1. introduction n = f(n ,nt-2,nt-3
TRANSCRIPT
1
Year
Pop
ulat
ion
inde
x
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
1810 1870 1930 1990
Time Series Analysis
Nt = f(Nt-1,Nt-2,Nt-3)1. Introduction
3. Global wader dynamics
2. Sarcoptic mange-fox dynamics
Year
Pop
ulat
ion
inde
x
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
1810 1870 1930 1990
Time Series Analysis 1. Introduction
-1
1
3
5
7
1820 1860 1900 1940 1980
Time Series Analysis (TSA)
1. Introduction
2
4
6
8
1965 1970 1975 1980 1985
grow
th ind
ex
2
4
6
8
1965 1970 1975 1980 1985
grow
th ind
ex Cassiope tetragonaCassiope tetragona
40
80
120
160
20 30 40 50 60 70 80
year
Julia
n fl
ower
dat
e TussilagoTussilago
Population Time Series
Time Series Analysis (TSA)
1. Introduction
-1
1
3
5
7
1820 1860 1900 1940 1980
Popu
latio
n si
ze (
stdz
)
year
• We have seen and expect changes in ecology parallel to changes in climate • Interesting as this may be, we need to go further - to go behind the patterns to expose the processes …
… direct, indirect, multi-trophic, cascading, feedback dynamics
temporally
spat
ially
16 pops
Time Series Analysis (TSA)
1. Introduction
• Time series analysis (TSA) is a pure statistical tool designed to disentangle the autocovariate patterns in time series
• We have seen and expect changes in ecology parallel to changes in climate • Interesting as this may be, we need to go further - to go behind the patterns to expose the processes …
… direct, indirect, multi-trophic, cascading, feedback dynamics
Time Series Analysis (TSA)
Analysis of the lynx 10-year cycle
• Boreal forest the Arctic Ocean
• Boreal forest to the Arctic Ocean • Food: berry (summer), birch, willow (winter)
• Food: snowshoe hare, squirrel, grouse
• Predators: lynx, fox, coyote, owl
Lepus americanus Lynx canadensis
2
Time Series Analysis (TSA)
An example: the lynx 10-year cycle
Distinct 10-year cycle (harvest data)
Processer?: obscure!
Sun spots lynx
Sun spots
+
EXTRINSIC (DID): Lunar cycles (moonlight quality), Weather, Forest-fire (plant) INTRINSIC (DD): Hare population, Plant-Hare, Lynx-Hare
Time Series Analysis (TSA)
Distinct 10-year cycle (harvest data)
Processer?: obscure!
Sun spots lynx
Sun spots
EXTRINSIC (DID): Lunar cycles (moonlight quality), Weather, Forest-fire (plant) INTRINSIC (DD): Hare population, Plant-Hare, Lynx-Hare
-
An example: the lynx 10-year cycle
Time Series Analysis (TSA)
Distinct 10-year cycle (harvest data)
Processer?: obscure!
Sun spots lynx
Sun spots
EXTRINSIC (DID): Lunar cycles (moonlight quality), Weather, Forest-fire (plant) INTRINSIC (DD): Hare population, Plant-Hare, Lynx-Hare
+
Pure correlation analyses not good!
An example: the lynx 10-year cycle
Time Series Analysis (TSA)
What to do?
An example: the lynx 10-year cycle
Nt = f(Nt-1,Nt-2,..., Nt-11)!...
Kluane indicates that hare-predator interactions are central.
lynx
hare
year
dens
ity
f(Nt-1,Nt-2) decrease Nt =
f(Nt-1,Nt-2) increase
… dynamics non-linear!
High dependence (80%) on hare density ...
An example: the lynx 10-year cycle
Time Series Analysis (TSA) Time Series Analysis (TSA)
1. Introduction
1935 1945 1955 1965year
abun
danc
e
An example: grouse population dynamics
3
Time Series Analysis (TSA)
1. Introduction
1935 1945 1955 1965year
abun
danc
e te
mpe
ratu
re
rtemp,grouse = 0.90
Does temperature explain 81%?
An example: grouse population dynamics
Time Series Analysis (TSA)
1. Introduction
1935 1945 1955 1965year
abun
danc
e te
mpe
ratu
re
rtemp,grouse = 0.01
at,t-1 = 0.46
An example: grouse population dynamics
Time Series Analysis (TSA)
1. Introduction
• TSA makes no sense without an ecological framework
(i) Scale is important: data must reflect biology (ii) Data often have an ”internal dependence” (populations and phenology)
John Maynard Smith ... ”mathematics without ecology are sterile” ...
Time Series Analysis (TSA)
1. Introduction
bt
tt aN
RNN
1
1
1 −
−
+=
Maynard Smith – Slatkin population model
Dynamics of Natural Populations
1. Introduction
R: fundamental net reproductive rate a: susceptibility of crowding b: degree of intra-specific competition
“a” gives the level about which fluctuations occur
bt
tt aN
RNN
1
1
1 −
−
+=
Time Series Analysis (TSA)
1. Introduction
bt
tt aN
RNN
1
1
1 −
−
+=
Maynard Smith – Slatkin population model
( )btett aNrXX 11 1log −− +−+=
1)1())(log( −−+−= tet XbabrXAutoregressive model (AR)
4
Time Series Analysis (TSA)
1. Introduction
intra-specific
predator
inter-specific
forage
inter-specific
Combine ecological theory with time series analysis
Nt = f(Nt-1, Nt-2, Nt-3)
Time Series Analysis (TSA)
1. Introduction
predator
forage
intra-specific
inter-specific
inter-specific
Nt = f(Nt-1, Nt-2, Nt-3)
Combine ecological theory with time series analysis
Time Series Analysis (TSA)
1. Introduction
predator
forage
intra-specific
inter-specific
inter-specific
Nt = f(Nt-1, Nt-2, Nt-3)
Combine ecological theory with time series analysis
Time Series Analysis (TSA)
1. Introduction
predator
forage
intra-specific
inter-specific
inter-specific
Nt = f(Nt-1, Nt-2, Nt-3)
Combine ecological theory with time series analysis
Statistical dimension of time series indicates number of trophic interactions!!!
Year
Pop
ulat
ion
inde
x
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
1810 1870 1930 1990
Time Series Analysis 2. Sarcoptic mange – fox dynamics
Time Series Analysis (TSA)
2. Sarcoptic mange – fox dynamics
• The disease sarcoptic mange is caused by the skin-dwelling mite (Sarcoptes scabiei var. vulpes) and has been reported in red fox populations in Europe, North America and Russia.
• Approximately one month after exposure, infected foxes commonly develop skin lesions characteristic of hyperkeratosis. Severe loss of hair and progressive deterioration of body condition then follows and, in the majority of observed cases, infected foxes eventually die from starvation.
5
Time Series Analysis (TSA)
2. Sarcoptic mange – fox dynamics
Time Series Analysis (TSA)
2. Sarcoptic mange – fox dynamics
12110 −− ++= ttt XXX βββDensity
dependence Effect of mange
Time Series Analysis (TSA)
2. Sarcoptic mange – fox dynamics
12110 −− ++= ttt XXX βββ
<1 0
1 <0
AR(1)
AR(2)
Density dependence
Effect of mange
Time Series Analysis (TSA)
2. Sarcoptic mange – fox dynamics
β1
β2
Time Series Analysis (TSA)
2. Prey dynamics with climate
Forchhammer et al. 1
Figure 1
fox: Yt = ln(Pt)
prey: Xt = ln(Nt)
Climate: Ut
a1(t-1)
b1(t-1)
b3(t)
b2(t) a2(t-2)
a3(t) Pt = Pt−1 exp(a0 + a1Yt−1 + a2X t−2 + a3Ut )
Nt = Nt−1 exp(b0 + b1Xt−1 + b2Yt + b3Ut )
Xt = (b0a0 − a1b0 )+ (2 + a1 + b1)Xt−1 + (b2a2 − b1 − a1 − a1b1 −1)Xt−2
+(b3 + b2a3)Ut + (−b3a1 − b3)Ut−1
Xt = β0 + β1Xt−1 + β2Xt−2 +ω1Ut +ω 2Ut−1ARMA (2,2):
Time Series Analysis (TSA)
2. Prey dynamics with climate
Forchhammer et al. 1
Figure 1
fox: Yt = ln(Pt)
prey: Xt = ln(Nt)
Climate: Ut
a1(t-1)
b1(t-1)
b3(t)
b2(t) a2(t-2)
a3(t)
14
Table 1: The statistical (i.e., ARMA in Equation 3) coefficients expressed as a
function of the ecological interaction coefficients in Figure 1 and their ecological
interpretation.
Statisticalcoefficients
Ecologicalcoefficients
Ecologicalinterpretation
β1 2+a1+b1 direct density dependence: function of intra-trophic interactions {a1,b1} only.
β2 b2a2-b1-a1-a1b1-1 delayed density dependence: function of inter-trophic interactions {a2,b2} and a complex function of intra-trophic interactions {a1,b1}.
ω1 b3+b2a3 additive effect of direct {b3} and indirect climatic influence through predator dynamics {a3,b2}.
ω2 -b3a1-b3 direct climatic influence {b3} relative to its interaction with fox density dependence {a1}.
Forchhammer et al.
6
Time Series Analysis (TSA)
2. Prey dynamics with climate
Forchhammer et al. 1
Figure 1
fox: Yt = ln(Pt)
prey: Xt = ln(Nt)
Climate: Ut
a1(t-1)
b1(t-1)
b3(t)
b2(t) a2(t-2)
a3(t)
Forchhammer et al. 2
Figure 2
-0.04
0
0.04
0.08
0.12
roe deer hare partridge
0.6
0.8
1
1.2
roe deer hare partridge-0.4-0.3-0.2-0.100.10.2(a)
(b)
Ɣ E1’ ż E2’
Ɣ Z1’ ż Z2’
Z1’
and
Z2’
E1’ E2’ *
** *
*
*
*
Time Series Analysis (TSA)
2. Prey dynamics with climate
Forchhammer et al. 1
Figure 1
fox: Yt = ln(Pt)
prey: Xt = ln(Nt)
Climate: Ut
a1(t-1)
b1(t-1)
b3(t)
b2(t) a2(t-2)
a3(t)
• Hare and partridge dynamics displayed delayed DD and climate mediated through fox.
• Roe deer dynamics displayed direct DD with both direct and indirect climatic effects.
23
Figure 5
Forchhammer et al. 5
Figure 5
hare:partial R2 = 0.52
partridge:partial R2 = 0.49
-0.05
0
0.05
0.1
0.15
0.2
-1 -0.5 0 0.5
partual R2 = 0.53
-0.05
0
0.05
0.1
0.15
01020304050
(a)
(b)
habitat quality(% sand in soil)
goodpoor
Z 2
Z 1
b 2
Forchhammer et al.
dire
ct D
D
Year
Pop
ulat
ion
inde
x
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
1810 1870 1930 1990
Time Series Analysis 3. Global wader dynamics
Time Series Analysis (TSA)
3. Global wader dynamics
H
L Cold & dry
warm & wet
H
L Cold & dry warm
& wet
High NAO winter
Low NAO winter
-6
-3
0
3
6
1850 1900 1950 2000
Time Series Analysis (TSA)
3. Global wader dynamics
-4
-2
0
2
4
1970 1980 1990 2000
Stdz
UK
win
ter
popu
latio
ns
Eurasian (Bar-tailed Godwit, Curlew, Dunlin, Grey Plover, Oystercatcher)
Nearctic (Black-tailed Godwit, Knot, Turnstone)
Mixed (Redshank, Ringed Plover, Sanderling)
Time Series Analysis (TSA)
3. Global wader dynamics
wintering breeding
migration
NAOt-1 NAOt
A model:
dd dd
other species
Nearctic Eurasian
7
Time Series Analysis (TSA)
3. Global wader dynamics
Partial Autocorrelation Plot
0 10 20 30Lag
-1.0
-0.5
0.0
0.5
1.0
Cor
rela
tion
Lag (years)
Corr
elat
ion,
ρ(N
t, N
t-n)
Black-tailed Godwit Autocovariate structure: lagged influence of N on N
truncation
AR(1) process!: Nt = f(Nt-1, NAOt, NAOt-1)
Time Series Analysis (TSA)
3. Global wader dynamics
Autocovariate structure: Nt = f(Nt-1, NAOt, NAOt-1)
β 1
(1st
ord
er A
R c
oefic
ient
)
stable (-2<β1≤1)
unstable (β1>1)
-0.4
0
0.4
0.8
1.2
Bar-tai
led G
odwit
Curlew
Dunlin
Grey Plov
er
Oysterc
atche
r
Black-t
ailed
God
witKno
t
Turns
tone
Redsh
ank
Ringed
Plover
Sande
rling
Eurasian Nearctic Mixed
Time Series Analysis (TSA)
3. Global wader dynamics
*
Covariate structure: Nt = f(Nt-1, NAOt, NAOt-1)
ω1 ω
2 (N
AO c
oeff
icie
nts)
-8-6-4-202468
Bar-tai
led G
odwit
Curlew
Dunlin
Grey Plov
er
Oysterc
atche
r
Black-t
ailed
God
witKno
t
Turns
tone
Redsh
ank
Ringed
Plover
Sande
rling
* *
* * * *
*
Eurasian Mixed Nearctic
Assumes a linear relationship between NAO and N!
Time Series Analysis (TSA)
3. Global wader dynamics
Non-linear effects of the NAO: a TAR model
-6
-4
-2
0
2
4
6
1970 1980 1990 2000
NAO
win
ter
inde
x
early (low NAO) phase
late (high NAO) phase
-0.5
0
0.5
1
early late
early late
early late
-12
-8
-4
0
4
8
β 1
ω1 ω
2
AR(1) NAOt NAOt-1
*
*
1989
Tree-regression analysis (reduce deviance by splitting)
Bar-tailed Godwit
Time Series Analysis (TSA)
3. Global wader dynamics
-8-6-4-202468
Bar-tai
led G
odwit
Curlew
Dunlin
Grey Plov
er
Oysterc
atche
r
Black-t
ailed
God
witKno
t
Turns
tone
Redsh
ank
Ringed
Plover
Sande
rling
*
* *
* * * *
*
Covariate structure: Nt = f(Nt-1, NAOt, NAOt-1)
Eurasian Mixed
ω1 ω
2 (N
AO c
oeff
icie
nts)
Nearctic
Time Series Analysis (TSA)
3. Global wader dynamics
Non-linear effects of the NAO: a TAR model
-6
-4
-2
0
2
4
6
1970 1980 1990 2000
NAO
win
ter
inde
x
-0.5
0
0.5
1
early late
early late
early late
-12
-8
-4
0
4
8
1989
early (low NAO) phase
late (high NAO) phase
β 1
ω1 ω
2
AR(1) NAOt NAOt-1
*
*
-1.5
-1
-0.5
0
0.5
1
early late
early late
early late
-4
-2
0
2
4AR(1) NAOt NAOt-1
*
β 1
ω1 ω
2
Bar-tailed Godwit
Redshank
8
Time Series Analysis (TSA)
3. Global wader dynamics
Direct negative temporal dependence was found in most populations (fluctuating stability)
ln(Nt-1)
ln(N
t) (i)
NAOt
ln(N
t)
High NAO current year winters decreased the number of wintering waders in 3 species (windier and wetter)
(ii)
NAOt-1
ln(N
t) High NAO in previous year’s winter
increased the number of waders (Eurasian) the following winter (improved conditions when arriving)
(iii)
Non-linear response to changes in the NAO
Time Series Analysis (TSA)
3. Global wader dynamics
-4
-2
0
2
4
1970 1980 1990 2000
Stdz
UK
win
ter
popu
latio
ns
Eurasian (Bar-tailed Godwit, Curlew, Dunlin, Grey Plover, Oystercatcher)
Nearctic (Black-tailed Godwit, Knot, Turnstone)
Mixed (Redshank, Ringed Plover, Sanderling)
Mads C. Forchhammer Aarhus University
Department of Bioscience Arctic Research Centre (ARC)
CIRCE
email: [email protected]