H x x

Wx j - Wa,b where a and b present location of neighbor nodes in columns and rows and H is the number of neighbor units, dependent upon the location of the map unit. The values were rescaled between 0 and 1 for the purpose of visual comparison. The matrix was presented as a grayscale picture based on the calculated values: bright areas with low values depict short distances while dark areas with high values represent long distances to the surrounding neighbors. The groups were produced with the dotted lines (Figure 4).

The lighter the gray scale between the map units, the smaller the relative distance between them. On the U-matrix, the nodes of the SOM tended to group at the lower and upper areas with stronger borderlines. Unpolluted areas, mainly consisting of CM sample sites, were located in the lower region of the SOM.

In addition, HD sample sites were also bounded by the U-matrix values in the upper right corner of the map.

Large-Scale Data

For sustainable ecosystem management, long-term, large-scale, and spatiotemporal surveys need to be performed. In order to fulfill the goal of the long-term study, a steady and consistent sampling program using a well-defined survey plan is necessary, which consequently produces a large amount of data. For establishing appropriate national policies for land management or water quality control, for example, a comprehensive understanding of large-scale data is necessary. The SOM has the advantage of processing this type of complex, large-scale data.

Figure 5 demonstrates application of the SOM to large-scale community data. Benthic macroinvertebrates were quantitatively collected at 1970 sample sites located in relatively clean to intermediately polluted areas in

Figure 3 Classification of sampling units by the trained SOM. Acronyms in units stand for samples: the first three letters represent sampling sites (see Figure 2), and the last three indicate the sampling season: SPR, spring; SUM, summer; AUT, autumn; and WIN, winter. Reproduced from Park Y-S, Chon T-S, Kwak I-S, and Lek S (2004) Hierarchical community classification and assessment of aquatic ecosystems using artificial neural networks. Science of the Total Environment 327: 105-122, with permission.

Figure 3 Classification of sampling units by the trained SOM. Acronyms in units stand for samples: the first three letters represent sampling sites (see Figure 2), and the last three indicate the sampling season: SPR, spring; SUM, summer; AUT, autumn; and WIN, winter. Reproduced from Park Y-S, Chon T-S, Kwak I-S, and Lek S (2004) Hierarchical community classification and assessment of aquatic ecosystems using artificial neural networks. Science of the Total Environment 327: 105-122, with permission.

South Korea from 1997 to 2002. A two-dimensional map was produced after the SOM training (Figure 5a).

Association of the patterned nodes could also be presented using cluster analysis (Figure 5b) in addition to the U-matrix (Figure 4). Figure 5b shows clustering based on Ward's linkage method. Depending upon the different degree of distances based on the clustering process, the clusters could be assigned appropriately (e.g., Clusters 1, 2, etc.). When the assigned cluster numbers were arranged on the map, the clusters represented geographic regions well (Figure 5c). This further indicated that the SOM could efficiently define ecoregions.

Clustering could be utilized for illustrating different levels of grouping through training (Figure 6). Depending upon the level of similarities, the larger grouping could appear in the map (Figure 6a). The three groups, 'Clusters 2 and 3' at the bottom-left corner, 'Clusters 4 and 8' at the bottom-right area, and the remaining clusters, were accordingly divided on the map (Figure 6b). The larger groups were based on sample site states: cleanness (lack of pollution) for 'Clusters 4 and 8' and geographical characteristics for 'Clusters 2 and 3'. After the initial grouping, the intermediate levels could also be observed (Figure 6b), being matched to the eight clusters as shown in Figure 5 a. Intermediate levels could be more finely divided into smaller clusters (Figure 6c). This type of clustering would be further useful for revealing detailed community organization at different organizational levels.

Figure 4 The clusters determined by the U-matrix. Reproduced from Park Y-S, Chon T-S, Kwak I-S, and Lek S (2004) Hierarchical community classification and assessment of aquatic ecosystems using artificial neural networks. Science of the Total Environment 327: 105-122, with permission.

elevations (e.g., Cluster 1; Figure 7a) correlated to metropolitan areas in Korea (Figure 5c). Cluster 2 was characterized by a high electrical conductivity (Figure 7b), due to the wide distribution of limestone in the area represented by Cluster 2 (Figure 5c).

Biological indices could correspondingly be presented in different clusters (Figure 8). 'Clusters 4 and 8' showed the high range of EPT richness (total species richness in Ephemeroptera, Plecoptera, and Trichoptera) and biological monitoring working party (BMWP) scores (Figures 8a and 8b). This was in accordance with the broad clustering of 'Clusters 4 and 8' (C group in Figure 6a) for showing the less-polluted sites. 'Clusters 2 and 3', another broad group indicated in Figure 6a (B group), also showed higher biotic indices' levels (Figures 8a and 8b), but they were characteristically located in the eastern geographic region of the Korean Peninsula (Figure 5c). The SOM could also visualize the occurrence of species corresponding to the individual groups. Profiles of different taxa would be accordingly presented on the map (Figure 9). The scopes of different species are presented in two dimensions. This type of visualization on the map provides useful data for establishing and monitoring ecosystem management policies.

Future Directions

Environmental factors could be also visualized using the SOM. Environmental variable profiles, such as altitude and conductivity, could be accordingly presented in different clusters (Figure 7). The levels of altitude, for instance, were high for Clusters 3, 4, and 8 (Figure 7a). These clusters correlated to the sample sites found in mountainous areas in Korea (Figure 5c). These areas correspondingly showed the lowest range of conductivities (Figure 7b). Clusters representative of lower

As illustrated above, SOM would be useful in providing comprehensive views of ecological data through data processing. Through ordination, clustering, and visualization, the pollution gradient affects the overall state of community changes, and the pollution gradient impacts were elucidated accordingly in response to environmental disturbances. The trained SOM readily accommodated the diverse scope of ecological systems exposed to various sources of stress and disturbance.

35 30 25

15 10 5

CD 2

"c3

u lju 1

SOM units

Figure 5 Classification of the samples according to the trained SOM. (a) The SOM units classified to eight clusters; (b) the dendrogram according to Ward's linkage method based on the Euclidean distance; (c) geographical location of the sampling sites matching to clusters according to the SOM (a). Reproduced from Park Y-S, Song M-Y, Park Y-C, Oh K-H, Cho E-C, and Chon T-S (2007) Community patterns of benthic macroinvertebrates collected on the national scale in Korea. Ecological Modelling 203: 26-33, with permission.

SOM units

Figure 5 Classification of the samples according to the trained SOM. (a) The SOM units classified to eight clusters; (b) the dendrogram according to Ward's linkage method based on the Euclidean distance; (c) geographical location of the sampling sites matching to clusters according to the SOM (a). Reproduced from Park Y-S, Song M-Y, Park Y-C, Oh K-H, Cho E-C, and Chon T-S (2007) Community patterns of benthic macroinvertebrates collected on the national scale in Korea. Ecological Modelling 203: 26-33, with permission.

(a) 3 clusters

(b) 8 clusters

(c) 16 clusters

(a) 3 clusters

(b) 8 clusters

(c) 16 clusters

Figure 6 Clustering in different levels with 3, 8, and 16 clusters according to Ward's linkage method based on the Euclidean distance. Reproduced from ParkY-S, Song M-Y, Park Y-C, Oh K-H, Cho E-C, and Chon T-S (2007) Community patterns of benthic macroinvertebrates collected on the national scale in Korea. Ecological Modelling 203: 26-33, with permission.

Figure 6 Clustering in different levels with 3, 8, and 16 clusters according to Ward's linkage method based on the Euclidean distance. Reproduced from ParkY-S, Song M-Y, Park Y-C, Oh K-H, Cho E-C, and Chon T-S (2007) Community patterns of benthic macroinvertebrates collected on the national scale in Korea. Ecological Modelling 203: 26-33, with permission.

CO i

350 300 250 200 150 100 50 0

150 100 50

Figure 7 Environmental variables in different clusters according to the SOM (Figure 5a): (a) altitude and (b) conductivity. Different alphabets indicate significant differences between the clusters based on the unequal N honestly significant difference (HSD) multiple comparison test (p = 0.05). Error bars indicate mean and standard error of each variable. Conductivity was not available at the samples in Cluster 1. Reproduced from ParkY-S, Song M-Y, Park Y-C, Oh K-H, Cho E-C, and Chon T-S (2007) Community patterns of benthic macroinvertebrates collected on the national scale in Korea. Ecological Modelling 203: 26-33, with permission.

Another advantage of the SOM is recognition. Since the models are based on learning processes, new data sets could be evaluated using the trained map. Evaluation of a new data set is possible using the previously patterned data. Figure 10 shows an example of recognition of the long-term survey data on the SOM separately trained with the

4 5 6 Cluster

Figure 8 Variation in biological indices in different clusters according to the SOM (Figure 5a): (a) EPT richness and (b) biological monitoring working party (BMWP) score. Different alphabets indicate significant differences between the clusters based on the unequal N honestly significant difference (HSD) multiple comparison test (p = 0.05). Error bars indicate mean and standard error of each variable. Reproduced from Park Y-S, Song M-Y, Park Y-C, Oh K-H, Cho E-C, and Chon T-S (2007) Community patterns of benthic macroinvertebrates collected on the national scale in Korea. Ecological Modelling 203: 26-33, with permission.

community data. The macroinvertebrate community data collected monthly at a sample site in the Suyong River from November 1992 to April 1995 were recognized in a sequence on the trained SOM (Figure 10a). In the early period (November 1992-November 1993), communities were mostly located in clusters III and IV, frequently crossing over the boundary between the two clusters. With respect to the low biological indices' values in clusters III and IV (Figure 10b), water quality appeared poor at this stage. As time progressed, the changes in community status were revealed with communities moving from the polluted state in cluster IV to the clean state cluster I in the later period (January 1994 and January 1995; Figure 10a). This is indicative of a temporal recovery of water quality in the winter of 1994 and 1995. The sample site in turn returned to the polluted state in cluster IV in the last period of survey in March 1995 (Figure 10a). Differences in biological and physicochemical indices obtained from newly recognized samples in different clusters were accordingly a b d

Figure 10 Monitoring of benthic macroinvertebrate communities collected at YCK in the Suyong stream from November 1992 to April 1995 according to the trained SOM. The sample was not collected in December 1994. (a) Recognition of the samples (November 1992-November 1993 (dots); January 1994-March 1995 (solid)). (b) Mean and SE of biological and physicochemical indices in different clusters defined in the SOM. The different alphabets indicate significant difference in the Mann-Whitney test (p <0.001). Reproduced from Song M-Y, Hwang H-J, Kwak I-S, etal. (2007) Self-organizing mapping of benthic macroinvertebrate communities implemented to community assessment and water quality evaluation. Ecological Modelling 203: 18-25, with permission.

Figure 10 Monitoring of benthic macroinvertebrate communities collected at YCK in the Suyong stream from November 1992 to April 1995 according to the trained SOM. The sample was not collected in December 1994. (a) Recognition of the samples (November 1992-November 1993 (dots); January 1994-March 1995 (solid)). (b) Mean and SE of biological and physicochemical indices in different clusters defined in the SOM. The different alphabets indicate significant difference in the Mann-Whitney test (p <0.001). Reproduced from Song M-Y, Hwang H-J, Kwak I-S, etal. (2007) Self-organizing mapping of benthic macroinvertebrate communities implemented to community assessment and water quality evaluation. Ecological Modelling 203: 18-25, with permission.

shown in Figure 10b. Biological indices such as EPT richness and BMWP were clearly differentiated based on statistical significance among the different clusters. Overall tracks recorded on the map demonstrated that states of communities collected on the regular basis could be continuously monitored using the trained SOM. This type of monitoring, based on the SOM recognition, would be efficient in estimating community states in the long-term survey.

In addition to recognition, integrative analysis of various taxa could be the future direction of the SOM implementation. As the monitoring horizon increases to cover multiple taxa, producers (e.g., algae), consumers (e.g., benthic macroinvertebrates), and decomposers (e.g., bacteria), the collected data need to be analyzed concurrently. The SOM is flexible in accommodating complex community structure and handling a large number of data in a nonlinear fashion.

Oplan Termites

Oplan Termites

You Might Start Missing Your Termites After Kickin'em Out. After All, They Have Been Your Roommates For Quite A While. Enraged With How The Termites Have Eaten Up Your Antique Furniture? Can't Wait To Have Them Exterminated Completely From The Face Of The Earth? Fret Not. We Will Tell You How To Get Rid Of Them From Your House At Least. If Not From The Face The Earth.

Get My Free Ebook


Post a comment