A review of the use of plant community data for assessing ecological quality

Abstract

• The use of community data for the assessment of quality is best developed in freshwater ecology. Methods such as RIVPACs use statistical relationships between environmental covariables and classifications of reference site communities to assess the relative quality of new sites.
• Although such ‘RIVPAC-like’ approaches can be conceived of for plant communities, existing sets of data most likely to be appropriate for defining reference sets (e.g. the NVC relevés) lack corresponding sets of environmental data. Environmental data could be extrapolated from extant high quality stands or remote-sensing data, but such approaches seem likely to lead to lower discriminatory ability as compared to actual co-located data. Other options include the 2007 Countryside Survey, which collected co-located environmental data but may not be representative of SSSI-quality habitat features, and the NPMS, which was purposefully biased towards higher quality sites, but which has not so far collected environmental data likely to be of use in a RIVPAC-like approach.
• Alternative approaches to assessing quality involve community data in isolation. That is, newly collected plant community data could be compared to reference sets denoting high and low ecological quality for a given habitat through multivariate methods such as classification and ordination.
• Challenges for this approach include: the selection of the reference samples to be used for any given habitat feature (potentially also taking into account regional variation); for ordination approaches, the number of axes to be assessed as part of a quality assessment, and how movement along these axes is assessed; and, for classification, the choice of clustering method to be used. ‘Fuzzy clustering’ in particular is increasingly used to derive measures of fit of new vegetation samples to existing reference classifications.

Full text

V1.4 3rd October 2016
A review of the use of plant community data for assessing ecological quality
O.L. Pescott & D.B. Roy, CEH Wallingford
Summary
 The use of community data for the assessment of quality is best developed in freshwater
ecology. Methods such as RIVPACs use statistical relationships between environmental
covariables and classifications of reference site communities to assess the relative quality of
new sites.
 Although such ‘RIVPAC-like’ approaches can be conceived of for plant communities, existing
sets of data most likely to be appropriate for defining reference sets (e.g. the NVC relevés)
lack corresponding sets of environmental data. Environmental data could be extrapolated
from extant high quality stands or remote-sensing data, but such approaches seem likely to
lead to lower discriminatory ability as compared to actual co-located data. Other options
include the 2007 Countryside Survey, which collected co-located environmental data but
may not be representative of SSSI-quality habitat features, and the NPMS, which was
purposefully biased towards higher quality sites, but which has not so far collected
environmental data likely to be of use in a RIVPAC-like approach.
 Alternative approaches to assessing quality involve community data in isolation. That is,
newly collected plant community data could be compared to reference sets denoting high
and low ecological quality for a given habitat through multivariate methods such as
classification and ordination.
 Challenges for this approach include: the selection of the reference samples to be used for
any given habitat feature (potentially also taking into account regional variation); for
ordination approaches, the number of axes to be assessed as part of a quality assessment,
and how movement along these axes is assessed; and, for classification, the choice of
clustering method to be used. ‘Fuzzy clustering’ in particular is increasingly used to derive
measures of fit of new vegetation samples to existing reference classifications.
Introduction
The assessment of the quality of ‘features’ on sites, whether habitat, species, or geological, is the
main purpose of the Commons Standards Monitoring (CSM) framework of the Joint Nature
Conservation Committee (JNCC 2004). JNCC (2004), quoting Brown (2000), define monitoring as “an
intermittent (regular or irregular) series of observations in time, carried out to show the extent of
compliance with a formulated standard or degree of deviation from an expected norm”; JNCC (2004)
also distinguish monitoring from surveillance, stating that “[surveillance] is repeated survey using a
standard methodology undertaken to provide a series of observations over time”. Note, however,
that the term “monitoring” is also used in a much more general sense throughout the ecological
literature (e.g. see Pescott et al. 2015 and references therein).
Monitoring then, in the context of this document, is intimately connected to some sort of clearly
defined reference state. The Common Standards approach to site monitoring is based upon
attributes (“characteristics of an interest feature that describe its condition, either directly or
indirectly”; JNCC, 2004) of habitats, selected species groups, or geological features
(http://jncc.defra.gov.uk/page-2199). The attributes of habitats chosen for CSM have been
developed by habitat experts and subsequently refined over time (Williams 2006). Within the CSM,
habitats are defined in terms of the UK National Vegetation Classification (Rodwell 1991 et seq.),
indeed, “[t]he National Vegetation Classification (NVC) is one of the key common standards
developed for the [UK’s] country nature conservation agencies” (http://jncc.defra.gov.uk/page-

V1.4 3rd October 2016
4259). Habitat attributes vary according to habitat type, but typically include things such as feature
extent, frequency or cover of indicator species, and physiognomic characteristics of vegetation (JNCC
2009). Full plot data, i.e. comprehensive quadrat data, are not normally required—historically this
approach has generally been considered to be too resource intensive to be implemented across all
Sites of Special Scientific Interest (SSSIs) containing a habitat feature, hence the current focus on
attributes, or indicators within CSM (Rowell 1993).
Aims
The aim of this document is to review an alternative but related approach to the assessment of site
quality, namely that of comparing newly collected plant community data to a reference set of
samples which themselves are taken as defining an agreed reference condition.
Existing methodologies for the estimation of reference states
‘Discrete’ approaches
Perhaps the best known ecological method for assessing quality based on a set of reference samples
is the River Invertebrate Prediction and Classification System (RIVPACS) approach (Clarke et al.
2003). RIVPACS proceeds by classifying a set of reference samples based on their macroinvertebrate
assemblages. These reference samples are “short river stretches, which are considered to be of high
ecological and chemical quality, and representative of the best examples of their particular river type
… carefully selected to encompass a wide range of physical types of running-water sites across a
geographical region” (Clarke et al. 2003). Hill’s Two-Way Indicator Species Analysis (TWINSPAN; Hill
1979) classification method (a hierarchical, divisive algorithm; Gauch 1982) has been the favoured
procedure for macroinvertebrate assemblage data to date in Britain (Moss et al. 1999; Moss 2000),
although similar implementations in Australia have used Unweighted Pair-Group Mean Averaging
(UPGMA) based on sample Bray-Curtis distance matrices (Clarke et al. 2003). Such approaches have
been termed ‘discrete’ due to their foundation in a set of discrete classes (Hermoso & Linke 2012).
Hermoso & Linke (2012) point out that these classes can either be formulated from the bottom up,
i.e. the community classes are based on assemblage data, as for the RIVPACS methodology described
here, or from the top down, where classes are erected using site environmental information only.
In RIVPACS, once a set of reference classes has been established multiple discriminant analysis
(MDA) is used to calculate the linear combination of environmental covariables that most efficiently
predict classification category membership; covariables are either fixed characteristics of sites (e.g.
altitude), or are measured at the time of community sampling (e.g. water chemistry; Clarke et al.
2003). The resulting equation can then be used to place new samples into the existing classes;
species data from the new samples can then be compared to the species data from the reference
samples to calculate a score of ‘ecological quality’ (Clarke et al. 2003).
‘Continuous’ approaches
Australian freshwater ecologists have recently developed an alternative to the RIVPACS approach, in
which site, rather than class, specific reference conditions are estimated (Linke et al. 2005). This
approach puts the continuous nature of variation in community composition to the fore, arguing
that reference conditions for a site are best estimated using a set of similar sites in terms of their
physical environment, irrespective of any a priori classification procedure. Hermoso & Linke (2012)
found this to be a superior method for estimating appropriate reference conditions for fish
assemblages of Iberian rivers. It is important to remember however, that the comparative
performance of discrete and continuous approaches will vary based on the amount of
environmental structuring of species assemblages in the real world; in those cases where variation in
communities is more highly structured by the environment, then discrete and continuous

V1.4 3rd October 2016
approaches are likely to give similar results. Ecologists should investigate the available options in the
context of regional biogeography and the likely structuring of species niches before deciding that
one approach is superior to another a priori.
For the UK, the north-west to south-east gradient in numerous correlated environmental variables
(e.g. rainfall, altitude, geology, land cover) create relatively strong structure at small scales (sensu
stricto), as demonstrated by numerous biogeographical studies of plants (Preston & Hill 1997;
Preston et al. 2013). At larger scales, i.e. regionally or locally, it is less obvious how strongly relatively
easily measured environmental covariables structure plant communities; although for many habitats
considerable information will exist on the relationship between the physical environment and plant
niches, we do not attempt to review this here.
Quality assessments for terrestrial plant communities
RIVPACS-like approaches
A comparable approach to RIVPACS for terrestrial ecosystems can easily be envisioned in theory, but
several challenges exist for any implementation. Historic vegetation community plot data (e.g. the
set of relevés used to construct the UK National Vegetation Classification) may form the basis for
defining sets of terrestrial reference conditions; however, for such datasets, originally collected for a
variety of research purposes, no standard set of measured environmental covariables is likely to
exist. The option to rely on remotely-sensed land-cover variables exists, but it seems unlikely, a
priori, that such variables will capture enough information to define appropriate reference
conditions, although modern remote sensing technologies suggest that such an avenue might be
profitably investigated. Even so, this option is unlikely to be available for historic samples, and
assumptions of equivalence between high-quality extant stands of vegetation and historic reference
samples would need to be made. The same assumption could also be made for newly collected field,
rather than remotely sensed, data. These approaches, however, could reduce the variation in
environmental conditions associated with a particular community reference class, potentially
reducing the classification accuracy of new samples.
Another challenge for an approach based entirely on the classification of sites from environmental
variables, regardless of whether that approach is based on a discrete or continuous model of
community variation, is the cost of collecting new environmental data for samples requiring
evaluation. In the case of terrestrial plant communities, this would be likely to require standard soil
measurements, such as pH, bulk carbon content, P, N, Ca, and other measurements to be made;
costs might be reduced if remotely sensed information could also be utilised. The most recent
Countryside Survey (Carey et al. 2008) collected data on most of these covariables, although the
stratified random design of this survey may not be representative of the high quality features
represented by the SSSI network. Alternatively, surveys which have sought to target higher quality
habitats (e.g. the National Plant Monitoring Scheme; Pescott et al. 2015) could be used as new
sampling frames for collecting environmental data. Note, however, that the effectiveness of this
approach would also depend on the representativeness of full survey plot data being collected
within the National Plant Monitoring Scheme (NPMS) relative to SSSI habitat features.
Community data only approaches
An approach based exclusively on community data, i.e. samples-by-species matrices, is one
alternative to the RIVPACS, or related, approaches. The community data-focused approach is well-
established in the literature for a variety of applications: species co-occurrence data is often used to
characterise environments, using the ecological theory that if most (or all) niches are filled, then the

V1.4 3rd October 2016
species present integrate important features of the environment, and therefore provide a strong
and reliable indication of the prevailing environment at a site (Goodall 1954; Gauch 1982 p. 136).
Fridley et al. (2007) use this approach to place species on a generalist-specialist gradient; Smart et al.
(2015) use a similar logic to derive measures of association between rare and common plants.
A community data approach to site quality assessment would rely upon:
(i) A set of community reference data, benchmarking the desired conception of high
ecological quality for all habitat features to be monitored; and, linked to this,
(ii) a clear conception of the relationship between any particular class of habitat feature to
be evaluated and the relevant subset of the reference data that will be used for the
evaluation.
(iii) A repeatable and robust methodology through which newly sampled stands may be
compared to the reference set. (Here, ‘robust’ is used in the sense of Gauch (1982), who
emphasises that a desirable characteristic of many multivariate methods used on
community data is the ability to obtain similar results given minor perturbations in
collected sample data.) The chosen method should also provide the ability to clearly
track change in stand quality through time.
These are discussed in turn below.
Community reference data and linking plant communities to reference sets (i, ii)
As quoted above, the UK NVC currently provides one of the key supports of the CSM system. An
obvious option therefore is to take advantage of the original NVC relevé set (Rodwell 2012) to
provide the reference data for quality assessment. The inclusion of other samples deemed to be high
enough quality could be decided by expert opinion, or through multivariate statistical approaches
(reviewed in Peet & Roberts 2013); however, even statistical approaches would likely require a
subjective decision on where to implement any particular cut-off for deciding whether the similarity
of sample x to reference subset y was close enough to merit inclusion in the high quality reference
set—parallel issues exist for evaluating condition (sensu CSM) with respect to the reference samples
(see iii below). Existing CSM habitat feature attributes may also be useful is these respects e.g. by
specifying certain cover or frequency thresholds for indicator species. The Centre for Ecology &
Hydrology are currently building a web portal for the hosting and collection of new and historic plot
samples (the ‘Plant Portal’)–chosen reference samples could, in theory, be tagged within this
database, making the reference set widely available and transparent.
In addition to the requirement for high quality reference samples, well defined low quality samples
would also be required, particularly if comparisons to high quality samples were to be carried out via
ordination approaches. The reason for this is that a multi-dimensional sample space defined only by
high quality reference samples is unlikely to be useful for ordinating low quality stands—ordination
is used for identifying important gradients in community data, for samples to be sensibly placed
along a gradient representing quality requires low, as well as high, quality reference points.
The main challenge for such an initiative would probably not be the accumulation of a nationwide
set of reference samples, but the related step of specifying which subset of samples were to be used
as the high and low quality reference sets for any stand of vegetation. Given that one of the main
principles of CSM was the creation of a clear set of criteria which could be applied consistently
across the UK (Rowell 1993), it will be seen that this decision is crucial for a consistent and
transparent evaluation of ecological quality (as defined in the context of CSM).

V1.4 3rd October 2016
To provide a brief example, the CSM guidance for upland habitats covers a wide range of plant
communities, including the habitat interest feature ‘blanket bog and valley bog (upland)’. This is
defined as covering the following NVC types (JNCC 2009): M1 Sphagnum auriculatum bog pool
community; M2 Sphagnum cuspidatum / recurvum bog pool community; M3 Eriophorum
angustifolium bog pool community; M17 Scirpus cespitosus – Eriophorum vaginatum blanket mire;
M18 Erica tetralix – Sphagnum papillosum raised and blanket mire; M19 Calluna vulgaris –
Eriophorum vaginatum blanket mire; M20 Eriophorum vaginatum blanket and raised mire; and M21
Narthecium ossifragum – Sphagnum papillosum valley mire.
The following NVC types are given as potentially indicating degraded blanket bog (where they occur
on peat deeper than 0.5 m): H9 Calluna vulgaris – Deschampsia flexuosa heath; H12 Calluna vulgaris
– Vaccinium myrtillus heath; M15 Scirpus cespitosus – Erica tetralix wet heath; M16 Erica tetralix –
Sphagnum compactum wet heath; M25 Molinia caerulea – Potentilla erecta mire; and, U6 Juncus
squarrosus – Festuca ovina grassland. Under the summary guidance table for the habitat interest
feature it is stated that “[w]here blanket bog communities are being replaced by either degraded
mire communities (M15, M16, M25), drier heath communities (H8, H12), or grassland type U6, and
where restoration back to blanket bog is considered to be feasible, then the degraded communities
should be assessed using the attributes and targets ascribed to blanket bog” (JNCC 2009). This
guidance implies that the use of relevés representing these high and low quality reference
communities would be appropriate for the ordination, or classification, of new stands of vegetation
for which ecological quality was to be assessed. However, this framing may be a somewhat naïve
view of the problem: amongst the reference NVC community types considerable floristic and
structural variation exists, ranging from bog pool communities (M1, M2, M3) to degraded, much
drier stands (e.g. H9, U6). It is therefore likely that reference ordinations of high and low quality
stands are likely to contain several significant environmental gradients—assessments of quality may
need to consider more than just the first two axes, or divisions, captured by multivariate statistical
techniques.
Potential methodological approaches (iii)
As noted above, any approach should be able to place a newly sampled stand in the context of a
priori high and low quality reference samples (or synoptic tables representing the composition of
these samples). The method should ideally also be able to clearly quantify change in a stand (i.e. a
habitat interest feature within the CSM program) through time.
Ordination
Ordination is a multivariate statistical tool for ordering m objects defined by n comparators in
multidimensional space. In vegetation science ordination is typically used to order vegetation
samples in two- or three-dimensional space defined by their floristic composition (Kent 2012). In the
context of CSM, ordination techniques could be used to produce an ordination of reference plots.
New samples requiring quality evaluation could then be projected onto these ordinations (for as
many axes as is desired); the projection of new samples onto existing ordinations, however, can only
take into account the influence of species contained within the original ordination: axes scores for
new samples are thus only based on species present in the reference plots. Note also that
researchers typically only display a subset of the extracted axes from any given ordination, typically
axis 1 versus axis 2 is displayed, although other axis combinations (e.g. axis 2 versus axis 3) are also
typically inspected. The selection of important axes can be based on objective criteria (Borcard et al.
2011), but expert judgement is typically also important.

V1.4 3rd October 2016
In theory then, new samples from habitat interest features can be projected onto existing reference
ordinations, with samples at different time points labelled appropriately. However, the results of this
approach are likely to become difficult to display with an increasing number of sampling points
(Poulin et al. 2013). In addition, although the trajectory of a vegetation stand relative to reference
samples over time may be assessed visually on any number of ordination axes, it is less clear how
useful the quantitative or graphical comparison of a sample’s axis score to the axis scores of a set of
high or low reference samples would be, particularly if the reference set axis scores exhibited high
variance on a given axis. For example, would the finding that a vegetation sample was moving closer
to the centre (e.g. the arithmetic mean) of a set of high quality reference samples on a particular axis
be a meaningful one if these samples displayed high variance and/or poor separation from low
quality reference samples on that axis? If such an approach was used, thorough investigation of the
environmental gradients represented by a given set of habitat reference samples would need to be
undertaken, in order that only ordination axes with interpretations relevant to environmental
gradients of interest were used for the assessment of quality. Ultimately this may be no harder than
the decisions that have already been made in specifying desirable attributes for habitat interest
features within the current CSM system, but this cannot be guaranteed a priori. We also note that
the assessment of multivariate distance could also be performed using an appropriate dissimilarity
metric, removing the need to assess individual axes separately (e.g. Smart, 2000).
Finally, we note here the existence of Principal Response Curves (van den Brink et al. 2008), a
multivariate technique invented for the purpose of efficiently extracting and displaying a measure of
temporal change within a community in comparison to a reference or experimental control. This
method proceeds on the basis of comparison to a single reference, which can either be monitored
once or repeatedly to produce a time series of samples (van den Brink et al. 2008). For
implementation within a CSM-like framework this would mean that a single reference site would
have to be chosen for every monitored stand of vegetation; this seems unlikely to be achievable
given the considerable effort implied by choosing a particular suitable reference sample for every
habitat interest feature to be monitored. The technique is more often used in spatially restricted
experimental or quasi-experimental settings (e.g. van den Brink et al. 2008; Poulin et al. 2013).
Fuzzy clustering
Clustering, or classification, techniques are also regularly used alongside ordination to aid in the
understanding of samples characterised by more dimensions (e.g. species) than can easily be
understood by the human mind (Gauch 1982; Kent 2012). Cluster analyses have traditionally worked
on the basis of assigning samples so that they are a member of one cluster only; such techniques are
typically referred to as ‘hard’ (De Cáceres & Wiser 2013). However, increasingly, ‘fuzzy’
classifications are being developed and used (De Cáceres et al. 2010; Wiser & De Cáceres 2013).
Approaches using fuzzy set theory allow for objects to be given probabilistic assignments to clusters
in multivariate space; typically these assignment scores are constrained to sum to 1 across clusters
(‘partitive clustering’; De Cáceres & Wiser 2013). Several different methods for fuzzy clustering have
been introduced to vegetation science, and have been shown to exhibit various strengths and
weaknesses in tests with real data (De Cáceres et al. 2010). Fuzzy clustering, as with other clustering
techniques, can be unsupervised, semi-supervised, or fully supervised. Fully supervised clustering
assigns samples to existing clusters; semi-supervised clustering allows for samples to be assigned to
existing clusters, but also allows new clusters to form if these are a better fit for some samples than
existing clusters; unsupervised clustering is simply clustering de novo, and starts with no predefined
clusters.

V1.4 3rd October 2016
One technique that has shown particular promise in the area of fuzzy clustering is ‘noise clustering’;
this is a semi-supervised fuzzy clustering technique (Tichý et al. 2014) that allows for samples
displaying a poor fit to existing clusters to form new clusters. This approach has been successfully
used to create and update vegetation classifications (De Cáceres et al. 2010; Wiser & De Cáceres
2013; Perrin 2015). Tichý et al. (2014) also present methods for semi-supervised classifications using
hard clustering approaches, noting that “crisp [i.e. hard] clustering is simpler to run and easier to
interpret than fuzzy clustering” and that in “many applications users need crisp classification that
defines site membership in groups as a categorical variable. For this purpose, fuzzy classifications are
often ‘defuzzified’ to facilitate interpretation” (e.g. see Perrin 2015). However, in the current context
of monitoring habitat interest features in relation to agreed standards, the ability of fuzzy
classification to produce sets of weights describing the affinity or otherwise of new stands to existing
clusters may fulfil the need for an objective method for assessing changes in ecological quality over
time. For monitoring purposes, noise clustering means that stands that do not show affinity to any of
the existing clusters can be investigated and dealt with appropriately; for example, a decision might
be made that a sample was not located within the habitat interest feature, or the sample could
represent a novel type of degradation that was not accounted for by the low quality reference set
e.g. the invasion of a new species, whether alien or native.
As explained above, semi-supervised clustering requires a set of existing cluster definitions. We note
here that in the context of the current monitoring application, where existing clusters would
represent low and high ecological quality in the context of CSM, these clusters could either be
defined initially by the synoptic tables of the NVC, or by relevés representing these community types
or degraded forms thereof. Note also that in creating these initial clusters, certain types may be
amalgamated into single clusters; this is not likely to be a serious deficiency of the approach as long
as low and high quality reference samples cluster separately. Finally, we note that the fuzzy
clustering method discussed here is implemented in the open source statistical computing
environment R (De Cáceres & Wiser 2013).
Other methods
Other methods for classifying plots according to existing classifications also exist, the best known in
Britain being the TABLEFIT method of Hill (1989), which provides goodness-of-fit scores for new plots
matched to the synoptic tables of the UK NVC (currently available at
http://www.ceh.ac.uk/services/tablefit-and-tablcorn). Alternatively, the method of Malloch (1996) is
included within the MAVIS package developed by the Centre for Ecology and Hydrology
(https://www.ceh.ac.uk/services/modular-analysis-vegetation-information-system-mavis). This
method uses the Czekanowski coefficient (Legendre & Legendre 2012) to match new plots to
existing NVC synoptic tables. Both TABLEFIT and MAVIS provide the user with the top ten matching
communities for a particular sample; however, both are limited to the existing NVC classification,
and are fully supervised methods. Another general drawback of these implementations is that they
are stand-alone softwares, with limited flexibility in terms of data input formatting, computational
pipelines, code documentation, analysis sharing, taxonomic change and future amendments to the
UK NVC.
Conclusions
Many of the decisions needed to implement a habitat interest feature quality assessment based only
on plot sample data are outlined above; it is clear that the decisions needed around defining
appropriate high and low quality reference sets for habitat interest features is one of the greatest
challenges, particularly if these sets are to be regionally appropriate (Smart, 2000). Existing lists of

V1.4 3rd October 2016
desirable CSM attributes for habitats would assist with this decision making. Decisions would also be
required to fit whatever quantitative outputs an assessment system provided into the existing CSM
classification of quality (i.e. favourable, unfavourable etc.); assessments of temporal change (e.g.
unfavourable, recovering) would require several sampling periods unless additional surveyor
assessments were also collected at the same time in the current fashion.
The advantages and disadvantages of moving from the existing indicator-based approach to a plot-
based approach are difficult to assess in the absence of data on the existing resources allocated to
CSM monitoring, but it seems highly likely that moving to a plot-based system will be more resource-
intensive. Citizen science initiatives based around recording plots could have the potential to assist
with this type of monitoring (e.g. Pescott et al. 2015), but it seems unlikely that the type of even
coverage needed to maintain a ‘common standard’ across all SSSIs is achievable. We note that this
unevenness in approach was the founding motivation behind the current implementation of CSM
(Rowell 1993). One advantage of a plot-based approach to ecological quality would be the
opportunity to implement a corresponding update of the UK NVC, perhaps putting this classification
onto a more sustainable footing (Wiser & De Cáceres 2013), as noted by Rodwell (2006) the
“published version of the NVC (Rodwell 1991 et seq.), is not meant as a static edifice”.

V1.4 3rd October 2016
References
Borcard, D., Gillet, F. & Legendre, P. (2011). Numerical Ecology with R. Springer.
van den Brink, P.J., den Besten, P.J., bij de Vaate, A. & ter Braak, C.J.F. (2008). Principal response
curves technique for the analysis of multivariate biomonitoring time series. Environmental
Monitoring and Assessment, 152, 271.
Brown, A. (2000). Habitat monitoring for conservation management and reporting. 3: Technical
guide. CCW, Bangor.
Carey, P., Wallis, S., Chamberlain, P.M., Cooper, A., Emmett, B.A., Maskell, L.C., McCann, T., Murphy,
J., Norton, L.R., Reynolds, B., Scott, A., Simpson, I.C., Smart, S.M. & Ullyett, J. (2008).
Countryside Survey: UK Results from 2007. URL
http://www.countrysidesurvey.org.uk/outputs/uk-results-2007 [accessed 28 April 2015]
Clarke, R.T., Wright, J.F. & Furse, M.T. (2003). RIVPACS models for predicting the expected
macroinvertebrate fauna and assessing the ecological quality of rivers. Ecological Modelling,
160, 219–233.
De Cáceres, M., Font, X. & Oliva, F. (2010). The management of vegetation classifications with fuzzy
clustering. Journal of Vegetation Science, 21, 1138–1151.
De Cáceres, M. & Wiser, S.K. (2013). How to use the vegclust package (ver. 1.6.1). https://cran.r-
project.org/web/packages/vegclust/vignettes/VegetationClassification.pdf
Fridley, J.D., Vandermast, D.B., Kuppinger, D.M., Manthey, M. & Peet, R.K. (2007). Co-occurrence
based assessment of habitat generalists and specialists: a new approach for the
measurement of niche width. Journal of Ecology, 95, 707–722.
Gauch, H.G. (1982). Multivariate Analysis in Community Ecology. Cambridge University Press,
Cambridge, UK.
Goodall, D. (1954). Objective methods for the classification of vegetation. III. An essay in the use of
factor analysis. Australian Journal of Botany, 2, 304–324.
Hermoso, V. & Linke, S. (2012). Discrete vs. continuum approaches to the assessment of the
ecological status in Iberian rivers, does the method matter? Ecological Indicators, 18, 477–
484.
Hill, M.O. (1989). Computerized Matching of Relevés and Association Tables, with an Application to
the British National Vegetation Classification. Vegetatio, 83, 187–194.
Hill, M.O. (1979). TWINSPAN: a FORTRAN program for arranging multivariate data in an ordered
two-way table by classification of the individuals and attributes. Section of Ecology and
Systematics, Cornell University.
JNCC. (2009). Common Standards Monitoring Guidance for Upland habitats. Version July 2009. JNCC,
Peterborough, UK.
JNCC. (2004). Common Standards Monitoring Introduction to the Guidance Manual. JNCC,
Peterborough, UK.
Kent, M. (2012). Vegetation Description and Data Analysis: A Practical Approach, 2nd edn. Wiley-
Blackwell, Chichester, UK.
Legendre, P. & Legendre, L.F.J. (2012). Numerical Ecology. Elsevier.
Linke, S., Norris, R.H., Faith, D.P. & Stockwell, D. (2005). ANNA: a new prediction method for
bioassessment programs. Freshwater Biology, 50, 147–158.

V1.4 3rd October 2016
Malloch, A.J.C. (1996). Match Version 2.0: A computer program to aid assignment of vegetation data
to the communities and subcommunities of the National Vegetation Classification. Unit of
Vegetation Science, Lancaster University.
Moss, D. (2000). Evolution of statistical methods in RIVPACS. Assessing the Biological Quality of
Freshwaters: RIVPACS and Other Techniques (eds J.F. Wright, D.W. Sutcliffe & M.T. Furse).
Freshwater Biological Association, Ambleside.
Moss, D., Wright, J.F., Furse, M.T. & Clarke, R.T. (1999). Α comparison of alternative techniques for
prediction of the fauna of running‐water sites in Great Britain. Freshwater Biology, 41, 167–
181.
Peet, R.K. & Roberts, D.W. (2013). Classification of Natural and Semi-natural Vegetation. Vegetation
Ecology (eds J. Franklin & E. van der Maarel), pp. 28–70. Wiley-Blackwell, New York.
Perrin, P.M. (2015). Irish Vegetation Classification Technical Progress Report No. 1. BEC Consultants
Ltd.
Pescott, O.L., Walker, K.J., Pocock, M.J.O., Jitlal, M., Outhwaite, C.L., Cheffings, C.M., Harris, F. & Roy,
D.B. (2015). Ecological monitoring with citizen science: the design and implementation of
schemes for recording plants in Britain and Ireland. Biological Journal of the Linnean Society,
115, 505–521.
Poulin, M., Andersen, R. & Rochefort, L. (2013). A New Approach for Tracking Vegetation Change
after Restoration: A Case Study with Peatlands. Restoration Ecology, 21, 363–371.
Preston, C.D., Harrower, C.A. & Hill, M.O. (2013). A comparison of distribution patterns in British and
Irish mosses and liverworts. Journal of Bryology, 35, 71–87.
Preston, C.D. & Hill, M.O. (1997). The geographical relationships of British and Irish vascular plants.
Botanical Journal of the Linnean Society, 124, 1–120.
Rodwell, J.S. (Ed.). (1991). British Plant Communities (5 Volumes). Cambridge University Press,
Cambridge, UK.
Rodwell, J.S. (2006). National Vegetation Classification: User’s Handbook. Joint Nature Conservation
Committee, Peterborough.
Rodwell, J. (2012). UK National Vegetation Classification Database. Biodiversity & Ecology, 4, 381–
381.
Rowell, T.A. (1993). Common Standards Monitoring for SSSIs, Volume 1. JNCC, Peterborough, UK.
Smart, S.M. (2000). Ecological assessment of vegetation from a nature reserve using regional
reference data and indicator scores. Biodiversity & Conservation, 9, 811–832.
Smart, S.M., Jarvis, S., Walker, K.J., Henrys, P.A., Pescott, O.L. & Marrs, R.H. (2015). Common plants
as indicators of habitat suitability for rare plants; quantifying the strength of the association
between threatened plants and their neighbours. New Journal of Botany, 5, 72–88.
Tichý, L., Chytrý, M. & Botta-Dukát, Z. (2014). Semi-supervised classification of vegetation:
preserving the good old units and searching for new ones. Journal of Vegetation Science, 25,
1504–1512.
Williams, J.M. (Ed.). (2006). Common Standards Monitoring for Designated Sites: First Six Year
Report. JNCC, Peterborough, UK.
Wiser, S.K. & De Cáceres, M. (2013). Updating vegetation classifications: an example with New
Zealand’s woody vegetation. Journal of Vegetation Science, 24, 80–93.