PAEQANN TOOLS
MANUAL FOR USERS
by
L.P. Bretin
Y.S, Park
M. Gevrey
S. Lek
EVK1-CT1999-00026 PAEQANN
Manual for users
1
PAEQANN User Manual
Contents
1.
Introduction ___________________________________________________________ 4
2.
Software Purpose _______________________________________________________ 5
3.
System Requirmement ___________________________________________________ 5
4.
Installation/Desinstallation _______________________________________________ 6
5.
HOW TO USE PAEQANN’s tools _________________________________________ 6
5.1. The welcome screen __________________________________________________________ 6
5.1.1. How to select an organism?_________________________________________________________ 6
5.1.2. How to select a country? ___________________________________________________________ 6
5.1.3. Command part __________________________________________________________________ 7
5.2. Visualisation screen __________________________________________________________ 8
5.2.1. Action part _____________________________________________________________________ 8
5.2.2. Visualisation part ________________________________________________________________ 8
5.3. Prediction screen ____________________________________________________________ 9
5.3.1. Sp/FFG/SR/Guild/... --> visualising results _____________________________________________ 9
5.3.2. Sp/FFG/SR/Guild/... --> Test new data _______________________________________________ 12
5.3.3. Community --> Visualising results __________________________________________________ 14
5.3.4. Community --> Test new data ______________________________________________________ 15
5.4. Ordination screen___________________________________________________________ 19
6.
Artificial Neural Networks (ANNs) _______________________________________ 25
6.1. Self-Organizing Maps (SOMs) _________________________________________________ 25
6.1.1. Development of SOMs ___________________________________________________________ 25
6.1.2. The SOM concept _______________________________________________________________ 26
6.1.3. Self-Organizing Map algorithm_____________________________________________________ 27
6.2. Back Propagation Artificial Neural Network (ANNbp) ____________________________ 29
6.2.1. Introduction ___________________________________________________________________ 29
6.2.2. Back Propagation Artificial Neural Network algorithms__________________________________ 29
6.3. PaD algorithms _____________________________________________________________ 32
7.
Organisms ____________________________________________________________ 32
7.1. Fish ______________________________________________________________________ 32
7.1.1. Belgium_______________________________________________________________________
7.1.1.1. Ordination ________________________________________________________________
7.1.1.2. Prediction _________________________________________________________________
7.1.2. France ________________________________________________________________________
7.1.2.1 Ordination _________________________________________________________________
7.1.2.2 Prediction__________________________________________________________________
5.3.1. Sp/FFG/SR/Guild/... --> visualising results ____________________________________________
5.3.2. Sp/FFG/SR/Guild/... --> Test new data _______________________________________________
5.3.3. Community --> Visualising results __________________________________________________
5.3.4. Community --> Test new data ______________________________________________________
7.1.3. Garonne ______________________________________________________________________
7.1.3.1 Ordination _________________________________________________________________
7.1.3.2 Prediction__________________________________________________________________
EVK1-CT1999-00026 PAEQANN
Manual for users
32
33
39
49
50
56
57
60
62
63
67
68
69
2
5.3.1. Sp/FFG/SR/Guild/... --> visualising results ____________________________________________
5.3.2. Sp/FFG/SR/Guild/... --> Test new data _______________________________________________
5.3.3. Community --> Visualising results __________________________________________________
5.3.4. Community --> Test new data ______________________________________________________
7.1.4. Italy__________________________________________________________________________
7.1.4.1 Ordination _________________________________________________________________
7.1.4.2 Prediction__________________________________________________________________
7.1.5. Luxembourg ___________________________________________________________________
7.1.5.1 Ordination _________________________________________________________________
70
72
74
75
79
80
81
82
82
7.2. Macroinvertebrate___________________________________________________________ 82
7.2.1. Austria _______________________________________________________________________
7.2.1.1 Ordination _________________________________________________________________
7.2.1.2 Prediction__________________________________________________________________
7.2.2. France ________________________________________________________________________
7.2.2.1 Ordination _________________________________________________________________
7.2.2.2 Prediction__________________________________________________________________
7.2.3. Luxembourg ___________________________________________________________________
7.2.31 Ordination _________________________________________________________________
7.2.32 Prediction __________________________________________________________________
7.2.4. The Netherlands ________________________________________________________________
7.2.4.1 Ordination _________________________________________________________________
7.2.4.2 Prediction__________________________________________________________________
7.2.5. Taxa/Species ___________________________________________________________________
82
83
83
83
83
83
83
83
83
83
84
84
84
7.3. Diatoms ___________________________________________________________________ 84
7.3.1. Environmental variables __________________________________________________________ 84
7.3.2. Taxa/Species ___________________________________________________________________ 86
8.
Authors ______________________________________________________________ 86
9.
References ____________________________________________________________ 87
EVK1-CT1999-00026 PAEQANN
Manual for users
3
1. Introduction
Historical factors aside, the structure and diversity of aquatic communities in running waters
are primarily dependent on a complex of physical, chemical and biotic factors. Physical and
chemical variables are themselves heavily dependent on climatic and catchment properties
(hydrology, geology, topography, etc.) which are in turn influenced by anthropogenic impacts
(hydraulic management, land use and agricultural practices, waste water discharge etc.).
During the past several decades, hydrobiological studies have identified the main factors which
determine freshwater communities, but very few have been able to establish deterministic links
between ecological factors and the structure of key aquatic communities. Ecological theories on
determinism of biocenosis structure in streams have put forward the effects of morphodynamic
properties of the river channel, and the importance of stream order, related to various channel
properties, to watershed surface area and to contribution of various sources of organic matter.
These theories have defined general schemes for explaining longitudinal variations in river
systems; however they do not allow prediction of community structure down to the species
level. A notable exception is the RIVPACS system (Wright, 1995), which has been developed in
the UK to predict the composition of undisturbed macroinvertebrate assemblages; it clearly
demonstrates the possibility of developing approaches for predicting community structure
from sets of environmental variables. Similarly, species richness in fish communities may be
predicted from watershed surface area, average discharge and net primary productivity (Guégan
et al., 1998).
A major difficulty is to distinguish the influence of natural characteristics, including natural
disturbances (storms, hydrological variability), from changes due to anthropogenic impacts.
Despite these uncertainties, key aquatic communities have been utilised, sometimes for decades,
to evaluate biological quality of streams and rivers. Practical methods for calculating “biotic
indices” have been designed, with the requirement that they be sufficiently simple for
application in routine surveys. As a result, a variety of standard methods have been suggested
and used for assessing water quality in different river orders, often without regard to a reference
to the natural state of the community. By contrast, recent approaches based on the concept of
ecosystem integrity and biodiversity are more promising, especially for integrated water
management.
The goal of the PAEQANN project was to develop general methodologies, based on advanced
modelling techniques, for predicting structure and diversity of key aquatic communities under
natural and man-made disturbances. This allowed the detection of the significance of various
environmental variables that structure these aquatic communities. These have been shown to
reveal predictable changes due to natural variability and human disturbances. Natural
conditions are described as undisturbed by human activities and man-made disturbances are
defined as various pollutants, discharge regulation, etc.
Such an approach to the analysis of aquatic communities made possible:
• to set up robust and sensitive ecosystem evaluation procedures that will work across a
large range of running water ecosystems throughout Europe,
o firstly, to point out the cause and effect relationships between environmental
conditions (physical, chemical, due to management actions) and certain relevant
aquatic communities (diatoms, macroinvertebrates, and fish)
o and then, to predict biocenosis structure in disturbed ecosystems, taking into
account all the relevant ecological variables
• to test ecosystem sensitivity to disturbance
EVK1-CT1999-00026 PAEQANN
Manual for users
4
•
to explore specific actions to be taken for restoration of ecosystem integrity
The long term aim of these investigations was therefore to help to define strategies for
conservation and restoration, compatible with local and regional development, and supported
by a strong scientific background.
The development of these general methodologies allowed:
1. the provision of predictive tools that can be easily applied to define the most effective
policies and institutional arrangements for resource management;
2. the application of the most effective and innovative techniques (mainly Artificial Neural
Networks) to identify problems in ecosystem functioning, resulting from ecosystem
degradation from human impact, and to model relevant biological resources;
3. the full exploitation of existing information, reducing the amount of field work (that is
both expensive and time consuming) needed in order to assess the health of freshwater
ecosystems;
4. the exploration of specific actions to be taken for restoration of ecosystem integrity;
5. the promotion of collaboration among scientists of different interested countries and
research fields, encouraging collaboration and dissemination of results and techniques.
2. Software Purpose
According to the main aim of the Paeqann project, we can describe the goal of this
programm as:
The main applied objective is to propose a set of tools for water management and water
policies in order to allow to easily assessing ecological quality and perturbations of stream
ecosystems. These tools will provide information about running water quality as well as
community structure. The assessment tools will allow identifying measures which should be
taken to restore biological integrity in running waters. Hopefully, the study can be considered
as a first step toward linking the improvement of water quality through specific management
measures (e.g. waste water treatment, habitat restoration, etc.) with the expected improvement in
ecological and biological value of running water systems.
Our program will allow you to consult the organism database and use the predicting
models for your own data.
3. System Requirmement
This program needs at least this configuration:
¾ Processor : 300 Mhz
¾ Ram : 32 Mo
¾ Display : 1024*768 16bits colors
¾ Os : Windows 98, Me, NT4, 2000, XP
But we recommand:
¾ Processor : 500 Mhz +
¾ Ram : 64 Mo +
¾ Os : Windows XP
EVK1-CT1999-00026 PAEQANN
Manual for users
5
4. Installation/Desinstallation
First of all, download PAEQANN.zip. Then run setup.exe. This file shows you all the steps
you need to install the software correctly. You can choose your installation path, and after let it
install alone. You do not need to restart your computer.
On some system you may experience “jet driver error” to fix it; you can download the
file daosdk.exe on the tools page of PAEQANN site. It wills fixe this problem.
If you want to unintall the software, two possibilities:
press start menu then choose PAEQANN folder and click uninstall
run control panel then click “Add or Remove Programs” and choose
PAEQANN.
5. HOW TO USE PAEQANN’s tools
5.1. The welcome screen
There are two ways to access the visualisation window for visualization of the dataset:
i. Selection of organism --> Selection of country ----> visualisation window
ii. Selection of country --> Selection of organism ----> visualisation window
The results of the models and the prediction of new data are accessible from the visualisation
window.
5.1.1. How to select an organism?
Three biological organisms significant to water quality assessment are considered: Fish,
Diatoms and Macroinvertebrates.
Click on the picture of your organism.
When an organism is selected, others become unavailable by changing the color (in this
example fish is selected).
5.1.2. How to select a country?
Eight European partners have collaborated in the PAEQANN project.
The flags of the countries with available data are activated.
Click on the flag corresponding to the country you want to explore.
EVK1-CT1999-00026 PAEQANN
Manual for users
6
When a country is selected, others become unavailable by changing the color.
After the couple selected, the visualisation window appears.
5.1.3. Command part
Cancel/Back: cancel the last action or return to the previous screen.
Help: clicking the help button on anytime while using the program activates the PAEQANN
help file and give the users direct context-related support.
EVK1-CT1999-00026 PAEQANN
Manual for users
7
Info: information about the PAEQANN's partners, programmers and project organizers.
Quit: exit the program.
5.2. Visualisation screen
5.2.1. Action part
Prediction: visualize results of predicting models and predicting new datasets.
Ordination: display classification of the datasets on a geographical map.
5.2.2. Visualisation part
Sampling sites are plotted on the geographic map with the water system network.
Clicking the left button of the mouse will display environmental variables and community
composition of the sampling site.
EVK1-CT1999-00026 PAEQANN
Manual for users
8
To see more precise position of the sampling site, use theses buttons:
: zoom in and zoom out
: move the geographic map on the left, down, right and up.
5.3. Prediction screen
The Prediction window:
In this part, community or species (Sp), functional feeding group (FFG), species richness (SR)
or guild, ... can be predicted using the available database of the PAEQANN project (visualising
results) or using new data (test new data).
5.3.1. Sp/FFG/SR/Guild/... --> visualising results
EVK1-CT1999-00026 PAEQANN
Manual for users
9
At the end of the Back Propagation modelling process, the sensitivity analysis using PaD
algorithm was applied to evaluate the contribution of the environmental variables for
predicting sp/FFG/SR/Guild/...
¾ Prediction results (Estimation vs. observation)
¾ Environmental variables
EVK1-CT1999-00026 PAEQANN
Manual for users
10
Available environmental variables using in the model.
¾ Partial derivatives of the ANN model response with respect to each environmental
variable, using PaD algorithm.
¾ Contribution of the environmental variables
This histogram is a classification of the environmental variables influence in percent, obtain
using PaD algorithm.
EVK1-CT1999-00026 PAEQANN
Manual for users
11
5.3.2. Sp/FFG/SR/Guild/... --> Test new data
¾ Input values
EVK1-CT1999-00026 PAEQANN
Manual for users
12
After the choice of the sp/FFG/SR/Guild/... to be predicted, user can enter new values of
environmental variables in the column "value" by simple click on the default value.
¾ Result
Before prediction of new data:
Result of new data prediction:
EVK1-CT1999-00026 PAEQANN
Manual for users
13
5.3.3. Community --> Visualising results
A site has first to be selected to see the observed community and the predicted community.
EVK1-CT1999-00026 PAEQANN
Manual for users
14
5.3.4. Community --> Test new data
EVK1-CT1999-00026 PAEQANN
Manual for users
15
Input values
EVK1-CT1999-00026 PAEQANN
Manual for users
16
¾ Prediction button: to predict the community, user can enter new values of
environmental variables in the column "value" by simple click on the default value and
press this button. The prediction results appear in the corresponding frame.
¾ Create data structure button: if the user want to predict the community of several sites,
the use of a file is convenient. Clicking on this button create the structure of the input
file needed by the program. This file can be open by any spreadsheet.
EVK1-CT1999-00026 PAEQANN
Manual for users
17
¾ Open input data button: when the input file is ready, clicking on this button open the
new data.
EVK1-CT1999-00026 PAEQANN
Manual for users
18
The following window displays the community prediction results of the selected file. As default,
the program display the results of the first sample (first line of the open file). The scrow list
offers the user to choose another sample.
Click on the "Save result" button create a file with all results.
5.4. Ordination screen
EVK1-CT1999-00026 PAEQANN
Manual for users
19
¾ Geographic map
This map displays the sampling sites with different colours which correspond to the clusters of
Self organizing map.
To see more precise position of the sampling site, use theses buttons:
: zoom in and zoom out
EVK1-CT1999-00026 PAEQANN
Manual for users
20
: move the geographic map on the left, down, right and up.
¾ Self organizing map
The Self organizing map was trained with community data, and then sampling sites were
classified (ordinate) on a SOM map.
The sampling sites can be visualised at two different cluster levels. At level 1, all units of the
map are visualised on the geographical map with different colours (types). At level 2, units of
the map are clustered in several groups according to their similarities (community similarity)
and the groups are visualised on the geographic map with different colours.
By cliking on a cell, more information is available.
If Site is on the scrow list, only the sites which are in the cell selected are on the geographic
map.
EVK1-CT1999-00026 PAEQANN
Manual for users
21
By cliking on the “All Sites” button, all sites are display on the map.
If Community is on the scrow list, the community structure of the selected cell appears.
¾ Environmental variables
After training the SOM with the community data, mean values of the environmental variables
were calculated in each unit of the SOM map.
The values were visualised on gray scale. Dark represents high value whereas light is low value.
In the scrow list displayed on the upper areas, user can choose concerning environmental
variables.
EVK1-CT1999-00026 PAEQANN
Manual for users
22
¾ Testing new data
User can test new community data with trained SOM by input corresponding values in the
column "density". The tested result is visualised on the corresponding unit of the SOM map
with a black circle.
EVK1-CT1999-00026 PAEQANN
Manual for users
23
The tested result:
EVK1-CT1999-00026 PAEQANN
Manual for users
24
6. Artificial Neural Networks (ANNs)
Over the last ten years, the use of Artificial Neural Networks (ANNs) in ecological modelling is
discussed (Colasanti, 1991), which resulted in different applications (e.g. Adams et al., Albiol et
al., 1995; Balls et al., 1996; Baran et al., 1996; Brey et al., 1996; Chon et al., 1996; French &
Recknagel, 1994; Komatsu et al., 1994; Lek et al., 1995, 1996a, 1996b, 1996c; Recknagel et al.,
1997; Scardi, 1996; Spitz et al., 1996). Typical ecological applications of ANNs include amongst
others:
¾ pattern recognition and classification in taxonomy (e.g. Nakano et al., 1991; Simpson et al.,
1992, 1993; Boddy et al., 2000),
¾ remote sensing (e.g. Carpenter et al., 1999; Civco, 1993; Gross et al., 1999; Keiner & Yan,
1998; Kimes et al., 1996; Mann & Benwell, 1996),
¾ GIS data analysis (e.g. Hilbert & Ostendorf, in press; Silveira et al., 1996).
¾ empirical models of ecological processes (Aoki & Komatsu, 1997; Aoki et al., 1999; Barciela
et al., 1999; Brey et al., 1996; Brey & Gerdes, 1998; Brosse et al., 1999; Lae et al., 1999;
Mastrorillo et al., 1997, 1998; Recknagel et al., 1997; Scardi, 1996, 2000; Scardi & Harding,
1999),
¾ tools for predicting community structure or population characteristics (Aussem & Hill,
1999; Baran et al., 1996; Guegan et al., 1998; Giske et al., 1998; Lek et al., 1996a, 1996; Scardi
et al., in press; Schleiter et al., 1999; Wagner et al., 2000)
¾ water management (e.g. Kastens & Featherstone, 1996),
¾ time series analysis and prediction (e.g. Recknagel, 1997; Chon et al., in press),
¾ ecosystem dynamics (e.g. with recurrent back-propagation algorithms by Pineda, 1987;
Chon et al., in press; Jeong et al., in press).
As for hydrological applications, although parametric statistical protocols and deterministic
models have been the traditional approaches in forecasting water quality variables in streams,
many recent efforts have shown that, when explicit information of hydrological subprocesses is
not available, ANNs can then be more effective (Zhu et al., 1994; Maier & Dandy, 2000).
A comprehensive overview of ANN applications in ecology and evolution is compiled by Lek &
Guegan (2000).
6.1. Self-Organizing Maps (SOMs)
6.1.1. Development of SOMs
The Self Organizing Map (SOM) algorithm has been proposed by Kohonen in the early
eighties (Kohonen, 1982). Since that time, SOMs have been used in a number of different
applications. As for ecological ones, SOMs have been used to reveal the relationships between
ecological communities, for instance to describe patterns in communities (Chon et al., 1996), to
analyse community data (Foody, 1999) or to model micro-satellite data (Giraudel et al., 2000).
SOMs are among the most well known neural networks with unsupervised learning rules. They
perform a topology-preserving projection of the data space onto a regular two-dimensional
space and can be used to effectively visualise clusters (Kohonen, 1995).
Already used in different areas, SOMs were able to recognize clusters in datasets where other
statistical algorithms failed to produce meaningful clusters (e.g. Cho, 1997). With the U-Matrix
EVK1-CT1999-00026 PAEQANN
Manual for users
25
method, SOMs can be used for clustering without prior knowledge of the number or size of the
clusters and then for studying Multivariate Time Series (Ultsch, 1999).
The detection and the visualisation of clusters are very straightforward and often outperform
the results obtained by classical classification methods. A drawback of the SOM algorithm is
that the size and the shape of the map have to be fixed in advance. Growing self-organising
networks have been proposed in order to deal with this problem (Villmann & Bauer, 1998), but
this approach remains still to be applied to ecological data.
6.1.2. The SOM concept
The Kohonen Self-Organizing Map (SOM) is one of the most popular unsupervised artificial
neural networks (Kohonen, 1995), it performs a topology-preserving projection of the data
space onto a regular two-dimensional space. SOM shares with the conventional ordination
methods the basic idea of displaying a high-dimensional dataset in a lower dimensional space
(usually a 2-dimensional space). This method is recommended for use in an exploratory
approach for data sets in which unexpected structures might be found. The complete
description of this method can be found in Kohonen (1995).
Let {X1 ;L, X p} be p vectors of Rn. With SOM, these p vectors will be projected in a non-linear
way onto a rectangular grid laid out on a hexagonal lattice with S hexagons: the Kohonen map
(figure 8.1).
Figure 8.1 A two-dimensional Self-Organizing Map. Each sphere symbolises each neuron in the input layer and in the
output layer (Kohonen map).
The goal of the SOM algorithm consists in putting the dataset on the map preserving the
neighbourhood, so the similar vectors should be mapped close together on the grid. For this
purpose, in each hexagon, a neuron will be considered. These neurons (mk )1≤k ≤S are in fact
reference vectors of Rn with components (wik )1≤i≤n,1≤k ≤S to be computed.
EVK1-CT1999-00026 PAEQANN
Manual for users
26
The modifications of the reference vectors are made through an Artificial Neural Network
(ANN). Modelling the human brain working, ANN has a learning ability: the components
(wik ) of each reference vector are computed during a training phase. The modifications of each
(wik ) take place by iterative adjustments. The Kohonen neural network consists of two layers:
the first one (input layer) is connected to each vector of the dataset, the second one (output
layer) forms a two-dimensional array of nodes (figure 8.1). In the output layer, the units of the
grid (reference vectors) give a representation of the distribution of the data set in an ordered
way. For learning, only input units are used, no expected-output data is given to the system: we
are referring to unsupervised learning.
6.1.3. Self-Organizing Map algorithm
¾
¾
¾
¾
¾
¾
Step 1: Epoch t=0, the reference vectors (mk )1≤k ≤S are initialised with random samples drawn
from the input dataset.
Step 2: A sample vector X j =(xij )1≤i≤n is randomly chosen as an input unit.
Step 3: The distances between X j and each reference vector are computed.
Step 4: The virtual unit mc closest to the input X j is chosen as the winning neuron. mc is
called the Best Matching Unit (BMU).
Step 5: The reference vectors (mk )1≤k ≤S are updated with the rule:
wik (t +1)=wik (t)+hck (t)[xij(t)−wik (t)] (1)
Step 6: Increase time t to t+1. If t<tmax then go to step 2 else stop the training.
In step 5, in the equation (1), the function hck (t) is called the neighbourhood function and
plays a very central role. During the learning process, the BMU defined in step 4 is not the
only updated unit. In the grid, a neighbourhood is defined around the BMU and all units
within this neighbourhood are updated. Several choices can be made for the definition of the
neighbourhood function. For instance, the neighbourhood can be written in terms of the
Gaussian function:
 rk − rc 

hc ( t ) =α ( t )⋅exp  −
 2 σ 2( t ) 


¾
¾
¾
(2)
rk −rc is the Euclidean distance on the map between the winning unit mc and each
reference vector mk .
σ is a decreasing function of time which defines the width of the part of the map affected
by the learning.
α is the ``learning-rate factor'', it is a decreasing function of the time.
and
both
converge towards 0.
The learning is broken down into two parts:
the ordering phase: during this phase, the virtual stations are widely modified in a large
neighbourhood of the Best Matching Unit. So, this occurs with large values for
and .
the tuning phase: when this second phase takes place, only the virtual units adjacent of the Best
Matching Unit are modified. This phase is much longer than the former one and
is
decreasing very slowly towards 0.
EVK1-CT1999-00026 PAEQANN
Manual for users
27
When the learning process is finished, the sample vectors can be mapped. For this purpose, the
BMU is computed for each sample vector and this one can be represented in the corresponding
hexagon.
EVK1-CT1999-00026 PAEQANN
Manual for users
28
6.2. Back Propagation Artificial Neural Network (ANNbp)
6.2.1. Introduction
The back propagation neural networks, also called multilayer feed-forward neural networks or
multilayer perceptron, are very popular and are used for a wide variety of problems more than
other types of neural networks. The ANNbp is based on the supervised procedure, i.e., the
network is built with a dataset where the outputs are known. ANNbp is a powerful system,
often capable of modelling complex relationships between variables. For a given input, one can
predict an output.
6.2.2. Back Propagation Artificial Neural Network algorithms
Structure
ANNbp is a layered feed-forward neural network, in which the non-linear elements (neurons)
are arranged in successive layers, and the information flows unidirectionally, from input layer
to output layer, through the hidden layer(s) (figure 9.1). As can be seen in this figure, neurons
from one layer are connected to all neurons in the adjacent layer, but no lateral connection
between neurons within one layer, or feedback connection are possible. The number of input
and output neurons depends on the number of explanatory and explained variables,
respectively. The hidden layer(s) is (are) an important parameter in the network.
Algorithms
ANNbp learning and update procedure is based on a relatively simple concept: if the network
gives the wrong answer, then the weights are corrected so that the error lessens, so future
responses of the network are more likely to be correct. The conceptual basis of the back
propagation algorithm was presented to a wide readership by Rumelhart et al. (1986).
In a training phase a set of input/target pattern pairs is used for training, which is presented to
the network many times. After training is stopped, the performance of the network is tested.
The ANNbp learning algorithm involves a forward-propagating step followed by a backwardpropagating step.
EVK1-CT1999-00026 PAEQANN
Manual for users
29
Hidden Layer
Input Layer
Output Layer
Figure 9.1 Schematic illustration of a three-layered feed-forward neural network, with one input layer, one hidden layer
and one output layer.
Forward-propagating step
Like a real neuron, the artificial neuron has many inputs, but only a single output, which can
stimulate many other neurons in the network. The neurons are numbered, for example the one
neuron in figure 9.2 is called j.
X0
X1
Wj0
Wj1
Neuron j
Xj
Wj2
X3
4
Wj
X4
Figure 9.2 Basic processing neuron in a network. Each input connection value (xi) is associated with a weight (wji).
The input the jth neuron receives from the ith neurons is indicated as x. Each connection to
the jth neuron is associated to a quantity called weight. The weight on the connection from the
ith neuron to the jth neuron is denoted wji. An input connection may be excitatory (positive
weight) or inhibitory (negative weight). A net input (called activation) for each neuron is the
sum of all its input values multiplied by their corresponding connection weights, expressed by
the formula:
EVK1-CT1999-00026 PAEQANN
Manual for users
30
a j =∑w ji xi +B j
i
where i is the total number of neurons in the previous layer, Bj is a bias term, which influences
the horizontal offset of the function. The bias Bj may be treated as the weight from the
supplementary input unit, which has a fixed output value of 1. Once the activation of the
neuron is calculated, we can determine the output value by applying a transfer function:
x j = f(a j)
We can use many different transfer functions, e.g. linear function, a threshold function, a
sigmoid function, etc. The sigmoid function is often used in ecology, its formula is:
x j = f ( a j ) = 1− a j
1+ e
The weight plays an important role in propagation of the signal in the network. They establish
a link between an input pattern and the associated output pattern, i.e. they contain the
knowledge of the neuronal network about the problem/solution relationship.
Backward-propagating step
The backward-propagating step begins with the comparison of the network output pattern to
the target value, when the difference (or error) is calculated. The backward-propagating step
then calculates error values and changes the incoming weights, starting with the output layer
and moving backward through the successive hidden layers.
The error signal associated with each processing unit indicates the amount of error associated
with that unit. This parameter is used during the weight-correction procedure, while learning is
taking place. A large value for the error signal indicates a large correction should be made to
the incoming weights; its sign reflects the direction in which the weights should be changed.
The adjustment of weight depends on three factors: the error value of the target unit, the
output value for the source unit and the learning rate. The learning rate commonly between 0
and 1, determine the rate of learning of the network.
Training the network
Before starting the training, the connection weights are set to small random values. Next the
input patterns are applied to the network to obtain the output. The differences between the
output calculations and the target expected are used to modify the weights. One complete
calculation is called an epoch or iteration. This processed is repeated until a suitable level of
error is achieved. Using a parameter called momentum, chosen generally between 0 and 1 allow
getting out of a local minimum.
Testing the network
A testing set of data serves to assess the performance of the network after training is complete.
The input patterns are fed into the network and the desired output patterns compared with
those given by the neural network. The agreement or the disagreement of these two sets gives an
indication of the performance of the neural network model. If it is possible, the best solution is
to divide the data set in the aim to use two different sets of data, one for the training and the
testing stage and the second one to validate the model (Mastrorillo et al. 1998). Different
partitioning procedures existed according to the size of the available dataset: k-fold crossvalidation or hold-out (Utans & Moody 1991; Efron & Tibshirani 1995; Kohavi & Wolpert
1996; Friedman 1997), leave-one-out (Efron, 1983; Kohavi 1995).
EVK1-CT1999-00026 PAEQANN
Manual for users
31
The network can be overtrained or overfitted, that is it loses its capacity to generalize. Three
parameters are responsible of this phenomenon: the number of epochs, the number of hidden
layers and the numbers of neurons in each hidden layer. It is very important to determine the
appropriate numbers of these elements in ANNbp modelling.
6.3. PaD algorithms
Two results can be obtained by this method. The first is a profile of the output variations for small
changes of each input variable and the second is being a classification of the relative contributions of
each variable to the network output.
To obtain the profile of the variations of the output for small changes of one input
variable, we compute the partial derivatives of the ANN output with respect to the input
(Dimopoulos et al., 1995; Dimopoulos et al., 1999). For a network with ni inputs, one hidden
layer with nh neurones, and one output (i.e. no=1), the partial derivatives of the output yj with
respect to input xj (with j=1,…,N and N the total number of observations) are:
nh
d ji = S j ∑ who I hj (1 − I hj )wih
h =1
(on the assumption that a logistic sigmoid function is used for the activation). When Sj is the derivative
of the output neuron with respect to its input, Ihj is the response of the hth hidden neuron, who and wih
are the weights between the output neuron and hth hidden neuron, and between the ith input neuron and
the hth hidden neuron.
A set of graphs of the partial derivatives versus each corresponding input variable can then be
plotted, and enable direct access to the influence of the input variable on the output. One example of
an interpretation of these graphs is that, if the partial derivative is negative then, for this value of the
studied variable, the output variable will tend to decrease while the input variable increases. Inversely, if
the partial derivatives are positive, the output variable will tend to increase while the input variable also
increases.
The second result of PaD concerns the relative contribution of the ANN output to the data set
with respect to an input. It is calculated by a sum of the square partial derivatives obtained per input
variable:
N
SSDi = ∑ ( d ji ) 2
j =1
One SSD (Sum of Square Derivatives) value is obtained per input variable. The SSD values
allow classification of the variables according to their increasing contribution to the output
variable in the model. The input variable that has the highest SSD value is the variable which
influences the output variable most.
7. Organisms
7.1. Fish
7.1.1. Belgium
The fish database of Belgium is composed of 804 streams-sampling stations. Four different
institutes sampled these stations: CSP (398 stations), FUNDP (153 stations), RIVO (36 stations),
IBW (217 stations).
EVK1-CT1999-00026 PAEQANN
Manual for users
32
7.1.1.1. Ordination
List of species used by the SOM:
Abrb abramis brama L.
Bream
Linnaeus, 1758
Alba
Alburnus Alburnus L.
Bleak
Linnaeus, 1758
Albb
Alburnus bipunctatus Bloch
Schneider
Bloch, 1782
Ictn
Ictalurus nebulosus Le Sueur
Brown bullhead
Anga Anguilla anguilla L.
European Eel
Aspa
Aspius aspius L.
Asp
Barb
Barbus barbus L.
Barbel
Noeb Noemacheilus barbatulus L.
Stone loach
Blib
Blicca bjoerkna L.
Silver bream
Cara
Carassius auratus L.
Goldfish
EVK1-CT1999-00026 PAEQANN
Manual for users
Linnaeus, 1758
Linnaeus, 1758
33
Carc
Carassius carassius L.
Crucian Carp
Linnaeus, 1758
Chon Chondrostoma nasus L.
Common Nase
Linnaeus, 1758
Cobt Cobitis taenia L.
Spiny loach
Cotg Cottus gobio L.
Bullhead
Cypc Cyprinus carpio L.
Common Carp
Linnaeus, 1758
Esol
Esox lucius L.
European Pike
Linnaeus, 1758
Gasa
Gasterosteus aculeatus L.
Three-spined Stickleback Linnaeus, 1758
Gobg Gobio gobio L.
Gudgeon
Linnaeus, 1758
Gymc Gymnocephalus cernua L.
Ruffe
Linnaeus, 1758
Ictp
Ictalurus punctatus Rafinesque Black bullhead
Lamp Lampetra planeri Bloch
Brook Lamprey
Bloch, 1784
Lepg
Lepomis gibbosus L.
Pumpkinseed
Linnaeus, 1758
Leuc
Leuciscus cephalus L.
Chub
Linnaeus, 1758
Leud Leucaspius delineatus L.
Rain bleak / Sunbleak
Leui
Leuciscus idus L.
Ide
Leul
Leuciscus leuciscus L.
Dace
Linnaeus, 1758
Lotl
Lota lota L.
Burbot
Lacepède, 1802
Misf
Misgurnus fossilis L.
Pond loach
Oncm Oncorhyncus mykiss Walbaum Rainbow trout
Osme Osmerus eperlanus L.
Smelt
Perf
Perch
Linnaeus, 1758
Phop Phoxinus phoxinus L.
Minnow
Linnaeus, 1758
Plaf
Platichtys flesus L.
Flounder
Psep
Pseudorasbora parva Schlegel
Perca fluviatilis L.
Punp Pungitius pungitius L.
Nine-spined Stickleback Linnaeus, 1758
Rhos Rhodeus sericeus amarus Bloch Bitterling
Pallas, 1776
Rutr
Rutilus rutilus L.
Roach
Linnaeus, 1758
Sala
Salvelinus alpinus L.
Artic char
Salf
Salmo trutta fario L.
Brown Trout
Salv
Salvelinus fontinalis Mitchill
Brook trout
Sals
Salmo salar L.
Atlantic Salmon
Salt
Salmo trutta trutta L.
Sea trout
Scae
Scardinius erythrophthalmus L. Rudd
Sanl
Sander lucioperca L.
Sander
Thyt
Thymallus thymallus L.
Grayling
Linnaeus, 1758
Tint
Tinca tinca L.
Tench
Linnaeus, 1758
Umbp Umbra pygmaea De Kay
EVK1-CT1999-00026 PAEQANN
Linnaeus, 1758
Linnaeus, 1758
Linnaeus, 1758
Stone moroko
Manual for users
34
¾ Geographic map
This map displays the sampling sites with different colours which correspond to the clusters of
Self organizing map.
To see more precise position of the sampling site, use theses buttons:
: zoom in and zoom out
EVK1-CT1999-00026 PAEQANN
Manual for users
35
: move the geographic map on the left, down, right and up.
¾ Self organizing map
The Self organizing map was trained with community data, and then sampling sites were
classified (ordinate) on a SOM map.
The sampling sites can be visualised at two different cluster levels. At level 1, all units of the
map are visualised on the geographical map with different colours (types). At level 2, units of
the map are clustered in several groups according to their similarities (community similarity)
and the groups are visualised on the geographic map with different colours.
By cliking on a cell, more information is available.
If Site is on the scrow list, only the sites which are in the cell selected are on the geographic
map.
EVK1-CT1999-00026 PAEQANN
Manual for users
36
By cliking on the “All Sites” button, all sites are display on the map.
If Community is on the scrow list, the community structure of the selected cell appears.
¾ Environmental variables
After training the SOM with the community data, mean values of the environmental variables
were calculated in each unit of the SOM map.
The values were visualised on gray scale. Dark represents high value whereas light is low value.
In the scrow list displayed on the upper areas, user can choose concerning environmental
variables.
EVK1-CT1999-00026 PAEQANN
Manual for users
37
¾ Testing new data
User can test new community data with trained SOM by input corresponding values in the
column "density". The tested result is visualised on the corresponding unit of the SOM map
with a black circle.
EVK1-CT1999-00026 PAEQANN
Manual for users
38
The tested result:
7.1.1.2. Prediction
List of environmental variables used by BP:
EVK1-CT1999-00026 PAEQANN
Manual for users
39
Name
Unit
Type
Source
surface area
M²
Field
FUNDP, CSP,
RIVO or IBW
river width
Meter
Field
FUNDP, CSP,
RIVO or IBW
On map
FUNDP, CSP,
RIVO or IBW
On map
FUNDP, CSP,
RIVO or IBW
Distance from source
Km
of the station
On map
FUNDP, CSP,
RIVO or IBW
Altitude of the station Meter
On map
FUNDP, CSP,
RIVO or IBW
Temperature in
January
°C
Laboratory
FUNDP, CSP,
RIVO or IBW
Temperature of july
°C
Laboratory
FUNDP, CSP,
RIVO or IBW
Mean of hqual and
wqual
FUNDP, CSP,
RIVO or IBW
watershed area
Slope of the station
‰
general quality
The Prediction window:
In this part, community or species (Sp), functional feeding group (FFG), species richness (SR)
or guild, ... can be predicted using the available database of the PAEQANN project (visualising
results) or using new data (test new data).
1. Sp/FFG/SR/Guild/... --> visualising results
EVK1-CT1999-00026 PAEQANN
Manual for users
40
At the end of the Back Propagation modelling process, the sensitivity analysis using PaD
algorithm was applied to evaluate the contribution of the environmental variables for
predicting sp/FFG/SR/Guild/...
¾ Prediction results (Estimation vs. observation)
EVK1-CT1999-00026 PAEQANN
Manual for users
41
¾ Environmental variables
Available environmental variables using in the model.
¾ Partial derivatives of the ANN model response with respect to each environmental
variable, using PaD algorithm.
¾ Contribution of the environmental variables
This histogram is a classification of the environmental variables influence in percent, obtain
using PaD algorithm.
EVK1-CT1999-00026 PAEQANN
Manual for users
42
2. Sp/FFG/SR/Guild/... --> Test new data
¾ Input values
EVK1-CT1999-00026 PAEQANN
Manual for users
43
After the choice of the sp/FFG/SR/Guild/... to be predicted, user can enter new values of
environmental variables in the column "value" by simple click on the default value.
¾ Result
Before prediction of new data:
Result of new data prediction:
3. Community --> Visualising results
EVK1-CT1999-00026 PAEQANN
Manual for users
44
A site has first to be selected to see the observed community and the predicted community.
4. Community --> Test new data
EVK1-CT1999-00026 PAEQANN
Manual for users
45
Input values
EVK1-CT1999-00026 PAEQANN
Manual for users
46
¾ Prediction button: to predict the community, user can enter new values of
environmental variables in the column "value" by simple click on the default value and
press this button. The prediction results appear in the corresponding frame.
¾ Create data structure button: if the user want to predict the community of several sites,
the use of a file is convenient. Clicking on this button create the structure of the input
file needed by the program. This file can be open by any spreadsheet.
EVK1-CT1999-00026 PAEQANN
Manual for users
47
¾ Open input data button: when the input file is ready, clicking on this button open the
new data.
EVK1-CT1999-00026 PAEQANN
Manual for users
48
The following window displays the community prediction results of the selected file. As default,
the program display the results of the first sample (first line of the open file). The scrow list
offers the user to choose another sample.
Click on the "Save result" button create a file with all results.
7.1.2. France
The database is a set coming from the database held by the Conseil Supérieur de la Pêche
(Banque Hydrobiologique et Piscicole), covering a 13 years survey period. The base is a set of
688 reference sites, which are fairly evenly distributed across French rivers and which contain
the occurrence of 40 most common fish species of France in relation to regional and local
EVK1-CT1999-00026 PAEQANN
Manual for users
49
environmental factors. The data were sampled during 1985-98 by the Conseil Supérieur de la
Pêche by electrofishing during low-flow periods (from August to October). Fish were
identified to species in the field, and released to the water.
7.1.2.1 Ordination
List of species used by the SOM:
Code Latin Name
Common Name
Authors
Abra Abramis brama
Carp bream
Linnaeus, 1758
Abrs Abramis spp.
Bream
Linnaeus, 1758
Albb Alburnoides bipunctatus
Schneider
Bloch, 1782
Alba Alburnus alburnus
Bleak
Linnaeus, 1758
Anga Anguilla anguilla
European eel
Linnaeus, 1758
Barb Barbus barbus
Barbel
Linnaeus, 1758
Barm Barbus meridionalis
Mediterranean barbel
Risso, 1826
Blef
Freshwater blenny
Asso, 1801
Cara Carassius auratus
Goldfish
Linnaeus, 1758
Carc Carassius carassius
Crucian carp
Linnaeus, 1758
Chon Chondrostoma nasus
Common nase
Linnaeus, 1758
Chot Chondrostoma toxostom
Soiffe
Vallot, 1837
Cotg Cottus gobio
Bullhead
Linnaeus, 1758
Cypc Cyprinus carpio
Common carp
Linnaeus, 1758
Esol
European pike
Linnaeus, 1758
Gama Gambusia affinis
Mosquitofish
Baird and Girard, 1853
Gasa Gasterosteus aculeatus
Three-spined stickleback Linnaeus, 1758
Gobg Gobio gobio
Gudgeon
Linnaeus, 1758
Gymc Gymnocephalus cernua
Ruffe
Linnaeus, 1758
Ictm Ictalurus melas
Black bullhead
Rafinesque, 1820
Lamp Lampetra planeri
Brook lamprey
Bloch, 1784
Lepg Lepomis gibbosus
Pumpkinseed
Linnaeus, 1758
Leud Leucaspius delineatus
Belica
Heckel, 1843
Leuc Leuciscus cephalus
Chub
Linnaeus, 1758
Leul Leuciscus leuciscus
Dace
Linnaeus, 1758
Leus Leuciscus soufia
Blageon
Risso, 1826
Lotl
Burbot
Lacepède, 1802
Mics Micropterus salmoides
Largemouth bass
Lacepède, 1802
Nemb Nemacheilus barbatulus
Stone loach
Linnaeus, 1758
Perf
Perch
Linnaeus, 1758
Phop Phoxinus phoxinus
Minnow
Linnaeus, 1758
Punp Pungitus pungitius
Nine-spined stickleback Linnaeus, 1758
Blennius fluviatilis
Escox lucius
Lota lota
Perca fluviatilis
EVK1-CT1999-00026 PAEQANN
Manual for users
50
Rhos Rhodeus sericeus
Bitterling
Pallas, 1776
Rutr
Rutilus rutilus
Roach
Linnaeus, 1758
Sals
Salmo salar
Atlantic salmon
Linnaeus, 1758
Salf
Salmo trutta fario
Brown trout
Linnaeus, 1758
Scae Scardinius erythrophthalmus Rudd
Linnaeus, 1758
Stil
Zander
Linnaeus, 1758
Thyt Thymallus thymallus
Grayling
Linnaeus, 1758
Tint
Tench
Linnaeus, 1758
Stizostedion lucioperca
Tinca tinca
¾ Geographic map
EVK1-CT1999-00026 PAEQANN
Manual for users
51
This map displays the sampling sites with different colours which correspond to the clusters of
Self organizing map.
To see more precise position of the sampling site, use theses buttons:
: zoom in and zoom out
: move the geographic map on the left, down, right and up.
¾ Self organizing map
The Self organizing map was trained with community data, and then sampling sites were
classified (ordinate) on a SOM map.
The sampling sites can be visualised at two different cluster levels. At level 1, all units of the
map are visualised on the geographical map with different colours (types). At level 2, units of
the map are clustered in several groups according to their similarities (community similarity)
and the groups are visualised on the geographic map with different colours.
EVK1-CT1999-00026 PAEQANN
Manual for users
52
By cliking on a cell, more information is available.
If Site is on the scrow list, only the sites which are in the cell selected are on the geographic
map.
By cliking on the “All Sites” button, all sites are display on the map.
If Community is on the scrow list, the community structure of the selected cell appears.
EVK1-CT1999-00026 PAEQANN
Manual for users
53
¾ Environmental variables
After training the SOM with the community data, mean values of the environmental variables
were calculated in each unit of the SOM map.
The values were visualised on gray scale. Dark represents high value whereas light is low value.
In the scrow list displayed on the upper areas, user can choose concerning environmental
variables.
¾ Testing new data
EVK1-CT1999-00026 PAEQANN
Manual for users
54
User can test new community data with trained SOM by input corresponding values in the
column "density". The tested result is visualised on the corresponding unit of the SOM map
with a black circle.
The tested result:
EVK1-CT1999-00026 PAEQANN
Manual for users
55
7.1.2.2 Prediction
List of environmental variables used by BP:
Name
Unit
Type
Source
Total catchment area
km2
of the basin
Measured
using a digital
planimeter on CSP
a 1:1000 000scale map
Surface area of the
catchment
km²
Measured
using a digital
planimeter on CSP
a 1:1000 000scale map
Distance from
headwater sources
km
Measured
using a digital
planimeter on CSP
a 1:1000 000scale map
Width
m
Field
CSP
Gradient
‰
Derived from
topographic
maps
CSP
Depth
m
Field
CSP
Sampling surface
m²
/
CSP
Altitude
m
Derived from
topographic
CSP
EVK1-CT1999-00026 PAEQANN
Manual for users
56
maps
Temperature in July °C
Meteorological
office for the
mean monthly
values and
CSP
interpolation
on the basis of
the National
Grid
Temperature in
January
°C
Meteorological
office for the
mean monthly
values and
CSP
interpolation
on the basis of
the National
Grid
°C
Meteorological
office for the
mean monthly
values and
CSP
interpolation
on the basis of
the National
Grid
Air temperature
The Prediction window:
In this part, community or species (Sp), functional feeding group (FFG), species richness (SR)
or guild, ... can be predicted using the available database of the PAEQANN project (visualising
results) or using new data (test new data).
5.3.1. Sp/FFG/SR/Guild/... --> visualising results
EVK1-CT1999-00026 PAEQANN
Manual for users
57
At the end of the Back Propagation modelling process, the sensitivity analysis using PaD
algorithm was applied to evaluate the contribution of the environmental variables for
predicting sp/FFG/SR/Guild/...
¾ Prediction results (Estimation vs. observation)
¾ Environmental variables
EVK1-CT1999-00026 PAEQANN
Manual for users
58
Available environmental variables using in the model.
¾ Partial derivatives of the ANN model response with respect to each environmental
variable, using PaD algorithm.
¾ Contribution of the environmental variables
This histogram is a classification of the environmental variables influence in percent, obtain
using PaD algorithm.
EVK1-CT1999-00026 PAEQANN
Manual for users
59
5.3.2. Sp/FFG/SR/Guild/... --> Test new data
¾ Input values
EVK1-CT1999-00026 PAEQANN
Manual for users
60
After the choice of the sp/FFG/SR/Guild/... to be predicted, user can enter new values of
environmental variables in the column "value" by simple click on the default value.
¾ Result
Before prediction of new data:
Result of new data prediction:
EVK1-CT1999-00026 PAEQANN
Manual for users
61
5.3.3. Community --> Visualising results
A site has first to be selected to see the observed community and the predicted community.
EVK1-CT1999-00026 PAEQANN
Manual for users
62
5.3.4. Community --> Test new data
EVK1-CT1999-00026 PAEQANN
Manual for users
63
Input values
EVK1-CT1999-00026 PAEQANN
Manual for users
64
¾ Prediction button: to predict the community, user can enter new values of
environmental variables in the column "value" by simple click on the default value and
press this button. The prediction results appear in the corresponding frame.
¾ Create data structure button: if the user want to predict the community of several sites,
the use of a file is convenient. Clicking on this button create the structure of the input
file needed by the program. This file can be open by any spreadsheet.
EVK1-CT1999-00026 PAEQANN
Manual for users
65
¾ Open input data button: when the input file is ready, clicking on this button open the
new data.
EVK1-CT1999-00026 PAEQANN
Manual for users
66
The following window displays the community prediction results of the selected file. As default,
the program display the results of the first sample (first line of the open file). The scrow list
offers the user to choose another sample.
Click on the "Save result" button create a file with all results.
7.1.3. Garonne
Relatively little work has been carried out to describe fish assemblages in the Garonne river
compared to other large rivers of France. This database has thus been carried out to pattern the
fish richness in the Garonne river basin. The data were collected between 1986 and 1996 by
the laboratoire d'ingénierie agronomique, ENSAT and the Paul Sabatier University of
EVK1-CT1999-00026 PAEQANN
Manual for users
67
Toulouse. All sites were sampled once by electrofishing, during low-flow periods. Fish were
identified to species in the field. The base is a set of 239 reference sites fairly evenly
distributed across the Garonne river basin containing the presence or absence of 44 fish
species in relation to environmental factors. The database constituted of 6 environmental
variables, 3 of which being geographical variables and the other 3 being physical variables
7.1.3.1 Ordination
List of species used by the SOM:
Code Latin Name
Common Name
Authors
Alba Alburnus alburnus
Bleak
Linnaeus, 1758
Aloa Alosa alosa
Allis Shad
Linnaeus, 1758
Alof Alosa fallax
Twaite Shad
Lacepède, 1803
Anga Anguilla anguilla
European Eel
Linnaeus, 1758
Barb Barbus barbus
Barbe
Linnaeus, 1758
Blef
Freshwater Blenny
Asso, 1801
Mics Micropterus salmoides
Large-Mouthed Bass
Lacepède, 1802
Rhos Rhodeus sericeus
Bitterling
Pallas, 1776
Abrb Abramis brama
Common Brea
Linnaeus, 1758
Blib
Blicca bjokna
Silver Bream
Linnaeus, 1758
Esol
Esox lucius
European Pike
Linnaeus, 1758
Carc Carassius carassius
Crucian Carp
Linnaeus, 1758
Cypc Cyprinus carpio
Common Carp
Linnaeus, 1758
Cotg Cottus gobio
Bullhead
Linnaeus, 1758
Leuc Leuciscus cephalus
Chub
Linnaeus, 1758
Gasa Gasterosteus aculeatus
Three-spined Stickleback Linnaeus, 1758
Punp Pungitus pungitius
Nine-spined Stickleback Linnaeus, 1758
Plaf
Flounder
Linnaeus, 1758
Gama Gambusia affinis
Mosquitofish
Baird & Girard, 1853
Rutr
Roach
Linnaeus, 1758
Gobg Gobio gobio
Gudgeon
Linnaeus, 1758
Gymc Gymnocephalus cernua
Ruffe
Linnaeus, 1758
Chon Chondrostoma nasus
Common Nase
Linnaeus, 1758
Nemb Nemacheilus barbatulus
Stone loach
Linnaeus, 1758
Petm Petromyson marinus
Sea Lamprey
Linnaeus, 1758
Lamp Lampetra planeri
Brook Lamprey
Bloch, 1784
Mugc Mugil cephalus
Mullet
Linnaeus, 1758
Sala
Arctic Charr
Linnaeus, 1758
Pacp Pachychilon pictum
Albanian roach
Heckel & Kner, 1858
Ictm Ictalurus melas
Black Bullhead
Rafinesque, 1820
Perf
Perch
Linnaeus, 1758
Blennius fluviatilis
Platichthys flesus
Rutilus rutilus
Salvelinus alpinus
Perca fluviatilis
EVK1-CT1999-00026 PAEQANN
Manual for users
68
Lepg Lepomis gibbosus
Pumpkinseed
Linnaeus, 1758
Psep Pseudorasbora parva
False Harlequin
Temminch et Schlegel, 1842
Scae Scardinius erythrophtalmus Rudd
Linnaeus, 1758
Salf
Salvelinus fontinalis
Brook Charr
Mitchill, 1815
sStil
Stizostedion lucioperca
Zander
Linnaeus, 1758
Sals
Salmo salar
Atlantic Salmon
Linnaeus, 1758
Oncm Oncorhynchus mykiss
Rainbow Trout
Walbaum, 1792
Tint
Tench
Linnaeus, 1758
Chot Chondrostoma toxostoma
Soiffe
Vallot, 1837
Salf
Salmo trutta fario
Brown Trout
Linnaeus, 1758
Salt
Salmo trutta trutta
Sea Trout
Linnaeus, 1758
Phop Phoxinus phoxinus
Minnow
Linnaeus, 1758
Leul Leuciscus leuciscus
Dace
Linnaeus, 1758
Tinca tinca
7.1.3.2 Prediction
List of environmental variables used by BP:
Name
Unit Type
Source
Altitude of the
sampling site
m
Recorded on several 1/25 000 scale maps
ENSAT/
University
Distance from the
source
km Recorded on several 1/25 000 scale maps
ENSAT/
University
Catchment area
km²
Measured with a digital planimeter on a 1/500 000 ENSAT/
scale map of the Garonne river basin
University
The Prediction window:
In this part, community or species (Sp), functional feeding group (FFG), species richness (SR)
or guild, ... can be predicted using the available database of the PAEQANN project (visualising
results) or using new data (test new data).
EVK1-CT1999-00026 PAEQANN
Manual for users
69
5.3.1. Sp/FFG/SR/Guild/... --> visualising results
At the end of the Back Propagation modelling process, the sensitivity analysis using PaD
algorithm was applied to evaluate the contribution of the environmental variables for
predicting sp/FFG/SR/Guild/...
¾ Prediction results (Estimation vs. observation)
EVK1-CT1999-00026 PAEQANN
Manual for users
70
¾ Environmental variables
Available environmental variables using in the model.
¾ Partial derivatives of the ANN model response with respect to each environmental
variable, using PaD algorithm.
¾ Contribution of the environmental variables
EVK1-CT1999-00026 PAEQANN
Manual for users
71
This histogram is a classification of the environmental variables influence in percent, obtain
using PaD algorithm.
5.3.2. Sp/FFG/SR/Guild/... --> Test new data
¾ Input values
EVK1-CT1999-00026 PAEQANN
Manual for users
72
After the choice of the sp/FFG/SR/Guild/... to be predicted, user can enter new values of
environmental variables in the column "value" by simple click on the default value.
¾ Result
Before prediction of new data:
Result of new data prediction:
EVK1-CT1999-00026 PAEQANN
Manual for users
73
5.3.3. Community --> Visualising results
A site has first to be selected to see the observed community and the predicted community.
EVK1-CT1999-00026 PAEQANN
Manual for users
74
5.3.4. Community --> Test new data
EVK1-CT1999-00026 PAEQANN
Manual for users
75
Input values
EVK1-CT1999-00026 PAEQANN
Manual for users
76
¾ Prediction button: to predict the community, user can enter new values of
environmental variables in the column "value" by simple click on the default value and
press this button. The prediction results appear in the corresponding frame.
¾ Create data structure button: if the user want to predict the community of several sites,
the use of a file is convenient. Clicking on this button create the structure of the input
file needed by the program. This file can be open by any spreadsheet.
EVK1-CT1999-00026 PAEQANN
Manual for users
77
¾ Open input data button: when the input file is ready, clicking on this button open the
new data.
EVK1-CT1999-00026 PAEQANN
Manual for users
78
The following window displays the community prediction results of the selected file. As default,
the program display the results of the first sample (first line of the open file). The scrow list
offers the user to choose another sample.
Click on the "Save result" button create a file with all results.
7.1.4. Italy
The fish database from the provinces of Vicenza and Belluno (NE Italy) includes 264
sampling sites. For each site 21 environmental variables were recorded: elevation in m, mean
depth, % of surface for runs, pools and riffles (3 variables), mean width in m, % of surface
with boulders, rocks and pebbles, gravel, sand, silt and clay (5 variables), stream velocity
EVK1-CT1999-00026 PAEQANN
Manual for users
79
(score 0-5), % of surface with vegetation covering, % of surface with shadow, anthropic
disturbance (score 0-4), water temperature, pH, conductivity in S cm-1, gradient in ‰,
distance from headwater source in km, surface of drainage basin in km². Fish community is
represented by 36 species (presence/absence data only). These data were obtained from two
small consulting cooperative enterprises (Aquaprogram scrl and Bioprogram scrl), that
collected them on behalf of the provincial Administration from 1987 to 1994. Other data will
be available in the next months from both existing databases and new field records.
7.1.4.1 Ordination
List of species used by the SOM:
Code Latin Name
Common Name
Authors
Abrab Abramis brama
Common Bream
Linnaeus, 1758
Alba Alburnus alburnus alborella
Bleak
De Filippi, 1844
Anga Anguilla anguilla
European Eel
Linnaeus, 1758
Barm Barbus meridionalis
Meriditerranean Barbel
Risso, 1826
Barp Barbus plebejus
Italian Barbel
Bonaparte, 1839
Carc Carassius carassius
Crucian Carp
Linnaeus, 1758
Chog Chondrostoma genei
South Europe Nase
Bonaparte, 1839
Cobt Cobitis taenia
Spined loach
Linnaeus, 1758
Cotg Cottus gobio
Bullhead
Linnaeus, 1758
Cypc Cyprinus carpio
Common Carp
Linnaeus, 1758
Esol
European Pike
Linnaeus, 1758
Gamh Gambusia holbrooki
Eastern mosquitofish
Girard, 1859
Gasa Gasterosteus aculeatus
Three-spined Stickleback
Linnaeus, 1758
Gobg Gobio gobio
Gudgeon
Linnaeus, 1758
Ictm
Black Bullhead
Rafinesque,
1820
Lamp Lampetra planeri
Brook Lamprey
Bloch, 1784
Lepg Lepomis gibbosus
Pumpkinseed
Linnaeus, 1758
Leuc Leuciscus cephalus
Chub
Linnaeus, 1758
Leus Leuciscus souffia
Blageon
Risso, 1826
Mics Micropterus salmoides
Large-Mouthed Bass
Lacepede, 1802
Oncm Oncorhynchus mykiss
Rainbow Trout
Walbaum, 1792
Orsp Orsinigobius punctatissimus
(Italian name: Panzarolo)
Canestrini, 1864
Padm Padogobius martensii
(Italian name: Ghiozzo di
fiume)
Günther, 1861
Perf
Perch
Linnaeus, 1758
Phop Phoxinus phoxinus
Minnow
Linnaeus, 1758
Rute Rutilus erythrophthalmus
(Italian name: Triotto)
Zerunian, 1982
Esox lucius
Ictalurus melas
Perca fluviatilis
EVK1-CT1999-00026 PAEQANN
Manual for users
80
Sabl
Sabanejewia larvata
Italian Loach
DeFilippi, 1859
Salm Salmo (trutta) marmoratus
Marble Trout
Cuvier, 1829
Salh
Salmo trutta hybr.
trutta/marmoratus
Sea Trout - Marble Trout
hybrid
Cuvier, 1829
Salt
Salmo (trutta) trutta
Sea Trout
Linnaeus, 1758
Salv
Salvelinus fontinalis
Brook Char
Mitchill, 1815
Scae Scardinius erythrophthalmus
Rudd
Linnaeus, 1758
Thyt Thymallus thymallus
Grayling
Linnaeus, 1758
Tint
Tench
Linnaeus, 1758
Tinca tinca
7.1.4.2 Prediction
List of environmental variables used by BP:
Name
Unit
Type Source
Elevation
m
Field Provincial Administration publications
Mean depth
m
Field Provincial Administration publications
Runs
surface, % Field Provincial Administration publications
Pools
surface, % Field Provincial Administration publications
Riffles
surface, % Field Provincial Administration publications
Mean width
m
Boulders
surface, % Field Provincial Administration publications
Rocks and peebles
surface, % Field Provincial Administration publications
Gravel
surface, % Field Provincial Administration publications
Sand
surface, % Field Provincial Administration publications
Silt and clay
surface, % Field Provincial Administration publications
Stream velocity
score, 0-5 Field Provincial Administration publications
Vegetation covering
surface, % Field Provincial Administration publications
Shade
%
Field Provincial Administration publications
Field Provincial Administration publications
Anthropic disturbance score, 0-4 Field Provincial Administration publications
pH
Field Provincial Administration publications
Conductivity
mS/cm
Field Provincial Administration publications
Gradient
%
Map Italian Army Geographical Institute
Catchment area surface km2
Map Italian Army Geographical Institute
Distance from source
Map Italian Army Geographical Institute
EVK1-CT1999-00026 PAEQANN
km
Manual for users
81
7.1.5. Luxembourg
The fish database of Luxembourg is composed of 34 stream-sampling stations. These stations were
sampled from 1994 to 1996. Diatom and macroinvertebrates were also sampled in those sites during
the same years. The Water and Forest Administration helped the CRP-GL to sample the fishes. The
fishes were sampled in the framework of the biocenotic study of the rhithral part of the Luxemburg
streams.
Physical variables (geographical position, altitude, geology, source distance, slope, river width, water
level, speed current, shading, temperature) were measured by the Centre de Recherche Public-Gabriel
Lippmann (CRP-GL) in the field and on maps (1/25000). Some chemical variables were also
measured in the field by the CRP-GL (pH, oxygen, conductivity), and in laboratory (PO42-). The other
variables were measured by the Ministère de l’Environnement (BOD5, NH4+, NO2-, NO3-, carbonate
hardness, total hardness, Cl-, Ptot, SO42-, K+).
7.1.5.1 Ordination
List of species used by the SOM:
Code Latin species name
Common name
Authors
Alba Alburnus alburnus
Bleak
De Filippi, 1844
Anga Anguilla anguilla
European Eel
Linnaeus, 1758
Cobt Cobitis taenia
Spined loach
Linnaeus, 1758
Cotg Cottus gobio
Bullhead
Linnaeus, 1758
Gasa Gasterosteus aculeatus Three-spined Stickleback Linnaeus, 1758
Lama Lampetra planeri
Brook Lamprey
Bloch, 1784
Nemb Nemachelius barbatulus Loach
Linnaeus, 1758
Oncm Oncorhyncus mykiss
Rainbow Trout
Walbaum, 1792
Phop Phoxinus phoxinus
Minnow
Linnaeus, 1758
Salf
Sea Trout
Linnaeus, 1758
Salmo trutta f. fario
7.2. Macroinvertebrate
7.2.1. Austria
The Upper Austrian macroinvertebrate database includes 225 sampling sites. All sampling
sites are within the Danube basin. The samples were taken between 1996 and 1999, once per
site, synchronously with diatom and ciliata samples. All samples and measurements were
taken by the Upper Austrian water authority (UAWA). Sampling of organism is part of a long
term survey program of the water authority. Due to some geographical differences, only data
from 102 sites can be used for combination with biological data. As a result of the agreement
on data combination within the PAEQANN project only a reduced number of sites could be
used for further calculation (77). The database includes 536 macroinvertebrate taxa and 25
chemical and physical variables (see parameter list in paragraph 2.1). Additional structural
data will be collected during the following months.
EVK1-CT1999-00026 PAEQANN
Manual for users
82
7.2.1.1 Ordination
7.2.1.2 Prediction
7.2.2. France
At the present time, the matrix contains 425 taxa with different levels of identification from
157 running-water sites. A total of 5 environmental variables are associated with each
sampling site:
•
•
•
•
•
“N° d’ordre” = Stream Order (Strahler classification), obtained from maps
“Altitude” = Elevation (metres), mostly obtained from maps
“Pente” = Slope (0/00), mostly obtained from maps
“ Température max” = annual maximal water temperature (°C)
“ Distance source” = Distance to source (kilometres), mostly obtained from maps
7.2.2.1 Ordination
7.2.2.2 Prediction
7.2.3. Luxembourg
The macroinvertebrate database of Luxembourg includes 147 different sampling stations.
Macroinvertebrate samplings were done in the same sampling sites as the diatoms samplings, and on
the same date. Each station was sampled twice: the first time being in autumn, and the second time in
spring. The samplings were carried out from 1994 to 1997. The macroinvertebrate database of
Luxembourg is composed of 292 records. The determinations are for all the taxonomic groups made at
a specific level. These records came from the small Luxembourg streams and were sampled in the
framework of the biocenotic study of the rhithral part of these streams. This study has 2 aims: to
establish the most complete faunistic inventory (to a specific level), and to provide a quantitative and
qualitative analysis with chemical and biological indices.
Physical variables (geographical position, altitude, geology, source distance, slope, river width, water
level, speed current, shading, temperature) were measured by the Centre de Recherche Public-Gabriel
Lippmann (CRP-GL) in the field and on maps (1/25000). Some chemical variables were also
measured in the field by the CRP-GL (pH, oxygen, conductivity), and in the laboratory (PO42-). The
other variables were measured by the Ministère de l’Environnement (BOD5, NH4+, NO2-, NO3-,
carbonate hardness, total hardness, Cl-, Ptot, SO42-, K+).
7.2.31 Ordination
7.2.32 Prediction
7.2.4. The Netherlands
The Dutch database consists of samples which were collected from 664 sites situated in the
province of Overijssel (The Netherlands); 609 sites were only visited in one season and 55
sites were visited in two seasons. The objective was to capture the majority of the species and
their relative abundances present at a given site. At each site, major habitats were selected
over a 10 to 30 m long stretch of the waterbody and were sampled with the same sampling
effort. The sampling effort was thus standardised for each site.
At shallow sites, vegetation habitats were sampled by sweeping a pond-net (200 mm * 300
mm, mesh size 0.5 mm) several times through each vegetation type over a length of 0.5-1 m.
EVK1-CT1999-00026 PAEQANN
Manual for users
83
Bottom habitats were sampled by vigorously pushing the pond net through the upper few
centimetres of each bottom type over a length of 0.5 to 1m. The habitat samples were then
combined for the site to give one sample with a standard area of 1.5 m2 1.2m 2 of vegetation
& 0.3 m2 of bottom). At sites lacking vegetation, the standard sampling was confined to the
bottom habitats. At deeper sites, five samples were taken with an Ekman-Birge sampler from
the bottom habitats. These five grabs were equivalent to one 0.5 m pond net bottom sample.
Vegetation habitats were sampled with a pond net as described above. Again the total
sampling area was standardised as 1.5 m2. Macroinvertebrate samples were taken to the
laboratory, sorted by eye, counted and identified to species level.
The sampling dates were spread over the four seasons as well as over several years (1981 up
to and including 1985). Season was taken into account by defining sampling periods as
nominal “environmental” variables within the analysis.
A data sheet was used to note a number of abiotic and some biotic variables in the field. Some
were measured directly (width, depth, surface area, temperature, transparency, percentage of
vegetation cover, percentage of sampled habitat), others (such as regulation, substratum, bank
shape) were classified. Field instruments were used to measure oxygen, electrical conductivity,
stream velocity and pH. Surface water samples were taken to determine chemical variables.
Other variables, like land-use, bottom composition, and distance from source, were gathered
from additional sources (data from water boards, maps). In total, 70 abiotic variables were
measured at each site.
7.2.4.1 Ordination
7.2.4.2 Prediction
7.2.5. Taxa/Species
cf. Annexe 1.
7.3. Diatoms
7.3.1. Environmental variables
DESCRIPTION
COMMENT
UNITS
Catchment area up to the site
km2
Distance to source
km
Stream order from the source
including all tributaries (quantitative value)
to the site
Slope
covering a length of 1 km
m km-1
Width
m
Altitude
m
Latitude (Greenwich)
°
Longitude (Greenwich)
°
Geology
EVK1-CT1999-00026 PAEQANN
Limestone, Sandstone, Granitic, Volcanic,
Quaternary sediments, Other, Mixed or Mudstone
(schist & shale)
Manual for users
84
Catchment area of the whole
river
km2
Discharge
low, mid or high
Hydropower installation
within 10 km upstream
yes or no
Water level
lowest water levels, mid levels or flood levels
River morphology
natural, partly channelized or totally channelized
m3 s-1
Reduction of flow within 10
yes or no
km upstream
Shading
closed, mid or opened
Current velocity
or 1 (low) < 0.2 m s-1; 3 (high) > 0.5 m s-1
m s-1
Alkalinity (HCO3- + CO32-)
meq l-1
Biological Oxygen Demand
5d
mg l-1
Ca2+
mg l-1
Cl-
mg l-1
Electric conductivity
at 20°C
µs cm-1
Dissolved oxygen
mg l-1
Dissolved organic carbon
mg l-1
NH4+
mg l-1 N
NO2-
mg l-1 N
NO3-
mg l-1 N
pH
PO43-
mg l-1 P
Temperature
°C
Depth
m
Percent of oxygen saturation
%
Alkalinity (HCO3- + CO32-)
ºF
Alkalinity (HCO3- + CO32-)
mg l-1
CaCO3
Biotic Index
The biotic Index was developed by the Univ.
Innsbruck (Pipp) and has 6 classes (1 is the best
and 6 the worsest) Near natural conditions are 1
and 2)
Biological Oxygen Demand
= 54% of BOD5
2d
mg l-1
CO32-
mg l-1
Chemical oxygen demand
mg l-1
Electric conductivity
at 25°C
Dissolved mineral nitrogen
EVK1-CT1999-00026 PAEQANN
µs cm-1
mg l-1 N
Manual for users
85
Hardness
mg l-1
CaCO3
Hardness
°dH
HCO3-
mg l-1
K+
mg l-1
Mg++
mg l-1
Na+
mg l-1
Azote Kejdahl
mg l-1
permanganate oxydability
mg O2 l-1
total phosphorus
mg l-1 P
SiO2
mg l-1
Suspended matter
mg l-1
SO42-
mg l-1
Total Organic Carbon
mg l-1
Turbidity
low, mid or high
Turbidity
FTU
7.3.2. Taxa/Species
cf. Annexe 2.
The bellow figure shows the cluster number of diatom community, using SOM method, as
implemented in the PAEQANN tool
8. Authors
This program have been designed and written from April 2002 to December 2002 at
CESAC by:
EVK1-CT1999-00026 PAEQANN
Manual for users
86
¾
Luc-Patrick Bretin
¾
Sebastien Launay
¾
Luc Bégault
It is avaible under the Gnu Public Licence.
9. References
Adams S.M., Jaworska J.S. & Ham K.D., 1996. Influence of ecological factors on the
relationship between MFO induction and fish growth: bridging the gap using neural
networks. Mar. Environ. Res., 42(1-4): p. 197-201.
Adkison M.D. & Peterman R.M., 1996. Results of Bayesian methods depend on details of
implementation: An example of estimating salmon escapement goals. Fish. Res., 25(2),
155-170.
Albiol J., Campmajo C., Casas C. & Poch M., 1995. Biomass estimation in plant cell cultures: a
neural network approach. Biotechn. Progr., 11, 88–92.
Andrews R., Diederich J. & Tickle A.B., 1995. A survey and critique of techniques for extracting
rules from trained artificial neural net-works. Knowl. Based Syst., 8, 373–389.
Aoki I., Komatsu T. & Hwang, K., 1999. Prediction of response of zooplankton biomass to
climatic and oceanic changes. Ecol. Model., 120 (2-3): 261-270.
Aoki I. & Komatsu T., 1997. Analysis and prediction of the fluctuation of the sardine
abundance using a neural network. Oceanol. Acta, 20(1): 81-88.
Aussem A. & Hill D., 1999. Wedding connectionist and algorithmic modelling towards
forecasting Caulerpa taxifolia development in the north-western Mediterranean sea. Ecol.
Model., 120 (2-3): 225-236.
Balls G.R., Palmer Brown D. & Sanders G.E., 1996. Investigating microclimatic influences on
ozone injury in clover (Trifolium subterraneum) using artificial neural networks. New
Phytol., 132, 271-280.
Baraldi A. & Parmiggiani F., 1997. Fuzzy combination of Kohonen's and ART neural network
models to detect statistical regularities in a random sequence of multi-valued input
patterns. In: Proceedings of ICNN'97, International Conference on Neural Networks,
volume I, pages 281--286. IEEE Service Center, Piscataway, NJ.
Baran P., Lek S., Delacoste M. & Belaud A., 1996. Stochastic models that predict trout
population density or biomass on a mesohabitat scale. Hydrobiologia, 337(1-3): p. 1-9.
Barciela R. M., Garcia E. & Fernandez E., 1999. Modelling primary production in a coastal
embayment affected upwelling using dynamic ecosystem models and artificial neural
networks. Ecol. Model., 120, 199-211.
Boddy L., Morris C. W., Wilkins M. F., Luan Al-Haddad G. A., Tarran R. R., Jonker P. &
Burkill H., 2000. Identification of 72 phytoplankton species by radial basis function
neural network analysis of flow cytometric data. Mar. Ecol. Prog. Ser., 195: 47-59.
Billheimer D., Cardoso T., Freeman E., Guttorp P., Ko H.W. & Silkey M., 1997. Natural
viability of benthic species composition in the Delaware Bay. Environ. Ecol. Stat., V4, 95115.
Breiman L., Friedman J., Olshen R. & Stone C.J. (1984) Classification and regression trees.
Chapman & Hall, New York.
Brey T. & Gerdes D., 1998. High Antarctic macrobenthic community production. J. Exp. Mar.
Biol. Ecol., 231: 191–200.
Brey T., Jarre-Teichmann A. & Borlich O., 1996. Artificial neural network versus multiple linear
regression: predicting P/B ratios from empirical data. Mar. Ecol. Prog. Ser., 140: 251-256.
EVK1-CT1999-00026 PAEQANN
Manual for users
87
Brosse S., Guegan J., Tourenq J. & Lek S., 1999. The use of artificial neural networks to assess
fish abundance and spatial occupancy in the littoral zone of a mesotrophic lake. Ecol.
Model., 120 (2-3): 299-311.
Cai J., Cottis R. A. & Lyon S. B., 1999. Phenomenological modelling of atmospheric corrosion
using an artificial neural network. Corros. Sci., 41, 1001-1030.
Carpenter G. A., Gopal S., Macomber S., Martens S., Woodcock C.E. & Franklin J. 1999. A
neural network method for efficient vegetation mapping. Remote Sens. Environ., 70: 326–
338.
Carpenter S.R., Kitchell J.F., Cottingham K.L., Schindler D.E., Christensen D.L., Post D.M. &
Voichick N., 1996. Chlorophyll variability, nutrient input, and grazing: Evidence from
whole-lake experiments. Ecology, 77(3), 725-735.
Chambers, J. & Hastie, T. (1991) Statistical Models in S, Wadsworth/Brooks Cole, Pacific
Grove.
Chen Y. & Fournier D., 1999. Impacts of atypical data on Bayesian inference and robust
Bayesian approach in fisheries. Can. J. Fish. Aquat. Sci., 59(9), 1525-1533.
Chen Y., Breen P.A. & Andrew N.L., 2000. Impacts of outliers and mis-specification of priors
on Bayesian fisheries-stock assessment. Can. J. Fish. Aquat. Sci., 57(11), 2293-2305.
Cho S.B. 1997. Self-organizing map with dynamical node-splitting: Application to handwritten
digit recognition. Neural Comput., 9, 1345-1355.
Chon T.S., Kwak I.S., Park Y.-S., Kim T.-H. & Kim Y.-S., in press. Patterning and the short-term
prediction of community dynamics of benthic macroinvertebrates in urbanized streams
with implementation of biological and environmental factors in a recurrent artificial
neural network. Ecol. Model.
Chon T.S., Park Y.S., Moon K.H. & Cha E.Y., 1996. Patternizing communities by using an
artificial neural network. Ecol. Model., 90: 69-78.
Chow M. Y. & Yee S. O., 1991. Robustness test of an incipient fault detector Artificial Neural
Network. IEEE, 73-78.
Civco D.L. 1993. Artificial neural networks for land cover classification and mapping.
International Journal of Geographic Information Systems, 7(2):173-186.
Colasanti R.L., 1991. Discussions of the possible use of neural network algorithms.Ecol. Model.
Binary 3, 13–15.
Cottingham K.L. & Schindler D.E., 2000. Effects of grazer community structure on
phytoplancton response to nutrient pulses. Ecology, 81(1), 183-200.
Cow-Rogers S., 1997. Inseason forecasting of skeena River sockeye run size using Bayesian
probability theory. Canadian Report of Fisheries and Aquatic Sciences, 0(2412), 1-43.
Dimopoulos I., Chronopoulos J., Chronopoulou Sereli A. & Lek S., 1999. Neural network
models to study relationships between lead concentration in grasses and permanent
urban descriptors in Athens city (Greece). Ecol. Model.,120 (2-3): 157-165.
Dimopoulos Y., Bourret P. & Lek S., 1995. Use of some sensitivity criteria for choosing
networks with good generalization ability. Neural Process. Lett., 2: 1-4.
Dobson A.J., 1983. An introduction to statistical modelling. Chapman & Hall.
Efron B., 1983. “Estimating the error rate of a prediction rule: improvement on crossvalidation.” Journal American Statistical Association 78: 316-330.
Efron B. and R. J. Tibshirani (1995). “Cross-validation and the bootstrap: estimating the error
rate of the prediction rule.” Rep. Tech. Univ. Toronto.
El-Keib A. A. & Ma X., 1995. Application of Artificial Neural Networks in Voltage Stability
Assessment. IEEE Transactions on Power Systems, 10: 1890-1896.
Engelbrecht A. P., Cloete I. & Zurada J. M., 1995. Determining the significance of input
parameters using sensitivity analysis, From natural to artificial neural computation.
Springer-Verlag, Malaga-Torremolinos, Spain.
EVK1-CT1999-00026 PAEQANN
Manual for users
88
Firth D., 1991. Generalized linear models. In Hinkley et al. Statistical Theory and Modelling.
London, Chapman & Hall.
Flood I. & Kartam N., 1994. Neural networks in civil engineering. I: Principles and
understanding. J. Comput. Civil Eng., 8 (2): 131–148.
Foody G. M., 1999. Applications of the self-organizing feature map neural-network in
community data-analysis. Ecol. Model., 120 (2-3): 97-107.
French M. & Recknagel F., 1994. Modeling algal blooms in freshwaters using artificial neural
networks, in Computer Techniques in Environmental Studies. Vol. 2. Environmental
Systems, P.e. Zannetti, Editor. Computational Mechanics Publications: Southampton,
Boston. p. 87-94.
Friedman J. H. 1997. “On bias, variance, o/1-loss and the curse-of-dimensionality.” Data
Mining and knowledge discovery 1: 55-77.
Garson G.D., 1991. Interpreting neural network connection weights. Artificial Intelligence Expert,
6, 47-51.
Geladi P. & Kowalski B.R., 1986. Partial Least Squares Regression: A tutorial, Anal. Chim. Acta
185: 1-17
Gelman, A., Carlin, J.B., Stern, H.S. and Rubin, D.B.(1995). Bayesian Data Analysis. London:
Chapman and Hall.
Giraudel J.L., Aurelle D. & Lek S., 2000. Application of the self-organizing mapping and fuzzy
clustering microsatellite data: how to detect genetic structure in brown trout (Salmo
trutta) populations. In Lek, S. & Gueguan, J.F., editors, {Artificial Neuronal Networks:
application to ecology and evolution}, Environmental Science, pages 187-201, SpringerVerlag.
Giske J., Huse G. & Fiksen O., 1998. Modelling spatial dynamics of fish. Rev. Fish. Biol. Fish. 8,
57–91.
Goh A. T. C., 1995. Back-propagation neural networks for modeling complex systems. Art.
Intell. Eng., 9: 143-151.
Gross L., Thiria S. & Frouin R., 1999. Applying artificial neural network methodology to ocean
color remote sensing. Ecol. Model.,120 (2-3): 237-246.
Grossberg, S. 1982. Studies of Mind and Brain: Neural Principals of Learning, Perception,
Development, Cognition, and Motor Control. Reidel Press, Boston.Guegan J.F., Lek S.
& Oberdorff T., 1998. Energy availability and habitat heterogeneity predict global
riverine fish diversity. Nature, 391: 382–384.
Guegan, J.-F., S. Lek, and T. Oberdorff. 1998. Energy availability and habitat heterogeneity
predict global riverine fish diversity. Nature 391:382-384.
Gunn S., 1998. Support vector machines for classification and regression. ISIS Technical
Report, Image-Speech & Intelligent Systems Group, University of Southampton.
Guo Z. & Uhrig R.E., 1992. Using modular neural networks to monitor assident conditions in
nuclear. SPIE, Applications of Artificial Neural Networks, III: 1709, 505-516.
Hammitt A.M. & Bartlett E.B., 1995. Determining Functional Relationships from trained
Neural Networks. Math. Comput. Model., 22: 83-103.
Hammond T., 1997. A Bayesian interpretation of target strength data from the Grand Banks.
Can. J. Fish. Aquat. Sci., 54(10), 2323-2333.
Hagan, M.T., H.B. Demuth, & M.Beale, 1996. Neural Network Design. PWS Publishing
Company, Boston.
Hastie, T. & Tibshirani, R., 1990. Generalized Additive Models, Chapman and Hall.
Haykin, S., 1999. Neural Networks: A Comprehensive Foundation (2nd edition). Prentice Hall,
New Jersey
Hecht-Nielsen, R. 1987. Counter propagation networks. Proc. of the Int. Conf. On Neural
Networks, II,19-32, IEEE Press, New York, June 1987.
EVK1-CT1999-00026 PAEQANN
Manual for users
89
Heiland I., 1990. Partial Least Squares Regression and Statistical Models, Scandinavian Journal
of Statistics, 17: 97-114
Helu S.L., Sampson D.B. & Yin Y., 2000. Application of statistical model selection criterion to
the Stock Synthesis assessment program. Can. J. Fish. Aquat. Sci., 57(9): 1784-1793.
Hilbert D.W. & Ostendorf B., in press. The utility of artificial neural networks for modelling
the distribution of vegetation in past, present and future climates. Ecol. Model.
Hilborn R. & Liermann M., 1998. Standing on the shoulders of giants: Learning from
experience in fisheries. Rev. Fish boil. Fisher., 8(3), 273-283.
Hoijtink, H. & Molenaar I.W., 1997. A multidimensional item response model: Constrained
latent class analysis using the Gibbs sampler and posterior predictive checks.
Psychometrika, 62: 171-190.
Hoijtink, H., 1998. Constrained latent class analysis using the Gibbs sampler and posterior
predictive p-values: Applications to educational testing. Statistica Sinica, 8: 691-712.
Hoijtink, H., 2001. Confirmatory latent class analysis: Model selection using Bayes factors and
(pseudo) likelihood ratio statistics. Multivariate Behavioral Research.
Hornik K & Stinchcombe M, White H, 1989 Multilayer feedforward networks are universal
approximators. Neural Networks 2: 359-366
Höskuldsson A., 1988. PLS Regression Methods. J. Chemometrics, 2: 211-228
Hosmer, D.W. and Lemeshow, S. 1989. Applied logistic regression. Wiley. New York.
Hsu C. T., Tzeng Y. M., Chen C. S. & Cho M. Y., 1995. Distribution feeder loss analysis by
using an artificial neural network. Electr. Pow. Syst. Res. 34, 85-90.
Jeong K.-S., Joo G.-J., Kim H.-W., Ha K. & Recknagel F., in press. Predicting dynamics of
phytoplankton in the flow regulated Nakdong River (South Korea) by means of
recurrent artificil neural networks. Ecol. Model.
Jon T., Cass A. & Richards L.J., 2000. A Bayesian decision analysis to set escapement goals for
Fraser River sockeye salmon (Oncorhynchus nerka). Can. J. Fish. Aquat. Sci., 57(5), 962979.
Jørgensen S.E. & Padisak J., 1996. Does the intermediate disturbance hypothesis comply with
thermodynamics ? Hydrobiologia 323, 9-21.
Jørgensen S.E., 1997. Integration of Ecosystem Theories: A Pattern. Kluwer Academic Publ.
Dordrecht, 388 pp.
Jørgensen S.E. & de Bernardi D., 1998. The Use of structural dynamic models to explain
successes and failures of biomanipulation. Hydrobiologia 359, 1-12 .
Jørgensen S.E. & Fath B., 2001. Growth and Development. Submitted to American Naturalist.
Kass, R.E., and Raftery, A.E. (1995). Bayes factors. Journal of the American Statistical Association,
90, 773-795.
Kastens T.L. & Featherstone A.M., 1996. Feedforward back-propagation neural networks in
prediction of farmer risk preference. Am. J. Agri. Econ. 78, 400–415.
Keiner L.E. & Yan X.-H., 1998. A neural network model for estimating sea surface chlorophyll
and sediments from Thematic Mapper imagery. Remote Sens. Environ., 66: 153–165.
Kim S.H., Yoon C. & Kim B.J., 2000. Structural Monitoring System Based on Sensitivity
Analysis and a Neural Network. Computer-aided Civil and Infrastructure Engineering, 155:
309-318.
Kimes D.S., Holben B.N., Nickeson J.E., & McKee W.A., 1996. Extracting forest age in a Pacific
Northwest forest from Thematic Mapped and topographic data. Remote Sens. Environ.,
56: 133-140.
Kinas P.G., 1996. Bayesian fishery stock assessment and decision making using adaptive
importance sampling. Can. J. Fish. Aquat. Sci., 53(2): 414-423.
Kohavi, R. 1995. A study of cross-validation and bootstrap for estimation and model selection,
Morgan Kaufmann Publishers.
EVK1-CT1999-00026 PAEQANN
Manual for users
90
Kohavi, R., and D. H. Wolpert. 1996. Bias plus variance decomposition for zero-one loss
functions., Saitta L .Machine Learning. Proceedings of the Thirteenth Ibnternational
Conference edition. Morgan Kaufmann Publishers INC., Bari.
Kohonen, T., 1982. Self-organized formation of topologically correct feature maps. Biol.
Cybern., 43: 59-69.
Kohonen T., 1982. Analysis of a simple self-organizing process. Biol. Cyb., 44 (2): 135-140.
Kohonen, T., 1984. Self-Organizing Maps, volume 30 of Springer Series in Information Science,
Springer, Berlin.
Kohonen T., 1995. Self-Organizing Maps, volume 30 of Springer Series in Information
Sciences. Springer, Berlin, Heidelberg. (Second Extended Edition 1997).
Kohonen T., 2000. An unsupervised learning method that produces organized representation
from real information. In Malmgren, H., Boga, M.,& Niklasson, L., editors, Artificial
Neural Networks in Medicine and Biology, Proceedings of the ANNIMAB-1
Conference, Göteborg, Sweden, 13--16 May 2000, pages 45-53.
Komatsu T., Aoki I., Mitani I. & Ishii T., 1994, Prediction of the catch of Japanese sardine
larvae in Sagami Bay using a neural network. Fish. Sci., 60: 385-391.
Kosko B., 1992. Neural Networks and Fuzzy Systems: A Dynamical Systems Approach to Machine
Intelligence. Prentice-Hall, Englewood Cliffs.
Kuikka S., Hilden M., Gislason H., Hansson S., Sparholt H. & Varis O., 1999. Modeling
environmentally driven uncertainties in Baltic cod (Gadus morhua) management by
Bayesian influence diagrams. Can. J. Fish. Aquat. Sci., 59(4): 629-641.
Lae R., Lek S. & Moreau J., 1999. Predicting fish yield of African lakes using neural networks.
Ecol. Model.120, 325-335.
Lamon E.C. & Clyde M.A., 2000: Accounting for model uncertainty in prediction of
chlorophyll a in Lake Okeechobee. Journal of Agricultural Biological and Environmental
Statistics, 5(3): 297-322.
Lee D., Shepard B., Sanborn B. & Ulmer L., 1996. Assessing extinction risks for westslope
cutthroat trout in the upper Missouri River Basin using BayVAM model. Bulletin of the
Ecological Society of America, 77: 258.
Lee D.A. & Rieman B.E., 1997. Population viability assessment of salmonids by using
probabilistic networks. North American Journal of Fisheries Management, 17(4): 1144-1157.
Lek S. & Guegan J-F [eds.], 2000. Artificial Neuronal Networks: Application to Ecology and
Evolution, Springer-Verlag, Berlin/Heidelberg.
Lek S., Belaud A., Dimopoulos I., Lauga J. & Moreau J., 1995. Improved estimation, using
neural networks, of the food consumption of fish populations. Mar. Freshwater Res., 46:
p. 1229-1236.
Lek S., Belaud A., Baran P., Dimopoulos I. & Delacoste M., 1996a. Role of some environmental
variables in trout abundance models using neural networks. Aquat. Living Resour., 9: 2329.
Lek S., Delacoste M., Baran P., Dimopoulos I., Lauga J. & Aulagnier S., 1996b. Application of
neural networks to modelling nonlinear relationships in ecology. Ecol. Model., 90: 39-52.
Lek S., Dimopoulos I. & Fabre A., 1996c. Predicting phosphorus concentration and phosphorus
load from watershed characteristics using backpropagation neural networks. Acta
Oecologica, 17(1): 43-53.
Liermann M. & Hilborn R., 1997. Depensation in fish stocks: A hierarchic Bayesian metaanalysis. Can. J. Fish. Aquat. Sci., 54(9): 1976-1984.
Lin, C.-T. and C.S.G. Lee. 1996. Neural Fuzzy Systems. Prentice Hall PTR, Upper Saddle River,
New Jersey, 797 pp.
Liong S. Y., Lim W. H., Kojiri T. & Hori T., 2000a. Advance flood forecasting for flood
stricken Bangladesh with a fuzzy reasoning method. Hydrolog. Process., 14: 431-448.
EVK1-CT1999-00026 PAEQANN
Manual for users
91
Liong S. Y., Lim W. H. & Paudyal G. N., 2000b. River Stage Forecasting in Bangladesh: Neural
Network Approach. J. Comput. Civil Eng., 14: 1-8.
Maier H. R. & Dandy G. C., 1998. The effect of internal parameters and geometry on the
performance of back-propagation neural networks: an empirical study. Environ. Modell.
Softw., 13: 193-209.
Maier H. R. & Dandy G. C., 2000. Neural networks for the prediction and forecasting of water
resources variables: A review of modelling issues and applications. Environ. Modell.
Softw., 15: 101-124.
Mann S.& Benwell G.L., 1996. The integration of ecological, neural and spatial modelling for
monitoring and prediction for semi-arid landscapes. Comput. Geosci., 22(9): 1003-1012.
Mastrorillo, S., F. Dauba, et al. (1998). “Predicting local fish species richness in the Garonne
River basin.” Comptes Rendus de l'Académie des Sciences Série III Sciences de la Vie
321(5): 423-428.
McCullagh P. & Nelder J.A. (1989) Generalized linear models. Second edition. London,
Chapman & Hall.
Nelder J.A. & Wedderburn R.W.M. (1972) Generalized additive models. Journal of Royal
Statistical Society, A, 135: 370-384.
Mastrorillo S., Dauba F., Oberdorff T., Guegan J.F. & Lek S., 1998. Predicting local fish species
richness in the Garonne River basin. Comptes Rendus de l'Academie des Sciences, Serie III
Sciences de la Vie, 321: 423-428.
Mastrorillo S., Lek S. & Dauba F., 1997. Predicting the abundance of minnow Phoxinus
phoxinus (Cyprinidae) in the river Ariege (France) using artificial neural networks.
Aquat. Living Resour., 10(3): 169-176.
Mau B., Newton M.A., Larget B., 1999. Bayesian phylogenetic inference via Markov chain
Monte Carlo methods. Biometrics, 55(1), 1-12.
McAllister M.K. & Ianelli J.N., 1997. Bayesian stock assessment using catch-age data and the
sampling importance resample algorithm. Can. J. Fish. Aquat. Sci., 54(2): 284-300.
McAllister M.K. & Kirkwood G.P., 1998. Bayesian stock assessment: A review and example
application using the logistic model. ICES J. Mar. Sci., 55(6): 1031-1060.
McAllister M.K. & Kirkwoood G.P., 1998. Using Bayesian decision analysis to help achieve a
precautionary approach for developing fisheries. Can. J. Fish. Aquat. Sci., 55(12): 26422661.
McAllister M.K. & Pikitch E.K., 1997. A Bayesian approach to choosing a design for surveying
fishery resources: Application to the eastern Bering Sea trawl survey. Can. J. Fish. Aquat.
Sci., 54(2): 301-311.
Meyer R. & Millar R., 1999. Bayesian stock assessment using a state-space implementation of
the delay difference model. Can. J. Fish. Aquat. Sci., 59(1): 37-52.
Myers R.A., Bowen K.G. & Barrowman N.J., 1999. Maximum reproductive rate of fish at low
population sizes. Can. J. Fish. Aquat. Sci., 56(12): 2404-2419.
Nakano, H., et al., 1991. Identification of plankton using a neural network with a function of
unknown species detection. Rep. Measurement Res., Group, IM-91-30: 47-56.
Newman K., 1997. Bayesian averaging of generalized linear models for passive integrated
transponder tag recoveries from salmonids in the Snake River. North American Journal of
Fisheries Management, 17(2): 362-377.
Ogle D.H., Pruitt R.C., Spangler G.R. & Cyterski M.J., 1996. A Bayesian approach to assigning
probabilities to fish ages determined from temporal signatures in growth increments.
Can. J. Fish. Aquat. Sci., 53(8): 1788-1794.
Ozesmi S. L. & Ozesmi U., 1999. An artificial neural network approach to spatial habitat
modelling with interspecific interaction. Ecol. Model., 116: 15-31.
EVK1-CT1999-00026 PAEQANN
Manual for users
92
Parker D.B., 1982. Learning logic. Invention report S81-64, File 1, Office of Technology
Licensing, Stanford University.
Patterson K.R., 1999. Evaluating uncertainty in harvest control law catches using Bayesian
Markov chain Monte Carlo virtual population analysis with adaptive rejection sampling
and including structural uncertainty. Can. J. Fish. Aquat. Sci., 56(2): 208-221.
Pella J.J., Masusa M.M. & Chen D.G., 1998. Forecast methods for inseason management of the
southeast Alaska chinook salmon troll fishery. In, Funk f., Quinn T.J., Heifetz J., Ianellí
J.N., Powers J., Schweigert J.F., Sullivan P., Zhang C.I., editors. Lowell Wakefield
Fisheries Symposium Series; Fishery stock assessment models. Alaska Sea Grant College
Program, Fairbanks.
Peterman R.M., Peters C.N., Robb C.A. & Frederick S.W., 1999. Bayesian decision analysis and
uncertainty in fisheries management. In Pitcher T.J., Hart P.J., Pauly D., editors.
Reinverting fisheries management. Kluwer Academic Publishers, Dordrecht .
Pineda F., 1987. Generalization of backpropagation to recurrent neural networks. Phys. Rev. Lett., 19 (59):
2229-2232.
Pinelalloul B., Niyonsenga T. & Legendre P., 1995. Spatial and environmental components of
fresh-water zooplankton structure. Ecoscience, V2: 1-19.
Punt A.E. & Hilborn R., 1997. Fisheries stock assessment and decision analysis: the Bayesian
approach. Rev. Fish Biol. Fisher., 7(1): 35-63.
Punt A.E. & Walker T.I., 1998. Stock assessment and risk analysis for the school shark
(Galeorhinus galeus) of southern Australia. Mar. Freshwater Res., 49(7): 719-731.
Qian S.S., Lavine M. & Stow C.A., 2000. Univariate Bayesian nonparametric binary regression
with application in environmental management. Environ. Ecol. Stat., 7(1): 77-91.
Recknagel F. & Lek S., in press. State of the art of ecological modelling by artificial neural
networks. Ecol. Model.
Recknagel F., 1997. ANNA - Artificial Neural Network model predicting blooms and
succession of blue-green Algae. Hydrobiologia, 349: 47-57.
Recknagel F., French M., Harkonen P. & Yabunaka K., 1997. Artificial neural network
approach for modelling and prediction of algal blooms. Ecol. Model., 96(1-3): 11-28.
Robb C.A. & Peterman R., 1998. Application of Bayesian decision analysis to management of a
sockeye salmon (Oncorhynchus nerca) fishery. Can. J. Fish. Aquat. Sci., 55(1): 86-98.
Rumelhart D.E., Hinton G.E. & Williams G.E., 1986. Learning representations by backpropagating errors. Nature, 323: 533-536
Scardi M., in press. Advences in neural network modeling of phytoplankton primary
production. Ecol. Model.
Scardi M., Lek S., Lim P., Di Dato P. & Oberdorff T., in press. Artificial neural networks as a
tool for predicting fish community composition in rivers. Proceedings of the Third
World Congress of Nonlinear Analysts (WCNA-2000), Catania (Italy), July 19-26, 2000.
Journal of Nonlinear Analysis.
Scardi M. & Harding L. W., 1999. Developing an empirical model of phytoplankton primary
production: a neural network case study. Ecol. Model., 120 (2-3): 213-223.
Scardi M., 1996. Artificial neural networks as empirical models of phytoplankton production.
Mar. Ecol. Prog. Ser., 139: 289-299.
Scardi M., 2000. Neuronal network models of phytoplankton primary production. p. 115-129.
In: Lek S, Guegan J-F [eds.], Artificial Neuronal Networks: Application to Ecology and
Evolution, Springer-Verlag, Berlin/Heidelberg.
Schleiter I.M., Borchardt D., Wagner R., Dapper T., Schmidt K., Schmidt H., & Werner H.,
1999. Modelling water quality, bioindication and population dynamics in lotic
ecosystems using neural networks. Ecol. Model.,120 (2-3): 271-286.
EVK1-CT1999-00026 PAEQANN
Manual for users
93
Schwenker F., Kestler Hans~A., & Palm G., 2000. Combination of supervised and unsupervised
learning for radial-basis-function networks. In Bothe, H. and Rojas, R., editors,
{Proceeding of the ICSC Symposia on Neural Computation (NC'2000) May 23-26,
2000 in Berlin, Germany}. University of Ulm, D-89069 Ulm, Germany, ICSC Academic
Press.
See L. & Openshaw S., 2000. A hybrid multi-model approach to river level forecasting. Hydrol.
Sci. J., 45 (4): 523-536
Seginer I., 1997. Some artificial neural network applications to greenhouse environmental
control. Comput. Electron. Agr., 18: 167-186.
Silveira V.F., Khator S.K. & Barcia R.M., 1996. An information management system for
forecasting environmental change. Computers ind. Engng, 31 (1-2): 289-292.
Simpson R., Williams R., Ellis R. & Culverhouse, P.F., 1992. Biological pattern recognition by
neural networks. Mar. Ecol. Prog. ser., 79 (3): 303-308.
Simpson R., Culverhouse P.F., Williams R. & Ellis R., 1993. Classification of Dinophyceae by
artificial neural networks. Dev. Mar. Biol., 3: 183-190.
Smith A.D.M. & Punt A.E., 1998. Stock assessment of gemfish (Rexea solandri) in eastern
Australia using maximum likelihood and Bayesian methods. In, Funk f., Quinn T.J.,
Heifetz J., Ianellí J.N., Powers J., Schweigert J.F., Sullivan P., Zhang C.I., editors. Lowell
Wakefield Fisheries Symposium Series; Fishery stock assessment models. Alaska Sea
Grant College Program, Fairbanks.
Spitz F., Lek S. & Dimopoulos I., 1996. Neural network models to predict penetration of wild
boar into cultivated fields. Journal of Biological Systems, 4: 433-444.
Steinberg L.J., Reckhow K.H. & Wolpert R.L., 1996. Bayesian model for fate and transport of
polychlorinated biphenyl in upper Hudson River. Journal of Environmental Engineering,
122 (5): 341-349.
Sung A. H., 1998. Ranking importance of input parameters of neural networks. Expert Syst.
Appl. 15, 405-411.
Titterington, D.M., Smith, A.F.M. & Makov, U.E., 1985. Statistical Analysis of Finite Mixture
Distributions. New York: Wiley.
Trigg D.J., Walley W.J. & Ormerod S.J., 2000. A prototype Bayesian belief network for the
diagnosis of acidification in Welsh rivers. In Ibarra-Berastegi G., Brebbia C., Zanetti P.,
editors. Proceedings of the ENVIROSOFT 2000. Development and Application of
Computer Techniques to Environmental Studies, Bilbao.
Ultsch A., 1999. Data mining and knowledge discovery with emergent self-organizing feature
maps for multivariate time series. In Oja, E. & Kaski, S., editors, Kohonen Maps, pages
33-45. Elsevier, Amsterdam.
Utans, J. and J. E. Moody (1991). “Selecting neural network architectures via the prediction
risk: application to corporate bond rating prediction.” Proceedings of the first
International Conference on Artificial Intelligence Applications on Wall Street, IEEE
Computer Society Press, Los Alamitos.
van Wijk M. T. & Bouten W., 1999. Water and carbon fluxes above European coniferous forests
modelled with artificial neural networks. Ecol. Model., 120: 181-197.
Vapnik V., 1995. The nature of statistical learning theory. Springer, New York.
Varis O. & Kuikka S., 1997. Joint use of multiple environmental assessment models by a
Bayesian meta-model: The Baltic salmon case. Ecol. Model.,102: 341-351.
Varis O., 1997. Bayesian decision analysis for environmental and resource management.
Environmental Modelling & Software with Environmental data News, 12: 177-185.
Vasko K., Toivonen H.T.T. & Korhola A., 2000. A Bayesian multinomal Gaussian response
model for organism-based environmental reconstruction. J. Paleolimnol., 24(3): 243-250.
EVK1-CT1999-00026 PAEQANN
Manual for users
94
Vignaux M., Vignaux G.A., Lizamore S. & Gresham D., 1998. Fine-scale mapping of fish
distribution from commercial catch and effort data using maximum entropy
tomography. Can. J. Fish. Aquat. Sci., 55(5): 1220-1227.
Viktor H. L., Engelbrecht A. P. & Cloete I., 1995. Reduction of Symbolic Rules from Artificial
Neural Networks using Sensitivity Analysis. IEEE: 1788-1793.
Villmann T.& Bauer H.U., 1998. Applications of the growing self-organizing map.
Neurocomputing, 21 (1): 91-100.
Wagner R., Dapper T. & Schmidt H.H., 2000. The influence of environmental variables on the
abundance of aquatic insects: a comparison of ordination and artificial neural networks.
Hydrobiologia, 422: 143-152.
Walley W.J. & Dzeroski S., 1996. Biological Monitoring: A Comparison Between Bayesian,
Neural and Machine Learning Methods of Water Quality Classification. In
Environmental Software Systems, London, Chapman and Hall, pp. 229-240.
Walley W.J. & Fontama V.N., 1997. Bio-Monitoring of Rivers: An AI Approach to Data
Interpretation. Toxicology & Ecotoxicology News, 4(6): 182-184.
Walley W.J. & Fontama V.N., 2000. New Approaches to River Quality Classification Based
Upon Artificial Intelligence. In Assessing the Biological Quality of Fresh Waters
Ambleside/GB. Freshwater Biological Association, pp. 263-280.
Webos P., 1974. Beyond regression: new tools for prediction and analysis in the behavioral
sciences. Thesis, Harvard University.
Wolfram S., 1984. Cellular Automata. PhysicaD, 10, 1-35.
Yao J., Teng N., Ph H.L. & Tan C. L., 1998. Forecasting and analysis of marketing data using
neural networks. Journal of Information Science and Engineering, 14: 843-862.
Yao X. & Liu Y., 1997. A new evolutionary system for evolving artificial neural networks. IEEE T. Neural
Networ., 8 (3): 694-713.
Zegers, S.L. and Karim, M.R. (1991). Generalized linear models with random effects: a Gibbs
sampling approach. Journal of the American Statistical Association, 86, 79-86.
Zhu M.L., Fujita M. & Hashimoto N. 1994. Application of neural networks to runoff
prediction. In: Stochastic and statistical methods in hydrology & environmental
engineering, ed. Hipel, K.W. et al. Kluwer Academic Publishers, Norwell, Mass.
Zurada J.M., 1992. Introduction to Artificial Neural Systems. West Publishing Co., St. Paul,
MN, USA. 785 pp.
Zurada J. M., Malinowski A. & Cloete I., 1994. Sensitivity Analysis for Minimization of Input
Data Dimension for Feedforward Neural Network. In "ISCAS'94. IEEE International
Symposium on circuits and Systems", Vol. 6, pp. 447-450, London.
EVK1-CT1999-00026 PAEQANN
Manual for users
95