Content uploaded by Leann Kanda
Author content
All content in this area was uploaded by Leann Kanda on Jun 12, 2020
Content may be subject to copyright.
Winter energetics of Virginia opossums Didelphis virginiana and
implications for the species’ northern distributional limit
L. Leann Kanda
Kanda, L. L. 2005. Winter energetics of Virginia opossums Didelphis virginiana and
implications for the species’ northern distributional limit. /Ecography 28: 731 /744.
While climatic limitations are widely recognized as primary factors determining the
distributions of many species, the physiological link between climate and species’
persistence is poorly understood. The Virginia opossum Didelphis virginiana is a species
for which winter energetics have been evaluated and a northern geographical limit has
been hypothesized. Expansion of opossum populations beyond this limit, however,
suggests that a previous winter energetics model requires modification. I update this
energetics model by incorporating random foraging success to estimate the probability
of opossum survival under varying winter temperature regimes. Estimation of opossum
‘‘success’’ for winters in Amherst, Massachusetts, since 1926 showed that juvenile
females, the key breeding component of the population, would survive at a rate high
enough to maintain a stable population in only 4 of the 77 yr. The model correctly
predicted the fate of 13 of 14 opossums monitored in the Amherst area during the
winters of 2000 /2003. The current energetics model does not correctly predict autumn
weight gain, but it does accurately estimate opossum winter survival. However, the
model predicts that opossums should be winter-limited in areas such as Amherst,
Massachusetts, where in fact they are well established. This discrepancy may be
explained in three ways: weather station data do not adequately reflect available
microclimates, opossums show high levels of flexibility in cold-weather foraging
behavior, and most likely, humans provide food and shelter that mitigate the effect of
winter.
L. L. Kanda (lkanda@bio.umass.edu), Graduate Program in Organismic and
Evolutionary Biology, 319 Morrill Science Center South, Univ. of Massachusetts,
Amherst 01003, MA, USA.
Climatic factors have long been acknowledged as
principal agents responsible for limiting the distributions
of species (Wallace 1876, Rosenzweig 1995, McNab
2002). As humans modify local and global climates,
understanding the mechanisms by which climate deter-
mines species’ persistence is increasingly important
(Humphries et al. 2002). In many cases, however, the
implications of climate change for species’ ranges are
based upon broad correlations, and proximate mechan-
isms are rarely examined (Humphries et al. 2002,
McNab 2002: Chapter 14).
The northern distribution of the Virginia opossum
Didelphis virginiana is perhaps one of the best-studied
cases of mammalian range limitation by climate. A
tropically evolved species, the opossum’s distribution
in North America has been expanding northward
in historical times (Gardner and Sunquist 2003). Corre-
lations have been reported between the opossum’s
northern range limit and the /78C January isotherm
(Tyndale-Biscoe 1973), and northern animals have been
observed with severely frostbitten tails and ears (Long
and Copes 1968). At the northern edge of their dis-
tribution opossums have been found dead of starvation
in early spring (Blair 1936), and a correlation of oposs-
um sightings in Ontario with warm winters is sugges-
tive of climatic limitations (Peterson and Downing 1956,
Accepted 2 May 2005
Copyright #ECOGRAPHY 2005
ISSN 0906-7590
ECOGRAPHY 28: 731 /744, 2005
ECOGRAPHY 28:6 (2005) 731
de Vos 1964). The winter energetics of opossums have
been examined in the laboratory (Brocke 1970, Pippitt
1976, Hsu et al. 1988); these studies indicate that
opossums have poor thermoregulatory ability in cold
weather, poor fur insulation, and that they reduce the
effect of cold winters primarily through behavioral
avoidance. For example, opossums do not forage on
nights below ca
/48C (Wiseman and Hendrickson 1950,
Brocke 1970). Brocke (1970) used energetic parameters
obtained in the laboratory to estimate the physiological
limitation of opossums, then linked this estimation with
climatic data to hypothesize that the Virginia opossum
was at or close to its potential northern distribution in
central and northeastern North America.
The current distribution of the opossum casts doubt
on the predicted winter physiological limit. Opossums
have expanded their range into areas where it was
believed they could not survive. The most generous
distributional estimate given by Brocke (1970) suggested
opossums could survive as far north as mid-Vermont
and New Hampshire. However, opossums currently are
reported throughout these states and into Canada
(Gardner and Sunquist 2003).
Given this contradiction in the expected and actual
distributions of opossums, I re-examined Brocke’s (1970)
energetic model to determine if any of its assumptions or
limitations reduce its applicability to northern opossum
population persistence. I modified the energetic para-
meters to form an updated model, and applied this
model to the Amherst, Massachusetts, region. Accord-
ing to all previous studies (Brocke 1970, Tyndale-Biscoe
1973, Pippitt 1976, Gardner 1982) Massachusetts should
fall within the physiological distributional limit of
Virginia opossums, and they first appeared in the state
in the early 1920s (Gardner and Sunquist 2003). How-
ever, recent road-kill and camera surveys indicate
opossums are not present in central Massachusetts
forests outside of the urbanized landscape, suggesting
that the urban-rural interface in Massachusetts is a
distributional edge (Kanda 2005). The new energetics
model described in this paper predicts the fate of
opossums given a range of pre-winter weights, success
in foraging, and actual climatic data from Amherst,
Massachusetts over the last 77 yr. I then tested the model
using data from 14 winter-monitored opossums living in
Amherst.
Brocke’s (1970) energetic model
The most detailed winter energetic parameters available
for the Virginia opossum come from Brocke’s (1970)
dissertation on animals from a Michigan population. He
estimated the energy intake and energy expenditure of
opossums under two scenarios: days in which opossums
leave their den to forage, and days in which opossums
remain in their dens. For each scenario, Brocke estimated
the energy intake and energy expenditure of captive
animals in outdoor pens. For energy intake if the animal
forages, Brocke measured the energy in daily ad libitum
intake of opossums feeding on carrion. For energy
expenditure, he measured weight loss of captive animals
over periods when they chose to be ‘‘highly active’’ (i.e.,
foraging) and periods when they remained inactive in
their dens. Brocke used observed weight loss in captive
animals, as well as condition of roadkill carcasses
throughout the winter, to estimate that the average
opossum could lose up to 42% of its pre-winter body
mass before starving to death. Because opossums
catabolize muscle as well as fat reserves, Brocke mea-
sured that the opossum converts one gram of body mass
to 4.4 kcal of energy (E
W
/4400 kcal kg
1
).
Brocke’s arrangement of the energetic equations is not
immediately intuitive; here I rearrange them slightly for
interpretation in general terms of energy input and
energy expenditure (using Brocke’s notations where
applicable; for parameter definitions and specific mea-
surements see Table 1). The total winter energy expen-
diture (E
T
) is the sum of energy expenditure on resting
(D
S
) and foraging (D
F
) days over winter:
ETDSESDFEF
On foraging days the animal will add to its energy stores
a daily intake E
I
, so total energy gained overwinter from
food (E
G
)is
EGDFEI
and therefore net energy loss (DE) overwinter is simply
the difference,
DEETEG
The net energy lost overwinter comes from the opos-
sum’s body reserves (W). The maximum net energy loss
permissible before starvation occurs is therefore the
energy obtainable from the reserve body mass
DEmax WEW
The maximum number of energy-expensive resting days
an opossum can tolerate overwinter, then, is determined
with
/substitution and rearrangement:
WEWETEGWEWDSESDFEFDFEI
DSWEW120 (EIEF)
ESEIEF
Brocke concluded from captive and road-killed sam-
ples that the average pre-winter opossum (A
(O)
) weighed
3.04 kg. Based on his energetic measurements for an
animal of this size (Table 1), Brocke calculated that a
3.04 kg opossum could theoretically survive up to DS
/
90 resting days, interspersed with 30 foraging days, if it
eats ad libitum on the foraging days. Brocke also
estimated that the average juvenile opossum would be
732 ECOGRAPHY 28:6 (2005)
2.59 kg (A
(J)
) before winter, and could therefore lose
1.09 kg (W
(J)
). Brocke found that such smaller animals
had slightly lower energy expenditure per day but could
not eat as much ad libitum food as the average opossum
(Table 1). Juvenile opossums should therefore be able to
tolerate no more than 85 forced resting days in a 120-d
winter.
What dictates the foraging behavior of an opossum in
winter? Like Wiseman and Hendrickson (1950), Brocke
had observed that winter opossum foraging activity is
strongly linked to ambient temperature; opossums gen-
erally stop foraging if night temperatures are 5
//48C.
(Pippitt [1976] found that opossums avoid foraging in
temperatures much below 08C because they lose thermo-
regulatory control in such temperatures.) Based upon
daily temperature relationships in Lansing, Michigan,
Brocke generalized that opossums would not forage if
the daily maximum temperature did not rise above
freezing.
In the end, Brocke realized that the opossum popula-
tion should fail before reaching the conditions that
restrict foraging to 30 d a winter, since ad libitum
foraging in winter conditions is not realistic. He
concluded that the average opossum should tolerate no
more than 70 d of enforced resting in the den over
winter, with 50 d of foraging. Though he built many
models exploring opossum energetics when foraging fails
to produce food, he had no empirical energetic evidence
for this final decision. It was probably influenced by the
fact that at the time the distributional limit of opossums
in Michigan coincided with the isotherm of 70 d/winter
with daily maximum temperatures below freezing.
Brocke (1970) further generalized the geographical
limitation as the southern edge of the pine-hemlock
ecotone. These two estimates (the isotherm and the
ecotone) are typically quoted as the geographical limit
imposed by the winter physiology of the opossum
(Gardner 1982, Hsu et al. 1988, McNab 2002: Chapter
14, Gardner and Sunquist 2003).
Methods
I formed a time-stepped computer model using
MATLAB Student (Anon. 1997) that incorporated
the energetic parameters measured by Brocke (1970)
(Table 1). Instead of using the ratio of foraging to non-
foraging days to estimate the theoretical range limit, I
used weather station temperature data for winters in
Amherst, Massachusetts (Northeast Regional Climate
Center, Ithaca, NY and National Climatic Data Center,
Asheville, NC) to estimate whether opossums would
forage on a given day, and thus generated predictions of
if and when an opossum would be expected to starve to
death during an Amherst winter.
The model was constructed in a time-incremented
fashion for two reasons. First, the order in which an
opossum encounters enforced non-foraging days should
be important (the animals may be able to survive a
certain number of non-foraging days on average, but not
if they all occur before the foraging days that replenish
energy stores). Second, if animals are not eating
ad libitum every foraging trip, the order in which they
encounter more or less food in foraging bouts over
winter could also be very important. This is also why the
random foraging success model is stochastic; it allows
the generation of a confidence interval on the model
prediction of death date or weight of live animals in the
spring, as well as weight of animals on any given date
over winter.
Table 1. Virginia opossum winter physiological parameters measured by Brocke (1970). Parameter names are mine, adapted from
Brocke where possible.
Parameter Measure Description
D
S
Number of days in a 120-d winter spent resting in the den.
D
F
120-D
S
Number of days in a 120-d winter on which foraging occurs.
E
s
Energy spent during a resting day.
E
S(O)
113 kcal Average opossum resting energy spent in 24 h.
E
S(J)
102 kcal Juvenile resting energy spent in 24 h.
E
F
Energy spent during a foraging day.
E
F(O)
158 kcal Average opossum foraging energy spent in 24 h.
E
F(J)
154 kcal Juvenile foraging energy spent in 24 h.
E
I
Energy taken in from food on a foraging day.
E
I(O)
315 kcal Ad libitum intake for the average opossum.
E
I(J)
270 kcal Ad libitum intake for a juvenile opossum.
L 131 kcal kg
1
Ad libitum intake in 24 h by body weight.
A Autumn opossum weight (1 December).
A
(O)
3.04 kg Autumn average opossum weight.
A
(J)
2.59 kg Autumn average juvenile weight.
A
(FJ)
2.37 kg Autumn average juvenile female weight.
P 0.42 Percent acceptable weight loss.
W AP Mass of animal convertable to energy.
E
W
4400 kcal kg
1
Opossum weight to energy conversion.
ECOGRAPHY 28:6 (2005) 733
The model requires input of the beginning opossum
weight and a series of daily maximum temperatures
representing a particular winter. In a loop procedure,
each daily temperature was used to determine whether
the opossum ‘‘foraged’’ or not. If the temperature was
]
/08C, the energy reserve parameter (the amount of
body mass currently available before starvation) was
increased by the net foraging gain; otherwise the resting
expenditure was subtracted. If the energy reserve
dropped to or below 0, the loop ended and the
hypothetical opossum was considered dead; otherwise,
the loop incremented to the next day.
Perfect foraging juvenile simulation
I first constructed the model to simulate the ad libitum
juvenile energetic model used by Brocke (1970). I began
the model with the estimate of the autumn mean juvenile
weight (Table 1), rather than the mean opossum weight,
because juveniles are the most important age class for
the upcoming breeding season. It is rare for an opossum
to reach its second or third winter, as opossums are
very short-lived for their size (Kanda and Fuller 2004).
In the favorable climate of Florida, no more than 26% of
1-yr-old females lived to be 2 yr old (Sunquist and
Eisenberg 1993). Farther north, Seidensticker et al.
(1987) observed only 8% survival from 1 to 2 yr old in
Virginia, and in Wisconsin no monitored females lived to
their second breeding season (Gillette 1980). Given the
apparent rarity of northern opossums successfully con-
tributing to reproduction after their first breeding season
(i.e. over-winter adults form a very small part of the
subsequent breeding population), I considered juveniles
to be the most important age class.
To replicate the ad libitum model in the time-stepped
form, I set all the parameters to the energetic parameters
measured and used by Brocke (1970) in his juvenile
model (Table 1). I calculated mass changes of a
hypothetical opossum that began with an input mean
juvenile weight (A
(J)
/2.59 kg), and which could lose
42% (P) of this weight before starvation (i.e., W
(J)
/
A
(J)
P/1.09 kg). Winter temperatures were defined in a
matrix T
i
where T
1
is the maximum temperature (8C)
from 1 December. W
i
represents the body mass available
for energy conversion on day i (i.e., W
0
/W
(J)
). Days
were incremented such that for i
/1 to 120,
if TiB0;WiWi1(ES(J)=EW)
if Ti]0;WiWi1[(EI(J)EF(J))=EW]
until either i
/120 or W5/0.
This simulation was run using temperature data from
Amherst winters between December 1926 and March
2003. As it is a deterministic model, it was run only once
for each winter data set. For each year, the expectation of
whether the opossum would live or starve to death was
recorded.
Normally distributed foraging success model
Ad libitum foraging probably is unlikely during northern
winters so I sought to simulate the random foraging
success that an opossum is more likely to encounter in
each foraging bout. Only anecdotal observations have
been made of the forage available to opossums (a few
direct observations or snow-tracks; Wiseman and Hen-
drickson 1950, Brocke 1970, Kanda unpubl.). Based on
these anecdotes, I knew that in a winter landscape it
would be possible for opossums to fail to find anything
to eat during some foraging bouts. However, opossums
also occasionally encountered sources such as carrion
which were probably sufficient for ad libitum foraging
opportunity. Finally, it seems reasonable that each of
these scenarios is less likely than the opossum locating
some intermediate amount of food (for example, insects
burrowed in a log). I selected a model that required
the simplest assumptions that would reflect this anecdo-
tal observation of foraging success: a normal distri-
bution where on average an opossum obtains 50% of its
ad libitum intake.
I therefore modeled foraging success as a normal
distribution with mean of 50% maximum ad libitum
intake, and standard deviation of 25% of the maximum
intake. I constrained the distribution between 0 and
131 kcal kg
1
of opossum. Brocke (1970) calculated
131 kcal kg
1
of opossum as the ad libitum intake; the
constant figure of 270 kcal (E
I(J)
) represents 131 kcal
kg
1
/2.06 kg (mean mid-winter juvenile opossum
weight). Because my model incorporates time, the
current opossum weight reenters the calculation. Thus
on any given foraging night, we generate success as
S
/(random from normal distribution, m/65, s/33)
kcal kg
1
opossum, and each foraging night recalculate
energy intake as a function of success and current
opossum weight:
EI;iS(AW0Wi1)
I ran each winter data set 1000 times, with foraging
success selected randomly at each foraging bout within a
run, and recorded the percentage of runs in which the
opossum was expected to survive.
Model parameter sensitivity
The energetic estimates used to parameterize the model
do not include variability estimates. I preserved the
deterministic nature of the energetic parameters in the
normally distributed foraging success model and eval-
uated the model multiple times under differing values of
each parameter. This approach allowed a clear inter-
pretation of the magnitude of effect on survival per unit
change in the parameters. I referred to data available in
Brocke (1970) for estimation of appropriate value ranges.
His study included both males and females, but he did
734 ECOGRAPHY 28:6 (2005)
not indicate animal ages. I assume that animal weight
influences energetic parameters, whereas age and sex
do not.
Resting energy expenditure, foraging energy expendi-
ture, the percentage of body weight available for energy
conversion, and the critical foraging decision tempera-
ture were each varied in turn. Resting energy expenditure
varied in Brocke’s experimental animals between 1.44
and 2.36 kcal kg
1
h
1
per animal, which translates to
101/112 kcal d
1
in 6 opossums weighing between 1.82
and 2.92 kg, and 141 kcal d
1
for a larger 3.06 kg male
(Brocke 1970: Chapter 8, Table 9). In the model, up to
10% variation from the original 102 kcal estimate is
examined. Calculation from Brocke’s (1970) table of
energy requirements for higher activity periods (Chapter
8, Table 10) indicates a range of 129 /206 kcal d
1
in
foraging energy for 8 animals. I examined the normally
distributed foraging success model with foraging energy
varying up to 20% from the original 154 kcal estimate
(123.2/184.8 kcal). Four experimental animals moni-
tored by Brocke (1970) for winter weight loss survived
after 39.5 to 44.7% weight loss; interestingly the animal
losing only 39.5% of its initial weight was noted to have
almost died. A 5th animal did die after losing 42% of its
initial weight. In the model, I varied the percentage of
body weight available for energy consumption between
36 and 48% of the initial weight. Foraging occurred only
at or above 08C in the original model; I also examined
the model when the foraging decision temperature was
/28,/18,18, and 28C. In each of these scenarios the
number of foraging days was recorded.
The percentage of animals surviving the winter after
1000 runs of each scenario was compared to the null
model (i.e., the original normally distributed foraging
success model parameterized from Brocke’s averages).
For each scenario, years with 100% or 0% survival were
ignored, as the difference from the null model is curtailed
at the end of the probability distribution. The remaining
data sets of change in survival probability by variant
were analyzed in linear regression models using SAS
ver.8 (Anon. 1999).
Opossum autumn weight
The model has been evaluated with the beginning weight
of the average juvenile (2.59 kg, as estimated by Brocke
1970). However, as opossums are polygynous and
sexually dimorphic (with males larger than females),
the critical group for over-winter survival is the female
juveniles. From winter weights, Brocke back-calculated
that female juveniles average 2.37 kg on 1 December
(220 g lighter than the juvenile average). Therefore, I
evaluated the model using beginning weights between
2.09 and 3.09 kg (9
/500 g), including 2.37 kg. As above,
the model was run 1000 times for every beginning weight
and the percentage predicted to survive in each year was
recorded. The change in probability of survival as the
autumn weight differs from 2.59 kg was examined with
linear regression (Anon. 1999).
Verification of normally distributed foraging success
model
During the winters of 2000, 2001, and 2002, I radio-
collared opossums in the Amherst, Massachusetts,
region. I captured opossums in wire cage traps in and
around the Univ. of Massachusetts, Amherst campus.
Animals were placed in a cloth bag and restrained by
hand. Animals were weighed, sexed, aged via tooth
eruption (Gardner 1982) and total, tail, hind foot, and
ear lengths were measured. All female opossums received
radio-collars (50 g), as did males if equipment was
available. Radio-collared individuals were located via
radio-telemetry once every 24 /72 h. In warm weather,
animals were recaptured once a month to adjust the
collar fit and obtain the weight of the animal. The first
winter, I attempted to recapture animals (#3 and #6);
however, I immediately observed that I was actively
disrupting their foraging opportunities with my capture
attempts. Because I was concerned that such disruptions
throughout winter could significantly reduce animal
survival by reducing energy acquisition, I ceased my
capture attempts (subsequent modeling showed that the
few days I did potentially disrupt foraging were not
sufficient to have altered the fate of these animals). In
subsequent winters, animals were recaptured in warm
spells, or not until spring. Radio-collars were equipped
with mortality switches that altered the radio signal if
the collar had not moved in 8 h. When animals died,
every attempt was made to retrieve the body to
determine death weight, condition, and cause of death.
All capture and monitoring procedures were conducted
in accordance with Univ. of Massachusetts, Amherst,
IACUC protocol.
For the three years combined, 18 animals were
monitored, beginning in December. Two were killed
(roadkill and predation) before the end of the month,
and two disappeared before spring. For the 14 remaining
animals, both pre-winter weight (Table 2) and winter fate
were known.
I assume that the energetic process described in the
model is fundamentally the same regardless of animal
age or sex per se, but instead depends upon the autumn
weight of the animal (which is correlated with age and
sex). Breeding begins late in the winter, but it is unclear
whether males seek females (Ryser 1992) or vice versa
(Pippitt 1976), so energetic demands of finding mates
may not be sex biased. I therefore include all the animals
for which I have winter data. In this sample, I have
known survival outcomes of animals that began at
ECOGRAPHY 28:6 (2005) 735
particular autumn weights. I tested the model by
simulating the winter for opossums at these weights
and comparing model predictions with known fates of
the animals. The normally distributed foraging success
model was used with inputs to simulate each animal: the
animal’s pre-winter weight was used as beginning weight,
and the first day of the temperature data was the date on
which the pre-winter weight was taken. With 1000 runs
of the model, the probability of survival for the animal
was calculated. Either the date of expected starvation
was recorded, or the expected spring weight of the
animal, should it live. If an animal was caught and
weighed mid-winter, the expected weight was also
calculated from the model for the date in question.
To examine the effect of parameter variation on
predictions of the normally distributed foraging success
model, I altered the model with the high and low
extremes previously examined for four parameters in
the model (resting energy, foraging energy, percent
weight loss, and foraging decision temperature). This
resulted in 8 additional predictions of survival prob-
ability for each animal.
Results
Perfect foraging
Using the perfect foraging model, average-sized juvenile
opossums would be expected to survive every Amherst
winter from 1926 to 2002. The number of days in
December through March with maximum temperature
not exceeding freezing varied from 7 to 61 (Fig. 1).
Over the last 77 yr, the number of winter days with
maximum temperatures below 08C declined (days below
freezing
//0.2(yr)/439); however, only a small
amount of variation in number of days below freezing
was explained by year (r
2
/0.16). Juvenile opossums
modeled with ad libitum intake on foraging days not
only would be expected to survive all these Amherst
winters, but to end March fatter than they were in
December (Fig. 2).
Table 2. Autumn and winter weights of opossums in the Amherst, Massachusetts, area monitored in the winters of 2000, 2001, or
2002.
Opossum Sex Age Pre-winter Mid-winter
Date Weight (kg) Date Weight (kg)
3 F J 3 Dec 2000 2.45 2 Feb 2001 1.31
6 F J 6 Nov 2000 2.20 27 Jan 2001 1.45
22 F A 11 Oct 2001 2.60 11 Jan 2002 3.16
27 F J 15 Sept 2001 2.80 30 Jan 2002 3.35
28 F A 20 Sept 2001 2.05
63 M J 17 Dec 2001 3.15
74 M J 27 Oct 2002 4.10
80 F J 11 Nov 2002 2.25 23 Dec 2002 2.00
82 M J 20 Dec 2002 2.05
86 F J 16 Oct 2002 1.13
88 F J 5 Nov 2002 2.35
89 F J 9 Nov 2002 2.75
92 M J 20 Dec 2002 2.85
100 M J 20 Dec 2002 2.80
0
10
20
30
40
50
60
70
1925 1945 1965 1985 2005
Year
Number of days with maximum ≤0°C
Fig. 1. The number of days between December and March with
maximum temperature 5
/08C recorded at the local weather
station for Amherst, Massachusetts, 1926 /2002.
Fig. 2. Model output for perfect foraging success by juvenile
opossums, for the winter of 1990 /1991.
736 ECOGRAPHY 28:6 (2005)
Normally distributed foraging success
The survival prospects of juvenile opossums having
random foraging success were considerably lower than
the survival of opossums that ate ad libitum (Fig. 3a). In
nearly half the years (35 of 77), B
/25% of average-sized
juveniles would be expected to survive. For 17 of these
years, the death toll from starvation was projected to be
100%. The probability of survival increased over the years
(survival probability
/0.64(yr)/1202), although year
explained only a small amount of the variation in survival
(r
2
/0.14). There are still years within the last decade with
climatic regimes severe enough to predict no or low
survival of opossums beginning winter at 5
/2.59 kg.
Model parameter sensitivity
Variation in each of the four parameters (resting energy
expenditure, foraging energy expenditure, percentage of
body weight available for consumption, and foraging
decision temperature) resulted in linear changes in
average survival probability, as expected from the model
structure. For any given year the effect of altered model
parameters on survival probability would be difficult to
predict, because the size of the effect was very different
for different years. However, for each parameter the
mean change in survival probability was strongly linear.
For every kcal increase in resting energy expenditure,
there was a mean reduction of 1.06% in survival (change
in percent survival
//1.06(resting energy)/107.97;
r
2
/0.70). Foraging energy expenditure had a larger
effect, with an average decrease of 2.67% in survival for
each additional kcal expended (change in percent
survival
//2.67(foraging energy)/406.20; r
2
/0.89).
The larger the percentage of the body weight that could
be utilized as energy, the higher the probability of
opossum survival (change in percent survival
/1.71(per-
cent weight loss)
/72.31; r
2
/0.73). A shift in the
foraging decision temperature had a large effect on
Fig. 3. The over-winter percent survival
of (a) juvenile (2.59 kg) opossums and
(b) female juvenile (2.37 kg) opossums
predicted by a normally distributed
random foraging success model under
Amherst, Massachusetts temperature
regimes. ]
/67% survival of the female
juveniles is estimated as necessary for
opossum population persistance.
(a)
(b)
0
10
20
30
40
50
60
70
80
90
100
1925 1945 1965 1985 2005
Percent survival
0
10
20
30
40
50
60
70
80
90
100
1925 1945 1965 1985 2005
Year
Percent survival
ECOGRAPHY 28:6 (2005) 737
survival probability; a change in minimum foraging
temperature of 18C yielded a change in survival prob-
ability of 21.5% (change in percent survival
/
/21.52(decision temperature)/1.5; r
2
/0.72). The
number of non-foraging days in a given year is deter-
mined by the minimum foraging temperature and the
winter temperature regime. Because the order of non-
foraging days is important, and is unique to each year,
the predicted survival probability for a particular
number of non-foraging days can vary widely (Fig. 4).
For example, foraging temperature-winter regime com-
binations resulting in 28 non-foraging days had survival
probabilities that ranged from 6 to 93%.
Opossum autumn weight
An opossum’s initial weight had a considerable effect on
its survival probability (change in percent survival
/
200.8 (opossum kg)/525.6; r
2
/0.91) (Fig. 5). A 25-g
change in the opossum weight changed the survival
probability by 5%.
If an average female juvenile begins winter at 2.37 kg,
the model predicted a low probability of survival in most
years (Fig. 3b). Fewer than 25% of females were
expected to survive the winter in 64 of the 77 yr. For
35 of these 64 yr the model predicted that no
2.37 kg female could survive the winter. Because of
limited reproduction (Hossler et al. 1994) and high
mortality the rest of the year (Wright 1989, Sunquist
and Eisenberg 1993, Hossler et al. 1994), Kanda and
Fuller (2004) estimated that at least 67% of over-
wintering juvenile females need to survive in order to
maintain a stable population. In contrast, the model
shows only four years (1974, 1986, 1997, and 2001) in
which at least 67% of the 2.37 kg females would be
expected to survive.
Model verification
Of the 14 live opossums for which I had over-winter
data, the fates of 13 of the animals were correctly
predicted by the normally distributed foraging success
model (Table 3). In winter 2000, two juvenile females (#3
and #6) that were predicted to starve in early February
did so, and their weights at starvation were within 110 g
of the weight expected after a 42% reduction from their
autumn weight (Fig. 6).
The winter of 2001 was particularly warm, and none
of the three monitored animals died of starvation. The
model correctly predicted survival for two of the three
females (Table 3). Of the two females correctly predicted
to survive, the mid-winter and spring weights of one
female were correctly predicted (#22), while the other
female’s weights were greatly overestimated in the model
(#27; 95% CI 3.57 /7.29 kg midwinter instead of 3.35 kg,
and 4.81/13.68 kg in spring instead of 3.30 kg) (Fig. 7).
The third female (#28) was weighed early in the autumn
and never recaptured, though she survived the winter.
The model predicted that she should not survive the
winter.
0
10
20
30
40
50
60
70
80
90
100
0 10203040506070
Non-foraging days in winter
Survival probability
Fig. 4. The survival probability for a 2.59 kg opossum under
different numbers of non-foraging days (produced through
combinations of years and foraging decision temperatures).
The range of survival probabilities possible under 28 non-
foraging days is highlighted.
Fig. 5. Linear regression with 95%
confidence intervals of change in
survival probability as input weight
is altered from the null weight of
2.59 kg.
738 ECOGRAPHY 28:6 (2005)
In 2003, all four smaller animals (females #80, #86, and
#88, and male #82) had high model probabilities of
starvation, and all died (Table 3). Animals #88 and #82
died in late January and late February, respectively, as pre-
dicted. Animal #82 was able to deplete his body stores
by 140 g more than anticipated by the model before he
died (Fig. 8). Animal #80 died a few days earlier than
expected by the model, and although death weight was
not obtained, her mid-winter weight was correctly pre-
dicted. Like #28, #86 was weighed early in the autumn.
While she did die as predicted, she did not do so until Feb-
ruary rather than November, and at death she weighed
considerably more than the predicted 0.66 kg (Fig. 8).
The larger animals monitored in 2003 (males #63,
#74, #92, #100, and female #89) all survived the winter,
as expected (Table 3). The model suggested large weight
gains for #74, inconsistent with an eventual weight in
late April of 2.80 kg. Animal #63 was killed at the end of
March and weighed 2.05 kg, slightly lower than the
model’s estimates of his spring weight. Spring weights of
both #89 (Fig. 8) and #100 were consistent with the
model.
Predictions under parameter variability
No model with altered parameter estimates consistently
predicted actual animal fates better than the null model
(Table 4a). The one animal for which the null model
failed to correctly predict its fate (#28) was correctly
predicted to live only by a model with low foraging
energy expenditure. Across all 14 animals, however, the
low-foraging-energy variant correctly predicted the fates
of only 10 of the animals. Alteration of the model to a
high-foraging-energy variant or a 28C change in either
direction in the foraging decision temperature also
resulted in models with reduced predictive capability
for these animals. Resting foraging energy or acceptable
percentage of weight lost could both be altered from the
null model without influencing the model’s fate predic-
tions. For animals that died, the null model correctly
predicted the date of death in 4 of the 6 cases. Of the
7 animals that lived and were weighed in spring, the null
model correctly predicted 3 weights. No model variant
increased the accuracy of predicting the death dates of
animals that starved or live weight of animals that
survived (Table 4b).
Discussion
While Amherst, MA, winters within the last three-
quarters of a century are increasingly amenable to
survival of small opossums, the model developed here
suggests that most Amherst winters are still too cold to
support a stable opossum population. The first records
of opossums in Massachusetts date from the early 1920s
(Gardner and Sunquist 2003); however, weather records
for the late 1920s would predict low juvenile opossum
survival (2.59 kg; B
/25%) and no survival of females
averaging 2.37 kg. The large number of years in the
1970/2002 interval that should have permitted survival
of young (2.59 kg) opossums cannot explain the previous
establishment of opossums in the area.
Sensitivity of the normally distributed foraging success
model to variations in the parameterization of the model
depended greatly on the year. Averaged across years,
the model was not very sensitive to changes in resting
energy or acceptable percentage weight loss. The model
results change more dramatically with foraging energy.
Table 3. Comparison of predicted and actual opossum fates. Predicted starvation weight is a deterministic 42% of the animal’s
autumn weight.
Opossum Fate Death date End weight (kg)
Predicted
survival
probability
Actual Predicted
(95% CI)
Actual Predicted
(95% CI)
Actual
a
%D
3 0.00 Starve 21 Feb (1 Feb/11 Mar) 7 Feb 1.42 1.31 /7.7
6 0.00 Starve 24 Jan (6 Jan /10 Feb) 5 Feb 1.28 1.23 /3.9
22 0.99 Lived
b
4.59 (2.06 /7.12) 2.75 /40.1
27 1.00 Lived 9.26 (4.84 /13.68) 3.30 /64.4
28 0.00 Lived 20 Dec (16 Nov /22 Jan) 1.19 n/a
63 1.00 Lived
b
3.65 (2.65 /4.65) 2.05 /43.8
74 1.00 Lived 10.35 (7.94 /12.76) 2.8 /72.9
80 0.00 Starve 13 Feb (22 Jan /6 Mar) 17 Jan 1.30 n/a
82 0.00 Starve 12 Mar (19 Feb/2 Apr) 25 Feb 1.19 1.05 /11.8
86 0.00 Starve 8 Nov (4 Nov/11 Nov) 28 Feb 0.66 1.02 54.5
88 0.06 Starve 23 Feb (26 Jan /22 Mar) 27 Jan 1.36 n/a
89 0.93 Lived 2.45 (1.64 /3.26) 1.75 /28.6
92 0.98 Lived 2.18 (1.67 /2.69) n/a
100 0.94 Lived 2.05 (1.60 /2.50) 2.25 9.8
a
#28 and #92 lived but were not recaptured in the spring; the bodies of #80 and #88 were not retrievable but circumstances
indicated death by starvation.
b
#22 and #63 died on 21 Feb and 22 Mar, respectively, from non-winter related causes.
ECOGRAPHY 28:6 (2005) 739
However, foraging energy measurements were originally
taken on captive animals in pens 2.4
/2.4 m, with food
at specific feeding platforms within the pens (Brocke
1970). The foraging energy expenditures measured there-
fore probably underestimate the energy expended by a
wild opossum that will have to travel longer in cold
ambient temperatures in search of food in unpredictable
locations. Because increased foraging expenditure corre-
lates to decreased survival probability, inaccuracy in the
model representation of foraging energy is most likely to
translate to over-optimistic survival expectations.
The temperature at which the foraging decision is
made can have a considerable effect upon opossum
survival, depending upon how the change in decision
temperature affects the number of foraging/non-foraging
days. The total number of non-foraging days has a very
large effect on opossum survival. A 2.59 kg opossum can
have no more than 52 non-foraging days in order to
survive over winter. With no more than 14 non-foraging
days survival probability is expected to be 100%. When
there are 14/52 non-foraging days, as is common in
Amherst, MA, the order in which these non-foraging
days occur may be very important, as documented by
the wide variety in expected survival probabilities for the
same number of non-foraging days calculated from
different year-foraging decision combinations.
The most important influence on the model is not the
potential variation in the energetic parameters of the
model but the input of opossum weight. The size
attained by the opossum by 1 December has a large
effect on the probability of survival through the winter.
Larger animals will have larger fat stores to draw upon
and will be capable of taking and consuming larger prey
(when available). Our sample included a number of such
animals; however most of the larger animals were males.
Fig. 6. Model output for simulation of animals #3 and #6, and
the maximum daily temperatures recorded in the winter of 2000.
For clarity, only 10 of the 1000 model runs for each animal is
illustrated. * denotes measured animal weights, with input
autumn weights provided in parantheses on the abscissa.
Predicted weights at starvation (reserve at 0 kg) are also given.
Fig. 7. Model output for simulation of animal #27 and the
maximum daily temperatures recorded in the winter of 2001.
For clarity, only 10 of the 1000 model runs for each animal is
illustrated. * denotes measured animal weights, with input
autumn weight provided in parantheses on the abscissa.
740 ECOGRAPHY 28:6 (2005)
The trouble for northern opossum populations is that
the important class that must survive the winters are
the juvenile females, which are the smallest individuals.
The model predicted
/67% survival of average-sized
(2.37 kg) females in only 4 of our 77 yr of time series,
suggesting that populations should decline in most
modeled years (Kanda and Fuller 2004). Of course,
females in the real population have an unknown variance
in size. However, if the female weights are normally
distributed, and the mean females are not predicted to
survive, then the 50% of the population that is smaller
than average would also not be expected to survive. If
stable demographics require ]
/67% survival, then those
years in which survival probability of the average female
is low should also be characterized by low population
persistence.
Replacement of lost animals via immigration is
unlikely. While female opossums are known to disperse
in the spring (with or without babies already in the
pouch), most opossum dispersal distances are B
/7km
(Gillette 1980), and the climate 10 km south of Amherst
is not sufficiently different from Amherst to be hosting a
source population. Further, immigration clearly cannot
explain the recent occurrence of opossums 250 km north
of Amherst (Gardner and Sunquist 2003).
The obvious discrepancy between these model predic-
tions and the fact that opossum populations remain
robust in and north of Amherst, MA, immediately calls
into question the normally distributed foraging success
model. However, comparison of the model with actual
opossums followed in the Amherst area shows the strong
predictive accuracy of the model. Given opossum
weights close to the beginning of winter (November or
December), the model correctly predicts the opossum’s
fate, including weight and date of death or the weight of
surviving animals.
Exceptions were animal #80, who died earlier than
anticipated by the model, and animal #63, who lived as
expected but weighed less than predicted. Animal #80’s
death was probably accelerated by a large snowstorm on
26 December 2002 that apparently blocked the animal’s
den entrance. There was no evidence of foraging activity
by #80 after the storm, and a mortality signal was
received from the den three weeks later. Animal #63
spent at least four days not foraging before his death on
22 April from injuries, however at 23 g expended per
resting day this accounts for only 92 g of the discrepancy
between the actual weight (2.05 kg) and expected weight
(2.65/4.65 kg).
Fig. 8. Model output for simulation of animals #86, #82, and
#89, and the maximum daily temperatures recorded in the
winter of 2002. For clarity, only 10 of the 1000 model runs for
each animal is illustrated. * denotes measured animal weights,
with input autumn weights provided in parantheses on the
abscissa. Predicted weights at starvation (reserve at 0 kg) are
also given for animals predicted to die.
ECOGRAPHY 28:6 (2005) 741
The model was less accurate when knowledge of the
pre-winter opossum condition was obtained earlier in
the autumn. In particular, juvenile animal #86 was
predicted to die of starvation before winter even began.
The model is designed to simulate winter foraging
conditions, and does not consider the growth of young
animals nor the foraging conditions available in the
autumn. Because the total foraging energy gained in
the model is dependant upon the size of the animal, the
model limits energy intake of small animals during this
pre-winter period. Instead of the autumn weight gain
normally seen, the model predicts decreasing weight and
starvation before winter despite foraging every night.
Male #82 was weighed six days after #86, his sibling,
and was 1.25 kg at the time (very similar to #86’s
1.13 kg). However, by mid-December, #82 weighed
2.05 kg. I expect that #86 gained weight in a similar
manner. This inability of the model to correctly simulate
autumn conditions also accounts for the inaccurate
predictions for the small adult #28.
The same principal model foraging success assump-
tion may also account for the extraordinary and
unrealistic weight gains predicted for two animals that
lived comfortably through their respective winters (#27
and #74). Because these animals had higher pre-winter
weights, the energy gained during each foraging bout
was proportionally higher. Obviously there is a natural
limit to the amount of energy gained in a bout (a
limitation either of the animal’s physical capacity or the
amount of food located), but the model does not impose
such a limit. Thus as the model animal gained weight,
the potential amount of food it gained also increased
without limit, and large animals rapidly increased in
weight.
Examination of predictions generated by variation of
the parameters confirms that the null model was the
most accurate. Reasonable variations in resting energy or
percentage of weight loss do not alter the model.
Changes in foraging energy expectations do have a
major impact upon the model’s predictive ability, as do
changes in foraging decision temperature; interestingly,
variations of these parameters tend to have lower
predictive ability than the null model. I did not allow
resting and foraging energy expenditures to vary with
opossum weight, although Brocke (1970) indicated that
heavier animals had higher expenditures; however, the
model variants with higher expenditures did not have
higher predictive power for the larger animals, and the
model with higher foraging energy expenditures had
decreased predictive power overall.
The reduced performance of the model with higher
foraging energy expenditure is particularly curious, as I
expect that wild opossum foraging expenditures tend to
be higher than the average measured in captivity.
Examination of models with differing foraging energy
expenditure was equivalent to examination of models
with constant foraging energy expenditure but differing
foraging success, since the energy gained and energy
Table 4. Accuracy of normally distributed foraging success model variants in predicting (a) animal fates, and (b) live animal spring
weights or starved animal death dates. âindicates accurate prediction.
(a)
3
(die)
6
(die)
22
(live)
27
(live)
28
(live)
63
(live)
74
(live)
80
(die)
82
(die)
86
(die)
88
(die)
89
(live)
92
(live)
100
(live)
Null âââ â â â ââââ â ââ
36% loss âââ â â â ââââ â ââ
48% loss âââ â â â ââââ â ââ
Resting 92 kcal âââ â â â ââââ â ââ
Resting 112 kcal âââ â â â ââââ â ââ
Foraging 123 kcal âââââ ââ âââ
Foraging 185 kcal ââ â â â ââââ
Decision
/28Cââ â â ââââââ ââ
Decision 28Câââ â â â ââââ
(b)
Live weight Death date
22 27 28 63 74 89 100 3 6 80 82 86 88
Null âââââââ
36% loss ââââââ
48% loss ââââ
Resting 92 kcal ââââââ
Resting 112 kcal âââââââ
Foraging 123 kcal ââ
Foraging 185 kcal ââââ
Decision
/28Cââââ
Decision 28Cââ ââââ
742 ECOGRAPHY 28:6 (2005)
expended on a foraging day are subtracted from one
another in the model. It is possible that the combination
of 50% ad libitum foraging success and the mean
measured foraging expenditure (i.e. the null model)
resulted in a foraging-day energy gain similar to a higher
foraging expenditure but also a higher mean intake in
the actual animals.
The accuracy of the model in predicting actual
opossum fates (when considering December through
March) supports the validity of the general model
predictions of poor survival for juveniles, and particu-
larly for smaller female juveniles. If this model does in
fact accurately represent the fates of opossums experien-
cing Amherst winters, then why do opossums exist in
Amherst? The only two input variables are opossum
autumn weight and winter temperatures. The winter
temperatures reflected by the weather station may not
represent the microclimates experienced by individual
opossums throughout the local area. The other way in
which the climate is represented in the model is through
changes in the foraging decision temperature. While
changing the foraging decision temperature rule did not
increase predictive ability of the model for this sample of
monitored animals, the influence of the decision rule
varies greatly by year. In some years, the particular
pattern of cold weather combined with a shift in the
foraging decision temperature could greatly affect an
opossum’s survival probability.
The model suggests that the most important influence
on opossum winter survival is the autumn weight of the
individual. It is possible that our small sample of females
does not accurately reflect the juvenile female popula-
tion, and larger individuals such as #27 and #89 may
better represent the population. If an animal has access
to an abundant or high quality food source, it will enter
winter in better condition and be better able to maintain
itself over the course of the winter. Monitored females
#27 and #89 suggest how this may occur: both animals
lived in urban environments and used human-related
resources, such as dining hall dumpsters, trash bins,
and areas where humans deliberately left out food for
wildlife.
The model supports the previous expectations that
winter temperatures are a limiting factor on the northern
range limit of the Virginia opossum. However, the
opossum populations have expanded north beyond
where regional winter temperatures alone would predict
species limitation. Amherst, MA, serves as an example
where winter temperatures should restrict opossum
population persistence, yet the opossum population is
well established in the urban areas. Despite it being one
of the best-understood mammalian species distributional
limits, closer examination suggests that mechanisms
underlying the northern limit of Virginia opossums
remain enigmatic. The model also suggests three major
areas of inadequate understanding: the microclimates
actually experienced by opossums, the exact relationship
of opossum foraging behavior with ambient temperature,
and most importantly the role of additional factors, such
as human-related resources, operating to mitigate the
restrictive climatic effects. These factors can only be
understood in greater detail through close monitoring of
the behavior of individual opossums as they face north-
ern winter conditions.
Acknowledgements /T. K. Fuller and P. R. Sievert supported
both the fieldwork and manuscript. E. M. Jakob and J. Podos
also provided valuable comments. J. T. Finn encouraged the
model programming. The manuscript was greatly improved by
suggestions from D. A. Kelt. The field work was supported by
the USGS Massachusetts Cooperative Fish and Wildlife
Research Unit, by a Grant-in-Aid of Research from Sigma Xi,
the Scientific Research Society, by Max McGraw Wildlife
Foundation, by a David J. Klingener Scholarship from the
Massachusetts Museum of Natural History, and by the
Cooperative State Research Extension, Education Service,
U.S. Dept of Agriculture, Massachusetts Agricultural
Experiment Station, under Project No. MAS0071. This is
Massachusetts Agricultural Experiment Station Publication
No. 3357.
References
Anon. 1997. Matlab student edition, ver. 5. /Mathworks,
Natick, MA, USA.
Anon. 1999. SAS, ver. 8. /SAS Inst., Cary, NC, USA.
Blair, W. F. 1936. An opossum dies of cold and hunger. /J.
Mammal. 17: 410.
Brocke, R. H. 1970. The winter ecology and bioenergetics of the
opossum, Didelphis marsupialis, as distributional factors in
Michigan. /Ph.D. thesis, Michigan State Univ.
de Vos, A. 1964. Range changes of mammals in the Great Lakes
region. /Am. Midl. Nat. 71: 210/231.
Gardner, A. L. 1982. Virginia opossum. /In: Chapman, J. A.
and Feldhamer, G. A. (eds), Wild mammals of North
America: biology, management, and economics. Johns
Hopkins Univ. Press, pp. 3 /36.
Gardner, A. L. and Sunquist, M. E. 2003. Opossum. /In:
Feldhamer, G. A., Thompson, B. C. and Chapman, J. A.
(eds), Wild mammals of North America: biology, manage-
ment, and conservation. Johns Hopkins Univ. Press, pp. 3 /
29.
Gillette, L. N. 1980. Movement patterns of radio-tagged
opossums in Wisconsin. /Am. Midl. Nat. 104: 1 /12.
Hossler, R. J., McAninch, J. B. and Harder, J. D. 1994. Maternal
denning behavior and survival of juveniles in opossums in
southeastern New York. /J. Mammal. 75: 60 /70.
Hsu, M., Harder, J. D. and Lustick, S. I. 1988. Seasonal
energetics of opossums (Didelphis virginiana ) in Ohio.
/Comp. Biochem. Physiol. 90A: 441 /443.
Humphries, M. M., Thomas, D. W. and Speakman, J. R. 2002.
Climate-mediated energetic constraints on the distribution
of hibernating mammals. /Nature 418: 313 /316.
Kanda, L. L. 2005. Factors influencing survival and reproduc-
tion of Virginia opossums (Didelphis virginiana) at their
northern distributional limit. /Ph.D. thesis, Univ. Massa-
chusetts Amherst.
Kanda, L. L. and Fuller, T. K. 2004. Demographic responses of
Virginia opossums to limitation at their northern boundary.
/Can. J. Zool. 82: 1126 /1134.
Long, C. A. and Copes, F. A. 1968. Note on the rate of
dispersion of the opossum in Wisconsin. /Am. Midl. Nat.
80: 283 /284.
ECOGRAPHY 28:6 (2005) 743
McNab, B. K. 2002. The physiological ecology of vertebrates.
/Cornell Univ. Press.
Peterson, R. L. and Downing, S. C. 1956. Distributional records
of the opossum in Ontario. /J. Mammal. 37: 431/435.
Pippitt, D. D. 1976. A radiotelemetric study of the winter
energetics of the opossum Didelphis virginiana Kerr. /Ph.D.
thesis, Univ. Kansas.
Ryser, J. 1992. The mating system and male mating success
of the Virginia opossum (Didelphis virginiana) in Florida.
/J. Zool. 228: 127 /139.
Rosenzweig, M. L. 1995. Species diversity in space and time.
/Cambridge Univ. Press.
Seidensticker, J., O’Connell, M. A. and Johnsingh, A. J. T. 1987.
Virginia opossum. /In: Novak, M. et al. (eds), Wild
furbearer management and conservation in North America.
Ontario Trappers Assoc., pp. 247 /261.
Sunquist, M. E. and Eisenberg, J. F. 1993. Reproductive
strategies of female Didelphis./Bull. Florida Mus. Nat.
His. 36: 109 /140.
Tyndale-Biscoe, H. 1973. Life of marsupials. /Edward Arnold.
Wallace, A. R. 1876. The geographical distribution of animals.
/Harper.
Wiseman, G. L. and Hendrickson, G. O. 1950. Notes on the life
history and ecology of the opossum in southeast Iowa. /J.
Mammal. 31: 331 /337.
Wright, D. D. 1989. Mortality and dispersal of juvenile
opossums, Didelphis virginiana./M.S. thesis, Univ.
Florida.
Subject Editor: Douglas Kelt.
744 ECOGRAPHY 28:6 (2005)
