Executive summary
The HORIZON Europe GUARDEN project successfully concluded its 36-month mission to bridge the gap between advanced environmental observation, AI-driven analytics, and practical decision-making. By leveraging Earth Observation (EO) data, in-situ monitoring, and citizen science, GUARDEN developed a suite of operational Decision Support Applications (DSAs). These integrated solutions were validated across four European pilots and one exploratory case in Madagascar, delivering measurable impacts on biodiversity governance, spatial planning, and the implementation of the EU Green Deal.
1. Core Technological Achievements & Suite of Tools
GUARDEN delivered an open, scalable, and interoperable suite of digital tools designed to quantify biodiversity changes and Ecosystem Services (ES).
AI-Driven Species Identification
AvesEcho: A cutting-edge deep learning model for audio-based bird community identification in European soundscapes, now integrated into the Xeno-canto database.
Pl@ntNet API & Plots: Improved identification of vegetation surveys through the management of multi-image HD quadrat plots (e.g., 50×50 cm), thus improving field reports for practitioners.
High-Resolution Spatial Modeling
DeepSDM-framework: A deep learning framework for training species distribution models (SDMs) using multi-modal inputs has been developed and completed with a tool predicting habitat labels based on EUNIS classification from plant assemblages (cf. Pl@ntBERT).
Data Hubs & Visualization Platforms
GeoPl@ntNet: A centralized web mapping application delivering OGC-compliant WMS layers for high-resolution species and indicator predictions (at a 50 m resolution) across Europe.
Integrated DSA: A modular, web-based system that acts as the central integrative layer. Designed with tailored interfaces for each pilot, the DSA consolidates heterogeneous citizen science data streams, high-resolution GeoPl@ntNet predictions, acoustic monitoring data, and Bayesian scenario models, into a unified, user-friendly environment to drive spatial planning and policymaking.
CirculAR: An Augmented Reality mobile application projecting complex 3D models and spatial indicators to boost stakeholder engagement.
The project systematically identified and mapped 13 Key Exploitable Results to the Horizon Results Platform to ensure long-term sustainability and market uptake. Exploitable results are categorized primarily as technological solutions, models/algorithms, and high-value occurrence datasets.
2. Real-world validation: Project case studies
GUARDEN deployed its systems across diverse geographic and regulatory contexts, proving the adaptability and operational readiness of the platform.
France (Conservation vs. Infrastructure)
Applied GUARDEN tools to the Avoid-Reduce-Compensate (ERC) framework for the Montpellier–Perpignan railway project. Evaluated 3 distinct land-use scenarios, demonstrating how predictive SDMs can inform sustainable spatial planning. Achieved a species agreement rate of 68.39% between AI plant species predictions and expert assessments.
Greece (Wind Farm Monitoring)
Deployed novel acoustic monitoring networks across wind energy parks in Evia and Thrace. AvesEcho AI model showcased excellent acoustic detection capabilities by identifying an initial pool of 275 bird species, successfully accelerating the expert validation process and demonstrating the value of keeping human-in-the-loop expertise paired with advanced AI.
Spain (Urban Coastal Management)
The BioPlatgesMet initiative used the MINKA suite in the Barcelona Metropolitan Area. By processing >6000 citizen observations, public administrators utilized the MINKA Dashboard to detect 24 invasive and 46 protected species, triggering real-time adaptive management decisions (e.g., swapping mechanical beach cleaning for manual interventions).
Cyprus (Industrial Expansion)
Supported the Vasilikos Area Spatial Master Plan (VASMP) using integrated DSAs to map ES trajectories against industrial growth. Evaluated habitat health indices and avifauna disruptions to directly support Environmental Impact Assessment (EIA) committees.
Madagascar (Global Transferability)
Conducted >400 systematic field plots across gradients of degraded and intact ecosystems. Fostered immense citizen participation with >8000 uploads via a localized Malagasy version of Pl@ntNet, proving the tools’ viability in high-biodiversity, data-scarce tropical regions.
3. Capacity building, outreach & Open science
GUARDEN established an impact maximization plan centered on generating reusable public knowledge:
Open AI Challenges
Co-organized the internationally recognized PlantCLEF 2023, 24 & 25 data science challenges to push the boundaries of AI-driven multi-species quadrat detection, attracting >500 followers and 38 competing global research teams.
Comprehensive Training
Executed >50 events (webinars, hybrid bootcamps, field walks) educating over 1000 cumulative participants on how to use spatial DSAs and interpret biodiversity indicators.
Open Access Outputs & Documentation
Produced 39 scientific publications, 30 open-source code releases.D6.3 (Capacity building) & D5.4 (Performance evaluation) act as permanent references and replication guidelines for scaling similar systems across EU municipalities.
Policy Impact and Future Sustainability
GUARDEN’s ultimate legacy lies in its capacity to translate complex environmental observations into concrete responses for the EU Biodiversity Strategy for 2030 and the Nature Restoration Law.
By successfully embedding digital observatories and predictive DSAs into established frameworks (like France’s ERC processes or Spain’s municipal monitoring programs), GUARDEN provides a replicable blueprint for regional governments. Through strict adherence to the interoperability standards, continuous interaction with the EU JRC (in particular with the Knowledge Centre for Biodiversity and the Food security Unit) and integration with GBIF, GUARDEN ensures that its tools will continuously serve as a backbone for evidence-based spatial management long after the project’s official conclusion.
