Sampling techniques to verify forest data gathered, and statistics involved

1
Sampling techniques to verify forest data
gathered, and statistics involved

• T.A. Groen, P. Van Laake and B. Turner

2
Outline

• Basic statistical issues -calculation of sample
  size
• IPCC directives-How they want to see
  calculations
• Sampling methods

3
Requirements

• Representative and large enough dataset
• Questions
• What is large enough?
• What is representative?

4
What is large enough?

• Quality of the dataset
• Variability

5
Source GOFC-GOLD sourcebook june 2008
6
Sources of errors

• Measurement error
• Wrong inference by use of models
• Upscaling over non homogeneous area
• Upscaling using remote sensing with
  classification errors
• Registration and calculation errors

7
What is large enough?

• Quality of the dataset
• Variability

Histogram
8
What is large enough?

• Quality of the dataset
• Variability

9
What is large enough?

• Quality of the dataset
• Variability
• Variance
• Standard deviation

10
What is large enough?

• Quality of the dataset

11
What is large enough?
12
What is large enough?
Confidence interval
Source http//www.answers.com/topic/normal-distri
bution
13
What is large enough?

• Margin of error
• Confidence level (e.g. 95)
• Look up z
• Calculate margin of error with

14
What is large enough?

• x350 Mg C ha-1
• s100 Mg C ha-1
• n40
• confidence level 95
• z 1.960
• m

15
What is large enough?

• m 31 Mg C ha-1
• Population mean is somewhere between
• 350 31 and 350 31 or
• 311 and 381

16
What is large enough?
17
What is large enough?

• Imagine we strive for a sampling error of 5
• Then m 0.05 350 17.5 Mg C ha-1
• Confidence interval 95 -gt z 1.960
• s s 100

18
Outline

• Basic statistical issues -calculation of sample
  size
• IPCC directives-How they want to see
  calculations
• Sampling methods

19
What is large enough?

• Specific GPG LULUCF requirements
• uncertainty defined as

Please note that There is no predetermined
level of precision uncertainty is assessed to
help prioritise efforts to improve the accuracy
of inventories in the future and guide decisions
on methodological choice. Uncertainties are also
of interest when judging the level of agreement
between national inventories and emission or
removal estimates made by different institutions
or approaches. IPCC 2006
20
(No Transcript)
21
Calculations with uncertainties

• For example C has been estimated in above ground
  biomass and below ground biomass
• Then total uncertainty is

22
Conservancy Issue

• the reductions in emissions should not be
  over-estimated (or at least the risk of
  overestimating should be minimized)
• IPCC GPG LULUCF

source
23
Outline

• Basic statistical issues -calculation of sample
  size
• IPCC directives-How they want to see
  calculations
• Sampling methods

24
What is representative?

• Sampling strategies

Forest
25
What is representative?

• Sampling strategies

Systematic Random Sampling
Starting point of regular grid is random
Forest
26
(No Transcript)
27
What is representative?

• Sampling strategies

Stratified Random Sampling
Each stratum is proportionally in the sample
Forest
28
What is representative?

• Stratified Random Sampling

29
What is representative?
30
Summary

• Good Practice Guidance is to
• Calculate and report uncertainty
• Try to improve large uncertainty by-take more
  samples-stratify the sampling-check potential
  sources of error (for example accuracy of used
  remote sensing image)