A brute‐force approach to vegetation classification

Aim: Introduction of a novel approach to the classification of vegetation data (species by plot matrices). This approach copes with a large amount of noise, groups irregularly shaped in attribute space and species turnover within groups. Method: The proposed algorithm (Isopam) is based on the classification of ordination scores from isometric feature mapping. Ordination and classification are repeated in a search for either high overall fidelity of species to groups of sites, or high quantity and quality of indicator species for groups of sites. The classification is performed either as a hierarchical, divisive method or as non‐hierarchical partitioning. In divisive clustering, resulting groups are subdivided until a stopping criterion is met. Isopam was tested on 20 real‐world data sets. The resulting classifications were compared with solutions from eight widely used clustering algorithms. Results: When looking at the significance of species fidelities to groups of sites, and at quantity and quality of indicator species, Isopam often achieved high ranks as compared with other algorithms.     

Vegetation mapping using hierarchical object‐based image analysis applied to aerial imagery and lidar data

Aims: The primary objective of this study is to map the distribution and quantify the cover of vegetation alliances over the entirety of San Clemente Island (SCI). To this end, we develop and evaluate the mapping method of hierarchical object‐based classification with a rule‐based expert system. Location: San Clemente Island, California, USA. Methods: We developed and tested an approach based on hierarchical object‐based classification with a rule‐based expert system to effectively map vegetation communities on SCI following the Manual of California Vegetation classification system. In this mapping approach, the shrub species defining each vegetation community and non‐shrub growth forms were first mapped using aerial imagery and lidar data, then used as input in an automated mapping rule set that incorporates the percent cover rules of a field‐based mapping rule set. Results: The final vegetation map portrays the distribution of 19 vegetation communities across SCI, with the largest areas comprised of California Annual and Perennial Grassland (35%) and three types of coastal sage scrub and maritime succulent scrub, comprising a combined 53% of the area. Map accuracy was assessed to be 79% based on fuzzy methods and 61% with a traditional accuracy assessment. The accuracy of tree identification was assessed to be 81%, but species‐level tree accuracy was 45%. Conclusions: Semi‐automated approaches to vegetation community mapping can produce repeatable maps over large spatial extents that facilitate ecological management efforts. However, some low‐statured shrub community types were difficult to differentiate due to patchy canopies of co‐occurring species including abundant non‐native grasses characteristic of complex disturbance histories. Species‐level tree mapping accuracy was low due to the difficulty of identifying species within poorly illuminated canyons, resulting from sub‐optimal image acquisition timing.     

Assignment of relevés to pre‐defined classes by supervised clustering of plant communities using a new composite index

Question: How does a newly designed method of supervised clustering perform in the assignment of relevé (species composition) data to a previously established classification. How do the results compare to the assignment by experts and to the assignment using a completely different numerical method? Material: Relevés analysed represent 4186 Czech grassland plots and 4990 plots from a wide variety of vegetation types (359 different associations or basal communities) in The Netherlands. For both data sets we had at our disposal an expert classification, and for the Czech data we also had available a numerical classification as well as a classification based on a neural network method (multi‐layer perceptron). Methods: Two distance indices, one qualitative and one quantitative, are combined into a single index by weighted multiplication. The composite index is a distance index for the dissimilarity between relevés and vegetation types. For both data sets the classifications by the new method were compared with the existing classifications. Results: For the Czech grasslands we correctly classified 81% of the plots to the classes of an expert classification at the alliance level and 71% to the classes of the numerical classification. Correct classification rates for the Dutch relevés were 64, 78 and 83 % for the lowest (subassociation or association), association, and alliance level, respectively. Conclusion: Our method performs well in assigning community composition records to previously established classes. Its performance is comparable to the performance of other methods of supervised clustering. Compared with a multi‐layer perceptron (a type of artificial neural network), fewer parameters have to be estimated. Our method does not need the original relevé data for the types, but uses synoptic tables. Another practical advantage is the provision of directly interpretable information on the contributions of separate species to the result.     

Numerical reproduction of traditional classifications and automatic vegetation identification

Questions: Is it possible to develop an expert system to provide reliable automatic identifications of plant communities at the precision level of phytosociological associations? How can unreliable expert‐based knowledge be discarded before applying supervised classification methods? Material: We used 3677 relevés from Catalonia (Spain), belonging to eight orders of terrestrial vegetation. These relevés were classified by experts into 222 low‐level units (associations or sub‐associations). Methods: We reproduced low‐level, expert‐defined vegetation units as independent fuzzy clusters using the Possibilistic C‐means algorithm. Those relevés detected as transitional between vegetation types were excluded in order to maximize the number of units numerically reproduced. Cluster centroids were then considered static and used to perform supervised classifications of vegetation data. Finally, we evaluated the classifier's ability to correctly identify the unit of both typical (i.e. training) and transitional relevés. Results: Only 166 out of 222 (75%) of the original units could be numerically reproduced. Almost all the unrecognized units were sub‐associations. Among the original relevés, 61% were deemed transitional or untypical. Typical relevés were correctly identified 95% of the time, while the efficiency of the classifier for transitional data was only 64%. However, if the second classifier's choice was also considered, the rate of correct classification for transitional relevés was 80%. Conclusions: Our approach stresses the transitional nature of relevé data obtained from vegetation databases. Relevé selection is justified in order to adequately represent the vegetation concepts associated with expert‐defined units.     

Numeric classification as an aid to spectral mapping of vegetation communities

This study demonstrates a vegetation mapping methodology that relates the reflectance information contained in multispectral imagery to traditionally accepted ecological classifications. Key elements of the approach used are (a) the use of cover rather than density or presence/absence to quantify the vegetation, (b) the inclusion of physical components as well as vegetation cover to describe and classify field sites, (c) development of an objective land cover classification from this quantitative data, (d) use of the field sample sites as training areas for the spectral classification, and (e) the use of a discriminant function to effectively tie the two classifications together. Land cover over 39000 ha of Australian chenopod shrubland was classified into nine groups using agglomerative hierarchical clustering, a discriminant function developed to relate cover and spectral classes, and the vegetation mapped using a maximum likelihood classification of multi-date Landsat TM imagery. The accuracy of the mapping was assessed with an independent set of field samples and by comparison with a map of land systems previously interpreted from aerial photography. Overall agreement between the digital classification and the land system map was good. The units that have been mapped are those derived from numeric vegetation classification, demonstrating that accepted ecological methods and sound image analysis can be successfully combined.     

Effects of different fidelity measures and contexts on the determination of diagnostic species

Question: How different are lists of diagnostic species of vegetation units, derived using various fidelity measures, in different contexts and with presence/absence versus cover data? Methods: Six different fidelity measures were calculated for vegetation units of two classified data sets covering contrasting types of Central European vegetation (beech forest and dwarf shrub vegetation). Both statistical and non‐statistical fidelity measures were used, and either species presence/absence or cover was considered. Each measure was calculated on four hierarchical levels and within two different contexts, either within the whole data set or within the next higher level of hierarchical classification. Average similarities of the diagnostic species lists derived from various combinations of fidelity measures and contexts were calculated and visualized using principal coordinate analysis (PCoA). Results: The correlations between fidelity values derived from non‐statistical and statistical measures were rather weak. Nevertheless, diagnostic species lists calculated for the same syntaxon by different measures usually had several species in common. Average similarity between pairs of fidelity measures or contexts (based on the Sørensen similarity index) ranged from 0.21 to 0.92. PCoA clustered individual combinations of fidelity measures and contexts mainly according to the context and the use of presence/absence versus cover data, rather than according to the fidelity measures. Conclusions: The strongest impact on the lists of diagnostic species was not the fidelity measure itself but the context of its application and the use of presence/absence or cover data. Despite the weak correlation between individual fidelity values, traditional (non‐statistical) and statistical measures produce quite similar lists of diagnostic species, provided that the context of the analysis is the same. Both approaches have their advantages and disadvantages, and the choice of the appropriate algorithm should depend on the focus of the study.     

Vegetation classification,mapping, and monitoring at Voyageurs National Park,Minnesota: An application of the U.S. National Vegetation Classification

Question: How can the U.S. National Vegetation Classification (USNVC) serve as an effective tool for classifying and mapping vegetation, and inform assessments and monitoring? Location: Voyageurs National Park, northern Minnesota, U.S.A and environs. The park contains 54 243 ha of terrestrial habitat in the sub-boreal region of North America. Methods: We classified and mapped the natural vegetation using the USNVC, with ‘alliance’and ‘association’as base units. We compiled 259 classification plots and 1251 accuracy assessment test plots. Both plot and type ordinations were used to analyse vegetation and environmental patterns. Color infrared aerial photography (1:15840 scale) was used for mapping. Polygons were manually drawn, then transferred into digital form. Classification and mapping products are stored in publicly available databases. Past fire and logging events were used to assess distribution of forest types. Results and Discussion: Ordination and cluster analyses confirmed 49 associations and 42 alliances, with three associations ranked as globally vulnerable to extirpation. Ordination provided a useful summary of vegetation and ecological gradients. Overall map accuracy was 82.4%. Pinus banksiana - Picea mariana forests were less frequent in areas unburned since the 1930s. Conclusion: The USNVC provides a consistent ecological tool for summarizing and mapping vegetation. The products provide a baseline for assessing forests and wetlands, including fire management. The standardized classification and map units provide local to continental perspectives on park resources through linkages to state, provincial, and national classifications in the U.S. and Canada, and to NatureServe's Ecological Systems classification.     

A new algorithm for the determination of differential taxa

Question: How can we determine differential taxa in a vegetation data set? Methods: The new algorithm presented here uses an intuitive fidelity threshold based on relative constancy differences. It is tested on a simulated and a real data set. The results of the proposed algorithm are discussed in comparison with other methods used for the determination of differential taxa. Results: The new algorithm defines each taxon in each group of relevés as: (1) positively differentiating, (2) positively‐negatively differentiating, (3) negatively differentiating, or (4) non‐differentiating. Each taxon in a data set may be: (1) positively, positively‐negatively or negatively differentiating for each group in the data set, (2) differentiating for some groups and non‐differentiating for the remaining groups, or (3) non‐differentiating for all groups in the data set. Conclusions: The new algorithm finds the relevé groups that are positively differentiated against other groups that are negatively differentiated. It reveals differentiating structures in the data set and thus makes quantification of the relations among and between different syntaxonomic ranks conceivable. As it distinguishes between different types of differential taxa, it might improve standards of typification in vegetation classification.     

The management of vegetation classifications with fuzzy clustering

Questions: Does fuzzy clustering provide an appropriate numerical framework to manage vegetation classifications? What is the best fuzzy clustering method to achieve this? Material: We used 531 relevés from Catalonia (Spain), belonging to two syntaxonomic alliances of mesophytic and xerophytic montane pastures, and originally classified by experts into nine and 13 associations, respectively. Methods: We compared the performance of fuzzy C‐means (FCM), noise clustering (NC) and possibilistic C‐means (PCM) on four different management tasks: (1) assigning new relevé data to existing types; (2) updating types incorporating new data; (3) defining new types with unclassified relevés; and (4) reviewing traditional vegetation classifications. Results: As fuzzy classifiers, FCM fails to indicate when a given relevé does not belong to any of the existing types; NC might leave too many relevés unclassified; and PCM membership values cannot be compared. As unsupervised clustering methods, FCM is more sensitive than NC to transitional relevés and therefore produces fuzzier classifications. PCM looks for dense regions in the space of species composition, but these are scarce when vegetation data contain many transitional relevés. Conclusions: All three models have advantages and disadvantages, although the NC model may be a good compromise between the restricted FCM model and the robust but impractical PCM model. In our opinion, fuzzy clustering might provide a suitable framework to manage vegetation classifications using a consistent operational definition of vegetation type. Regardless of the framework chosen, national/regional vegetation classification panels should promote methodological standards for classification practices with numerical tools.     

Evaluating the stability of the classification of community data

We propose a method for a posteriori evaluation of classification stability which compares the classification of sites in the original data set (a matrix of species by sites) with classifications of subsets of its sites created by without‐replacement bootstrap resampling. Site assignments to clusters of the original classification and to clusters of the classification of each subset are compared using Goodman‐Kruskal's lambda index. Many resampled subsets are classified and the mean of lambda values calculated for the classifications of these subsets is used as an estimation of classification stability. Furthermore, the mean of the lambda values based on different resampled subsets, calculated for each site of the data set separately, can be used as a measure of the influence of particular sites on classification stability. This method was tested on several artificial data sets classified by commonly used clustering methods and on a real data set of forest vegetation plots. Its strength lies in the ability to distinguish classifications which reflect robust patterns of community differentiation from unstable classifications of more continuous patterns. In addition, it can identify sites within each cluster which have a transitional species composition with respect to other clusters.     

Formalized reproduction of an expert‐based phytosociological classification: A case study of subalpine tall‐forb vegetation

Abstract. Delimitation of vegetation units in phytosociology is traditionally based on expert knowledge. Applications of expert‐based classifications are often inconsistent because criteria for assigning relevés to vegetation units are seldom given explicitly. Still, there is, e.g. in nature conservation, an increasing need for a consistent application of vegetation classification using computer expert systems for unit identification. We propose a procedure for formalized reproduction of an expert‐based vegetation classification, which is applicable to large phytosociological data sets. This procedure combines Bruelheide's Cocktail method with a similarity‐based assignment of relevés to constancy columns of a vegetation table. As a test of this method we attempt to reproduce the expert‐based phytosociological classification of subalpine tall‐forb vegetation of the Czech Republic which has been made by combination of expert judgement and stepwise numerical classification of 718 relevés by TWINSPAN. Applying the Cocktail method to a geographically stratified data set of 21794 relevés of all Czech vegetation types, we defined groups of species with the statistical tendency of joint occurrences in vegetation. Combinations of 12 of these species groups by logical operators AND, OR and AND NOT yielded formal definitions of 14 of 16 associations which had been accepted in the expert‐based classification. Application of these formal definitions to the original data set of 718 relevés resulted in an assignment of 376 relevés to the associations. This assignment agreed well with the original expert‐based classification. Relevés that remained un‐assigned because they had not met the requirements of any of the formal definitions, were subsequently assigned to the associations by calculating similarity to relevé groups that had already been assigned to the associations. A new index, based on frequency and fidelity, was proposed for calculating similarity. The agreement with the expert‐based classification achieved by the formal definitions was still improved after applying the similarity‐based assignment. Results indicate that the expert‐based classification can be successfully formalized and converted into a computer expert system.     

瑞典河漫滩草甸植被的数量分类和排序

本文用目前各国学者广泛使用的一些数量分类和排序方法对瑞典河漫滩草甸样地资料进行了分类和排序。所用的方法包括多元等级聚合分类(TABORD程序),多元等级分划分类(TWINSPAN程序),PCA排序(ORDINA程序),RA和DCA排序(DECORANA程序)。研究结果表明可以把28个样地分为6个群落类型,它们的分布格局是与土壤水分梯度密切相关的。此外本文还对数量分类和排序方法在植物群落学研究中的应用以及所用方法的比较进行了讨论。     

Applying randomization tests to cluster analyses

Abstract. In applying randomization tests to hierarchical cluster analyses, we have noted a potentially misleading result: within a significant group, linkages are often identified as significant even when species are randomly distributed among the group's sites. We demonstrate this through a cluster analysis of a constructed matrix with two groups of 20 sites that share no species, and within each group species are randomly distributed among sites. A randomization test identified both of the groups and all linkages within them as significant, while the same test found all linkages non‐significant in the cluster analysis of a matrix containing just one of the two groups of 20 sites. In general, a non‐random distribution of species within a data set shortens linkages relative to distances in null distributions derived from randomized versions of the data. This confounds efforts to identify significant sub‐groups within a significant group. However, the significance of sub‐groups possibly could be tested by comparing linkage distances to a null distribution derived from the randomization and clustering of a sub‐matrix containing only the sites within the larger group. In essence, this comparison tests the null hypothesis that within the significant group, sites represent random assemblages of species. When applied to actual data sets, an approach involving sequential randomization tests could allow the evaluation of all nodes in a classification, increasing the utility of randomization tests and strengthening the interpretation of groups produced by cluster analysis.     

The use of the Cole/Hurlbert C8 association coefficient in inverse ecological classification

Abstract. Inverse classification is routinely used in vegetation surveys to produce groups of sociologically similar species. However, no classification methods have been proposed specifically for this purpose, nor has any evaluation been made of the suitability of existing methodsforthe purpose. Anewvariant of Cluster Analysis is introduced, i. e. using the Cole/Hurlbert association measure C₈ as coefficient, and named for convenience Cole Cluster Analysis. This method, and four standard ones, were used on saltmarsh vegetational data from New Zealand. Ecophysiological data were obtained from salt-tolerance experiments. These data, and distributional information, were used as independent criteria against which to compare the inverse vegetation classifications. Information Analysis did not prove suitable for inverse analysis in this test. Nor did Cluster Analysis with the Simple Matching Coefficient or with Jaccard's coefficient. Indicator Species Analysis was more suitable, but the new Cole Cluster Analysis seemed the most effective on these data.     

Comparison of floristic and structural classification of vegetation

A comparison is made between floristic and structural-physiognomic classifications of a tropical dry, semi-evergreen forest and thicket vegetation of south-eastern India. The classifications are strikingly similar in their main groupings which are ecologically meaningful; the differences between the classifications are virtually limited to allocation of some stands to different subcommunities. It is concluded that the use of structural-physiognomic criteria allows a detailed and ecologically significant classification of vegetation. Fairly advanced calculation facilities are necessary, however, to reach such a classification, since the structural-physiognomic differences between the resulting groupings are largely of a quantitative and not of a qualitative nature. This is a consequence of the general occurrence in all stands of the vegetation of the very great majority of the characters used in this study.Nomenclature follows Sprangers & Balasubramanian (1978) who give a complete list of authorities.     

Can vegetation be discretely classified in species-poor environments? Testing plant community concepts for vegetation monitoring on sub-Antarctic Marion Island

The updating and rethinking of vegetation classifications is important for ecosystem monitoring in a rapidly changing world, where the distribution of vegetation is changing. The general assumption that discrete and persistent plant communities exist that can be monitored efficiently, is rarely tested before undertaking a classification. Marion Island (MI) is comprised of species-poor vegetation undergoing rapid environmental change. It presents a unique opportunity to test the ability to discretely classify species-poor vegetation with recently developed objective classification techniques and relate it to previous classifications. We classified vascular species data of 476 plots sampled across MI, using Ward hierarchical clustering, divisive analysis clustering, non-hierarchical kmeans and partitioning around medoids. Internal cluster validation was performed using silhouette widths, Dunn index, connectivity of clusters and gap statistic. Indicator species analyses were also conducted on the best performing clustering methods. We evaluated the outputs against previously classified units. Ward clustering performed the best, with the highest average silhouette width and Dunn index, as well as the lowest connectivity. The number of clusters differed amongst the clustering methods, but most validation measures, including for Ward clustering, indicated that two and three clusters are the best fit for the data. However, all classification methods produced weakly separated, highly connected clusters with low compactness and low fidelity and specificity to clusters. There was no particularly robust and effective classification outcome that could group plots into previously suggested vegetation units based on species composition alone. The relatively recent age (c. 450,000 years B.P.), glaciation history (last glacial maximum 34,500 years B.P.) and isolation of the sub-Antarctic islands may have hindered the development of strong vascular plant species assemblages with discrete boundaries. Discrete classification at the community-level using species composition may not be suitable in such species-poor environments. Species-level, rather than community-level, monitoring may thus be more appropriate in species-poor environments, aligning with continuum theory rather than community theory.     

Decadal‐scale aspen changes: evidence in remote sensing and tree ring data

Question: (1) Which remote sensing classification most successfully identify aspen using multitemporal Landsat 5 TM images and airborne lidar data? (2) How has aspen distribution changed in southwestern Idaho? (3) Are topographic variables and conifer encroachment correlated with aspen changes? Location: Reynolds Creek Experimental Watershed in southwestern Idaho, USA. Methods: Multi‐temporal Landsat 5 TM and lidar data were used individually and fused together. The best classification model was compared with a 1965 aspen map and tree ring data. Conifer encroachment was examined via image‐based change detection and field mapping. Lidar‐derived topographic variables were correlated with aspen change patterns using quantile regression models. Results: The best Landsat 5 TM classification was a normalized difference vegetation index (NDVI)‐based approach with 92% overall accuracy. The lidar classification of tree presence/absence performed with 100% overall accuracy. Fusing the lidar classification with various Landsat 5 TM classifications improved overall accuracies 3 to 6%. Among the fusion models, the NDVI‐lidar fusion performed best with 96% overall accuracy. Change detection indicated 69% decline in aspen cover, but 179% increase in aspen cover in other areas of the watershed. Conifers have completely replaced 17% of the aspen, while 93% of the remaining aspen stands have young Douglas‐fir and western juniper trees underneath the aspen canopy. Aspen significantly decreased (P‐values <0.05) with increasing elevation (up to 2150 m) and decreasing slope. Conclusions: Landsat 5 TM data used with a NDVI‐based approach provide an accurate method to classify aspen distribution. Landsat 5 TM classifications can be further improved via fusion with lidar data. Aspen change patterns are spatially variable: while aspen is drastically declining in some parts of this watershed, aspen is increasing in other areas.     

Towards unification of national vegetation classifications: A comparison of two methods for analysis of large data sets

Abstract. In European phytosociology, national classifications of corresponding vegetation types show considerable differences even between neighbouring countries. Therefore, the European Vegetation Survey project urgently needs numerical classification methods for large data sets that are able to produce compatible classifications using data sets from different countries. We tested the ability of two methods, TWINSPAN and COCKTAIL, to produce similar classifications of wet meadows (Calthion, incl. Filipendulenion) for Germany (7909 relevés) and the Czech Republic (1287 relevés) in this respect. In TWINSPAN, the indicator ordination option was used for classification of two national data sets, and the extracted assignment criteria (indicator species) were applied crosswise from one to the other national data set. Although the data sets presumably contained similar community types, TWINSPAN revealed almost no correspondence between the groups derived from the proper classification of the national data set and the groups defined by the assignment criteria taken from the other national data set. The reason is probably the difference in structure between the national data sets, which is a typical, but hardly avoidable, feature of any pair of phytosociological data sets. As a result, the first axis of the correspondence analysis, and consequently the first TWINSPAN division, are associated with different environmental gradients; the difference in the first division is transferred and multiplied further down the hierarchy. COCKTAIL is a method which produces relevé groups on the basis of statistically formed species groups. The user determines the starting points for the formation of species groups, and groups already found in one data set can be tested for existence in the other data set. The correspondence between the national classifications produced by COCKTAIL was fairly good. For some relevé groups, the lack of correspondence to groups in the other national data set could be explained by the absence of the corresponding vegetation types in one of the countries, rather than by methodological problems.