Landscape Ecology

1
Landscape Ecology
2
This Talk

• Background - landscape ecology
• Major landscape experiments
• Riverina restoration study (or what we got for
  our phone company)

3
Landscape Ecology

• The study of ecology at large-scales -
• usually means the study of species or groups of
  species responding to something at large scales
• Natural landscapes
• Fragmented landscapes
• Corridors connecting patches
• Restoration in landscapes

4
(No Transcript)
5
(No Transcript)
6
(No Transcript)
7
(No Transcript)
8
(No Transcript)
9
Australia The Continent
Same size as mainland USA 20 million people 7 to
10 of worlds species Oldest, most isolated
continent oldest living life forms
10
Biodiversity

• 30,000 sp flowering plants (85 endemic)
• 820 sp. Eucalypts 800 sp. Acacias
• gt300,000? Invertebrate taxa (gt95 endemic)

11
Biodiversity

• Centre of marsupial radiation

85 mammals endemic
one sixth of worlds parrots
12
Biodiversity

• 89 reptiles endemic
• 25 new species per year! (eg Jervis Bay)
• Deserts worlds highest reptile diversity

• Greatest diversity of
• front-fanged snakes, pythons, sea-snakes
• skinks, goannas and geckos

93 of amphibians endemic
13
The impact on the land

• massive changes in 200 years
• extensive land clearing (gt500 million ha yr)
• high land degradation rates
• gt 25 million ha salt-affected by 2050
• widespread soil degradation 20 tonnes/ha/yr

14
Bush Bashing

• some vegetation types 95-99 cleared
• many woodlands lt1 remain, often on roadsides and
  public lands (e.g. rubbish tips, cemeteries)
• no patches gt 3 ha, most ltlt 1 ha
• temperate grasslands NE Victoria lt 0.01 left

15
Impacts

• leads the world in mammal extinctions
• 30 mammals from deserts gone
• 99 range contraction of some mammals
• 50 of woodland birds extinct in 100 years
• 15 of frogs extinct, endangered, threatened

16
Impacts

• Overall threatened species (total)
• 14 of mammals, 12 of plants
• (next highest USA/Mexico 12, 8)

17
Landscape Research and Conservation

• Many conceptual modellers
• Many simulation modellers
• Few large-scale empirical studies
• Field data appropriately analyzed can change
  policy and improve conservation

18
5 Key Programs

• Central Highlands of Victoria
• Tumut Fragmentation Expt
• Nanangroe Longitudinal Study
• Jervis Bay Ecological Burning Study
• Riverina Restoration Study
• Approx. 6 major sub-projects per program

19
Background

• Research Team
• - 1 x Ecologist (Lindenmayer)
• - 1 x Statistician (Cunningham)
• - 3 x Field Staff (Crane, Michael, McGregor)
• Field staff locally based

20
Background

• A270-300K per year
• Highly productive
• Issues beyond science
• Strong community links
• Communication with key parties

21
Central Highlands of Victoria

• 1983- (20 years)
• Native forest - wood prod. conservation
• Array of field sites - varying age classes,
  disturb. hist.
• ID logging effects and impact mitigation
• Mammals, birds, reptiles

22
(No Transcript)
23
(No Transcript)
24
(No Transcript)
25
The Tumut Fragmentation Natural Experiment

• Commenced 1995 - ongoing
• Eucalypt patches in pines
• plantation design conservation
• Mammals, birds, reptiles, frogs, inverts,
  vascular plants, mosses, weeds etc

26
The Tumut Fragmentation Natural Experiment

• Many features of an experiment
• 166 sites, matched controls and matrix sites
• X-sectional landscape context fragmentation
  effects

27
(No Transcript)
28
Tumut Outcomes

• New fragmentation perspectives
• Transformed plantation design and establishment
  in south-eastern Australia
• Forest and woodland clearing patch size down to 1
  ha

29
The Nanangroe Longitudinal Study

• Commenced 1997- ongoing
• Woodland fragments in new pine landscape
  cross-matched sites
• Changing matrix effects on fragments
• Prospective longitudinal 62561010 138
  sites by at least 8 time points
• Mammals, birds, reptiles, frogs

30
(No Transcript)
31
Jervis Bay Ecological Burning Study

• Commenced Dec. 2002
• 112 sites
• 8 veg types x 4 age since burning classes
  restrospective study
• 60 of sites to be prescribed burned in
  before/after longitudinal study
• Mammals, birds and reptiles

32
(No Transcript)
33
Outcomes

• New species !!!!
• New experiment implemented
• Capacity-building in local indigenous community

34
The Riverina Restoration Study
35
Background

• What is the value of restoration?
• (What did we get for the Telco?)
• Biodiversity values not known
• What species use replanting?
• Is it related to replanting size, shape etc

36
Complex Issues

• Replantings and remnant
• vegetation on same farm (additive?)
• Is replanting use related to how much
• vegetation cover in broader landscape
• (influences if planting conc or dispersed
• Is there a threshold vegn cover affect
• - and how would we test for it?

37

• Many declining small birds in here eg. Red-Capped
  Robin

Plantings near Ladysmith
38
A typical planting in the study near Nangus
39
A typical remnant in the study near Gundagai
40
Why is conserving wildlife important?

• Wildlife provide many environmental services
  (pest control)
• Wildlife is part of our Natural Heritage
• Wildlife can potentially provide future resources
  (food medicines)

41
Extinct from the SW Slopes
Bilby (Walla Walla and Cootamundra)
Bridle Nail-Tail Wallaby (Wagga Wagga)
Australian Bustard
42
Objectives of the study

• To estimate the effects of tree planting on
  presence and abundance of vertebrates( birds,
  mammals, reptiles) and assess whether the effects
  are consistent across different farm types and
  landscape types.
• Compare vertebrates across a range of vegetation
  types within different landscapes.
• Explore relationships between biota and
  covariates at the different levels.
• Are identified relationships of practical use as
  predictors to help guide landscape restoration.
• At what scale, if any, does tree planting
  increase wildlife diversity?

43
Experimental Design

• 2 Regions (Murray Murrumbidgee Catchments)
• 21 Landscapes (10 000ha area)
• 2 Farms per landscape(1000ha area)
• 4 sites per farm (total 168)
• 3 plots per site (observational unit)

44
10 Landscapes (10000ha) of the Murray Region
(showing remnant veg)

• Henty

• Holbrook

• Albury

45
Landscape with plantings (10000ha)

• Plantings

• Two Farms per Landscape

• Remnants

46
Treatment Design _at_ Landscape level

• Planting
• High Low
• High remveg 1 4
• Low remveg 4 4 reps
• Plus
• 8 control landscapes
• 4 low remveg no plantings
• 4 high remveg no plantings

47
Treatments

• Site level
• Four (4) sites per farm
• 3 plantings and 1 remnant
• or
• 4 remnants (on farms with no
• plantings- natural regrowth, old growth,
  coppiced)
• Plot Level (measurement)
• 3 plots 2 observers ( bird counts)
• transect spotlight searches for possums
• 3 lots of hair tubes for ground mammals
• 3 sets refuges for reptiles

48
A farm without plantings (1000ha)
It is expected that many species found in
remnants and plantings will be effected by what
makes up the surrounding area (eg. paddock
trees, cropping paddocks and other plantings
and remnants). These variables will be measured
directly(satellite images) around the site,
across the farm and the landscape.

• Paddock Trees

49
Some covariates at Site level

• Sites types
• Plantings
• (block or strip small or large)
• Remnants
• (large or small heavily grazed, lightly grazed
• natural regeneration)

50
Treatment Structure (Actual)

• Growth Type
• plantg regrowth coppiced
  oldgrth
• Land_Type Farm_Type
• Planting Planting 39 3
  2 8
• No Planting 0
  14 12 27
• No Planting Planting 0 0
  0 0
• No Planting 0
  18 12 33

51
Statistical Model

• The general form for a binary response was
  linear logistic
• logit pXbZu
• Fixed factors (X) included the treatments
  Landscape type, Farm Type, Growth Type
  covariates at all levels.
• Fixed effects are represented by b.
• Random effects (Z) were associated with
  Landscape, Farm unit and Site.
• The vector of random effects is represented by
  u and
• var (u)G, a function of unknown variance
  components.
• Regression coefficients and variance components
  were jointly estimated, using weighted least
  squares and REML, or the method of estimation for
  generalised linear mixed models (Schall, 1991).

52
Results Number of species of bird per site

• Estimated variance components (Scale effects)
• Random term Component Component s.e.
  ChiSq
• No covariates Full model
• Landscape 0.290 0.181
  0.435 1
• Farm 0.846 0.973
  0.520 14
• Site 3.134 2.950 0.401

53
Results Number of Species per site

• Table of predicted means for
  LandscapeType.Farm_Type (p0.02)
• Farm Type
• Planting No Planting Margin
• LandscapeType
• Planting 4.377
  2.844 3.611
• No Planting
  2.718 2.718
• Standard error of differences Average
  0.6138

54
Results Number of Species per site

• Table of predicted means for Farm Type effect
  within Planted Landscapes for natural veg sites
  (p0.009)
• Farm_Type
• Planting No Planting
• Planting
• (Site level)
• Natural 4.975 3.379
• Yes 3.824
• Standard error of differences Average
  0.5438

55
Results Number of Species per site

• Table of predicted means for GrowthType
• p-value for planting v natural 0.037
• p-value for effect of growth type in natural
  0.056
• Growth_Type
• planting regrowth coppiced oldgrowth
• 2.494 3.021 4.138 3.756
• Average standard error of difference0.53

56
Results Number of Species per farm

• Richness data are extensive ( Aggregate
  quantities have the same well-defined physical
  meaning)
• Table of predicted means for LandscapeType.Farm_Ty
  pe (plt0.001)
• Farm_Type
• Planting No Planting Margin
• LandscapeType
• Planting 18.93
  13.62 16.28
• No Planting
  14.91 14.91
• Standard error of differences Average
  1.109
• Landscape component of variance 2.858 ChiSQ3
• Residual variance 6.339

57
ResultsProbability of detecting Ringtail Possum

• Estimated variance components and dispersion
• Random term Component
  s.e.
• No covariates
• Landscape 0.231
  0.562
• Farm 1.095 0.520
  
• Dispersion 0.76 0.095

58
ResultsProbability of detecting Ringtail Possum

• Estimated variance components and dispersion
• (excluding Plantings - no Ringtails)
• Random term Component s.e.
• no covariates
  
• Landscape 0.4513 0.6055
• Farm Unit 0.6074 0.6904
• Dispersion 0.830 0.1239

59
ResultsProbability of detecting Ringtail Possum

• Detection rates()
• Landscape Type Planting No Planting
• Growth_Type Farm_Type
• planting Planting 0
• No Planting
• regrowth Planting 33
• No Planting 36 39
• coppiced Planting 50
• No Planting 0 33
• oldgrowth Planting 25
• No Planting 52 30

60
ResultsProbability of detecting Common Brushtail
Possum

• Estimated variance components and dispersion
• Random term Component
  s.e.
• No covariates
• Landscape 1.286
  0.78
• Farm 0.819 0.604
  
• Dispersion 0.728 0.091

61
ResultsProbability of detecting Common Brushtail
Possum

• Detection rates()
• Landscape Type Planting No Planting
• Growth_Type Farm_Type
• planting Planting 5
• No Planting
• regrowth Planting 0
• No Planting 50 28
• coppiced Planting 0
• No Planting 33 50
• oldgrowth Planting 38
• No Planting 26 64

62
ResultsProbability of detecting Common Brushtail
Possum

• p-value for Landscape Type.Growth Type0.04
• Estimated probabilities
• 
  LandscapeType
• Planting No
  Planting
• Growth_Type
• planting Excluded
• natural regrowth 0.3897 0.2572
• coppiced 0.3846 0.4598
• oldgrowth 0.2211 0.6195

63
Some statistical issues in the design of
quasi-experiments in landscape ecology

• experimental units are often large and
  heterogeneous
• size is important!
• replication is expensive but fundamental
• random assignment not possible but random
  (haphazard) choice maybe possible
• identify key factors and control (stratify) for
  them or at least, measure them
• adjacent experimental units not spatially
  independent
• not possible to have balance(orthogonality)

64
Some statistical issues in the design of
quasi-experiments in landscape ecology

• measurement is often a problem-low detection
  rates
• ensure equal size and equal effort at each
  experimental unit
• there are many possible covariates at different
  levels
• data are inherently multivariate and are often
  binary or counts - inflated zeros
• statistical methods are often sophisticated-mixed
  models
• language of ecologists? Sampling means taking an
  observation-measurement protocols in the field?
  Not about selection of units.Census means to
  count birds ( or animals) etc

65
Summary

• Common misconception is that design is about
  treatment structures only.
• Behind the simplest design lies a wealth of
  thought and discussion. (Effective replication
  Control- identifiable factors strata etc
  Randomisation - average effects of uncontrollable
  factors)
• Statistical science does make a difference when
  collaboration is effective
• Output is more than the sum of the parts -
  synergism
• The development of experimental design is one of
  the great contributions of statistical science to
  science and technology. Yet almost nobody knows
  anything about it. John Nelder, 1999