HAL Id: hal-02804111
https://hal.inrae.fr/hal-02804111
Submitted on 5 Jun 2020
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of sci-entific research documents, whether they are pub-lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Sampling strategy and stratification
Nicolas Picard, Laurent Saint-André, Matieu Henry
To cite this version:
Nicolas Picard, Laurent Saint-André, Matieu Henry. Sampling strategy and stratification. Training Material Sampling Strategy, UN-Reducing Emissions from Deforestation and forest Degradation (UN-REDD). Genève, CHE.; Food and Agriculture Organization (FAO). ITA., Jun 2012, Hanoi, Vietnam. 22 p. �hal-02804111�
Sampling Strategy and
Stratification
Pham Cuong, Inoguchi Akiko
Hanoi, June 18 - 22
th2012
Authors : N. Picard (CIRAD), L. Saint-André
(CIRAD - INRA), and M. Henry (FAO)
Context
Forests are basically heterogonous, in tree size, tree species,
ecological growth conditions etc…
So which sampling strategy
should we adopt ?
Turn heterogeneity into
Homogeneity
Global rules: stratify the variability ! (to get homogeneous entities on
which measurements will be done)
For example, at the scale of a country
With the objective of quantifying C stocks,
Then the stratification should be made according to :
-ecological regions,
-site index within each region, -age within each site index class,
-and stand density within each age class…..
For example, at the scale of an eddy-correlation site
With the objective of comparing NEP estimates (from the flux tower and from
destructive measurements)
Then the stratification should be done according to :
-the age of the stand,
-the season within each studied year, -the foot print of the flux tower….
Turn heterogeneity into
Homogeneity
Global rules : Drawing of the sampling chain highlighting the objectives of
the study
For example:
Country Ecological situations Stands Measured Trees AllometricEquation
Volume/Biomass of a single tree
Volume/Biomass of the average tree Volume/Biomass of the selected stand
Volume/Biomass of an other stand ?
Prediction
This step is then objective-dependent
and the availability of time, staff and money (!)
Practically, it is the result of two constraints:
the wanted accuracy
Turn heterogeneity into
Homogeneity
More specifically, the optimum sampling design can not be reached
for volume and biomass equations …. The choice for the appropriate
equation is generally done “a posteriori”
Sampling strategy
Some theoretical considerations, case study relating tree volume to D
2H
This model makes the hypothesis that the
relationship between tree volume and D
2H is
linear and that the error distribution
ε
follows a
normal law of mean 0 and variance
σ
2It can be fitted using n trees on which volume and D
2H were measured –
This sample also gives
The volume prediction for one single tree (knowing D* and H*) is then given by :
the average of D
2H on the sample
Estimated volume
Sampling strategy
Some theoretical considerations, case study relating tree volume to D
2H
If now, we want to assess how many trees (n) we should sample to have a
precision of E on this tree (D* and H* are known)
Making the choice of sampling randomly the trees into the stand, then
With
µ
=
and
τ
the standard deviation of D2H on the sample
Sampling strategy
Some theoretical considerations, case study relating tree volume to D
2H
Taking E=5%, a tree volume of 2m
3,
µ
=5m
3and
τ
=1m
3……
n=98 trees
Is there a way to lower this number of trees, ie, to optimize the sampling ?
Minimize this value (but
difficult to do in practice
on the field
Maximize this value =
sampling the
extreme trees….
Sampling strategy
Some theoretical considerations, case study relating tree volume to D
2H
BUT, strong drawbacks if……
Sampling strategy
Practical applications: biomass equations for an eucalyptus clone in
Congo
Reference: Laclau 1996
30 sampled trees
Models fitted with 6, 12, 18, 24 trees equally spread out in all classes of tree basal area, 20 repetitions for each fitting
R2 and biomass estimates variations decrease with the number of sampled trees. 12 to 18 trees is a good
compromise between time consuming and the wanted accuracy.
Sampling strategy
Practical applications: generic biomass equations for eucalyptus in
Congo
Stand selection :
Time from savannahafforestation Savannah
0 11 19 25 years
High Forest C1 Coppice C1 High Forest C2
Plantations Plantations
Cycle N°1 Cycle N°2
G1 G2 G3A G3B G3C G3D Time from savannah
afforestation Savannah
0 11 19 25 years
High Forest C1 Coppice C1 High Forest C2
Plantations Plantations
Cycle N°1 Cycle N°2
G1 G2 G3A G3B G3C G3D
Tree selection : 4 to 12 trees per stand according to their basal area
Reference: Saint-André et al. 2005
Field Constraint
Sampling strategy
Practical applications : volume tables for the main species in France
Reference: Vallet et al, 2006
Eawy
Camp Cusson Côte aux Hêtraux
Camp Souverain Camp Cusson 2 Camp Cusson 1 1929 1934 1949 Retz
Forêt Peuplement Placette Eclaircie
… …
… …
…
Forêt de provenance des échantillons Forêt Nonbre de Peuplement Nombre de Jeux Bellême 3 12 Blois 4 10 Champenoux 2 7 Haye 1 1 Tronçais 5 12 C h ê n e TOTAL 15 42 Barres 2 2 St Just d'Avray 1 5 D o u g la s TOTAL 3 7 Clusaz 1 2 Doubs 1 1 Noirmont 1 2 Risol 1 2 Sixt 1 1 E p ic éa TOTAL 5 8
Sampling strategy
Some Indications (Pardé et Bouchon, 1988):
Mono-specific and even-aged stands : 30 trees
A 15ha group of stand : 100 trees
Forest of 1000ha : 400 trees
region scale : 800 trees
Ecological range of a given species : 2000 to 3000 trees
Chave et al. 2004:
One tropical rain forest in Panama: 300 trees
Sampling strategy
Selecting trees on the field
Tree selection after a first inventory
It is recommended to use the tree-basal-area bar charts (equal width of basal area classes) because
the largest trees contribute more to the total stand volume than the smallest ones (if volumes are normally distributed)
and because the variance of volume, biomass or nutrient content increase with the tree size
It is then better to sample the largest trees to catch this variability
Note: for multiple-stem trees, use the sum of stem basal area to get the average diameter of the
trees; in some particular cases (ex: dry zones) it is better to use the base of the trunk than the breast height 0 10 20 30 40 50 60 70 80 90 100 0 20 40 60 80 100
% of trees (ranked by tree size)
% o f to ta l st an d v o lu m e The 20% largest trees provide 50% total stand volume
Sampling strategy
Selecting trees on the field
Same number of trees by basal area classes ? Or proportionality to the number of trees within each class ?
This is the usual case when the objective is to get a robust model
(ex: one equation for several ages):
Extreme trees have to be sampled to act upon the building and the fitting of the model
Same number of trees
This is the usual case when the objective is to get an equation
devoted to one particular purpose (ex: tree volume at age 7years)
Proportionality to the number of trees within each class
Practically, we often use a mixture of these two options
Base with the same number of trees by classes + complementarySampling strategy
Selecting trees on the field
Case of non-homogeneous stands
Here again, the sampling to be applied is a function of the wanted accuracy and the available means
Frequency (% of trees)
Classes of basal area
Specie 1
Specie 2
-add a new level of stratification ?
-how many trees to be felled for each specie ?
If so
Equal ?
Proportional ?
if no
The total histogramSampling strategy
Selecting trees on the field
Case of non-homogeneous stands
Example: coppices of Quercus Ilex in Morocco
One stand, 20 trees to be felled
14% of multiple-stem trees
86% of single-stem trees
17 trees
3 trees
Histograms of tree basal areas
0 0.05 0.1 0.15 0.2 0.25 0.3 0 100 200 300 400 500 600 700
Tree basal area (cm2)
F re q u e n cy ( % o f tr ee s) Multiple-stem Single-stem
Sampling strategy
Selecting trees on the field
It is also preferable to avoid particular trees (top broken, sinuous shapes, etc…)
EXCEPT if these trees represent a big proportion of the stand, or if the objective is to quantify these defects
Lastly it is necessary to avoid trees located near the roads or near the forest limits
Sometimes, field constraints change drastically the
sampling designed on the computer
It is preferable to select trees within plot that satisfy two main criteria: no high rates of mortality to ensure an homogeneous and representative tree growth
Sampling strategy
Estimation of the overall stand volume or biomass
Once the biomass equation is built, the question of the sampling strategy for forest inventory remains
with and N the total number of trees measured in the plot It can be shown that the previous equation can be rewritten as follows
with e denoting the values calculated on the sample used to calibrate the equation
And the confidence interval of total volume is given by
Sampling strategy
Estimation of the overall stand volume or biomass
Without stratification, the variance of total volume is given by
With stratification, the variance of total volume is given by
Where h denotes the strata, Nh the total number of trees measured in the strata, nh the number of trees felled in the strata for building the allometric equation
Sampling strategy
Estimation of the overall stand volume or biomass
Size of the plot to be inventoried ? Smaller is the plot= smaller is the representativity in natural forests 0 2 4 6 8 10 12 14 16 18 20 225 325 425 525 625 0 10 20 30 40 50 60 50 250 450 650 850 1250 0 100 200 300 400 500 600 125 375 750 1250 1750 2250 2750 4750 AGB (Mg ha-1) Plot size = 20 x 20 m
Plot size = 10 x 10m Plot size = 100 x 100m
Chave, J., R. Condit, et al. (2003). "Spatial and temporal variation of biomass in a tropical forest: results from a large census plot in Panama." Journal of Ecology 91: 240–252.
Henry, M., S. Adu-Bredu, et al. (submitted). "Structure and functioning of rain forests ecosystems in Ghana " International Journal of forestry Research.
Sampling strategy
Estimation of the overall stand volume or biomass
Size of the plot to be inventoried ?
In general, for a given sampling effort :
Number of plots
Plot area
Zone area
If the stand is heterogeneous, it is better to install numerous small plots; while in homogeneous stands, it is better to install few large plots