Supplementary Information - Spatial networks of habitats, populations, and communities: connecting approaches to keep cutting edges

Spatial networks of habitats, populations and communities: connecting approaches to keep cutting edges Supplementary Information Savary, Paul∗1 1Université Marie et Louis Pasteur, CNRS, ThéMA, 32 rue Mégevand F-25000 Besançon, France October 14, 2025 Methodological details Web Of Science query The full query used in June 2025 in the Web of Science was the following: ALL = (((habitat* graph*) OR (habitat* network*) OR (patch* graph*) OR (patch* network*) OR (patch-based graph*) OR (patch-based network*) OR (graph-based ind*) OR (graph-based metric*) OR (graph-based habitat) OR (landscape graph*) OR (landscape network*) OR (graphab) OR (conefor) OR (network* of habitat*) OR (network* of patch*) OR (graph* of habitat*) OR (graph* of patch*) OR (habitat spatial model*) OR (spatial* graph*) OR (spatial* network*) OR (population* graph*) OR (population* network*) OR (communit* graph*) OR (communit* network*) OR (network* of community*) OR (network* of population*) OR (graph* of community*) OR (graph* of population*) OR (metacommunit* graph*) OR (metacommunit* network*) OR (metapopulation* graph*) OR (metapopulation* network*) OR (dispersal* graph*) OR (dispersal* network*) OR (genetic graph*) OR (genetic network*)) AND (connectivity OR movement* OR dispers* OR coloniz* OR migration OR (gene flow) OR (range expan*) OR (range shift)) AND (ecolog* OR evolution* OR (landscape genetic*) OR (population genetic*) OR conservation OR biodiversity)) AND PY=(2000-2024) ∗Corresponding author: [email protected] 1 Results were limited to the following: •Document types: Article or Review Article or Book Chapters •Web of Science Categories: Ecology or Biodiversity Conservation or Environmental Sciences or Evolutionary Biology or Multidisciplinary Sciences or Environmental Studies or Biology or Zoology or Genetics Heredity or Marine Freshwater Biology or Geography or Urban Studies or Biochemistry Molecular Biology or Forestry or Regional Urban Planning or Mathematical Computational Biology or Entomology or Microbiology or Plant Sciences or Mathematics Interdisciplinary Applications or Remote Sensing or Computer Science Information Systems or Computer Science Interdisciplinary Applications or Fisheries or Ornithology or Limnology or Soil Science or Statistics Probability or Behavioral Sciences or Parasitology or Infectious Diseases Some publication sources were excluded from the results because their inclusion would have limited the comparability of their publications with the other included publications. Their thematic scope was either too different from the fields of ecology, evolution, and conservation (e.g., ESTUAR COAST SHELF S or PHARMACOGENOMICS J), too specialized (e.g., ANN ZOOL FENN), or they used the query terms in different contexts (e.g., works on genetic regulation networks published in GENOME BIOL). The excluded sources are the following: INT J ENVIRON RES, J ENVIRON PLANN MAN, AQUAT SCI, MAR ECOL PROG SER, SPAT STAT-NETH, ADV APPL PROBAB, INT J PLANT SCI, SYST BIOL, ARCH BIOL SCI, PACHYDERM, S AFR GEOGR J, STOCH PROC APPL, MED IMAGE ANAL, SYST ENTOMOL, GEOGR COMPASS, GISCI REMOTE SENS, INT J ENV RES PUB HE, PEST MANAG SCI, APPL NETW SCI, CHINESE GEOGR SCI, CATENA, CHAOS SOLITON FRACT, ECOL SOC, PEERJ COMPUT SCI, MATH BIOSCI, ENVIRON MONIT ASSESS, POPUL SPACE PLACE, GEOGR ANAL, THEOR ECOL-NETH, ECOL ENTOMOL, J BIOL DYNAM, INTEGR ZOOL, FRONT MICROBIOL, BANGL J PLANT TAXON, ECOL PROCESS, FOREST SYST, COMMUN NONLINEAR SCI, IEEE T MOBILE COMPUT, J LAND USE SCI, WATER RES, IEEE ACCESS, ECOL ECON, INT J MOL SCI, AQUACULT REP, LAND DEGRAD DEV, NORTH-WEST J ZOOL, J TRANSP GEOGR, WETLANDS, IEEE NETWORK, MSYSTEMS, BIOLOGIA, ISME J, EPJ DATA SCI, INT J SUST DEV WORLD, J MICROBIOL, BIOSYSTEMS, J HAZARD MATER, ITAL J ZOOL, J N AM BENTHOL SOC, FRONT ENV SCI-SWITZ, INFECT GENET EVOL, FEMS MICROBIOL ECOL, AFR J MAR SCI, ZOOKEYS, ENVIRON SCI POLICY, IMETA, INFORM SCIENCES, POL J ENVIRON STUD, FRONT CELL INFECT MI, COMPLEXITY, ENVIRON RES, FOOD WEBS, ZOOLOGIA-CURITIBA, APPL SOIL ECOL, TRANSBOUND EMERG DIS, J HERPETOL, ENVIRON GEOCHEM HLTH, FRESHW SCI, MAR ENVIRON RES, J ARID ENVIRON, HOUSING STUD, APPL ENVIRON MICROB, MATH COMPUT SIMULAT, NEUROBIOL LEARN MEM, IEEE T SYST MAN CY C, J BIOSCIENCES, ISPRS INT J GEO-INF, HELIYON, FRONT ENV SCI ENG, HABITAT INT, J PHYS-COMPLEXITY, STUD AVIAN BIOL, MATH MODEL NAT PHENO, BMC GENOMICS, IEEE ACM T COMPUT BI, BMC SYST BIOL, J EXP BOT, GENE DEV, EUR J ENTOMOL, GLOBAL FOREST FRAGMENTATION, ANNU REV BIOCHEM, PHARMACOGENOMICS J, BEHAV ECOL SOCIOBIOL, CABI INVASIVE SER, FRONT MAR SCI, NEW ZEAL J ECOL, ENVIRON SCI POLLUT R, J COMPUT BIOL, ECOSCIENCE, J BIOL CHEM, ANIM BIOL, FUZZY SET SYST, J GENET GENOMICS, METHODS MOL BIOL, ANN ZOOL FENN, BIOESSAYS, POLAR BIOL, REV BIOL TROP, PLANT ECOL, WIRES WATER, COMPUT COMMUN, J FISH BIOL, IFOREST, ENVIRON BIOL FISH, BIOINFORMATICS, ENVIRON REV, MICROORGANISMS, PLANT PHYSIOL, SYST APPL ACAROL-UK, FUNCT PLANT BIOL, ANNU PLANT REV, J ENVIRON DEV, J CELL BIOCHEM, EPIDEMICS-NETH, HERPETOL J, PLANT SCI, MOL CELL BIOL, HUM MOL GENET, FUNCT INTEGR GENOMIC, KSII T INTERNET INF, PLANT J, ESTUAR COAST SHELF S, J INTEGR PLANT BIOL, PROG BIOCHEM BIOPHYS, MICROB ECOL, ORNITOL NEOTROP, ECOSYST SERV, COMPUT-AIDED CIV INF, ANN BOT-LONDON, PARASITOL RES, MOL PLANT, INTEGR COMP BIOL, FEBS J, SEX TRANSM INFECT, B MATH BIOL, NORTHWEST SCI, J BIOMED INFORM, FRONT APPL MATH STAT, RIVER RES APPL, NAT CONSERVACAO, HEMOGLOBIN, ECOL ENG, INT J PRIMATOL, DEV SO AFR, FRONT PLANT SCI, MATRIX BIOL, COMPUT BIOL CHEM, NEW ZEAL J MAR FRESH, ACTA CHIROPTEROL, PLOS PATHOG, PATHOGENS, SIMUL-T SOC MOD SIM, GENOME BIOL EVOL, AUST J ZOOL, CHEM SENSES, CERNE, ENTOMOL FENNICA, INT J DATA MIN BIOIN, ISR J ECOL EVOL, GENOME BIOL, ACTA BIOTHEOR, WILEY SER COMPU QUAN, J RURAL STUD, LAND-BASEL, CITIES, REMOTE SENS-BASEL, ANIMALS-BASEL, PHYCOLOGYBASEL, SUSTAINABILITY-BASEL, FORESTS, PLANTS-BASEL, GENES-BASEL, DIVERSITY-BASEL, REMOTE 2 SENS ENVIRON, GENETIKA-BELGRADE, ACS EST WATER, MBIO, MOL PSYCHIATR, BOTHALIA, J INVERTEBR PATHOL Criteria defining the different types of spatial networks Spatial networks usually fall into the following categories, differing by the structural or biological nature of their nodes and links (Figure 1): Defined from structural landscape features: Habitat networks: Their nodes correspond to discrete habitat patches and their links to potential movement paths (e.g., dispersal or migration movements) (Urban & Keitt,2001). They are referred to as "habitat (patch) networks", "landscape graphs", or "patch-based graphs" (Galpern et al.,2011), among others. These networks are commonly used to assess habitat connectivity, compute and visualize corridors or connectivity metrics, prioritize conservation measures, or estimate impacts. Software programs used to build these networks include Conefor Sensinode (Saura & Torné,2009) and Graphab (Foltête et al.,2012). River networks: Their links correspond to river branches, making them dendritic networks (i.e., tree-like), whereas their nodes are most often river junctions or river segments. These nodes can be directly associated with populations or communities. These networks often serve as a basis to investigate metapopulation or metacommunity dynamics or the respective influence of community assembly processes in riverine ecosystems (Brown & Swan,2010;Campbell Grant et al.,2007;Fagan,2002). Defined from biological entities: Population genetic networks: Their nodes correspond to populations or individuals of the same species, while their links are estimated from their pairwise genetic differentiation or other proxies of gene flow (e.g., landscape distances, group assignment probabilities). They are used to describe population genetic structure and investigate landscape genetic relationships, as initially proposed by Dyer & Nason with the "popgraph" approach (Dyer & Nason,2004). Metapopulation networks: Their nodes are populations subject to extinction-colonization dynamics and their links describe the dispersal connections among such populations. These networks are a graph-based representation of spatially-explicit metapopulations, as conceptualized by the seminal work of Ilka Hanski and collaborators (Hanski,1994,1999). They can serve as a basis to (i) estimate the parameters of metapopulation models (e.g., incidence, colonization, or extinction local functions, metapopulation extinction time or capacity) or (ii) to simulate metapopulation dynamics in varying network structures. Metacommunity networks: Their nodes are communities whereas their links represent potential dispersal paths giving rise to metacommunity dynamics, i.e., a combination of species sorting, biotic interactions, and dispersal movements shaping diversity patterns at landscape scale. The works of Economo & Keitt (Economo & Keitt,2008;Economo,2011) set the ground for the use of this approach; see (Savary et al.,2024) for a recent synthesis. Meta-networks: These networks are made from a combination of at least one spatial network and another network, spatial or not, linked in an explicit way to the former. They include networks in which a species interaction network (e.g., food web, pollination network) is embedded within each spatial node, spatial networks connected to other networks representing human infrastructures or discrete time steps, spatial networks having different types of nodes and links (e.g., habitat or movement types) or connections to aspatial entities (e.g., species) (Figure 5). Landscape data and simulation details For both types of simulations, species populations occupied 610 forest patches larger than 5 hectares identified in the same study area considered for making Figure 1 (43°40’N; 0°43’W, 500 km2). The land cover data used for making the map (Figure 1A) are public data provided by the French Institut national de l’information géographique et forestière as part of the land cover database OCS GE and the cartographic database BD TOPO (hydrographic networks and vegetation types). In the simple proofof-concept simulations made for Box 2, I accounted for landscape structure to compute pairwise costdistances among forest patches in the same way as commonly done when building habitat networks. I used the Graphab software (Foltête et al.,2021). I created a raster land cover map of 20 m spatial 3 resolution including 11 land cover types. I assigned a cost value to each land cover type, reflecting common resistance schemes used for woody habitat species (see Table S1). Table S1: Cost values assigned to each land cover type to compute pairwise cost-distances among the 610 forest patches of the study area Land cover type Cost Broadleaf forests 1 Coniferous forests 1 Mixed forests 1 Bushland 50 Other woody vegetation 50 Low vegetation 100 Barren land 200 Non-built gravel road 500 Non-built impervious land 1000 Water bodies 1000 Built-up areas 10000 The pairwise cost-distance matrix obtained was used to compute dispersal probabilities during the simulations, and for creating the habitat network displayed on Figure 1D based on a minimum planar graph topology. A metric distance of 5 km corresponded approximately to 1,200 cost-distance units, according to a log-log linear regression of cost-distances against metric distances. The node-level metric "Flux" was computed on this graph, using the following formula: Fi= n X j=1 aj×exp−α×dij with n= 610 the number of patches in the network, ajthe area of patch j, and αthe parameter setting the exponential decay of dispersal probabilities for an increasing distance dij between patches iand j. The value of αwas set such that pij = 0.01 or 0.05 for a cost-distance dij equal to 1,200 or 2,400 cost-distance units, resulting in four series of values for these different parameters. Besides, the Betweenness Centrality metric was computed to describe the role of habitat patches for the "traversability" of the network. This property is quantified by this metric as the number of times a patch is crossed by the shortest paths between every pair of two patches in the networks, using the following formula: BCi=PjPkajake−α×djk j, k ∈ {1, . . . , n}, k < j, i ∈Pjk (1) with Pjk the pairs of patches to consider, that is, all patch pairs jk (i! = j, i! = k) connected by a shortest path across the network stepping through patch i. The BC metric was computed with the same values of the αparameter as the Flux metric. Population genetic simulation The population genetic simulations reproduced the effect of dispersal-driven gene flow and genetic drift (i.e., a random loss of genetic diversity inversely proportional to the size of a population) in populations occupying 610 habitat patches. The populations included 18,300 individuals with at last 10 individuals per population (i.e. an average of 30 individuals per population). The number of individuals assigned to each population was proportional to the logarithm of their total area, so that large patches included more individuals than small ones. The initial individuals started the simulation with a randomly assembled diploid genotype counting 20 loci and 20 different possible alleles per locus. Mutations could then happen at a rate µ= 5 ×10−4. The sex-ratio was equal to 1 and every female individual could have an average of 3 offspring per generation, according to a Poisson distribution. After their birth, the individuals either remained in the natal patch or disperse. They could live up to 5 generations. The dispersing individuals could disperse from patch ito another patch jwith a probability pij decreasing exponentially with the cost-distance dij, so that pij =exp−α×dij . I used 6 4 dispersal intensities by setting the proportion of dispersers pto 0.001, 0.05, and 0.5, and simultaneous setting αvalues so that the probability pij of covering a distance equivalent to 2,400 cost-distance units was also equal to 0.001, 0.005, 0.01, 0.05, 0.1, or 0.5. Beyond the set number of individuals per population, supernumerary individuals died after dispersal, keeping local population sizes constant over time. At the next generation, individuals could breed with individuals within their patch. I simulated this dynamics for 250 generations, reaching a stable state. At the end of each simulation, I computed population allelic richness as the average number of alleles per locus. The codes were adapted from the simulation functions of the PopGenReport package (Adamack & Gruber,2014) and are available online at: https://gitlab.com/psavary3/SpatialHabNetworksReview Metacommunity simulation The metacommunity simulations aimed at reproducing patterns of species diversity shaped by the action of neutral processes alone (i.e., dispersal and stochastic extinction, without niche differentiation or heterogeneous competition among individuals from different species) or the joint action of dispersal, stochastic extinction, species sorting, and heterogeneous competition among individuals of the same or different species (stabilizing competition). For that purpose, I used the simulation framework of Khattar et al. (2024), based on generalized Lotka-Volterra competition and a discrete Beverton-Holt population growth model. All 610 habitat patches included a community made of at most 30 species. Initially, each species could randomly colonize a patch with a probability equal to 0.5. Then, the local performance of a species was computed as a function of the distance of the species niche optimum to the local conditions, and of the local competitive interactions. In neutral simulations, every patch had an environmental value corresponding to species optimum and every individual, regardless of its species, competed with other individuals according to a competition coefficient αequal to 0.045. The maximum growth rate Kwas equal to 2. These parameters made it possible to compute the local abundance of every species in each patch based on its local performance. For more details about the population growth model, see Khattar et al. (2024). A random draw of individuals according to a Poisson distribution having for average the expected abundance included some stochasticity in the simulation and made local extinctions possible. After this recruitment step, dispersal could happen among communities according to the set proportion of dispersers and to their ability to cover a given cost-distance. The same dispersal parameters as for the genetic simulation were used (p=0.001, 0.005, 0.01, 0.05, 0.1, or 0.5). After 50 generations, I computed the local species richness in every community. When species sorting and heterogeneous competition were included, species niche optima were randomly drawn from a uniform distribution ranging from 0.2 to 0.8. The niche breadth of all species was equal to 0.1. Local environmental conditions at the patch level were computed as the proportion of open areas in a radius of 2 km around the forest patches, ranging from about 0.3 to 0.8. This distinguished the virtual species according to a gradient of forest/open area specialization. Stabilizing competition was implemented with a competition matrix with intra-specific coefficients αii equal to 0.006 and inter-specific coefficients αij equal to 0.003, thereby favoring local coexistence. I considered 6 scenarios, combining 3 dispersal rates and probabilities and 2 settings (neutral vs with species sorting and heterogeneous competition). Assessment of the relationship between diversity and graph-based metrics At the end of all simulations (population genetics and metacommunities), I computed the Pearson correlation coefficient between the diversity response (allelic richness or species richness) and the following metrics computed at the node level: area (i.e., node size), Flux, and Betweenness Centrality (for all αvalues). The results are presented as heatmaps (see Figure 6). 5 Supplementary tables Table S2: Total number, and number and proportion of studies adopting an experimental approach for each research branch using a distinct type of spatial network. Network type Nb. papers Nb. experimental % experimental Habitat network 313 4 1.3 (Meta)population network 128 7 5.5 Meta-network 82 2 2.4 Population genetic network 60 0 0 River networks 34 7 20.6 (Meta)community network 33 5 15.2 Dispersal network 21 0 0 Spatial network 6 0 0 Supplementary figures Identification of studies via Web of Science IdentificationScreeningInclusion Records identified from the Web of Science: n = 1274 Records removed before screening: Records other than articles or book chapt.: n = 14 Records from journals outside the relevant field: n = 269 Records screened: n = 991 Records excluded: (missing DOI, discontinued journal) n = 2 Records excluded: No use of spatial networks nor in the relevant field: n = 283 Exclusion at second screening: n = 27 Records assessed for eligibility: n = 989 Records included in review: n = 679 Figure S1: PRISMA flowchart showing the identification, screening, and inclusion steps of the bibliographic query on the Web of Science, following Page et al. (2021). 6 Movement Ecology J. of Theoretical Biology The American Naturalist Molecular Ecology Oikos Ecology Letters Ecography PNAS Ecology & Evolution Ecology Freshwater Biology Ecological Modelling Procs.Roy.Soc.B Ecological Applications PLoS ONE Scientific Reports J. of Animal Ecology Journal of Applied Ecology Methods in Ecology & Evolution Biological Conservation Sci. of the Total Environment Diversity and Distributions Conservation Biology Landscape Ecology J. of Nature Conservation J. of Environmental Management Ecological Indicators Landscape and Urban Planning Biodiversity and Conservation Animal Conservation Ecosphere Habitat network 2000 - 2014 2015 - 2025 Metapopulation network Meta-network Population genetic network River network Metacommunity network Movement Ecology J. of Theoretical Biology The American Naturalist Molecular Ecology Oikos Ecology Letters Ecography PNAS Ecology & Evolution Ecology Freshwater Biology Ecological Modelling Procs.Roy.Soc.B Ecological Applications PLoS ONE Scientific Reports J. of Animal Ecology Journal of Applied Ecology Methods in Ecology & Evolution Biological Conservation Sci. of the Total Environment Diversity and Distributions Conservation Biology Landscape Ecology J. of Nature Conservation J. of Environmental Management Ecological Indicators Landscape and Urban Planning Biodiversity and Conservation Animal Conservation Ecosphere AB Figure S2: Bipartite networks showing the proportion of papers about each type of spatial network published in the journals having published at least 6 papers included in the final corpus between (A) 2000 and 2014 or (B) 2015 and 2025. Journals are sorted in the same order as on Figure 3. Papers about "spatial networks" (generic term), "dispersal networks" and reviews are excluded. Urban et Keitt, 2001. Ecology Saura & Pascual−Hortal, 2007, Landscape and Urban Planning Urban et al., 2009, Ecology Letters Taylor et al., 1993, Oikos Saura & Torné, 2009, Environmental Modelling & Software Galpern et al., 2011, Biological Conservation McRae et al., 2008, Ecology Calabrese & Fagan, 2004, Front. Ecology Environ. Adriaensen et al. 2003. Landscape and Urban Planning Rayfield et al. 2011. Ecology Hanski, 1998. Nature Hanski & Ovaskainen, 2000. Nature Fahrig, 2003. Annual Reviews Ecol. Evol. System. Hanski, 1994. Journal of Animal Ecology Dale & Fortin, 2010. Annual Reviews Ecol. Evol. System. Levins, 1969. Bulletin of the Entomological Society of America Hanski, 1999. Metapopulation Ecology Leibold et al., 2004. Ecology Letters Foltête et al., 2012. Environmental Modelling & Software Dyer & Nason, 2004. Molecular Ecology Csardi & Nepusz, 2006. InterJournal Garroway et al., 2008. Evolutionary Applications Carrara et al., 2012. PNAS Fahrig, 2013. Journal of Biogeography Campbell Grant et al., 2007. Ecology Letters Hubbell, 2001. The Unified Neutral Theory of Biodiv. and Biogeo. Fortuna et al., 2009. PNAS Economo & Keitt, 2010. Oikos Altermatt, 2013. Aquatic Ecology Economo & Keitt, 2008. Ecology Letters MacArthur & Wilson, 1967. The Theory of Island Biogeography Guimerà & Nunes Amaral, 2005. Nature Manel et al., 2003. Trends in Ecology & Evolution Mouquet & Loreau, 2003. The American Naturalist Brown & Swan, 2010. Journal of Animal Ecology Chesson, 2000. Annual Reviews Ecol. Evol. System. Dunning et al., 1992. Oikos Habitat network (Meta)population network Population genetic network (Meta)community network River networks Meta−network Dispersal network Spatial network 0 1 2 −1 0 1 2 Dim. 1 (32.24 %) Dim.2 (26.4 %) Figure S3: Results of the Correspondence Analysis of the contingency table of the number of citations of the references listed in Table 4 and below in each research branch using a distinct type of spatial network. The name of the network type is displayed at the centroid of the coordinates of the references cited in papers using these networks. 7 Papers from 2015-2024 Papers from 2000-2014 citing Habitat network % citations Metapopulation network Meta-network Population genetic network Metacommunity network River network 2 residuals % citations 2 residuals 93.6 3.6 0.4 0.5 0.5 1.4 5.4 -4.7 -3.3 -7.7 -2.9 -3.2 39.2 46.4 3.2 1.6 3.2 6.4 -5.0 15.5 1.1 -2.3 2.0 1.3 50.0 13.0 8.7 0.0 28.3 0.0 -2.2 1.3 3.4 -1.8 13.8 -1.2 71.1 7.2 12.4 5.2 2.1 2.1 -0.8 -0.2 7.8 -0.7 0.3 -0.7 37.2 7.0 0.0 2.3 2.3 51.2 -3.1 -0.2 -0.9 -1.2 0.3 17.4 40.4 3.0 1.8 53.6 0.0 1.2 -5.6 -2.2 -0.0 22.9 -1.7 -1.5 Figure S4: Bipartite networks in which a paper published between 2015 and 2024 (higher nodes) is connected to the papers published between 2000 and 2014 (lower nodes) that are cited within it. The node colors of both layers correspond to the type of spatial network studied in the papers. The links are colored according to the type of network considered in the paper citing the reference. The central panels include all connections whereas the side panels focus on each type of spatial network. The proportion of papers about each spatial network that is cited within the papers about the focal spatial network is indicated on top of the latter panels, as well as the residuals of the χ2test performed on the corresponding contingency table. This provides a preferential citation index independent of the imbalance of paper frequencies. Positive χ2indicate a preference and are displayed in bold. Only the papers included in the final corpus are considered here. Papers about "spatial networks" (generic term), "dispersal networks" and reviews are excluded. Papers from 2015-2024 Papers from 2000-2014 citing Habitat network Metapopulation network Meta-network Population genetic network Metacommunity network River network Figure S5: Bipartite network showing the proportion of papers published between 2015 and 2024 (higher nodes) about each type of spatial network citing papers published between 2000 and 2014 (lower nodes) about each type of spatial network. The node colors of both layers correspond to the type of spatial network. The links are colored according to the type of network considered in the paper citing the reference. Only the papers included in the final corpus are considered here. Papers about "spatial networks" (generic term), "dispersal networks" and reviews are excluded. 8 Species richness Allelic richness Flux metric (log-scale) Flux metric (log-scale) 5 10 15 0 10 20 30 10 12 14 10 12 14 10 12 14 p = 0.001 p = 0.05 p = 0.5 10 12 14 10 12 14 10 12 14 p = 0.001 p = 0.05 p = 0.5 Neutral dynamics Neutral dynamics Species sorting, competition Figure S6: Relationships between the Flux metric computed at the node-level for each forest patch displayed on Figure 1 (in log-scale) and either the allelic richnness of populations (top panels) or species richness of communities (bottom panels) simulated in these patches under the action of different ecological or evolutionary processes. Vertical panels distinguish simulations performed with varying dispersal rates and probabilities (p= 0.001, 0.05 or 0.5). Neutral population genetic simulations (black trends) included the effect of genetic drift and dispersal-driven gene flow. Neutral metacommunity simulations (orange trends) included the effect of dispersal, stochastic extinction (i.e., ecological drift), and equal intraand inter-specific competition. The non-neutral metacommunity simulations (blue trends) included species with different niches subject to species sorting, stronger intrathan inter-specific competition (stabilizing), as well as dispersal and stochastic extinction. Each trend summarizes the linear relationship between the Flux metric and the diversity response for n= 610 patches. See simulation details in the sections above. 9