NATURE FIRST: DEMO OF SOLUTIONS FOR BIODIVERSITY MONITORING
On April 26th, Nature FIRST had the opportunity to showcase achievements, solutions and technologies that focus on biodiversity preservation. We covered various topics such as taxonomy crossovers, ecosystem base maps, the 3edata Habitat Mapping method, TrapTagger for Species Recognition, and the Nature FIRST Knowledge Graph. We also had a Q&A session where participants engaged in a lively discussion about the ongoing challenges in biodiversity preservation, and we received valuable feedback from key stakeholders. If you weren’t able to tune in for the demo, you can still gain valuable insights by watching the recording. If you need a quick refresher, join us as we do an overview of what was discussed.
What is Nature FIRST?
The core of the Nature FIRST project is biodiversity. Our main challenges are to achieve near real-time biodiversity monitoring, to detect changes and trends in a very early stage, and to translate predictions into actionable and timely interventions.
We are working with four nature reserves spread over five countries: Ancares-Courel in Spain, Danube Delta in Romania, Maramures Transboundary Area in Romania and Ukraine, and the Stara Planina Mountain in Bulgaria. These field sites cover quite extensive territories and are inhabited by vulnerable species like the bear and the wolf, and icon species like the pelican and the sturgeon. The reserves face problems which can also be found all over Europe in Natura 2000 sites. These include – but are not limited to – habitat loss and fragmentation, fire, human-wildlife conflicts, poaching, and other related issues.
However, addressing these challenges is not easy. Firstly, mapping and monitoring changes in nature conservation areas is very costly, and as a result, it's not done very frequently, sometimes only every 5 to 10 years. Another problem is the absence of a common language. Taxonomies for classifying habitats, species and interventions vary from one region to another, which makes it difficult to aggregate data from various sites, observe general trends and draw overarching conclusions. Finally, many monitoring programs often result in information that is not sufficiently actionable, because the analyses are too late.
WIth this in mind, we aim for continuous near real-time monitoring of these areas to provide actionable information. This is achieved by combining remote sensing images (satellite-based and on-site), with environmental forensics, which is a science focused on extracting insights from field evidence. You can use environmental forensics to find causes and effects, but what is more important, you can use it proactively, to stop processes or crimes affecting a nature site as soon as possible. Another crucial part of this project is the development of digital twins, which are copies of the ecosystems. We step up from just mapping an area into near real-time monitoring it, and forecasting data. In this way, we get predictive information that can be used to prevent issues before they occur.
Most importantly, the Nature FIRST project is very practice-oriented. All the solutions that we are presenting were tested in the field and are ready to be used. After the end of the project or even during its implementation, these solutions will be available for Natura 2000 managers, researchers, analysts, policy-makers and other stakeholders that are interested or work with biodiversity-related issues.
Ecosystem base maps
Nature FIRST aims to obtain a comprehensive understanding of natural areas, particularly Natura 2000 sites, through ecosystem base maps. The goal is to transform the existing abundance of environmental data into actionable information.
A base map consists of multiple layers that serve as the building blocks for practical applications. These layers are made accessible so that they can create value and be directly utilised in the field. Currently, environmental data is being collected from the four designated field sites, and in turn, the analysis and management of the data are carried out in a generic manner, allowing the methods to further be applied to any Natura 2000 site.
A base map consists of different layers, including static (e.g., Digital Elevation Model) and dynamic (e.g. Normalised Difference Vegetation Index) layers. Static layers offer important information influencing various factors including species distribution, species abundance, human traffic, and more. Dynamic layers capture changes over time. Vegetation undergoes changes throughout the seasons, so this layer is regularly updated to ensure the availability of the latest information, while also maintaining an archive of collected data.
The base maps can also be used for Risk & Suitability Mapping, to assess the probability of illegal activities, or for habitat mapping. Additionally, the goal is to develop a Digital Twin, where the base map serves as a foundation for accessing real-time information about nearby features and potential risks in the field. In this way, you can walk into the field with the app and collect data, while at the same time being able to request all the data that is already available, such as: where is the nearest river, the nearest forest, and how large is it? What is the risk of encountering a poacher? What is the likelihood of encountering an elephant? The base map is a foundational layer as we aim to have all this information available in real time.
Habitat mapping method
The Nature FIRST project builds upon previous developments and focuses on three main objectives: generating habitat maps for protected areas and assessing habitat changes using remote sensing technologies, as well as obtaining parameters for assessing habitat conservation status.
The project plans to map the four field sites, which cover a large area of 29,800km2, by creating a generic model for habitat mapping. The model consists of training areas for supervised classification, using images primarily from the Sentinel satellite, and classification tools like machine learning models and thematic layers.
The habitat classification process involves classifying a 10 km square using Sentinel images; creating training areas with the correct habitats verified in the field; running the model; refining and generalising the results; and generating the final habitat map. This map uses the EUNIS classification system and helps assess conservation status and manage protected areas.
The project has developed a generic model and obtained training areas for the Ancares y Courel site. Next steps include integrating the habitat mapping tool into the Sensing Clues platform, establishing training areas in other field sites, and creating specific rules for each site. Future plans also involve developing a tool for detecting habitat changes and automating training area updates.
Camera-trap image recognition
The Nature FIRST project and other conservation organisations utilise camera traps to monitor biodiversity. Since manually checking and annotating the captured images is time-consuming, the project offers an AI-based solution.
The AI models can recognize humans, vehicles, animals, and even individual species, making the process more efficient. The models filter images, allowing users to focus on relevant ones. With AI, the time spent classifying images can be reduced by 98.44%, while only losing about 0.23% in precision.
Images can be uploaded to the platform to then be processed using AI models or manually annotated using the TrapTagger tool. Traptagger is an open source software solution developed in South Africa, which can be used by everyone for free. Once the images have been annotated, the results can be exported and imported back into the platform for analysis and reporting. If the results are also integrated with the Sensing Clues platform, this enables data consolidation, analysis, reporting, risk mapping and digital twins.
So what are our next steps? We are now testing the connection between TrapTagger and Sensing Clues, so we can import the data correctly. We will then also run some pilots and improve the models' performance in recognizing specific species in challenging conditions (such as bad light). Additionally, we will also improve animal counting methods.
Taxonomy crossovers
Implementing taxonomies is crucial for efficient monitoring and analysis. Without a shared understanding of concepts, monitoring becomes time-consuming, leading to low-frequency data collection and delayed insights. Analysts spend most of their time on data engineering instead of analysis, resulting in reactive protection measures that are often too late to prevent biodiversity decline.
However, using taxonomies resolves these challenges. By establishing a unified language and linking different versions of taxonomies, analytics can become the primary focus, enabling real-time monitoring and reporting. This shift leads to high-frequency data collection, low-latency insights, and timely action to reverse biodiversity decline.
Taxonomy alignment is achieved through crossovers, which entails, for instance, comparing concepts between different versions using spreadsheets. This creates a "Source of Truth" (used to ensure everyone is using the same data) that interlinks habitat descriptions, allowing machines and humans to interpret them consistently. This means that even though habitats might have a different name, if they semantically mean the same thing, this will be interpreted both by the machines and the humans in the same way. Exact match and close match relations are used to handle different naming variations.
The data is presented in a graph format, making it accessible and machine-readable. Project linking connects concepts across domains, such as habitats and species, facilitating comprehensive analysis. By converting data into RDF format and utilising unified views, disparate data sources are integrated into a knowledge graph, enabling researchers and developers to make complex queries and gain valuable insights.
The knowledge graph makes it possible to answer previously unattainable questions, such as identifying sites with declining species or finding species in specific habitats. It also enables recommendations based on typical species and facilitates the attraction of certain animals to specific habitats.
Future plans involve visualising information and further enriching the knowledge graph to enhance data analysis and decision-making processes.
Nature FIRST Knowledge Graph
The Nature FIRST knowledge graph is a comprehensive system that combines information about habitats, species, locations, and human influences in one place. It not only describes the data but also includes important metadata, such as data source and reliability. The metadata helps assess the quality of observations, considering factors like data age and seasonal variations. By integrating metadata and observations, the graph enables better understanding and utilisation of the information.
The knowledge graph overcomes challenges in data engineering by linking various data sources regardless of their structure or format. It can incorporate diverse data types, including images, spreadsheets, and app-based observations. Its multilingual and adaptable nature allows it to present information in different languages and classification systems, facilitating cross-domain collaboration.
Furthermore, the graph empowers users to answer questions, explore relationships, and make informed decisions. It can identify species from the red list in specific habitats or determine species distribution in adjacent areas. This knowledge is crucial for policy-making and conservation efforts.
In addition, the graph supports the development of technologies to address biodiversity loss, including early warning systems for human-wildlife conflicts and the use of environmental forensics to combat poaching.
We aim to make all of these solutions affordable and bring them closer to real-time implementation. In conclusion, along with the ecosystem base maps, the habitat mapping method, the TrapTagger tool, and the taxonomy crossovers, the NF knowledge graph represents a powerful tool for tackling the complex challenges of biodiversity conservation and management. It integrates diverse data sources, incorporates metadata, and enables cross-domain collaboration.
We hope you enjoyed learning more about Nature FIRST’s demonstrated solutions and technologies for biodiversity monitoring and preservation. If anything is unclear you can always watch the recording and download the slides here. To stay up to date about our next public demo, sign up for our newsletter!
