The STREAMES Project: Linking Heuristic And Empirical Knowledge Into An Expert System To Assess Stream Managers

Abstract

The increase in stream nutrient loads from anthropogenic sources has become a serious problem, especially in developed regions. Humans affect streams by modifying the landscape in ways that increase the transport of nutrients to surface waters, by directly dumping urban or industrial sewage into the stream, or by modifying streams in ways that reduce their ability to respond to increased nutrient loads. In Mediterranean regions these problems are compounded by the scarcity of water. The decision-making processes involved in water quality management require extensive human expertise or extensive computation with large data sets. In this sense, the STREAMES project aims to develop a knowledge-based environmental decision support system (EDSS) to support and advice water managers in the management of human-altered streams. This EDSS will integrate an Expert System (ES), concretely a rule-based reasoning system (RBS), with a Geographical Information System to address spatial information for the appropriate stream management actions, and a numerical model to estimate point and non-point nutrient sources from middle size catchments. The RBS will be developed by integrating heuristic knowledge from experts in surface water management, as well as empirical knowledge from stream scientists, based both on previous studies and on data directly acquired from experimental sampling. This paper will present the objectives of the STREAMES project with emphasis in the knowledge acquisition and development of the RBS.

Full text

The STREAMES Project:
Linking Heuristic And Empirical Knowledge Into An
Expert System To Assess Stream Managers
J. Comas
1
, E. Llorens
1
, M. Poch
1
, G. Markakis
2
, T. Battin
3
, S. Gafny
4
, E.Maneux
10
, E. Martí
5
, M. Morais
6
,
M.A. Puig
5
, M. Pusch
7
, J.L. Riera
8
, F. Sabater
8
, A.G. Solimini
9
and P. Vervier
10
1
Chemical and Environmental Engineering Laboratory (LEQUIA), University of Girona, Campus Montilivi s/n,
E-17071 Girona, Catalonia, EU, e-mail: {quim/esther}@lequia.udg.es
2
Natural History Museum of Crete, University of Crete, Knossou Avenue, 714 09 Heraklion, Greece, EU
3
Department of Limnology, Institute of Ecology and Conservation Biology, Althanstrasse 14, 1090 Viena,
Austria, EU
4
Institute for Nature Conservation Research, Ramat Aviv, 699778 Tel Aviv, Israel
5
Centre d’Estudis Avançats de Blanes (CSIC), Camí de Sta. Bàrbara s/n, 17300 Blanes, Catalonia, EU
6
Centro de Ecologia Aplicada, Universidad de Evora, Largo dos Colegiais 2, 7002-554 Evora, Portugal, EU
7
Institut fuer Gewaesseroekologie und Binnenfisherei, Mueggelseedamm, 310, 12587 Berlin, Germany, EU
8
Departament d’Ecologia, Universitat de Barcelona, Diagonal 645, Barcelona 08028, Catalonia, EU
9
Dipartamento di Biologia, Università di Roma "Tor Vergata", Via della Ricerca Scientifica, 00133 Roma, Italy,
EU
10
Centre d'Ecologie des Systemes Aquatiques Continentaux – UMR C5576 – Centre National de la Recherche
Scientifique - Universite Paul Sabatier, Rue Jeanne Marvig 29, 31055 Toulouse, France, EU
Abstract: The increase in stream nutrient loads from anthropogenic sources has become a serious problem,
especially in developed regions. Humans affect streams by modifying the landscape in ways that increase the
transport of nutrients to surface waters, by directly dumping urban or industrial sewage into the stream, or by
modifying streams in ways that reduce their ability to respond to increased nutrient loads. In Mediterranean
regions these problems are compounded by the scarcity of water. The decision-making processes involved in
water quality management require extensive human expertise or extensive computation with large data sets. In
this sense, the STREAMES project aims to develop a knowledge-based environmental decision support system
(EDSS) to support and advice water managers in the management of human-altered streams. This EDSS will
integrate an Expert System (ES), concretely a rule-based reasoning system (RBS), with a Geographical
Information System to address spatial information for the appropriate stream management actions, and a
numerical model to estimate point and non-point nutrient sources from middle size catchments. The RBS will be
developed by integrating heuristic knowledge from experts in surface water management, as well as empirical
knowledge from stream scientists, based both on previous studies and on data directly acquired from
experimental sampling. This paper will present the objectives of the STREAMES project with emphasis in the
knowledge acquisition and development of the RBS.
Keywords: expert system; rule-based system, environmental decision support system; stream management; river
water quality
444

1. INTRODUCTION
Nowadays, poor river water quality has become a
serious problem, especially in developed regions, due
to the high nutrient (nitrogen and phosphorus) loads
from anthropogenic sources dumped into the rivers.
Humans affect streams:
• By modifying the land uses in ways that increase
the transport of nutrients to surface waters (large
inputs of nutrients from non-point sources, i.e.
agricultural activity),
• by directly dumping inputs from point sources
(urban or industrial sewage),
• by modifying the streams in ways that reduce
their ability to respond to increased nutrient loads
(e.g. elimination of meanders, destruction of the
riparian vegetation, etc…).
Pristine stream ecosystems can cope with a certain
degree of pollution, whereas heavily polluted and
modified streams cannot retain and transform
excessive nutrient loads. Nevertheless, the response of
streams to anthropogenic impacts is not invariable but
different according to the type of river and water uses.
In Mediterranean regions these problems are
compounded by the scarcity of water.
Therefore, the decision-making processes involved in
stream reach management require extensive human
expertise from people involved directly with day-to-
day stream problems (water managers), empirical
knowledge from scientific research and elaborated
calculation over large sets of numerical and symbolic
data.
Anthropogenic
impacts
Non-point
sources
Point
sources
Responses of streams
Several
water uses
“integrated” management
requires:
ü Experts from diverse fields
ü Empirical knowledge
ü Heuristic
River catchment
Anthropogenic
impacts
Non-point
sources
Point
sources
Responses of streams
Several
water uses
“integrated” management
requires:
ü Experts from diverse fields
ü Empirical knowledge
ü Heuristic
River catchment
Figure 1. The management problem
Thus, the stream water quality management becomes
a typical complex and ill-defined problem whose
optimal management requires an integrated and
multidisciplinary approach (figure 1). This integrated
management can be reached with a tool built upon the
concepts and methods of human reasoning, an
intelligent tool. In this sense, the STREAMES
(STream REach Management, an Expert System
1
)
project appears as an attempt to develop and implement
a knowledge-based environmental decision support
system (KB-EDSS or simply, EDSS) to help Water
Managers (WM) in taking decisions.
This paper describes the objectives of STREAMES with
major emphasis in the bottleneck of the RBS
development: the knowledge acquisition step.
The organisation of this paper follows the next sections:
Section 1 introduces the problem of nutrient retention.
Section 2 gives a brief description of the Streames
project. The EDSS characteristics are introduced in
Section 3. Section 4 addresses the issue of the
knowledge acquisition as a key step for developing a
user-friendly tool. Section 5 presents the future work
with respect to the EDSS and finally, some conclusions
are given in Section 6.
2. THE STREAMES PROJECT
The application of Artificial Intelligence in
environmental sciences is a relatively new discipline.
Papers began to be published sporadically in the middle
of 80s, but it was not until the 90s when the number of
papers experienced a significant increase [Cortés et al.,
2000]. Some interesting environmental applications
include MEDEX [Hadjimichael et al., 1996], an
intelligent tool to assist Mediterranean weather forecast;
CHARADE [Avesani et al., 1993], a decision-making
system for environmental emergencies; FRAME [Calori
et al., 1994], an ES designed to aid in the selection of air
pollution models; and DAI-DEPUR, a distributed
knowledge-based system to supervise wastewater
treatment plant management ([Sànchez et al., 1996] and
[R.-Roda et al., 2002] for a recent vision of the
application). Within the river water domain,
mathematical models, simulation models and decision
support systems are used for catchment management, to
reduce nutrient loads or to solve eutrophication
problems ([Rekolainen et al., 1999], [Young et al.,
1995] and [Davis et al., 1998]). However, none of these
tools implies cooperation between stream managers and
scientists. Moreover, they only consider descriptive
parameters. A new approach integrating expertise,
existing knowledge and new empirical knowledge about
1
(“Human effects on nutrient cycling in fluvial
ecosystems: The development of an ES to assess
stream water quality management at reach scale”,
EVK1-2000-00081, Vth Framework Programme EC;
www.streames.org)
445

nutrient retention capacity concerning not only
descriptive and structural but also functional
parameters is proposed. In this context, STREAMES
aims at evaluating the effect of substantial nutrient
loads on in-stream retention. Furthermore,
STREAMES will examine the relationships between
nutrient retention and selected physical, chemical and
biological structural or functional parameters that may
constrain (i.e., nutrient sources from the catchment) or
control (i.e., in-stream processes) nutrient retention
capacity in human altered streams. Particular
emphasis will be on Mediterranean streames.
In a practical sense, the final goal of STREAMES
project is to develop and validate an EDSS for stream
managers from either private or public water quality
agencies. So the end product will be the EDSS, a
useful decision support system to give diagnosis about
the stream quality and propose solutions to the
problems. These solutions should allow to attain not
only a water quality improvement but also a good
ecological state for the river, according to the Water
Framework Directive (2000/60/EEC).
This project involves 17 partners (10 from scientific
research centres or universities and 7 from water
agencies) from 8 European countries and the project is
divided into five workpackages (WP). Workpackage 1
analyses, at the catchment scale, the relationship
between land uses and nutrient loads to ecosystems
and the relative importance of point and non-point
nutrient sources. WP2 analyses the effects of high
nutrient loads on stream nutrient transport,
transformation and retention at reach scale. WP3
analyses the role of stream biota on the control of
nutrient retention at the sub-reach scale. Finally, WP4
focuses on the development of the EDSS and WP5
promotes the utility and application of the EDSS to
other end-users.
3. THE ENVIRONMENTAL DECISION
SUPPORT SYSTEM
3.1. Structure of the EDSS
The EDSS to be developed in the STREAMES project
integrates a rule-based reasoning component (the core
of the system) with a Geographic Information System
(GIS) and linked with a numerical model to estimate
point and non-point sources. This model will be the
MONERIS model [Behrendt et al., 2001], modified
for Mediterranean regions.
The EDSS must be fed with all the available
information relevant to the final outcome. This
information, both qualitative and quantitative, is stored
in a database, where it remains easily accessible to
software requirement. Additional information can be
required to the user to reach a deeper conclusion. Then
the user would be responsible for gathering these new
measurements. The amount and quality of this
information will condition the quality of the EDSS’s
outcomes.
The EDSS will provide 3 types of outputs: diagnosis,
actions and prognosis. Figure 2 shows the conceptual
framework of the EDSS with its three-level outcome,
based on the specific requirements of the WM.
Concerning the diagnosis step, the EDSS will be able to
infer the river state related to functionality features (for
example, the self-purification capacity of the stream
reach relative to its potential capacity, the nutrient
uptake length or the recovery time). Once this diagnosis
step is reached, the user can require the EDSS to
propose a list of suitable management strategies to
maximise self-purification at the stream reach.
DATABASE
Quantitative
Qualitative
END-USER
OUTCOME
DIAGNOSIS
% of self-purification capacity
SOLUTIONS
List of management strategies
PROGNOSIS
Simulation of scenarios
RULE-BASED SYSTEM
MODEL
G.I.S.
requirement
data gathering
input
extra information
DATABASE
Quantitative
Qualitative
END-USER
OUTCOME
DIAGNOSIS
% of self-purification capacity
SOLUTIONS
List of management strategies
PROGNOSIS
Simulation of scenarios
RULE-BASED SYSTEM
MODEL
G.I.S.
requirement
data gathering
input
extra information
Figure 2. How the EDSS will support the decision
making tasks on stream management.
Finally, the EDSS will offer a prognosis output to the
end-user, providing several scenarios to simulate the
effect of the different actions proposed as solutions and
giving the percentage of success according to the
adopted solution.
3.2. Structure of the Rule-Based System
The structure of any RBS presents two main
independent modules: the Knowledge Base (KB) and
the Inference Engine (IE). The KB contains the overall
knowledge of the process (in our case, stream nutrient
management) codified by means of heuristic rules (a
rule is a set of conditions and conclusions linked to a
446

given hypothesis). The bottleneck of the KB
development is the knowledge acquisition process. In
the first half of the project, G2 [Gensym, 2000] will
be used as inference engine to build the first prototype
of the RBS. However, the final product must be
codified over a new shell (including the inference
engine) that must also be developed during the course
of the project by WP4.
3.3. Requirements of the EDSS
The EDSS will support WM in two ways: First, it will
help them to evaluate the sources and magnitude of
nutrient (nitrogen and phosphorus) loads affecting the
stream reach of interest. Second, it will help WM to
decide on the best strategy for stream amelioration at
the particular reach, with special emphasis on actions
directed towards increasing nutrient retention and
transformation within the stream.
The EDSS must detect any problems caused for a
wastewater treatment plant (WWTP). It must predict
which site is the best place to build a new WWTP and
which wastewater treatment has to be implemented.
Moreover, the EDSS will evaluate the WWTP impact
on the studied reach and give a set of actions to
subdue the impacts caused by the WWTP or other
inputs. In order to prevent confusions, the EDSS will
only focus on the reach scale. Of course, in some
problems, the EDSS must work on the catchment
scale.
4. KNOWLEDGE ACQUISITION
The knowledge acquisition process is the key process
to build a complete knowledge base, in this case, a
manual of operation on stream management, in the
development of the rule-based system, the main
module of the EDSS. We are interested on acquiring
the whole knowledge necessary to identify the stream
quality problems and to solve them. This knowledge is
not only concerned on how the WM deal with their
rivers but also on how the scientists think that stream
management could be improved. Three types of
knowledge can be distinguished: general knowledge
related to the domain, heuristic knowledge and
empirical knowledge, more specific since contains
distinguishing features related to the specific sites
where the RBS will be applied (Figure 3).
Literature
Scientists and water
managers
Empirical data from
experimental campaigns
data processing
interviews
review
General Knowledge
Empirical Knowledge
(specific)
Knowledge
Base
Heuristic Knowledge
Literature
Scientists and water
managers
Empirical data from
experimental campaigns
data processing
interviews
review
General Knowledge
Empirical Knowledge
(specific)
Knowledge
Base
Knowledge
Base
Heuristic Knowledge
Figure 3. Knowledge acquisition in the RBS
development.
In order to obtain as much as possible of knowledge of
quality about nutrient retention, different sources will be
used:
• literature
• survey among stream experts and face-to-face
interviews
• study of the historical data and sampling sites of
eleven different scenarios
While the general and heuristic knowledge of the
process can be mainly extracted from literature and
human experts, the empirical knowledge can only be
acquired by processing empirical data directly
collected from the experimental campaigns in each
study site (in our case from WP1, WP2 and WP3
tasks) (figure 3).
4.1. Literature review
A bibliographical review about stream reach
management, especially referred to nutrient pollution,
and the state-of-the-art of decision support systems
applied to river or stream management will be done.
4.2. Questionnaire and face-to-face interviews
A questionnaire will be sent to the WM and scientists.
The purpose of this questionnaire is to obtain a general
view of stream management. The objective is to acquire
as much information as possible to understand the
reasoning mechanisms of WM (and also from scientists)
when tackling any problem about stream nutrient
retention. Therefore, information about data availability,
minimal specifications, expectations and minimal
capabilities of the RBS are required. With the results of
this survey, we will prepare the questions for the
personal interviews.
Each questionnaire consists of 4 parts:
447

• Part 1 intends to obtain general information
about river basins of each participant country.
• Part 2 refers to the stream water quality.
Availability and format of historical data and
relevance of most important parameters to
identify the stream water quality are requested.
• Part 3 relates to the main stream quality
problems that WM have to face when
managing rivers and the criteria used to identify
them. This part intends to identify the minimum
data and knowledge used for WM (minimum
required data for the RBS) to tackle any
problem.
• Part 4 questions about requirements and
capabilities of the whole EDSS. It asks about
the preferences of WM with respect to the
possible outputs of the EDSS: diagnosis, range
of solutions and types of prognosis.
4.3. Empirical data from sampling sites
The behaviour of pristine rivers and polluted rivers
will be studied within the experimental campaigns
(eleven scenarios with different conditions are
considered). Both types of rivers can react differently
against high nutrient loading. We expect to find some
relationships between functional and descriptive
parameters from the processing and study of data
obtained from sampling sites. Some inductive
machine learning tools can be used to improve
knowledge discovering from empirical data ([Comas
et al., 2001] and [R.-Roda et al., 2001]).
5. FUTURE STEPS
Once all knowledge will be gained (empirical from
experimental campaigns, questionnaires and internal
forums of discussion and general from questionnaires
and literature research), efforts will concentrate on
(see figure 4):
• redefine the conceptual framework of the EDSS
according to the WM (end-users) final specific
requirements in order to build a useful support
tool,
DISCUSSION GROUPS
of EXPERTS
EMPIRICAL DATA
+
STREAMES project
CONCEPTUAL FRAMEWORK
DECISION TREES
NEW RULES
EDSS
DISCUSSION GROUPS
of EXPERTS
EMPIRICAL DATA
+
STREAMES project
CONCEPTUAL FRAMEWORK
DECISION TREES
NEW RULES
EDSS
Figure 4. Translating knowledge into a useful EDSS.
• structuring and representing the knowledge in a
decision tree (DT) fashion as a previous step to
build the KB. Afterwards, each branch of the trees
will be easily codified by means of a set of heuristic
“if-then” rules. We are also evaluating the use of
some automatic tools to translate the DT into rules.
Every decision tree will refer to a defined problem:
eutrophication, excess of ammonia, excess of
phosphorus, suspended solids, organic matter
pollution…but probably most of them will be
interrelated to others.
• After the development of the KB, the RBS will be
integrated with spatial information within the
context of a GIS system. The final EDSS must be
able to communicate with different external
applications (e.g. an external database).
• Finally, a global validation of the prototype will be
done. This validation will be conducted at two
different levels: (a) a validation of the technical
aspects of the prototype in relation to the use of this
tool by final end-users; and (b), a validation of the
outputs of the EDSS.
6. CONCLUSIONS
This paper presents a European project whose main
deliverable is an EDSS prototype to support and advice
water agencies in the management of human-altered
streams. The complex management of environmental
systems requests this kind of tools to be applied. This
project presents an important example of application of
science into the real world.
As far as we know, this will be the first time that an
EDSS is proposed to estimate and evaluate uncertainties
in the prediction of stream water quality and to support
the decision of management actions. Another important
issue is the integration of a Geographic Information
System to address spatial information for the
appropriate stream management actions.
Additionally, other innovative aspects of this project are
a) the conduction of field research that will support the
448

development of the RBS, b) the close involvement of
WM/authorities throughout the planning,
implementation and validation of this tool, c) the tight
collaboration that will exist among scientist, managers
and RBS developers during the course of the project.
This collaboration will ensure the consideration of the
stream water quality problem from several
perspectives and, therefore, the resultant product will
benefit from it. We are aware that this may be a
challenge, but at the same time we strongly believe
that it is only through a close dialogue between
academic/research and management institutions that
stream management strategies can be developed and
applied successfully.
7. REFERENCES
Avesani, P., A. Perini, and F. Ricci, Combining CBR
and constraint reasoning in planning forest
fire fighting. Proceedings of the First
European Workshop on Case-Based
Reasoning, 235-239, 1993.
Behrendt, H., P. Huber, M. Ley, D. Opitz, O.
Schmoll, G. Scholz, and R. Uebe, Nutrient
emissions into river basins of Germany.
UBA-Text and CD-ROM, 75/99, 288 S,
1999.
Calori, G., F. Colombo, and G. Finzi, Frame: a
knowledge-based tool to support the choice
of the right air pollution model, Computer
Support for Environmental Impact
Assessment, Guariso G. and B. Page (Eds.),
211-222, 1994.
Comas, J., Dzeroski, S., Gibert, K., R.-Roda, I. and
Sànchez, M., Knowledge Discovery by
means of Inductive Methods in Wastewater
Treatment Plant Data, AI communications,
14(1), 45-62, 2001.
Cortés, U., M. Sànchez-Marrè, L Cecaronni, I. R.-
Roda, and M. Poch, Environmental Decision
Support Systems, Applied Intelligence 13(1),
77-91, 2000.
Davis, J. R., T.F.N. Farley, W.J. Young, and S.M.
Cuddy, The experience of using a decision
support system for nutrient management in
Australia, Water Science and Technology,
37(3), 209-216, 1998.
Gensym. G2 Classic 5.1 Rev.7, Gensym Corporation,
Cambridge, USA, 2000.
Hadjimichael, M., P. Arunas, L. Kuciauskas, and R.
Brody, MEDEX: a fuzzy system for
forecasting Mediterranean gale force winds.
Proceedings FUZZ-IEEE96, International
Conference on Fuzzy systems, 529-534, 1996.
Rekolainen, S., J. Grönroos, I. Bärlund, A. Nikander,
and Y. Laine, Modelling the impacts of
management practices on agricultural
phosphorus losses to surface waters of Finland,
Water Science and Technology, 39(12), 265-
272, 1999.
R.-Roda, I., Comas, J., Poch, M., Sánchez-Marrè, M.
and Cortés, U., Automatic Knowledge
Acquisition from Complex Processes for the
development of Knowledge Based Systems,
Industrial and Engineering Chemistry
Research, 40(15), 3353-3360, 2001.
R.-Roda, I., Comas, J., Colprim, J., Poch, M., Sànchez-
Marrè, M., Cortés, U., Baeza, J. and Lafuente
J., A hybrid supervisory system to support
wastewater treatment plant operation:
implementation and validation, Water Science
& Technology, 45(4-5), 289-297, 2002.
Sànchez, M., U. Cortés, J. Lafuente, I. R.-Roda, and
M. Poch., DAI-DEPUR: a Distributed
Architecture for Wastewater Treatment Plants
Supervision. Artificial Intelligence in
Engineering, 10(3), 275-285, 1996.
Young, W.J., T.F. Farley, and J.R. Davis, Nutrient
management at the catchment scale using a
decision support system, Water Science and
Technology (River basin management for
sustainable development), 32(5-6), 277-282,
1995.
449