Research Paper

INTRODUCTION

A well-documented and long-term population decline in Rusty Blackbirds (Euphagus carolinus; e.g. Link and Sauer 1996, Greenberg and Droege 1999, Avery 2000, Niven et al. 2004) has attracted attention to, and concern for the species (e.g. Greenberg et al. 2011). Evidence for that decline comes from historical data (Greenberg and Droege 1999), evaluation of Breeding Bird Survey data (Sauer et al. 2017), and evaluation of Christmas Bird Count (hereafter CBC) data (e.g. Niven et al. 2004). Although the main driver of the population decline is not known, habitat loss in the overwintering range may be a contributing factor (Greenberg et al. 2011). But, as Greenberg and Matsuoka (2010) suggest, population fluctuations outside the core overwintering area have made the species difficult to study. Additional difficulties may be due to the species’ change in habitat use over the past decades and weather-dependent changes in yearly habitat use. For example, Luscier et al. (2010) found that Rusty Blackbirds are somewhat flexible in their overwintering habitat use, and Newell Wohner et al. (2018) found that changes in winter habitat use are driven by changes in temperature and precipitation. They found that Rusty Blackbirds were more likely to use wetland forests on warmer days when there had been no rain during the previous five days. During cooler winters Rusty Blackbirds use pecan orchards but in warmer years they do not. Because habitat use by Rusty Blackbirds is influenced by weather within and between years, multiple different winter habitats are required to support overwintering populations. Further, because habitat use is weather-dependent, and because availability of different habitats may vary regionally, populations of Rusty Blackbirds could fluctuate regionally among years even if habitat availability remained constant. Thus, our ability to evaluate population change requires knowledge of weather and habitat availability in a region, so we cannot simply look at the numbers of Rusty Blackbirds counted.

We chose to compare population trends of Rusty Blackbirds among CBCs done in five of Arkansas’s physiographic regions: the Ozark Plateaus, Arkansas River Valley, Ouachita Mountains, Gulf Coastal Plain, and Mississippi Alluvial Plain. Those regions differ in the availability of habitat that is used by Rusty Blackbirds (James and Neal 1986). The two upland regions (Ozarks and Ouachitas) contain less flooded forest and bottomland hardwood habitat than the three lowland regions (Arkansas River Valley, Gulf Coast, and Mississippi), although Rusty Blackbirds are regularly recorded on CBCs in all regions (James and Neal 1986). Because of variations in the distribution and extent of available habitat across these regions, we hypothesized the presence of distinct trends in overwintering blackbird populations among the regions. Additionally, we opted to investigate the potential influence of temperature, precipitation, and cloud cover on population dynamics across these regions. Given findings of Newell Wohner et al. (2018), which indicated that temperature and precipitation impacted Rusty Blackbird habitat utilization, we anticipated dissimilar counts of blackbirds across the various regions within Arkansas. Our primary objective was to assess the interplay of temperature, precipitation, and habitat characteristics among regions on distribution and numbers of overwintering Rusty Blackbirds in Arkansas. We also wanted to determine the relative importance of temperature, precipitation, and cloud cover in influencing numbers of Rusty Blackbirds detected during Christmas Bird Counts.

METHODS

We analyzed data collected by volunteers participating in CBCs (National Audubon Society 2020). Christmas Bird Counts are conducted within a 24.14 km diameter circle. The count circle is usually divided into several sections and each section is given to a different party where each party is supposed to count all the species and individual birds in their section of the count circle. Latitude and longitude for each count circle are available along with the amount of time spent counting by each party (Bock and Root 1981). Data from 13 CBC circles in which Rusty Blackbirds were counted between 1965 and 2020 were used for our analysis (Table 1). We excluded 23 CBC circles conducted in Arkansas in which Rusty Blackbirds were recorded fewer than 11 times and two CBC circles that were not surveyed over the entire time span between 1965 and 2020 (i.e., they were either established well after 1965 or ceased being surveyed well before 2020). We chose 11 as the cutoff for the number of years in which Rusty Blackbirds were detected because we wanted to exclude count circles in which Rusty Blackbirds were unlikely to be encountered. Further, we chose 1965 as the initial count year so that we had consistent representation across Arkansas (between 1925 and 1965, 11 CBC circles were present in Arkansas but Rusty Blackbirds were reported in only 1 to 5 of those circles). Ten of the CBC circles that we included were surveyed for more than 50 years including two locations that were surveyed every year during the period; three count circles were missing between 1 and 22 survey years (Table 1). We considered using physiographic region of each count circle as a predictor, but five of the count circles spanned more than one region (Fig. 1). In addition, four of the count circle locations changed between 1965 and 2019 (Fig. 1). By using latitude and longitude of the center of each respective count circle, we were able to account for shifts in position over time.

Modeling

The counts of Rusty Blackbirds were overdispersed (variance to mean ratio = 235.0) and zero-inflated. We explored several modeling techniques to account for those modeling challenges and used Akaike’s Information Criterion (Akaike 1973) to select the best technique. Ultimately, we chose to use negative binomial generalized additive models (hereafter GAMs) because they had smaller Akaike Information Criterion (Akaike 1973) scores corrected for small sample size (hereafter AICc) and were faster and more flexible than other modeling techniques. Further, results from the other models were qualitatively similar and would not have led to different conclusions. To produce GAMs, we employed the R package “mgcv.” Our analysis aimed to unravel associations between the number of Rusty Blackbirds tallied on CBCs, and a set of predictor variables including year, effort (measured in party hours), as well as weather and habitat characteristics. We did not use numbers per party hour as our response variable for reasons outlined in Link and Sauer (1999), which we discuss in more detail below. Generalized additive models are easy to use and are able to fit complex, non-linear trends (Fewster et al. 2000, Zuur et al. 2009, Kangas et al. 2010). Further, GAMs are appropriate to evaluate count-based surveys (Sauer et al. 2004, Zuur et al. 2009). Examples of GAMs used in analyses of count data include work on farmland birds (Fewster et al. 2000), Ruffs (Calidris pugnax; Rakhimberdiev et al. 2011), Goldcrest (Regulus regulus), Greenfinch (Chloris chloris; Knape 2016), bats (Ingersoll et al. 2013), and jellyfish (Carybdea marsupialis; Canepa et al. 2017). We only included pairs of predictors whose correlation was less than 0.5 in models that contained more than one predictor. Further, we specified the restricted maximum likelihood smoothing method instead of the default generalized approximate cross validation (hereafter GACV). Restricted maximum likelihood (REML) is not sensitive to local minima (Reiss and Ogden 2009), which produced models that explained less of the deviance than did GACV. However, the trends revealed by using REML were easier to interpret than the “wigglier” trends produced by using the default GACV smoothing method. For each GAM we report deviance explained, which is a measure of fit, and effective degrees of freedom (edf), which indicates how complex (wiggly) the relationship is.

Negative binomial generalized additive models use the parameter theta to account for overdispersion (Hilbe 2011). We determined theta by sequentially fitting models with theta that varied between 0.15 and 0.3 in increments of 0.01 and selected the value of theta that minimized AIC for each model. To ensure that our models were fit using appropriate vector space (i.e., had enough basis dimensions) we used the gam.check function in R. We followed Meehan et al. (2019) in removing three extreme outliers that were > 3 standard deviations of the mean number of birds counted after log transformation. Three additional records were removed because of erroneous party hours (637.4 party hours), or probable errors (192 and 120 party hours that are > 3 sd from the mean party hours for their respective counts).

We used party hours, which is the number of hours groups spent counting birds summed for all the groups, to account for effort in all of our models. Link and Sauer (1999) pointed out that the relationship between effort and the number of birds should have an asymptote because the number of birds is limited. They proposed a Box-Cox transformation of party hours to account for the non-linear relationship between count effort and number of birds recorded. However, because GAMs are not constrained by linearity, we chose to add untransformed party hours as a covariate and allowed the model to “find” the relationship between party hours and birds counted. An advantage of using GAMs to model the effect of effort on count numbers is that GAMs can model very complex relationships. For species like Rusty Blackbirds, the relationship between count and effort might be complex because their numbers vary substantially among years (e.g., Newell Wohner et al. 2018) and because they mix with huge flocks of other species of blackbirds and starlings, which influences detectability (Strassburg 2011).

Our predictors were year, number of party hours, temperature, precipitation, and cloud cover, forest cover, and location of the CBC circles in Arkansas. To determine the relative importance of the different weather variables we compared a series of models that each contained one weather variable and party hours. Akaike Information Criterion (Akaike 1973) was used to rank models wherein we used a ΔAIC of 2 to identify the better variables (Burnham and Anderson 2002). To reduce the impact of large numbers of model parameters on AIC values we used AIC corrected for small sample size (hereafter AICc) and used package “AICcmodavg” to obtain AICc values. We compared AICc values for those models, to assess the relative importance of cold versus warm temperatures, precipitation, and cloud cover.

We used multivariate GAMs to assess the influence of temperature, precipitation, and habitat characteristics among regions on numbers of Rusty Blackbirds. Latitude and longitude of each count circle’s center were used to indicate the location, and as above, number of party hours was included to represent effort. Other than the intercepts and categorical variables, all terms were fit as smooths to account for non-linear relationships. We looked for an interaction between weather and location because temperature and precipitation influence relative use of forest and urban habitat by Rusty Blackbirds (e.g., Newell-Wohner et al. 2018) and because habitat availability varies among regions in Arkansas (James and Neal 1986). We also looked for an interaction between minimum November temperature (which was our best weather variable) and forest density because Newell-Wohner et al. (2018) found that use of forest habitat by Rusty Blackbirds was influenced by temperature. We used the “ti” option in the mgcv package in R to evaluate that interaction.

To identify the best multivariate model for evaluating the influence of weather, location, effort, habitat, and year, we used a combination of forward, backward, and bi-directional stepwise selection. The final model was produced by including variables that were selected by at least two of the selection routines. To avoid concurvity due to correlation among variables, we only used variables for the selection process whose correlations were less than 0.5 (we started with minimum November temperature because it was identified as the best variable from the simple models mentioned above). To evaluate effect size for variables that exhibited a linear relationship, we inserted their linear coefficients into this formula: 1-exp(coefficient*10). We used 10 units (for example, 10 degrees in temperature, 10% in forest density) as the increment because it proved to be more effective than the conventional 1 unit increase to illustrate effect size. For variables that exhibited non-linear monotonic relationships with counts of Rusty Blackbirds, we estimated effect size by estimating the difference in blackbirds predicted at the beginning and end of the sampling period.

We obtained monthly summaries of weather data by using the county time series feature in the “Climate at a Glance” section of the National Oceanic and Atmospheric Administration (NOAA) website: (https://www.ncdc.noaa.gov/cag/county/time-series). Daily weather variables were recorded in the morning and evening on the day of the count and are available on the CBC website. However, daily weather was not recorded for 108 counts. Consequently, we used the NOAA website listed above to obtain values for the missing temperature data. Cloud cover in the morning was also evaluated. Missing cloud cover data were not available on the above site so we converted missing values of cloud cover to “unknown.” We also combined partially cloudy, and partly clear into partly cloudy and partly foggy and foggy into foggy. As mentioned above, we expected weather to influence counts differently across Arkansas because habitat availability varies regionally (James and Neal 1986) and because differences in temperature and precipitation influence habitat use by Rusty Blackbirds (Newell Wohner et al. 2018). We evaluated regional variation in Rusty Blackbird counts by pooling the data collected within each circle by decade. This approach allowed us to model the distribution of the pooled data for each decade, as we lacked sufficient yearly data to analyze distributions across Arkansas annually. The regional analysis was done by using latitude and longitude as smoothed terms in a GAM and including decade as an interaction term. We produced heat maps for each decade to illustrate the changing regional counts of Rusty Blackbirds. To determine if cold and warm temperatures influenced regional distribution of Rusty Blackbirds, we classified years as warm if their averaged November minimum temperatures (averaged over all counts) was ≥ 90^th percentile; similarly, cold temperatures were ≤10^th percentile of averaged November minimum temperatures. We used November minimum temperatures as our criterion because it had the lowest AICc value among all the simple weather models that we used to rank the variables discussed above. We then modeled latitude and longitude as predictors and included classified years (i.e., cold and warm years) as an interaction term in the model (as above we included party hours to control for differences in effort). We produced heat maps to visualize the regional distributions during warm and cold years. Finally, we evaluated the proportion of forested area of each county as an influence on distribution of Rusty Blackbirds. We obtained forested area for each county from the U.S. Forest Service by searching publications on this site: https://www.fs.usda.gov/research/treesearch; and county area from this site: http://www.usa.com/rank/arkansas-state--land-area--county-rank.htm. Forest area of each county was updated in 1978 and 1995, which gave us three different forest acreages for each county (1965–1977, 1978–1994, and 1995–2020).

RESULTS

Influence of weather

All of the temperature variables followed an inverse relationship with counts of Rusty Blackbirds (see Fig. 2 as an example). Further, all had relatively small (0.28–1.15) effective degrees of freedom indicating that their relationships were relatively linear. Finally, they all accounted for less than 4.2% of the deviance in blackbird numbers. November minimum temperature was better than any other single weather variable in explaining numbers of Rusty Blackbirds (Table 2). A 10-degree increase in minimum November temperature resulted in a 73% decrease in predicted counts of Rusty Blackbirds. Precipitation was not as important as low and average temperatures in influencing counts of Rusty Blackbirds. The ΔAICc values for models of November and December precipitation were greater than 2 compared to the models of minimum and average temperatures (Table 2). Most of the other models were ranked within 2 ΔAICc values of each other and were similar in relative importance (Table 2).

Combined effects

The stepwise procedure produced a model that included November and December minimum temperature, morning cloud cover, percent forest by county, party hours, count year, and latitude and longitude by decade. This model accounted for 24% of the deviance. A simpler model without percent forest area had a slightly lower AICc compared to the more complex model (ΔAICc = 2.25) but accounted for less of the deviance (22.2 instead of 24.3).

Warmer temperatures, larger proportions of forest area per county, and cloudy conditions on the morning of the CBC were associated with lower counts of Rusty Blackbirds. An interaction term between forest area and minimum November temperature did not improve the model (ΔAICc = 2.48). The relationship for minimum November temperature was nonlinear (edf = 2.3) but was monotonic revealing a decrease of approximately 30 birds as minimum November temperature increased from 4 to 10 degrees C (Fig. 2A). The relationship between minimum December temperature and number of Rusty Blackbirds was almost linear (Fig. 2B) and also inverse, with a 10 degree increase in minimum December temperature causing an approximate decrease of 30 blackbirds (Fig. 2B).

Forest density was linearly related to the log of blackbird counts (edf = 1.001) and had a slope of -1.44 (modeled linearly). Therefore, a 10% increase in forest density would result in a predicted decrease of 13.4%. We found that forest density ranged from a low of around 10% to a high of around 87%. An increase in forest density of that magnitude is predicted to result in a 67.2% decrease, which corresponds to approximately 14.1 fewer blackbirds.

Cloud cover on the morning of the count was the only weather condition during the count that was included in the model. Cloudy conditions were associated with fewer detections of Rusty Blackbirds and foggy conditions had the strongest impact on those counts; foggy mornings were associated with a 58.3% reduction in expected numbers of blackbirds counted compared to clear skies.

The trend in Rusty Blackbird counts was relatively flat between 1965 and 1995 after which the numbers trend upwards (edf for year = 3.5; Fig. 3). Our model showed an increase of just over 75 birds since 1995. The model also contained a complex relationship between decade and region that is discussed below. A linear model of November minimum temperatures over the entire span of the study indicates that temperatures are increasing (slope = 0.01, F_1,660= 5.95, p = 0.015). That finding contradicts our finding that numbers of Rusty Blackbirds are inversely related to temperature, because we also found that Rusty Blackbirds have increased in Arkansas starting in about 1995. However, minimum November temperatures have trended down starting in 2004; a linear model for the period between 2004 and 2020 confirms the decrease (slope = -0.13, F_1,221 = 23.29, p < 0.000).

The relationship in the combined model between effort and number of birds counted is complex (edf = 6.659) and did not follow the expected asymptotic pattern (Fig. 4). The model predicted a peak of Rusty Blackbirds at about 60 party hours after which numbers declined. Although the decline in numbers detected is unexpected, the confidence interval widens dramatically for values of party hours > 90 because of small sample size and the more realistic asymptotic relationship fits within those confidence limits (Fig. 4).

Regional variation

The distribution of Rusty Blackbirds across Arkansas varied by decade (Fig. 5). Between 1975 and 1985, Rusty Blackbirds were most abundant in Northwestern Arkansas. Starting in 1985 they shifted slightly south along the western part of the state; after 2005 they became more common in Central Arkansas (Fig. 5). The deviance accounted for by this model was 13.7%, which was substantially larger than any of the simple weather models (Table 2) and the effective degrees of freedom for latitude and longitude by decade was 2.0 for 3 decades and 5.4, 6.3, and 7.7 for three other decades. Thus, the distribution of blackbirds varied nonlinearly across Arkansas. We also found that the pattern of abundance across the decades was influenced by temperature. As mentioned above, minimum November temperatures declined in Arkansas after about 2005. During that period, Rusty Blackbirds shifted in abundance from Northwest to Central Arkansas (Fig. 5). That pattern is also evident in Figure 6, which shows that in relatively warm years, Rusty Blackbirds were more common in the northwestern tier of Arkansas (Fig. 6b), whereas during cooler years they became more abundant in the southeastern portion (Fig. 6a). Minimum November temperatures varied across Arkansas (median values were 4.97 °C Alluvial Plain, 5.22 °C Gulf Coast, 4.78 °C Ouachita, 3.00 °C Ozarks, 4.17 °C River Valley). The Ozarks were significantly cooler than all of the other regions (p < 0.001 Dunn’s Test); and the Gulf Coast was warmer that all of the other regions (p ≤ 0.023, Dunn’s Test). Finally, the Alluvial Plain was warmer than the Ozarks and the River Valley (p < 0.001, Dunn’s Test).

DISCUSSION

Trend in count numbers over time

We found that Rusty Blackbirds have been counted more frequently in Arkansas starting in the mid-1990s. Time and effort only accounted for a small proportion of the variation in counts. Other factors that we did not assess played a more substantial role in driving variation in counts of Rusty Blackbirds in Arkansas. Savard et al. (2011) found that the number of Rusty Blackbirds migrating south was influenced by climate, abundance of red-backed voles (Clethrionomys gapperi), and productivity of Boreal Owls (Aegolius funereus) on the breeding grounds. Recruitment and conditions during migration likely have a larger impact on overwintering numbers than weather conditions during the overwintering period. We did not incorporate variables associated with productivity or numbers that migrated south, which limited our models’ explanatory power. Perhaps the reason that November temperatures tended to have more influence on counts than did December temperatures is because of the timing of migration. Johnson et al. (2012) found that Rusty Blackbirds initiate migration from Alaskan breeding grounds in mid-September and spend 70–84 days migrating to overwintering areas, which means that they do not stop migrating until between 24 November and 10 December.

Our finding that CBCs of Rusty Blackbirds have increased over time is counter to the overall trend indicated by analyses of Christmas Bird Counts (e.g., Niven et al. 2004, Hamel et al. 2009, Bochert 2015). However, our findings are consistent with Strassburg (2011) who found an increase in numbers of Rusty Blackbirds recorded on CBCs starting in the mid-1990s in 10 southeastern states from Texas to Tennessee. Hamel et al. (2009) also found an increasing trend in the western population of Rusty Blackbirds starting in the 1990s. Groups of Rusty Blackbirds that winter west of the Appalachians are drawn from a huge area of Canada and Alaska (Hobson et al. 2010) and given differences in population stability of Rusty Blackbirds within that region (Machtans et al. 2007), we suspect that different trends also exist within the non-breeding range. However, in an extensive analysis of CBC data, Niven et al. (2004) found consistent population declines among populations overwintering on both sides of the Appalachians. But, Niven et al. (2004) did not have access to data after 2003. Our data suggested that most of the increase we found came after 2003. Adding the last 20 years of data to the analysis performed by Niven et al. (2004) could determine whether a recent increase has also occurred across the entire geographic range of Rusty Blackbirds.

Although our analyses indicated that numbers of Rusty Blackbirds counted on CBCs have increased in Arkansas since 1995, that does not mean that the overall population is increasing. Our analysis only indicates that Rusty Blackbirds are being encountered more frequently in Arkansas. Just as our analysis showed that weather influenced where Rusty Blackbirds overwintered in Arkansas, we would expect that weather also influences where in the United States Rusty Blackbirds overwinter. The Audubon Society’s climate change model (https://www.audubon.org/field-guide/bird/rusty-blackbird) predicts that as temperatures increase, the winter range of Rusty Blackbirds will shift northward away from the Gulf Coast of North America. However, although we found that numbers of Rusty Blackbirds counted in Arkansas have increased starting around 1995 and that November and December temperatures cooled over that same period, the relationship is not particularly strong, and we conclude that factors other than weather probably also contribute to the upward trend in count numbers of blackbirds in Arkansas. Perhaps Rusty Blackbirds have become more detectable. Luscier et al. (2010) found that Rusty Blackbirds are not as specialized in habitat use as previously reported; Rusty Blackbirds may be more detectable in open agricultural settings than in forested habitats, and if they are using agricultural settings more now than in the past (Luscier et al. 2010, Strassburg 2011, Borchert 2015, Newell Wohner et al. 2018), increasing counts may simply be due to increasing detectability of Rusty Blackbirds. In addition, a growing awareness of the Rusty Blackbirds among birders may cause more effort to be exerted in detecting them among flocks of other blackbirds.

Rusty Blackbird numbers fluctuate widely between years (Greenberg and Matsuoka 2010) and this was seen in our data as well. Large fluctuations could mask the impact of year as an explanatory variable and hinder one’s ability to identify trends over time. Thompson (2002) suggested that detecting population declines in abundant species requires very large decreases in populations. He noted that this constraint limits our ability to be proactive in management. Rusty Blackbird numbers varied between 0 and 12,000 among locations and years in Arkansas. That wide variation in numbers increased the difficulty of revealing a trend in numbers of Rusty Blackbirds in Arkansas. By using a GAM we detected a trend in the data that was difficult to see in a plot of the data. We suggest that GAMs are effective in identifying subtle trends in count data, even for species whose populations fluctuate between high abundance and absence. By using GAMs to model populations, we may see trends sooner and be able to adjust management sooner as well.

Regional distribution

One of our most significant findings pertains to changing regional distributions of Rusty Blackbirds across different decades that were associated with changes in temperature. During warmer years, we observed an increase in Rusty Blackbird numbers in the northern part of Arkansas, consistent with prior research indicating avian range shifts towards northern or polar regions due to rising temperatures (e.g., Thomas and Lennon 1999, Parmesan and Yohe 2003, Root et al. 2003, La Sorte and Thompson 2007). However, the Rusty Blackbird's response was more complex; they exhibited northwesterly movement in warmer years and southeasterly movement during cooler years. Complex shifts in distribution associated with climate change are less frequently documented (Martin 2001, Wilson et al. 2013). The Audubon Society’s Survival by Degrees climate change model (https://www.audubon.org/field-guide/bird/rusty-blackbird) predicts a northward expansion and western range restriction linked to temperature increases. The disparity between our finding of a northwestern shift instead of a simple northward shift and the eastward shift projected by the Audubon model may be attributed to differences in scale and resolution. Our finding of a westward shift within Arkansas, is consistent with the Audubon model because the blackbirds remain in Arkansas, which was also predicted by the Survival by Degrees model.

We hypothesize that the Rusty Blackbird’s observed geographic shift is associated with an interaction between temperature and habitat use, which is why they did not simply move north and south with increasing and decreasing temperatures. The regional shift results because habitat in Arkansas varies regionally and because habitat use by Rusty Blackbirds is influenced by temperature (Newell Wohner et al. 2018, Ohanyan 2021). Eastern Arkansas features bands of bottomland hardwood forests along the Mississippi and White rivers that are frequently used by Rusty Blackbirds (sightings available in ebird: https://ebird.org). Eastern Arkansas also has extensive pecan orchards located along the Mississippi River (University of Arkansas Cooperative Extension Services 2003). The eastward shift in abundance linked to lower temperatures may be connected to Rusty Blackbirds’ utilization of bottomland hardwoods prevalent along the Mississippi River. However, our analysis does not allow for the identification of specific habitats, and despite the Mississippi River’s prominence in Eastern Arkansas, agricultural landscapes dominate that region. Moreover, Newell Wohner et al. (2018) found that Rusty Blackbirds were less inclined to utilize bottomland hardwoods during cold years, challenging our hypothesis that the shift into Eastern Arkansas was primarily driven by access to bottomland hardwoods. Both Newell Wohner (2018) and Ohanyan (2021) reported Rusty Blackbirds used pecan groves during cold years but not during warm years. Consequently, the shift toward the Mississippi River during cold years could also have been associated with use of extensive pecan orchards located along the Mississippi River. Our findings suggest that the regional distributions of Rusty Blackbirds in Arkansas are influenced by temperature, with the likely underlying cause being regional differences in habitat availability. However, our data do not allow for a robust assessment of such an interaction because we did not include detailed habitat measurements within the count circles in our model. The interaction that we assessed between proportion of forest land in the count circle’s county and November minimum temperature failed to improve our model, probably because proportion of forest in the county in which the count circle was located did not represent habitat that was used by Rusty Blackbirds in the count circle.

The regional changes in abundance exhibited by Rusty Blackbirds increase the difficulty in evaluating their abundance because even within a relatively small portion of their range, like Arkansas, blackbirds may be increasing in one area and decreasing in another. Consequently, a complete understanding of population dynamics requires a very broad-scale evaluation. Wilson et al. (2013) found something similar for Western (Aechmophorus occidentalis) and Clark’s Grebes (A. clarkii). Populations of both species of grebes declined in the northern portion of their winter range while increasing dramatically in the south. We suspect that the northward shift of Rusty Blackbirds is responsible for fewer birds being counted in Arkansas during warm years because birds may have wintered north of Arkansas or because only two count circles are located in the northern tier of Arkansas (Fig. 1). As a result, in warm years Rusty Blackbirds are either located north of, or in a part of Arkansas where they are less likely to be counted. Therefore, assessing population changes in Rusty Blackbirds necessitates extensive geographic sampling covering a wide range of habitats and regions.

Weather

We found that temperature was a more influential variable than precipitation in influencing CBCs of Rusty Blackbirds. In contrast, Newell Wohner et al. (2018) found that precipitation influenced Rusty Blackbird relative use of forested wetlands and pecan groves. We suspect that precipitation may have less influence on how far south Rusty Blackbirds overwinter than does temperature. Although, precipitation may have influenced habitat selection by Rusty Blackbirds (see discussion in Ohanyan 2021), the effect of precipitation may have been overwhelmed by a stronger effect of temperature on Rusty Blackbirds.

Forest area

We found slightly fewer Rusty Blackbirds in count circles that were located in counties that had a higher proportion of forest land. This finding might seem unexpected. However, we think our finding is consistent with the biology of Rusty Blackbirds. Rusty Blackbirds use bottomland hardwood forests and are absent from other types of forests (Luscier et al. 2010). The most heavily forested counties in Arkansas are located in the Ouachitas, Ozarks, and Gulf Coast Plain. Forestland in the Ozarks and Ouachitas is mostly upland forest (James and Neal 1986) that is inappropriate for Rusty Blackbirds. The forestland in the Gulf Coast Plain is dominated by pines (James and Neal 1986), which is also inappropriate for Rusty Blackbirds. Finally, the proportion of forestland in a county may not actually represent what is present in the count circle. A more accurate representation of forest habitat within the count circles would increase our model’s performance. Thus, including covariates to distinguish between bottomland forests and other forests would improve our ability to assess effects of forest on the abundance and distributions of Rusty Blackbirds during winter. Some possible covariates include presence of large rivers in the count circle to indicate potential for bottomland hardwood forests, topographic and edaphic features associated with bottomland hardwood forests. For example, we might have used digital elevation maps to identify bottomland hardwood forests (e.g., Lang et al. 2013) within the count circle.

CONCLUSION

This study enhances our understanding of factors that influence patterns of non-breeding distribution of Rusty Blackbirds. We showed that distributions of Rusty Blackbirds varied across Arkansas and that shifts in regional distributions were associated with changes in minimum November and December temperatures. Warmer temperatures are often associated with shifts northward by many birds and other organisms (e.g., Thomas and Lennon 1999, Parmesan and Yohe 2003, Root et al. 2003, La Sorte and Thompson 2007). However, we did not find a simple temperature-mediated north-south shift in distribution; rather, Rusty Blackbirds shifted to the east-central portion of the state during cooler years and northwest Arkansas during warm years. We hypothesize that the shift in distribution is associated with temperature-dependent changes in habitat use that have been observed in Rusty Blackbirds (Newell Wohner et al. 2018, Ohanyan 2021). Habitat in eastern Arkansas includes the Mississippi and White rivers that support bands of bottomland hardwood forest. The region also supports extensive pecan orchards, both of which are used by Rusty Blackbirds (Mettke-Hofmann et al. 2015, Newell Wohner et al. 2018).

Rusty Blackbirds exhibited an increasing trend in Arkansas starting in 1995. We do not know whether the trend in Arkansas is representative of other portions of the range but we think it warrants further investigation. Finally, we hope that the approach we took in using GAMs to produce heat maps of overwintering distribution would lead others to consider similar analyses for other species. Such an analysis encompassing the entire range of the Rusty Blackbird would doubtless lead to other conclusions and insights for the species.

LITERATURE CITED

Akaike, H. 1973. Information theory and an extension of the maximum likelihood principle. Pages 267-281 in B. Petrov and F. Cazakil editors. Proceedings of the 2nd International Symposium on Information Theory. Akadémiai Kiadó, Budapest, Hungary.

Avery, M. L. 2000. Rusty Blackbird (Euphagus carolinus), version 1.0. In A. F. Poole, editor. Birds of the world. Cornell Lab of Ornithology, Ithaca, New York, USA. https://doi.org/10.2173/bow.rusbla.01

Bock, C. E., and T. Root. 1981. The Christmas Bird Count and avian ecology. Studies in Avian Biology 6:17-23.

Borchert, S. M. 2015. Site-specific habitat and landscape associations of Rusty Blackbirds wintering in Louisiana. Thesis. Louisiana State University, Baton Rouge, Louisiana, USA. https://digitalcommons.lsu.edu/gradschool_theses/1841

Burnham, K. P., and D. R. Anderson. 2002. Model selection and inference: a practical information-theoretic approach. Second edition. Springer-Verlag, New York, New York, USA.

Canepa, A., V. Fuentes, M. Bosch-Belmar, M. Acevedo, K. Toledo-Guedes, A. Ortiz, E. Durà, C. Bordehore, and J.-M. Gili. 2017. Environmental factors influencing the spatio-temporal distribution of Carybdea marsupialis (Lineo, 1978, Cubozoa) in South-Western Mediterranean coasts. PLoS ONE 12(7): e0181611. https://doi.org/10.1371/journal.pone.0181611

Fewster, R. M., S. T. Buckland, G. M. Siriwardena, S. R. Baillie, and J. D. Wilson. 2000. Analysis of population trends for farmland birds using generalized additive models. Ecology 81:1970-1984. https://doi.org/10.1890/0012-9658(2000)081[1970:AOPTFF]2.0.CO;2

Greenberg, R., D. W. Demarest, S. M. Matsuoka, C. Mettke-Hofmann, D. Evers, P. B. Hamel, J. Luscier, L. L. Powell, D. Shaw, M. L. Avery, K. A. Hobson, P. J. Blancher, and D. K. Niven. 2011. Understanding declines in rusty blackbirds. Pages 107-126 in J. V. Wells, editor. Boreal birds of North America: a hemispheric view of their conservation links and significance. Vol. 41 in Studies in Avian Biology. University of California Press, Berkeley, California, USA. https://doi.org/10.1525/9780520950580-012

Greenberg, R., and S. Droege. 1999. On the decline of the Rusty Blackbird and the use of ornithological literature to document long-term population trends. Conservation Biology 13:553-559. https://doi.org/10.1046/j.1523-1739.1999.97478.x

Greenberg, R., and S. M. Matsuoka. 2010. Rangewide ecology of the declining Rusty Blackbird: mysteries of a species in decline. Condor 112:770-777. https://doi.org/10.1525/cond.2010.100153

Hamel, P., D. DeSteven, T. Leininger, and R. Wilson. 2009. Historical trends in Rusty Blackbird nonbreeding habitat in forested wetlands. Pages 341-353 in T. Rich, C. Arizmendi, D. Demarest, and C. Thompson, editors. Tundra to tropics: connecting birds, habitats and people. Proceedings of the 4th International Partners in Flight Conference, McAllen, Texas, USA, 13-16 Feb. 2008.

Hilbe, J. M. 2011. Negative binomial regression. Second edition. Cambridge University Press, Cambridge, UK. https://doi.org/10.1017/CBO9780511973420

Hobson, K. A., R. Greenberg, S. L. Van Wilgenburg, and C. Mettke-Hofmann. 2010. Migratory connectivity in the Rusty Blackbird: isotopic evidence from feathers of historical and contemporary specimens. Condor 112:778-788. https://doi.org/10.1525/cond.2010.100146

Ingersoll, T. E, B. J. Sewall, and S. K. Amellon. 2013. Improved analysis of long-term monitoring data demonstrates marked regional declines of bat populations in the Eastern United States. PLoS ONE 8(6):e65907. https://doi.org/10.1371/journal.pone.0065907

James, D. A., and J. C. Neal. 1986. The birds of Arkansas, their distribution and abundance. University of Arkansas Press, Fayetteville, Arkansas, USA.

Johnson, J. A., S. M. Matsuoka, D. F. Tessler, R. Greenberg, and J. W. Fox. 2012. Identifying migratory pathways used by Rusty Blackbirds breeding in Southcentral Alaska. Wilson Journal of Ornithology 124:698-703. https://doi.org/10.1676/1559-4491-124.4.698

Kangas, K., M. Luoto, A. Ihantola, E. Tomppo, and P. Siikamaki. 2010. Recreation-induced changes in boreal bird communities in protected areas. Ecological Applications 20:1775-1786. https://doi.org/10.1890/09-0399.1

Knape, J. 2016. Decomposing trends in Swedish bird populations using generalized additive mixed models. Journal of Applied Ecology 53:1852-1861. https://doi.org/10.1111/1365-2664.12720

La Sorte, F. A., and F. R. Thompson III. 2007. Poleward shifts in winter ranges of North American birds. Ecology 88:1803-1812. https://doi.org/10.1890/06-1072.1

Lang, M., G. McCarty, R. Oesterling, I.-Y. Yeo. 2013. Topographic metrics for improved mapping of forested wetlands. Wetlands 33:141-155. https://doi.org/10.1007/s13157-012-0359-8

Link, W. A., and J. R. Sauer. 1996. Extremes in ecology: avoiding the misleading effects of sampling variation in summary analyses. Ecology 77:1633-1640. https://doi.org/10.2307/2265557

Link, W. A., and J. R. Sauer. 1999. Controlling for varying effort in count surveys: an analysis of Christmas Bird Count data. Journal of Agricultural, Biological, and Environmental Statistics 4:116-125. https://doi.org/10.2307/1400592

Luscier, J. D., S. E. Lenhen, and K. G. Smith. 2010. Habitat occupancy by Rusty Blackbirds wintering in the lower Mississippi Alluvial Valley. Condor 112:841-848. https://doi.org/10.1525/cond.2010.100147

Machtans, C. S., S. L. Van Wilgenburg, L. A. Armer, and K. A. Hobson 2007. Retrospective comparison of the occurrence and abundance of Rusty Blackbird in the Mackenzie Valley, Northwest Territories. Avian Conservation and Ecology - Écologie et conservation des oiseaux 2(1):3. https://doi.org/10.5751/ACE-00106-020103

Martin, T. E. 2001. Abiotic vs. biotic influences on habitat selection of coexisting species: climate change impacts? Ecology 82:175-188. https://doi.org/10.1890/0012-9658(2001)082[0175:AVBIOH]2.0.CO;2

Meehan, T. D., N. L. Michel, and H. Rue. 2019. Spatial modeling of Audubon Christmas Bird Counts reveals fine-scale patterns and drivers of relative abundance trends. Ecosphere 10(4):e02707. https://doi.org/10.1002/ecs2.2707

Mettke-Hofmann, C., P. B. Hamel, G. Hofmann, T. J. Zenzal, Jr., A. Pelligrini, J. Malpass, M. Garfinkel, N. Schiff, and R. Greenberg. 2015. Competition and habitat quality influence age and sex distribution in wintering rusty blackbirds. PLoS ONE 10(5):e0123775. https://doi.org/10.1371/journal.pone.0123775

National Audubon Society. 2020. The Christmas Bird Count historical results by count. National Audubon Society, New York, New York, USA. https://netapp.audubon.org/CBCObservation/Historical/ResultsByCount.aspx

Newell Wohner, P. J., R. J. Cooper, S. H. Schweitzer, and R. S. Greenberg. 2018. Rusty Blackbird patch use during winter in suburban landscapes. Journal of Wildlife Management 82:1736-1745. https://doi.org/10.1002/jwmg.21548

Niven, D. K., J. R. Sauer, G. S. Butcher, and W. A. Link. 2004. Christmas Bird Count provides insights into population change in land birds that breed in the boreal forest. American Birds 58:10-20.

Ohanyan, A. 2021. Overwinter ecology of rusty blackbirds (Euphagus carolinus) in Arkansas. Thesis. Arkansas Tech University, Russellville, Arkansas, USA.

Parmesan, C., and G. Yohe. 2003. A globally coherent fingerprint of climate change impacts across natural systems. Nature 421:37-42. https://doi.org/10.1038/nature01286

Rakhimberdiev, E., Y. I. Verkuil, A. A. Saveliev, R. A. Väisänen, J. Karagicheva, M. Y. Soloviev, P. S. Tomkovich, and T. Piersma. 2011. A global population redistribution in a migrant shorebird detected with continent-wide qualitative breeding survey data. Diversity and Distributions 17:144-151. https://doi.org/10.1111/j.1472-4642.2010.00715.x

Reiss, P. T., and R. T. Ogden. 2009. Smoothing parameter selection for a class of semiparametric linear models. Journal of the Royal Statistical Society Series B: Statistical Methodology 71:505-523. https://doi.org/10.1111/j.1467-9868.2008.00695.x

Root, T. L., J. T. Price, K. R. Hall, S. H. Schneider, C. Rosenzweig, and J. A. Pounds. 2003. Fingerprints of global warming on wild animals and plants. Nature 421:57-60. https://doi.org/10.1038/nature01333

Sauer, J. R., W. A. Link, and J. A. Royle. 2004. Estimating population trends with a linear model: technical comments. Condor 106:435-440. https://doi.org/10.1093/condor/106.2.435

Sauer, J. R., D. K. Niven, J. E. Hines, D. J. Ziolkowski, Jr, K. L. Pardieck, J. E. Fallon, and W. A. Link. 2017. The North American breeding bird survey, results and analysis 1966-2015. Version 2.07.2017. USGS Patuxent Wildlife Research Center, Laurel, Maryland, USA.

Savard, J.-P. L., M. Cousineau, and B. Drolet. 2011. Exploratory analysis of correlates of the abundance of Rusty Blackbirds (Euphagus carolinus) during fall migration. Ecoscience 18:402-408. https://doi.org/10.2980/18-4-3401

Strassburg, M. D. 2011. The evaluation of Christmas Bird Counts as an indicator of population trends and habitat selection in blackbirds and starlings. Thesis. North Dakota State University, Fargo, North Dakota, USA. http://hdl.handle.net/10365/19385

Thomas, C. D., and J. J. Lennon. 1999. Birds extend their ranges northwards. Nature 399:213. https://doi.org/10.1038/20335

Thompson, W. L. 2002. Towards reliable bird surveys: accounting for individuals present but not detected. Auk 119:18-25. https://doi.org/10.1093/auk/119.1.18

University of Arkansas Cooperative Extension Services. 2003. Crop profile for pecans in Arkansas. University of Arkansas Cooperative Extension Services, Little Rock, Arkansas, USA. https://ipmdata.ipmcenters.org/documents/cropprofiles/ARpecans.pdf

Wilson, S., E. M. Anderson, A. S. G. Wilson, D. F. Bertram, and P. Arcese. 2013. Citizen science reveals an extensive shift in the winter distribution of migratory Western Grebes. PLoS ONE 8(6):e65408. https://doi.org/10.1371/journal.pone.0065408

Zuur, A., E. N. Ieno, N. J. Walker, A. A. Saveliev, G. M. Smith. 2009. Mixed effects models and extensions in R. Springer, New York, New York, USA. https://doi.org/10.1007/978-0-387-87458-6