Semi‐supervised classification of vegetation: preserving the good old units and searching for new ones

Tichý, Lubomír; Chytrý, Milan; Botta‐Dukát, Zoltán

doi:10.1111/jvs.12193

Cited by 46 publications

(64 citation statements)

References 44 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As such, one would be reluctant to develop a quantitatively based national classification until there was comprehensive coverage of vegetation plot data (the term 'plot' is used here in a broad sense to incorporate both quadrat and transect data) collected in a consistent way. A framework has recently been developed, termed 'semisupervised clustering' (Tichý et al 2014), that allows new data to be incorporated into a pre-existing classification, while retaining types defined in the original classification (De Cáceres et al 2010). For New Zealand's woody vegetation, adopting this framework allowed the 17 vegetation alliances described by Wiser et al (2011) to be retained and also related to a finer thematic level when new data were analysed, and 12 new alliances and 79 associations to be defined (Wiser & De Cáceres 2013; the names 'alliance' and 'association' follow usage in Europe and North America; Peet & Roberts 2013).…”

Section: Introductionmentioning

confidence: 99%

Expanding an existing classification of New Zealand vegetation to include non-forested vegetation

Wiser¹,

Thomson²,

Cáceres³

2016

NZ J Ecol

View full text Add to dashboard Cite

Abstract:We produced the first national-scale quantitative classification of non-forest vegetation types, including shrubland, based on vegetation plot data from the National Vegetation Survey Databank. Semisupervised clustering with the fuzzy classification algorithm Noise Clustering was used to incorporate these new data into a pre-existing quantitative classification of New Zealand's woody vegetation. Fuzzy classification allows plots to be designated as transitional when they are similar to multiple vegetation types; the Noise Clustering algorithm allows plots having unique composition to be designated as outliers. We combined plot data collected using two different methods by transforming abundances to relative ranks and showed our classification results were robust to this. Of the 6362 plots analysed, 505 were assigned to previously defined woody vegetation types. Using the remaining 5857 plots, we defined vegetation types at two hierarchical levels comprising 25 alliances and 56 associations. Ten of the alliances are tussocklands, six are grasslands, four are stonefields or gravelfields, two are herbfields, one is rushland, and two are newly defined woody alliances. The classification defined compositional differences among well-known widely distributed short and tall tussock grasslands of the South Island. Notably it distinguished Chionochloa pallens, C. crassiuscula and C. oreophila tussocklands in wetter western regions from those dominated by C. rigida and C. macra in the east, and the domination of eastern South Island short tussock grasslands by Festuca novae-zelandiae and Poa colensoi. We demonstrate the distinctiveness of the vegetation of four naturally uncommon ecosystems -coastal turfs, northern gumlands, granite sand plains and braided riverbeds. Insufficient plot data precluded the definition of North Island Chionochloa rubra grassland types and many wetland and coastal communities. The 1846 plots designated as outliers mainly occur on warmer, wetter and less invaded sites than classified plots. Semi-supervised clustering allowed us to progress the development of an extendable, plot-based, quantitative classification of all New Zealand's vegetation despite data gaps.

show abstract

Section: Introductionmentioning

confidence: 99%

Expanding an existing classification of New Zealand vegetation to include non-forested vegetation

Wiser¹,

Thomson²,

Cáceres³

2016

NZ J Ecol

View full text Add to dashboard Cite

show abstract

“…We performed a semi‐supervised K‐means classification (Tichý et al. ). This algorithm uses a priori information about group membership for some plots to define centroids of clusters representing previously established vegetation units.…”

Section: Methodsmentioning

confidence: 99%

Diversity of lowland hay meadows and pastures in Western and Central Europe

Rodríguez-Rojo

Jíménez-Alfaro

Jandt

et al. 2017

Applied Vegetation Science

Self Cite

View full text Add to dashboard Cite

Questions Which are the main vegetation types of lowland hay meadows and pastures in Western and Central Europe? What are the main environmental gradients that drive patterns of species composition? Is it possible to classify these grasslands to phytosociological alliances that reflect management practices? Location Western and Central Europe (excluding the Alps and Carpathians). Methods A database of 21 400 vegetation plots of mesic grasslands across Western and Central Europe was compiled. After geographically stratified resampling, semi‐supervised classification based on the K‐means algorithm was applied to assign a subset of plots into 32 a priori association‐level vegetation types and to search for new types within the subset of non‐assigned plots. The vegetation plots assigned into the final vegetation types were submitted to another K‐means classification to show the grouping into higher‐level vegetation types. Results A total of 36 associations were distinguished in the resampled subset of 8277 vegetation plots and were grouped into four large groups: (1) eutrophic and intensively managed hay meadows and permanent pastures; (2) nutrient‐rich grasslands developed from recently abandoned fields or managed under irregular practices of mowing and manuring; (3) non‐eutrophic lowland and submontane hay meadows; (4) extensively managed pastures and Atlantic grazed hay meadows. A PCoA of the associations of these four groups showed that extensively managed pastures were floristically more similar to non‐eutrophic hay meadows than to permanent intensively managed pastures, which was more obvious in the Atlantic region than in Central Europe. Species composition of the lowland hay meadows was clearly differentiated according to biogeographic sectors. Other floristic differences were related to climate, altitude, soil base status and topography. Conclusions This analysis challenges the traditional concept of mesic grassland alliances separating hay meadows from pastures. New classification should be based mainly on the differences in management intensity rather than in management practice. Consequently, nutrient‐poor extensive pastures, which currently are not considered in the European Habitats Directive, should receive the same conservation attention as low‐intensive hay meadows, because both types of vegetation can be equally species‐rich and do not differ substantially in floristic composition from each other.

show abstract

“…Penalized information criteria have also recently been shown to have good properties for choosing between competing vegetation classification solutions (Lyons et al., ). Other internal metrics related to species composition are also an option (Tichý et al., ), but we opted to stick with a model‐based approach. Uncertainties in parameter estimates and predictions were calculated via bootstrapping using the Bayesian bootstrap (Rubin, ).…”

Section: Methodsmentioning

confidence: 99%

Simultaneous vegetation classification and mapping at large spatial scales

Lyons

Foster

Keith

2017

Journal of Biogeography

View full text Add to dashboard Cite

Aim Multivariate mixture models offer the ability to streamline the typically multi‐stage process of vegetation classification and mapping into a single, simultaneous analytical step. Our aim is to demonstrate the model's utility over large and diverse areas, with comparison to existing classification and mapping conventions. Location New South Wales, Australia. Methods We demonstrate a statistical model that uses both multivariate species data and environmental covariate data to simultaneously classify observations into groups and predict those groups out into environmental and geographical space. We used two large data sets to demonstrate the method: an ~810,000 km2 area with 4,715 sites and 488 species and a ~220,000 km2 area with 5,183 sites and 446 species. A range of topographic, terrain, climate and soil predictors were used as environmental variables, including future projected climate variables. Models can be fit with the R package “RCPmod.” Results The method results in probabilistic memberships to vegetation assemblages, for both the classification and map. There was general agreement between our approach and existing vegetation classification conventions, but we explore some notable differences. We were able to examine the environmental gradients that define the predicted vegetation distributions, as well as make predictions about how the distribution of species assemblages might shift as a result of climate change. Main conclusions Our approach tightens the link between description of biodiversity patterns and their depiction in space, by considering both biotic and spatially explicit abiotic information simultaneously. The method allows uncertainty to be quantified objectively across a consistent set of groups for both vegetation classification and mapping, which is rarely the case in traditional multi‐stage approaches. A simultaneous modelling approach increases capacity to make predictions into varying spatial, temporal and environmental dimensions, providing new ecological insights and increasing capacity for evidence‐based decision making.

show abstract

Semi‐supervised classification of vegetation: preserving the good old units and searching for new ones

Cited by 46 publications

References 44 publications

Expanding an existing classification of New Zealand vegetation to include non-forested vegetation

Expanding an existing classification of New Zealand vegetation to include non-forested vegetation

Diversity of lowland hay meadows and pastures in Western and Central Europe

Simultaneous vegetation classification and mapping at large spatial scales

Contact Info

Product

Resources

About