Ecological Cartography: Constructing Plant Species Reference Landscapes From Co‐Occurrence Networks
Flemming SkovABSTRACT
Aims
This study presents EcoNetMap, a network‐based framework that uses species co‐occurrence alone to build a fixed reference landscape with a common coordinate system in which species, plots, habitat types and datasets can be located and compared. It functions analogously to a geographic map: Distances encode ecological similarity, and directions correspond to environmental gradients.
Location
Demonstrated using Danish data, but the framework is applicable to any vegetation monitoring dataset.
Methods
Applied to Denmark's NOVANA dataset (43 terrestrial habitat types, > 140,000 plots), 12 candidate reference maps were constructed from stratified and random subsamples of the data using two layout algorithms (Fruchterman–Reingold and classical multidimensional scaling) and six sampling strategies. Each candidate was evaluated against four complementary evaluation metrics on independent data, and a weighted composite score was used to select the optimal configuration. In the selected map, species occurring in at least 10 plots formed the core network, whereas sub‐threshold species were positioned on the basis of their associations with core taxa. The resulting reference landscape served as a fixed coordinate system for analysing the complete dataset.
Results
Environmental gradients in light, moisture, nitrogen and soil reaction emerged from the co‐occurrence data alone, with no environmental variables used in construction (mean adjusted R 2 = 0.686, p < 0.001), and life forms, community clusters, forest shade categories and habitat types all occupied coherent, ecologically interpretable regions of the landscape. The reference network placed 1019 species in a single frame: a densely connected core of 819 species (> 125,000 edges) with moderate modularity (Q = 0.143), consistent with the continuous, gradient‐like nature of Danish vegetation rather than sharply bounded communities. The same frame resolved ecological structure across scales, from individual species through habitat types to whole dataset and biogeographic profiles. Of 12 candidate maps, the Fruchterman–Reingold layout on a stratified 750‐plot‐per‐habitat sample gave the best composite score (0.496); a classical multidimensional scaling comparison confirmed the layout choice rather than the sampling was decisive.
Conclusion
The EcoNetMap framework contributes to vegetation analysis by translating complex co‐occurrence patterns into a visual reference landscape that reveals ecological structure difficult to detect with traditional methods, including community modularity, core‐periphery specialisation and continuous gradients within and between habitat types. Requiring only standard species occurrence data, it works alongside existing ordination and classification approaches and integrates readily with established monitoring programmes, offering a scalable tool for comparative vegetation analysis and conservation assessment, and a stable frame against which compositional change can be tracked over time.