Input from capture mark recapture methods to the understanding of population biology

Transcription

1 Input from capture mark recapture methods to the understanding of population biology Roger Pradel, iostatistics and Population iology team CEFE, Montpellier, France 1

2 Why individual following? There are phenomenons for which it is not sufficient to observe an individual once. survival 2

3 Why individual following? There are phenomenons for which it is not sufficient to observe an individual once. survival Trade-offs Carry-over effects 3

4 Detection in the wild Individual recognition marks (artificial or natural) Account for imperfect detection specific models 4

5 A current unifying paradigm: the multievent framework An upshot of 5 years of development (Cormack 1964, Jolly 1965, Seber 1965, Arnason 1972,1973, Pradel 25) 5

6 A current unifying paradigm: the multievent framework Central idea : Disconnect state and event 6

7 The simplest CR data (Cormack-Jolly-Seber model) European dipper individual marks K occasions encountered not encountered 7

8 The simplest CR data (Cormack-Jolly-Seber model) encountered code 1 not encountered code each individual has an encounter history Time : Individual # Individual # Individual # Individual # i... 1 Individual # Individual # 255 8

9 The simplest CR data (Cormack-Jolly-Seber model) what we know: encountered or not encountered phenomenon of interest? survival Consider the two states: alive dead survival = transition from state alive to state dead 9

10 The simplest CR data (Cormack-Jolly-Seber model) what we know: encountered or not encountered what we want to know? survival A distinction is made between what we perceive: what is hidden: the events the states 1

11 The simplest CR data (Cormack-Jolly-Seber model) the state process takes place between times t and t+1: alive at t dead at t alive at t+1 dead at t+1 dead at t+1 with probability f with probability 1-f with probability 1 the state process depends on the last state reached. (first-order Markov process) 11

12 The simplest CR data (Cormack-Jolly-Seber model) the event process takes place at time t: alive at t encountered with probability p not encountered with probability 1-p dead at t not encountered with probability 1 the event process depends on the current state 12

13 The simplest CR data (Cormack-Jolly-Seber model) Vocabulary The probabilities that describe the state process are called transition probabilities. Those that describe the event process are called event probabilities. 13

14 more complex CR data intering site fidelity in Canada Geese ranta canadensis Hestbeck et al. 1991) 3 sites (large areas along East coast of US) Carolinas, Chesapeake, Mid-Atlantic 6 years (winter captures+resightings)

15 A more complex CR data (Conditional Arnason-Schwarz model) Time Canada goose data History # History # History # History # h History # History # 624

16 Studied phenomenon: movements Events () Not encountered (1) Encountered in NA (2) Encountered in Ch (3) Encountered in Ca

17 Studied phenomenon: movements (partially) hidden states Dead Alive at NA Alive at Ch Alive at Ca Events and states are closely related, but not entirely

18 States generate events States Dead Alive at NA Alive at Ch Alive at Ca Events () Not encountered (1) Encountered in NA (2) Encountered in Ch (3) Encountered in Ca

19 The state process It is described by the transition matrix: Φ t NA Ch Ca NA Ch Ca f f f f f f f f f f 1f 1f f f f f f f

20 The event process It is described by the event matrix: p p p p p p t NA Ch Ca 1 2 3

21 Preferred alternative parameterization for the state process Transition between successive wintering sites can be viewed as involving two successive distinct processes: S : survival followed by Y : movement conditional on survival 21

22 Step 1: Survival IN ROW: IN COLUMN: Ai : Alive at site i; D dead SFi : Survivor from site i; D dead SF1 SF2 SF3 D A1 s 1 1 s 1 A2 s 2 1 s 2 D A3 s 3 1 s 3 1

23 Step 2: Movement IN ROW: IN COLUMN: SFi : Survivor from site i; D dead Ai : Alive at site i; D dead A1 A2 A3 D SF SF SF D 1

24 Some cases with more state uncertainty reeding status (Kendall et al., 23; Pradel 25) Sex membership (Fujiwara & Caswell, 22; Nichols et al., 24; Pradel et al., 28) Hidden heterogeneity (Pledger et al., 23) Epidemiology (Conn & Cooch, 29)

25 reeding status 4 events: Not encountered () Found, ascertained as breeder (1) Found, ascertained as non-breeder (2) Found, status unknown (3) 3 states: breeding () non-breeding (N) Dead ( )

26 The model construction States reeding Non-breeding Dead Events () Not encountered (1) Found, identified as (2) Found, identified as N (3) Found, status unknown Each live state can generate 3 different events

27 reeding status: initial state probabilities probability that a newly encountered individual is a breeder Π t N 1

28 transition parameters 1 1 ) (1 1 ) (1,,,, N N N N N t f f f f f f Φ N N

29 transition parameters or equivalently (two steps) Step 1: Survival (t) N(t) S SN f f N 1f 1f 1 N Intermediate states: S = surviving breeder ; SN = surviving non-breeder Step 2: reeding S SN (t+1) N(t+1), N, 1 1, N, 1

30 event parameters 1 ) (1 1 ) (1 1 N N N N N t p p p p p p N N

31 Step 1: Encounter event parameters or equivalently (two steps) N 1 p 1 p 1 NE E E N Intermediate events: NE = not encountered ; E = encountered N p p N Step 2: reeding assessment NE E E N N 1 1 N

32 The generic multievent model b, probabilities of events conditional on current state E 1 E 2 E 3 S 1 S 2 S 3 events states, initial state probabilities f probabilities of transition (Hidden Markov Model Structure)

33 Unleashing the imagination! Same data Different questions different states

34 Phoenicopterus ruber roseus DATA 3 Events: = not seen 1= captured and ringed fledgling 2= seen at the colony breeder example capture history: Ringed as chicks No recruitment before 3 years old (age sexual maturity) Only breeders are resighted at the colony

35 Phoenicopterus ruber roseus DATA 3 Events: = not seen 1= captured and ringed fledgling 2= seen at the colony breeder Studying accession to reproduction 3 States: NR= not recruited yet R= recruited D= dead

36 Phoenicopterus ruber roseus DATA 3 Events: = not seen 1= captured and ringed fledgling 2= seen at the colony breeder Studying breeding propensity 3 States: N= not breeding = breeding D= dead

37 Studying the role of breeding experience in breeding propensity (Pradel et al. PLOS One 212) 7 States: N= not breeding, no previous experience at the onset of the season, = breeding, no previous experience at the onset of the season, N1= not breeding, one previous experience at the onset of the season, 1= breeding, one previous experience at the onset of the season, N2= not breeding, > 1 previous experience at the onset of the season, 2= breeding, > 1 previous experience at the onset of the season D= dead

38 Studying the role of breeding experience in breeding propensity (Pradel et al. PLOS One 212) 2+ 1 Experience level

39 Use in behavioural studies

40 foraging specialization in a generalist species (Sanz-Aguilar et al. Ecology 215) White storks wintering in southern Spain feed on rice field and dumps

41 foraging specialization in a generalist species (Sanz-Aguilar et al. Ecology 215) known residents and known immigrants seem to have different habits

42 foraging specialization in a generalist species (Sanz-Aguilar et al. Ecology 215) DATA 3 Events (8 days): = not seen 1= observed at a ricefield 2= observed at a dump example capture histories: known residents, 711 known immigrants, 782? Invasive crayfish in ricefields available after ploughing Mortality negligible during the study period

43 foraging specialization in a generalist species (Sanz-Aguilar et al. Ecology 215) 6 States: R1= resident, ricefield specialist, R2= resident, dump specialist, R3= resident, generalist, I1= immigrant, ricefield specialist, I2= immigrant, dump specialist, I3= immigrant, generalist No change of state allowed Interest lies in proportions of the different strategies in the 2 groups

44 foraging specialization in a generalist species (Sanz-Aguilar et al. Ecology 215) IS: specifying the states nature Step 1: Origin Step 2: strategy π g fixed to 1 for known residents, for known immigrants

45 foraging specialization in a generalist species (Sanz-Aguilar et al. Ecology 215) Events: specifying the states nature Step 1: use of ricefields α fixed to 1 for R1 and I1, to for R2 and I2 Step 2: resighting

46 foraging specialization in a generalist species (Sanz-Aguilar et al. Ecology 215) specialists generalists ricefields dumps Residents 72% % Immigrants 16% 24% 6%

47 Use of indirect information

48 estimating pair fidelity (Culina et al. Ecology & Evolution 213) Even when focal individual not captured, we may know something about its state Use of events such as: «Previous partner captured with another bird»

49 estimating pair fidelity (Culina et al. Ecology & Evolution 213) Using only certain status observations leads to overestimation of fidelity ψ Results Great Tits, Wytham woods, 3 years of data faithful individuals survive better (.6 >.4) and remain more faithful (.4>.3)

50 Studying reproductive skipping behavior (Sanz-Aguilar et al. Ecological Applications 212) Using nest inspection

51 Studying reproductive skipping behavior (Sanz-Aguilar et al. Ecological Applications 212) Results: Cory s Shearwater individuals temporarily absent have lower future reproduction and breeding-site fidelity nest information allows estimation of temporary emigration + new biological insights (intracolony dispersal )

52 getting the best from your data

53 Adult survival selection in relation to multilocus heterozygosity (Cézilly et al. Oecologia 216) A difficult question: heterozygosity-fitness correlations expected to be low. A relatively modest sample size: 391 individuals, 7 years Logit(S) = a + b * MLH search of a parsimonious modeling of detection

54 recapture rate Adult survival selection in relation to multilocus heterozygosity (Cézilly et al. Oecologia 216),9,8, ,6 12,5,4, ,2, effort (days) Logit(p) = t+sex+ind detection is improved by additional fieldwork days late occasions lie above trend

55 survival rate Adult survival selection in relation to multilocus heterozygosity (Cézilly et al. Oecologia 216),9,8,7,6,5,4,3,2,1,36,56,76,96 Weak but significant effect of MLH detected (percentage deviance explained is 1.14%) MLH Logit(S) = a + b * MLH Logit(p) = sex+ b*log(effort)+c*trend+ε 2 parameters instead of 12 with t

56 Conclusion

57 Other multievent models All previous models program E-SURGE (downloadable from CEFE web site) 1. robust design (Souchay et al. 214, Kendall et al. 212) 2. epidemiology (false positive/negative) (Conn & Cooch 29) 3. individual heterogeneity (Pledger et al. 23) 4. sources of mortality (Schaub & Pradel 24, Tavecchia et al. 212) 5. senescence (Marzolin et al. 211) 6. sex or species uncertainty (Runge et al. 27, Genovart et al. 212) 7. stopover duration (Rivalan et al. 26) 8. memory models (Rouan et al. 29) 9. mixture of information (Duriez et al. 29) 1. dispersal among habitats (Cayuela et al. 217)

58 Models requiring other tools spatially explicit CR models (orchers J ornith 21), integrated population models Use ugs called from R (Kéry & Schaub, 212. Academic Press, urlington)