Find research datasets worth reusing

University of Turin IRIS-registered Publications for co-authorship networks

Data extracted from the University of Turin (UNITO) IRIS publication database (available at <a href="https://iris.unito.it/">www.iris.unito.it</a>) regarding publication authorship. This data was generated in order to produce co-authorship networks of authors inside of UNITO. The JSON-formatted dataset includes data regarding UNITO affiliated authors that have published from the year 2012 to roughly September 2023, along with a list of all of their publications in the same time period.The dataset encompasses 15807 authors and 66313 articles. The JSON has the following structure:<ul><li>`authors` (list): A list of objects representing authors, each with the following structure:<ul><li>`name` (string): The author's given name, all in lowercase letters. This field is always populated;</li><li>`surname` (string): The author's family name, all in lowercase letters. This field is always populated;</li><li>`affiliation` (string): The string "university of turin";</li><li>`department` (string or NULL): Empty (`null`) or with a string indicating the author's current (as of data collection, roughly September 2023) work department;</li><li>`id` (string): Unique UUID4 of the author.</li></ul></li><li>`papers` (list): A list of object representing published articles, each with the following structure:<ul><li>`id` (string): Unique IRIS ID of the publication. This field is always populated;</li><li>`title` (string): Title of the publication. This field is always populated;</li><li>`year` (numeric): Year of the article's publication date. This field is always populated;</li><li>`authors` (list): A non-empty list of strings. Each string is one ID of one of the authors in the `authors` list.</li></ul></li></ul>The data was kindly provided by the IRIS office in September 2023 and preprocessed by Luca Visentin to a digestible JSON.

opencc-zeroNov 2023View details →

Supporting information for: Accounting for the topology of road networks to better explain human-mediated dispersal in terrestrial landscapes

Human trade and movements are central to biological invasions worldwide. Human activities not only transport species across biogeographical barriers but also accelerate their post-introduction spread in the landscape. Thus, by constraining human movements, the spatial structure of road networks might greatly affect the regional spread of invasive species. However, few invasion models have accounted for the topology of road networks so far, and its importance for explaining the regional distribution of invasive species remains mostly unexplored. To address this issue, we developed a spatially explicit and mechanistic human-mediated dispersal model that accounts and tests for the influence of transport networks on the regional spread of invasive species. Using as a model the spread of the invasive ant Lasius neglectus in the middle Rhône valley (France), we show that accounting for the topology of road networks improves our ability to explain the current distribution of the invasive ant. In contrast, we found that using human population density as a proxy for the frequency of transport events decreases models' performance and might thus not be as appropriate as previously thought. Finally, by differentiating road networks into sub-networks, we show that national and regional roads are more important than smaller roads for explaining spread patterns. Overall, our results demonstrate that the topology of transport networks can strongly bias regional invasion patterns and highlight the importance of better incorporating it into future invasion models. The mechanistic modelling approach developed in this study should help invasion scientists explore how human-mediated dispersal and topography shape invasion dynamics in landscapes. Ultimately, our approach could be combined with demographic, natural dispersal and environmental suitability models to refine spread scenarios and improve invasive species monitoring and management at regional to national scales.

opencc-zeroNov 2023View details →

Simulated results from an agent-based model examining inequality and innovation in social networks

Theories of innovation often balance contrasting views that either smart people create smart things or smartly constructed institutions create smart things. While population models have shown factors including population size, connectivity, and agent behavior as crucial for innovation, few have taken the individual-central approach seriously by examining the role individuals play within their groups. To explore how network structures influence not only population-level innovation but also performance among individuals, we studied an agent-based model of the Potions Task, a paradigm developed to test how structure affects a group's ability to solve a difficult exploration task. We explore how size, connectivity, and rates of information sharing in a network influence innovation and how these have an impact on the emergence of inequality in terms of agent contributions. We find, in line with prior work, population size has a positive effect on innovation, but that large and small populations perform similarly per capita; that many small groups outperform fewer large groups; that random changes to structure have few effects on innovation; and that the highest performing agents tend to occupy more central network positions. Moreover, we show that every network factor which facilitates innovation leads to a proportional increase in inequality of performance, creating "genius effects" among otherwise "dumb" agents in both idealized and real-world networks.

Additional resources for "Day ahead electricity price forecasting with neural networks - one or multiple outputs?"

Open the record for dataset details and reuse information.

Application of artificial neural network to forecast indoor air temperature in a building with artificial ventilation: impact of early stopping.

Indoor air temperature prediction can facilitate energy-saving actions without compromising the indoor thermal comfort of occupants. The aim of this study was to analyse the performance of various artificial neural networks with a view to proposing an optimal approach for predicting the indoor temperature of a tertiary building with artificial ventilation. The MLP, CNN, LSTM models and the CNN-LSTM combination (long short-term memory network) were used and coupled with the optimisation algorithms (Adam, SGD) and the independent hyper-parameters early stopping and dropout. The parameters used are outdoor ambient temperature, outdoor relative humidity, indoor relative humidity, wet bulb temperature, black globe temperature and mean radiant temperature. The data is collected in an artificially ventilated building in Yaoundé, Cameroon. A numerical code was developed in Python to run the simulations. In order to study the impact of the parameters on the prediction, two scenarios were distinguished in this work: (1) all the parameters are input to the network, (2) only the parameters whose absolute value of the correlation coefficient was greater than or equal to 0.5 were used. The impact of early stopping is assessed by distinguishing two case studies: the first without early stopping, the second with early stopping. The results showed that without early stopping, the MLP, CNN, LSTM and CNN-LSTM networks are adequate for predicting the temperature with the second scenario, mainly with both the SGD and Adam algorithms, and CNN-LSTM is the most appropriate model because the MSE and MAE values obtained in this case were closer to 0. With early stopping, the learning time is reduced and the learning curves are improved; the models optimised better with the SGD algorithm in general, but the best neural network model was obtained with the Adam algorithm and the LSTM network for the performances MSE=0.0005, MAE=0.0130 with the second scenario.Keywords: prediction, indoor temperature, artificial neural network, early stopping, artificially ventilated building.

Vesuvio cGNSS Network - Rinex data quality control (2020)

For each cGNSS station, the file contains a summary report with information about Rinex observation data. For each day, the summary line shows the following information:<ol><li>cGNSS station name (Name),</li><li>the start time of the window, the time format is year (Y), month (M), day (D), hour and minutes (Hour), day of year (DOY), modified Julian date (M J Date) and GNSS week (Week),</li><li>the end time of the window, the time format is year (Y), month (M), day (D), hour and minutes (Hour), day of year (DOY), modified Julian date (M J Date), and GNSS week (Week),</li><li>the start and end times of the window (time format is year month day hour min),</li><li>window time laps (Hrs),</li><li>observation interval (OI),</li><li>the number of possible observations (#expt) above the elevation mask,</li><li>the number of complete observations (#obs),</li><li>the ratio of complete to possible observations as a percent (DCP),</li><li>the RMS "multipath combinations" values MP1 and MP2, in meters, limited by the elevation mask (MP1, MP2) rounded to two decimal points,</li><li>cycle slips (CS),</li><li>the ratio of complete observations to cycle slips (obs/CS).</li></ol>A full description of cGNSS network is reported in: - De Martino P, Dolce M, Brandi G, Scarpato G, Tammaro U (2021). The Ground Deformation History of the Neapolitan Volcanic Area (Campi Flegrei Caldera, Somma–Vesuvius Volcano, and Ischia Island) from 20 Years of Continuous GPS Observations (2000–2019). Remote Sensing. 13(14):2725. doi:10.3390/rs13142725.Please cite this when using the dataset

Campi Flegrei cGNSS Network - Rinex data quality control (2020)

opencc-zeroNov 2023View details →

Data from: Clots reveal anomalous elastic behavior of fiber networks

The adaptive mechanical properties of soft and fibrous biological materials are relevant to their functionality. The emergence of the macroscopic response of these materials to external stress and intrinsic cell traction from local deformations of their structural components is not well understood. Here, we investigate the nonlinear elastic behavior of blood clots by combining microscopy, rheology, and an elastic network model that incorporates the stretching, bending, and buckling of constituent fibrin fibers. By inhibiting fibrin crosslinking in blood clots, we observe an anomalous softening regime in the macroscopic shear response as well as a reduction in platelet-induced clot contractility. Our model explains these observations from two independent macroscopic measurements in a unified manner, through a single mechanical parameter, the bending stiffness of individual fibers. Supported by experimental evidence, our mechanics-based model provides a framework for predicting and comprehending the nonlinear elastic behavior of blood clots and other active biopolymer networks in general.

Keyterm Co-Occurence Network

A bibliometric map displaying various nodes (word items) and the links by which they are connected conceived on VOSviewer. In this publication, we focused on words indicative of deficit thinking (at risk, disorder, risk, deficit, psychopathology, impairment, abnormality, difficulty, problem, risk factor, delay, dysfunction, cognitive impairment, conduct disorder, brain abnormality, cognitive deficit, failure, anomaly, attention problem, behavior problem, cognitive dysfunction, environmental risk factor, neuropsychological deficit, executive dysfunction). This map can be viewed and explored by uploading the attached json file on app.vosviewer.com. Readers can evaluate specific nodes by using the "find" feature on the left-handside menu.

Robustness assessment of a C++ implementation of the LeNet-5 convolutional neural network.

The architecture of the LeNet-5 convolutional neural network (CNN) was defined by LeCun in its paper "Gradient-based learning applied to document recognition" (https://ieeexplore.ieee.org/document/726791) to classify images of hand written digits (MNIST dataset).This architecture has been customized to use Rectified Linear Unit (ReLU) as activation functions instead of Sigmoid.It consists of the following layers:<ul><li>conv1: Convolution 2D, 1 input channel (28x28), 3 output channels (28x28), kernel size 5, stride 1, padding 2.</li><li>relu1: Rectified Linear Unit (3@28x28).</li><li>max1: Subsampling buy max pooling (3@14x14).</li><li>conv2: Convolution 2D, 3 input channels (14x14), 6 output channels (14x14), kernel size 5, stride 1, padding 2.</li><li>relu2: Rectified Linear Unit (6@14x14).</li><li>max2: Subsampling buy max pooling (6@7x7).</li><li>fc1: Fully connected (294, 147)</li><li>fc2: Fully connected (147, 10)</li></ul>The fault hypotheses for this work include the occurrence of:<ul><li>BF: single bit-flip faults</li><li>S0: single, double-adjacent and triple-adjacent stuck-at-0 faults</li><li>S1: single, double-adjacent and triple-adjacent stuck-at-1 faults</li></ul>In the memory cells containing all the parameters of the CNN:  <ul><li>w: weights (float32)</li><li>b: biases (float32)</li></ul>Images 200 to 249 from the MNIST dataset have been used as workload.This dataset contains the raw data obtained from running exhaustive fault injection campaigns for all considered fault models, targeting all considered locations and for all the images in the workload.<h3>Files information</h3><ul><li>golden_run.csv: Prediction obtained for all the images considered in the workload in the absence of faults (Golden Run). This is intended to act as oracle to determine the impact of injected faults.</li><li>single_faults/bit_flip folder: Prediction obtained for all the images considered in the workload in presence of single bit-flip faults. There is one file for each parameter of each layer.</li><li>single_faults/stuck_at_0 folder: Prediction obtained for all the images considered in the workload in presence of single stuck-at-0 faults. There is one file for each parameter of each layer.</li><li>single_faults/stuck_at_1 folder: Prediction obtained for all the images considered in the workload in presence of single stuck-at-1 faults. There is one file for each parameter of each layer.</li><li>double_adjacent_faults/stuck_at_0 folder: Prediction obtained for all the images considered in the workload in presence of double adjacent stuck-at-0 faults. There is one file for each parameter of each layer.</li><li>double_adjacent_faults/stuck_at_1 folder: Prediction obtained for all the images considered in the workload in presence of double adjacent stuck-at-1 faults. There is one file for each parameter of each layer.</li><li>triple_adjacent_faults/stuck_at_0 folder: Prediction obtained for all the images considered in the workload in presence of triple adjacent stuck-at-0 faults. There is one file for each parameter of each layer.</li><li>triple_adjacent_faults/stuck_at_1 folder: Prediction obtained for all the images considered in the workload in presence of triple adjacent stuck-at-1 faults. There is one file for each parameter of each layer.</li></ul><h3>Methodology information</h3>First, the CNN was used to classify all the images of the workload in the absence of faults to get a reference to determine the impact of faults. This is golden_run.csv file.After that, one fault injection experiment was executed for each of the 16 most significant bits bit of each element of each parameter of the CNN, as previous fault injection experiments showed that the occurrence of the considered faults in the 16 least significant bits does not impact the behaviour of the network.Each experiment consisted in:<ul><li>Affecting the bits (inverting it in case of bit-flip faults, setting it to 0 or 1 in case of stuck-at-0 or atuck-at-1 faults) identified by the mask.</li><li>Classifying all the images of the workload in the presence of this fault. The obtained output was stored in a given .csv file.</li><li>Removing the fault from the CNN by restoring the affected bits to its previous value.</li></ul><h3>List of variables (Name : Description (Possible values))</h3><ul><li>IMGID: Integer number identifying the considered image (200-249).</li><li>TENSORID: Integer number identiying the parameter affected by the fault (0 - No fault, 1 - conv1.w, 2 - conv1.b, 3 - conv2.w, 4 - conv2.b, 5 - fc1.w, 6 - fc1.b, 7 - fc2.w, 8 - fc2.b).</li><li>ELEMID: Integer number identiying the element of the parameter affected by the fault (-1 - No fault, [0-2] - conv1.b, [0-74] - conv1.w, [0-5] - conv2.b, [0-149] - conv2.w, [0-146] - fc1.b, [0-43217] - fc1.w, [0-9] - fc2.b, [0-1469] - fc2.w).</li><li>MASK: 8-digit hexadecimal number identifying those bits affected by the fault ([00000000 - No fault, FFFFFFFF - all 32 bits faulty]).</li><li>FAULT: String identiying the type of fault (NF - No fault, BF - bit-flip, S0 - Stuck-at-0, S1 - Stuck-at-1).</li><li>SOFTMAX: 10 decimal numbers obtained after applying the softmax function to the provided output. They represent the probability of the image of belonging to the corresponding category for classification.</li><li>PRED: Integer number representing the category predicted for the processed image.</li><li>LABEL: integer number representing the actual category for the processed image.</li></ul>

Combined bioinformatics and machine learning methodologies reveal prognosis-related ceRNA network and propose ABCA8, CAT, and CXCL12 as independent protective factors against osteosarcoma

Supplementary Table 1. Basic traits of the seven microarray datasets from the Gene Expression Omnibus and The Cancer Genome Atlas.Supplementary Table 2. Basic characteristics of the nine differentially expressed circRNAsSupplementary Table 3. Index of concordance (C-index) and variance inflation factor (VIF) of ABCA8, CXCL12, and CAT.Supplementary Table 4. LASSO and cox analysis of ceRNA with coef/se(coef) < 0.01 and P-value of proportional hazards assumption (PH) > 0.05.Supplementary Table 5. Robust rank aggregation analysis of ABCA8, CXCL12, and CAT. LogFC in the four datasets and RRA score of the three RNAs.Supplementary Figure 1. Competitive endogenous RNA in osteosarcoma.Supplementary Figure 2. Protein–protein interaction (PPI) network of genes in the competitive endogenous RNA networkSupplementary Figure 3. Proportional hazards assumption (left) and linearity assumption (right).Supplementary Figure 4. Survival analysis for competitive endogenous RNA in osteosarcoma.

Data and code from Lamb et al., "Evaluating conservation units using network analysis: a sea duck case study"

This file consists of data and code used to construct network models for scoters in North America and is associated with the manuscript "Evaluating conservation units using network analysis: a sea duck case study" published in Frontiers in Ecology and the Environment.  Continental scoter network mapping is the R script used to run analyses. duck_nodes is the main datafile. Columns are organized as follows:id - unique identifierspecies - species from which the centroid was obtained (BLSC = black scoter, SUSC = surf scoter, WWSC = white-winged scoter)stage - period of the annual cycle to which the centroid belongs (W = winter, B = breeding, S = spring staging, M = fall staging and molt, WM = winter migration, BM = breeding migration, MM = molt migration, SM = spring migration)site - position of centroid within season (i.e., W1 = first site occupied during winter, W2 = second site occupied, etc.)cycle - number of annual cycles following transmitter attachment (1 = first cycle after attachment, 2 = second cycle after attachment, etc.)sex - sex of individual (M = male, F = female)age - age of individual (HY = hatch year, SY = second year, TY = third year, ASY = after second year, ATY = after third year, AHY = after hatch yearcapture_reg - general area where individual was capturedcapture_subreg - specific region within capture region where individual was capturedlon - longitude of centroidlat - latitude of centroidduration - number of days spent at centroidstart - date of arrival at centroidend - date of departure from centroidjstart - Julian date of arrival at centroidjend - Julian date of departure from centroidseason - season of annual cycle in which centroid occurred (W = winter, F = fall, B = breeding, S = spring)year - calendar year in which centroid beganto - node in which centroid is groupedfrom - node in which previous centroid is grouped (i.e., node in which indiviual was located before moving to present node)to_sea - season of annual cycle in which centroid occurredfrom_sea - season of annual cycle in which previous centroid occurredtype - movement type to centroid; the first letter represents the season (coded as in "season" column), and the second represents the nature of the movement (WD = dispersal within a season, M = migration among seasons)type2 - same as "type", but with dispersal movements coded by the stage in which they occur (W = winter, B = breeding, SM = spring migration, WM = winter migration)ew - capture location in eastern (east; Atlantic and Great Lakes) or western (west; Pacific) North Americacount_ind_sp - total number of tracked individuals of the species represented by centroidwt - base centroid weight (all centroids equal, deployments excluded)wt_sp - species-adjusted centroid weight: for centroid x in species s, weightx = (N centroids) × (1 / (n centroids in s))wt_dur - duration-adjusted centroid weight: for centroid x, weightx = (days at centroid location) × 365-1wt_ind -  individual-adjusted centroid weight: for centroid x in individual j, weightx = 1 / (n centroids in j)wt_ind_sp - individual and species adjusted centroid weight: for centroid x, individual j, and species s, weightx = (N centroids / (N species * n individuals in s)) × (1 / (n centroids in j))wt_cap - capture location adjusted centroid weight: for centroid x and capture location c, weightx = (N centroids / N capture locations) / n centroids in cwt_ew - east-west adjusted centroid weight: for for centroid x and region r, weightx = (N centroids / N regions) / n centroids in rwt_spew - species and east-west adjusted centroid weight: for centroid x species s, and region r, weightx = (N centroids / (N species × N regions)) / n centroids for species s in region rwt_indspew - individual, species, and east-west adjusted centroid weight: for centroid x, individual j, species s, and region r, weightx = N centroids / (N species × N regions × n centroids for individual j × n individuals for species s in region r)wt_spewcap - species, east-west, and capture location adjusted centroid weight: for centroid x, species s, capture location c, and region r, weightx = N centroids / (N species × N regions × n centroids for species s in capture location c × n capture locations for species s in region r)

Data from: Network-level changes in the brain underlie fear memory strength

The strength of a fear memory significantly influences whether it drives adaptive or maladaptive behavior in the future. Yet, how mild and strong fear memories differ in underlying biology is not well understood. We hypothesized that this distinction may not be exclusively the result of changes within specific brain regions, but rather the outcome of collective changes in connectivity across multiple regions within the neural network. To test this, rats were fear conditioned in protocols of varying intensities to generate mild or strong memories. Neuronal activation driven by recall was measured using cfos immunohistochemistry in 12 brain regions implicated in fear learning and memory. The interregional coordinated brain activity was computed and graph-based functional networks were generated to compare how mild and strong fear memories differ at the systems level. Our results show that mild fear recall is supported by a well-connected brain network with small-world properties in which the amygdala is well-positioned to be modulated by other regions. In contrast, this connectivity is disrupted in strong fear memories and the amygdala is isolated from other regions. These findings indicate that the neural systems underlying mild and strong fear memories differ, with implications for understanding and treating disorders of fear dysregulation.

opencc-by-4.0May 2022View details →

Designing Optimal Convolutional Neural Network Architecture Using Differential Evolution Algorithm

Convolutional Neural Networks (CNNs) are widely used deep learning models for solving various tasks such as computer vision, speech recognition, among others. However, CNNs are developed manually based on problem-specific domain knowledge and tricky settings, which are laborious, time-consuming and challenging. To address these issues, this study proposes an Improved Differential Evolution of Convolutional Neural Network algorithm, namely IDECNN, to design CNN layer architectures for image classification task.

opencc-by-4.0Oct 2021View details →

Keynote: Bringing Reinforcement learning Into Radio Light Network for Massive Connections

<blockquote>3GPP standardization has been progressing at an astonishingly rapid phase, where Release 15 and Release 16 have set the foundations of the 5G system, while Release 17 provides enhancements and optimizations to enable support for further use cases. In parallel to 5G standardization efforts, several initiatives worldwide endeavour to drive and support the evolution of smart networks and services. Among others Europe is establishing the Joint Undertaking on Smart Networks and Services in the frame of the Horizon Europe programme for research and innovation. Other initiatives are complementing the European initiative, such as Secure 5G & Beyond Act in the U.S., roadmap towards 6G in Japan, MSIT 6G programme in S. Korea, and MIIT 6G programme in China.</blockquote><blockquote>The workshop will provide an opportunity for reflection and discussion about requirements and architectural considerations for future generations of mobile systems. The focus will be on presenting version 4.0 of the Architecture white paper developed by the 5G PPP architecture working group. It will allow to move from 5G and beyond towards a fully-fledged 6G architecture.</blockquote>

Patterns of liana diversity and host interaction networks in selectively-logged and unlogged forests of Uppangala, Western Ghats, India

Lianas shape tropical forest species composition, structure, and dynamics. Increasing climate fluctuation and anthropogenic disturbances increase liana abundance. Despite the increasing number of liana studies in India, only a few have examined the distribution and association of hosts with lianas, or liana-host interaction networks to determine their functional significance and conservational value. Therefore, our objective was to fill the knowledge gap about the diversity, abundance, and network structure of liana-host interactions in response to logging disturbance in a wet evergreen forest of Uppangala in central Western Ghats, India. We sampled lianas ≥1 cm in diameter at 1.3 m from the base and their host trees in thirty 20m x 20m plots in selectively-logged and unlogged forest management regimes. We evaluated liana-host tree interactions in logged and unlogged forests and retrieved community-level measures (nestedness, connectance, modularity, and network specialization index) and species-level indicators (species specialization index). Diversity and abundance of liana species were considerably greater in the selectively logged forest site. The logged forest site had compartmentalization, anti-nestedness, and network specialization, while unlogged forests were not showing any significant network structure. Most species of lianas and hosts were peripherals, but others were structurally important (connectors, module hubs, and network hubs) in the two forest sites. Forest management regimes had distinct structurally significant species.

Taxonomically revised dataset of acanthomorphic acritarch species of the Doushantuo Formation: dataset for rarefaction, NMDS, and network analysis

This is supplementary material for the article entitled "Silicified microfossils from the Ediacaran Doushantuo Formation along a shelf margin-slope-basin transect in Hunan Province, South China, with stratigraphical implications" by Ouyang et al. The dataset contains (1) taxonomic revisions of published acanthomorph specimens from the Doushantuo Formation from 55 previous publications that provide clear, pulished, microfossil images and clearly designated stratigraphic horizons; and updated occurrence data of Doushantuo acanthomorphic acritarchs and certain sphaeromorphic taxa considered as stratigraphically useful (i.e., Schizofusa zangwenlongii Grey, 2005) based on the revised taxonomy, with each occurrence assigned to a specific fossil collection; (2) acritarch relative abundance data from this and six previous publications for rarefaction analysis in this study; (3) presence/absence data of the Doushantuo acritarchs (based on occurrence data in sheet 1) for NMDS and network analysis with R in this study; (4) species abbreviations in (3); (5) species loadings generated by the NMDS analysis and plotted in Fig. 41.3 of this study.

Data from: Latest Ordovician (Hirnantian) brachiopod faunal lists used for non-matric multidimensional scaling (NMDS) and network analyses

A total of 107 brachiopod genera of Hirnantian age among 42 localities worldwide are compiled into a binary dataset (Table S1; presence =1, absence = 0). The majority of the faunal lists was derived from the well-screened Hirnantian brachiopod faunal data of Rong et al. (2020). In this study, the Hirnantian faunal lists are updated for the following localities: Anticosti Island, eastern Canada; Edgewood region, American Mid-Continent; Mackenzie Mountains, northwestern Canada. Detailed discussions on these faunal update and references are provided in the main paper (section on Paleobiogeography of the Mackenzie Mountains Hirnantian fauna).

Unraveling the cavity-nesting network at large spatial scales: The biogeographic role of woodpeckers as ecosystem engineers

Aim: Cavities are usually a limiting resource for several forest-dwelling vertebrates, with effects that propagate through ecological networks. Although diverse assemblages of primary excavators (e.g., woodpeckers) are assumed to increase cavities, other forest resources can also limit populations of primary excavators and cavity users, thus undermining the ecological role of excavators over different scales. We aim to test the biogeographical-scale relationships between primary excavators and cavity users by distinguishing the contribution of forest characteristics. Location: Southern South America Methods: We used species distribution models, which combine bioclimatic and remote sensing derived variables, to map the richness of vertebrates composing the cavity network of temperate and Mediterranean forests of South America. Based on a resampling procedure for ensuring spatial independence, we fitted structural equation models to estimate causal relationships between forest characteristics and cavity-user vertebrates. Results: Secondary cavity users (obligated, habitat generalists and forest specialists) were positively and strongly influenced by the richness of primary excavators, while mammal richness was more influenced by tree richness. The richness of trees and Nothofagus tree species influenced positively the richness of primary excavators and secondary cavity users. Canopy height and net primary productivity affected positively secondary cavity users. Main conclusions: Our results confirm the role of primary excavators as ecosystem engineers but highlight the importance of considering large spatial scales when analyzing cavity-nesting networks. Biogeographical patterns of tree diversity and forest structure can be important drivers of cavity-nesting networks that remain hidden when studies are conducted over fine spatial scales.