Philippa Kaur

These authors contributed equally: Yannick Garcin, Enno Schefuß, Greta C. Dargie, Simon L. LewisAix Marseille University, CNRS, IRD, INRAE, CEREGE, Aix-en-Provence, FranceYannick Garcin & Ghislain GassierInstitute of Geosciences, University of Potsdam, Potsdam, GermanyYannick GarcinMARUM—Center for Marine Environmental Sciences, University of Bremen, Bremen, GermanyEnno SchefußSchool of Geography, University of Leeds, Leeds, UKGreta C. Dargie, Bart Crezee, Dylan M. Young, Andy J. Baird, Paul J. Morris & Simon L. LewisSchool of Geography and Sustainable Development, University of St Andrews, St Andrews, UKDonna Hawthorne, Ian T. Lawson & George E. BiddulphIFP Energies Nouvelles, Earth Sciences and Environmental Technologies Division, Rueil-Malmaison, FranceDavid SebagInstitute of Earth Surface Dynamics, Geopolis, University of Lausanne, Lausanne, SwitzerlandDavid SebagFaculté des Sciences et Techniques, Université Marien Ngouabi, Brazzaville, Republic of the CongoYannick E. Bocko & Y. Emmanuel Mampouya WeninaÉcole Normale Supérieure, Université Marien Ngouabi, Brazzaville, Republic of the CongoSuspense A. IfoÉcole Normale Supérieure d’Agronomie et de Foresterie, Université Marien Ngouabi, Brazzaville, Republic of the CongoMackline MbembaFaculté de Gestion des Ressources Naturelles Renouvelables, Université de Kisangani, Kisangani, Democratic Republic of the CongoCorneille E. N. Ewango & Joseph Kanyama TabuFaculté des Sciences, Université de Kisangani, Kisangani, Democratic Republic of the CongoCorneille E. N. EwangoInstitut Supérieur Pédagogique de Mbandaka, Mbandaka, Democratic Republic of the CongoOvide Emba & Pierre BolaSchool of Geography, Geology and the Environment, University of Leicester, Leicester, UKGenevieve Tyrrell, Arnoud Boom & Susan E. PageSchool of Water, Energy and Environment, Cranfield University, Bedford, UKNicholas T. GirkinBritish Geological Survey, Centre for Environmental Geochemistry, Keyworth, UKChristopher H. VaneInstitute of Earth Sciences, University of Lausanne, Lausanne, SwitzerlandThierry AdatteNEIF Radiocarbon Laboratory, Scottish Universities Environmental Research Centre (SUERC), Glasgow, UKPauline GulliverSchool of Biosciences, University of Nottingham, Nottingham, UKSofie SjögerstenDepartment of Geography, University College London, London, UKSimon L. Lewis More

Automated tracking of four social insect speciesFifty queenright colonies were used in the tracking experiments (Table 1). Honeybee colonies (subspecies A. mellifera carnica) were housed in the campus apiary of the University of Lausanne. Colonies of L. niger were raised from single mated queens collected on campus. T. nylanderi colonies were collected from the University of Lausanne campus, and L. acervorum colonies collected from Anzeindaz, Switzerland. These four species were chosen because of their abundance and easy availability in Switzerland, and because they – or closely-related species – have previously been used as model systems for the study of spatial organisation in social insects20,21,23,27,44. The colony sizes used in our experiments (Table 1) fell within the natural range of sizes experienced by these species in nature, either as recently founded colonies (L. niger colonies are founded by a single queen and progressively grow from a few workers to mature sizes of up to 40,000 workers over the course of several years; new honeybee colonies are founded by swarms counting 2400–41,000 bees61) or as mature colonies (all colonies of T. nylanderi and L. acervorum used in our experiments were mature colonies collected whole from the field).In all species, a paper tag bearing a unique two-dimensional barcode was glued to the thorax of individuals to allow automated tracking of their movements (Fig. S1). In the ants, tagging of all individuals was performed in a single session two days before the beginning of the experiment, whilst in the bees, newly-emerged workers (one-day-old or less) were tagged every 3 days over the 21 days prior to the beginning of the experiment (Supplementary Note 1).Tagged colonies were kept in glass observation nests with a single entrance (internal nest dimensions, A. mellifera: 69 × 45 × 4 cm, L. niger: 70 × 40 × 8 mm; L. acervorum: 63 × 42 × 2 mm, T. nylanderi: 63 × 42 × 1.5 mm). The honeybee observation nests also included a 64 × 44 cm wooden frame enclosing a double-sided wax comb containing honey, pollen, and developing brood20. Bees were free to move between both sides of the comb. In all species, individuals were allowed to freely exit and enter the nest. Ants were provided with ad libitum food (Drosophila, sugar solution) and water in a foraging arena, while bees foraged on natural resources outside. Both the ant and honeybee observation nests were exposed to diurnal cycles of temperature and light (Supplementary Note 1).High resolution digital video cameras operating at two frames per second were used to identify the location and orientation of each tag across successive images22. All colonies were continuously tracked for three days, which corresponded to the inter-cohort time in the honeybee colonies. The trajectories of each worker, and the physical contacts between workers (Fig. S18 and Supplementary Note 14) were extracted using an existing software pipeline62.Building bipartite site-visit networksTo quantify the spatial preferences of individual ants and bees, the interior of the nest was discretised into a regular hexagonal lattice (Fig. 1a, b). Because the worker body lengths of our four study species span an order of magnitude (from ~ 1.5 mm for T. nylanderi to ~ 15 mm for A. mellifera), the width of the hexagonal bins were defined as 1/4 of the mean worker body-length.To characterise the spatial preferences of different individuals to different parts of the nest, we counted the number of times ({n}_{i}^{s}) that each individual i visited each hexagonal site s. A visit by individual i to site s began when i crossed the border into s, and was terminated when i crossed the border out of s, regardless of the amount of time spent inside. To prevent stationary individuals located on the border between two adjacent sites from rapidly accumulating many single-frame visits to the two sites, successive visits to a same site were only counted when at least 20s elapsed between the end of the previous visit and the start of the next.The site-visit data were used to construct a bipartite network, in which individuals (layer 1) were connected by undirected edges to the sites (layer 2) they visited (Fig. 1c, d). Because individuals typically made multiple visits to the same sites, each edge i–s was weighted according to the total number of times individual i visited site s, that is, ({n}_{i}^{s}).Partitioning site-visit networks into modulesThe extent to which the site-visit networks were partitioned into discrete ‘modules’ (i.e., set of workers with similar space-use patterns and the set of sites that they exhibit strong ties to) was assessed using the DIRTLPAwb+ algorithm for partitioning weighted bipartite networks39. This algorithm searches for the partition that maximises the number and strength of the links within modules, whilst minimizing connections between modules. The number of modules was not specified a priori by the user, but was identified by the algorithm. All site-visit networks had positive modularity (Fig. S3), indicating that they could be partitioned into a set of well-separated modules (Figs. 1e–h, S2, and S4–S5). The modules in each partition were then assigned functional labels according to the following rules. First, the module whose sites were on average closest to the nest entrance was labelled ‘forager’ module. Second, the module or modules with the greatest spatial overlap with the brood pile in the ant colonies or the broodnest(s) in the honeybee colonies were labelled ‘nurse’ module(s). After defining the forager and nurse modules, the remaining modules (if any) were labelled as follows. If there was only one module remaining after identifying the nurse and forager modules, as was typically the case in honeybee colonies, it was labelled ‘peripheral’. If there were two modules remaining, as was typically the case in ant colonies, then the module whose sites were on average closer to the nest borders (i.e., to the periphery of the nest) was labelled ‘peripheral’, and the remaining module labelled ‘intermediate’. In some cases, the DIRTLPAwb+ algorithm identified five or more modules (9.0% of all iterations across all species and colonies). In these cases, the supernumerary modules never contained more than 1 or 2 individuals, and as they could not be unequivocally assigned using our labelling scheme, they were left unclassified for these iterations.Validating network modulesAs a network constructed by a purely random process could exhibit apparent modular structure by chance, we tested whether the discovered modules represent statistically significant entities. To do so, we produced 1000 null model random networks for each observed network using an established permutation method for bipartite networks63 (Supplementary Note 2). Comparisons between the maximum modularity of the observed networks with that of the corresponding random networks showed that, in all four species, the observed modularity was significantly greater than expected by chance (Fig. S3).Constructing worker task profilesA unique labour profile for each ant and each honeybee was constructed by estimating the activity of each worker in the following four tasks:1. Entrance guarding: workers were classed as guarding when they were (i) within two body lengths of the entrance, (ii) roughly facing the entrance, i.e., with a body alignment diverging from the direct heading to the entrance by no more than π/2 radians, and (iii) ‘on station’ at the entrance, as defined by trajectory coordinates with an associated first passage time (ref. 64; time taken for the individual to pass beyond a circle centred on its current location with a radius of two body-lengths) in excess of 500s.2. Patrolling: workers were classed as patrolling65 when they were (i) active, and (ii) ‘roaming’, as defined by first passage times of More

Survey sites and data collectionBenthic and fish surveys were conducted at randomly stratified sites throughout the entirety of the FRT by NOAA’s National Coral Reef Monitoring Program (NCRMP). Sites were categorized into three biogeographic regions, including Dry Tortugas (DRTO, n = 228), Florida Keys (FLKs, n = 322), and Southeast Florida (SEFL, n = 173) (Fig. 1). The Florida Keys were further classified into the following four sub-regions: Lower Keys (LK, n = 103), Middle Keys (MK, n = 46), Upper Keys (UK, n = 140), Biscayne (BISC, n = 33). Within each region/sub-region (except for SEFL), reefs were categorized according to reef types. For DRTO, this included bank, forereef, and lagoon reef sites. For the LK, MK, UK, and BISC, reef types were categorized as inshore, mid-channel, and offshore. Data were collected throughout the region in 2014, 2016, and 2018.Fish and benthic surveys were conducted in accordance with NCRMP methodologies34 (Table S2). The protocol used for the fish surveys was developed from a modified Reef Visual Census (RVC) method35 and was performed using a stratified random sampling design. Divers surveyed two 15 m diameter cylinders, spaced 15 m apart. Fish species were identified to the lowest taxonomic level for a period of five minutes. This was followed by an additional five minutes dedicated to recording species abundances and sizes (10 cm bins).Surveys were used to quantify the benthic cover at each site. The protocol for these surveys followed a standard line point-intercept sampling design. At each site, a 15 m weighted transect was draped along the reef surface. Surveyors recorded benthic composition at 15 cm intervals along the transect (i.e., 100 equidistant points). The benthic composition from these 100 points was then transformed to percent cover of ecologically important functional groups (scleractinian coral [species-specific], gorgonians, hydrocoral, CCA, macroalgae, turf algae, sponges, bare/dead substrate, sand/sediment).Carbonate budget analysisPlanar benthic surveys were adjusted to account for the three-dimensional complexity (i.e., rugosity) of each site using light detection and ranging (LiDAR) data (1 m horizontal resolution; 15 cm vertical resolution) from topobathymetric mapping surveys of the South Florida eastern coastline conducted by NOAA’s National Geodetic Survey. A 15 m x 15 m region of interest (ROI) was placed around the GPS coordinates of each site using ArcGIS Pro with 3D and Spatial Analyst extensions (ESRI). The ROI was then overlaid with existing multibeam echosounder (MBES) and LiDAR bathymetry data. Within the ROI, LiDAR was extracted using the Clip Raster function from ArcPy (ArcGIS’s python coding interface), and the Surface Volume tool was used to calculate the 3D surface area. Rugosity was calculated by dividing the 3D surface area by the 2D surface area of the ROI.The methodology for standardizing reef carbonate budgets to topographic complexity (i.e., rugosity) diverged from that of the ReefBudget approach by using site-specific rugosity rather than species-specific rugosity17. This was a necessary limitation of this analysis as transect rugosity at 1 m increments was not measured using the NCRMP benthic survey protocol. To ensure that reef topographic complexity was still accounted for, however, rugosity of the entire reef site, calculated from LiDAR bathymetry data, was used in this analysis. While rugosity of the site rather than of each benthic component, specifically for corals, can lead to an under or overestimation of carbonate production rates, we note that site and species rugosity (i.e., encrusting and massive coral morphologies) was low for the vast majority of sites and species surveyed, thereby reducing the probability of an under or overestimation.Reef carbonate budget analysis was performed following a modified version of the ReefBudget approach17. Coral carbonate production was derived from species-specific linear extension rates (cm year−1), skeletal density (g cm−3), coral morphology (branching, massive, sub-massive, encrusting/plating), and percent cover. Carbonate production by CCA and other calcareous encrusters was similarly calculated as a function of surface area, literature reported linear extension rates, and skeletal density17. Gross carbonate production at each survey site was measured as the sum total of carbonate production by all calcareous organisms found at each site and was standardized to site-specific reef rugosity.Gross carbonate erosion for each survey-site was calculated as the sum total of erosion by four bioeroding groups: parrotfish, microborers, macroborers, and urchins. The calculations roughly followed the ReefBudget methodologies17 (Table 1). Parrotfish size frequency distributions from NCRMP surveys were multiplied by size and species-specific bite rates (bites min−1), volume removed per bite (cm3), and proportion of bites leaving scars to calculate total parrotfish erosion17. The substrate density (1.72 g cm−3) used in these calculations followed that of the ReefBudget protocol17. Microbioerosion was calculated from the percent cover of dead coral substrate, which was multiplied by a literature-derived rate17 of − 0.240 kg CaCO3 m−2 year−1. Macroboring was calculated as the percent cover of clionid sponges multiplied by the average erosion rate of all Caribbean/Atlantic clionid sponges17 (-6.05 kg CaCO3 m−2 year−1). External bioerosion by urchins was calculated using Diadema urchin abundance collected from the benthic surveys. Due to the lack of test size data from the NCRMP benthic surveys, urchin abundance was multiplied by the bioerosion rate of an average test sized36 (66 mm) Caribbean/Atlantic Diadema urchin (-0.003 kg CaCO3 m-2 year−1). While using an average test sized Diadema urchin for this analysis may have led to an under or overestimation of urchin erosion, the abundance of Diadema urchins measured in the surveys was minimal, as they appeared to be functionally irrelevant across the FRT.Model validationAs the survey methodologies and data sources employed in this analysis were modified from that of the standard ReefBudget approach17, we chose to validate our model through a fine scale temporal comparison of annual ReefBudget surveys conducted by NOAA at Cheeca Rocks (UK) to three nearby NCRMP sites used in our analysis. Since the NCRMP surveys were performed in 2014, 2016, and 2018, this study focused exclusively on these three survey years from the NOAA Cheeca Rocks dataset. Temporal trends related to reef growth/erosion were visually compared to see if survey types provided comparable results (SI Figure S6).Statistical analysisAll model calculations and statistical analyses were performed using R37 with the R Studio extension38. Generalized linear models (GLMs) were run on response variables involved in habitat production (i.e., net carbonate production, gross carbonate production, and gross carbonate erosion) to evaluate spatial trends related to reef development across sub-regions and reef types. Each GLM was performed with reef type being nested within sub-region. The best fit distribution for each variable was determined using the fitdistrplus R package39. Linear regression analysis was used to evaluate the relationship between net carbonate production and both live coral cover and parrotfish biomass. All plots were created using ggplot2 R package40 and edited for style with Adobe Illustrator41. More

Sket, B. Can we agree on an ecological classification of subterranean animals?. J. Nat. Hist. 42, 1549–1563. https://doi.org/10.1080/00222930801995762 (2008).Article 

Google Scholar 
Mammola, S. et al. Scientists’ warning on the conservation of subterranean ecosystems. Bioscience 69, 641–650. https://doi.org/10.1093/biosci/biz064 (2019).Article 

Google Scholar 
Boulton, A. J., Fenwick, G. D., Hancock, P. J. & Harvey, M. S. Biodiversity, functional roles and ecosystem services of groundwater invertebrates. Invertebr. Syst. 22, 103–116. https://doi.org/10.1071/IS07024 (2008).Article 

Google Scholar 
Danielopol, D. L. & Griebler, C. Changing paradigms in groundwater ecology—From the ‘living fossils’ tradition to the ‘new groundwater ecology’. Int. Rev. Hydrobiol. 93, 565–577. https://doi.org/10.1002/iroh.200711045 (2008).Article 

Google Scholar 
Griebler, C., Malard, F. & Lefébure, T. Current developments in groundwater ecology—From biodiversity to ecosystem function and services. Curr. Opin. Biotechnol. 27, 159–167. https://doi.org/10.1016/j.copbio.2014.01.018 (2014).Article 
PubMed 

Google Scholar 
Fišer, C. Niphargus—A model system for evolution and ecology. In Encyclopedia of Caves (eds Culver, D. C. et al.) 746–755 (Academic Press, 2019).Chapter 

Google Scholar 
Riddle, M. R. et al. Insulin resistance in cavefish as an adaptation to a nutrient-limited environment. Nature 555, 647–651. https://doi.org/10.1038/nature26136 (2018).Article 
ADS 
PubMed 
PubMed Central 

Google Scholar 
Gibert, J. et al. Assessing and conserving groundwater biodiversity: Synthesis and perspectives. Freshw. Biol. 54, 930–941. https://doi.org/10.1111/j.1365-2427.2009.02201.x (2009).Article 

Google Scholar 
Trontelj, P. et al. A molecular test for cryptic diversity in ground water: How large are the ranges of macro stygobionts?. Freshw. Biol. 54, 727–744. https://doi.org/10.1111/j.1365-2427.2007.01877.x (2009).Article 

Google Scholar 
Cooper, J. E. Ecological and Behavioral Studies in Shelta Cave, Alabama, with Emphasis on Decapod Crustaceans (University of Kentucky, 1975).
Google Scholar 
Voituron, Y., de Fraipont, M., Issartel, J., Guillaume, O. & Clobert, J. Extreme lifespan of the human fish (Proteus anguinus): A challenge for ageing mechanisms. Biol. Lett. 7, 105–107. https://doi.org/10.1098/rsbl.2010.0539 (2011).Article 
PubMed 

Google Scholar 
Poulson, T. L. Cave adaptation in amblyopsid fishes. Am. Midl. Nat. 70, 257–290. https://doi.org/10.2307/2423056 (1963).Article 

Google Scholar 
Venarsky, M. P., Huryn, A. D. & Benstead, J. P. Re-examining extreme longevity of the cave crayfish Orconectes australis using new mark–recapture data: A lesson on the limitations of iterative size-at-age models. Freshw. Biol. 57, 1471–1481. https://doi.org/10.1111/j.1365-2427.2012.02812.x (2012).Article 

Google Scholar 
Culver, D. C., Kane, T. C. & Fong, D. W. Adaptation and Natural Selection in Caves: The Evolution of Gammarus minus (Harvard University Press, 1995).Book 

Google Scholar 
Niemiller, M. L. & Poulson, T. L. Subterranean fishes of North America: Amblyopsidae. In Biology of Subterranean Fishes (eds Trajano, E. et al.) 169–280 (CRC Press, 2010).Chapter 

Google Scholar 
Fišer, C., Zagmajster, M. & Zakšek, V. Coevolution of life history traits and morphology in female subterranean amphipods. Oikos 122, 770–778. https://doi.org/10.1111/j.1600-0706.2012.20644.x (2013).Article 

Google Scholar 
Purvis, A., Gittleman, J. L., Cowlishaw, G. & Mace, G. M. Predicting extinction risk in declining species. Proc. Biol. Sci. 267, 1947–1952. https://doi.org/10.1098/rspb.2000.1234 (2000).Article 
PubMed 
PubMed Central 

Google Scholar 
Pearson, R. G. et al. Life history and spatial traits predict extinction risk due to climate change. Nat. Clim. Change 4, 217–221. https://doi.org/10.1038/nclimate2113 (2014).Article 
ADS 

Google Scholar 
Niemiller, M. L., Bichuette, E. & Taylor, S. J. Conservation of cave fauna in Europe and the Americas. In Ecological Studies: Cave Ecology (eds Moldovan, O. T. et al.) 451–478 (Springer, 2018).Chapter 

Google Scholar 
Niemiller, M. L. & Taylor, S. J. Protecting cave life. In Encyclopedia of Caves (eds Culver, D. C. et al.) 822–829 (Academic Press, 2019).Chapter 

Google Scholar 
Niemiller, M. L., Taylor, S. J., Slay, M. E. & Hobbs, H. H. III. Biodiversity in the United States and Canada. In Encyclopedia of Caves (eds Culver, D. C. et al.) 163–176 (Academic Press, 2019).Chapter 

Google Scholar 
Hortal, J. et al. Seven shortfalls that beset large-scale knowledge of biodiversity. Annu. Rev. Ecol. Evol. Syst. 46, 523–529. https://doi.org/10.1146/annurev-ecolsys-112414-054400 (2015).Article 

Google Scholar 
MacKenzie, D. I. et al. Occupancy Estimation and Modeling: Inferring Patterns and Dynamics of Species Occurrence (Academic Press, 2018).MATH 

Google Scholar 
Roberto, P. & Pietro, B. Species rediscovery or lucky endemic? Looking for the supposed missing species Leistus punctatissimus through a biogeographer’s eye (Coleoptera, Carabidae). ZooKeys 740, 97–108. https://doi.org/10.3897/zookeys.740.23495 (2018).Article 

Google Scholar 
Chu, C., Mandrak, N. E. & Minns, C. K. Potential impacts of climate change on the distributions of several common and rare freshwater fishes in Canada. Divers. Distrib. 11, 299–310. https://doi.org/10.1111/j.1366-9516.2005.00153.x (2005).Article 

Google Scholar 
Larson, E. R. & Olden, J. D. Latent extinction and invasion risk of crayfishes in the southeastern United States. Conserv. Biol. 24, 1099–1110. https://doi.org/10.1111/j.1523-1739.2010.01462.x (2010).Article 
PubMed 

Google Scholar 
Filipe, A. F. et al. Selection of priority areas for fish conservation in Guadiana River Basin, Iberian Peninsula. Conserv. Biol. 18, 189–200. https://doi.org/10.1111/j.1523-1739.2004.00620.x (2004).Article 

Google Scholar 
Mammola, S. et al. Fundamental research questions in subterranean biology. Biol. Rev. 95, 1855–1872. https://doi.org/10.1111/brv.12642 (2020).Article 
PubMed 

Google Scholar 
Domínguez-Domínguez, O., Martínez-Meyer, E., Zombrano, L. & de León, G. P. Using ecological-niche modeling as a conservation tool for freshwater species: Live-bearing fishes in central Mexico. Conserv. Biol. 20, 1730–1739. https://doi.org/10.1111/j.1523-1739.2006.00588.x (2006).Article 
PubMed 

Google Scholar 
Mammola, S. & Leroy, B. Applying species distribution models to caves and other subterranean habitats. Ecography 41, 1194–1208. https://doi.org/10.1111/ecog.03464 (2018).Article 

Google Scholar 
Castellarini, F., Malard, F., Dole-Olivier, M. & Gibert, J. Modelling the distribution of stygobionts in the Jura Mountains (eastern France). Implications for the protection of ground waters. Divers. Distrib. 13, 213–224. https://doi.org/10.1111/j.1472-4642.2006.00317.x (2007).Article 

Google Scholar 
Foulquier, A., Malard, F., Lefébure, T., Douady, C. J. & Gibert, J. The imprint of Quaternary glaciers on the present-day distribution of the obligate groundwater amphipod Niphargus virei (Niphargidae). J. Biogeogr. 35, 552–564. https://doi.org/10.1111/j.1365-2699.2007.01795.x (2008).Article 

Google Scholar 
Johns, T. et al. Regional-scale drivers of groundwater faunal distributions. Freshw. Sci. 34, 316–328. https://doi.org/10.1086/678460 (2015).Article 

Google Scholar 
Camp, C. D., Wooten, J. A., Jensen, J. B. & Bartek, D. F. Role of temperature in determining relative abundance in cave twilight zones by two species of lungless salamander (family Plethodontidae). Can. J. Zool. 92, 119–127. https://doi.org/10.1139/cjz-2013-0178 (2014).Article 

Google Scholar 
Korbel, K. L., Hancock, P. J., Serov, P., Lim, R. P. & Hose, G. C. Groundwater ecosystems vary with land use across a mixed agricultural landscape. J. Environ. Qual. 42, 380–390. https://doi.org/10.2134/jeq2012.0018 (2013).Article 
PubMed 

Google Scholar 
Español, C. et al. Does land use impact on groundwater invertebrate diversity and functionality in floodplains?. Ecol. Eng. 103, 394–403. https://doi.org/10.1016/j.ecoleng.2016.11.061 (2017).Article 

Google Scholar 
Christman, M. C. et al. Predicting the occurrence of cave-inhabiting fauna based on features of the earth surface environment. PLoS One 11, e0160408. https://doi.org/10.1371/journal.pone.0160408 (2016).Article 
PubMed 
PubMed Central 

Google Scholar 
Zagmajster, M. et al. Geographic variation in range size and beta diversity of groundwater crustaceans: Insights from habitats with low thermal seasonality. Glob. Ecol. Biogeogr. 23, 1135–1145. https://doi.org/10.1111/geb.12200 (2014).Article 

Google Scholar 
Poff, N. L. Landscape filters and species traits: Towards mechanistic understanding and prediction in stream ecology. J. North Am. Benthol. Soc. 16, 391–409. https://doi.org/10.2307/1468026 (1997).Article 

Google Scholar 
Stevenson, R. J. Scale-dependent determinants and consequences of benthic algal heterogeneity. J. North Am. Benthol. Soc. 16, 248–262. https://doi.org/10.2307/1468255 (1997).Article 
ADS 

Google Scholar 
U.S. Geological Survey. NLCD 2011 land cover. Multi-Resolution Land Characteristics. https://www.mrlc.gov/data/nlcd-2011-land-cover-conus (2011).Adamski, J. C. Geochemistry of the Springfield Plateau Aquifer of the Ozark Plateaus Province in Arkansas, Kansas, Missouri and Oklahoma, USA. Hydrol. Process. 14, 849–866. https://doi.org/10.1002/(SICI)1099-1085(20000415)14:5%3c849::AID-HYP973%3e3.0.CO;2-7 (2000).Article 
ADS 

Google Scholar 
Woods, A. J. et al. Ecoregions of Oklahoma (Color Poster with Map, Descriptive Text, Summary Tables, and Photographs) (U.S. Geological Survey, 2005).
Google Scholar 
Unklesbay, A. G. & Vineyard, J. D. Missouri Geology: Three Billion Years of Volcanoes, Seas, Sediments, and Erosion (University of Missouri Press, 1992).
Google Scholar 
Eigenmann, C. H. A new blind fish. In Proceedings of the Indiana Academy of Science 1897 (ed Waldo, C. A.) 231 (1898).Graening, G. O., Fenolio, D. B., Niemiller, M. L., Brown, A. V. & Beard, J. B. The 30-year recovery effort for the Ozark cavefish (Amblyopsis rosae): Analysis of current distribution, population trends, and conservation status of this threatened species. Environ. Biol. Fish. 87, 55–88. https://doi.org/10.1007/s10641-009-9568-2 (2010).Article 

Google Scholar 
Niemiller, M. L., Near, T. J. & Fitzpatrick, B. M. Delimiting species using multilocus data: Diagnosing cryptic diversity in the southern cavefish, Typhlichthys subterraneus (Teleostei: Amblyopsidae). Evolution 66, 846–866. https://doi.org/10.1111/j.1558-5646.2011.01480.x (2012).Article 
PubMed 

Google Scholar 
Hobbs, H. H. Jr. & Brown, A. V. A new troglobitic crayfish from northwestern Arkansas (Decapoda: Cambaridae). Proc. Biol. Soc. Wash. 100, 1040–1048 (1987).
Google Scholar 
Graening, G. O., Slay, M. E., Brown, A. V. & Koppelman, J. B. Status and distribution of the endangered Benton cave crayfish, Cambarus aculabrum (Decapoda: Cambaridae). Southwest. Nat. 51, 376–381. https://doi.org/10.1894/0038-4909(2006)51[376:SADOTE]2.0.CO;2 (2006).Article 

Google Scholar 
Faxon, W. Cave animals from southwestern Missouri. Bull. Mus. Comp. Zool. 17, 225–240 (1889).
Google Scholar 
Graening, G. O., Hobbs, H. H. III., Slay, M. E., Elliott, W. R. & Brown, A. V. Status update for bristly cave crayfish, Cambarus setosus (Decapoda: Cambaridae), and range extension into Arkansas. Southwest. Nat. 51, 382–392. https://doi.org/10.1894/0038-4909(2006)51[382:SUFBCC]2.0.CO;2 (2006).Article 

Google Scholar 
Hobbs, H. H. III. Cambarus (Jugicambarus) subterraneus, a new cave crayfish (Decapoda: Cambaridae) from northeastern Oklahoma, with a key to the troglobitic members of the subgenus Jugicambarus. Proc. Biol. Soc. Wash. 106, 719–727 (1993).
Google Scholar 
Graening, G. O. & Fenolio, D. B. Status update of the Delaware County cave crayfish, Cambarus subterraneus (Decapoda: Cambaridae). Proc. Okla. Acad. Sci. 85, 85–89 (2005).
Google Scholar 
Hobbs, H. H. Jr. & Cooper, M. R. A new troglobitic crayfish from Oklahoma (Decapoda: Astacidae). Proc. Biol. Soc. Wash. 85, 49–56 (1972).
Google Scholar 
Graening, G. O. et al. Range extension and status update for the Oklahoma cave crayfish, Cambarus tartarus (Decapoda: Cambaridae). Southwest. Nat. 51, 94–99 (2006).Article 

Google Scholar 
Hobbs, H. H. III. A new cave crayfish of the genus Orconectes, subgenus Orconectes, from southcentral Missouri, USA, with a key to the stygobitic species of the genus (Decapoda, Cambaridae). Crustaceana 74, 635–646. https://doi.org/10.1163/156854001750377911 (2001).Article 

Google Scholar 
Miller, B. V. The Hydrology of the Carroll Cave-Toronto Springs System: Identifying and Examining Source Mixing Through Dye Tracing, Geochemical Monitoring, Seepage Runs, and Statistical Methods (Western Kentucky University, 2010).
Google Scholar 
Mouser, J. B., Brewer, S. K., Niemiller, M. L., Mollenhauer, R. & Van Den Bussche, R. Comparing visual and environmental DNA surveys for detection of stygobionts. Subterr. Biol. 39, 79–105. https://doi.org/10.3897/subtbiol.39.64279 (2021).Article 

Google Scholar 
Longmire, J. L., Maltbie, M. & Baker, R. J. Use of “Lysis Buffer” in DNA Isolation and Its Implication for Museum Collections (Museum of Texas Tech University, 1997).Book 

Google Scholar 
Mouser, J. B., Mollenhauer, R. & Brewer, S. K. Relationships between landscape constraints and a crayfish assemblage with consideration of competitor presence. Divers. Distrib. 25, 61–73. https://doi.org/10.1111/ddi.12840 (2019).Article 

Google Scholar 
U.S. Geological Survey. 1 Arc-second digital elevation models (DEMs)—USGS national map 3DEP downloadable data collection. https://data.usgs.gov/datacatalog/data/USGS:35f9c4d4-b113-4c8d-8691-47c428c29a5b (U.S. Geological Survey, 2017).Oak Ridge National Laboratory Distributed Active Archive Center. MODIS and VIIRS land products global subsetting and visualization tool. Oak Ridge National Laboratory Distributed Active Archive Center. https://doi.org/10.3334/ORNLDAAC/1379 (2018).Horton, J. D., San Juan, C. A. & Stoeser, D. B. The state geologic map compilation (SGMC) geodatabase of the conterminous United States. U.S. Geol. Surv. https://doi.org/10.3133/ds1052 (2017).Article 

Google Scholar 
MacKenzie, D. I. et al. Estimating site occupancy rates when detection probabilities are less than one. Ecology 83, 2248–2255. https://doi.org/10.1890/0012-9658(2002)083[2248:ESORWD]2.0.CO;2 (2002).Article 

Google Scholar 
Tyre, A. J. et al. Improving precision and reducing bias in biological surveys: Estimating false negative error rates. Ecol. Appl. 13, 1790–1801. https://doi.org/10.1890/02-5078 (2003).Article 

Google Scholar 
Gelman, A. & Hill, J. Data Analysis Using Regression and Multilevel Hierarchical Models (Cambridge University Press, 2007).
Google Scholar 
Kéry, M. & Royle, J. A. Applied Hierarchical Modeling in Ecology: Analysis of Distribution, Abundance and Species Richness in R and BUGS (Academic Press, 2016).MATH 

Google Scholar 
Plummer, M. JAGS: A program for analysis of Bayesian graphical models using Gibbs sampling. In Proceedings of the 3rd International Workshop on Distributed Statistical Computing (eds Hornik, K. et al.) 1–10 (Austrian Science Foundation, 2003).
Google Scholar 
R Core Team. R: A Language and Environment for Statistical Computing (R Foundation for Statistical Computing, 2020).
Google Scholar 
Kellner, J. jagsUI: A wrapper around ‘rjags’ to streamline ‘JAGS’ analyses. https://CRAN.R-project.org/package=jagsUI (R-project, 2019).Brooks, S. P. & Gelman, A. General methods for monitoring convergence of iterative simulations. J. Comput. Graph. Stat. 7, 434–455. https://doi.org/10.1080/10618600.1998.10474787 (1998).Article 
MathSciNet 

Google Scholar 
Kruschke, J. K. Doing Bayesian Data Analysis: A Tutorial with R, JAGS, and Stan (Academic Press, 2015).MATH 

Google Scholar 
Hobbs, N. T. & Hooten, M. B. Bayesian Models (Princeton University Press, 2015). https://doi.org/10.1515/9781400866557.Book 

Google Scholar 
Conn, P. B., Johnson, D. S., Williams, P. J., Melin, S. R. & Hooten, M. B. A guide to Bayesian model checking for ecologists. Ecol. Monogr. 88, 526–542. https://doi.org/10.1002/ecm.1314 (2018).Article 

Google Scholar 
Allan, J. D. Landscapes and riverscapes: The influence of land use on stream ecosystems. Annu. Rev. Ecol. Evol. Syst. 35, 257–284. https://doi.org/10.1146/annurev.ecolsys.35.120202.110122 (2004).Article 

Google Scholar 
Paul, M. J. & Meyer, J. L. Streams in the urban landscape. Annu. Rev. Ecol. Evol. Syst. 32, 333–365. https://doi.org/10.1146/annurev.ecolsys.32.081501.114040 (2001).Article 

Google Scholar 
Wicks, C., Kelley, C. & Peterson, E. Estrogen in a karstic aquifer. Groundwater 42, 384–389. https://doi.org/10.1111/j.1745-6584.2004.tb02686.x (2004).Article 

Google Scholar 
Buřič, M., Kouba, A., Máchová, J., Mahovská, I. & Kozák, P. Toxicity of the organophosphate pesticide diazinon to crayfish of differing age. Int. J. Environ. Sci. Technol. 10, 607–610. https://doi.org/10.1007/s13762-013-0185-4 (2013).Article 

Google Scholar 
Sohn, L., Brodie, R. J., Couldwell, G., Demmons, E. & Sturve, J. Exposure to a nicotinoid pesticide reduces defensive behaviors in a non-target organism, the rusty crayfish Orconectes rusticus. Ecotoxicology 27, 900–907. https://doi.org/10.1007/s10646-018-1950-4 (2018).Article 
PubMed 
PubMed Central 

Google Scholar 
Noltie, D. B. & Wicks, C. M. How hydrogeology has shaped the ecology of Missouri’s Ozark cavefish, Amblyopsis rosae, and southern cavefish, Typhlichthys subterraneus: Insights on the sightless from understanding the underground. Environ. Biol. Fish. 62, 171–194. https://doi.org/10.1023/A:1011815806589 (2001).Article 

Google Scholar 
Kuhajda, B. R. & Mayden, R. L. Status of the federally endangered Alabama cavefish, Speoplatyrhinus poulsoni (Amblyopsidae), in Key Cave and surrounding caves, Alabama. Environ. Biol. Fish. 62, 215–222. https://doi.org/10.1023/A:1011817023749 (2001).Article 

Google Scholar 
Hutchins, B. T. The conservation status of Texas groundwater invertebrates. Biodivers. Conserv. 27, 475–501. https://doi.org/10.1007/s10531-017-1447-0 (2018).Article 

Google Scholar 
Niemiller, M. L. et al. Discovery of a new population of the federally endangered Alabama cave shrimp, Palaemonias alabamae Smalley, 1961, in northern Alabama. Subterr. Biol. 32, 43–59. https://doi.org/10.3897/subtbiol.32.38280 (2019).Article 

Google Scholar 
Abell, R., Allan, J. D. & Lehner, B. Unlocking the potential of protected areas for freshwaters. Biol. Conserv. 134, 48–63. https://doi.org/10.1016/j.biocon.2006.08.017 (2007).Article 

Google Scholar 
Liu, Y. et al. A review on effectiveness of best management practices in improving hydrology and water quality: Needs and opportunities. Sci. Total Environ. 601–602, 580–593. https://doi.org/10.1016/j.scitotenv.2017.05.212 (2017).Article 
ADS 
PubMed 

Google Scholar  More

All methods were performed in accordance with the local relevant guidelines, regulations and legislation.InstrumentsLI-6400 photosynthesis system (LI-6400 Inc., Lincoln, NE, USA) and PAM-2500 portable chlorophyll fluorescence apparatus (PAM-2500, Walz, Germany) were used in the study.MaterialsAbout 90 living E. brevicornu plants were collected from Taihang Mountains in October 2018. The E. brevicornu was not in endangered or protected. The collection of these E. brevicornu plants was permitted by local government. These plants were averagely planted in nine plots of 2 m2. The roots of E. pubescens were planted 6–8 cm below ground. These plots were placed on farmland near Taihang Mountains and covered with sunshade net (about 70% light transmittance). These plants were timely irrigated after planting to ensure that they grew well but not fertilized.Determination of photosynthetic characteristicsThe photosynthetic characteristics of mature leaves on the E. brevicornu plants were determined between June 6–8, 2019 with the Li-6400 photosynthesis system. The diurnal variation of photosynthesis in three leaves of three plants was determined. When the light response curve was determined, the temperature of the leaf chamber was set at 28 °C, and the concentration of CO2 in the leaf chamber was set at 400 µmol mol−1. When determining the CO2 response curve, the light intensity in the leaf chamber was set at 1000 µmol m−2 s−1, and the temperature of the leaf chamber was set at 28 °C. The light response curve and CO2 response curve were determined three times in three leaves of three different plants.Determination of chlorophyll fluorescence characteristicsThe fluorescence characteristics of chlorophyll in E. brevicornu leaves were determined with PAM-2500 portable chlorophyll fluorescence apparatus between June 8–9, 2019. The leaves underwent dark adaptation for 30 min before determining slow kinetics of chlorophyll fluorescence. Then the light curves of chlorophyll fluorescence were determined. All of these determinations were repeated three times on three mature leaves of three plants.The data was analysed with SPSS (Statistical Product and Service Solutions, International Business Machines Corporation, USA). The light response curves were fitted with following modified rectangular hyperbola model11,12.$${text{Photo}}, = ,{text{E}}cdotleft( {{1} – {text{M}}cdot{text{PAR}}} right)cdotleft( {{text{PAR}} – {text{LCP}}} right)/({1}, + ,{text{N}}cdot{text{PAR}})$$PAR is the value of light intensity in leaf chamber of Li-6400 photosynthesis system. Photo is net photosynthetic rate. LCP is the light compensation point. E is the apparent quantum yield. M and N are parameters. The dark respiration rate under the LCP is calculated according to E·LCP. The light saturation point (LSP) is calculated according to (((M + N) ·(1 + N·LCP)/M)½)/−1)/N.The net photosynthetic rate under the light saturation point (LSP) can be calculated according to the above model.The CO2 response curves were fitted with below modified rectangular hyperbola model11,12.$${text{Photo}}, = ,{text{E}}cdotleft( {{1} – {text{M}}cdot{text{PAR}}} right)cdotleft( {{text{PAR}} – {text{CCP}}} right)/({1}, + ,{text{N}}cdot{text{PAR}})$$PAR is the value of light intensity in leaf chamber of Li-6400 photosynthesis system. Photo is net photosynthetic rate. CCP is CO2 compensation point. E is also the apparent quantum yield. M and N are parameters. The dark respiration rate under the CO2 calculated according to E·CCP. The CO2 saturation point (CSP) is calculated according to (((M + N) ·(1 + N·CCP)/M)½)/−1)/N.The net photosynthetic rate under the CO2 saturation point (CSP) can be alternatively calculated according to the above model.The light curves of chlorophyll fluorescence were fitted according to the below model of Eilers and Peeters12,13.$${text{ETR}}, = ,{text{PAR}}/({text{a}}cdot{text{PAR}}^{{2}} , + ,{text{b}}cdot{text{PAR}}, + ,{text{c}})$$ETR is the electron transport rate of photosynthetic system II. PAR is fluorescence intensity. The letters a, b and c are parameters. More

Balvanera, P. et al. Bioscience 64, 49–57 (2014).Article 

Google Scholar 
Hooper, D. U. et al. Ecol. Monogr. 75, 3–35 (2005).Article 

Google Scholar 
Balvanera, P. et al. Ecol. Lett. 9, 1146–1156 (2006).Article 
PubMed 

Google Scholar 
Cardinale, B. J. et al. Am. J. Bot. 98, 572–592 (2011).Article 
PubMed 

Google Scholar 
Cardinale, B. J. B. J. et al. Nature 486, 59–67 (2012).Article 
CAS 
PubMed 

Google Scholar 
Manning, P. et al. in Advances in Ecological Research (eds Eisenhauer N. et al.) 323–356 (Academic, 2019).Le Provost, G. et al. Nat. Ecol. Evol. https://doi.org/10.1038/s41559-022-01918-5 (2022).Felipe-Lucia, M. R. et al. Proc. Natl Acad. Sci. USA 117, 28140–28149 (2020).Article 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Foley, J. A. et al. Science 309, 570–574 (2005).Article 
CAS 
PubMed 

Google Scholar 
Cardinale, B. J. et al. Ecology 94, 1697–1707 (2013).Article 
PubMed 

Google Scholar 
Teles da Mota, V. & Pickering, C. J. Outdoor Recreat. Tour. 30, 100295 (2020).Article 

Google Scholar 
Mitchell, M. G. E. et al. Trends Ecol. Evol. 30, 190–198 (2015).Article 
PubMed 

Google Scholar 
Raudsepp-Hearne, C. & Peterson, G. D. Ecol. Soc. 21, 16 (2016).Article 

Google Scholar 
Chaplin-Kramer, R. & Kremen, C. Ecol. Appl. 22, 1936–1948 (2012).Article 
PubMed 

Google Scholar  More

Dataset acquisitionPlant-frugivore network data were obtained through different online sources and publications (Supplementary Table 1). Only networks that met the following criteria were retrieved: (i) the network contains quantitative data (a measure of interaction frequency) from a location, pooling through time if necessary; (ii) the network includes avian frugivores. Importantly, we removed non-avian frugivores from our analyses because only 28 out of 196 raw networks (before data cleaning) sampled non-avian frugivores, and not removing non-avian frugivores would generate spurious apparent turnover between networks that did vs. did not sample those taxa. In addition, the removal of non-avian frugivores did not strongly decrease the number of frugivores in our dataset (Supplementary Fig. 20a) or the total number of links in the global network of frugivory (Supplementary Fig. 20b). Furthermore, non-avian frugivores, as well as their interactions, were not shared across ecoregions and biomes (Supplementary Fig. 21), so their inclusion would only strengthen the results we found (though as noted above, we believe that this would be spurious because they are not as well sampled); (iii) the network (after removal of non-avian frugivores) contains greater than two species in each trophic level. Because this size threshold was somewhat arbitrary, we used a sensitivity analysis to assess the effect of our network size threshold on the reported patterns (see Sensitivity analysis section in the Supplementary Methods and Supplementary Figs. 22–24); and (iv) network sampling was not taxonomically restricted, that is, sampling was not focused on a specific taxonomic group, such as a given plant or bird family. Note, however, that authors often select focal plants or frugivorous birds to be sampled, but this was not considered as a taxonomic restriction if plants and birds were not selected based on their taxonomy (e.g., focal plants were selected based on the availability of fruits at the time of sampling, or focal birds were selected based on previous studies of bird diet in the study site). The first source for network data was the Web of Life database42, which contains 33 georeferenced plant-frugivore networks from 28 published studies, of which 12 networks met our criteria.We also accessed the Scopus database on 04 May 2020 using the following keyword combination: (“plant-frugivore*” OR “plant-bird*” OR “frugivorous bird*” OR “avian frugivore*” OR “seed dispers*”) AND (“network*” OR “web*”) to search for papers that include data on avian frugivory networks. The search returned a total of 532 studies, from which 62 networks that met the above criteria were retrieved. We also contacted authors to obtain plant-frugivore networks that were not publicly available, which provided us a further 110 networks. The remaining networks (N = 12) were obtained by checking the database from a recently published study12. In total, 196 quantitative avian frugivory networks were used in our analyses.Generating the distance matrices to serve as predictor and response variablesEcoregion and biome distancesWe used the most up-to-date (2017) map of ecoregions and biomes3, which divides the globe into 846 terrestrial ecoregions nested within 14 biomes, to generate our ecoregion and biome distance matrices. Of these, 67 ecoregions and 11 biomes are represented in our dataset (Supplementary Figs. 1 and 2). We constructed two alternative versions of both the ecoregion and biome distance matrices. In the first, binary version, if two ecological networks were from localities within the same ecoregion/biome, a dissimilarity of zero was given to this pair of networks, whereas a dissimilarity of one was given to a pair of networks from distinct ecoregions/biomes (this is the same as calculating the Euclidean distance on a presence–absence matrix with networks in rows and ecoregion/biomes in columns).In the second, quantitative version, we estimated the pairwise environmental dissimilarity between our ecoregions and biomes using six environmental variables recently demonstrated to be relevant in predicting ecoregion distinctness, namely mean annual temperature, temperature seasonality, mean annual rainfall, rainfall seasonality, slope and human footprint38. We obtained climatic and elevation data from WorldClim 2.143 at a spatial resolution of 1-km2. We transformed the elevation raster into a slope raster using the terrain function from the raster package44 in R45. As a measure of human disturbance, we used human footprint—a metric that combines eight variables associated with human disturbances of the environment: the extent of built environments, crop land, pasture land, human population density, night-time lights, railways, roads and navigable waterways26. The human footprint raster was downloaded at a 1-km2 resolution26. Because human footprint data were not available for one of our ecoregions (Galápagos Islands xeric scrub), we estimated human footprint for this ecoregion by converting visually interpreted scores into the human footprint index. We did this by analyzing satellite images of the region and following a visual score criterion26. Given the previously demonstrated strong agreement between visual score and human footprint values26, we fitted a linear model using the visual score and human footprint data from 676 validation plots located within the Deserts and xeric shrublands biome – the biome in which the Galápagos Islands xeric scrub ecoregion is located – and estimated the human footprint values for our own visual scores using the predict function in R45.We used 1-km2 resolution rasters and the extract function from the raster package44 to calculate the mean value of each of our six environmental variables for each ecoregion in our dataset. Because biomes are considerably larger than ecoregions (which makes obtaining environmental data for biomes more computationally expensive) we used a coarser spatial resolution of 5-km2 for calculating the mean values of environmental variables for each biome. Since a 5-km2 resolution raster was not available for human footprint, we transformed the 1-km2 resolution raster into a 5-km2 raster using the resample function from the same package.To combine these six environmental variables into quantitative matrices of ecoregion and biome environmental dissimilarity, we ran a Principal Component Analysis (PCA) on our scaled multivariate data matrix (where rows are ecoregions or biomes and columns are environmental variables). From this PCA, we selected the scores of the four and three principal components, which represented 89.6% and 88.7% of the variance for ecoregions and biomes, respectively, and converted it into a distance matrix by calculating the Euclidean distance between pairs of ecoregions/biomes using the vegdist function from the vegan package46. Finally, we transformed the ecoregion or biome distance matrix into a N × N matrix where N is the number of local networks. In this matrix, cell values represent the pairwise environmental dissimilarity between the ecoregions/biomes where the networks are located. The main advantage of using this quantitative approach is that, instead of simply evaluating whether avian frugivory networks located in distinct ecoregions or biomes are different from each other in terms of network composition and structure (as in our binary approach), we were also able to determine whether the extent of network dissimilarity depended on how environmentally different the ecoregions or biomes are from one another.Local-scale human disturbance distanceTo generate our local human disturbance distance matrix, we extracted human footprint data at a 1-km2 spatial resolution26 and calculated the mean human footprint values within a 5-km buffer zone around each network site. For the networks located within the Galápagos Islands xeric scrub ecoregion (N = 4), we estimated the human footprint index using the same method described in the previous section for ecoregion- or biome-scale human footprint. We then calculated the pairwise Euclidean distance between human footprint values from our network sites. Thus, low cell values in the local human disturbance distance matrix indicate pairs of network sites with a similar level of human disturbance, while high values represent pairs of network sites with very different levels of human disturbance.Spatial distanceThe spatial distance matrix was generated using the Haversine (i.e., great circle) distance between all pairwise combinations of network coordinates. In this matrix, cell values represent the geographical distance between network sites.Elevational differenceWe calculated the Euclidean distance between pairwise elevation values (estimated as meters above sea level) of network sites to generate our elevational difference matrix. Elevation values were obtained from the original sources when available or using Google Earth47. In the elevational difference matrix, low cell values represent pairs of network sites within similar elevations, whereas high values represent pairs of network sites within very different elevations.Network sampling dissimilarityWe used the metadata retrieved from each of our 196 local networks to generate our network sampling dissimilarity matrices, which aim to control statistically for differences in network sampling. There are many ways in which sampling effort could be quantified, so we began by calculating a variety of metrics, then narrowed our options by assessing which of these was most related to network metrics. We divided the sampling metrics into two categories: time span-related metrics (i.e., sampling hours and months) and empirical metrics of sampling completeness (i.e., sampling completeness and sampling intensity), which aim to account for how complete network sampling was in terms of species interactions (Supplementary Table 2).We selected the quantitative sampling metrics to be included in our models based on (i) the fit of generalized linear models evaluating the relationship between number of sampling hours and sampling months of the study and network-level metrics (i.e., bird richness, plant richness and number of links), and (ii) how well time span-related metrics, sampling completeness and sampling intensity predicted the proportion of known interactions that were sampled in each local network (hereafter, ratio of interactions) for a subset of the data. This latter metric, defined as the ratio between the number of interactions in the local network and the number of known possible interactions in the region involving the species in the local network, captures raw sampling completeness. Therefore, ratio of interactions estimates, for a given set of species, the proportion of all their interactions known for a region that are found to occur among those same species in the local network. To calculate this metric, we needed high-resolution information on the possible interactions, so we used a subset of 14 networks sampled in Aotearoa New Zealand, since there is an extensive compilation of frugivory events recorded for this country48. After this process, we selected number of sampling hours, number of sampling months and sampling intensity for inclusion in our statistical models (Supplementary Figs. 7 and 8; Supplementary Table 2). We generated the corresponding distance matrices by calculating the Euclidean distance between metric values. Similarly, we generated a Euclidean distance matrix for differences in sampling year between pairs of networks, which aims to account for long-term changes in the environment, species composition and network sampling methods. We obtained the sampling year of our local networks from the original sources and calculated the mean sampling year value for those networks sampled across multiple years.Because sampling methods, such as sampling design, focus (i.e., focal taxa, which determines whether a zoocentric or phytocentric method was used), interaction frequency type (i.e., how interaction frequency was measured) and coverage (total or partial) might also affect the observed plant-frugivore interactions49, we combined these variables into a single distance matrix to estimate the overall differences in sampling methods between networks. Because most of these variables were categorical with multiple levels (Supplementary Table 3), we generated our method’s dissimilarity matrix by using a generalization of Gower’s distance method50, which allows the treatment of different types of variables when calculating distances. For this, we used the dist.ktab function from the ade4 package51. We ran a Principal Coordinates Analysis (PCoA) on this distance matrix, selected the first four axes, which explained 81.2% of the variation in method’s dissimilarity, and calculated the Euclidean distance between pairs of networks using the vegdist function from the vegan package46 in R45.Network dissimilarityWe generated three network dissimilarity matrices to be our response variables in the statistical models. In the first, cell values represent the pairwise dissimilarity in species composition between networks (beta diversity of species; βS)27. Second, we measured interaction dissimilarity (beta diversity of interactions; βWN), which represents the pairwise dissimilarity in the identity of interactions between networks27. Importantly, we did not include interaction rewiring (βOS) in our main analysis because this metric can only be calculated for networks that share interaction partners (i.e., it estimates whether shared species interact differently)27, which limited the number and the spatial distribution of networks available for analysis (but see the Rewiring analysis section for an analysis on the subset of our dataset for which this was possible). Metrics were calculated using the network_betadiversity function from the betalink package52 in R45.Finally, we calculated a third dissimilarity matrix to capture overall differences in network structure. We recognize that there are many potential metrics of network structure, and that many of these are strongly correlated with one another53,54,55,56. We therefore chose a range of metrics that captured the number of links, their relative weightings (including across trophic levels), and their arrangement among species, then combined these into a single distance matrix. Specifically, we quantified network structural dissimilarity using the following metrics: weighted connectance, weighted nestedness, interaction evenness, PDI and modularity.Weighted connectance represents the number of links relative to the number of possible links, weighted by the frequency of each interaction55, and is therefore a measure of network-level specialization (higher values of weighted connectance indicate lower specialization). Importantly, it has been suggested that connectance affects persistence in mutualistic systems54. We measured nestedness (i.e., the pattern in which specialist species interact with proper subsets of the species that generalist species interact with) using the weighted version of nestedness based on overlap and decreasing fill (wNODF)57. Notably, nested structures have been commonly reported in plant-frugivore networks33. Interaction evenness is Shannon’s evenness index applied for species interactions and represents how evenly distributed the interactions are in the network21,58. This metric has been previously demonstrated to decline with habitat modification as a consequence of some interactions being favored over others in high-disturbance environments21. PDI (Paired Difference Index) is a measure of species-level specialization on resources and a reliable indicator not only of specialization, but also of absolute generalism59. Thus, this metric contributes to understanding of the ecological processes that drive the prevalence of specialists or generalists in ecological networks59. In order to obtain a network-level PDI, we calculated the weighted mean PDI for each local network. Finally, we calculated modularity (i.e., the level of compartmentalization within networks) using the DIRTPLAwb+ algorithm60. Modularity estimates the extent to which species within modules interact more with each other than with species from other modules61, and it has been demonstrated to affect the persistence and resilience of mutualistic networks54. All the selected network metrics are based on weighted (quantitative) interaction data, as these have been suggested to be less biased by sampling incompleteness62 and to better reflect environmental changes21. All network metrics were calculated using the bipartite package63 in R45.We ran a Principal Component Analysis (PCA) on our scaled multivariate data matrix (N × M where N is the number of local networks in our dataset and M is the number of network metrics), selected the scores of the three principal components, which represented 89.9% of the variance in network metrics, and converted it into a network structural dissimilarity matrix by calculating the Euclidean distance between networks. In this distance matrix, cell values represent differences in the overall architecture of networks (over all the network metrics calculated), and therefore provide a complementary approach for evaluating how species interaction patterns vary across large-scale environmental gradients.Statistical analysisWe employed a two-tailed statistical test that combines Generalized Additive Models (GAM)29 and Multiple Regression on distance Matrices (MRM)30 to evaluate the effect of each of our predictor distance matrices on our response matrix. With this approach, we were able to fit GAMs where the predictor and responsible variables are distance matrices, while accounting for the non-independence of distances from each local network by permuting the response matrix30. The main advantage of using GAMs is their flexibility in modeling non-linear relationships through smooth functions, which are represented by a sum of simpler, fixed basis functions that determine their complexity29. Using GAM-based MRM models allowed us to obtain F values for each of the smooth terms (i.e., smooth functions of the predictor variables in our model), and test statistical significance at the level of individual variables. The binary versions of ecoregion and biome distance matrices (with two levels, “same” or “distinct”) were treated as categorical variables in the models, and t values were used for determining statistical significance. We fitted GAMs with thin plate regression splines64 using the gam function from the mgcv package29 in R45. Smoothing parameters were estimated using restricted maximum likelihood (REML)29. Our GAM-based MRM models were calculated using a modified version of the MRM function from the ecodist package65, which allowed us to combine GAMs with the permutation approach from the original MRM function (see Code availability). All the models were performed with 1000 permutations (i.e., shuffling) of the response matrix.We explored the unique and shared contributions of our predictor variables to network dissimilarity using deviance partitioning analyses. These were performed by fitting reduced models (i.e., GAMs where one or more predictor variables of interest were removed) using the same smoothing parameters as in the full model and comparing the explained deviance. We fixed smoothing parameters for comparisons in this way because these parameters tend to vary substantially (to compensate) if one of two correlated predictors is dropped from a GAM.Assessing the influence of individual studies on the reported patternsBecause our dataset comprises 196 local frugivory networks obtained from 93 different studies, and some of these studies contained multiple networks, we needed to evaluate whether our results were strongly biased by individual studies. To do this, we followed the approach from a previous study66 and tested whether F values of smooth terms and t values of categorical variables (binary version of ecoregion and biome distances) changed significantly when jackknifing across studies. We did this by dropping one study from the dataset and re-fitting the models, and then repeating this same process for all the studies in our dataset.We found a number of consistent patterns within different subsets of the data (Supplementary Figs. 15 and 16); however, some of the patterns we observed appear to be driven by individual studies with multiple networks, and hence are less representative. For instance, the study with the greatest number of networks in our dataset (study ID = 76), which contains 35 plant-frugivore networks sampled across an elevation gradient in Mt. Kilimanjaro, Tanzania67, had an overall high influence on the results when compared with the other studies. By re-running our GAM-based MRM models after removing this study from our dataset, we found that the effect of biome boundaries on interaction dissimilarity is no longer significant, whereas the effects of ecoregion boundaries, human disturbance distance, spatial distance and elevational differences remained consistent with those from the full dataset (Supplementary Table 33). Nevertheless, all the results were qualitatively similar to those obtained for the entire dataset when using network structural dissimilarity as the response variable (Supplementary Table 34).Rewiring analysisInteraction rewiring (βOS) estimates the extent to which shared species interact differently27. Because this metric can only be calculated for networks that share species from both trophic levels, we selected a subset of network pairs that shared plants and frugivorous birds (N = 1314) to test whether interaction rewiring increases across large-scale environmental gradients. Importantly, since not all possible combinations of network pairs contained values of interaction rewiring (i.e., not all pairs of networks shared species), a pairwise distance matrix could not be generated for this metric. Thus, we were not able to use the same statistical approach used in our main analysis, which is based on distance matrices (see Statistical analysis section). Instead, we performed a Generalized Additive Mixed-effects Model (GAMM) using ecoregion, biome, human disturbance, spatial, elevational, and sampling-related distance metrics as fixed effects and network IDs as random effects (to account for the non-independence of distances) (Supplementary Table 35). We also performed a reduced model with only ecoregion and biome distance metrics as predictor variables (Supplementary Table 36). The binary version of ecoregion and biome distance metrics (with two levels, “same” or “distinct”) were used as categorical variables in both models. Interaction rewiring (βOS) was calculated using the network_betadiversity function from the betalink package52 in R45. Although it has been recently argued that this metric may overestimate the importance of rewiring for network dissimilarity68, our main focus was not the partitioning of network dissimilarity into species turnover and rewiring components, but rather simply detecting whether the sub-web of shared species interacted differently. In this case, βOS (as developed by ref. 27) is an adequate and useful metric68. We fitted our models using the gamm4 function from the gamm4 package69 in R45. Smoothing parameters were estimated using restricted maximum likelihood (REML)29.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article. More

Experiment results and regression modelThe simulation experiment results based on the design scheme are presented in Table 4, including 24 analysis factors and 7 zero-point experiments for estimating the errors. Quadratic multiple regression analysis of the results in Table 4 was performed using the Design-Expert software, and the regression models between the influencing factors and evaluation indices were established as follows:$$ Y_{{1}} = {1767.57} – {64.29}X_{{1}} + {117.46}X_{{2}} + {324.46}X_{{3}} + {107.87}X_{{4}} – {21.81}X_{{1}} X_{{2}} + {17.94}X_{{1}} X_{{3}} – {41.44}X_{{1}} X_{{4}} + {16.69}X_{{2}} X_{{3}} – {41.19}X_{{2}} X_{{4}} + {73.56}X_{{3}} X_{{4}} + {23.2}{X_{{1}}^{{2}}} – {82.42}{{X_{{2}}}^{{2}}} – {13.17}{{X_{{3}}}^{{2}}} – {53.67}{{X_{{4}}}^{{2}}} $$$$ Y_{{2}} = {1968.14} + {636.42}X_{1} + {34.42}X_{2} + {66}X_{3} + {115.17}X_{{4}} + {28.63}X_{{1}} X_{{2}} + {9.13}X_{{1}} X_{{3}} – { 45.87}X_{{1}} X_{{4}} + {1}0X_{{2}} X_{{3}} + {30.5}X_{{2}} X_{{4}} – {1.75}X_{{3}} X_{{4}} + {55.03}{X_{{1}}^{{2}}} – {8.1}{{X_{{2}}}^{{2}}} – {72.72}{{X_{{3}}}^{2}} + {61.03}{{X_{{4}}}^{{2}}} $$Table 4 Experiment schemes and results.Full size tableThe relationship between the actual values of the efficiency of conveying-soil and the distance of throwing-soil and the predicted values of the regression model is shown in Fig. 7. It can be seen from Fig. 7 that the actual values are basically distributed on the predicted curve, consistent with the trend of the predicted values, and linearly distributed.Figure 7Scatter plot. (a) Scatter plot of actual and predicted distance of throwing-soil. (b) Scatter plot of actual and predicted efficiency of conveying-soil.Full size imageVariance analysis and discussionThe F-test and analysis of variance (ANOVA) were performed on the regression coefficients in the regression models of the evaluation indices Y1 and Y2, and the results are shown in Table 5. According to the significance values P of the lack of fitting in the regression models of the objective functions Y1 and Y2 in Table 5, PL1 = 0.1485  > 0.05 and PL2 = 0.2337  > 0.05 (both were not significant), indicating that no loss factor existed in the regression analysis, and the regression model exhibited a high fitting degree.Table 5 ANOVA results of regression model.Full size tableAccording to the ANOVA, the significance values P of each influencing factor in the test could be determined28. For the evaluation index Y1, the factors X1, X2, X3, X4, X3X4, X22, X42 had extremely significant influences, while the factors X1X4, X2X4 had a significant influence. For the evaluation index Y2, the factors X1, X3, X4, X1X4, X12, X32, X42 had extremely significant influences, and the factors X2, X1X4 had a significant influence. Within the level range of the selected factors, according to the F value of each factor as shown in Table 5, the weight of the factors affecting the efficiency of conveying-soil is feeding speed  > helix angle of auger  > rotating speed of auger  > slope angle. And the weight of the factors affecting the distance of throwing-soil is slope auger  > rotating speed of auger  > feeding speed  > helix angle of auger.In addition, it is obvious that there are interactions between the feeding speed and rotating speed of the auger, slope auger and rotating speed of auger, helix angle of the auger and rotating speed of the auger on the efficiency of conveying-soil Y1. For the distance of throwing-soil Y2, there is an interaction between the slope angle and the rotating speed of the auger.Analysis of response surfaceThe fitting coefficient of the efficiency of conveying-soil is R2 = 0.9714, R2adjust = 0.9263, R2pred = 0.8082, the difference between R2adjust and R2pred is less than 0.2. The fitting coefficient of the distance of throwing-soil is R2 = 0.9873, R2adjust = 0.9742, R2pred = 0.9355, the difference between R2adjust and R2pred is smaller than 0.2. It is indicated that the response surfaces of the two models established have good consistency and predictability for the experimental results29.The response surface is created directly using the Design-Expert software. After entering the data, select “Analysis” module. In the “Model-Graph” menu bar, select “3D-surface” to switch to the 3D view. To express the interactive influence of each factor on the efficiency of conveying-soil Y1 and distance of the throwing-soil Y2, the above two quadratic regression equations of the evaluation indices were subjected to the dimensionality reduction treatment. Two of the factors was set to level 0, while the other two underwent interaction effect analysis to study the influence law on the evaluation indices Y1 and Y2, and the corresponding response surfaces were generated, as illustrated in Fig. 8.Figure 83D response diagram effect of evaluation indices. (a) Effect of interaction between X1 and X2 on efficiency of conveying-soil. (b) Effect of interaction between X2 and X4 on efficiency of conveying-soil. (c) Effect of interaction between X3 and X4 on efficiency of conveying-soil. (d) Effect of interaction between X3 and X4 on distance of throwing-soil.Full size imageIt can be seen in Fig. 8a, when the slope angle was constant, the efficiency of conveying-soil increased with the rotating speed of the auger to a certain value, then the efficiency increase changed more gently. The reasons for this phenomenon are described as follows. On the one hand, the greater the kinetic energy of the soil when leaving the original position, and the thinner the soil was cut, resulting in the smaller the probability of blockage in the spiral blade space. On the other hand, the centrifugal force of soil arriving at the pit mouth is greater, so it does not obstruct in the pit mouth. However, if the rotation speed of the auger was too high and the soil layer cut was too thin, the subsequent soil’s driving effect to the front would be weakened, or even the flow would be interrupted, so the vertical rising speed of the soil would be reduced. When the rotational speed of the auger was constant, the efficiency of conveying-soil decreased with the increase of slope and then slightly increased. With the increase of slope, the time of slope cutting process increased, and there was more soil backfilling on the side of high altitude, which leaded to the reduction of soil discharge efficiency. However, with the increase of slope, the amount of soil slide at the pit mouth was increased, improving the efficiency of soil discharge. Further analysis demonstrated that the response surface for Y1 changed more rapidly in the direction of the rotating speed than in that of the slope angle, indicating that the rotating speed of auger X4 had a more significant influence than the slope angle X1.As can be seen in Fig. 8b, when the helix angle of the auger was fixed, the efficiency of conveying-soil continued to increase with the increase of the rotation speed. When the rotating speed of auger was fixed, the efficiency of conveying-soil increased with the increase of the helix angle and tends to decrease when it reached a certain value. The spiral blades space was the channel of soil movement. This phenomenon was caused by the increase of the gap between the two spiral blades with the increase of the helix angle of the auger, the soil was not easy to produce blockage. Meanwhile, the movement distance of soil was shorter, and the soil with higher kinetic energy was discharged more quickly from the pit. When reaching the pit mouth, the angle of soil throwing was larger and the soil backfilling rate was reduced. However, if the helix angle of auger was too large, the upward support ability and friction of the spiral blade surface to the soil would be reduced. Further analysis demonstrated that the response surface for Y1 changed more rapidly in the direction of the helix angle than the rotating speed of the auger, indicating that the helix angle of the auger X2 had a more significant influence than the rotating speed of the auger X4.When the feeding speed was fixed, the efficiency of throwing-soil continued to increase with the increase of the rotating speed. When the rotating speed of auger was fixed, the efficiency of the throwing-soil with the increase of the feeding speed (see in Fig. 8c). The phenomenon was caused by the faster the feeding speed of the auger, the thickness of soil cut per unit time increased. Furthermore, the subsequent driving force of soil increased, and the soil kinetic energy increased. However, in the actual production, excessive feeding speed would cause soil blockage on the surface of spiral blades. The reason is due to in the simulation process, the soil would not stop moving because of blockage. Further analysis demonstrated that the response surface for Y1 changed more rapidly in the direction of the rotating speed than in that of the feeding speed, indicating that the rotating speed of auger X4 had a more significant influence than the feeding speed X3.When the slope was fixed, the distance of the throwing-soil increased with the increase of rotation speed of the auger, and the increase amplitude increased gradually, as shown in Fig. 8d. The reason for this phenomenon was that the soil had more kinetic energy when it left its original position and the centrifugal force it received when it reaching the pit mouth is greater. When the rotation speed was too low, the soil layer was thin and the subsequent soil driving force was insufficient, resulting in the soil mass per unit area at the pit mouth was light and then the kinetic energy was small. When the rotating speed of auger was fixed, the distance of the throwing-soil increased continuously with the increase of the slope. As the slope increased, the time of soil swipe down process increased and then the rolling distance on the slope increased. Further analysis demonstrated that the response surface for Y2 changed more rapidly in the direction of the slope angle than in that of the rotating speed of auger, indicating that the slope angle X1 had a more significant influence than the rotating speed X3.Comprehensive optimal designAs relative importance and influencing rules of various experimental factors on evaluation indexes were different from each other, evaluation indexes should be taken into comprehensive consideration30. The optimization equation is obtained by the Design-Expert software multi-objective optimization method with Y1 and Y2 as the optimization objective function.$$25le {X}_{1}le 45$$$$10le {X}_{2}le 22$$$$0.04le {X}_{3}le 0.1$$$$30le {X}_{4}le 120$$$${{Y}_{1}}_{mathrm{max}}({X}_{1},{X}_{2},{X}_{3},{X}_{4})$$$${{Y}_{2}}_{min}({X}_{1},{X}_{2},{X}_{3},{X}_{4})$$In practice, the best combination of parameters needs to be selected according to the terrain slope. When the slope was fixed, the Design-Expert software was applied to optimize and solve the above mathematical model. The optimal combination of working parameters affecting the efficiency of conveying-soil Y1 and distance of throwing-soil Y2 for the auger were obtained and are shown in Table 6. If the ground preparation was required before the digging operation, the digging parameters can be designed according to values of Group 6 in Table 6.Table 6 Optimal parameter combinations of several terrain slopes.Full size tableDisturbance of soilA soil disturbance is defined as the loosening, movement and mixing of soil caused by an auger passing through the soil16. In the interface of the EDEM Analyst, add a “Clipping plane” to show the movement of the auger inside the pit. The kinetic energy, soil particle velocity vector, and velocity value of soil particles is observed when the auger in the middle of the soil bin31,32, as shown in Fig. 9.Figure 9The disturbance of the soil effect by spiral blade.Full size imageThe soil was lifted to the surface and then dropped to the lower side. In addition to the volume occupied by the spiral blades, the disturbed area also included the out-of-pit disturbed area caused by the compression of the cutting end of the spiral blade, as shown in the lower left corner of the auger.The kinetic energy and velocity of soil decreased firstly and then increased along the opposite direction of the auger feeding. The cutting end of the auger and the soil-throwing section occurred in the region with high kinetic energy and velocity. This was because the maximum kinetic energy was obtained at the cutting end of the auger, which was gradually consumed in the process of rising. After reaching the dumping end, the soil lost the restraint of the pit wall. When the centrifugal force of soil lost the reaction force, the kinetic energy of soil increased. Too much kinetic energy, however, can cause the soil to spread too far, causing subsequent trouble. The kinetic energy of the soil at the cutting end was related to the rotational speed of the auger. The spiral angle affected the angle between the force and gravity, and then the kinetic energy consumption in the process of soil increased.Verification experimentsTo verify the accuracy of the optimization model for auger working, as well as to evaluate the rationality of the working parameter combination optimized by the virtual experiment, performance verification tests were carried out on the EDEM software. According to the optimized process parameter setting test (as shown in Table 6), the relative error between the theoretical value and the experimental value was obtained. The verification test results are summarized in Table 7. The average relative errors of the efficiency of conveying-soil and the distance of throwing-soil between the Theoretical value and text value were only 4.4%, 9.1%. The simulation model is fairly accurate. The field performance verification experiments were carried out in slope. Figure 10 illustrates the field test and working conditions.Table 7 Results and comparison of validation test.Full size tableFigure 10Operation diagram at the experiment site.Full size image More