detecting trends in dragonfly data - difficulties and opportunities - arco van strien statistics...

30
Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduct ion

Upload: spencer-paul-weaver

Post on 13-Jan-2016

216 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

Detecting trends in dragonfly data

- Difficulties and opportunities -

Arco van Strien

Statistics Netherlands (CBS)

Introduction

Page 2: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

Statistics Netherlands’ involvement in monitoring

National:

Breeding birds

Waterbirds

Flora

Amphibians

Reptiles

Butterflies

Dragonflies

Bats and a few other mammals

Lichens

Fungi

• Close co-oparation with NGO’s for each species group

• NGO’s responsible for field work (mainly by volunteers)

• Statistics Netherlands responsible for statististical analysis

Introduction

International:

European Wild Bird Indicators

European Grassland Butterfly Indicator

European Bats (in development)

Page 3: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

Structure of talk• Difficulty 1: Statistical power• Difficulty 2: Bias

Three monitoring alternatives:• No standardisation of field method• Strong standardisation & analysis method TRIM • Medium standardisation & analysis method Occupancy

modeling

• Combining trends (European perspective)• Conclusions

Introduction

Page 4: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

Difficulties in trend detection:• Inability to detect existing trends (low statistical

power)• Even more worse: Biased trend estimates

(increase or decline larger or smaller than in reality)

• Sufficient statistical power to detect trends

• No or negligible bias in trend estimates

Monitoring schemes need to have:

Introduction

Page 5: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

Statistical power is low if: • Yearly fluctuations in abundance are high • Sites have different year-to-year changes

Dragonflies have considerable fluctuations in abundance …

Power

Page 6: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

Power

Page 7: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

Statistical power is low if: • Yearly fluctuations in abundance are high • Sites have different year-to-year changes

• Sufficient sampling effort (= no. of sites; exact no. sites required depends on a.o. field method)

• Use longer detection period (= wait longer)

Remedies:

Power

Page 8: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

Length of time series of smooth snake Coronella austria

A longer detection period leads to more accurate trend estimates

Years

Power

Page 9: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

Risk of bias is higher if: • Sampling effort per site is not constant across years• Detectability of species is not constant during the season

and between years• Inadequate sampling strategy applied e.g. dragonfly-rich

areas oversampled

Bias

Sufficient statistical power to detect trends• No or negligible bias in trend estimates

Monitoring schemes need to have:

Page 10: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

Increasing sampling effort leads to artificial increase

Bias

High sampling effort

Low sampling effort

Index of species

Page 11: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

Risk of bias is higher if: • Sampling effort per site is not constant across years• Detectability of species is not constant during the season & years• Inadequate sampling strategy applied e.g. dragon-rich areas

oversampled

• Standardize sampling effort (field method)• Take into account variation in detectability

during the season (= multiple visits) • Apply adequate sampling strategy (or adjust a

posteriori any bias due to unequal sampling)

Remedies:

Bias

Page 12: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

Structure of talk Difficulty 1: Statistical power Difficulty 2: Bias

Three monitoring alternatives:• No standardisation of field method• Strong standardisation & analysis method TRIM • Medium standardisation & analysis method Occupancy

modeling

• Combining trends (European perspective)• Conclusions

Alternatives

Page 13: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

Field method Data collection Example

no standardisation

presence data e.g. per grid cell (e.g. 5x5 km2)

Comparison of distribution between two periods

strong standardisation

count data per site Dutch dragonlfly scheme

medium standardisation

presence and absence data per grid cell or site

?

Monitoring alternatives

Alternatives

Page 14: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

No standardisation of field method

• As in Atlas studies or studies to compile Red Lists• No fixed sites • Sampling efforts vary between years• No prescription of field method• No formal sampling strategy • Collecting presence data only

>>>>• Statistical power low (only sensitive to pick up strong declines

& increases in distribution) • Risk of bias considerable due to not constant sampling efforts• Statistical analysis: simple comparison of distribution data (or a

statistical method)

Alternatives

Page 15: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

Strong standardisation of field method

• As in dragonfly scheme in the Netherlands• Fixed sites (500 m - 1 km long)• Yearly surveys• Multiple visits per year (during the season)• Detailed prescription of field method (fixed sampling effort per

site) • Sampling strategy: preferably (stratified) random choice of sites• Collecting count data

>>>>• Statistical power high • Risk of bias low• Statistical analysis: TRIM

TRIM

Page 16: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

TRIM version 3 Poisson regression (loglinear models, GLM) for count data

Pannekoek, J. & A van Strien, 2001. TRIM 3. Statistics Netherlands, Voorburg TRIM

Page 17: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

TRIM: Trends and Indices for Monitoring data

• Specially developed by Statistics Netherlands for wildlife monitoring based on count data

• Statistical heart of wildlife monitoring data analysis• Internationally accepted and in use in many European

countries• Easy to use• Freeware• Calculates yearly indices

TRIM

Page 18: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

Site

Year 1 Year 2 Year 3 Year 4 Year 5

1 20 10 8 2 3 2 20 10 12 3 2 3 16 8 10 3 3 4 8 4 6 6 5 5

10 5 7 7 8

Sum

74 37 43 21 21

Index 100 50 58 28 28

INDEX: the total (= sum of al sites) for a year divided by the total of the base year

TRIM

Page 19: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

Statistical characteristics of TRIM

• Produces yearly indices and overall trends per species • Produces confidence intervals • Include overdispersion & serial correlation in models• Goodness-of-fit tests for comparing models• Covariates to test trends between sets of sites • Weight factors may be included to improve representativeness if sites are not randomly selected• Imputation of missing values

TRIM

Page 20: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

Site

Year 1 Year 2 Year 3 Year 4 Year 5

1 20 10 8 2 3 2 20 10 12 3 2 3 16 (7.5) ? 10 3 3 4 8 4 6 (2.3) ? 5 5

10 5 7 7 8

Sum

74 36 43 17 21

Index 100 49 58 23 28

Imputation of missing counts required to compute correct indices

TRIM

Page 21: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

Medium standardisation of field method

To be developed, but think of:

• Preferably fixed sites• Survey per site once every 2-3 years• Multiple visits per year (during the season)• Limited prescription of sampling effort per site, e.g. 1 hour field work• Sampling strategy: preferably (stratified) random choice of sites• Collecting presence/absence data per site per visit (or abundance

categories)

>>>• Power not high • Risk of bias low• Statistical analysis: Occupancy modeling to adjust for bias due to

limited standardisation

Occupancy

Page 22: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

MacKenzie, D.I., J.D. Nichols, J.A. Royle, K.H. Pollock, L.L. Bailey & J.E. Hines, 2006. Occupancy estimation and modeling. Elsevier, Amsterdam.

Occupancy modeling: Recent developments in statistical methods make it possible to estimate area of occupancy while taking into account the detectability of species (which may differ according to e.g. not constant sampling efforts)

Based on absence/presence data from repeated visits (capture-recapture)

Statistical method is in development

Freeware (PRESENCE, MARK)

Occupancy

Page 23: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

Occupancy modeling uses capture histories per site to separate occupancy and detectability

Site Visit 1 Visit 2

1 0 1

2 1 0

3 0 1

4 1 0

Occupancy

Site Visit 1 Visit 2

1 1 1

2 1 1

3 0 0

4 0 0

area of occupancy 100%detection probability per visit 50%

area of occupancy 50%detection probability per visit 100%

Simple example

Page 24: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

Statistical characteristics of Occupancy modeling

• Produces estimate of area of occupancy per year (or period), taking into account detectability of species• Comparing area of occupancy per period >> trend • Produces confidence intervals• Allows missing values• Covariates to allow for effect of e.g. temperature during visit, incompleteness of survey etc. • Weighting procedure (if sites are not randomly selected) to be developed

Occupancy

Page 25: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

Structure of talk Difficulty 1: Statistical power Difficulty 2: Bias

Three monitoring alternatives: No standardisation of field method Strong standardisation & analysis method TRIM Medium standardisation & analysis method Occupancy

modeling

• Combining trends (European perspective)• Conclusions

Combining trends

Page 26: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

European population trend of species A

European population trend of species A

A. van Strien, J. Pannekoek & D. Gibbons, 2001. Bird Study 48:200-213Combining trends

Yearly population size of species A in country 1

Yearly population size of species A in country 1

Combining TRIM results per country, weighted by population sizes, is well-developed

Yearly population size of species A in country 2

Yearly population size of species A in country 2

Yearly population size of species A in country 3

Yearly population size of species A in country 3

Yearly population size of species A in country 4

Yearly population size of species A in country 4

Page 27: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

40

60

80

100

120

140

1980 1985 1990 1995 2000 2005Year

Pop

ulat

ion

Inde

x (1

980=

100)

Common farmland birds (19)

Other common birds (25)

Common forest birds (33)

Gregory R.D., van Strien, A., Vorisek P. et al., 2005. Phil. Trans. R. Soc. B. 360: 269-288 Gregory, R.D., Vorisek, P., van Strien, A. et al., 2007. Ibis, 49, s2, 78-97

Combining trends

Example of combining TRIM results of countries: Pan-Euromonitoring Common Bird Monitoring project producing Farmland Wild Bird Indicator (EU biodiversity indicator)

Page 28: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

European trend in occupancy area ofspecies A

European trend in occupancy area ofspecies A

Yearly occupancy area of species A in country 1

Yearly occupancy area of species A in country 1

Combining areas of occupancy per country, weighted by areas, appears possible (but needs to be developed)

Yearly occupancy area of species A in country 2

Yearly occupancy area of species A in country 2

Yearly occupancy area of species A in country 3

Yearly occupancy area of species A in country 3

Yearly occupancy area of species A in country 4

Yearly occupancy area of species A in country 4

Combining trends

Page 29: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

Scores of alternatives

Field method Power to detect trends

Risk of bias

Sampling effort needed

no standardisation (presence data)

strong standardisation

(count data)

medium standardisation

(pres/absence data)

Conclusions

Page 30: Detecting trends in dragonfly data - Difficulties and opportunities - Arco van Strien Statistics Netherlands (CBS) Introduction

Conclusions

• Statistical enemies of monitoring: low power and bias• Standardisation of sampling effort helps to increase

power and to reduce bias• Monitoring based on strong standardisation: high power

& little bias. But it requires considerable sampling efforts • If strong standardisation is not feasible, consider medium

standardisation: lower power, but again little bias (if detection probabilities are taken into account)

• For both alternatives statistical methods are available • Both alternatives enable to combine trends across

countries

Conclusions