Destruction of habitats through direct human exploitation is the greatest contemporary cause of terrestrial biodiversity declines (Newbold et al., 2015). Land-use is considered to be one of the most important drivers of biodiversity, affecting patterns of species diversity (Jetz et al., 2007), distributions (Wilcove et al., 1998; Opdam and Wascher, 2004; Tingley and Beissinger, 2013; Regos et al., 2018), and ecological processes (Dickinson et al., 1991; Dale, 1997; Allan, 2004). However, the influence of recent climate change is also exerting a clear influence on species' populations and biodiversity (Parmesan, 2006; Both et al., 2009; Gutiérrez Illán et al., 2014) and these effects are expected to intensify over the coming decades (Thomas et al., 2004).

Until recently, the effects of these two anthropogenic stressors have been considered in isolation, but there are strong theoretical reasons to expect that they can interact synergistically to drive biodiversity declines (Travis, 2003; Opdam and Wascher, 2004). First, populations that are already in decline due to habitat loss may be less likely to behaviorally adapt to climate change; lower recruitment reduces the number of natal dispersers, and therefore the potential for prospecting and colonization of newly available habitat at the range edge (Holt and Keitt, 2000). Second, fragmentation of native habitats can directly reduce functional connectivity of landscapes (Taylor et al., 1993), thereby preventing the potential for metapopulation persistence (Hanski and Parmesan, 1999) and range shifting (Mair et al., 2014). Third, loss or degradation of certain habitat types (e.g., forest) can cause regional rainfall reductions, thereby intensifying negative climate effects (Lawrence and Vandecar, 2015).

Unfortunately, the relative roles of climate and land-use in affecting species distributions and population trends over relatively short time periods (<50 years) is not well-known (Mantyka-Pringle et al., 2012). It has been hypothesized that climate influences distributions at broad scales and over the long-term, whereas the influence of land-use is shorter term and at finer spatial scales (Lemoine et al., 2007; Soberón, 2007). However, this hypothesis has been difficult to test. Previous efforts to understand the cumulative effects of climate and land-cover changes on biodiversity have typically relied on a niche modeling approach whereby current or past species distributions are statistically linked to climate (Peterson et al., 2002a), land-cover (Lawler et al., 2014), or both (Jetz et al., 2007; Ponce-Reyes et al., 2013; Sohl, 2014). These associations are then projected forward under various climate change and land-use scenarios to provide estimates of species vulnerability. Changes in climate and land-cover are implicitly assumed to cause change in species ranges (Jantz et al., 2015), with potential impacts on natural system functioning and ecosystem services (Botkin et al., 2007).

Several hurdles have limited empirical testing of these projections. First, available past, current and projected land-cover data are typically available only in coarse-resolution human-defined categories (e.g., forest, agriculture, urban). These categorical data may have little bearing on how individual species are distributed across landscapes and often predict occurrence poorly (Shirley et al., 2013; Betts et al., 2014). Further, vegetation data must be at sufficiently fine resolutions to match the spatial scales experienced by species (Wiens, 1989), but such data are rarely available at broad spatial scales associated with species' ranges and climate data.

Second, and most importantly, climate and land-cover tend to be highly correlated over short time periods and broad spatial scales; precipitation and temperature are well-known to influence vegetation type and cover, which makes it challenging to attribute the cause of species distributions to either climate or land-cover change (Pearson and Dawson, 2003). Darwin (1859) foreshadowed this problem over 150 years ago: “… When we travel from south to north, or from a damp region to a dry, we invariably see some species gradually getting rarer and rarer, and finally disappearing; and the change of climate being conspicuous, we are tempted to attribute the whole effect to its direct action. But this is a very false view …”.

Accelerated rates of both climate (Karl et al., 2015) and land-use change (Hansen et al., 2013) offer the potential to test the independent roles of each of these factors in affecting population trends. Though climate and land-cover are usually highly correlated, this is not necessarily the case for changes in these potential stressors; areas with the most habitat loss (e.g., tropical forest; Hansen et al., 2013) have not necessarily experienced the greatest accelerations in climate change (Chen et al., 2011; Mantyka-Pringle et al., 2012). Temporal autocorrelation in both climate and land cover are likely to result in overly optimistic estimates of the role of these variables in driving species abundance and distributions. Locations in the landscape that have relatively high abundance in time t are likely to still have high abundance in time t₂. For example, predicting long-lived tree species abundance distributions as a function of climate variables and then testing these predictions on data 75 years later should result in a high degree of concordance between predicted and observed distributions (Dobrowski et al., 2011). Predicting population trajectories as a function environmental change is substantially more challenging (Ehrlen and Morris, 2015) because current population sizes provide no information about the direction of future change (i.e., both large and small populations may increase or decrease). Thus, if a model is successful at predicting population change, it is more likely to reflect the drivers of species' abundance distributions.

Here, we capitalize on this decoupling between climate and land-cover changes to build models that predict the abundance of 101 bird species over 27 years (1983–2010) in the western United States as a function of climate, land-cover and their combined effects. We then provide, to our knowledge, the first empirical test of the degree to which predicted changes in species abundances, as a function of land cover and climate, reflect observed trends. Predicting animal numbers, rather than distributions, is of critical importance because it is abundance, rather than occupancy, that is the greater determinant of extinction risk (Ehrlen and Morris, 2015). Specifically, we asked: (1) Do land-cover, climate or their combined effects best predict bird population trends over time? (2) Are there spatial “hotspots” where declines or increases are more common across species? (3) Are life history and/or ecological traits associated with sensitivity to climate vs. land-cover change? (4) Which species attributes are associated with high vs. low-performing models?

Overall, climate, land-cover, and combined models all described bird species abundances with some accuracy; correlations between predicted and observed bird data in the 1980s time period were high (climate ρ¯ = 0.74 [0.28–0.99], land-cover ρ¯ = 0.75 [0.12–0.98], combined ρ¯ = 0.82 [0.25–0.99] (Figure 2).

Predicting trends in bird abundance over the 30-year period was more challenging; changes in climate significantly predicted observed abundance trends for 44/108 species (ρΔ¯ = 0.37 ± 0.09 [SD]) and land-cover changes alone significantly predicted trends for 47/108 species (ρΔ¯ = 0.36 ± 0.09). Land-cover trend predictions were rarely superior to climate models; only 8 species had statistically significant observed trend correlations that were best predicted by land-cover. However, the synergistic effects of climate and land-cover significantly predicted trends for 59/108 species (ρΔ¯ = 0.37 ± 0.11) and for 37 of these species, including information on land-cover increased trend prediction success over climate data alone (Figure 3). We rarely detected significant spatial autocorrelation in residuals of observed vs. predicted trend correlations (Figure S2) and the magnitude of autocorrelation was consistently low (Moran's I < 0.2).

Accurate models for predicting future changes in species distributions and population trends are essential for understanding the effects of global change and in policy development (Mazziotta et al., 2015; Thomas, 2017; Titeux et al., 2017). Such models are particularly relevant considering that the Earth is likely experiencing a sixth mass extinction event, mostly due to anthropogenic factors (Barnosky et al., 2011; Ceballos et al., 2015; Thomas, 2017).

Considerable efforts have been made in recent years to assess the effects of the diverse aspects of global change on the ecosystems (Pearson and Dawson, 2003). However, most of these studies focused on the evaluation of each component of global change in isolation (Keith et al., 2014; Ostberg et al., 2015; Pacifici et al., 2015, but see Marshall et al., 2018). Changes in climate and land use may exert a greater joint effect on ecosystems biota and ecosystem services than when considered individually (Gutiérrez Illán et al., 2012; Mantyka-Pringle et al., 2012; McCauley et al., 2015; Anteau et al., 2016). If this is the case for most species, predictions about the real effects of anthropogenic global change, regardless of their spatio-temporal scale, may have been underestimated (Stanton et al., 2012). This is why it is surprising that most projections of future species distributions for different parts of the planet are primarily based on climate change scenarios, without taking into account land-use change (e.g., Bakkenes et al., 2002; Peterson et al., 2002b; Thomas et al., 2004), let alone the synergistic effects of climate and land use. Thus, there is a pressing need to develop species distribution models that reflect and predict the impact of these potentially synergistic factors. In the present study, we examined the potential synergistic effects of climatic and land-cover change factors on populations of 108 bird species of the Pacific Northwest in North America over the last three decades. To our knowledge, ours is the first study to formally test the prediction success of land cover and climate models using a model-validation forecasting approach.

Our results indicate that future and observed effects of global change on avian biodiversity in the Pacific Northwest may be greater than the sum of the effects caused by climate and land-use changes separately. As expected, the inclusion of land-cover change improved the capacity of our models to estimate bird abundance in a single time period. If such models were validated using only data from the same place and time, it is possible that inclusion of more variables simply resulted in overfitting. Indeed, our results from cross-validation in the same time period (1983–86) showed relatively high model performance (Figure 2). However, our most rigorous validation tested whether models could predict species abundance trends, a much more challenging task. For 37 species, including land-cover information improved abundance change predictions.

Overall, synergistic effects of climate and land-cover were able to significantly predict abundance changes for more than half of the analyzed species (59 out of 108). Although this is cause for some optimism, it is important to note that modeled predictions for 45% of species did not even significantly correlate with observed changes. Such a result should temper existing unvalidated projections about species responses to climate and land-cover change (e.g., Thuiller et al., 2004; Jetz et al., 2007, Ponce-Reyes et al., 2013; Sohl, 2014; Jantz et al., 2015). This underscores the reality that even though a model may be well-validated in a single place and time (e.g., via cross-validation), or even in different regions (Betts et al., 2006), this is not necessarily an indication that models will effectively predict future trends.

The relatively low explanatory power of our models in relation to previous studies could stem from the following: (i) most of the studies have looked at occupancy predictions, which constitutes only a pseudo-validation. Predicting actual trends is a much more challenging task (Regos et al., 2018). (ii) The spatial scale at which the species' responses to changes in climate and land-use change occur are undoubtedly finer resolution than we were able to measure with current data (e.g., microclimate under canopy, fine-resolution habitat associations, cold-air pooling, etc.; Frey et al., 2016; Betts et al., 2018). Mismatches between predictions and reality likely stem from missing key climate or habitat variables that were perhaps initially correlated with the variables used in the analysis, but have become decoupled over time (for example, perhaps a fine-scale habitat feature such as cavity trees were associated with a certain reflectance band in time t, but this is no longer the case in time t +1). (iii) There are obviously other factors, well known to influence populations (e.g., dispersal, species interactions, disease), that were not included in our models.

It is reassuring, to some extent, that the capacity of our models to predict bird population changes was improved for species with declining, rather than increasing, populations. This indicates that we may have a better idea about population stressors for species of potential conservation concern, which at least offers the potential for management actions to reverse declines. We hypothesize that this finding results from the rather deterministic process of decline and local extinction of species if the climate changes to exceed the species' niche. However, colonization and population increases in a landscape may be more dependent on less predictable processes that are often a function of landscape configuration such as dispersal (Gustafson and Gardner, 1996). Thus, species with expanding ranges and populations may be inherently more difficult to predict.

All authors conceived the ideas and designed methodology. MB, JG, and ZY led the data analyses. MB wrote the first draft of the manuscript with major contributions from JG. All authors contributed critically to the drafts and gave final approval for publication.