Biodiversity Informatics, 19, 2025, pp. 7-32 7 STOPOVER HOTSPOTS FOR MIGRATORY BIRDS IN NORTH AND CENTRAL AMERICA Shi Feng1, Qinmin Yang1*, Huijie Qiao2*, Luis E. Escobar3, Xuan Yan4 1State Key Laboratory of Industrial Control Technology, College of Control Science and Engineering, Zhe- jiang University, Hangzhou, 310007, PR China. 2State Key Laboratory of Animal Biodiversity Conservation and Integrated Pest Management, Institute of Zoology, Chinese Academy of Sciences, Beijing, 100101, PR China. 3Department of Fish and Wildlife Conservation, 1015 Life Science Cir, Virginia Tech, Blacksburg, VA 24061, USA. 4Independent researcher, 7912 Heritage Palms TRL, Mckinney, TX 75070, USA. Abstract. Despite the large body of literature on avian migratory behavior, there is little information about stopover sites during bird movement, including the population-level drivers of breeding grounds and wintering grounds. Stopovers play an essential role in bird migratory site chains for energy supply and rest. There is an urgent need to identify and protect stopover sites to secure the long-term sustainability of migratory network connectivity and stability. To address this challenge, we reconstructed a migration network and identified geo- graphic hotspots denoted as stopover sites. And we analyzed the high-density population movements of 52 focal migratory bird species using comprehensive observation data from eBird through PageRank algorithm. Furthermore, potential alternative stopover sites were explored using a word embedding technique based on geo-functional similarity. Our study was conducted in North and Central America during a three-year period and revealed three key stopover areas, including Florida peninsula and its inland, the region of Central America, and the region near Puget Sound. Results from this study can be used for conservation prioritization guidance, active surveillance of bird pathogens, and bird management. Keywords: Central America, bird conservation, eBird, migration network, stopover Introduction Understanding migration pathways is critical for bird management and conservation. Migratory birds travel thou- sands of miles between breeding and wintering grounds, facing numerous threats along their routes, including habitat loss, climate change, and hunting (Nemes et al., 2023). By analyzing migration pathways, conservationists can identify key stopover sites and habitats that are essential for bird survival. Information about key stopover sites enables the cre- ation of targeted protection strategies, such as establishing protected areas, restoring habitats, and implementing interna- tional agreements to safeguard migratory routes (Higuchi et al., 2012). Additionally, migration pathway studies can help predict the impacts of environmental changes on bird populations for proactive measures to mitigate potential threats (La Sorte et al., 2016) and to implement precision epidemiology of avian influenza. Usually, migration depends on a suit of interconnected sites (Runge et al., 2015), and their conservation requires a deep understanding of the connectivity in migration networks. That is, untangling migration patterns requires to answer how migratory species connect their breeding and non-breeding grounds through their trajectories. Migratory connectiv- ity can describe the spatiotemporal link of individuals and populations between sites caused by migratory movements, which can influence both long-term evolutionary responses and short-term population dynamics (Webster et al., 2002). Research on bird migratory connectivity generally focuses on geographical patterns that describe linkages between breeding and wintering grounds and the influencing drivers, such as resource requirements and ecological relationships (Kramer et al., 2018). Stopover sites, such as wetlands, forest fragments or grasslands, are essential for birds to rest and *Corresponding authors: Qinmin Yang, Email: qmyang@zju.edu.cn, Huijie Qiao, Email: qiaohj@ioz.ac.cn mailto:qmyang@zju.edu.cn Shi Feng et al. – Stopover Hotspots in North and Central America 8 refuel during their migration journey. These sites are con- sidered to be key nodes in migration network connectivity and require targeted attention (Guo et al., 2024). Tradi- tionally, migration connectivity research relies on tracking techniques. For example, Knight et al. (2018) constructed a migration network for Tree Swallow (Tachycineta bicol- or) with 133 geolocators to assess the spatial connection between wintering and breeding areas to help develop op- timal conservation strategies in North America. Similarly, Xu et al. (2020) applied high-resolution GPS tracking data of 10 whooper swans (Cygnus cygnus), 81 swan geese (An- ser cygnoides), 93 bar-headed geese (Anser indicus), and 54 greater white-fronted geese (Anser albifrons) with GPS loggers during 2005-2018 to study the migration network connectivity in the Central and East Asian-Australasian Flyways. Lastly, Catry et al. (2024) followed the migra- tory trajectories of 20 grey plovers (Pluvialis squatarola) with tracking devices to highlight important stopover sites and potential bottlenecks. These studies focused on the individual-level flyway and network, limiting the insight into population-level movements and being restricted to tracking costs and bird body size. Tracking a few indi- viduals is not suitable for reconstructing the movement of a species. Similarly, individual tracking covers limited species due to the difficulty of capturing and carrying the track devices for small-sized ones. Based on this data limitation of animal movement, broad-scale monitoring, such as radar or citizen science data, can provide comprehensive data for wider views of signals about the biogeography of bird movement. For ex- ample, Bonter et al. (2009) identified critical stopover sites with weather-surveillance radar images from 2000 to 2001 in the Great Lakes basin. Guo et al. (2024) applied five years of weather surveillance radar data to map the stop- over densities of land birds during spring and autumn mi- gration across the eastern United States. Radar data, how- ever, fail to identify specific bird species and only provide information on total bird biomass. This coarse-scale esti- mation limits conservation and management strategies for specific bird species, and is subjected to the radar coverage region. Ai et al. (2024) developed a portable stereo vision observer for bird flocks in field scenarios based on feature and sensor methods. In their study, Ai et al. captured birds natural flocking behaviors such as foraging and conver- gent flying in mudflats and seashores, which is helpful for stopovers detection. Nevertheless, observer activities are assumed to be scenario-oriented and can be disturbed by temperature and wind, and this observer focuses on short- term movement of specific individuals within the observa- tion site, failing to capture dynamics of the long-distance migration process. eBird (https://ebird.org/) is the world’s largest bird- watching data repository and has comprehensive informa- tion on the distribution and movement patterns of avian species at the population level. Lin et al. (2020) used eBird data to discover Priority Stopover Sites (PSSs) and quan- tified the potential benefit for resident species in terms of species abundance focusing on three North American countries, including Canada, the United States (US), and Mexico. Nicol et al. (2023) proposed a hidden semi-Mar- kov model to infer crucial stopover nodes based on count data from eBird to estimate the most likely migration links among regions for an endangered shorebird named Far Eastern curlew (Numenius Madagascariensis) in the East Asian-Australasian Flyway. The current literature on stopover sites, however, evaluates the importance of sites based on population abundance or potential distributions combining the site protection status in isolation, but not from a systematic perspective. The spatial and temporal information of linkages among stopover sites for multiple species within a migratory network is usually overlooked. For example, Zhang et al. (2023) explored the backbone nodes of migration network among different regions us- ing data from the Global Biodiversity Information Facility (GBIF, https://www.GBIF.org/) for 1862 species in 26 bird orders. In their study, Zhang et al identified the relative importance of nodes by betweenness (Wang et al., 2008) which is the number of shortest paths passing through the individual nodes in a network. The Zhang et al study, how- ever, focused on the role that nodes play in the shortest paths of the network rather than paths across the network. Therefore, methods to comprehensively reconstruct over- all stopover site connection are needed for biologically realistic migration network analysis (Wyborn and Evans, 2021). PageRank algorithm and word embedding tech- nique can help address the need of more accurate network modeling methods for stopover detection. PageRank was originally used for webpage ranking in Google (Page et al., 1999), where PageRank considers the number of connections to a webpage (represents a node in the users’ clicking network). PageRank emphasizes the position and link relationship of nodes in the overall structure of the network. It measures node importance by analyzing the relationship between nodes, accounting for multiple indicators such as linkage quantity and quality, link distribution, damping factor, and initial PageRank, among others. Through the comprehensive calculation of these indicators, PageRank can determine the relative im- portance of nodes in the network. Usually, nodes linked to a high-PageRank node will have an increasing PageRank value and are considered more important in the network. In human society networks, PageRank values have been Shi Feng et al. – Stopover Hotspots in North and Central America 9 used to identify important user social nodes (Hong et al., 2023), and analyze urban mobility network (Wang et al., 2017). Word embedding is a natural language processing (NLP) technique used to represent words as dense vec- tors (Mikolov et al., 2013). These vectors capture seman- tic relationships and contextual information about words, allowing computers to process and understand human lan- guage more effectively. Word embedding methods have been widely used for vector representations of words in documents, social analyses, and location analyses in hu- man society (Jin et al., 2014). Mikolov et al. (2013) in- troduced word2vec algorithm to represent each word as a vector using large-scale document datasets. Words with similar contexts will have similar representations in the vector space. It is popular in the NLP applications, in- cluding machine translation (Qian et al., 2019), citation visualization (Berger et al., 2016), and sentiment analy- sis (Liu et al., 2020; Xiong et al., 2018). Meanwhile, the method has also been widely applied in human mobility data for geo-functional similarity exploration based on se- mantic similarity (Zhu et al., 2019), where a location can be defined as a word, and a set of successive and previous locations in a trajectory can be defined as its context. Con- sidering that trajectory data is a type of sequential data, exploring the function of a location through its context is similar to understanding the meaning of a word in a sen- tence. For example, Zhu et al. (2019) proposed a location representation Location2vec method based on word2vec to capture the similarity relationships among locations. Based on these, the aim of this work was to investigate the stopover hotspots during the bird migration process and identify alternative stopover sites to increase the site con- nectivity along bird migratory networks. We assessed the importance of stopover sites with the network-level link- age quantity and quality through the PageRank algorithm. This study provides a comprehensive population-level un- derstanding for migration trajectories for diverse taxa us- ing eBird data. We introduced the word embedding tech- nique to analyze trajectory sequence contexts and detected potential alternative sites based on geo-functional similar- ity. Discovery of specific stopover hotspots are expected to inform migratory bird conservation, surveillance, and management for North and Central America birds. Materials and Methods Data acquisition and preparation Bird occurrence data for 52 focal migratory bird spe- cies (Table S1) were collected from the eBird (https:// ebird.org/) citizen-science database for the period 2017- 2022. The 52 migratory species are selected from 82 focal species list (Schrimpf et al., 2021) where observation data are abundant for analysis. Data from 2017-2019 worked as the study dataset, and the data from 2020-2022 were used for network validation. Observation data covered continental North and Central America with metadata in- cluding date and location (longitude and latitude) for each record. The study was carried out in MATLAB R2021b for PageRank value calculation, and Python 3.10 for trajecto- ry estimation and word2vec analysis. The specific packag- es are shown below for the full process: pandas (version 1.4.3; Reback et al., 2022), numpy (version 1.23.5; Harris et al., 2020), datetime (version 4.4), matplotlib (version 3.5.3; Hunter, 2007), xlwt (version 1.3.0; Machin, 2017), pyproj (version 3.3.1; Whitaker, 2022), pygam (version 0.8.0; Marín, 2018), scikit-learn (version 1.1.2; Grisel et al., 2022), basemap (version 1.3.4; Whitaker, 2022), tqdm (version 4.65.0; Yorav-Raphael & da Costa-Luis, 2024), xlrd (version 2.0.1; Withers, 2020), openpyxl (version 3.1.2; Gazoni & Clark, 2023), genism(version 4.3.1; Re- hurek, 2023), and xlwings(version 0.30.6; Zumstein,2023). We randomly subsampled the original dataset to retain 100 records per day to keep the balance between computa- tion cost and information available. Migration trajectories of high-density populations for each species during each migration cycle were estimated using a minimum cost analysis (Feng et al., 2021; Somveille et al., 2021). Data were first preprocessed through mean location interpola- tion for observations of missing dates, using rolling-win- dow smoothing (Zivot and Wang, 2007), and outliers were detected and removed using Space Local Deviation Factor algorithm (Zhang and Wang, 2011). Observations were then discretized by Mean-shift clustering algorithm (Der- panis, 2005) for dense population centroids. Centroids were grouped according to the minimum cost principle and trajectories were fitted using a Generalized Additive Model (Hastie and Tibshirani, 1990). Migration trajectory sequence conversion To convert the migration trajectories into geographic cell-ID sequences, we set an origin point in position (0.1° N, 134.2° W) which was (10799.54, -14796209.33) in the Spherical Mercator map (Fig.1), and built a grid coordi- nate system for 35 x 36 cells divided by: (-14796209.33 + 200000m, 10799.54 + 200000n) where m ϵ [1,35], n ϵ [1,36] to cover the continental North and Central America (Fig.1). Then the number for each cell was calculated by: m + 35 * (n – 1). Shi Feng et al. – Stopover Hotspots in North and Central America 10 Migration network construction and stopover hotspots mining In this study, each geographic cell was defined as a node in the migration network with links resulting from cells with occurrences in adjacent positions in trajectory sequences. Node importance in the bird migration network was assessed through birds’ movement between cells with the PageRank algorithm. We set the initial PageRank val- ues for all nodes as 1/N, where N was the total number of nodes (Rogers, 2002). In order to avoid infinite itera- tion, it is usually necessary to set a convergence threshold. That is, if the PageRank value of each web page changes less than a preset threshold (0.0001 in our case), its value is considered to have converged, and the iteration can be stopped. Alternatively, by setting the maximum number of iterations as an iteration termination condition (100 in our case). The damping factor is constant with an empirical value of 0.85, which is used to simulate the movement probability between different nodes and avoid the problem of infinite loops. After setting the movement relationship between nodes, the PageRank value calculation result of each node can be obtained according to the set threshold and the maximum number of iterations. PageRank values could represent the relative importance of each node in the migration network, which was used for node ranking. The mathematical expression for PageRank algorithm is: ( ) ( ) (1 ) ( ) ( ) i i B A i B PR node PR node d d L node = − + ∑ where, PR(nodeA) is the PageRank of node A; d is the damping factor; PR ( ) iBL node are the PageRank values of nodes that link to node A; ( ) iBL node is the number of outbound links on node Bi. The cells in the grid coordinate system after geographic partition in this study were considered as nodes and the sequence of trajectories formed links between different nodes to construct a migration network. We defined nodes with the sojourn time between 5-20 days in the trajectory sequences as stopover sites based on the empirical dis- tribution of bird stopover days during migration (Kaiser, 1999; Fig.S1). This stopover period allowed us to differ- entiate stopover sites from breeding and wintering sites. Potential alternative stopover sites exploration Cell-ID sequences were analyzed using a word em- bedding technique to explore the potential alternative sites for stopover hotspots identified through PageRank algo- rithm. A cell-ID in the trajectory sequence was defined as a word, while a trajectory sequence was defined as a sentence, with multiple trajectories defined as a document corpus. This ensemble of cells allowed us to encode a mi- gratory trajectory as: tr = {cell – ID1, cell – ID2, ..., cell – IDk, k = 1,2,...} Figure 1. Schematic diagram for geographic partition by building 35×36 cells to cover continental North and Central America. The number for each cell is 1-35 in the bottom line, then 36-70 for the second line, and so on. Through the grid coordinate system after geographic partition, trajectories are presented as geographic cell-ID sequences according to daily locations. For example, the ellipse trajectory in the diagram above can be represented by a sequence from point A to point B and back to point A as follows (sojourn time ignored): 543-544-580-615-650-684-719-754-788-822- 856-891-854-818-783-748-714-679-644-610-576-542-543. Shi Feng et al. – Stopover Hotspots in North and Central America 11 All the trajectory sequences were encoded as: TR = {tr1, tr2,...trn, n = 1,2,...} Then we confined the context of word cell – IDi as: ( 1) 1 1 ( 1)( ) { , , , ,..., , }i i m i m i i i m i mcontext cell ID cell ID cell ID cell ID cell ID cell ID cell ID− − − − + + − +− = − − − − − − ( 1) 1 1 ( 1)( ) { , , , ,..., , }i i m i m i i i m i mcontext cell ID cell ID cell ID cell ID cell ID cell ID cell ID− − − − + + − +− = − − − − − − where m is the window size. Figure 2 shows a case when the window size is set as five. We applied the Skip-Gram model (McCormick, 2016), which is part of word2vec (Mikolov et al., 2013), to build word representations (embeddings) that can predict the surrounding context words for a given target word. Skip- Gram model can help capture the semantic relationships between words, discovering the geo-functional similar- ity relationship between sites in the migration network. The model takes a target word as input and predicts the previous and following words expected to appear in the context. Word prediction is conducted using a user-spec- ified window size. Model training involves adjusting the word embeddings to maximize the probability of correctly predicting the context words surrounding the target word. Here, we defined the object of the Skip-Gram model to maximize the average log probability as: log ( ( ) | ) cell ID A L p context cell ID cell ID − ∈ = − −∑ where A is all the words in the document corpus, context( cell ID− ) is the context of the word cell ID− with its size 2 1m + .We then moved a contextual window of length 2 1m + across the documents to maximize the co-occurrence probability of the words that appeared within a window. We assumed that the sequence of words was identically distributed and independent, where the probability of its contextual words was: ( ) context( ) (context( ) ) ccell ID c cell ID cell ID cell IDp cell ID cell ID p ∈− − = −− −− ∏∣ ∣ where ccell ID− is a contextual word of cell ID− . The probability of ( )ccell ID cell IDp − −∣ was calcu- lated with SoftMax function: ( )| cell ID cell ID T c c T u u W p cell I D v vD cell I v v − ∈ − =− − ∑ where , ,c ucell ID cell ID cell ID− − − are the vectors of word , ,c ucell ID cell ID cell ID− − − , W denote the set of all words. ucell ID− is one of the total words, and “T” is the transposition for the vector of ucell ID− , then the dot products of the vector cell ID− and cell ID− can be calculated. In the Skip-Gram model, the computational complexity of the output layer was high because it required calculating the probability distribution over the entire vocabulary. Here, the vocabulary represented the entire datasets of cell-IDs, which was large and in turn, time-consuming. We applied a negative sampling to selectively update model parameters (Mikolov et al., 2013), allowing the model to update only a small subset of parameters in each training step. The negative samples accelerated the training process. For the target word represented by cell ID− in the formula, we selected its contextual words context(P) as positive samples and P words that did not belong to context( cell ID− ) as negative samples (Zhu et al., 2019). A logistic regression was applied to train the model using positive and negative samples, aiming to maximize the prediction probability for positive samples while minimizing it for negative samples. We defined the object function as: ( ) ( ) L = log ( ) log(1 ( )) p T T x cell ID p cell ID cell ID A x context cell ID cell ID NEG cell ID v v v vσ σ− − − ∈ ∈ − − ∈ −   + −     ∑ ∑ ∑ Figure 2. A diagram for the context of a word with the window size is five. The context of a word icell ID− with its window size is five can be defined as: 5 1 1 5( ) { ,..., , ,..., }i i i i icontext cell ID cell ID cell ID cell ID cell ID− − + +− = − − − − , where a contextual word ( ccell ID ) can be one of 5icell ID −− , 4icell ID −− , 3icell ID −− , 2icell ID −− , 1icell ID −− , 1icell ID +− , 2icell ID +− , 3icell ID +− , 4icell ID +− , 5icell ID +− . Shi Feng et al. – Stopover Hotspots in North and Central America 12 where: 1(x)= (1 exp( ))x σ + − , ,x p cell IDv v v − is the vectors of word , ,x pcell ID cell ID cell ID− − − . We use xcell ID− to present a word included in the con- text of cell ID− and use pcell ID− to present a word not included in the context of cell ID− . “T” also presents the transposition of the vector for dot products between vec- tors. Finally, a high-dimensional vector for each word was got after optimizing the object function on the trajectory documents. The similarity of two vectors ,a bv v was then calculated with the cosine distance in the vector space as: ( ) 2 2 Similarity , a b a b a b v vv v v v ⋅ = ⋅   The cosine distance calculation results for similarity rep- resentation between each two high-dimensional vectors in the trajectory corpus were shown in the heat map for geographic interpretation. Geo-functional similarity min- ing was done by finding the two cell-IDs closest to the important stopover sites (hotspots got through PageRank algorithm) in the vector space with the minimum cosine distances as the potential alternative stopover sites. When applying word2vec for vector presentation, the following two parameter settings need to be noted: (1) Vector dimensionality: Low dimensionality usually means information loss. We set 100 as the embedding di- mension to facilitate information preserving and compact- ness according to empirical knowledge and experiments. (2) Window Size: A larger window tends to capture more overall information, and a smaller window gets local syn- tactic contexts (Levy and Goldberg, 2014). For mobility data, a larger window reveals mobility behaviors, and a smaller window captures geographical similarity (Zhu et al., 2019). According to the length of trajectories, we set the window size as 5 to focus on the local geographical similarity of neighbor cells. Analytical framework The overall analytical framework is described in Fig- ure.3. Migratory trajectories were estimated from field oc- currence data in the form of geographic coordinates from the eBird repository (https://ebird.org/; Feng et al., 2021). Geographic coordinates of migration trajectories were an- alyzed as spatio-temporal words using a situation-aware Figure 3. Overall structure for the method. (a) Trajectory estimation based on observations from eBird. The modelling process begins with es- timating bird migration trajectories based on citizen-science observation data stored in eBird platform. The data collection forms the foundation for understanding the migration paths of birds. (b) Geographic partition- ing for migration network construction. After estimating the trajectories, the migration routes are geographically partitioned, and a migration net- work can be constructed between different cells which form nodes, and edges represent the migration paths between these cells. (c) Stopover hotspots mining based on PageRank algorithm. The PageRank algorithm is then applied to the migration network to identify important stopover hotspots. PageRank ranks nodes based on their linkage quantity and qual- ity, thereby identifying critical nodes, especially key stopover sites in the migration network. (d) Potential alternative sites exploration based on word embedding technique. A word embedding technique (word2vec) is used to explore potential alternative sites that have similar geo-functional properties to the identified stopover hotspots. This technique identifies locations that could serve as alternative stopover sites based on spatial and functional similarities. (e) Stopover protection prioritization guid- ance. Stopover protection prioritization can be inferred based on the stop- over hotspots and alternative stopover sites. Right: (a)-(b) Preparation. The first two steps are part of the data preparation phase, which focuses on trajectory estimation and preparation for network construction. (c) Stopover hotspots based on PageRank. The third step utilizes the PageR- ank algorithm to rank and identify stopover hotspots. (d) Geo-functional similarity sites based on word2vec. The fourth step explores alternative sites with similar geo-functional characteristics using the word2vec tech- nique. (e) Protection guidance. The final step is for stopover protection prioritization guidance based on sites predicted. Shi Feng et al. – Stopover Hotspots in North and Central America 13 analysis (Zhu et al., 2019). The situation-aware analysis assessed dynamic bird mobility across a grid coordinate system of continental North and Central America after geographic partition. Trajectory coordinates were convert- ed to cell-ID (the number for each cell) sequences. Cell-ID sequences were used to build a migratory network with each cell as a node. Stopover hotspots were identified us- ing the PageRank algorithm (Page et al., 1999) and poten- tial stopover alternative hotspots were determined using a word embedding technique (Mikolov et al., 2013). Stop- over hotspots and alternative stopover sites were projected on maps to identify areas for protection prioritization and analysis for migratory birds. Stopover hotspots characteristics The effect of landscape configuration on the network was explored based on landscape variables and stopover hotspots. We overlaid the artificial light (https://www. nasa.gov/image-article/earth-night/), the topographic map (https://apps.nationalmap.gov/viewer/), and the World Database on Protected Areas (https://www.protectedplan- et.net/en/thematic-areas/wdpa?tab=WDPA) in North and Central America for protection prioritization reference. And we used the coverage of the total species and trajecto- ry numbers to evaluate the stopover hotspots importance, and the overlapped species to assess the results of geo- graphic functional similarity. Then we used the datasets of the Global Land Cover Estimation (GLanCE; Stanimirova et al., 2023) product and Human Footprint (HFP) datasets (Mu et al., 2022) to analyze the land cover classes and the interference degree of human activities on the ecosys- tem about the stopover hotspot conditions quantitatively. Finally, we combined the stopover hotspot areas with the Important Bird and Biodiversity Areas (IBAs) in North and Central America (Donald et al., 2019; https://data- zone.birdlife.org/) for protection strategy comparison. Results Migration network structure and PageRank value results We modeled 52 migratory species during the period from 2017 to 2019, and estimated 540 migration trajecto- ries in 2017, 435 in 2018, and 435 in 2019. The migration network structure resulted in a total of 1410 sequences during the study period (Fig.4), including their respective PageRank values at the cell level (Fig.5). The top 10 im- portant stopovers nodes with the highest PageRank val- ues were: cells 237, 589, 554, 520, 519, 952, 271, 624, 236,and 918. Potential alternative stopover sites A heatmap of cosine distance revealed calculation re- sults between each two high-dimensional vectors of cells by word embedding technique (Fig.6). We performed geo-functional similarity mining for the top 10 important stopover cell-IDs (i.e., 237, 589, 554, 520, 519, 952, 271, 624, 236, 918) and the results were shown in Table 1. Figure 4. Migration network structure for 2017-2019. Each node can be located and show the in-degree and out-degree. Take node 1299 in the black square for example, it represents cell with number 1299 in the geographic partition map (Fig.1) with its in-degree and out-degree are 14. Arrows show movement direction between nodes. Inset: The in (out) degree can reflect the number of arrows move to (from) each node. The figure above can help us obtain the degree and structure infor- mation for each node in the migration network which works as the basis of PageRank algorithm. https://www.nasa.gov/image-article/earth-night/ https://www.nasa.gov/image-article/earth-night/ https://datazone.birdlife.org/ https://datazone.birdlife.org/ Shi Feng et al. – Stopover Hotspots in North and Central America 14 Figure 5. The PageRank value results. (a) The PageRank value curve of 1091 cell-IDs included in the avian trajectory sequences, which is ordered from the largest to smallest. The y-axis is the PageRank values, and the x-axis is cell-IDs; (b) The PageRank value curve for the top 10 important stopovers cells, where 10 cells: 237, 589, 554, 520, 519, 952, 271, 624, 236, 918 are included. Similarly, the y-axis is the PageRank values, and the x-axis is the cell-IDs of top 10 important stopovers. (a) Curve for PageRank values of each cell (b) Curve for the top 10 PageRank values of stopover cells Figure 6. Heatmap of cosine distance between vectors of cells based on word2vec. The image on the left is a partial eagle eye diagram of the right one for a clearer presentation. The color of the grid can show the similarity between two cells. The smaller the cosine distance is, the higher the similarity is. Table 1. Top 10 important stopovers and their potential alternative sites with cosine distances analysis for 2017-2019 Top 10 important stopovers Potential alternative stopover sites (cosine distance based) 237 236(0.0367) 165(0.0406) 589 659(0.1367) 624(0.1454) 554 519(0.1330) 624(0.1437) 520 519(0.0661) 485(0.0749) 519 520(0.0661) 485(0.1238) 952 917(0.0498) 987(0.0606) 271 237(0.0610) 236(0.0618) 624 659(0.1240) 694(0.1291) 236 235(0.0261) 237(0.0367) 918 883(0.0493) 952(0.0611) Shi Feng et al. – Stopover Hotspots in North and Central America 15 Post modeling interpretation The map for the stopover hotspots and their potential alternative sites in North and Central America (Fig.7) re- vealed three areas highlighted for higher protection prior- itization: Florida peninsula and its inland (FL), the region of Central America (CA), and the region near Puget Sound (PS) rich in national parks and forests. The overlapped map with the artificial light map and topographic map can be found in Fig.S2. In the stopover hotspots result assessment for 2017- 2019, the cells with top 10 (i.e., 10/1091; 0.9%) PageRank values were extracted, covering 777 trajectories of 46 spe- cies (Table S2). Trajectories accounted for 88.5% (46/52) species and 55.1% (777/1410) trajectory sequences. Independent data from 2020-2022 were used for model evaluation focusing on key stopover sites detection. The top 10 important cells covered 729 (729/1353; 53.9%) tra- jectories of 41 (41/52; 78.9% ) species (Table S3). These important cells revealed that the sites during our three- year study period also stand out as important stopovers in the test years. Also, we assessed the results of geographic function- al similarity through the overlapped species for 2017- 2019. The number of overlapped species revealed that the more species overlap, the higher the similarity is (Tables 2 and S2). Data from 2020-2022 revealed the effectiveness of geo-functional similarity. That is, the similarity results re- vealed that the study period and the testing period share almost the same outcomes for potential alternative sites exploration (Tables 3 and S3). Furthermore, Fig.8 showed different landcover classed in the three hotspots areas, including water, developed, barren/sparsely vegetated, tress, shrub and herbaceous with all the three areas having the most percentages of trees, where “CA” represented the Central America area, “FL” represented the Florida peninsula and its inland, and “PS” represented the region near Puget Sound. Fig 9 showed the HFP levels of the stopover hotspots: (1) PS: the distribution was concentrated at a low value with its median about 10, which indicated that human interference was relatively low and stable; (2) The Florida peninsu- la and its inland (FL): its median was about 17 with its maximum HFP value could be greater than 35, indicating that there was a rather strong human interference to parts of this area; (3) Central America (CA): its median was about 15 locating between FL and PS indicating moder- ate interference levels. Finally, the overlapping percentage between the stopover hotspot cells and IBAs was 13/18 (72.2%, grids in red and yellow) with 5/18 (27.8%) not identified as IBAs (Fig.10). Figure 7. Stopover hotspots and potential alternative sites for them over- laid on the protected area map. The red solid circles represent the top 10 important stopovers acquiring the top 10 PageRank values. The orange circles represent potential alternative stopover sites. Green polygons de- note terrestrial and inland water protected areas. Blue polygons represent marine protected areas. Three areas are highlighted and recommended as higher protection prioritization sites for targeted attention, including Florida peninsula and its inland, the region of Central America, and the region near Puget Sound. Table 2. Top 10 important stopovers and their potential alternative sites with overlapped species analysis for 2017-2019 Top 10 important stopovers (included species number) Potential alternative stopover sites (overlapped species number) 237(10) 589(25) 554(19) 520(9) 519(15) 952(13) 271(12) 624(19) 236(10) 918(20) 236(8) 659(17) 519(13) 519(9) 520(9) 917(11) 237(9) 659(18) 235(3) 883(14) 165(2) 624(17) 624(14) 485(6) 485(7) 987(10) 236(9) 694(17) 237(8) 952(13) Table 3. Top 10 important stopovers and their potential alternative sites with overlapped species analysis for 2020-2022 Top 10 important stopovers (included species number) Potential alternative stopover sites (overlapped species number) 237(9) 589(20) 554(20) 520(14) 519(16) 952(11) 271(11) 624(19) 236(7) 918(18) 236(7) 659(17) 519(15) 519(11) 520(13) 917(10) 237(9) 659(18) 235(3) 883(17) 165(1) 624(18) 624(15) 485(7) 485(7) 987(10) 236(7) 694(15) 237(7) 952(10) Shi Feng et al. – Stopover Hotspots in North and Central America 16 Figure 8. Landcover classes analysis with the GLanCE product. “CA” represents the Central America area, “FL” represents the Florida peninsula and its inland, and “PS” represents the region near Puget Sound. The figure above can show different landcover classed in the three hotspots areas, including water, developed, barren/sparsely vege- tated, tress, shrub and herbaceous. And all the three areas have the most percentages of trees. Figure 9. Human disturbance analysis with the Human Footprint dataset. The gray dotted lines represent the 0%, 1%, 5%, 25%, 50%, 75%, 95%, 99%, and 100% quantiles of the dataset which are 0.006, 0.316, 5.440, 10.511, 15.911, 24.520, 30.927, and 41.259 respectively. For the three stopover hotspots: (1) The region near Puget Sound (PS): the distribution is concentrated at a low value with its median about 10, which indicates that human interference is relatively low and stable; (2) The Florida peninsula and its inland (FL): its median is about 17 with its maximum HFP value can be greater than 35, indicating that there is a rather strong human interference to parts of this area; (3) Central America (CA): its median is about 15 locating between FL and PS indicating moderate interference levels. Shi Feng et al. – Stopover Hotspots in North and Central America 17 Discussion This study combined the booming of public citizen sci- ence observation data with the urgent need to understand bird migration (Rosenberg et al., 2019). In this article, we presented a systematic method considering the network structure during migration for multi-species at the popu- lation level to help develop precision conservation strat- egies. Effective stopover hotspots mining and potential alternative sites exploration can help manage migration network connectivity and decrease migration risks, es- pecially for long-distance migratory birds (Zurell et al., 2018). We applied the PageRank algorithm for stopover sites importance assessment considering both the linkage quantity and quality across the migration network for stop- over hotspots discovery. We also explored the alternative sites for each hotspot based on geo-functional similarity using a word embedding technique. Our study effectively mined both the spatiotemporal and semantic information of the trajectory sequences and offered effective and pro- spective guidance for protection prioritization decisions. Our method employed 52 focal migratory species in North and Central America during 2017-2019, and generated rigorous estimations for stopover hotspots and potential alternative sites during migration. Species analysis for stopover hotspots and their potential alternative sites The larger number of species included in a stopover site can be used as a proxy of the high importance of the site. The more species overlapped between sites, the high- er the similarity between the sites. That is to say, species sharing a stopover site could be a proxy of biogeographic analogy even in spatially disparate zones. And the species overlapped can indicate the similarity of geographic func- tions, which means areas attract more same species, the more similar their geographic functions is. We note that cell 237 acquires a higher PageRank value for its crucial location as the “bottleneck” of the American flyway. This area is around the Central American isthmus, includes plains and Central volcanic ridge covered with rich forests and some farmlands as well as both water and well-pro- tected land resources, which can provide support and ev- idence for the inference of Central America’s significant value for migratory bird network connectivity (Bayly et al., 2018). Detection of relevant cells can give refined and scientific targets for Central America protection against potential threats from forest loss and expansion of land conversion (Cohen et al., 2007). Figure 10. The stopover hotspots overlapped with the IBAs map. The overlapping percentage between the stopover hotspot cells and IBAs is 13/18 (72.2%, grids in red and yellow) with 5/18 (27.8%) not identified as IBAs. Shi Feng et al. – Stopover Hotspots in North and Central America 18 Stopover hotspots characteristics analysis With the growing attention to migratory bird protec- tion, there is a great need for informed protection of stop- over hotspots across migratory paths. We found fundamen- tal stopover sites across a migration network and potential alternative sites for further management accounting for human activity. Our finding mirrored other efforts for da- ta-driven biodiversity conservation Studies have explored relevant decision-making plans. For example, Thomson et al. (2020) developed a landscape-scale spatial conser- vation action planning tool (SCAP) to offer advances for conservation actions in heterogenous landscapes. Guo et al. (2024) discussed the relationship between protection status and light pollution with the stopover hotspots distri- bution in the eastern US. Migratory birds need water to drink and feed. Thus, areas close to water sources, such as coastlines, freshwa- ter lakes, rivers, and wetlands, are often crucial stopovers (Boere et al., 2006). Different species of migratory birds have different habitat needs, Thus, heterogenous habitats can attract more migratory species (Tu et al., 2020). The three stopover hotspot areas (FL, CA, and the region near PS) were found close to water sources and food sources like forests. The stopover sites CA and PS are currently surrounded by terrestrial and marine protected areas. The regions around FL, however, are not under sufficient pro- tection. Considering the high-level population density and economic development (Mu et al., 2022), it may be nec- essary to involve proper urban development planning and private lands involvement for improved stopover protec- tion in the eastern US. Results highlighted the importance of forest conservation for bird migration (Guo et al., 2023; Mehlman et al., 2005). Also, the land cover of CA, FL and the region near PS is dominated by trees, calling for more attention while making bird conservation decisions (Fig.8). HFP index levels of the region near PS suggest a lower level of hu- man disturbance. CA shows moderate but widespread human presence. In contrast, FL shows a wide range of HFP values, with many cells experiencing high levels of anthropogenic impact, which needs more protection activ- ities (Fig.9). Lastly, the three hotspot areas overlap with the pro- tected areas in North and Central America (Fig.10, 72.2% overlap). McClure et al. (2018) suggested enhancing co- operation through policy mechanisms that account for pro- tected areas for raptor species along the Central America area. Kirby et al. (2008) called for attention to agricultural intensification, human infrastructure development, and forest protection based on protected areas for migratory landbird and waterbird species. It is unclear whether birds select migration stopovers in protected areas or if protect- ed areas are established in sites recognized as important for seasonal bird congregation. Future directions One important limitation of our work is the limited ca- pacity to detect areas with small PageRank values that are still relevant to migration connectivity. For example, some endangered species could have low eBird records resulting in weak signal of stopover hotspots, but could be the spe- cies in higher need of management and protection, such as Kirtland’s Warbler (Setophaga kirtlandii) breeding in the Great Lakes. Areas not covered by eBird observations are not assessed and, instead, oversampled areas provide more information at the cost of bias. Assisting with individual tracking data, human movement data or other observation datasets (Ai et al., 2024) could complement our stopover hotspots map. Additionally, PageRank algorithm and word embedding technique only consider the link relationship between nodes, and ignored other information, such as hu- man and bird behavioral data. Finally, eBird data has the biases and limitations that citizen science data have, such as the bias from regional differences in reporting rates that depend heavily on species’ overlap with eBird users’ ac- tivities (Robinson et al., 2022). To address this data limita- tion, information from different databases and additional taxa during longer periods would help detect and mitigate systematic bias from a single data platform. Conclusion Our novel analytical approach to access migratory pat- terns revealed important stopover nodes in the migratory network for a comprehensive range of bird taxa in North and Central America. We extracted important stopover signals based on the network-level evaluation method by PageRank algorithm, and explored alternative stopover sites that accounted for spatial and temporal context in- formation of trajectory sequences between sites with word embeddings, avoiding studying the sites as isolated ones. Targeted protection and recharging of areas near water and forests were recommended in the sites identified as hot stopover sites of bird migration, especially in developed land in the eastern United States. Discoveries and analyt- ical approaches presented here could be useful for migra- tory connectivity protection in the full migration cycle and inform multinational conservation prioritization. Data and Code Availability Data and code used to perform this study are avail- able at GitHub (https://github.com/ash0920-git/stop- over-hotspots). Shi Feng et al. – Stopover Hotspots in North and Central America 19 Acknowledgements This work is supported by the National Key R&D Program of China (2022YFF0802300), National Nat- ural Science Foundation of China (U21A20478), Zhe- jiang High-Level Talents Special Support Program (2021R52002). LEE was supported by National Science Foundation CAREER (2235295) and HEGS (2116748) awards, NIH K01AI168452 award, Virginia Tech DA PPP, CeZAP, and ICTAS grants, and the Chinese Academy of Sciences PIFI project 2024PVC0085. The content is sole- ly the responsibility of the authors and does not necessari- ly represent the official views of the National institutes of Health. Competing Interests The authors have declared that no competing interests exist. References Ai, Y., Zhai, H., Sun, Z., Yan, W., and Hu, T., 2024. Flock- Seer: a portable stereo vision observer for bird flock- ing. IET Cyber Syst. Robot. 6(1): e12118. https://doi. org/10.1049/csy2.12118 Artificial light map, 2024. https://www.nasachina.cn/ apod/19232.html (accessed June 14, 2024) Bayly, N. J., Rosenberg, K. V., Easton, W. E., Gomez, C., Carlisle, J. A. Y., Ewert, D. N., and Goodrich, L., 2018. Major stopover regions and migratory bottle- necks for Nearctic-Neotropical landbirds within the Neotropics: a review. Bird Conserv. Int. 28(1): 1-26. https://doi.org/10.1017/S0959270917000296 Berger, M., McDonough, K., and Seversky, L. M., 2016. cite2vec: citation-driven document explora- tion via word embeddings. IEEE Trans. Vis. Com- put. Graph. 23(1): 691-700. https://doi.org/10.1109/ TVCG.2016.2598667 Boere, G. C., Galbraith, C. A., and Stroud, D. A. (Eds.)., 2006. Waterbirds around the world: a global overview of the conservation, management and research of the world’s waterbird flyways. Bonter, D. N., Gauthreaux Jr, S. A., and Donovan, T. M., 2009. Characteristics of important stopover locations for migrating birds: remote sensing with radar in the Great Lakes Basin. Conserv. Biol. 23(2): 440-448. https://doi.org/10.1111/j.1523-1739.2008.01085.x Catry, T., Correia, E., Gutiérrez, J. S., Bocher, P., Robin, F., Rousseau, P., and Granadeiro, J. P., 2024. Low mi- gratory connectivity and similar migratory strategies in a shorebird with contrasting wintering population trends in Europe and West Africa. Sci. Rep. 14(1): 4884. https://doi.org/10.1038/s41598-024-55501-y Cohen, E. B., Barrow Jr, W. C., Buler, J. J., Deppe, J. L., Farnsworth, A., Marra, P. P., and Moore, F. R., 2017. How do en route events around the Gulf of Mexico in- fluence migratory landbird populations? Condor Orni- thol. Appl. 119(2): 327-343. https://doi.org/10.1650/ CONDOR-17-20.1 Derpanis, K. G., 2005. Mean-Shift clustering. Lecture Notes. [Online]. Available:http://www.cse.yorku. ca/%7Ekosta/CompVis_Notes/mean_shift.pdf (ac- cessed 10 March 2023) Donald, P. F., Fishpool, L. D., Ajagbe, A., Bennun, L. A., Bunting, G., Burfield, I. J., ... and Wege, D. C., 2019. Important Bird and Biodiversity Areas (IBAs): the development and characteristics of a global inventory of key sites for biodiversity. Bird Con- serv. Int. 29(2), 177-198. https://doi.org/10.1017/ S0959270918000102 eBird, 2023. https://ebird.org/ (accessed April 22, 2023) Feng, S., Yang, Q., Hughes, A. C., Chen, J., and Qiao, H., 2021. A novel method for multi-trajectory re- construction based on LoMcT for avian migration in popu-lation level. Eco. Inform. 63: 101319. https:// doi.org/10.1016/j.ecoinf.2021.101319 Gazoni, E. and Clark, C, 2023. openpyxl 3.1.2. Available from: https://pypi.org/project/openpyxl/ (accessed 10 Feb 2024) GBIF. Available from https://www.gbif.org/ (accessed 20 August 2023) Grisel, O., Mueller,A., … and Eren.K., 2022. scikit-learn/ scikit-learn: scikit-learn 1.1. 2. Available from: https:// zenodo.org/records/6968622 (accessed 10 Feb 2023) Guo, F., Buler, J. J., Smolinsky, J. A., and Wilcove, D. S., 2023. Autumn stopover hotspots and multi- scale habitat associations of migratory landbirds in the eastern United States. Proc. Natl. Acad. Sci. 120(13): e2203511120. https://doi.org/10.1073/ pnas.2203511120 Guo, F., Buler, J. J., Smolinsky, J. A., and Wilcove, D. S., 2024. Seasonal patterns and protection status of stop- over hotspots for migratory landbirds in the eastern United States. Curr. Biol. 34(1): 235-244. https://doi. org/10.1016/j.cub.2023.11.033 Harris, C.R., Millman, K.J., van der Walt, S.J. et al., 2020. Array programming with NumPy. Nature 585, 357– 362. https://doi.org/10.1038/s41586-020-2649-2 Hastie, T. J., and Tibshirani, R. J., 1990. Generalized Ad- ditive Models (Vol. 43). CRC Press, London. Higuchi, H., 2012. Bird migration and the conservation of the global environment. J. Ornithol. 153(Suppl 1), 3-14. https://doi.org/10.1007/s10336-011-0768-0 Shi Feng et al. – Stopover Hotspots in North and Central America 20 Hong, L., Qian, Y., Gong, C., Zhang, Y., and Zhou, X., 2023. Improved Key Node Recognition Method of Social Network Based on PageRank Algorithm. Com- put. Mater. Continua. 74(1). https://doi.org/10.32604/ cmc.2023.029180 Hunter, J.D. 2007. Matplotlib: a 2D Graphics Environ- ment. Comput. Sci. Eng. 9 (3), 90-95. https://doi.org/ 10.1109/MCSE.2007.55 Jin, Y. T., You, J., Wakamiya, S., and Kwon, H. Y., 2024. Analyzing user reactions using relevance between lo- cation information of tweets and news articles. EPJ Data Sci. 13(1): 44. https://doi.org/10.1140/epjds/ s13688-024-00465-2 Kaiser, A., 1999. Stopover strategies in birds: a review of methods for estimating stopover length. Bird Study. 46 (Supplement): S299-S308. https://doi. org/10.1080/00063659909477257 Kirby, J. S., Stattersfield, A. J., Butchart, S. H., Evans, M. I., Grimmett, R. F., Jones, V. R., ... and Newton, I., 2008. Key conservation issues for migratory land- and waterbird species on the world’s major flyways. Bird Conserv. Int., 18(S1), S49-S73. https://doi. org/10.1017/S0959270908000439 Knight, S. M., Bradley, D. W., Clark, R. G., Gow, E. A., Bélisle, M., Berzins, L. L., ...and Norris, D. R., 2018. Constructing and evaluating a continent-wide migra- tory songbird network across the annual cycle. Ecol. Monogr. 88(3): 445-460. https://doi.org/10.1002/ ecm.1298 Kramer, G. R., Andersen, D. E., Buehler, D. A., Wood, P. B., Peterson, S. M., Lehman, J. A., ... and Streby, H. M., 2018. Population trends in Vermivora warblers are linked to strong migratory connectivity. Proc. Natl. Acad. Sci. 115(14): E3192-E3200. https://doi. org/10.1073/pnas.1718985115 La Sorte, F. A., Fink, D., Hochachka, W. M., and Kelling, S., 2016. Convergence of broad-scale migration strat- egies in terrestrial birds. Proc. R. Soc. B. 283(1823): 20152588. https://doi.org/10.1098/rspb.2015.2588 Levy, O., and Goldberg, Y., 2014. Dependency-based word embeddings. Proc. 52nd Annu. Meet. Assoc. Comput. Linguist. https://doi.org/10.3115/v1/P14-2050 Lin, H. Y., Schuster, R., Wilson, S., Cooke, S. J., Rode- wald, A. D., and Bennett, J. R., 2020. Integrating sea- son-specific needs of migratory and resident birds in conservation planning. Biol. Conserv. 252: 108826. https://doi.org/10.1016/j.biocon.2020.108826 Liu, F., Zheng, L., and Zheng, J., 2020. HieNN-DWE: a hierarchical neural network with dynamic word embeddings for document-level sentiment classi- fication. Neurocomputing. 403: 21-32. https://doi. org/10.1016/j.neucom.2020.04.084 Machin, J., 2017. xlwt 1.3.0. Available from: https://pypi. org/project/xlwt/ (accessed 12 Feb 2022) Marín, S. D., 2018. pygam 0.8.0. Available from: https:// pypi.org/project/pygam/ (accessed 11 Feb 2022) MATLAB, R2021b. version 9.11.0.1873467, Natick, Mas- sachusetts: The MathWorks Inc. McClure, C. J., Westrip, J. R., Johnson, J. A., Schulwitz, S. E., Virani, M. Z., Davies, R., ... and Butchart, S. H., 2018. State of the world’s raptors: Distributions, threats, and conservation recommendations. Biol. Conserv., 227, 390-402. https://doi.org/10.1016/j.bio- con.2018.08.012 McCormick, C., 2016. Word2vec tutorial-the skip- gram model. [Online]. Available: http://mc- cormickml.com/2016/04/19/word2vec-tutori- al-the-skip-gram-model Mehlman, D. W., Mabey, S. E., Ewert, D. N., Duncan, C., Abel, B., Cimprich, D., ... and Woodrey, M., 2005. Conserving stopover sites for forest-dwelling migra- tory landbirds. The Auk, 122(4), 1281-1290. https:// doi.org/10.1093/auk/122.4.1281 Mikolov, T., Chen, K., Corrado, G., and Dean, J., 2013. Efficient estimation of word representations in vector space. arXiv preprint arXiv:1301.3781. https://doi. org/10.48550/arXiv.1301.3781 Mikolov, T., Sutskever, I., Chen, K., Corrado, G. S., and Dean, J., 2013. Distributed representations of words and phrases and their compositionality. NeurIPS. 26. https://doi.org/10.48550/arXiv.1310.4546 Mu, H., Li, X., Wen, Y., Huang, J., Du, P., Su, W., ... and Geng, M., 2022. A global record of annual terrestri- al Human Footprint dataset from 2000 to 2018. Sci. Data, 9(1), 176. https://doi.org/10.1038/s41597-022- 01284-8 Nemes, C. E., Cabrera-Cruz, S. A., Anderson, M. J., DeG- roote, L. W., DeSimone, J. G., Massa, M. L., and Co- hen, E. B., 2023. More than mortality: consequences of human activity on migrating birds extend beyond direct mortality. Ornithol. Appl. 125(3), duad020. https://doi.org/10.1093/ornithapp/duad020 Nicol, S., Cros, M. J., Peyrard, N., Sabbadin, R., Trépos, R., Fuller, R. A., and Woodworth, B. K., 2023. Fly- wayNet: a hidden semi-Markov model for inferring the structure of migratory bird networks from count data. Methods Ecol. Evol. 14(2): 265-279. https://doi. org/10.1111/2041-210X.14011 Page, L., Brin, S., Motwani, R., and Winograd, T., 1999. The PageRank citation ranking: bringing order to the web. Stanford Infolab. [Online]. Available: http://il- pubs.stanford.edu:8090/422/1/1999-66.pdf Python 3.10. Python Software Foundation, 2021. Shi Feng et al. – Stopover Hotspots in North and Central America 21 Qian, M., Liu, J., Li, C., and Pals, L., 2019. A comparative study of English-Chinese translations of court texts by machine and human translators and the Word2Vec based similarity measure’s ability to gauge human evaluation biases. Proc. MT Summit XVII: Transla- tor, Project and User Tracks. pp: 95-100. https://acl- anthology.org/W19-6714.pdf Reback, J., Jbrockmendel, McKinney, W., van Den Boss- che, J., Roeschke, M., … and Augspurger, T. pandas dev/pandas: Pandas1.4.3. Available from: https://ze- nodo. org/record/3509134 (accessed 2 Feb 2023) Rehurek, R., 2023. gensim 4.3.1. Available from: https:// pypi.org/project/gensim/ (accessed 2 Feb 2024) Robinson, O. J., Socolar, J. B., Stuber, E. F., Auer, T., Ber- ryman, A. J., Boersch-Supan, P. H., ... and Johnston, A., 2022. Extreme uncertainty and unquantifiable bias do not inform population sizes. Proc. Natl. Acad. Sci. 119(10): e2113862119. https://doi.org/10.1073/ pnas.2113862119 Rogers, I., 2002. The Google Pagerank algorithm and how it works. [Online]. Available: https://cs.wmich.edu/ gupta/teaching/cs3310/lectureNotes_cs3310/Pager- an%20Explained%20Correctly%20with%20Exam- ples_www.cs.princeton.ed_~chazelle_courses_BIB_ pagerank.pdf Rosenberg, K. V., Dokter, A. M., Blancher, P. J., Sauer, J. R., Smith, A. C., Smith, P.A., and Marra, P. P., 2019. Decline of the North American avifauna. Science, 366(6461):120-124. https://doi.org/10.1126/science. aaw1313 Runge, C. A., Watson, J. E., Butchart, S. H., Hanson, J. O., Possingham, H. P., and Fuller, R. A., 2015. Protected areas and global conservation of migratory birds. Sci- ence, 350(6265): 1255-1258. https://doi.org/10.1126/ science.aac9180 Schrimpf, M. B., Des Brisay, P. G., Johnston, A., Smith, A. C., Sánchez-Jasso, J., Robinson, B. G., and Koper, N., 2021. Reduced human activity during COVID-19 alters avian land use across North America. Sci. Adv. 7(2): eabf5073. https://doi.org/10.1126/sciadv. abf5073 Somveille, M., Bay, R. A., Smith, T. B., Marra, P. P., and Ruegg, K. C., 2021. A general theory of avian mi- gratory connectivity. Ecol. Lett. 24(9): 1848-1858. https://doi.org/10.1111/ele.13817 Stanimirova, R., Tarrio, K., Turlej, K., McAvoy, K., Stone- brook, S., Hu, K. T., ... and Friedl, M. A., 2023. A global land cover training dataset from 1984 to 2020. Sci. Data, 10(1), 879. https://doi.org/10.1038/s41597- 023-02798-5 Thomson, J., Regan, T. J., Hollings, T., Amos, N., Geary, W. L., Parkes, D., ... and White, M., 2020. Spa- tial conservation action planning in heterogeneous landscapes. Biol. Conserv. 250, 108735. https://doi. org/10.1016/j.biocon.2020.108735 Tu, H. M., Fan, M. W., and Ko, J. C. J., 2020. Different habitat types affect bird richness and evenness. Sci. Rep., 10(1), 1221. https://doi.org/10.1038/s41598- 020-58202-4 USGS geographic map, 2024. https://apps.nationalmap. gov/viewer/ (accessed June 17, 2024) Wang, H., Hernandez, J. M., and van Mieghem, P., 2008. Betweenness centrality in a weighted network. Phys. Rev. E Stat. Nonlin. Soft Matter Phys., 77(4), 046105. https://doi.org/10.1103/PhysRevE.77.046105 Wang, M., Yang, S., Sun, Y., and Gao, J., 2017. Discov- ering urban mobility patterns with PageRank based traffic modeling and prediction. Physica A: Stat. Mech. Appl. 485: 23-34. https://doi.org/10.1016/j. physa.2017.04.155 Webster, M. S., Marra, P. P., Haig, S. M., Bensch, S., and Holmes, R. T., 2002. Links between worlds: unraveling migratory connectivity. Trends Ecol. Evol. 17(2), 76-83. https://doi.org/10.1016/S0169- 5347(01)02380-1 Whitaker, J., 2022. basemap 1.3.4. Available from: https:// pypi.org/project/basemap/ (accessed 3 Feb 2023) Whitaker, J., 2022. pyproj 3.3.1. Available from: https:// zenodo.org/record/3509134 (accessed 7 Feb 2023) Withers, C., 2020. xlrd 2.0.1. Available from: https://pypi. org/project/basemap/ (accessed 13 Feb 2022) World Database on Protected Areas, 2024. https:// www.protectedplanet.net/en/thematic-areas/wd- pa?tab=WDPA (accessed May 15, 2024) Wyborn, C., and Evans, M. C., 2021. Conservation needs to break free from global priority mapping. Nat. Ecol. Evol. 5(10): 1322-1324. https://doi.org/10.1038/ s41559-021-01540-x Xiong, S., Lv, H., Zhao, W., and Ji, D., 2018. Towards Twitter sentiment classification by multi-level sen- timent-enriched word embeddings. Neurocomput- ing, 275:2459-2466. https://doi.org/10.1016/j.neu- com.2017.11.023 Xu, Y., Si, Y., Takekawa, J., Liu, Q., Prins, H. H., Yin, S., ... and de Boer, W. F., 2020. A network approach to pri- oritize conservation efforts for migratory birds. Con- serv. Biol. 34(2): 416-426. https://doi.org/10.1111/ cobi.13383 Yorav-Raphael, N. and and da Costa-Luis, C., 2024. Tgdm: a fast, extensible progress bar for python and cli, https://github.com/tqdm/tgdm, Version 4.65.0. https://doi.org/10.1111/ele.13817 https://pypi.org/project/basemap/ https://pypi.org/project/basemap/ https://zenodo.org/record/3509134 https://zenodo.org/record/3509134 https://pypi.org/project/basemap/ https://pypi.org/project/basemap/ https://doi.org/10.1038/s41559-021-01540-x https://doi.org/10.1038/s41559-021-01540-x https://doi.org/10.1016/j.neucom.2017.11.023 https://doi.org/10.1016/j.neucom.2017.11.023 https://doi.org/10.1111/cobi.13383 https://doi.org/10.1111/cobi.13383 Shi Feng et al. – Stopover Hotspots in North and Central America 22 Zhang, T. Y., and Wang, X. L., 2011. Outlier detection algorithm based on space local deviation factor. Comput. Eng. 14. https://doi.org/10.3969/j.issn.1000- 3428.2011.14.096 (in Chinese with English abstract) Zhang, W., Wei, J., and Xu, Y., 2023. Prioritizing global conservation of migratory birds over their migration network. One Earth, 6(11): 1340-1349. https://doi. org/10.1016/j.oneear.2023.08.017 Zhu, M., Chen, W., Xia, J., Ma, Y., Zhang, Y., Luo, Y., and Liu, L., 2019. Location2vec: a situation-aware rep- resentation for visual exploration of urban locations. IEEE Trans. Intell. Transp. Syst. 20(10): 3981-3990. https://doi.org/10.1109/TITS.2019.2901117 Zivot, E., and Wang, J., 2007. Modeling financial time series with S-Plus® (Vol. 191). Springer Science & Business Media. Zumstein,F.,2023. xlwings 0.30.6. Available from: https://pypi.org/project/xlwings/ (accessed 13 Feb 2024) Zurell, D., Graham, C. H., Gallien, L., Thuiller, W., and Zimmermann, N. E., 2018. Long-distance migratory birds threatened by multiple independent risks from global change. Nat. Clim. Chang. 8(11): 992-996. https://doi.org/10.1038/s41558-018-0312-9 https://doi.org/10.1016/j.oneear.2023.08.017 https://doi.org/10.1016/j.oneear.2023.08.017 https://pypi.org/project/xlwings/ Shi Feng et al. – Stopover Hotspots in North and Central America 23 Figure S1. Frequency distribution of minimum stopover length during autumn migration periods Figure S2. Stopover hotspots overlapped with the artificial light map and topographic map (a) Stopover hotspots overlapped with the artificial light map (b) Stopover hotspots overlapped with the topographic map Shi Feng et al. – Stopover Hotspots in North and Central America 24 Table S1. Species list and their habitats information* Number Species name 1 Accipiter cooperii Breeds in forested areas; more common in suburban areas. 2 Aix sponsa Found in wetlands and flooded woods. 3 Anas platyrhynchos Found anywhere with water, including city parks, backyard creeks, and various wetland habitats. 4 Archilochus colubris Readily comes to sugar water feeders and flower gardens. 5 Ardea alba Ponds, marshes, and tidal mudflats. 6 Ardea Herodias Occurs in almost any wetland habitat, from small ponds to marshes to saltwater bays. 7 Branta canadensis Occurs in any open or wetland habitat. 8 Bucephala albeola Found in bays, estuaries, reservoirs, and lakes in winter. Travels to boreal forest and nests in cavities in summer. 9 Buteo jamaicensise Dges of trees. 10 Buteo lineatus Often in forested areas. 11 Catharus ustulatus Breeds in the boreal forest. 12 Charadrius vociferus Often in fields with short grass or barren dirt. 13 Colaptes auratus Often seen feeding on the ground in open areas, foraging for ants and worms. 14 Cyanocitta cristata Pairs or small groups travel through mature deciduous or coniferous woodlands. 15 Dumetella carolinensis Especially thickets or second-growth at the edge of forests, often near water. 16 Egretta thula Found in a variety of wetland habitats, especially shallow marshy pools and mudflats. 17 Fulica americana Ponds, city parks, marshes, reservoirs, lakes, ditches, and saltmarshes. 18 Geothlypis trichas Found in shrubby wet areas, including marshes, forest edges, and fields. 19 Haliaeetus leucocephalusnear Near bodies of water. 20 Hirundo rustica Especially large fields and wetlands. 21 Icterus galbula Breeds in deciduous trees in open woodlands, forest edges, orchards, riversides, parks, and backyards. 22 Larus delawarensis Found along lakes, rivers, ponds, and beaches. 23 Leiothlypis celata Found in scrubby areas, woodland edge, and thickets. Shi Feng et al. – Stopover Hotspots in North and Central America 25 Number Species name 24 Leiothlypis ruficapilla Breeds in coniferous or mixed forests. 25 Mareca strepera Typically found in pairs or small flocks in shallow wetlands, ponds, or bays. 26 Megaceryle alcyon Edges of streams, lakes, and estuaries. 27 Melospiza melodia Especially edges of fields, often near water. 28 Molothrus aterwoods Farmland, and stockyards. 29 Myiarchus crinitus Deciduous forests. 30 Pandion haliaetus On top of channel markers, utility poles and high platforms near water. 31 Passerina cyaneathe Edge of forests and fields. 32 Pheucticus ludovicianus Especially in deciduous forests. 33 Pipilo erythrophthalmus Inhabits scrubby areas and forest edges. 34 Podilymbus podiceps Occurs on ponds and marshes. 35 Polioptila caerulea Breed in deciduous woodlands, often near water. 36 Quiscalus quiscula Forages in fields, scrubby areas, and open woods. 37 Sayornis phoebe Woodland edge, brushy fields, or edges of ponds. 38 Setophaga americana Breeds in mature coniferous or deciduous forests, especially near water. 39 Setophaga coronata Mixed forests, often near clearings or edges. In migration and winter, found in any woodland or open shrubby area, including coastal dunes, fields, parks, and residential areas. 40 Setophaga palmarum Breeds in bogs and clearings in the boreal forest. 41 Setophaga petechia Near water, often foraging in shrubs fairly low to the ground. 42 Sialia sialis Favors fields and open woods. 43 Spatula discors Usually found in shallow wetlands or marshes. 44 Spinus tristis Found in weedy fields, cultivated areas, roadsides, orchards, and backyards. 45 Spizella passerina Usually found in open woodlands, scrubby areas, or even in suburban settings. 46 Stelgidopteryx serripennis Often seen near water, sometimes in mixed flocks with other swallows. Shi Feng et al. – Stopover Hotspots in North and Central America 26 Number Species name 47 Sturnus vulgaris Often abundant, gathering in large flocks in open agricultural areas and towns and cities. 48 Toxostoma rufum Shrubby habitats, especially second-growth woodland, thickets, and forest edge. 49 Troglodytes aedon Open or semiopen habitats, including suburbs, parks, rural farmland, and woodland edge with thick tangles. 50 Turdus migratorius In gardens, parks, yards, golf courses, fields, pastures, and many other wooded habitats. 51 Zenaida macroura Found in a variety of habitats from agricultural fields to lightly wooded areas. 52 Zonotrichia leucophrys Breeds in brushy areas or thickets in open forest, often with conifers. *The habitat information is collected from eBird (https://ebird.org/). Shi Feng et al. – Stopover Hotspots in North and Central America 27 Table S2. Cell-IDs with the included species for 2017-2019 Top 10 important stopovers Potential alternative stopover sites Cell-ID Species Cell-ID Species Cell-ID Species 237 Catharus ustulatus Egretta thula Hirundo rustica Myiarchus crinitus Pandion haliaetus Pheucticus ludovicianus Setophaga petechia Sialia sialis Spatula discors Ardea alba 236 Catharus ustulatus Hirundo rustica Myiarchus crinitus Pandion haliaetus Pheucticus ludovicianus Setophaga petechia Sialia sialis Spatula discors Troglodytes aedon Icterus galbula 165 Pheucticus ludovicianus Setophaga petechia 589 Archilochus colubris Ardea alba Buteo jamaicensis Dumetella carolinensis Egretta thula Geothlypis trichas Leiothlypis ruficapilla Myiarchus crinitus Pandion haliaetus Passerina cyanea Pheucticus ludovicianus Podilymbus podiceps Polioptila caerulea Quiscalus quiscula Sayornis phoebe Setophaga americana Setophaga palmarum Spatula discors Stelgidopteryx serripennis Bucephala albeola Catharus ustulatus Charadrius vociferus Icterus galbula Hirundo rustica Setophaga petechia 659 Buteo lineatus Geothlypis trichas Icterus galbula Myiarchus crinitus Pandion haliaetus Polioptila caerulea Sayornis phoebe Setophaga americana Setophaga coronata Setophaga palmarum Dumetella carolinensis Podilymbus podiceps Quiscalus quiscula Stelgidopteryx serripennis Buteo jamaicensis Catharus ustulatus Hirundo rustica Leiothlypis ruficapilla Megaceryle alcyon Pheucticus ludovicianus Spizella passerina 624 Buteo jamaicensis Buteo lineatus Geothlypis trichas Leiothlypis ruficapilla Myiarchus crinitus Pandion haliaetus Pheucticus ludovicianus Podilymbus podiceps Polioptila caerulea Sayornis phoebe Setophaga americana Setophaga palmarum Stelgidopteryx serripennis Dumetella carolinensis Icterus galbula Spatula discors Catharus ustulatus Hirundo rustica Quiscalus quiscula 554 Archilochus colubris Buteo jamaicensis Dumetella carolinensis Egretta thula Geothlypis trichas Myiarchus crinitus Pandion haliaetus Passerina cyanea Pheucticus ludovicianus Podilymbus podiceps Polioptila caerulea Setophaga americana Setophaga palmarum Anas platyrhynchos Catharus ustulatus Leiothlypis ruficapilla Hirundo rustica Setophaga petechia Spatula discors 519 Buteo jamaicensis Fulica americana Myiarchus crinitus Pandion haliaetus Passerina cyanea Podilymbus podiceps Setophaga americana Setophaga palmarum Catharus ustulatus Egretta thula Leiothlypis ruficapilla Pheucticus ludovicianus Stelgidopteryx serripennis Hirundo rustica Setophaga petechia 624 Buteo jamaicensis Buteo lineatus Geothlypis trichas Leiothlypis ruficapilla Myiarchus crinitus Pandion haliaetus Pheucticus ludovicianus Podilymbus podiceps Polioptila caerulea Sayornis phoebe Setophaga americana Setophaga palmarum Stelgidopteryx serripennis Dumetella carolinensis Icterus galbula Spatula discors Catharus ustulatus Hirundo rustica Quiscalus quiscula Shi Feng et al. – Stopover Hotspots in North and Central America 28 Top 10 important stopovers Potential alternative stopover sites Cell-ID Species Cell-ID Species Cell-ID Species 520 Myiarchus crinitus Pandion haliaetus Passerina cyanea Podilymbus podiceps Setophaga americana Setophaga palmarum Leiothlypis ruficapilla Pheucticus ludovicianus Setophaga petechia 519 Buteo jamaicensis Fulica americana Myiarchus crinitus Pandion haliaetus Passerina cyanea Podilymbus podiceps Setophaga americana Setophaga palmarum Catharus ustulatus Egretta thula Leiothlypis ruficapilla Pheucticus ludovicianus Stelgidopteryx serripennis Hirundo rustica Setophaga petechia 485 Buteo jamaicensis Passerina cyanea Podilymbus podiceps Setophaga americana Myiarchus crinitus Pheucticus ludovicianus Setophaga palmarum 519 Buteo jamaicensis Fulica americana Myiarchus crinitus Pandion haliaetus Passerina cyanea Podilymbus podiceps Setophaga americana Setophaga palmarum Catharus ustulatus Egretta thula Leiothlypis ruficapilla Pheucticus ludovicianus Stelgidopteryx serripennis Hirundo rustica Setophaga petechia 520 Myiarchus crinitus Pandion haliaetus Passerina cyanea Podilymbus podiceps Setophaga americana Setophaga palmarum Leiothlypis ruficapilla Pheucticus ludovicianus Setophaga petechia 485 Buteo jamaicensis Passerina cyanea Podilymbus podiceps Setophaga americana Myiarchus crinitus Pheucticus ludovicianus Setophaga palmarum 952 Aix sponsa Anas platyrhynchos Bucephala albeola Geothlypis trichas Larus delawarensis Leiothlypis celata Megaceryle alcyon Melospiza melodia Setophaga coronata Spinus tristis Turdus migratorius Zonotrichia leucophrys Branta canadensis 917 Anas platyrhynchos Bucephala albeola Buteo lineatus Geothlypis trichas Larus delawarensis Leiothlypis celata Melospiza melodia Setophaga coronata Setophaga palmarum Turdus migratorius Zonotrichia leucophrys Aix sponsa Branta canadensis Leiothlypis ruficapilla Molothrus ater 987 Aix sponsa Bucephala albeola Geothlypis trichas Haliaeetus leucocephalus Leiothlypis celata Megaceryle alcyon Melospiza melodia Setophaga coronata Turdus migratorius Zonotrichia leucophrys Branta canadensis 271 Ardea alba Catharus ustulatus Hirundo rustica Icterus galbula Myiarchus crinitus Pandion haliaetus Pheucticus ludovicianus Setophaga petechia Sialia sialis Spatula discors Archilochus colubris Spizella passerina 237 Catharus ustulatus Egretta thula Hirundo rustica Myiarchus crinitus Pandion haliaetus Pheucticusludovicianus Setophaga petechia Sialia sialis Spatula discors Ardea alba 236 Catharus ustulatus Hirundo rustica Myiarchus crinitus Pandion haliaetus Pheucticus ludovicianus Setophaga petechia Sialiasialis Spatula discors Troglodytes aedon Icterus galbula Shi Feng et al. – Stopover Hotspots in North and Central America 29 Top 10 important stopovers Potential alternative stopover sites Cell-ID Species Cell-ID Species Cell-ID Species 624 Buteo jamaicensis Buteo lineatus Geothlypis trichas Leiothlypis ruficapilla Myiarchus crinitus Pandion haliaetus Pheucticus ludovicianus Podilymbus podiceps Polioptila caerulea Sayornis phoebe Setophaga americana Setophaga palmarum Stelgidopteryx serripennis Dumetella carolinensis Icterus galbula Spatula discors Catharus ustulatus Hirundo rustica Quiscalus quiscula 659 Buteo lineatus Geothlypis trichas Icterus galbula Myiarchus crinitus Pandion haliaetus Polioptila caerulea Sayornis phoebe Setophaga americana Setophaga coronata Setophaga palmarum Dumetella carolinensis Podilymbus podiceps Quiscalus quiscula Stelgidopteryx serripennis Buteo jamaicensis Catharus ustulatus Hirundo rustica Leiothlypis ruficapilla Megaceryle alcyon Pheucticus ludovicianus Spizella passerina 694 Buteo lineatus Egretta thula Geothlypis trichas Icterus galbula Megaceryle alcyon Myiarchus crinitus Pandion haliaetus Pheucticus ludovicianus Podilymbus podiceps Polioptila caerulea Sayornis phoebe Setophaga americana Setophaga coronata Setophaga palmarum Spizella passerina Stelgidopteryx serripennis Dumetella carolinensis Pipilo erythrophthalmus Quiscalus quiscula Toxostoma rufum Buteo jamaicensis Catharus ustulatus Hirundo rustica Passerina cyanea Sialia sialis 236 Catharus ustulatus Hirundo rustica Myiarchus crinitus Pandion haliaetus Pheucticus ludovicianus Setophaga petechia Sialia sialis Spatula discors Troglodytes aedon Icterus galbula 235 Hirundo rustica Catharus ustulatus Setophaga petechia 237 Catharus ustulatus Egretta thula Hirundo rustica Myiarchus crinitus Pandion haliaetus Pheucticus ludovicianus Setophaga petechia Sialia sialis Spatula discors Ardea alba 918 Aix sponsa Anas platyrhynchos Bucephala albeola Geothlypis trichas Larus delawarensis Leiothlypis celata Leiothlypis ruficapilla Megaceryle alcyon Melospiza melodia Pandion haliaetus Setophaga coronata Sturnus vulgaris Turdus migratorius Zonotrichia leucophrys Branta canadensis Colaptes auratus Spinus tristis Ardea herodias Mareca strepera Molothrus ater 883 Aix sponsa Anas platyrhynchos Ardea herodias Bucephala albeola Geothlypis trichas Leiothlypis celata Leiothlypis ruficapilla Pandion haliaetus Setophaga coronata Setophaga palmarum Sturnus vulgaris Zonotrichia leucophrys Branta canadensis Melospiza melodia Turdus migratorius Molothrus ater 952 Aix sponsa Anas platyrhynchos Bucephala albeola Geothlypis trichas Larus delawarensis Leiothlypis celata Megaceryle alcyon Melospiza melodia Setophaga coronata Spinus tristis Turdus migratorius Zonotrichia leucophrys Branta canadensis Shi Feng et al. – Stopover Hotspots in North and Central America 30 Table S3. Cell-IDs with the included species for 2020-2022 Top 10 important stopovers Potential alternative stopover sites Cell-ID Species Cell-ID Species Cell-ID Species 237 Catharus ustulatus Myiarchus crinitus Pandion haliaetus Pheucticus ludovicianus Setophaga petechia Spatula discors Hirundo rustica Icterus galbula Ardea alba 236 Catharus ustulatus Hirundo rustica Icterus galbula Pheucticus ludovicianus Setophaga petechia Myiarchus crinitus Spatula discors 165 Setophaga petechia 589 Archilochus colubris Buteo jamaicensis Dumetella carolinensis Geothlypis trichas Myiarchus crinitus Pandion haliaetus Pheucticus ludovicianus Polioptila caerulea Quiscalus quiscula Sayornis phoebe Setophaga americana Setophaga palmarum Setophaga petechia Spatula discors Ardea herodias Catharus ustulatus Icterus galbula Leiothlypis ruficapilla Podilymbus podiceps Stelgidopteryx serripennis 659 Ardea herodias Buteo lineatus Dumetella carolinensis Geothlypis trichas Icterus galbula Myiarchus crinitus Pandion haliaetus Pheucticus ludovicianus Podilymbus podiceps Polioptila caerulea Quiscalus quiscula Sayornis phoebe Setophaga americana Setophaga coronata Setophaga palmarum Setophaga petechia Spizella passerina Leiothlypis ruficapilla Sialia sialis Spatula discors Stelgidopteryx serripennis Toxostoma rufum Egretta thula 624 Buteo jamaicensis Buteo lineatus Dumetella carolinensis Geothlypis trichas Icterus galbula Leiothlypis ruficapilla Myiarchus crinitus Pandion haliaetus Pheucticus ludovicianus Polioptila caerulea Quiscalus quiscula Sayornis phoebe Setophaga americana Setophaga palmarum Setophaga petechia Spatula discors Ardea herodias Podilymbus podiceps Stelgidopteryx serripennis 554 Archilochus colubris Buteo jamaicensis Dumetella carolinensis Egretta thula Myiarchus crinitus Pandion haliaetus Passerina cyanea Pheucticus ludovicianus Setophaga americana Setophaga palmarum Setophaga petechia Spatula discors Stelgidopteryx serripennis Ardea herodias Catharus ustulatus Fulica americana Icterus galbula Leiothlypis ruficapilla Polioptila caerulea Quiscalus quiscula 519 Archilochus colubris Buteo jamaicensis Leiothlypis ruficapilla Myiarchus crinitus Passerina cyanea Podilymbus podiceps Setophaga americana Setophaga petechia Stelgidopteryx serripennis Ardea herodias Catharus ustulatus Pheucticus ludovicianus Setophaga palmarum Dumetella carolinensis Fulica americana Icterus galbula 624 Buteo jamaicensis Buteo lineatus Dumetella carolinensis Geothlypis trichas Icterus galbula Leiothlypis ruficapilla Myiarchus crinitus Pandion haliaetus Pheucticus ludovicianus Polioptila caerulea Quiscalus quiscula Sayornis phoebe Setophaga americana Setophaga palmarum Setophaga petechia Spatula discors Ardea herodias Podilymbus podiceps Stelgidopteryx serripennis Shi Feng et al. – Stopover Hotspots in North and Central America 31 Top 10 important stopovers Potential alternative stopover sites Cell-ID Species Cell-ID Species Cell-ID Species 520 Buteo jamaicensis Egretta thula Leiothlypis ruficapilla Passerina cyanea Setophaga americana Setophaga petechia Catharus ustulatus Myiarchus crinitus Podilymbus podiceps Fulica americana Icterus galbula Megaceryle alcyon Pandion haliaetus Setophaga palmarum 519 Archilochus colubris Buteo jamaicensis Leiothlypis ruficapilla Myiarchus crinitus Passerina cyanea Podilymbus podiceps Setophaga americana Setophaga petechia Stelgidopteryx serripennis Ardea herodias Catharus ustulatus Pheucticus ludovicianus Setophaga palmarum Dumetella carolinensis Fulica americana Icterus galbula 485 Buteo jamaicensis Passerina cyanea Setophaga americana Setophaga petechia Catharus ustulatus Fulica americana Pandion haliaetus 519 Archilochus colubris Buteo jamaicensis Leiothlypis ruficapilla Myiarchus crinitus Passerina cyanea Podilymbus podiceps Setophaga americana Setophaga petechia Stelgidopteryx serripennis Ardea herodias Catharus ustulatus Pheucticus ludovicianus Setophaga palmarum Dumetella carolinensis Fulica americana Icterus galbula 520 Buteo jamaicensis Egretta thula Leiothlypis ruficapilla Passerina cyanea Setophaga americana Setophaga petechia Catharus ustulatus Myiarchus crinitus Podilymbus podiceps Fulica americana Icterus galbula Megaceryle alcyon Pandion haliaetus Setophaga palmarum 485 Buteo jamaicensis Passerina cyanea Setophaga americana Setophaga petechia Catharus ustulatus Fulica americana Pandion haliaetus 952 Aix sponsa Bucephala albeola Buteo lineatus Geothlypis trichas Leiothlypis celata Megaceryle alcyon Melospiza melodia Spinus tristis Turdus migratorius Zonotrichia leucophrys Anas platyrhynchos 917 Anas platyrhynchos Branta canadensis Bucephala albeola Buteo lineatus Leiothlypis celata Melospiza melodia Spinus tristis Turdus migratorius Zonotrichia leucophrys Aix sponsa Megaceryle alcyon 987 Aix sponsa Branta canadensis Bucephala albeola Colaptes auratus Geothlypis trichas Icterus galbula Leiothlypis celata Megaceryle alcyon Melospiza melodia Spinus tristis Turdus migratorius Zonotrichia leucophrys Anas platyrhynchos Haliaeetu leucocephalus Larus delawarensis 271 Catharus ustulatus Hirundo rustica Icterus galbula Myiarchus crinitus Pandion haliaetus Pheucticus ludovicianus Setophaga petechia Spatula discors Stelgidopteryx serripennis Ardea alba Spizella passerina 237 Catharus ustulatus Myiarchus crinitus Pandion haliaetus Pheucticus ludovicianus Setophaga petechia Spatula discors Hirundo rustica Icterus galbula Ardea alba 236 Catharus ustulatus Hirundo rustica Icterus galbula Pheucticus ludovicianus Setophaga petechia Myiarchus crinitus Spatula discors Shi Feng et al. – Stopover Hotspots in North and Central America 32 Top 10 important stopovers Potential alternative stopover sites Cell-ID Species Cell-ID Species Cell-ID Species 624 Buteo jamaicensis Buteo lineatus Dumetella carolinensis Geothlypis trichas Icterus galbula Leiothlypis ruficapilla Myiarchus crinitus Pandion haliaetus Pheucticus ludovicianus Polioptila caerulea Quiscalus quiscula Sayornis phoebe Setophaga americana Setophaga palmarum Setophaga petechia Spatula discors Ardea herodias Podilymbus podiceps Stelgidopteryx serripennis 659 Ardea herodias Buteo lineatus Dumetella carolinensis Geothlypis trichas Icterus galbula Myiarchus crinitus Pandion haliaetus Pheucticus ludovicianus Podilymbus podiceps Polioptila caerulea Quiscalus quiscula Sayornis phoebe Setophaga americana Setophaga coronata Setophaga palmarum Setophaga petechia Spizella passerina Leiothlypis ruficapilla Sialia sialis Spatula discors Stelgidopteryx serripennis Toxostoma rufum Egretta thula 694 Ardea herodias Buteo lineatus Dumetella carolinensis Egretta thula Geothlypis trichas Icterus galbula Megaceryle alcyon Myiarchus crinitus Pandion haliaetus Podilymbus podiceps Polioptila caerulea Quiscalus quiscula Sayornis phoebe Setophaga americana Setophaga petechia Spizella passerina Leiothlypis ruficapilla Setophaga coronata Sialia sialis Toxostoma rufum Setophaga palmarum Turdus migratorius 236 Catharus ustulatus Hirundo rustica Icterus galbula Pheucticus ludovicianus Setophaga petechia Myiarchus crinitus Spatula discors 235 Catharus ustulatus Hirundo rustica Icterus galbula 237 Catharus ustulatus Myiarchus crinitus Pandion haliaetus Pheucticus ludovicianus Setophaga petechia Spatula discors Hirundo rustica Icterus galbula Ardea alba 918 Aix sponsa Ardea herodias Branta canadensis Bucephala albeola Colaptes auratus Geothlypis trichas Megaceryle alcyon Melospiza melodia Pandion haliaetus Spinus tristis Sturnus vulgaris Turdus migratorius Zonotrichia leucophrys Anas platyrhynchos Leiothlypis celata Leiothlypis ruficapilla Molothrus ater Setophaga coronata 883 Aix sponsa Anas platyrhynchos Ardea herodias Branta canadensis Bucephala albeola Colaptes auratus Geothlypis trichas Megaceryle alcyon Molothrus ater Spinus tristis Sturnus vulgaris Turdus migratorius Zonotrichia leucophrys Leiothlypis celata Leiothlypis ruficapilla Melospiza melodia Setophaga coronata 952 Aix sponsa Bucephala albeola Buteo lineatus Geothlypis trichas Leiothlypis celata Megaceryle alcyon Melospiza melodia Spinus tristis Turdus migratorius Zonotrichia leucophrys Anas platyrhynchos