Preserving biodiversity, encompassing species and their habitats, is gaining significant attention and becoming a central concern, alongside the focus on climate change. Climate change directly impacts biodiversity and is a prominent aspect of Environmental, Social, and Governance (ESG) criteria. At the EU level, designated areas called Natura 2000 sites have been established for protection and conservation, aimed at safeguarding habitats and species. However, the data regarding these sites, habitats, and species is currently dispersed and isolated, resulting in limited usefulness. To address this issue, we introduce our work on a Knowledge Graph (KG) for biodiversity, known as Nature First KG. This KG aims to connect various data silos, including information about sites, species, and habitats, through cross-references called crossovers. Combining it with a digital twin, we empower recommender use cases such as: preventing human-wildlife conflicts, facilitating species reproduction, and combating illegal poaching to name a few.
CEUR ceur-ws.org
Climate change is one of the main challenges that has preoccupied mankind in the recent
decades. The efects of climate change have critically altered ecosystems and biodiversity all
around the world, changes in ecosystem range and distribution, ecosystem composition, local
species extinctions or mass mortality events of plants and animals have been observed [
Furthermore, at the European Union (EU) level, designated areas called Natura 2000 sites CEUR Workshop Proceedings have been established for protection and conservation, aimed at safeguarding habitats and species. However, the data regarding these sites, habitats, and species is currently dispersed and isolated, resulting in limited usefulness. Data is provided by diferent organizations and classification systems such as EUNIS2 and IUCN 3 in diferent structure, format and completeness; with IUCN reporting mainly on threatened species aka. “Red List Species.” The problem is further exacerbated due to existence of diferent versions of habitat taxonomies alongside habitat names and codes that have changed over time but in fact are equivalent, namely habitat “Subarctic and alpine dwarf Salix scrub” with code S21 in EUNIS ver. 2021 versus “Subarctic and alpine dwarf willow scrub” with code F2.1 in EUNIS ver. 2012. Domain experts have created spreadsheets maintaining the relationships between habitats – describing how one habitat maps to another using crossovers, i.e., if they are equivalent (=), superset (>), subset (<) or overlap (#) to the designated habitats in other versions. Despite those eforts, the data is not linked and contextualized to a larger context, and the semantics of such relations are only known to those experts. In addition, the occurrences of species that are known to exist in habitats are written using latin names as strings (Ursus arctos), without further connection to the source of truth species URIs for looking up and dereferencing them as things (https: //eunis.eea.europa.eu/species/1568). Similarly, data about sites have connections to habitats and species, and are also of spatial form that are provided in shapefiles along with geometric coordinates. This opens new challenges in terms of querying and performing geo-calculations with polygons, in addition to having relations to habitats and species as triple patterns by using GeoSPARQL. For a better presentation, we have summarised all the discussed problems and challenges in Table 1.
In this paper, we present our work towards a Knowledge Graph (KG) for biodiversity in the context of Nature FIRST research project4—dubbed Nature FIRST KG—that connects silos of data, namely sites, species and habitats by using cross-references, so-called crossovers. The imported data is heterogeneous, ranging from shapefiles to tabular data, which are then mapped, integrated and consolidated in a KG; afterwards the entities in KG are linked using relations based on crossovers, constituting Linked Open Data (LOD). The crossover relations are based on SKOS relations with well-defined meaning, namely for habitat mapping e x a c t M a t c h , b r o a d M a t c h , n a r r o w M a t c h , or c l o s e M a t c h ; in other cases we use a bespoke OWL (object) property e.g., h a s D i a g n o s t i c S p e c i e s specifying indicator species for a habitat. The relation from a site to a habitat also contains the coverage information in percentage that has motivated us to use the RDF*5 (RDF-star) data model for representing the percentages in the relation itself in a compact way akin to property graphs. We summarize our contributions: • Provide a KG that semantically links disparate information, allowing to traverse and get new insights powering new use cases pertaining to biodiversity; • Methodology for creating relations from crossovers in KG; • Publish and Consume KG using LOD frontends, graph view and SPARQL endpoint to comply with FAIR principles. 2European Nature Information System of the European Environment Agency (EUNIS/EEA), https://eunis.eea.europa. eu/ 3The International Union for Conservation of Nature, https://www.iucn.org/ 4https://www.naturefirst.info/ 5https://www.w3.org/2021/12/rdf-star.html
Problem
Silo-ed data Dataset versions String occurrences
Shape files Natura 2000 sites
Authority sources EUNIS/EEA, IUCN habitats & species
EUNIS/EEA habitats EUNIS/EEA sites, habitats, species
EUNIS/EEA sites EUNIS/EEA sites
Challenges completeness, mapping mapping, semantics
entity extraction GeoSPARQL computations reified statements
The data ingested in the KG comprises of habitats, species and Natura 2000 sites. There are various data authorities when it comes to habitat and species data, such as EUNIS and IUCN. The data sources are in diferent formats, schemas and completeness (c.f. Table 2). In addition, within EUNIS there exist diferent version of habitats (ver. 2017, 2021) that map to a legacy one (ver. 2012). The requirement is to consolidate the data into a KG, with each version having a crossover link to the source of truth or legacy version. The advantage of this approach is that one can report data that is already described using a taxonomy by specifying another taxonomy that is interlinked. Each version has its own description, codes and granularity in terms of broader relationships in the SKOS hierarchy. It is worth mentioning that Red List Species contained the taxonomic rank (S p e c i e s - > G e n u s - > F a m i l y - > O r d e r - > C l a s s - > P h y l u m - > K i n g d o m ), whereas EUNIS only contained ‘genus’ as a parent relationship. In both cases, s k o s : b r o a d e r relationships were created in order to create the hierarchy.
As seen from Table 2, some of the data is already in RDF, while others are non-RDF and
need to be transformed using ETL (Extract-Transform-Load). For the transformation, we used
UnifiedViews [
Regarding URI management — by following the Linked Data principles [
7https://www.eea.europa.eu/data-and-maps/data/natura-14/natura-2000-spatial-data 8https://www.eea.europa.eu/data-and-maps/data/eunis-habitat-classification/eunis-habitat-classification-review-2017 generate the URI, we made sure that it conforms to our Linked Data frontend so that it becomes dereference-able, e.g., https://sensingclues.poolparty.biz/HabitatClassificationScheme/237.
In Table 2 is shown the current snapshot of Nature First KG. Per each project (taxonomy) are given the stats such as the input data, number of total concepts, the crossovers with respect to other projects, and the total number of relations with respect to other projects. The relations are only materialized in direct relationships without storing the inverse relations - as also seen from the ‘no value’ (-) for #1 EUNIS Species.
No # Project (Taxonomy) 1 EUNIS Species 2 EUNIS Habitats 2012 3 EUNIS Habitats 2017 4 EUNIS Habitats 2021 5 Habitats Annex I 6 General habitats 7 IUCN Species 8 IUCN Habitats 9 Natura 2000 10 Corine Land Cover
Input data
RDF RDF XLS XLS XLS XLS RDF
CSV CSV, shapefile
RDF
The Linked Data frontend can be used to browse the projects and is accessible here9, whereas the SPARQL endpoint for a specific project, e.g. for ‘EUNIS Habitats 2012’ can be accessed here10. Moreover, the graph visualisation for all the projects is accessible using GraphViews application11. The aforementioned URIs ensure that the approach complies with FAIR.
We created explicit g e o n a m e s : n e a r b y relationships between Natura 2000 sites that in addition have relations to EUNIS species and ‘General habitats’ via the ontological relationships : s i t e H a s S p e c i e s and : s i t e H a s H a b i t a t resp. Moreover, the percentage coverage has been included to specify the percentage of habitat that the site contains using RDF*. We provide such a query in the following that combines n e a r b y relations and percentage of habitats in RDF*, which computes the TOP 5 largest habitats that are close to : A T 1 1 0 1 1 1 2 area12. PREFIX geonames : < h t t p s : / / www. geonames . o r g / o n t o l o g y # > PREFIX s i t e : < h t t p s : / / s e n s i n g c l u e s . p o o l p a r t y . b i z / S i t e O n t o l o g y / > PREFIX : < h t t p s : / / s e n s i n g c l u e s . p o o l p a r t y . b i z / N a t u r a 2 0 0 0 S i t e / > SELECT ? l a b e l (SUM ( ? p e r c e n t a g e ) a s ?sum ) ( g r o u p _ c o n c a t ( ? p e r c e n t a g e ) a s ? c n t ) WHERE { : AT1101112 geonames : n e a r b y ? s i t e s .
<<? s i t e s s i t e : s i t e H a s H a b i t a t ? l a b e l >> s i t e : p e r c e n t a g e C o v e r ? p e r c e n t a g e .
10https://sensingclues.poolparty.biz/PoolParty/sparql/Habitats 11https://sensingclues.poolparty.biz/GraphViews/ 12https://sensingclues.poolparty.biz/PoolParty/sparql/Natura2000Site } g r o u p by ? l a b e l o r d e r by d e s c ( ? sum ) l i m i t 5
Similarly, one can exploit : s i t e H a s S p e c i e s relations in order to build recommender systems
that can predict Ursus arctos movement in respect to sites, based on preferred habitats and
species. On top of this, one can use SPARQL query federation using SERVICE keyword in order
to query diferent SPARQL endpoints and join results based on common variables. This system
combined with a digital twin [
We have created a first version of Nature FIRST KG that can be used to power diferent use cases
that pertain biodiversity, addressing the problems reported in Table 1. We plan to enrich our
KG and ingest new sources that are related to site conservation, threats, treatment actions and
plans. Similarly, we are planning to add Ecological Networks [
We thank Boris Hinojo and Emil Zegers for their assistance and constructive feedback.