon stocks of young plantations. However, further studies conducting drone flights in different conditions are needed to make this approach more scalable.