123
addition, Duque et al. (2016), as part of their research, present the evaluations of
strategies for reducing air pollution in urban ecosystems. According to Llop et al.
(2012), traffic intensity, green areas, and individual household fireplaces have a
great influence on air quality and the reduction of lichen diversity. Previously conducted research had shown identical results (Conti and Cecchetti 2001; Loppi 2014;
Washburn and Cullen 2006; Perlmutter 2010; Sujetoviene 2015; Sujetoviene and
Galinyte 2016).
4.5.4 Statistical Analyses
The statistical analysis of data obtained by field research can be performed with the
help of several widely accepted static tools. The application of artificial neural networks has proven to be very effective in determining the spatial distribution patterns
of lichens. In our previous research, Kohonen’s self-organizing maps were used
(self-organizing maps, SОМ; Kohonen 1982, 2001), which represent a type of artificial neural network, i.e., an unsupervised algorithm (Ristić et al. 2019; Ristić et al.
2020). The SОМ method is a technique of modeling and visualization of linear and
nonlinear connections in a high-dimensional data set in the form of a lowdimensional space (neural network). Self-organizing maps differ from other types
of neural networks in that they store information about the topological properties of
inputs using the function of neighboring neurons. The Kohonen network consists of
a number of neurons, each of which contains a set of weight factors of equal length
as the input data vector. The structure of the SOM consists of an input and an output layer.
The data is entered into the analysis using an input matrix. The input matrix for
displaying the spatial distribution of lichens consists of a certain number of rows
and columns. The number of rows represents the number of investigated points,
while the number of columns represents the number of ascertained taxa. A selforganizing map can be interpreted as a set of neurons, each of which represents one
group of data, with the data belonging to the group represented by the “winning”
neuron. Upon completion of network training, all localities were assigned to the
appropriate neurons. The resolution of the SOM network depends on the type of
analysis.
As the SOM analysis does not show statistical indications of the species responsible for cluster separation, this type of analysis is often accompanied by the application of the IndVal method (the indicator values; Dufrêne and Legendre 1997). In
order to identify the significant species with IndVal parameter values greater than
25%, the Monte Carlo significance test with 1000 permutations is applied. Such
species are representative of the group of localities with a relative frequency and
density of no less than 50%. Species that have an IndVal value of less than 25%, and
are statistically significant (p < 0.05), represent taxa that are important for the group.
These species provide additional information about the group but are not
4 Lichens as the Main Indicator in Biological Monitoring of Air Quality
addition, Duque et al. (2016), as part of their research, present the evaluations of
strategies for reducing air pollution in urban ecosystems. According to Llop et al.
(2012), traffic intensity, green areas, and individual household fireplaces have a
great influence on air quality and the reduction of lichen diversity. Previously conducted research had shown identical results (Conti and Cecchetti 2001; Loppi 2014;
Washburn and Cullen 2006; Perlmutter 2010; Sujetoviene 2015; Sujetoviene and
Galinyte 2016).
4.5.4 Statistical Analyses
The statistical analysis of data obtained by field research can be performed with the
help of several widely accepted static tools. The application of artificial neural networks has proven to be very effective in determining the spatial distribution patterns
of lichens. In our previous research, Kohonen’s self-organizing maps were used
(self-organizing maps, SОМ; Kohonen 1982, 2001), which represent a type of artificial neural network, i.e., an unsupervised algorithm (Ristić et al. 2019; Ristić et al.
2020). The SОМ method is a technique of modeling and visualization of linear and
nonlinear connections in a high-dimensional data set in the form of a lowdimensional space (neural network). Self-organizing maps differ from other types
of neural networks in that they store information about the topological properties of
inputs using the function of neighboring neurons. The Kohonen network consists of
a number of neurons, each of which contains a set of weight factors of equal length
as the input data vector. The structure of the SOM consists of an input and an output layer.
The data is entered into the analysis using an input matrix. The input matrix for
displaying the spatial distribution of lichens consists of a certain number of rows
and columns. The number of rows represents the number of investigated points,
while the number of columns represents the number of ascertained taxa. A selforganizing map can be interpreted as a set of neurons, each of which represents one
group of data, with the data belonging to the group represented by the “winning”
neuron. Upon completion of network training, all localities were assigned to the
appropriate neurons. The resolution of the SOM network depends on the type of
analysis.
As the SOM analysis does not show statistical indications of the species responsible for cluster separation, this type of analysis is often accompanied by the application of the IndVal method (the indicator values; Dufrêne and Legendre 1997). In
order to identify the significant species with IndVal parameter values greater than
25%, the Monte Carlo significance test with 1000 permutations is applied. Such
species are representative of the group of localities with a relative frequency and
density of no less than 50%. Species that have an IndVal value of less than 25%, and
are statistically significant (p < 0.05), represent taxa that are important for the group.
These species provide additional information about the group but are not
4 Lichens as the Main Indicator in Biological Monitoring of Air Quality