Professional Documents
Culture Documents
BioOne Complete (complete.BioOne.org) is a full-text database of 200 subscribed and open-access titles
in the biological, ecological, and environmental sciences published by nonprofit societies, associations,
museums, institutions, and presses.
Your use of this PDF, the BioOne Complete website, and all posted and associated content indicates your
acceptance of BioOne’s Terms of Use, available at www.bioone.org/terms-of-use.
Usage of BioOne Complete content is strictly limited to personal, educational, and non - commercial use.
Commercial inquiries or rights and permissions requests should be directed to the individual publisher as
copyright holder.
BioOne sees sustainable scholarly publishing as an inherently collaborative enterprise connecting authors, nonprofit
publishers, academic institutions, research libraries, and research funders in the common goal of maximizing access to
critical research.
Gundersen, H. & Andreassen, H.P. 1998: The risk o f m oose Alces alces col
lision: A predictive logistic model for moose-train accidents. - Wildl. Biol.
4: 103-110.
We used logistic m odels to estim ate the risk of m oose-train collisions for the
Rprosbanen railway in Norway. During 1990-1997, a total of 13,506 train
departures w ere registered along R prosbanen during the m onths w hen the
risk o f collision was highest (Decem ber to March). The statistical model
selected to predict the risk o f moose-train collisions included train route,
tim e o f day, lunar phase and average train speed, as w ell as two climatic
covariables, i.e. snow depth and temperature. Trains running at night, in the
m orning or in the evening experienced a higher risk o f collision with moose
A lces alces than day trains. The probability of collision was also higher dur
ing nights of full m oons than during nights o f h alf or no m oons. As observed
previously with trains in Norw ay m oose-kills increased with increasing
snow depth and decreasing tem peratures. To test the predictability o f the
model, we used a logistic model based on train departures during 1990-1996
to predict the num ber of m oose-train accidents during w inter 1996/97.
Although the model had a satisfactorily high predictability, the best models
would probably be those based on a com bination o f both tem poral and spa
tial aspects. We discuss how logistic m odels may be applied to introduce
rem edial actions on high-risk routes or during high-risk periods.
Key words: A lces alces, gam e-vehicle accidents, logistic m odel, railway
Table 2. The logit model chosen to best explain the risk of moose-train collisions. (AIC = 3228.2; AIC without train speed = 3228.7; AIC
for other models > 3230.6).
Predictor Estimate df xJ P
o 0.03
in the m o d el (h ereafter te rm ed ro u te factor). T he NH
C/D
0.02 '
m o st p arsim o n io u s m o d el ac co rd in g to A k a ik e ’s 3
In fo rm a tio n C rite rio n (A IC v alu es; B u rn h am & -4 0.01 '
1
A n d erso n (1993)) w as ch o sen to p red ic t the risk o f o 0 '
u N ight M o rn in g D ay E vening
m o o se -tra in collisions. z T IM E O F DAY
B
Results 0.04
w
C/D
O u t o f a to tal o f 13,506 train d ep artu res alo n g the O
O 0.03
R p ro sb a n en railw ay in the perio d s an alysed, there
w ere 4 0 6 m o o se -tra in collisio n s, w hich k illed 466
fa
m o o se. P re lim in ary analy ses sh o w ed th a t th e p ro b a o 0.02
bility o f co llid in g w ith a m o o se w as 0 .030 (95% C l = F u ll H a lf New
H
0.0 2 8 , 0 .0 3 4 ), b u t it v arie d co n sid erab ly b etw een L U N A R PH A S E
ro u tes (u n iv ariate m odel: x \ , i 3,498 = 174.73, P < 0.001) 0.1 0 1
CO
(Table 1). T he ro u te sp ecific risk o f k illin g a m oose
2
s
co u ld m o st prob ab ly b e ascrib ed b oth to v aria tio n s in
0.05 -
th e d istan ce o f the ro u te (u n iv ariate m odel: x \ 13,504 =
153.56, P < 0 .001), alth o u g h the slo p e o f the linear fa
m o d el w as ra th e r flat (d istan ce in km , estim a te =
0 .0 1 0 (S E = 0 .0 0 1 )), an d spatial v ariatio n s in co lli
sion risk along th e local areas trav e rsed by the d iffe r A V E R A G E T R A IN SP E E D
en t ro utes (e.g. due to d ifferen t fo o d av ailab ility and (k m /h )
to p o g rap h y as show n by G u n d ersen et al. (in press)).
To attain co n v erg en ce fo r a global m o d el (w ith no Figure 2. Probability of moose-train collisions in relation to A) the
time of day, B) lunar phase and C) average train speed (see statis
in teractio n s) w hen including th e ro u te factor, w e had
tics in Table 2).
to ex c lu d e th ree ro u tes w ith o u t any train kills, c o n
sistin g o f 190 d ep artu res (see T able 1). T he m odel
sh o w ed n o sig n ific an t asso ciatio n b etw e en m oose- m o o se-k ills an d av erag e train sp eed ( x \ I3,303 = 2.53,
k ills and train ty p e ( x \ ,3,302 = 0.02, P = 0 .857), or P = 0.111). T h e m o st p arsim o n io u s m o d el, w ith only
Table 3. The logit model used to predict the number of train-killed moose for trains running in the morning, day, evening and at night the
whole distance between Elverum and Rpros in winter 1996/97. (AIC = 1752.2; AIC without train speed = 1752.3; AIC for other models
> 1754.0).
Predictor Estimate df x2 P
No of accidents
Train departure No o f departures Observed Predicted G P
statistically significant term s, included route, tim e o f the night, m orning o r ev en in g than during the d ay
day, lu n ar phase and the covariables snow depth and tim e (Fig. 2A ), and 1.3 tim es h ig h er during periods
tem p erature (Table 2). H ow ever, according to A IC , w ith full m o o n s than during p erio d s w ith new o r h alf
this m odel w as in distinguishable from a m odel w hich m oons (see Fig. 2B). T h e pro b ab ility o f m oo se-train
also included average train speed (see Table 2). co llisio n s in creased w ith in creasin g train speed,
T h e risk o f collision w as 5-6.8 tim es higher during although n ot significantly so (see Fig. 2C). F or in
stance, an increase in train speed
from 50 to 100 k m /h o u r d oubled the
LU N A R PH A SE risk o f collision. T h e p ro b ab ility o f
H a lf/N e w
m oose-train co llisions also increased
w ith increasing snow depth and d e
creasin g tem p eratu re (see T able 2 for
estim ates, and G und ersen et al. in
press).
Train specification
T h e above analyses included all train
departures in 0 ste rd a le n . To depict
the risk o f different departures, w e
m ade a m o d el containing only p a s
senger trains running the w hole d is
tance betw een E lverum and R pros
(there are tw o trains (one going south
and one g oing north) in the m orning,
day, ev en in g and night). W e included
snow depth, daily average tem p era
tu re , lu n a r p h ase (fu ll m o o n vs
half/new m oon), train speed and train
(i.e. tim e o f day) in the m odel. To test
the p redictability o f the m odel w e
ex clu d ed the w inter o f 1996/97 and
used the m odel b ased on the data
from 1990 to 1996 to p red ict the
n u m b e r o f train -k illed m o o se fo r
-30j.V ^ each train (m orning, day, even in g and
-20 n ig h t trains) during w in ter 1996/97.
ofS
<0 * F ro m 1990 to 1996 th ere w ere
<c>
5,371 departures o f the train s in clu d
ed in our analyses, and 229 m oose
w ere killed. T he best m o d el acco rd
Figure 3. Probability of moose-train collisions of passenger trains running the whole dis
tance between Elverum and Rpros in the morning, day, evening and at night, in relation ing to the A IC criteria in clu d ed all
to lunar phase, snow depth and temperature (see statistics in Table 3). p red icto r variables (Table 3). In w in
108 © W IL D L IF E B IO LO G Y ■ 4 :2 (1998)
110 © W IL D L IF E B IO LO G Y ■ 4 :2 (1998)