Periodic Reporting for period 1 - FORSAID (Forest surveillance with artificial intelligence and digital technologies - FORSAID)
Periodo di rendicontazione: 2024-09-01 al 2025-08-31
Forests and the critical ecosystem services they provide are vital to life on our planet. Protecting them from environmental threats is therefore critical, especially as pests are becoming a growing challenge that requires new and effective tools to address. This view is at the core of the FORSAID project, which intends to build a comprehensive early detection framework for forest pests by both improving existing digital solutions and developing new ones. These efforts rely on a combination of tools and approaches, including remote sensing devices such as satellites and drones, smart traps, automatic identification systems, environmental DNA analysis, as well as the integration of citizen science tools and active stakeholder engagement to ensure that the solutions developed are practical and meet real-world needs.
A wide range of monitoring methods has been tested at various scales to improve how forest pests can be detected and monitored.
At large geographical scales, satellite and aerial images were used to study pests such as the pine processionary moth in France and Italy, the Norway spruce bark beetle in the Alpine region, and the oak lace bug in France and Slovenia. Historical aerial images helped reconstruct past outbreaks, and recent data allowed the mapping of tree damage over wide areas. At a smaller scale, drones and aircraft are being used to spot early signs of disease and stress in trees before visible damage occurs.
In parallel, the project focused on developing and testing tools to improve the detection and identification of forest pests at ground level. Smart traps equipped with cameras were tested in several countries for the monitoring of bark beetles and processionary moths. Moreover, systems based on the combined use of the Entomoscope and deep learning models were developed to automatically identify several groups of insect pests with high accuracy. Complementing these approaches, DNA-based methods are being created to detect pests from trap liquids and water flowing down tree trunks, allowing the presence of pests to be confirmed even when they are not directly observed.
In addition, the potential contribution of citizen science for early detection is being assessed using multiple methods, for example by analysing iNaturalist observations and by evaluating mobile apps and web-based platforms using citizen science and artificial intelligence for pest identification.
Stakeholder engagement is a key part of the project. A dedicated Stakeholder Committee brought together 26 forest managers, authorities, and other practitioners, whose feedback helped guide field testing and ensured that tools were assessed under realistic conditions.
At large geographical scales, satellite and aerial images were used to study pests such as the pine processionary moth in France and Italy, the Norway spruce bark beetle in the Alpine region, and the oak lace bug in France and Slovenia. Historical aerial images helped reconstruct past outbreaks, and recent data allowed the mapping of tree damage over wide areas. At a smaller scale, drones and aircraft are being used to spot early signs of disease and stress in trees before visible damage occurs.
In parallel, the project focused on developing and testing tools to improve the detection and identification of forest pests at ground level. Smart traps equipped with cameras were tested in several countries for the monitoring of bark beetles and processionary moths. Moreover, systems based on the combined use of the Entomoscope and deep learning models were developed to automatically identify several groups of insect pests with high accuracy. Complementing these approaches, DNA-based methods are being created to detect pests from trap liquids and water flowing down tree trunks, allowing the presence of pests to be confirmed even when they are not directly observed.
In addition, the potential contribution of citizen science for early detection is being assessed using multiple methods, for example by analysing iNaturalist observations and by evaluating mobile apps and web-based platforms using citizen science and artificial intelligence for pest identification.
Stakeholder engagement is a key part of the project. A dedicated Stakeholder Committee brought together 26 forest managers, authorities, and other practitioners, whose feedback helped guide field testing and ensured that tools were assessed under realistic conditions.
The project goes beyond existing forest pest monitoring practices by demonstrating how multiple digital tools can be effectively combined into an integrated early-warning approach.
New deep learning tools and a QGIS plugin for training deep learning models that further support species mapping and defoliation assessment were developed. Sentinel-1 and PALSAR-2 SAR data were processed to distinguish forest cover changes and identify causes. A multi-sensor monitoring strategy was drafted, combining Sentinel-1/2, PlanetScope, and UAV data for validation.
The optimisation of smart traps for bark beetles and processionary moths represents another significant improvement over standard monitoring techniques, with automated counting and real-time data access significantly reducing the workload and enabling faster responses to emerging threats. The technology is ready for use by the stakeholders and it will be demonstrated in the coming meeting.
The combined use of the Entomoscope imaging device and deep learning models enabled the accurate identification of several Agrilus species, and work is ongoing to extend this approach to more insect groups. The open-source nature of this technology, together with the relatively low cost of setting up and operating an Entomoscope, makes AI-based identification systems highly suitable for technology transfer initiatives. This approach could become a valuable tool for phytosanitary personnel, foresters, and environmental agencies, supporting more efficient monitoring and management of pest species.
To identify taxonomic and spatial gaps in citizen science data, a list of 1,781 native and non-native pest species was compiled and over 1.3 million iNaturalist observations were downloaded, with plans to integrate this dataset with records from GBIF. Preliminary results showed that search interest often peaked during or before outbreaks, indicating potential for early detection, but also highlighting challenges such as data context-dependence and lack of demographic detail.
Stakeholders were consulted to assess their needs, expectations, and constraints regarding the adoption of digital technologies for monitoring forest pests. The results revealed a strong alignment between the pest species prioritised by the FORSAID project and those considered important by stakeholders, underscoring the project's relevance and potential impact. Additionally, the survey provided insights into the digital tools currently employed by stakeholders, highlighting existing gaps and areas where improvements are needed to enhance detection, identification, and monitoring practices. Overall, the findings validated the strategic direction of FORSAID and provided valuable guidance for shaping the next steps of the project.
New deep learning tools and a QGIS plugin for training deep learning models that further support species mapping and defoliation assessment were developed. Sentinel-1 and PALSAR-2 SAR data were processed to distinguish forest cover changes and identify causes. A multi-sensor monitoring strategy was drafted, combining Sentinel-1/2, PlanetScope, and UAV data for validation.
The optimisation of smart traps for bark beetles and processionary moths represents another significant improvement over standard monitoring techniques, with automated counting and real-time data access significantly reducing the workload and enabling faster responses to emerging threats. The technology is ready for use by the stakeholders and it will be demonstrated in the coming meeting.
The combined use of the Entomoscope imaging device and deep learning models enabled the accurate identification of several Agrilus species, and work is ongoing to extend this approach to more insect groups. The open-source nature of this technology, together with the relatively low cost of setting up and operating an Entomoscope, makes AI-based identification systems highly suitable for technology transfer initiatives. This approach could become a valuable tool for phytosanitary personnel, foresters, and environmental agencies, supporting more efficient monitoring and management of pest species.
To identify taxonomic and spatial gaps in citizen science data, a list of 1,781 native and non-native pest species was compiled and over 1.3 million iNaturalist observations were downloaded, with plans to integrate this dataset with records from GBIF. Preliminary results showed that search interest often peaked during or before outbreaks, indicating potential for early detection, but also highlighting challenges such as data context-dependence and lack of demographic detail.
Stakeholders were consulted to assess their needs, expectations, and constraints regarding the adoption of digital technologies for monitoring forest pests. The results revealed a strong alignment between the pest species prioritised by the FORSAID project and those considered important by stakeholders, underscoring the project's relevance and potential impact. Additionally, the survey provided insights into the digital tools currently employed by stakeholders, highlighting existing gaps and areas where improvements are needed to enhance detection, identification, and monitoring practices. Overall, the findings validated the strategic direction of FORSAID and provided valuable guidance for shaping the next steps of the project.