Survey about Plot-based Vegetation Classification Methods

Why a survey about classification methods? 
Vegetation classification is an exercise of abstraction from observed vegetation patterns to vegetation units. This kind of exercise is artificial in the sense that there is not a unique (or natural) way of classifying plant communities. The purpose of this survey is to determine what are the most usual aims and approaches followed in plot-based vegetation classification exercises around the world.  

Who is gathering this information?
This questionnaire has been designed by Miquel De Cáceres (Centre Tecnològic Forestal de Catalunya, Spain) and Milan Chytrý (Masaryk University, Czech Republic) on behalf of the Vegetation Classification Committee of the International Association for Vegetation Science (http://www.iavs.org).

What will we do with the results? 
The Vegetation Classification Committee is planning to organize a one-day workshop around this topic in 2013 and your responses may be extremely valuable to obtain a range of opinions on vegetation classification that can be discussed at this workshop.  Please contact Miquel De Cáceres (miquelcaceres@gmail.com) if you have questions or comments about the survey. 

The period to answer the questionnaire has expired. Thank you for your collaboration!


Summary of responses

Number of respondants: 236

Geographical origin of respondants

Country
# %
Algeria 1 0.42
Argentina 2 0.85
Australia 12 5.08
Austria 3 1.27
Belgium 1 0.42
Bosnia-Hercegovina 2 0.85
Canada 9 3.81
Croatia 1 0.42
Czech Republic 6 2.54
Estonia 2 0.85
Finland 4 1.69
France 13 5.51
Georgia 1 0.42
Germany 31 13.1
Greece 2 0.85
Hungary 5 2.12
India 1 0.42
Iran 1 0.42
Ireland 4 1.69
Israel 1 0.42
Italy 22 9.32
Japan 3 1.27
Latvia 1 0.42
Mexico 5 2.12
Namibia 1 0.42
Netherlands 7 2.97
New Zealand 3 1.27
Nigeria 1 0.42
Norway 4 1.69
Pakistan 1 0.42
Panama 1 0.42
Poland 1 0.42
Portugal 2 0.85
Puerto Rico 1 0.42
Romania 2 0.85
Russia 5 2.12
Russian 1 0.42
Scotland 1 0.42
Slovak Republic 2 0.85
Slovenia 1 0.42
South Africa 3 1.27
South Korea 1 0.42
Spain 10 4.24
Sweden 2 0.85
Switzerland 5 2.12
Turkey 2 0.85
UK 7 2.97
Ukraine 1 0.42
USA 38 16.1

Continent

Europe14963%
Asia115%
North America5222%
South America31%
Africa63%
Australia156%
Antarctica00%


Experience with plot-based classification of vegetation

Number of studies where you have used numerical classification techniques (approximate)


1 to 59641%
5 to 104820%
More than 109239%

Scale of the study area in your classification exercises


Local (few square kms)16068%
Regional (one or several)17775%
Whole country7431%
Several countries4820%
Other63%
People could select more than one checkbox, so percentages may add up to more than 100%.
Continent(s) where you have done this research


Europe15365%
Asia4318%
North America6528%
South America2611%
Africa2611%
Australia219%
Antarctica00%
People could select more than one checkbox, so percentages may add up to more than 100%.


Purpose and general criteria for vegetation classification

What are the main purposes of vegetation classification based on plot records?

Vegetation mapping15167%
Describe existing biodiversity in a region/country18482%
Determine habitats with special conservation status11049%
Other6027%
People could select more than one checkbox, so percentages may add up to more than 100%.
What are the most important levels of abstraction that can be derived from plot data?

Low-level units based on floristics (e.g., associations, alliances)22596%
High-level units based on floristics (e.g., orders, classes)8938%
High-level units based on physiognomy (e.g. broad formations)4419%
High-level units based on physiognomy and floristics6729%
Physiognomico-climatic units applicable worldwide3013%
Functional classes7934%
Other2611%
People could select more than one checkbox, so percentages may add up to more than 100%.
What are the most elemental 'objects' in vegetation classification?

Complete vegetation stands19589%
Specific strata within the stand (i.e. synusiae)4621%
Other2712%
People could select more than one checkbox, so percentages may add up to more than 100%.


Sampling design and/or selection of plots from vegetation databases

If you work in the field, what strategy of plot selection do you use in field sampling?

Preferential sampling11753%
Simple random sampling5424%
Systematic sampling7032%
Stratified random sampling13260%
Other2110%
People could select more than one checkbox, so percentages may add up to more than 100%.

If you work in the field, please describe briefly the most rellevant details of your approach to select sampling locations.




























No answer (including the previous question) 9 4%
Preferential sampling after preliminary survey of the study area (or unspecified preferential) 68 29%
Preferential sampling after inspection of aerial photographs and remote sensing images 8 3%
Combination of stratification and preferential sampling within each stratum 30 13%
Preferential sampling with information from previous surveys 9 4%
Simple random sampling 36 15%
Stratified random sampling (unspecified stratum definition) 34 14%
Stratification using geological and/or soil data 31 13%
Stratification using topographic and/or climatic conditions 51 22%
Stratification using current or historic land use, habitat or physiognomy classes 86 36%
Stratification using disturbance regime or successional stage 6 3%
Geographical stratification (or unspecified systematic sampling) 53 22%
Gradient oriented transects (gradsects) 3 1%
Purposedly include transitional plots 2 1%
Include opportunistic plots for discovered diversity 3 1%
Other sampling schemes 4 2%
All (depending on the study objective) 7 3%

This was a free text question where people could indicate more than one method, so percentages do not add up to 100%.

If you work with vegetation-plot databases, what strategy of plot selection from the database do you use?

Based on dominance of selected species9347%
Based on occurrence of groups of diagnostic species10151%
Based on plot record assignment to units of a previous classifications7840%
Based on the year of sampling4824%
Based on the season of sampling168%
Based on the geographical location11156%
Other3920%
People could select more than one checkbox, so percentages may add up to more than 100%.
If you work with vegetation-plot databases, do you use some kind of stratified resampling before data analysis?

Based on geographical location of plots (e.g., using a geographical grid)9055%
Based on environmental characteristics of plots9357%
Based on species composition of plots8653%
Other2616%
People could select more than one checkbox, so percentages may add up to more than 100%.


Homogenization of plots of mixed provenance

When merging vegetation plot data coming from different authors/surveys/databases, do you check...

(1) taxonomic nomenclature (e.g., spelling errors, synonyms, ...) ?18790%
(2) differences in taxonomic resolution?15072%
(3) differences in plot sizes?14570%
(4) differences in measurement scales/units?14068%
(5) differences in the definition of vegetation strata (tiers)?7838%
Other3115%
People could select more than one checkbox, so percentages may add up to more than 100%.
If you marked (1) or (2) in the first question, please explain briefly how do you normally deal with taxonomic issues

1. Problems with taxonomic nomenclature (among 193 answers to the previous question)



















Automated correction using standardized species lists or databases 38 20%
Manual correction using authoritarive source (published floras or checklists) 97 50%
Manual correction using home-made synonimy list 7 4%
Consultation with experts 11 6%
Take vouchers of speciments in the field 2 1%
Compare records of other data sets to detect changes in nomenclature 2 1%
Keep original names 1 1%
Use aggregated taxon concepts when necessary 16 8%
Unclear answer 8 4%
No answer 40 21%

This was a free text question where people could indicate more than one method, so percentages do not add up to 100%.


2. Problems with taxonomic resolution (among 152 answers to the previous question)

Merge to broader taxonomic level (e.g. from subspecies to  species) 52 34%
Assign low-level taxa a posteriori (e.g. adding subspecies after checking with surveyers) 5 3%
Remove plant records with low taxonomic resolution (e.g. genera) (or remove plot records) 8 5%
Allow different levels of resolution for different taxa 8 5%
Unclear answer 9 6%
No answer 78 51%

This was a free text question where people could indicate more than one method, so percentages do not add up to 100%.


If you marked (3), (4) or (5) in the first question, please explain briefly how do you normally deal with problems of heterogeneity in field methods.

3. Differences in plot sizes (among 150 answers to the previous question)

Only comparable plot sizes 22 15%
Exclude extreme plot sizes 50 33%
Use species area curves to quantify bias 4 3%
Use nested plots or sub-sampling 5 3%
Trust robustness to different plot sizes 10 7%
Unclear answer 30 20%
No answer 36 24%
 
This was a free text question where people could indicate more than one method, so percentages do not add up to 100%.



4. Differences in measurement scales/units (among 147 answers to the previous question)



Only comparable measurement scales 11 7%
Transform to coarser resolution (e.g. to presence-absence) 25 17%
Transform to a single metric (cover scale or percentage cover) 45 31%
Unclear answer 27 18%
No answer 37 25%

This was a free text question where people could indicate more than one method, so percentages do not add up to 100%.


5. Differences in the definition of vegetation strata (among 87 answers to the previous question)



Only comparable strata 7 8%
Aggregate values across strata (e.g. using maximum value) 19 22%
Keep some strata as separate (e.g. cryptogam layers) 5 6%
Convert to common structure using species or plant atributes 3 3%
Convert to common structure using strata definition 4 5%
Unclear answer 18 21%
No answer 28 32%


This was a free text question where people could indicate more than one method, so percentages do not add up to 100%.






Defining ressemblance between stands (or strata within stands)

What are the vegetation attributes that you normally use for vegetation classification?

Floristic composition (with presence/absences only)12154%
Floristic composition (with abundance/importance values)20792%
Vertical structure of the community (e.g. cover values for each strata)8136%
Morphological or functional attributes of the species6328%
Other42%
People could select more than one checkbox, so percentages may add up to more than 100%.
What kind of scale(s) of measurement do you normally use for these attributes?

Measurements for plant abundance



Presence-absence 11 4,7%
Estimated percentage cover 70 29,7%
Cover (or cover-abundance) classes 149 63,1%
Basal area 3 1,3%
Plant biomass 3 1,3%
Counts or density 19 8,1%
Density classes 2 0,8%
Point intercept (quadrat/transect) 8 3,4%
Relative abundances 2 0,8%
This was a free text question where people could indicate more than one method, so percentages do not add up to 100%.






What kind of data-transformation and/or resemblance measure do you normally use?

Data transformations

van der Maarel 1979 6 2,5%
Square root 46 19,5%
Cubic root 1 0,4%
Logarithmic 30 12,7%
Arcsin 6 2,5%
Species maximum 3 1,3%
Plot maximum 1 0,4%
Total abundance 2 0,8%
Double Wisconsin 1 0,4%
Floating cut levels 1 0,4%
Pseudo-species 1 0,4%
This was a free text question where people could indicate more than one method, so percentages do not add up to 100%.




Ressemblance measures (some include implicit transformations)

Euclidean distance 14 5,9%
Manhattan distance 1 0,4%
Hellinger or Chord distances 14 5,9%
Bray-Curtis (Percentage difference) 76 32,2%
Similarity Ratio 2 0,8%
Kulczynski 1 0,4%
Morisita 1 0,4%
Gower 1 0,4%
Czekanowski Mean Character Diff. 2 0,8%
Chi-Square 6 2,5%
Ochiai for P/A 2 0,8%
Jaccard/Sorensen for P/A 26 11,0%
Probabilistic indices (e.g. Goodall) 1 0,4%

This was a free text question where people could indicate more than one method, so percentages
 do not add up to 100%.





Unsupervised classification methods

Indicate your preferred hierarchical or non-hierarchical classification method(s)


Manual sorting 7 3,0%
Ordination (PCA/DCA/CCA/NMDS…) 23 9,7%
TWINSPAN/Modified TWINSPAN 76 32,2%
Non-hierarchical (unspecified) 5 2,1%
Partitioning K-means/OPTPART 11 4,7%
TABORD 1 0,4%
PAM/ISOPAM 6 2,5%
Minimum spanning tree 1 0,4%
Fuzzy clustering 4 1,7%
Hierarchical agglomerative (unspecified) 39 16,5%
Beta-flexible 16 6,8%
Ward's 24 10,2%
Complete/Proportional linkage 4 1,7%
UPGMA / UPGMC 15 6,4%

This was a free text question where people could indicate more than one method, so percentages do not add up to 100%.



How do you deal with vegetation units previously-defined in your study area?



No answer/not applicable 57 24%
Build classification independently (perhaps comparing afterwards) 75 32%
Try to follow previous classifications (only create new ones if necessary) 72 31%
Follow classifications from other areas or at broader level of abstraction 15 6%
Unclear answer/Other 28 12%

This was a free text question where people could indicate more than one method, so percentages do not add up to 100%.



What are the most important criteria to validate the resulting vegetation units ?


Existence of indicator species15269%
Environmental distinction15470%
Robustness of solution to sampling variations7032%
Other3114%
People could select more than one checkbox, so percentages may add up to more than 100%.


Indicate what kind of statistical techniques you prefer to validate vegetation units


No answer/only expert review 107 45%
Evaluation of mapping results 6 3%
Discriminant analysis 30 13%
Indicator species analyses 27 11%
Unconstrained ordination (DCA, PCA, NMDS,...) 41 17%
MRPP test 10 4%
Non-linear regression models with environmental variables 5 2%
Linear models (including ANOVA and MANOVA) with environmental variables 11 5%
Constrained ordination with environmental variables 24 10%
Isolation (Silhouettes, internal vs. external variation) or crispness of result 8 3%
Ellenberg indicator values 1 0%
Analysis of Concentration 1 0%
Stability to sampling (e.g., bootstrap or cross validation) 3 1%
Consensus of classification methods 1 0%
Other 2 1%
Unclear answer 13 6%

What aspects are the most important to characterize vegetation units before publication?


Species composition (e.g. constancy values)20694%
Environmental ranges16073%
Management (past or current)9242%
Amount of invasive species188%
Diagnostic species12959%
List of prototypical plot records (e.g. typical relevés)4219%
Other2411%
People could select more than one checkbox, so percentages may add up to more than 100%.





Supervised classification methods

Indicate your preferred supervised classification method(s) (e.g., neural networks).



Plot matching to vegetation units using similarity measures 5 2,1%
K-means, PAM and their fuzzy variants 5 2,1%
Decision trees 5 2,1%
Neural networks 4 1,7%
Cocktail 10 4,2%
Assessment of diagnostic species 2 0,8%
Other 3 1,3%
Only for spatial modelling (GIS modelling, remote sensing) 8 3,4%
Unclear reply 13 5,5%
No experience/no reply 183 76,9%









Indicate your experience with on-line expert systems for vegetation classification



Developing/developed one 11 4,7%
Using one 8 3,4%
Unclear reply 7 3,0%
Limited experience 2 0,9%
No experience/no reply 207 88,1%