Skip to main content

Environmental drivers of autumn migration departure decisions in midcontinental mallards

Abstract

Background

The timing of autumn migration in ducks is influenced by a range of environmental conditions that may elicit individual experiences and responses from individual birds, yet most studies have investigated relationships at the population level. We used data from individual satellite-tracked mallards (Anas platyrhynchos) to model the timing and environmental drivers of autumn migration movements at a continental scale.

Methods

We combined two sets of location records (2004–2007 and 2010–2011) from satellite-tracked mallards during autumn migration in the Mississippi Flyway, and identified records that indicated the start of long-range (≥ 30 km) southward movements during the migration period. We modeled selection of departure date by individual mallards using a discrete choice model accounting for heterogeneity in individual preferences. We developed candidate models to predict the departure date, conditional on daily mean environmental covariates (i.e. temperature, snow and ice cover, wind conditions, precipitation, cloud cover, and pressure) at a 32 × 32 km resolution. We ranked model performance with the Bayesian Information Criterion.

Results

Departure was best predicted (60% accuracy) by a “winter conditions” model containing temperature, and depth and duration of snow cover. Models conditional on wind speed, precipitation, pressure variation, and cloud cover received lower support. Number of days of snow cover, recently experienced snow cover (snow days) and current snow cover had the strongest positive effect on departure likelihood, followed by number of experienced days of freezing temperature (frost days) and current low temperature. Distributions of dominant drivers and of correct vs incorrect prediction along the movement tracks indicate that these responses applied throughout the latitudinal range of migration. Among recorded departures, most were driven by snow days (65%) followed by current temperature (30%).

Conclusions

Our results indicate that among the tested environmental parameters, the dominant environmental driver of departure decision in autumn-migrating mallards was the onset of snow conditions, and secondarily the onset of temperatures close to, or below, the freezing point. Mallards are likely to relocate southwards quickly when faced with snowy conditions, and could use declining temperatures as a more graduated early cue for departure. Our findings provide further insights into the functional response of mallards to weather factors during the migration period that ultimately determine seasonal distributions.

Background

Avian seasonal migration is an energetically costly series of movements that may cover great distances, and is fundamental to the ecology of many bird species [1, 2]. Migration decisions can influence fitness in both the breeding and non-breeding portion of the year. The timing, distance and speed of migratory movements can influence and be influenced by body condition, reproductive success, and population composition [3, 4] and has been shown to influence long-term processes including continental or global distribution and speciation [5].

Understanding migratory movements in waterfowl has been considered particularly important, due to their status as popular gamebirds with associated population and habitat management programs [2, 6] as well as their role in the spread of zoonotic diseases [7]. Many waterfowl, and most ducks, depend on wetlands that are particularly threatened by anthropogenic climate and land use change [8, 9]. Phenological shifts in waterfowl migration have been connected to altered habitat conditions and weather patterns in both breeding and wintering ranges [9,10,11].

Although an increasing number of studies have recently been published about the proximal factors influencing migration chronology in waterfowl [12,13,14,15,16,17,18], much remains unclear. There is a lack of quantitative knowledge about the parameters that drive migration timing, distance, and choice of target location among waterfowl [2, 19], with the information scarcity more pronounced for autumn than spring migration [18]. Response to decreasing photoperiod is thought to be the principal external cue for autumn departure in many bird species, especially song birds [2, 20, 21], but this has not yet been clearly demonstrated in waterfowl. Among northern hemisphere waterfowl, decision to embark on southwards migration is thought to principally depend on energy budget considerations. With declining food availability under autumn and winter conditions, metabolic costs increase and food becomes harder to acquire, until staying at high latitudes is more costly than expending energy to relocate southward [12, 22, 23]. The effect of decreasing temperatures together with the occurrence of snow and ice cover are principal migration cues in ducks and geese, both in the short term and as cumulative measures over longer periods [2, 6, 12, 15, 18].

On a proximate time scale, decision to migrate is frequently influenced by how energetically favorable weather conditions are on a given day for long-distance movements [1]. Flight may be made costly by headwinds or facilitated by tailwinds [17, 24], precipitation may impede flight and increase thermoregulation costs [1, 17, 25], cloud cover may obstruct the view of visual cues used for navigation [26], and pressure differences may serve as cues for impending weather changes [1] or facilitate departure [27]. Flight weather is a prominent migration driver in passerines [19, 25, 28], but with the notable exception of wind direction [17, 18, 29, 30], it plays less of a role in waterfowl [31, 32]. However the majority of existing studies focused on shifts in waterfowl abundance at the population level, and were generally restricted to few chosen locations at particular latitudes. There is still much less information on what drives movement decisions at the individual level, and across the latitudinal range between breeding and wintering grounds. However location records from satellite-tracked birds are increasingly used to address this information gap [13, 14, 16, 18, 33].

The mallard (Anas platyrhynchos) is the most numerous species among ducks migrating through the Mississippi Flyway and considered a priority in the development of wetland management plans and hunting regulations [34, 35], with an average of 2.8 million individuals each year wintering in the Mississippi Alluvial Valley alone [36]. Consequently, the timing and drivers of mallards’ migratory movements into and through the flyway are of great interest for the development of accurate population models for conservation planning in the region [16, 37, 38]. In this study we investigated the autumn migration movements of satellite-tagged mallards within the Mississippi Flyway. Our objective was to evaluate individual-level decisions to embark on migration movements as a function of environmental drivers. As the migration process depends on bird behaviour over a sequence of directional movements, and extended stays at stopover sites are common for many waterfowl and especially ducks [2, 39], we were interested to study departure and stopover events both inside and outside the wintering range. We used a discrete choice modeling framework [40] to develop resource selection functions [41], while including environmental covariates that could account for the impact of winter conditions on local habitat and for the short-term energetic efficiency of undertaking migration flights. Based on the findings reported in the literature, we hypothesized that mallard’s decisions to relocate would be principally driven by the onset of longer-term winter conditions (snow and falling temperatures) and modified by short-term conditions that may facilitate or impede embarking on a long-distance flight (wind, precipitation, and visibility), and that these relationships would be applicable across the migration range.

Methods

Location data

We used two existing sets of location records from satellite-tracked mallards with a combined size of 220 individuals, of which 43 were eventually used for analysis. Data set A consisted of 180 mallards of both sexes that were captured and tagged in several locations in Arkansas in February–March 2004 (23 female: 10 male), February 2005 (27 female: 21 male), January–February and November–December 2006 (39 female: 15 male), and January–February 2007 [45 female]. Transmitter units weighed 22–35 g (1.8–2.6% of body mass at capture [mean ± S.D.: 1098 ± 132 g]) and used the CLS-Argos (Toulouse, France) satellite system to monitor movement [42]. For details on tagging and transmitters see [14].

Data set B was gathered from 40 mallard hens, of which 20 were captured in the same location in Saskatchewan in September 2010, and 20 in February 2011 in multiple location in Arkansas. Transmitter units weighed 28 g (2.4–2.7% of body mass at capture [mean ± S.D.: 1099 ± 71.5 g]) and used the GPS satellite system to monitor movement. We used a version of the dataset that was already censored for dead individuals and failed transmitters as described in [16].

We first censored the sets of records invalidated by death of individuals and transmitter failures. This entirely removed 9 individuals from set A and yielded a total of 211 individuals (168 females: 43 males) (for details see Additional file 1: Methods S1). Both data sets were then further processed and filtered before analysis. Additional file 1: Table S1 provides a breakdown of resultant sample sizes and individual numbers throughout the process.

First, successive records for each individual were combined to no more than a single movement per 24 h to prevent masking of long-range daily movements by high recording frequency. In such a case, a straight-line movement from the start point of the first movement to the end point of the last movement was assumed. This process excised 0.5% of records in data set A and 63% of records in data set B, because mean recording frequency in the latter was higher.

We then identified records that could be interpreted as the starting point of individual migration-scale movements during the autumn migration period, based on the following criteria (for details see Additional file 1: Methods S2): (1) movement distance to following recorded location ≥ 30 km, based on the approximate empirical breakpoint between local and migration flights in mallards [43]; (2) time period 1 September–31 December of each year; (3) time difference to next record ≤ 48 h; and (4) movement had a southward component.

We chose to pool the two processed data sets for further analysis, since the required location accuracy for the investigation of the migration-scale movements of interest (30+ km) was present in both sets. The pooled data set consisted of 269 relocation records representing a total of 82 individuals (48 female and 13 males from set A, 21 females from set B). The identified relocation records were used as the basis for constructing the set of alternative departure dates for discrete choice analysis.

Discrete choice modeling

We conceptualized migration as a choice that individual animals make based on ambient environmental conditions. To model the selection of departure dates by individual mallards, we used discrete choice models [41, 44, 45]. A choice set included a date that an animal selected to migrate and a matched suite of available alternative dates the duck did not migrate. Conceptually, animals assign separate utilities U to each alternative date in a choice set. The alternative with the highest utility has the greatest probability that the animal will select that alternative to migrate. Although the utility of any given alternative can be negative, the above concept remains unchanged: alternative departure dates with higher utilities have an increased probability of departure compared to alternatives with lower utilities. We modeled utility as an additive linear combination of covariates, wherein U of alternative j in choice set i by animal a based on k = 1…K covariates can be written as:

$$U_{aij} = z_{aij1} b_{a1} + \cdots + z_{aijK} b_{aK} + \varepsilon_{aij} = \underline{z}_{aij}^{\prime } \underline {b}_{a} + \varepsilon_{aij}$$

with errors εaij following the Gumbel extreme value distribution. In generalized linear terminology, U is the linear predictor. In our discrete choice model, observed data values are Y = 1 for dates on which an animal migrated and Y = 0 for dates on which an animal did not migrate. In each choice set, only one alternative is selected. The expected relative probability that alternative j in choice set i will be selected by animal a is then the discrete choice model [40, 44]:

$$P\left( {Y_{aij} = 1|\underline {z}_{aij} } \right) = \int {\frac{{\exp \left( {\underline{{z_{aij}^{\prime } }} \underline {b}_{a} } \right)}}{{\mathop \sum \nolimits_{j = 1}^{7} \exp \left( {\underline {z}_{aij}^{\prime } \underline {b}_{a} } \right)}}f\left( {\underline {b}_{a} } \right){\text{d}}\underline {b}_{a} }$$

where \(\mathop \sum \nolimits_{j = 1}^{7} P\left( {Y_{aij} = 1|\underline {z}_{aij} } \right) = 1.0\). ba denotes individual-level coefficients that account for inter-individual variance in selection patterns and serve to relax the assumption of independence from irrelevant alternatives (IIA) [44]. Observed heterogeneity in this variance component can be accommodated by including individual-specific covariates (e.g., sex), or a common distribution can be employed. Given our data structure and in the absence of specific information, we assumed a normal distribution f(ba) with mean ßk and standard deviation sk [45].

Rather than the commonly used selection among spatially distinct resources such as habitat patches or food sources (e.g., 48–51), we employed a time series of successive records (dates) at the same location as the choice set, and modeled the utility of each date as a temporal alternative—that is, as the time at which a relocation could be undertaken. We constructed choice sets of J alternatives by selecting an identified relocation record and the J-1 non-relocation records directly preceding it. The observed chosen alternative (identified relocation) was thus always on the last day (day J). We used J = 7 as a set size that provided a balance between number of alternatives (n = 7), number of available choice sets containing at least this number of alternatives (n = 73), and number of individual mallards represented by these choice sets (43; 52% of total individuals) (Additional file 1: Table S1). Although each choice set consisted of exactly 7 records, the period between separate records could be longer than one calendar day (see above), resulting in choice sets longer than 7 days. Among sets, 36/73 (49%) covered a period of 7 calendar days, 68/73 (93%) covered up to 14 days, and the longest period was 35 days.

The final data set thus consisted of 43 individuals (39 female: 4 male), 73 relocation dates and 73 × 7 = 511 total alternative dates. For analysis, data were specified as panel data (i.e., possible repeated choices for decision makers) at the individual level to account for the fact that ~ 50% of mallards (21/43) were represented by two or more choice sets (1–4 sets per individual, with a mean of 2).

Environmental data

After censoring, location records were matched with daily mean environmental parameter records from the National Oceanic and Atmospheric Administration’s (NOAA) North American Regional Reanalysis (NARR) database [49], using a nearest-neighbor approach based on their position within a grid of 32 × 32 km cells. This resolution corresponded closely to our chosen minimal range of 30 km for analyzed movements. We then matched each location record with the corresponding daily parameters of its grid cell. We selected 10 potentially informative variables based on their performance in previous population-level studies [12, 17, 18, 25]. Seven of these were used to describe short-term (daily) conditions at the location:

  • Surface air temperature (temp; °C) was measured at 2 m above ground level.

  • Depth of snow cover (snow; m)

  • Difference in barometric pressure (press diff; Pa) was calculated between successive records.

  • Total precipitation (precip; kg/m2) represented water in any form (rain, snow, freezing rain, or hail) amassed throughout the day.

  • Cloud cover (cloud; %) represented the combination of low-, mid- and high-level clouds.

  • Headwind and tailwind speed (head / tail; m/s). Wind speed was provided as separate meridional and zonal speed components, which were combined to yield a directional wind vector. This vector was then classified relative to the mean direction of all relocations (162°), with directions within 60° to either side (a 120° arc) classified as “tailwind” and the rest (a 240° arc) classified as “headwind”, and the associated wind speed assigned to the respective parameter while the other was set to 0.

We also computed three cumulative parameters descriptive of multi-day conditions:

  • Frost days (frost days; d) were consecutive days of mean temperature < 0 °C, calculated for each individual based on the rounded number of sequential 24 h periods spent in one location while the condition “temperature < 0 °C” prevailed. Location was considered to change whenever a bird had moved a cumulative straight distance of ≥ 30 km over any number of records (i.e., left a circle of radius 30 km around the last location). If the interval between two records spanned multiple calendar days with both bounding records at the same location, we assumed that the bird remained at the location for the entire period. Movement to a new location or temperatures > 0 °C reset the counter to zero.

  • Snow days (snow days; d) were consecutive days of snow depth ≥ 2.54 cm [1 inch], calculated equivalently to frost days.

  • Ice cover (ice cover; yes/no) was a binary index recording the assumed presence of ice ≥ 1 cm thick on shallow water bodies. We based the calculation of ice growth on empirical formulae that reported thin ice growth of 1 cm per 3.3 freezing degree days (FDD Celsius) and melting of 1 cm per 1.3 thawing degree days (TDD Celsius) [50, 51]. This applies from initial ice formation, the speed of which depends on the heat capacity of the water body; because dabbling ducks typically feed in shallow water bodies, we assumed a conservative two days of freezing temperatures for initial ice formation. At a given location, tracking of ice thickness was thus triggered after two calendar days of mean temperature < 0 °C and reset to zero if thickness was estimated to drop below 1 cm.

Daily means were used for the computation of all parameters. We did not include seasonal effect, in the form of Julian date, as a covariate because the structure of our model implied that this parameter would monotonically increase within each choice set (see below).

To rescale environmental covariates for analysis, all were centered on the mean and divided by two standard deviations [52]. Model coefficients (βK) therefore represent the expected change in utility for an alternative departure day given an increase of two standard deviations of the independent variable.

Candidate models and analysis

We used the environmental parameters as covariates to formulate a set of candidate models representing hypotheses about mallards’ choice of relocation day (Table 1). We included univariate models for each of the 10 covariates, and a full model containing all covariates. We also included separate models for testing the covariates quantified on a daily scale (model daily scale) and the covariates quantified over multiple days (model multi-day scale). Finally, we included a model with covariates that represented winter conditions, i.e., low temperatures and presence of snow and ice (model winter conditions). We also included a statistical null model that assumed each alternative had the same utility and thus that all alternatives had identical relative probabilities of selection. To assess multicollinearity, we calculated variance inflation factors (VIF) for each multivariate model; the maximum value (3.1) was well below the suggested threshold of 10 for assuming potentially problematic collinearity between predictors [53]. At the individual level, sex was available as a covariate. We ran all models both with inter-individual variance in covariates as driven by the discrete sex covariate, or as normally distributed across the range of each population-level covariate.

Table 1 Candidate models for discrete choice analysis

Resource selection probability functions (RSPFs) model the probability that an individual will select a resource unit with a given set of covariates when encountered, whereas resource selection functions (RSFs) model a dependent variable that is proportional to the RSFP. Assuming an exponential RSF, we calculated a selection index for the top model [w(z) = exp(z1ß1 +  + zKßK)] which is proportional to the selection of a given alternative at the population level [44, 54]. To investigate the interactions between covariates at different magnitudes, we predicted both utilities and selection indices over the recorded range of one covariate while fixing the others either at their mean, or at the 25th (“low” value) or 75th percentile (“high” value) of their range for the top model.

We evaluated the predictive ability of models with leave-one-out cross-validation, fitting models to the data excluding a single choice set, and excluding each of the 73 choice sets in turn [55]. We then calculated the mean percentage of models where the observed day (day 7) of migration exhibited the highest utility within the choice set (% correct) across all validation sets. Although the percent correct does not constitute a goodness-of-fit statistic [40], we provide it as a useful index of predictive power. We ranked models by Bayesian Information Criterion (BIC) rather than by Akaike’s Information Criterion (AIC) in the interest of more conservative ranking [53]. The population-level parameter estimates ßK and the standard deviations sK were counted as parameters for the purpose of BIC calculation. Inferences on relocation day choice by mallards were made from the top-ranked model. We fit discrete choice models with package gmnl [45] in R v4.0.3 [56]. We used 500 draws of simulated probabilities in the maximum likelihood estimation, after having determined that parameter estimates remained static above this number (at the level of three significant digits).

Results

Models with air temperature and snow conditions were highly ranked, while models conditional on wind speed, precipitation, presence of ice, pressure difference, and cloud cover received much lower support (Table 2). Model selection among the set of candidate models indicated a single top model, winter conditions, with a relative BIC model weight of 1.0. This model also shared the highest predictive power (60% correct over 73-fold cross-validation), together with the lower-ranked full model. The null model, representing equal selection probability for each of the seven alternatives in a set, had the expected predictive power of 14% (1/7). The highest-ranked and most predictive single-parameter model was temperature (ΔBIC 21.2, 52% correct). Headwind speed and ice cover had the lowest predictive power (7% and 4% respectively). We found no support for sex as a predictor of inter-individual preference, as the resultant model set showed the same rank order but globally raised BIC values (data not shown) compared to the set assuming a normal distribution of preferences across individuals. All inferences are therefore based on the highest-ranked model (winter conditions) of the latter set.

Table 2 Model selection table

The winter conditions model included temperature, depth of snow cover, number of frost days experienced, number of snow days experienced, and presence of ice cover (≥ 1 cm thick) (Table 1). Specifically, temperature exhibited a negative relationship with the probability of departure (\(\hat{\beta }\) =  − 6.04, SE = 2.59), indicating lower temperatures increased the probability of a duck choosing to migrate (Figs. 1A, 2A). In contrast, snow cover (\(\hat{\beta }\) = 5.23, SE = 3.04) and frost days (\(\hat{\beta }\) = 3.71, SE = 2.86) exhibited positive relationships with the probability of departure, indicating that greater snow cover and number of frost days increased the probability of migrating (Figs. 1B, C, 2B, C). The presence of ice cover ≥ 1 cm thick had the smallest positive relationship with probability of departure (\(\hat{\beta }\) = 1.65, SE = 1.90) (Figs. 1E, 2E). The number of snow days had the largest effect on probability of departure (\(\hat{\beta }\) = 10.61, SE = 8.68) with additional snow days generating a greater probability of migrating, but also exhibited high uncertainty (Figs. 1D, 2D); this parameter also had the greatest VIF (2.6) in the model. Substantial snow (greater numbers of snow days or greater snow depth) dominated choice of migration date to the extent that temperature parameters (frost days and temperature) had minimal impact, whereas under low snow conditions, migration probability depended on temperature conditions (Fig. 3). Inter-individual variability in preference among mallards was pronounced for temperature (ŝ = 9.34, SEŝ = 3.41) and frost days (ŝ = 8.20, SEŝ = 3.13), lower for depth of snow cover (ŝ = 5.71, SEŝ = 5.63) and snow days (ŝ = 2.42, SEŝ = 5.38), and almost absent for ice cover (ŝ = 0.33, SEŝ = 3.03) (Additional file 1: Fig. S1). Residuals of the top model (median = 0.045, SD = 0.275) fit a Gumbel extreme values distribution poorly (Kolomogorov-Smirnov test, p < 0.001). However, residuals were small relative to utility estimates (median = 0.152, SD = 5.344), indicating that the assumption was not overly restrictive [40]. There was no discernable difference between the distribution of all relocation latitudes (n = 73) and the distribution of latitudes where relocation was predicted correctly (Fisher-Pitman permutation test, 105 samples; n = 45, Z = 0.129, p = 0.898) or incorrectly (n = 28, Z =  − 0.173, p = 0.863) (Fig. 4).

Fig. 1
figure 1

Mean predicted utility ± 1 standard error (shaded area) for a resource unit as a function of covariate value over the observed range, based on the top model. All other covariates were held at mean values to calculate predictions. A Air temperature; B depth of snow cover; C accumulated days of frost; D accumulated days of snow ≥ 2.54 cm deep; E presence of ice cover ≥ 1 cm thick. Note different y-axis ranges. Negative slope values represent a decrease in the probability that the associated alternative is selected as the date for relocation

Fig. 2
figure 2

Mean predicted selection index ± 1 standard error (shaded area) as a function of covariate values in the top model (winter conditions). All other covariates were held at mean values to calculate predictions. The selection index is proportional to an alternative’s selection probability given this set of covariate values. A air temperature; B depth of snow cover; C accumulated days of frost; D accumulated days of snow ≥ 2.54 cm deep; E presence of ice cover ≥ 1 cm thick. Note different y-axis ranges

Fig. 3
figure 3

Mean predicted utility ± 1 standard error (shaded area) for a resource unit in the top model (winter conditions) as a function of temperature or frost days. A utility as a function of temperature, given high (green) or low (blue) values of snow days; B utility as a function of temperature, given high (green) or low (blue) values of snow cover depth. C, D equivalent for utility as a function of frost days. All other covariates were held at mean values to calculate predictions

Fig. 4
figure 4

Location of start points of observed mallard relocations (n = 73), indicating whether the day of relocation (day 7) was correctly predicted from the relevant choice set

Discussion

We investigated the decision of individual mallards during autumn migration in the Mississippi Flyway to relocate south, as a function of environmental covariates that were descriptive of foraging habitat conditions and flight weather. Our results indicated that relocation probability was primarily influenced by winter conditions characterized by snow cover in the previous few days and current snow cover depth. Mallards reacted most strongly to experienced snow days, with even a single snow day resulting in very high relative relocation probability (Fig. 2D). Responses to snow cover depths above 7 cm were similarly pronounced (Fig. 2B). At low snow cover or in absence of snow days in the recent past, temperatures dropping below 5 °C were an increasing incentive to relocate; whereas under snow conditions, relocation became likely at any temperature (Fig. 3A, B). Similar dynamics were predicted for experienced frost days (Fig. 3C, D). Among evaluated variables, the effect of temperature had the smallest uncertainty.

These results indicate a ranking of responses that has been suggested by several recent studies [12, 18, 57, 58]. Decreasing temperatures and the onset of frost may act as an early warning sign that elicits a response before foraging is actively impacted. Schummer et al. [12] developed a weather severity index composed of parameters similar to those in our top model (location-based cumulative snow and frost days, and current temperature and snow cover), and found that dabbling duck abundance at Missouri sites was principally correlated to the daily temperature component. Similarly, Xu and Si [18] found that greater white-fronted and swan geese timed their southwards departure from Northern Asian stopover sites by the onset of freezing temperatures, generally leaving before snow conditions became a factor. This more pronounced reaction to frost than to snow conditions was not present in our model results; however, food availability for mallards is directly diminished by snow cover on the ground in certain regions [59], and when encountered may force birds to relocate rapidly to avoid the loss of foraging opportunities [2].

Models yielded relatively large error estimates for the snow parameters (snow days and snow cover depth) due to the strongly right-skewed distribution of these data, with 99.5% of records showing values of respectively 0–3 days and 0–10 cm, and only a few records of up to 7 days and 15 cm (data not shown). While this resulted in substantial uncertainty in prediction at higher values (Figs. 1, 2), it has little impact on interpretation due to the rarity of these instances and mallards’ strong response to lower-value snow metrics (Fig. 2B, D). Allowing for this caveat, our results are consistent with a strong and rapid reaction to snow cover, and a more graduated response to temperature cues. Mallards may have a more pronounced tendency than other dabbling ducks to remain on location in the face of worsening conditions, and sometimes leave only in the event of snowfall [12, 60]. In addition to having a comparatively high body weight, mallards can also take advantage of non-wetland food sources like harvested fields [6] and are thus less dependent on ice-free shallow water than wetland-obligate dabbling ducks. This may partly explain why presence of ice cover, included in the top model as a putative “winter conditions” parameter, was found to have no notable effect on selection of relocation date (Figs. 1E, 2E); another likely cause was the low accuracy of the heuristics we used to calculate this metric, and the absence of a method to estimate fractional ice coverage of water bodies.

We found greater inter-individual variance in selection for temperature-related metrics (temperature, ŝ = 9.34; frost days, ŝ = 8.20) than for snow-related metrics (snow cover, ŝ = 5.71; snow days, ŝ = 2.42), possibly reflecting differences in condition ranges tolerated by individual mallards during the evaluated time periods close to relocation (Additional file 1: Fig. S1). Thus individual mallards might tolerate a variety of above- or below-freezing temperatures and none to several frost days before relocating, but rarely more than two snow days or a light snow cover. Differences in body mass or condition may play a role in this regard. Previous studies have shown that female mallards tended to remain at stopover sites longer [14] and migrated farther and arrived later than males [61], indicating a possibly greater tolerance for worsening weather conditions. However, we found no correlation between individual preferences and sex in our model, presumably because the data set was heavily skewed towards females (39:4), making it unlikely that sex differences could be reliably distinguished.

We hypothesized that mallards’ functional responses to environmental parameters would apply at any point during migration and in the wintering range, and therefore included relocation instances from across the whole modeled migration period (September-December) in our model. We found no discernable latitudinal differences between the set of all relocations and those of correctly or incorrectly predicted relocations (Fig. 4). This suggests that the identified responses to environmental conditions hold equally for departure from the summer range, stopover sites, and early relocations within the winter range. Similarly, van den Elsen [58] and Schummer et al. [10] found that the Schummer et al. [12] weather severity index was applicable throughout the latitudinal range of several duck species including mallards, although prediction could be further improved by adding a latitude parameter to the model [58].

Across the set of candidate models, the temperature and snow covariates, which could be considered descriptive of habitat availability, performed much better than the flight weather covariates that would be expected to influence short-term departure decisions (Table 2). In contrast, Xu and Si [18], who included snow and frost days and vegetation indices together with wind and precipitation in their goose migration model, concluded that tailwind conditions played a substantial role in predicting departure from stopover sites. O’Neal et al. [17] examined the influence of numerous flight weather parameters on the departure probability of a diverse group of dabbling ducks from an Illinois site, and found that the dominant predictors consisted of tailwind, absence of precipitation, and low cloud cover; however, the only multi-day or habitat-related parameter tested was a vegetation index. The departure-promoting effects of tailwinds in particular are well established for several goose species [29, 30, 62, 63], if less so for ducks. It is likely that in dabbling ducks, as in other bird species, large-scale environmental cues like habitat conditions drive willingness to depart at a daily or larger scale, while flight weather influences timing at a daily or smaller scale [2, 25, 64]. Our data were not well suited to model this distinction because the dynamic positioning of mallard departure locations (rather than recording at fixed sites) required the use of environmental data at a relatively coarse resolution (32 × 32 km, daily averages), which likely affected flight weather data more than multi-day habitat-related metrics and favored the impact of the latter. It is notable that while the candidate model bundling the short-term parameters (daily scale) was ranked well below the multi-day parameter models, it still achieved 53% predictive power (Table 1). We did not model interactions between the two classes of parameters because of our relatively low sample size (73 choice sets); investigating these relationships using a larger individual-based data set might be of considerable interest.

A benefit of data derived from individual tracked birds is the availability of cumulative parameters in each individual’s frame of reference rather than that of a visited location, which may be more relevant to the decision-making process. We thus calculated sequential snow and frost days as experienced by each mallard from the time of arriving at a location, in contrast to the location-continuous values used by Schummer et al. (2010, 2017) [10, 12]; however, differences between these two types of metrics were minimal in our choice sets (present in 3/73 sets; not shown) because the requirement for mallards to remain on location for multiple days prior to departure selected against locations that were already under winter conditions on arrival.

Snow as a principal migration driver is likely to be affected by warming global temperatures in the coming decades. In the Midwest region of North America, regional annual projections for the mid-twenty-first century average an increase of 2.3–2.9 °C [65], with winter temperatures from December to March increasing by 1.1–3.9 °C [66]; various regional models project reduced snowfall that may lower the number of days with snow cover of at least 1 cm by between 5 and 60 days annually [67]. The majority of investigations into the effects of climate change on duck migration have focused on spring migration, where warm conditions earlier in the year may allow better body condition and earlier arrival in the breeding range, but may also lead to a phenological mismatch between breeding stages and peak food availability [9]. However, the weather during autumn migration and the mid-winter period is also likely to be affected by climatic changes. Recent northwards shifts in the wintering distribution of diverse waterfowl and wader species have been documented (e.g. [68, 69]). Schummer et al. [10] found that the extent of areas in the Mississippi and Atlantic flyways that had winter conditions severe enough to cause mallards to relocate in the period December-February had declined from 1979 to 2013 [10]. Sauter et al. [60] reported that mid-winter movement distances of European mallards decreased between 1952 and 2004, likely due to less frequent occurrence of harsh winter conditions [60]. In North America, some studies have reported evidence that the winter distributions of North American mallards are shifting northwards [11, 70] (although these findings are based on less accurate convenience-sampled data; see also [71]). If birds remain longer at more northern latitudes in autumn and winter, this may increase foraging pressure in northerly areas and require adaptation in regional conservation planning to meet changing nutritional needs at the landscape level [72].

A substantial percentage of relocation choices remained unexplained by modeled parameters. The chosen temporal and spatial scales may have masked smaller-scale variations in weather and environmental conditions. Although our study was the first to examine and conceptualize waterfowl migration as an individual choice, the dataset and model structure limited our ability to include choice set-level parameters such as measures of wetland flooding status or vegetation cover, which may strongly affect habitat suitability for foraging [43]. The presence or absence of conspecifics may have influenced departure decisions through competition and food depletion, or by participation in flock movements. Disturbance by hunters also has been shown to factor into the timing of long-range relocation movements [73, 74]. Finally, we had no information of the body condition of individuals, which may have influenced length of stay at productive foraging sites, willingness to embark in energetically costly flight weather, and flight duration. The collection of metabolic data from free-living birds, while still complicated and costly, is becoming more feasible with the ongoing miniaturization of implantable bio-loggers [75]. Inferences about body condition and energy budget derived from such data can provide valuable additional information for the interpretation of movement records from tracked animals [76, 77]. Future research into North American waterfowl migration focusing on combining high-resolution location records with disturbance and metabolic data would be helpful for developing a comprehensive picture of the drivers of migratory decisions.

Conclusions

Our results show that among the tested environmental parameters influencing departure decision in autumn-migrating mallards, the dominant driver was the onset of snow conditions, and secondarily the onset of temperatures lowering close to or below the freezing point. Mallards are likely to relocate southwards quickly when faced with foraging impeded by snow, and use declining temperatures as a more graduated early cue for departure. The effects of short-term weather conditions presumed to be related with flight efficiency could not be distinguished in the model. Our findings provide further insights into the response of mallards to weather and climate factors during the migration period, and will be useful in the prediction and simulation of dabbling duck migratory movements under changing climatic conditions.

Availability of data and materials

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

References

  1. 1.

    Richardson WJ. Timing of bird migration in relation to weather: updated review. In: Gwinner E, editor. Bird migration: physiology and ecophysiology. Berlin: Springer; 1990.

    Google Scholar 

  2. 2.

    Newton I. Bird migration. William Collins; 2020.

  3. 3.

    Marra PP, Cohen EB, Loss SR, Rutter JE, Tonra CM. A call for full annual cycle research in animal ecology. Biol Lett. 2015;11(8):20150552.

    PubMed  PubMed Central  Google Scholar 

  4. 4.

    Rushing CS, Hostetler JA, Sillett TS, Marra PP, Rotenberg JA, Ryder TB. Spatial and temporal drivers of avian population dynamics across the annual cycle. Ecology. 2017;98(11):2837–50.

    PubMed  Google Scholar 

  5. 5.

    Winker K. Migration and speciation. Nature. 2000;404(6773):36–36.

    CAS  PubMed  Google Scholar 

  6. 6.

    Baldassarre G. Ducks, geese, and swans of North America. Baltimore: JHU Press; 2014. (Revised and updated).

    Google Scholar 

  7. 7.

    Olsen B, Munster VJ, Wallensten A, Waldenström J, Osterhaus AD, Fouchier RA. Global patterns of influenza: a virus in wild birds. Science. 2006;312(5772):384–8.

    CAS  PubMed  Google Scholar 

  8. 8.

    Field CB, Barros V, Stocker TF, Dahe Q. Managing the risks of extreme events and disasters to advance climate change adaptation: special report of the intergovernmental panel on climate change. New York: Cambridge University Press; 2012.

    Google Scholar 

  9. 9.

    Guillemain M, Pöysä H, Fox AD, Arzel C, Dessborn L, Ekroos J, et al. Effects of climate change on European ducks: what do we know and what do we need to know? Wildl Biol. 2013;19(4):404–19.

    Google Scholar 

  10. 10.

    Schummer ML, Coluccy JM, Mitchell M, Van Den Elsen L. Long-term trends in weather severity indices for dabbling ducks in eastern North America. Wildl Soc Bull. 2017;41(4):615–23.

    Google Scholar 

  11. 11.

    Meehan TD, Kaminski RM, Lebaron GS, Michel NL, Bateman BL, Wilsey CB. Half-century winter duck abundance and temperature trends in the Mississippi and Atlantic Flyways. J Wildl Manag. 2021;85:713–22.

    Google Scholar 

  12. 12.

    Schummer ML, Kaminski RM, Raedeke AH, Graber DA. Weather-related indices of autumn–winter dabbling duck abundance in Middle North America. J Wildl Manag. 2010;74(1):94–101.

    Google Scholar 

  13. 13.

    Krementz DG, Asante K, Naylor LW. Spring migration of mallards from Arkansas as determined by satellite telemetry. J Fish Wildl Manag. 2011;2(2):156–68.

    Google Scholar 

  14. 14.

    Krementz DG, Asante K, Naylor LW. Autumn migration of Mississippi Flyway mallards as determined by satellite telemetry. J Fish Wildl Manag. 2012;3(2):238–51.

    Google Scholar 

  15. 15.

    van Wijk RE, Kölzsch A, Kruckenberg H, Ebbinge BS, Müskens GJDM, Nolet BA. Individually tracked geese follow peaks of temperature acceleration during spring migration. Oikos. 2012;121(5):655–64.

    Google Scholar 

  16. 16.

    Beatty WS, Kesler DC, Webb EB, Raedeke AH, Naylor LW, Humburg DD. Quantitative and qualitative approaches to identifying migration chronology in a continental migrant. PLoS ONE. 2013;8(10):e75673.

    CAS  PubMed  PubMed Central  Google Scholar 

  17. 17.

    O’Neal BJ, Stafford JD, Larkin RP, Michel ES. The effect of weather on the decision to migrate from stopover sites by autumn-migrating ducks. Mov Ecol. 2018;6(1):23.

    PubMed  PubMed Central  Google Scholar 

  18. 18.

    Xu F, Si Y. The frost wave hypothesis: how the environment drives autumn departure of migratory waterfowl. Ecol Indic. 2019;101:1018–25.

    Google Scholar 

  19. 19.

    Van Doren BM, Horton KG. A continental system for forecasting bird migration. Science. 2018;361(6407):1115–8.

    PubMed  Google Scholar 

  20. 20.

    Gwinner E. Circadian and circannual programmes in avian migration. J Exp Biol. 1996;199:10.

    Google Scholar 

  21. 21.

    Gordo O. Why are bird migration dates shifting? A review of weather and climate effects on avian migratory phenology. Clim Res. 2007;35:37–58.

    Google Scholar 

  22. 22.

    Lovvorn JR. Distributional responses of canvasback ducks to weather and habitat change. J Appl Ecol. 1989;26(1):113.

    Google Scholar 

  23. 23.

    Newton I. Weather-related mass-mortality events in migrants. Ibis. 2007;149(3):453–67.

    Google Scholar 

  24. 24.

    Liechti F. Birds: blowin’ by the wind? J Ornithol. 2006;147(2):202–11.

    Google Scholar 

  25. 25.

    Erni B, Liechti F, Underhill LG, Bruderer B. Wind and rain govern the intensity of nocturnal bird migration in Central Europe: a log-linear regression analysis. Ardea. 2002;90:155–66.

    Google Scholar 

  26. 26.

    Åkesson S, Walinder G, Karlsson L, Ehnbom S. Reed warbler orientation: initiation of nocturnal migratory flights in relation to visibility of celestial cues at dusk. Anim Behav. 2001;61(1):181–9.

    PubMed  Google Scholar 

  27. 27.

    Sapir N, Wikelski M, Avissar R, Nathan R. Timing and flight mode of departure in migrating European bee-eaters in relation to multi-scale meteorological processes. Behav Ecol Sociobiol. 2011;65(7):1353–65.

    Google Scholar 

  28. 28.

    Schaub M, Liechti F, Jenni L. Departure of migrating European robins, Erithacus rubecula, from a stopover site in relation to wind and rain. Anim Behav. 2004;67(2):229–37.

    Google Scholar 

  29. 29.

    Fox AD, Glahder CM, Walsh AJ. Spring migration routes and timing of Greenland white-fronted geese: results from satellite telemetry. Oikos. 2003;103(2):415–25.

    Google Scholar 

  30. 30.

    Green M. Flying with the wind-spring migration of Arctic-breeding waders and geese over South Sweden. Ardea. 2004;92(2):145–59.

    Google Scholar 

  31. 31.

    Miskimen M. Meteorological and social factors in autumnal migration of ducks. Condor. 1955;57(3):179–84.

    Google Scholar 

  32. 32.

    Hagy HM, Yetter AP, Stodola KW, Horath MM, Hine CS, Ward MP, et al. Stopover duration of mallards during autumn in the Illinois River valley. J Wildl Manag. 2014;78(4):747–52.

    Google Scholar 

  33. 33.

    Edwards T, Fuqua D, James D, Kreher T, Link PT, Naylor LW, et al. Allocation of waterfowl habitat objectives within the Mississippi Alluvial Valley: an analytical framework and results. Lower Mississippi Valley Joint Venture; 2012.

  34. 34.

    Drilling N, Titman R, McKinney F. Mallard (Anas platyrhychos). In: Pool A, Gill F, editors. The birds of North America. Philadelphia: Academy of Natural Sciences; American Ornithologists’ Union; 2002. p. 658.

    Google Scholar 

  35. 35.

    Herbert JA, Chakraborty A, Naylor LW, Beatty WS, Krementz DG. Effects of landscape structure and temporal habitat dynamics on wintering mallard abundance. Landsc Ecol. 2018;33(8):1319–34.

    Google Scholar 

  36. 36.

    USFWS. Mid-winter waterfowl survey [Internet]. US Fish & Wildlife Service, Division of Migratory Bird Management; 2016. https://migbirdapps.fws.gov/mbdc/databases/mwi/mwidb.asp?opt=mwidb.

  37. 37.

    Soulliere GJ, Loges BW, Dunton EM, Luukkonen DR, Eichholz MW, Koch ME. Monitoring waterfowl in the midwest during the non-breeding period: challenges, priorities, and recommendations. J Fish Wildl Manag. 2013;4(2):395–405.

    Google Scholar 

  38. 38.

    Soulliere GJ, Al-Saffar MA, Coluccy J, Eichholz M, Gates B, Hagy H, et al. Upper Mississippi River and Great Lakes region joint venture waterfowl habitat conservation strategy-2017 revision [Internet]. Bloomington: U.S. Fish and Wildlife Service; 2017. p. 170. https://umgljv.org/docs/2017JVWaterfowlStrategy.pdf.

  39. 39.

    Bellrose FC, Crompton RD. Migrational behavior of mallards and black ducks as determined from banding. Ill Nat Hist Surv Bull. 1970;30(3).

  40. 40.

    Train K. Discrete choice methods with simulation. 2nd ed. New York: Cambridge University Press; 2009.

    Google Scholar 

  41. 41.

    Cooper AB, Millspaugh JJ. The application of discrete choice models to wildlife resource selection studies. Ecology. 1999;80:566–75.

    Google Scholar 

  42. 42.

    Collecte Localisation Satellite. ARGOSWEB user manual [Internet]. ARGOS; 2018. https://www.argos-system.org/wp-content/uploads/2018/11/User-manual-ArgosWeb-V5-2.pdf.

  43. 43.

    Beatty WS, Webb EB, Kesler DC, Raedeke AH, Naylor LW, Humburg DD. Landscape effects on mallard habitat selection at multiple spatial scales during the non-breeding period. Landsc Ecol. 2014;29(6):989–1000.

    Google Scholar 

  44. 44.

    Duchesne T, Fortin D, Courbin N. Mixed conditional logistic regression for habitat selection studies. J Anim Ecol. 2010;79(3):548–55.

    PubMed  Google Scholar 

  45. 45.

    Sarrias M, Daziano R. Multinomial logit models with continuous and discrete individual heterogeneity in R: the gmnl package. J Stat Soft [Internet]. 2017 [cited 2021 Feb 22];79(2). http://www.jstatsoft.org/v79/i02/.

  46. 46.

    Bonnot TW, Wildhaber ML, Millspaugh JJ, DeLonay AJ, Jacobson RB, Bryan JL. Discrete choice modeling of shovelnose sturgeon habitat selection in the Lower Missouri River: discrete choice modeling of shovelnose sturgeon habitat selection. J Appl Ichthyol. 2011;27(2):291–300.

    Google Scholar 

  47. 47.

    Güthlin D, Knauer F, Kneib T, Küchenhoff H, Kaczensky P, Rauer G, et al. Estimating habitat suitability and potential population size for brown bears in the Eastern Alps. Biol Conserv. 2011;144(5):1733–41.

    Google Scholar 

  48. 48.

    Beatty WS, Jay CV, Fischbach Anthony S, Grebmeier JM, Taylor RL, Blanchard AL, et al. Space use of a dominant Arctic vertebrate: effects of prey, sea ice, and land on Pacific walrus resource selection. Biol Conserv. 2016;203:25–32.

    Google Scholar 

  49. 49.

    Mesinger F, DiMego G, Kalnay E, Mitchell K, Shafran PC, Ebisuzaki W, et al. North American regional reanalysis. Bull Am Meteorol Soc. 2006;87(3):343–60.

    Google Scholar 

  50. 50.

    Ashton GD. Predicting lake ice decay. Hanover: Cold Regions Research and Engineering Lab; 1983.

    Google Scholar 

  51. 51.

    Ashton GD. Thin ice growth. Water Resourc Res. 1989;25(3):564–6.

    Google Scholar 

  52. 52.

    Gelman A. Scaling regression inputs by dividing by two standard deviations. Stat Med. 2008;27(15):2865–73.

    PubMed  Google Scholar 

  53. 53.

    Quinn GP, Keough MJ. Experimental design and data analysis for biologists. Cambridge: Cambridge University Press; 2002.

    Google Scholar 

  54. 54.

    Manley SW, Kaminski RM, Reinecke KJ, Gerard PD. Waterbird foods in winter-managed ricefields in Mississippi. J Wildl Manag. 2004;68(1):74–83.

    Google Scholar 

  55. 55.

    Arlot S, Celisse A. A survey of cross-validation procedures for model selection. Statist Surv [Internet]. 2010 [cited 2021 Jul 3];4. https://projecteuclid.org/journals/statistics-surveys/volume-4/issue-none/A-survey-of-cross-validation-procedures-for-model-selection/10.1214/09-SS054.full.

  56. 56.

    R Core Team. R: a language and environment for statistical computing [Internet]. R Foundation for Statistical Computing; 2021. https://www.R-project.org/.

  57. 57.

    Nuijten RJM, Kölzsch A, van Gils JA, Hoye BJ, Oosterbeek K, de Vries PP, et al. The exception to the rule: retreating ice front makes Bewick’s swans Cygnus columbianus bewickii migrate slower in spring than in autumn. J Avian Biol. 2014;45(2):113–22.

    Google Scholar 

  58. 58.

    Van Den Elsen L. Weather and photoperiod indices of autumn and winter dabbling duck abundance in the Mississippi and Atlantic Flyways of North America. MSc Thesis. The University of Western Ontario; 2016.

  59. 59.

    Jorde DG, Krapu GL, Crawford RD. Feeding ecology of mallards wintering in Nebraska. J Wildl Manag. 1983;47(4):1044.

    Google Scholar 

  60. 60.

    Sauter A, Korner-Nievergelt F, Jenni L. Evidence of climate change effects on within-winter movements of European Mallards Anas platyrhynchos. Ibis. 2010;152(3):600–9.

    Google Scholar 

  61. 61.

    Cristol DA, Baker MB, Carbone C. Differential migration revisited: Latitudinal segregation by age and sex class. In: Nolan V, Ketterson ED, Thompson CF, editors. Current ornithology [Internet]. Boston: Springer US; 1999 [cited 2021 Jun 26]. p. 33–88. https://doi.org/10.1007/978-1-4757-4901-4_2.

  62. 62.

    Beason RC. The influences of weather and topography on water bird migration in the southwestern United States. Oecologia. 1978;32(2):153–69.

    PubMed  Google Scholar 

  63. 63.

    Green M, Alerstam T, Clausen P, Drent R, Ebbinge BS. Dark-bellied Brent Geese Branta Bernicla bernicla, as recorded by satellite telemetry, do not minimize flight distance during spring migration. Ibis. 2002;144(1):106–21.

    Google Scholar 

  64. 64.

    Kölzsch A, Müskens GJDM, Kruckenberg H, Glazov P, Weinzierl R, Nolet BA, et al. Towards a new understanding of migration timing: slower spring than autumn migration in geese reflects different decision rules for stopover use and departure. Oikos. 2016;125(10):1496–507.

    Google Scholar 

  65. 65.

    Vose R, Easterling DR, Kunkel K, Wehner M. Temperature changes in the United States. In: Wuebbles DJ, Fahey DW, Hibbard KA, Dokken DJ, Stewart BC, Maycock TK, editors. Climate science special report: fourth national climate assessment. Washington, DC: U.S. Global Change Research Program; 2017. p. 185–206.

    Google Scholar 

  66. 66.

    Notaro M, Lorenz D, Hoving C, Schummer M. Twenty-first-century projections of snowfall and winter severity across Central-Eastern North America. J Clim. 2014;27(17):6526–50.

    Google Scholar 

  67. 67.

    Weiskopf SR, Ledee OE, Thompson LM. Climate change effects on deer and moose in the Midwest. J Wildl Manag. 2019;83(4):769–81.

    Google Scholar 

  68. 68.

    Maclean IMD, Austin GE, Rehfisch MM, Blew J, Crowe O, Delany S, et al. Climate change causes rapid changes in the distribution and site abundance of birds in winter. Glob Change Biol. 2008;14(11):2489–500.

    Google Scholar 

  69. 69.

    Lehikoinen A, Jaatinen K, Vähätalo AV, Clausen P, Crowe O, Deceuninck B, et al. Rapid climate driven shifts in wintering distributions of three common waterbird species. Glob Change Biol. 2013;19(7):2071–81.

    Google Scholar 

  70. 70.

    La Sorte FA, Thompson FR. Poleward shifts in winter ranges of North American birds. Ecology. 2007;88(7):1803–12.

    PubMed  Google Scholar 

  71. 71.

    Green AW, Krementz DG. Mallard harvest distributions in the Mississippi and Central Flyways. J Wildl Manag. 2008;72(6):1328–34.

    Google Scholar 

  72. 72.

    Lancaster JD, Wilson BC, Gulf Coast Joint Venture Waterfowl Working Group. Gulf Coast Joint Venture wintering waterfowl population and habitat objective model refinement 2021. Lafayette: Gulf Coast Joint Venture; 2021. p. 75.

  73. 73.

    Cox RR, Afton AD. Predictable interregional movements by female Northern pintails during winter. Waterbirds. 2021;23:258–69.

    Google Scholar 

  74. 74.

    Jensen GH, Tombre IM, Madsen J. Environmental factors affecting numbers of pink-footed geese Anser brachyrhynchus utilising an autumn stopover site. Wildl Biol. 2016;22(5):183–93.

    Google Scholar 

  75. 75.

    Forin-Wiart M-A, Enstipp MR, Le Maho Y, Handrich Y. Why implantation of bio-loggers may improve our understanding of how animals cope within their natural environment. Integr Zool. 2019;14(1):48–64.

    PubMed  Google Scholar 

  76. 76.

    Butler PJ, Woakes AJ. Seasonal hypothermia in migrating geese. J Exp Biol. 2001;204:7.

    Google Scholar 

  77. 77.

    Wascher CAF, Kotrschal K, Arnold W. Free-living greylag geese adjust their heart rates and body core temperatures to season and reproductive context. Sci Rep. 2018;8(1):2142.

    PubMed  PubMed Central  Google Scholar 

Download references

Acknowledgements

We thank everyone who assisted with the capture, handling, and application of transmitters to the animals used in this study, including K. Gardner, R. Theis, J.P. Fairhead, J. Jackson, J. Pagan, R. Malecki and D. Kostersky. We also thank Mariëlle van Toor, two anonymous reviewers, and the handling editor, N. Sapir, for their helpful comments on earlier versions of this manuscript.

Funding

Funding for this study was provided in part by the Natural Resource Conservation Service Conservation Effects Assessment Project (Grant No. NR183A750023C003). Arkansas Game and Fish Commission, Missouri Department of Conservation, Ducks Unlimited, Inc., and Ducks Unlimited Canada also provided additional financial and/or logistical support. The Missouri Cooperative Fish and Wildlife Research Unit is jointly sponsored by the Missouri Department of Conservation, the University of Missouri, the United States Geological Survey, the USFWS, and the Wildlife Management Institute. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

Author information

Affiliations

Authors

Contributions

EBW, FW and WSB conceived the study and designed the methodology. WSB, EBW, DCK, DGK, KA, and LWN collected the data. FW, WSB and EBW peformed the analysis. FW led the writing of the manuscript. All authors contributed to the drafts and gave final approval for publication.

Corresponding author

Correspondence to Florian G. Weller.

Ethics declarations

Ethics approval and consent to participate

During capture, banding and marking with backpack GPS satellite transmitters of mallards, all reasonable efforts were made by field personnel to minimize animal suffering [14, 16]. Mallards captured for the Beatty et al. [16] study were banded under United States federal banding permit 06569.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Additional file 1

. Supplementary Methods (censoring of positional data sets, and relocation record selection criteria) and Results (inter-individual selection variance in the top model).

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Weller, F.G., Beatty, W.S., Webb, E.B. et al. Environmental drivers of autumn migration departure decisions in midcontinental mallards. Mov Ecol 10, 1 (2022). https://doi.org/10.1186/s40462-021-00299-x

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s40462-021-00299-x

Keywords

  • Autumn migration
  • Mallard
  • Satellite tracking
  • Discrete choice model
  • Mississippi Flyway