Periodic Reporting for period 2 - GUARDEN (safeGUARDing biodivErsity aNd critical ecosystem services across sectors and scales)
Periodo di rendicontazione: 2024-05-01 al 2025-10-31
Biodiversity governance is shaped by multiple legal frameworks, strategies and action plans at global and EU level, but implementation is often constrained by cross‑sector and cross‑scale trade‑offs (e.g. conservation vs. development, and EU targets translated into national/local decisions). GUARDEN addressed this gap by strengthening evidence-based decision making through sustained Multi‑Stakeholder Partnerships (MSPs) and user‑oriented Decision Support Applications (DSAs). The project combined environmental observations with Earth Observation and AI (notably deep learning) to increase the availability of standardised, geolocalised biodiversity data; produce predictive models and high‑resolution maps of biodiversity, habitats and ecosystem services under multiple pressures; and deliver policy-relevant indicators usable in planning and management workflows.
GUARDEN activated and sustained MSPs across five pilot contexts (France, Greece, Spain, Cyprus, plus an exploratory deployment in Madagascar), evolving from initial stakeholder mapping to co‑creation and validation through workshops, focus groups, surveys and iterative feedback loops.
The GUARDEN system architecture and specifications were delivered and strengthened after the mid‑term review, consolidating system‑level requirements (performance, security, usability, authentication, harmonised data exchange/export) and clarifying component boundaries and the separation of observed vs. predicted data.
GUARDEN finalised and released AvesEcho, an open bird sound recognition model built via transfer learning using Xeno‑canto and pre‑trained sound models (BirdNET/AudioSet). AvesEcho was deployed via Naturalis’ “Arise” system and integrated into Xeno‑canto for suggesting species on unknown European recordings, and it processed GUARDEN case‑study audio data (with additional expert validation workflows in Greece to improve reliability).
Pl@ntNet’s vegetation survey capability was upgraded and operationalised as an enhanced plots/surveys workflow (second version), available as a developer REST/JSON service (authorised access) and an end‑user interface for plot/quadrat images with automated multi‑species predictions and cover estimates. Pl@ntNet’s model scale expanded to 86k species (Aug 2025 release).
Open modelling components were consolidated, including MALPOLON (high‑resolution SDM framework) and HDM‑framework (species‑to‑habitat classification with EUNIS labels), supported by transformer-based advances such as Pl@ntBERT and by semantically enriched Species Interaction Networks (knowledge graphs and reproducible pipelines).
EU‑ and case‑study‑scale mapping was scaled into production: GUARDEN generated and served thousands of plant species distribution maps at 50 m across continental Europe (≈15 TB) via interoperable WMS/GeoTIFF, integrated into GeoPl@ntNet, and produced EU‑wide biodiversity indicators and habitat predictions (SDM→habitat), alongside case‑study LULC/NDVI dynamics and ecosystem‑service mapping.
GUARDEN delivered a mature ICT ecosystem: GeoPl@ntNet became a stable Europe‑wide mapping platform (OGC‑compliant WMS, reporting and area-based analysis tools); CirculAR progressed to mobile deployment with pilot‑tailored AR content (Greek bioacoustics; Spanish dune biodiversity) and Pl@ntNet API integration; and MINKA was extended with a tailored dashboard (alerts), multilingual chatbot, and community forum.
Finally, GUARDEN completed, deployed and validated the Integrated GUARDEN System and four DSAs (France, Greece, Spain, Cyprus), integrating observations (e.g. GBIF/MINKA/Pl@ntNet), predictive layers and indicators (GeoPl@ntNet), landscape dynamics, ecosystem‑service scenarios, and (where relevant) acoustic monitoring outputs, enabling exploration, comparison and export for real planning/management use.
The GUARDEN system architecture and specifications were delivered and strengthened after the mid‑term review, consolidating system‑level requirements (performance, security, usability, authentication, harmonised data exchange/export) and clarifying component boundaries and the separation of observed vs. predicted data.
GUARDEN finalised and released AvesEcho, an open bird sound recognition model built via transfer learning using Xeno‑canto and pre‑trained sound models (BirdNET/AudioSet). AvesEcho was deployed via Naturalis’ “Arise” system and integrated into Xeno‑canto for suggesting species on unknown European recordings, and it processed GUARDEN case‑study audio data (with additional expert validation workflows in Greece to improve reliability).
Pl@ntNet’s vegetation survey capability was upgraded and operationalised as an enhanced plots/surveys workflow (second version), available as a developer REST/JSON service (authorised access) and an end‑user interface for plot/quadrat images with automated multi‑species predictions and cover estimates. Pl@ntNet’s model scale expanded to 86k species (Aug 2025 release).
Open modelling components were consolidated, including MALPOLON (high‑resolution SDM framework) and HDM‑framework (species‑to‑habitat classification with EUNIS labels), supported by transformer-based advances such as Pl@ntBERT and by semantically enriched Species Interaction Networks (knowledge graphs and reproducible pipelines).
EU‑ and case‑study‑scale mapping was scaled into production: GUARDEN generated and served thousands of plant species distribution maps at 50 m across continental Europe (≈15 TB) via interoperable WMS/GeoTIFF, integrated into GeoPl@ntNet, and produced EU‑wide biodiversity indicators and habitat predictions (SDM→habitat), alongside case‑study LULC/NDVI dynamics and ecosystem‑service mapping.
GUARDEN delivered a mature ICT ecosystem: GeoPl@ntNet became a stable Europe‑wide mapping platform (OGC‑compliant WMS, reporting and area-based analysis tools); CirculAR progressed to mobile deployment with pilot‑tailored AR content (Greek bioacoustics; Spanish dune biodiversity) and Pl@ntNet API integration; and MINKA was extended with a tailored dashboard (alerts), multilingual chatbot, and community forum.
Finally, GUARDEN completed, deployed and validated the Integrated GUARDEN System and four DSAs (France, Greece, Spain, Cyprus), integrating observations (e.g. GBIF/MINKA/Pl@ntNet), predictive layers and indicators (GeoPl@ntNet), landscape dynamics, ecosystem‑service scenarios, and (where relevant) acoustic monitoring outputs, enabling exploration, comparison and export for real planning/management use.
- Community-level vegetation surveys from plot images (Pl@ntNet plots/surveys): operational multi‑species inference (with cover estimates) directly from plot/quadrat images, embedded in a widely used platform.
- Unprecedented EU-scale mapping: plant species maps at 50 m over continental Europe (≈15 TB) served via open geospatial standards; SDM V2 adds multimodal inputs and improved operational performance.
- Reusable open pipelines: MALPOLON and HDM‑framework provide modular, open-source training/inference/evaluation workflows that support reproducibility and external reuse.
- Pl@ntBERT: transformer-based habitat modelling from species assemblages (EUNIS prediction) underpinning SDM→habitat mapping workflows; published in Nature Plants (https://doi.org/10.1038/s41477-025-02105-7(si apre in una nuova finestra)).
- Citizen science to action: MINKA + validation + dashboard alerts enabled an institutionally embedded chain from observations to management interventions in the Spanish pilot, extending citizen science beyond outreach.
- Unprecedented EU-scale mapping: plant species maps at 50 m over continental Europe (≈15 TB) served via open geospatial standards; SDM V2 adds multimodal inputs and improved operational performance.
- Reusable open pipelines: MALPOLON and HDM‑framework provide modular, open-source training/inference/evaluation workflows that support reproducibility and external reuse.
- Pl@ntBERT: transformer-based habitat modelling from species assemblages (EUNIS prediction) underpinning SDM→habitat mapping workflows; published in Nature Plants (https://doi.org/10.1038/s41477-025-02105-7(si apre in una nuova finestra)).
- Citizen science to action: MINKA + validation + dashboard alerts enabled an institutionally embedded chain from observations to management interventions in the Spanish pilot, extending citizen science beyond outreach.