Abstract

Midcontinent populations of arctic nesting geese (hereafter, arctic geese), including greater white-fronted geese Anser albifrons frontalis, lesser snow geese Anser caerulescens caerulescens, and Ross’s geese Anser rossii have increased in abundance and shifted their winter distribution in recent decades. Consequently, the number of arctic geese wintering in the Mississippi Alluvial Valley (MAV) has increased since the 1980’s. Stored endogenous nutrients are critically important to the life cycle of arctic geese as they are used to complete long-distance migration events, survive harsh winters, and supplement nutrients needed for reproduction. This study tracked temporal changes in body condition of arctic geese during the wintering period. Arctic geese were collected from October–February 2015–2016 and 2016–2017 in eastern Arkansas, USA. Proximate analysis was used to determine size of lipid and protein stores as an index of body condition. Protein stores were more stable through time than lipids, but a slight increase was observed in all species as winter progressed. Mean lipid stores were dynamic and were highest in November (x̄ greater white-fronted = 442.6 g, x̄ snow = 308.0 g, x̄ Ross’s = 268.25 g) and lowest in February (x̄ greater white-fronted = 215.1 g, x̄ snow = 142.0 g, x̄ Ross’s = 103.0 g). Greater white-fronted geese arrived earliest to the MAV and experienced an increase in endogenous lipid stores during early winter when high-energy food resources were most abundant. Conversely, snow and Ross’s geese arrived to the MAV later and did not appear to increase their lipid stores upon arrival. All three species experienced a decline in stored lipid mass as winter progressed; the decline may have been driven by a combination of factors such as resource depletion, a shift in dietary needs, physiological factors, hunting pressure, and increased energetic demands. An improved understanding of the role that “non-traditional” wintering grounds exert on the nutrient dynamics of arctic geese may aid in the management of growing and shifting populations. Keywords: agriculture, Arkansas, endogenous, greater white-fronted goose, lesser snow goose, lipids, proximate analysis, Ross’s goose

Statistical Methods

Principle Components Analysis

Methods: As an intermediate step, we performed a principle component analysis (PCA) using five morphological measurements (culmen, head length, tarsus length, middle toe length, and wing chord) to create a composite variable (i.e., principal component) to represent structural size (Alisauskas and Ankney 1987). We conducted a PCA for each species using prcomp in the Program R statistical package (R Core Development Team) and generated eigenvalues to assess the proportion of total variance in structural size explained by each of the principle components. Subsequently, we conducted a regression analysis to examine the relationship of the principle component that explained the most variation in structural size (PC1) with lipid and protein content (Ankney and Alisauskas 1987).

Results: Principle component analysis for each species effectively reduced five morphological measurements to a single composite variable used as an index of body size. For greater white-fronted geese, PC1 accounted for 67.2% of the variation in the structural measurements. For snow and Ross’s geese PC1 accounted for 73.0% and 70.1% of the variation, respectively. For all three species, PC2 accounted for less than 15.0% of the variation in the structural measurements, thus we elected to use PC1 as the index of structural size.

Regression analysis (PCA)

GWFG

model	term	estimate	std_error	statistic	p_value
protein ~ PC1	(Intercept)	469.48	4.09	114.79	0
protein ~ PC1	PC1	23.82	2.24	10.65	0

model	term	estimate	std_error	statistic	p_value
fat ~ PC1	(Intercept)	321.75	8.88	36.23	0.0000
fat ~ PC1	PC1	10.80	4.86	2.23	0.0274

ROGO

model	term	estimate	std_error	statistic	p_value
protein ~ PC1	(Intercept)	273.19	4.71	58.04	0
protein ~ PC1	PC1	-13.21	2.55	-5.18	0

model	term	estimate	std_error	statistic	p_value
fat ~ PC1	(Intercept)	147.68	9.64	15.32	0.0000
fat ~ PC1	PC1	-4.62	5.22	-0.88	0.3826

SNGO

model	term	estimate	std_error	statistic	p_value
protein ~ PC1	(Intercept)	429.67	3.42	125.69	0
protein ~ PC1	PC1	-20.64	1.80	-11.50	0

model	term	estimate	std_error	statistic	p_value
fat ~ PC1	(Intercept)	205.69	6.36	32.35	0.0000
fat ~ PC1	PC1	-8.89	3.34	-2.66	0.0087

Results: Total lipids (g) were correlated with body mass for all species but only weakly related to PC1 representing structural size. Total protein (g) was highly correlated with total body mass and structural size (PC1) for all species. To control for differing body size among individuals, total lipid and protein were corrected for structural size (PC1) by species.

Table 2 in manuscript

Calculate size-corrected protein and fat

Methods: If a significant linear relationship (p ≤ 0.05) was detected, we applied a correction factor to adjust lipid and protein for within-species differences in structural size (Ankney and Alisauskas 1991). The correction factor derived by Ankney and Alisauskas (1991) used the residuals from the linear regression equation to correct for structural size. The corrected values were used in further analyses. We used the Ankney and Alisauskas (1991) equation:

y1=yobs-[a+b(PC1)]+Ӯobs

y1 = corrected value
a = y-intercept of regression line
b = the slope of the regression line
yobs = actual value
Ӯobs = mean of actual values

GWFG

Figures show mass of interest (protein or fat) relationship with PC1 before and after correction. If correction worked properly, there should be no relationship between mass and PC1 after correction.

ROGO

SNGO

Lipid Analysis

Methods: Using data pooled for all species, we tested the effects of demographic variables and time of collection. We used body size corrected lipid mass as a continuous response variable. We used species, age, sex, year of study, and month of collection as categorical predictor variables, and day of year of collection as a continuous predictor variable. Species contained three levels (greater white-fronted, snow, and Ross’s), age contained two levels (adult and juvenile), sex contained two levels (male and female), year of study contained two levels (winter 2015-2016 and winter 2016-2017), and month of collection contained five levels (October, November, December, January, and February) where samples from both years of the study were pooled by month. For the continuous predictor variable, we used the day of year on which the specimen was collected, wherein 2 October was equal to day 275 and the year of the study was not considered (both years were pooled). We used forward stepwise selection to investigate contributive predictors and relevant interactions between fixed effects based on presence in the top-performing model. We fit linear models using Program R and determined top models using corrected Akaike Information Criteria (AICc). Models with a ∆AICc < 2 were considered equivalent, and from equivalent models, we selected the most parsimonious model.

Model selection

We first applied the same models with season(year of collection) as a random effect, but was unimportant, so we left out random effects.

Top model is corrected fat ~ month + species (no interaction)

Model comparisons

Table 3 in manuscript

## 
## Model selection based on AICc:
## 
##                  K    AICc Delta_AICc AICcWt Cum.Wt       LL
## lipid.mospp      8 4098.33       0.00   0.99   0.99 -2040.96
## lipid.moxspp    16 4107.92       9.59   0.01   1.00 -2037.16
## lipid.moyear     7 4188.90      90.57   0.00   1.00 -2087.29
## lipid.month      6 4194.76      96.43   0.00   1.00 -2091.26
## lipid.dayssince  3 4241.39     143.06   0.00   1.00 -2117.66
## lipid.spp        4 4260.06     161.73   0.00   1.00 -2125.97
## lipid.sppsex     5 4260.72     162.39   0.00   1.00 -2125.27
## lipid.sppagesex  6 4262.26     163.92   0.00   1.00 -2125.01
## lipid.year       3 4384.24     285.91   0.00   1.00 -2189.09
## lipid.age        3 4394.30     295.97   0.00   1.00 -2194.11
## lipid.sex        3 4394.48     296.15   0.00   1.00 -2194.21
## lipid.agesex     4 4396.29     297.96   0.00   1.00 -2194.09

## 'log Lik.' -2040.958 (df=8)

## 'log Lik.' -2037.156 (df=16)

## 'log Lik.' -2087.289 (df=7)

## 'log Lik.' -2091.258 (df=6)

## $deltaAIC
## [1]  0.00  9.59 94.98 96.43
## 
## $rel.LL
## [1] 1.000000000000000000000000000 0.008270998827770234193557108
## [3] 0.000000000000000000002373313 0.000000000000000000001149454
## 
## $weights
## [1] 0.991796849421052284512256847 0.008203150578947734569701389
## [3] 0.000000000000000000002353845 0.000000000000000000001140025

Selected model summary

Top model is corrected fat ~ month + species (no interaction)

Table 4 in manuscript

term	estimate	std.error	statistic	p.value
intercept	328.93	9.64	34.13	0.0000
pred_data_species_rogo	-142.15	14.54	-9.78	0.0000
pred_data_species_sngo	-75.84	9.92	-7.64	0.0000
pred_data_collection_month_november	101.10	15.53	6.51	0.0000
pred_data_collection_month_december	16.93	14.60	1.16	0.2471
pred_data_collection_month_january	-54.93	13.39	-4.10	0.0001
pred_data_collection_month_february	-109.50	14.05	-7.79	0.0000

Figure associated with best model

Figure 2 in manuscript

Fig. 2 Temporal trends in the lipid stores (g) of greater white-fronted geese Anser albifrons frontalis, lesser snow geese Anser caerulescens caerulescens, and Ross’s geese Anser rossii collected in southeast Arkansas from October–February 2015–2016 and 2016–2017. The x-axis represents the day of year on which the specimen was collected, wherein October 2nd is equal to day 275 and March 1st is equal to day 60.

Tukey multiple comparisons

Methods: We conducted an analysis of variance to test for differences in corrected lipid and protein content among species and sex/age classes and within species. Tukey-Kramer least square means multiple comparison tests were used to quantify differences between groups.

## 
##   Simultaneous Tests for General Linear Hypotheses
## 
## Multiple Comparisons of Means: Tukey Contrasts
## 
## 
## Fit: aov(formula = predfat ~ collection_month, data = pred_data)
## 
## Linear Hypotheses:
##                          Estimate Std. Error t value Pr(>|t|)    
## November - October == 0     92.59      17.77   5.212  <0.0001 ***
## December - October == 0    -24.57      16.05  -1.530   0.5386    
## January - October == 0    -111.21      13.59  -8.186  <0.0001 ***
## February - October == 0   -150.69      15.19  -9.918  <0.0001 ***
## December - November == 0  -117.16      18.18  -6.444  <0.0001 ***
## January - November == 0   -203.81      16.04 -12.703  <0.0001 ***
## February - November == 0  -243.28      17.43 -13.960  <0.0001 ***
## January - December == 0    -86.65      14.12  -6.135  <0.0001 ***
## February - December == 0  -126.12      15.68  -8.045  <0.0001 ***
## February - January == 0    -39.47      13.14  -3.004   0.0232 *  
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## (Adjusted p values reported -- single-step method)

## 
##   Simultaneous Tests for General Linear Hypotheses
## 
## Multiple Comparisons of Means: Tukey Contrasts
## 
## 
## Fit: aov(formula = predfat ~ species, data = pred_data)
## 
## Linear Hypotheses:
##                  Estimate Std. Error t value Pr(>|t|)    
## ROGO - GWFG == 0  -174.08      17.55  -9.920  < 0.001 ***
## SNGO - GWFG == 0  -116.06      10.92 -10.629  < 0.001 ***
## SNGO - ROGO == 0    58.02      17.86   3.247  0.00365 ** 
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## (Adjusted p values reported -- single-step method)

## 
##   Simultaneous Tests for General Linear Hypotheses
## 
## Multiple Comparisons of Means: Tukey Contrasts
## 
## 
## Fit: aov(formula = predfat ~ agesex, data = pred_data)
## 
## Linear Hypotheses:
##                Estimate Std. Error t value Pr(>|t|)
## A_M - A_F == 0   -4.835     14.008  -0.345    0.985
## J_F - A_F == 0  -14.227     23.619  -0.602    0.928
## J_M - A_F == 0   21.078     22.275   0.946    0.771
## J_F - A_M == 0   -9.392     22.946  -0.409    0.976
## J_M - A_M == 0   25.913     21.561   1.202    0.614
## J_M - J_F == 0   35.305     28.749   1.228    0.597
## (Adjusted p values reported -- single-step method)

Figure 3 in manuscript

Fig. 3 Results of Tukey-Kramer multiple comparisons test of lipid stores (g) by collection month for arctic geese collected in southeast Arkansas from October–February 2015–2016 and 2016–2017. For greater white-fronted geese Anser albifrons frontalis, mean monthly lipid mass was higher in November than all other months and lower in January and February than all other months. For lesser snow Anser caerulescens caerulescens and Ross’s geese Anser rossii mean monthly lipid mass was lower in January and February than in November and December. Black error bars represent statistically significant differences between mean monthly values (p < 0.05).

Protein Analysis

Methods: Using data pooled for all species, we tested the effects of demographic variables and time of collection. We used body size corrected protein mass as a continuous response variable. We used species, age, sex, year of study, and month of collection as categorical predictor variables, and day of year of collection as a continuous predictor variable. Species contained three levels (greater white-fronted, snow, and Ross’s), age contained two levels (adult and juvenile), sex contained two levels (male and female), year of study contained two levels (winter 2015-2016 and winter 2016-2017), and month of collection contained five levels (October, November, December, January, and February) where samples from both years of the study were pooled by month. For the continuous predictor variable, we used the day of year on which the specimen was collected, wherein 2 October was equal to day 275 and the year of the study was not considered (both years were pooled). We used forward stepwise selection to investigate contributive predictors and relevant interactions between fixed effects based on presence in the top-performing model. We fit linear models using Program R and determined top models using corrected Akaike Information Criteria (AICc). Models with a ∆AICc < 2 were considered equivalent, and from equivalent models, we selected the most parsimonious model.

Model selection

We first applied the same models with season(year of collection) as a random effect, but was unimportant, so we left out random effects.

Top model is corrected protein ~ species + age*sex

Model comparisons

Table 3 in manuscript

## 
## Model selection based on AICc:
## 
##                     K    AICc Delta_AICc AICcWt Cum.Wt       LL
## protein.sppagexsex  7 3686.07       0.00   0.82   0.82 -1835.87
## protein.sppagesex   6 3689.10       3.04   0.18   1.00 -1838.43
## protein.mospp       8 3725.44      39.38   0.00   1.00 -1854.51
## protein.sppsex      5 3738.34      52.27   0.00   1.00 -1864.08
## protein.moxspp     16 3741.61      55.54   0.00   1.00 -1854.00
## protein.spp         4 3741.87      55.80   0.00   1.00 -1866.88
## protein.agesex      4 4057.33     371.26   0.00   1.00 -2024.61
## protein.year        3 4060.45     374.38   0.00   1.00 -2027.19
## protein.age         3 4060.45     374.39   0.00   1.00 -2027.19
## protein.moyear      7 4061.44     375.37   0.00   1.00 -2023.56
## protein.sex         3 4064.34     378.27   0.00   1.00 -2029.13
## protein.dayssince   3 4068.86     382.79   0.00   1.00 -2031.40
## protein.month       6 4069.53     383.47   0.00   1.00 -2028.65

## 'log Lik.' -1835.872 (df=7)

## 'log Lik.' -1838.431 (df=6)

## 'log Lik.' -1854.514 (df=8)

## 'log Lik.' -1864.083 (df=5)

## $deltaAIC
## [1]  0.00  3.03 39.37 52.27
## 
## $rel.LL
## [1] 1.000000000000000000 0.219808184847789689 0.000000002824314942
## [4] 0.000000000004463892
## 
## $weights
## [1] 0.819801022900782050 0.180198974780182181 0.000000002315376279
## [4] 0.000000000003659504

Selected model summary

Top model is corrected protein ~ species + age*sex

Table 4 in manuscript

term	estimate	std.error	statistic	p.value
intercept	470.13	4.76	98.80	0.0000
pred_data_species_rogo	-200.64	7.83	-25.63	0.0000
pred_data_species_sngo	-41.71	4.90	-8.51	0.0000
pred_data_age_j	-29.16	8.67	-3.36	0.0009
pred_data_sex_m	15.31	5.15	2.97	0.0032
pred_data_age_j_pred_data_sex_m	-26.44	11.74	-2.25	0.0250

Figure associated with best model

Figure 4 in manuscript

Fig. 4 Temporal trends in protein stores by month of collection for greater white-fronted geese Anser albifrons frontalis, lesser snow geese Anser caerulescens caerulescens, and Ross’s geese Anser rossii collected in southeast Arkansas and used in our analysis of winter body condition from October–February 2015–2016 and 2016–2017.

Tukey multiple comparisons

## 
##   Simultaneous Tests for General Linear Hypotheses
## 
## Multiple Comparisons of Means: Tukey Contrasts
## 
## 
## Fit: aov(formula = predpro ~ collection_month, data = pred_data)
## 
## Linear Hypotheses:
##                          Estimate Std. Error t value Pr(>|t|)
## November - October == 0    14.622     14.894   0.982    0.861
## December - October == 0   -20.484     13.458  -1.522    0.544
## January - October == 0    -11.259     11.389  -0.989    0.858
## February - October == 0    -3.261     12.737  -0.256    0.999
## December - November == 0  -35.107     15.242  -2.303    0.143
## January - November == 0   -25.881     13.450  -1.924    0.302
## February - November == 0  -17.883     14.609  -1.224    0.734
## January - December == 0     9.226     11.840   0.779    0.935
## February - December == 0   17.223     13.142   1.311    0.681
## February - January == 0     7.998     11.014   0.726    0.949
## (Adjusted p values reported -- single-step method)

## 
##   Simultaneous Tests for General Linear Hypotheses
## 
## Multiple Comparisons of Means: Tukey Contrasts
## 
## 
## Fit: aov(formula = predpro ~ species, data = pred_data)
## 
## Linear Hypotheses:
##                  Estimate Std. Error t value      Pr(>|t|)    
## ROGO - GWFG == 0 -196.293      8.458 -23.209 <0.0000000001 ***
## SNGO - GWFG == 0  -39.810      5.263  -7.564 <0.0000000001 ***
## SNGO - ROGO == 0  156.483      8.611  18.173 <0.0000000001 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## (Adjusted p values reported -- single-step method)

## 
##   Simultaneous Tests for General Linear Hypotheses
## 
## Multiple Comparisons of Means: Tukey Contrasts
## 
## 
## Fit: aov(formula = predpro ~ agesex, data = pred_data)
## 
## Linear Hypotheses:
##                Estimate Std. Error t value Pr(>|t|)  
## A_M - A_F == 0  21.8800     8.6952   2.516   0.0550 .
## J_F - A_F == 0 -18.1099    14.6617  -1.235   0.5923  
## J_M - A_F == 0 -17.6786    13.8275  -1.279   0.5643  
## J_F - A_M == 0 -39.9899    14.2439  -2.808   0.0253 *
## J_M - A_M == 0 -39.5585    13.3837  -2.956   0.0162 *
## J_M - J_F == 0   0.4314    17.8460   0.024   1.0000  
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## (Adjusted p values reported -- single-step method)

Figure 5 in manuscript

Fig. 5 Results of Tukey-Kramer multiple comparisons test of protein stores (g) by collection month for arctic geese collected in southeast Arkansas and used in our analysis of winter body condition from October–February 2015–2016 and 2016–2017. For greater white-fronted geese Anser albifrons frontalis, mean monthly protein mass was higher in January and February than in October. There was no difference in monthly mean protein mass for lesser snow Anser caerulescens caerulescens or Ross’s geese Anser rossii. Black error bars represent statistically significant differences between mean monthly values (p < 0.05).

Data is available at: Carlson, Lindsay Gray; Massey, Ethan; Osborne, Douglas (2019), Temporal Trends in Body Condition of Arctic Geese Wintering in the Mississippi Alluvial Valley, Dryad, Dataset, https://doi.org/10.5061/dryad.0gb5mkkwq

Temporal Trends in Body Condition of Arctic Geese Wintering in the Mississippi Alluvial Valley: Statistical Analysis

LGCarlson

2/17/2019

Abstract

Statistical Methods

Principle Components Analysis

Regression analysis (PCA)

GWFG

ROGO

SNGO

Calculate size-corrected protein and fat

y1=yobs-[a+b(PC1)]+Ӯobs

GWFG

ROGO

SNGO

Lipid Analysis

Model selection

Top model is corrected fat ~ month + species (no interaction)

Model comparisons

Selected model summary

Top model is corrected fat ~ month + species (no interaction)

Figure associated with best model

Tukey multiple comparisons

Protein Analysis

Model selection

Top model is corrected protein ~ species + age*sex

Model comparisons

Selected model summary

Top model is corrected protein ~ species + age*sex

Figure associated with best model

Tukey multiple comparisons