Migratory species, by definition, occupy different ranges across the annual cycle. Focusing on only one stage of the annual cycle unintentionally limits our ability to obtain accurate predictions regarding the change of future species’ distributions. We found that SDMs for migratory birds do exactly that – the majority of SDMs from our review considered data from only one period of the annual cycle. Migratory species often have different habitat requirements and nutritional needs and are exposed to different environmental factors and threats during different seasons of their annual cycles (Small-Lorenz et al., 2013; Fink et al., 2020; Lamb et al., 2020). Therefore, an SDM focusing exclusively on modeling a species’ distribution during the breeding period, for example, without data from other periods of the annual cycle provides likely inaccurate information about the species’ range-wide distributional shifts in response to global change.

Surprisingly, we did not find a seasonal bias towards SDMs focused on the breeding period compared to the stationary non-breeding period. However, we did find support for a bias towards stationary periods of the annual cycle (breeding and nonbreeding) compared to non-stationary periods (spring and autumn migration). Mortality is often highest during non-stationary periods (Sillett and Holmes, 2002; Klaassen et al., 2014; Rushing et al., 2017; but see Conklin et al., 2017; Senner et al., 2019) and certain threats (e.g., collisions with buildings and power lines, cats) can impact migratory birds heavily during migration, whereas these risks pose less of a danger during stationary periods (Loss et al., 2015). Future research with tracking technologies should focus on uncovering patterns of range-wide migratory connectivity for migratory birds so that we can better identify population linkages and exposures to climate and environmental threats for connected populations.

A host of competing abiotic and biotic effects throughout the annual cycle differentially affect population dynamics and operate over large spatiotemporal scales to determine patterns of distribution and abundance (Gaston, 2009; Sexton et al., 2009). Seasonal interactions (lethal or nonlethal events in one season that carry-over and influence individuals or populations in a subsequent season) are inherent properties of these systems and are vital for understanding how events and selective pressures operate throughout the annual cycle to ultimately drive population dynamics and species distributions (Norris and Marra, 2007). Yet, most studies focus solely on one stage of the annual cycle, which remains the dominant framework for investigating species distributions. In large part, this is likely driven by a heavy breeding-centric seasonal bias in where research is conducted, and also when occurrence/monitoring data is collected (Marra et al., 2015). Although these biases are improving, we still found that of the 201 SDMs examined, nearly 55% only included factors from one stage of the annual cycle and of those, most were conducted during the breeding season only. Yet, the one study that did incorporate seasonal interactions (Aagaard et al., 2022) found that seasonal interactions were critical - environmental conditions operating outside of the season of study played an important role in influencing the abundance and distribution of their study species.

By ignoring seasonal interactions, studies that examine and attempt to predict future species distributions are missing critical drivers, hindering accurate prediction (Illán et al., 2014; Lee-Yaw et al., 2022). More to the point, they potentially provide flawed guidance for climate adaptation and conservation planning (Small-Lorenz et al., 2013). Environmental conditions experienced across the annual cycle vary drastically from season to season and are often undergoing different rates of climatic change. For example, in the Neartic-Neotropical migratory flyway, conditions on the temperate breeding grounds have warmed significantly leading to advances in the timing of breeding for many species (Brown et al., 1999; Dunn and Winkler, 1999; Townsend et al., 2013; Shipley et al., 2020) and have been suspected of driving species ranges north (Langham et al., 2015). Yet on the tropical nonbreeding grounds, continued and progressive drought (Neelin et al., 2006; Herrera and Ault, 2017) is delaying the start of migration (Studds and Marra, 2011; Rockwell et al., 2012) and strongly influencing survival patterns (Rockwell et al., 2017; Dossman et al., 2023a). These differences in climate change and thus climate exposure across different periods of the annual cycle almost certainly differentially impact changes in abundance and distribution of a species on the breeding grounds (Carey, 2009). The overall direction and magnitude in these distributional changes are likely to represent the cumulative effect of full annual cycle events on the demography of a species across its range (Dossman et al., 2023b). Accurately predicting complex future patterns of range shifts and distributional changes will require integration of full annual cycle events at the level of single species. Complex life history events drive the natural history of these species and ignoring these features dangerously oversimplifies future predictions. To date, no species distribution models have formally integrated seasonal interactions into their design and as such remain a key limitation towards advancing our understanding of how species distributions will respond in the face of climate change.

We found evidence for a geographic bias in current SDMs towards migratory birds in North America and Europe. Our findings are consistent with results from other studies spanning various disciplines within ecology and conservation research (Feldman et al., 2021). This geographic bias likely stems from a global imbalance in access to research and funding. Due to a multitude of social, political, and economical factors (Culumber et al., 2019), conservation and climate change research activity and funding occurs primarily in the United States, Canada, and European nations (Pasgaard et al., 2015; Di Marco et al., 2017). This is not to imply that research is not occurring in other regions of the world (Soares et al., 2023). In fact, there have been recent calls to acknowledge work from underrepresented regions where the ability to conduct and disseminate research is frequently inhibited due to language hegemony, journal access, and disproportionately lower access to research funding (Soares et al., 2023). Indeed, we conducted our literature review in English, so we may have missed studies on species distribution modeling for migratory birds published in other languages.

As a result of this geographic bias, much of what we understand about the distributional responses of migratory birds to global change comes from North American and European models (Stephens et al., 2016). These models typically conclude that the breeding ranges of migratory birds are shifting poleward (i.e., north) or to higher elevations via tracking of their preferred niche (Hitch and Leberg, 2007; Chen et al., 2011; Brommer et al., 2012). Although this likely holds true for some migratory species, recent evidence suggests the opposite pattern (i.e., southward range shifts) could be occurring for others due to climate change on the nonbreeding grounds (Dossman et al., 2023b), or that the southern range limit could be contracting with no concurrent shifts in the northern range limit (Rushing et al., 2020). Instead of moving, migratory birds may also track their preferred niche by altering the phenological onset of key life history events (e.g., breeding; Socolar et al., 2017).

A corollary of the geographic bias towards North America and Europe is that current SDMs also show a bias towards modeling terrestrial migratory birds versus marine species. Historically, building SDMs for seabirds in general – and particularly migratory seabirds – proved challenging due to sparse distributional data (toward breeding season and nesting sites), limited environmental information, and a comparatively poorer understanding of movement patterns compared to terrestrial species (Engler et al., 2017). Fortunately, these limitations have become increasingly less problematic thanks to the collection and dissemination of publicly-available citizen science and tracking data (e.g., Fauchald et al., 2019). Moreover, commonly used environmental data layers like sea surface temperature can now be combined with more advanced environmental data derived from remote sensing and satellite imagery (Engler et al., 2017). Combining these datasets should facilitate the development of better informed SDMs for migratory seabirds.

Most SDMs modeled songbirds, waterfowl, and shorebirds, which reflect the groups of birds most abundant globally (Callaghan et al., 2021) and the avian orders with the highest species richness - 60% of all avian species are in the Order Passeriformes (Avonet database; Tobias et al., 2022). However, both waterfowl and shorebirds make up 2% and 3%, respectively, of all avian species globally (Avonet database; Tobias et al., 2022). As a result, while we have a relatively better understanding of the drivers influencing species distributions for songbirds, waterfowl, and shorebirds, this taxonomic bias underscores the gap left for more speciose orders.

In addition to showing a taxonomic bias, the majority of SDM studies were conducted for obligate migratory species. The bias towards obligate migratory species combined with the geographic bias towards North American and European breeding species is not surprising; long-distance migration is common in strongly seasonal environments characteristic of temperate zones (Winger et al., 2014; Winger et al., 2019). This bias may also be partially explained by the dearth of studies on species that may engage in less dramatic (e.g., within rather than across continents) or predictable migrations (Jahn et al., 2012). While the majority of studies on migration focus on obligate, long-distance migrations and geographic regions (reviewed by Jahn et al., 2004), species that are partially migratory, irruptive, or migrate over shorter distances represent most species globally (Levey, 1994; Chapman et al., 2011; Hsiung et al., 2018). Thus, current SDMs are overlooking the vast majority of migratory bird species.

Collectively, most SDMs in this review either prioritized obligate migratory birds or, by the lack of explicitly defining what type of species they modeled (whether migrant, partial, resident, etc.), do not acknowledge the importance of specifying migratory behavior in modeling where species may be across space and time. This is exacerbated by the fact that many of these studies oversimplify the variation in migratory behavior present in species across their range or annual cycle (Berthold, 2001). Different populations of the same species may differ in when, how far, and if they even migrate at all in a given year (e.g.,Taylor and Norris, 2007; Jahn et al., 2019). This variability in potential migratory behaviors has major ramifications for accurately predicting where and when a species is exposed to a particular climate or threat (Nilsson et al., 2006). Furthermore, over a third of all studies we reviewed modeled multiple species simultaneously, many of which exhibit interspecific variation in migratory behavior, which was not accounted for in models. Interspecific variation in migration influences species’ exposure and resilience to global change by nature of differences in species-level ecological or life-history traits (Jiguet et al., 2007; Stevens et al., 2023), migratory connectivity (Patchett et al., 2018), and whether a species may be facultative or obligate across its range (Gilroy et al., 2016). Understanding how individuals and populations of species may be spatially and temporally connected throughout the year is critical to determining a species’ risk to threats, which ultimately shape population dynamics (Bauer et al., 2016). Thus, including variation in migratory behavior and connectivity within and among individuals and populations of the same and different species will greatly improve SDMs, and therefore also our ability to predict, target threats, and effectively manage and conserve species in the face of climate change and other drivers of global change.

Human-driven changes such as habitat loss, climate change and other direct sources of mortality are major drivers of decline in migratory bird populations (Robbins et al., 1989; Loss et al., 2015; Rigal et al., 2023) and are increasingly expected to play an important role in influencing the distribution of bird species (Jetz et al., 2007). Despite the potential impact of anthropogenic activities on migratory bird distributions, there were no examples of SDMs that incorporated threats from connected, spatially-explicit populations throughout the annual cycle, and only one third of the SDMs incorporated anthropogenic factors at all. Furthermore, of those SDMs that incorporated anthropogenic factors, the vast majority only incorporated habitat loss and/or climate change as potential predictor variables. By contrast, direct sources of mortality were only represented in 5% of SDMs that included anthropogenic factors. Interestingly, a number of studies assessing the exposure of bird populations to direct sources of anthropogenic threats did not incorporate anthropogenic factors directly into their SDMs, but instead, overlaid their predicted species distributions (i.e., SDM output) and threat layers in separate spatial analyses (Loring et al., 2014; Pereira et al., 2018). In spite of growing recognition of the impact of human influence on the distribution of species, most models do not appear to directly incorporate anthropogenic factors, which could undermine forecasts of future distributions (Yates et al., 2018).

Challenges exist for incorporating anthropogenic factors into SDMs. For many species, the relationship between anthropogenic activities and species distributions are not well understood and likely varies heterogeneously across spatiotemporal scales (Guisan and Thuiller, 2005; Zuckerberg et al., 2016). For example, researchers have found that artificial light at night can drive migratory bird distributions but the effect varies seasonally and across spatial scales (La Sorte et al., 2017; McLaren et al., 2018). Therefore, care must be taken to ensure that the resolution of data sources match the scale of human impact on species distribution. Additionally, standardized datasets on anthropogenic factors, particularly direct sources of mortality, are not widely available, limiting their potential incorporation as predictors into SDM projects (Loss et al., 2015). Datasets for habitat attributes and climate variables exist across multiple spatiotemporal scales and resolutions, which could explain why we found the majority of SDMs incorporating anthropogenic factors addressed these two categories of threats (Wüest et al., 2020; Zurell et al., 2023). Despite the paucity of standardized datasets for anthropogenic factors, over one third of studies (39%) were able to include anthropogenic factors by incorporating proxies of these threats. For example, Araujo et al. (2022) used distance to port as a proxy for disturbance from fishing boats and Adams and Root (2022) used building density as a proxy for disturbance from window collisions. Given the rapidly increasing pressure from a growing human population, the omission of anthropogenic processes from SDMs represents a major limitation to our ability to understand and forecast the distribution of species across multiple scales.

Full annual cycle SDMs are not a novel concept. Researchers have long acknowledged the importance of accounting for the annual cycle in work involving migratory species (Temple, 1988; Faaborg et al., 2010). Recently, the Cornell Lab of Ornithology unveiled a lineup of full annual cycle SDMs for hundreds of bird species in North America with sufficient observation data from eBird (Fink et al., 2020; Fink et al., 2022). These eBird full annual cycle SDMs address many of the limitations we describe above. Namely, they include relevant threat layers affecting all bird species (e.g., nighttime lights, Cao et al., 2014; land cover and land use classifications, Friedl and Sulla-Menashe, 2019) in addition to other standard predictor variables, can be applied to any species with sufficient observation data, and model the relative abundance of the species throughout the annual cycle using ensemble modeling (i.e., multiple, overlapping models based on specific time frames). While this novel approach represents a monumental achievement for migratory bird SDMs, it still lacks two critical aspects affecting the distribution of migratory birds around the world – migratory connectivity and seasonal interactions.

Only one study in our review integrated the effects of seasonal interactions into SDMs (Aagaard et al., 2022), making an important advance to developing full annual cycle SDMs for migratory birds. Notably, this study did not use any novel modeling designs to integrate seasonal interactions – they simply modified an existing mechanistic SDM framework (i.e., energetics-based individual based model; Aagaard et al., 2022). Moreover, Aagaard et al. (2022) reported that seasonal interactions played an important role in influencing distributional change for their study species, North American Mallards (Anas platyrhynchos). The fact that Aagaard et al. (2022) modeled mallards, a well-studied migratory bird species, using an existing SDM framework suggests that the key limitation inhibiting the development of full annual cycle SDMs is not necessarily a lack of modeling tools but instead may be missing abundance data across the annual cycle, information on patterns of migratory connectivity, and what controls distributional change for migratory birds (Howard et al., 2014). Including a seasonal interaction in an SDM is as simple as, for example, linking a winter covariate to each breeding observation in a SDM modeling the breeding distribution of a migratory species. Once we have these relevant data on the full annual cycle biology for a given migratory species, developing full annual cycle SDMs could be as straightforward as modifying existing SDM frameworks. We describe one of these potential modifications below.

Bayesian hierarchical modeling frameworks seem particularly promising for building the next generation of full annual cycle SDMs. Bayesian hierarchical models allow for the specification and inclusion of multiple simultaneous processes and stochasticity intrinsic to ecological systems – e.g., the probability that an American Redstart (Setophaga ruticilla) occurs in a given environment within their breeding range depends on breeding season environmental variables like forest and insect availability, but also on climate on their nonbreeding grounds (Dossman et al., 2023b). Studies focused on integrated population models use Bayesian hierarchical modeling frameworks routinely (Hostetler et al., 2015), but instead of modeling distributional change, they model population dynamics. In fact, we excluded several papers from our review that did incorporate seasonal interactions using a Bayesian hierarchical framework but did so with population models rather than SDMs (Almaraz et al., 2012; Rushing et al., 2016). The main difference between a population model and a full annual cycle SDM is that the former focuses on modeling population change, while an SDM models the probability of occurrence during a specified time period and relies exclusively on spatially-referenced occurrence and environmental data. Given this subtle difference, it should be straightforward to tweak these population modeling frameworks and apply them as full annual cycle SDMs in cases where spatially-referenced occurrence and environmental data as well as data on migratory connectivity are available.

Bayesian hierarchical models crucially allow for the specification of informative priors. The capability to specify priors in a modeling framework serves as an ideal system to incorporate previous knowledge not only about what influences range dynamics for a species (e.g., nonbreeding rainfall; Dossman et al., 2023b), but also information about their migratory connectivity. Including information about migratory connectivity helps reconcile one of the key assumptions of all SDMs – that all individuals respond equally and predictably to environmental change throughout their range – by allowing for different populations to respond to different drivers of global change as they would naturally according to their patterns of migratory connectivity. Using priors, it could be possible to model a species distribution using joint SDMs where you (1) divide a species into regional populations which the model considers separate ‘species’ and (2) limit different populations to respond only to the environment in regions where they occur throughout the annual cycle ( Figure 5 ).

Developing full annual cycle SDMs that connect populations within a species’ range throughout the annual cycle and incorporate the effects of seasonal interactions remains an important frontier for macroecology. Challenges include first defining patterns of migratory connectivity, then determining appropriate ways for incorporating the effects of seasonal interactions, customizing and tailoring full annual cycle SDMs to the diversity of migratory behaviors exhibited by birds around the globe, and gathering relevant, global environmental data on threats. Meanwhile, we face an increasing time constraint as the drivers of global change continue to quicken, strengthen, and become more volatile (IPCC 2023). How current environmental and anthropogenic processes impact future species distributions and abundance is one of the major questions and challenges of our time. Building realistic modeling frameworks based on accurate species biology is imperative to us making advances toward these challenges.