Marcello M. Veiga and John A. Meech
University of British Columbia,
Department of Mining & Mineral Process Engineering,
Vancouver, BC, V6T 1R9, Canada
ABSTRACT
Informal gold mining operations in the Amazon emit about 100 tonnes of mercury annually because of poor amalgamation practice. Lack of information and the complexity of mercury transformations hinder effective action to prevent such pollution and/or monitor its extent.
In the past, mathematical equations have been formulated to deal with the correlation of natural variables and mercury in biota but often, the error range is too great to provide useful models. Projection of physico-chemical variations in an environment subjected to mercury emissions onto the extent of bioaccumulation is fraught with many uncertainties and unknowns such as internal correlations between variables and the site-specific nature of biota contamination. Often assumptions about closed systems are invalid when applied to natural environments.
This work presents a heuristic approach to this problem in which amalgamation methods and natural variables are dealt with using an IF-THEN rule-based system that concludes about levels of emission and risk of bioaccumulation. The system is able to handle uncertain or vague data and yet still provide useful output. Belief in mercury oxidation or dangerous conditions are examined using Fuzzy Logic techniques and Inference Equations. These belief levels are derived from thermodynamic equilibrium data together with field observations of natural variables. This procedure can reduce the need for sophisticated analytical equipment and can provide information and diagnosis to non-technical people.
INTRODUCTION
Use of mercury in informal mining operations in the Amazon has been an efficient and cheap method for gold recovery, however improper use is causing pollution at rates of about 1 kg per kg of gold produced. These emissions result from inadequate amalgam distillation practice (80%) and from the dumping of amalgamation tailings (20%) (Farid et al., 1991). Some miners who burn amalgam show signs of mercurialism and fish-eating people living distant from mining activities have shown high Hg concentrations in blood (Malm, 1991; GEDEBAM, 1992).
Transformation of mercury in the environment depends on natural variables that contribute to chemical and biochemical reactions. Methylation has been reported to be mediated by either aerobic or anaerobic bacteria, although it is usually enhanced under anaerobic conditions (Adriano, 1986; D'Itri, 1990). Metallic mercury dumped in watercourses must be oxidized to be available for methylation. The mercury biogeochemical cycle is complex and becomes even less clear when we consider that some bacteria strains can convert methylmercury back to the metallic form (Billen et al, 1974). Sediment adsorption can also reduce the availability of oxidized forms to methylation (Veiga and Fernandes, 1992). So how can we analyze whether an environment is propitious to methylation? Which natural variables are more important? What is the relationship between these variables and mercury accumulation in fish?
These questions have lead to studies in countries where mercury pollution has been significant. In some projects, equations were formulated that correlate natural variables with mercury in biota, but the error range is too large to provide useful models. Projection of physico-chemical variations onto the extent of bioaccumulation is fraught with uncertainties and unknowns such as internal correlation between variables and the site-specific nature of biota contamination. Often assumptions about closed systems are invalid when applied to natural environments. For example, it is reported that Hg(II) compounds are only stable at Eh levels above 500 mV (Hem, 1970) at neutral pH, but this only applies to a system at equilibrium. In natural waters, non-equilibrium conditions are common with very slow transformation rates to more stable compounds (Baeyens et al, 1979). So other redox conditions can favour Hg(II) stability.
Much of this research has identified the key environmental factors that affect bioaccumulation. In the Amazon, mathematical correlation has not been successful at characterizing monitoring programmes, but considerable heuristic information about emission and bioaccumulation exists. Despite its subjectivity, this expertise may be useful in making reasonable conclusions about hazardous situations. In this case, the heuristic observations frequently neglected by statistical approaches may be the best and only source of knowledge.
This work focuses on Education as a measure to help reduce and hopefully eliminate eventually the emission of mercury from Amazonian gold operations. We believe the provision of expert knowledge to health workers, engineers, environmentalists, inspectors and miners about mercury behaviour in the environment can be an important contribution to solving this potentially devastating problem (Meech et al, 1993; Veiga and Meech, 1993).
Our main motivation has been to develop an Expert System (HgEX) that could assess risk situations (Veiga and Meech, 1992) using methods that accommodate vague data to accumulate certainty in conclusions based on field observations and expertise about the environmental behaviour of mercury. A user may enter uncertainty about all variables into a knowledge base that contains IF-THEN rules with inference equations that calculate belief in conclusions about mercury emission, bioavailability and bioaccumulation.
The reasoning adopted to conduct a risk assessment is based on a balance of synergistic and antagonistic factors that contribute to bioaccumulation. In other words, the intensity of a specific mercury emission must be combined with factors that enhance methylation and those which decrease bioavailability to establish the risk level for organisms and man. Models have been developed using Fuzzy Logic techniques and Weighted Inference Equations to combine pieces of evidence in rules that conclude about potential risk.
When knowledge is intricate and relationships between variables are difficult to correlate, Expert Systems are a useful way to gather dispersed information and accumulate certainty about facts. Field observations are the main source of information that allow evaluation of critical situations. The HgEX system conducts its diagnosis in three steps:
Each of these issues with respect to the Amazon will now be reviewed in brief followed by an examination of the methods used to accumulate certainty in each steps toward human poisoning.
MERCURY EMISSIONS
The mining method and the amalgamation procedure determine how mercury is released into the environment. At "garimpos", mercury entering the atmosphere can represent as much as 45% of that introduced into the amalgamation process, when retorts are not used (Figure 1).
Figure 1. Mercury balance in amalgamation in Poconé, Brazil. (adapted from Farid et al, 1991)
Mercury evaporated or burnt in these operations can travel long distances with subsequent precipitation by tropical rain storms. As rain water is rich in Hg (II) species formed by oxidation of Hg gas, pollution of fish has been discovered in remote areas of the Amazon where conditions are ripe for methylation (GEDEBAM, 1992).
When metallic mercury is discharged with amalgamation tailing, its relatively low mobility in natural watercourses creates points with very high concentrations ("hot spots"). When amalgamation is conducted on dredging barges, some miners dump the contaminated tailing into the river also forming "hot spots". These spots can be identified either at the bottom of creeks or close to their margins. But metallic mercury must be oxidized to create conditions for methylation. When "hot spots" are dried or dredged and exposed to temperatures above 30 °C, evaporation can be an important way in which Hg is dispersed through the forest. In some cases, it may be preferred to leave the "hot spot" alone rather than disturbing the sediment and dispersing the mercury during supposed removal.
BIOACCUMULATION
Dark water rivers play an important role for Hg incorporation in biota (Mannio et al, 1986). Unfortunately, little is known about the process by which organic substances in solution transform Hg into methylmercury (Me-Hg). Formation of fulvic acid complexes with Hg is a plausible explanation. We have conducted some preliminary experiments in which fulvic acid solutions exposed to metallic mercury increase Hg levels in solution by 100 times. As most dark water rivers are not directly impacted by gold mining activities, natural Hg from organic soils or atmospheric Hg must be contributing to the high levels in the biota from these environments.
In Sweden, a tendency for increased mercury in fish over the years has been recorded. Correlations between geographical, physical and chemical variables and Hg in fish was established by Håkanson et al, (1988) based on very scattered data from 1386 lakes. They formulated equations correlating Hg in fish and natural variables to attempt to predict Hg-bioaccumulation based on environmental factors. The data were reduced to 57 lakes to generate an empirical equation which was statistically significant at the 99.5% level with an r2 of 0.78 between predicted and observed Hg levels. But in fact, even with their heuristically-filtered approach, this equation had a error range of about ± 50%. So the application of statistical analyses to masses of data requires significant data reduction and only generates empirical results that can used only with extreme caution. Why not use the heuristics directly to generate logical conclusions which can be explained in terms of a knowledge base search pattern?
Lindqvist et al, (1991) have also reported on monitoring programmes in 83,000 small Swedish lakes. Correlations were important to set up remedial procedures, which have been conducted since 1978. A number of controversies about the effect of some natural variables have been raised (Richman, 1988; Verta et al, 1986)but the influence on bioaccumulation of pH, humosity, conductivity, biomass, solids in suspension and Hg in sediments would seem to be applicable to most environments.
HUMAN RISK AND HEALTH PROBLEMS
Mercury accumulation in humans has two main pathways in the Amazon: a) occupational exposure to vapours and b) methylmercury transferred by the fish.
Inhalation of mercury vapours is more significant for "garimpeiros" and gold shop workers. Once in the lungs, Hg is readily oxidized forming Hg (II) complexes which are soluble in many body fluids. As well, metallic Hg is also soluble in lipids allowing a rapid diffusion through the cell membranes (alveolar walls) followed by transport by blood lipids to sensitive tissues, such as the brain. The ultimate effect of Hg and related compounds is the inhibition of enzyme action.
From knowledge of metabolism and human experience with Hg inhalation, the critical organs are: (Suzuki, 1979)
Total elimination can take several years. The half-life of mercury in the brain is longer than in the kidney, thus urine Hg levels would not be expected to correlate with neurological findings once exposure has stopped. Short-term exposure to high levels causes clinical symptoms which mainly involve the respiratory tract. Hg levels in the urine of new workers should be lower than those of workers with a longer duration of exposure. (Suzuki, 1979; Stopford, 1979).
The symptoms usually associated with mercurialism are erethism (exaggerated emotional response), gingivitis, and muscular tremors. A person suffering from a mild case of Hg poisoning is usually unaware of the illness because the symptoms are psycho-pathological, such as irritability. Such ambiguous symptoms can lead to incorrect diagnosis (Jones, 1971; Cassidy and Furr, 1978).
A level of 60,000 µg/m³ was measured by Malm (1991) in the air when amalgam is burnt in pans. Urine samples have shown Hg levels >20 ppb for "garimpeiros" burning amalgam daily, whereas levels between 10 and 20 ppb were observed for those burning amalgams 2 or 3 times per week. Normal levels are below 10 ppb. Symptoms such as visual constriction, irritability, decreased memory and metallic taste were detected among the workers of gold shops in Alta Floresta (CETEM, 1991).
When intoxication takes places via food, in particular fish, most mercury is accumulated in methylated form (Huckabee, 1979, Padberg, 1990). High Hg levels (21 to 206 µg/l) have been found in the blood of individuals living distant of mining activities in the Amazon (GEDEBAM, 1992). Normal blood levels for unexposed people is 6 to 12 µg/l. A level of 200 µg/l (ppb) is reported as the lowest blood level observed in the Niigata incident in Japan, at which significant symptoms were observed (Nelson et al., 1971).
Fish caught in one Amazonian city showed Hg levels ranging from 0.009 to 2.75 mg/kg (ppm). If fish containing 0.5 ppm Me-Hg are eaten daily, the Allowable Daily Intake of 30 µg/day would be reached for a 70-kg person by the daily consumption of 60 g of fish. Considering the Canadian limit of 13 µg/day of any kind of Hg, 60 g of a fish with 0.2 ppm of Hg are enough to reach the limit (Kurland, 1973; GEDEBAM, 1992; CWQG, 1987).
Poisoning is not restricted to miners and high Hg concentrations in the blood of children (>100 ppb Hg) have been measured. Mercurialism symptoms are not clearly identified owing to differences in the amount of Hg burnt, fish consumed, fish origin as well other masking effects such as tropical diseases, alcohol consumption, etc. A level of 6 µg/g Hg in hair is reported as the typical concentration derived from weekly ingestion of 0.2 mg Me-Hg or a daily dose of 20 µg for adults. Fish-eating people living in remote areas in the Amazon not impacted directly by mining activity, show levels as high as 150 µg/g Hg in hair, but no classical Minamata disease has yet been recognized (Malm, 1991).
WEIGHTED INFERENCING OF MERCURY EMISSION POTENTIAL
To predict the extent of mercury emission, a number of events involved with mining and amalgamation have been examined (Figure 2), derived from observations at different Amazon operations. These events must be combined to predict the degree of mercury emissions.
Figure 2. Steps involved in mining and amalgamation practiced in the Amazon.
Rules can be used to comprise all situations adopted by the miners with conclusions about the degree of emissions obtained from a suitable method that combines all pieces of evidence.
We examined four methods to combine the evidence:
The Minimum Degree of Belief method is the simplest and most often used in rule-based systems. Unfortunately the method does not transfer smoothly changes in the Degree of Belief (DoB) values from the premises to conclusion. In addition, a large number of rules and possibilities must be established to represent the expertise precisely. If all options listed in Figure 2 are considered, there are over 70,000 combinations. Although not all combinations are actually practiced by the miners, representation of sensible situations is not a simple matter.
The MYCIN method can accumulate evidence using a unique rule for each step and event (Forsyth, 1984).This reduces the rules required to 31 for each of the two areas of study (region or single site). The Certainty Factor (CF) value associated with each rule must be formulated by the Expert and represents a weight associated with each input event. The method associates information in a very synergistic fashion and although convergence to full belief is approached when all the evidence is known, significant uncertainty can still exist with much accumulated evidence. The method distributes the relationship across the knowledge base as separate rules that are slower to activate than a single equation. It is more difficult to provide satisfactory explanations and justification. A single equation can be formulated for this technique but the number or terms in such an equation increases combinatorially with the number of variables.
Bayesian Logic requires establishing a prior probability for each conclusion option (50% can be chosen for convenience). The likelihood of the conclusion being observed in the presence and absence of each event must also be determined. So, twice as many relationship factors must be considered for each input, the levels of each having significant impact on the final outcome. Our experience has been that Experts generally balk at establishing the relationships in this way.
The method we have used in this work is adapted from the basic neural equation which propagates weighed evidence to a conclusion. The model used is based on the now famous Perceptron neural network developed by Rosenblatt in 1957 (Minsky and Papert, 1969). All inputs to a node in the network are summed after multiplying by a "suitable" weighting value between 0 and 1. In Rosenblatt's original network, the summation output is passed through a "hard" threshold limit which causes the node output to be either true or false.
In our system, the output emerges from a single node as a Degree of Belief (DoB) in the output signal ranging from 0 to 100. The actual output displayed to the user is filtered through two functions: one raises the DoB by an exponent (alpha-factor) which can also vary from 0 to 100; the second function involves linguistic defuzzification of the final DoB into terminology which characterizes the extent of pollution potential from non-existent to extremely high. The alpha-factor is determined from a separate knowledge base that examines socio-political, economic and technical issues about the situation in question. For the Amazon, alpha = 1; for North America today alpha = 100; for North America and the Amazon 150 years ago, alpha = 0.01.
This approach is akin to the Fuzzy Neural Rules proposed by Meech and Kumar (1992) and Khan (1993) and the Neural-Fuzzy Expert Systems (Kosko, 1992). The Comdale/X software used to build this system allowed us to experiment in the rule base design with the myriad of options presented above. The MYCIN and Bayesian methods can be made to work as well as the Weighted Inference approach but we believe their drawbacks are significant.
Rule-based reasoning begins by establishing levels of Hg recovery for the amalgamation method being used. When all amalgamation steps are conducted properly but the amalgam is burnt in pans, emission is considered "high" for large and medium operations. But for a small mining activity, Hg emission is considered to be "not-so-high". So, using a Weighted Inference process, it is possible to combine different events to conclude about amalgamation practice and Hg recovery without making use of 70,000 rules. For example, we can say that when "amalgams are burnt in pans", belief in a high Hg emission is 3 times greater than when "amalgamation tailing is left in pools". Surely this affirmation is more important for large rather than small operations. This reasoning is based on Expert knowledge and the combination of weights for all of the variables considered are used to establish belief levels.
A term has been coined called "High Emission Factors" or HEF to represent the collection of evidence that there is a high level of mercury emission. The following Weighted Inference Equation provides the DoB in high Hg levels when the importance of each variable to emission is considered:
where:
The main advantages of this method are:
Three facts are critical in the evaluation of mercury emission and must be answered by the user otherwise the DoBHEF is not calculated:
Other information can contribute to the analysis but is not required for the system to make a conclusion. Table 1 shows the linguistic terms used to characterize the High Emission Factor.
Table 1 - Linguistic Output for High Emission Factors (alpha = 1)
Eight real cases used to formulate the Weighted Inference Equation to conclude about DoBHEF are outlined in Table 2. These cases represent most of the "garimpeiros" behaviour and mining activity in the Amazon. While exceptions may exist, it is considered that the Inference Model used in this work will be able to cope with any unusual cases. The system has been designed so that the DoBHEF is more sensitive to minor changes for single sites than it is for regions. As most regions have a variety of mining and amalgamation methods, when all these events are combined, the emission prediction is undoubtedly high for nearly all regions. Of course, this is to be expected since the total emission rate in the Amazon is about 70 tonnes Hg/year and so a high DoBHEF for a large mining region is obvious.
Table 2. Correlation of the Model Output with Observations at Mining Sites
This model was applied to a much different site located in British Columbia. A small R&D company was using a pilot plant to evaluate gold in several placer deposits. The operation began in 1990 with screening of gravels (2 mm) and amalgamation of the undersize material using a processing unit called an Electrostatic Amalgamator. This is a sophisticated amalgamation pot with an electrode in the mercury which supposedly improves fine gold particle attraction. All material is soaked in a proprietary solution containing potassium permanganate to reduce mercury surface tension. In the next step, the material passes through a 15 cm mercury layer and is discharged. Mercury oxidation definitely occurs in this apparatus and mercury droplets are dragged to the tailing.
The resulting DoBHEF for this operation was 65%, i.e. "Mercury Emission is High" for an Amazon situation. In the North American context, clearly "extremely-high" would be more appropriate. This led us to establish the alpha-factor methodology described above.
In spite of the experimental character of this operation, fortunately it was discontinued after one month of operation when the potential risk was pointed out to the owners. Other examples likely exist today considering the fact that there are a number of popular books, magazines and companies promoting mercury usage in North America for the small-time prospector. The "Gold Panner's Manual" (Basque, 1991) sold in Canada and the US, actually suggests a simple way of separating gold and mercury by "baking" amalgam in the cavity of a scooped potato. It is obvious that isolated examples exist but we believe they are not significant enough to consider that Hg pollution is a rampant problem in Western North America... at least we hope not!
WEIGHTED INFERENCING OF BIOACCUMULATION
Methylation and bioaccumulation are controlled by a number of environmental factors. Variables contributing to enhanced methylation and bioaccumulation are combined into Dangerous Environmental Factor (DEF) while variables capable of reducing mercury availability are defined as Mercury Adsorption Factor (MAF).
To conclude about the Dangerous Factor, the following variables are evaluated :
The following variables determine the importance of Mercury Adsorption Factor:
A methodology similar to that used for DoBHEF is applied to determine DoBDEF and DoBMAF. Figure 3 shows how the factors are used to establish the Potential Bioaccumulation Risk (PBR).
This approach mirrors the simple concept that risk of bioaccumulation is a function of the level of mercury emission and the degree of transformation in the environment. Mercury abatement by adsorption on fine ferruginous sediments is the only way in which the risk can decrease.
Conclusions based on DoBPBR are defuzzified into linguistic warnings such as:
"This condition is highly favourable for bioaccumulation. Check the fish. We suggest remedial procedures" or "It seems that bioaccumulation is being controlled by natural factors".
The structure of an Expert System can provide for input of different types of data. When linguistic variables are used, the system can become easy to use and more accessible to non-technical people. Fuzzy Logic devised by Zadeh (1965) employs human analysis to provide an approximate and yet effective means to describe the behaviour of situations which are too complex or too ill-defined to allow precise mathematical analysis.
Variables such as conductivity, water transparency, Hg concentration, conductivity, pH, Eh, etc. are requested from the user to be transformed into expressions to be handled in rules. For example, the Eh and pH conditions in the sediments will determine the potential for oxidation of Hg into species which can be readily methylated.
Figure 4 shows the generally accepted equilibrium situation for Hg°(aq) and HgCl2(aq) at pH<6 and [Cl-] = 10-4 (typical Amazonian situation for dark water creeks). The full line represents the case where the ratio [Hg°(aq)]:[HgCl2(aq)] = 1, i.e. a situation in which we are fully certain that a dangerous Hg level exists. The dotted line indicates the case where this ratio = 1000. We consider this situation as the limit that can be used to exclude any environmental problem caused by mercuric species, since the accepted safe limit is 0.1 ppb (Pommen, 1991). This limit actually applies to total Hg, but we believe that only mercuric species should be considered dangerous since biomethylating agents work only with mercuric species, so there is unlikely to be danger at this concentration.
Now, the presence of organic acids in the sediment can drop the potential boundary between Hg°(aq) and any complex formed with Hg(II) and so our belief in "oxidizing Eh" must change. As most natural organic acids do not have consistent chemical composition, thermodynamic data on complexes with Hg are not common in the literature. Ramamoorthy and Kushner (1975) report an equilibrium constant K = 5.4 x 1011 for the complex formed between Hg(II) and fulvic acid (say HgFA). This results in a Eh of 390 mV for [Hg°(aq)]:[HgFA]= 1, if we consider an organic concentration of 10-1 M in the interstitial water. For a concentration ratio [Hg°(aq)]:[HgFA]= 1000, the Eh value drops to 300 mV. Lövgren and Sjöberg (1989) obtained experimentally an equilibrium constant of 2.34 x 1010 for a ligand-complex formed between mercuric species and dissolved organic matter from bog-waters, say HgL (it is not specified if the organic part is fulvic acid). This give us Eh values of 340 and 250 mV for [Hg°(aq)]:[HgL] of 1 and 1000 respectively.
Mercury has a very strong affinity for sulphur-containing functional groups frequently found in organic molecules. Whether adsorption can minimize the methylation mechanisms by reducing the availability of Hg(II) in the aquatic system (Andren and Harris, 1975; Bubb et al., 1991) or can promote more methylation by reactions between Hg and organics (Rogers, 1977; Verta, 1986), is not well understood. So, the Inference Equation adopts a conservative approach and organic matter in the polluted environment produces an increase in bioavailability.
The ultimate effect of bioaccumulation is human poisoning. So the DoBPBR is combined with other information to trigger a need for biological samples collected from human subjects. The HgEx system focuses on rationalizing the sampling programme to be implemented if required.
WEIGHTED INFERENCING OF HEALTH FACTORS
Based on the environmental picture established by belief in a Potential Bioaccumulation Risk, a questionnaire is available in the program to evaluate human risk. The conclusions do not yield a definitive clinical diagnosis but rather provide advice on situations where hair, urine or blood samples should be collected. If analyses of such samples are available, the system will compare them with normal ranges and based on the symptoms observed will suggest whether an individual has mercurialism or not.
Information on the procedures necessary to conduct an initial screening study are provided. When human fluids are sampled indiscriminately without prior environmental evaluation, confusion and suspicion can spread among affected communities, since analytical results rarely come back to the sample donors and neither does treatment methods for Hg intoxication. In addition, the cost involved in collecting and transporting chilled samples to laboratories is high. Rationalization and optimization of this operation must be applied to ensure reliable data.
The questionnaire investigates undue occupational exposure of workers and possible indirect intoxication of ordinary people. Data treatment also uses a Weighted Inference Equation considering all causes and symptoms related to mercurialism. Their relative importance has been obtained from the literature and experts in the field. A factor called "Biological Samples Recommended" is coined, similar to the other environmental factors described above. This factor is the result of the difference between the Degrees of Belief in events and symptoms which suggest mercury poisoning and those facts which can contribute to a misinterpretation of intoxication symptoms such as malaria, alcoholism, gasoline handling, etc.
As before, linguistic defuzzification is used to provide conclusions such as:: "Urine samples are recommended", "Blood samples are possibly required" or "Samples are not indicated", etc.
CONCLUSIONS
Expert Systems can play an important role in transferring heuristic knowledge to non-technical people who may be exposed to mercury. Since the subject is fraught with uncertainties about the effects of natural variables, the knowledge base accommodates imprecise data based on field observations. Fuzzy Logic and Weighted Inference Equations are suitable techniques to mimic the Expert reasoning, conferring elasticity to the Degree of Belief calculated for conclusions. In this particular case, linguistic terms have the same practical effect as complex mathematical models which usually demand high costs, much data and skill to quantify the relationships between each factor and bioaccumulation. Rapid diagnosis can bring rapid decisions.
The HgEX system has been designed to accumulate common-sense knowledge with field samples and observations about variables which usually correlate positively with mercury pollution. The system alleviates the difficulties of evaluating monitoring programmes by accepting vague data and by estimating levels of importance of each variable in the final conclusion based on Expert opinion.
ACKNOWLEDGMENT
We wish to thank the Natural Sciences and Engineering Research Council for partial support of this work through Operating Grant No. . Discussions with D. Tromans, C. van Netten, R. Meadowcroft, D. Dreisinger and R. Lawrence who critiqued the methods and made many suggestions are gratefully acknowledged.
REFERENCES
Adriano, D.C., 1986. Trace Elements in the Terrestrial Environment. Springer-Verlag, New York, 533p.
Basque, G., 1991. Gold Panner's Manual. Sunfire Pub. Ltd., Langley, B.C. Canada. 108p.
Jones,H.R., 1971. Mercury Pollution Control. New Jersey, Noyes Data Co., 251p.
Kosko, B., 1992. Neural Networks and Fuzzy Systems. Prentice-Hall Inc. New Jersey, 449 p.
Minsky, M. and Papert, S.1969. Perceptrons. MIT Press, Cambridge, MA, USA.
Nelson, N. et al., 1971. Hazards of Mercury. Environmental Research, 4, p.1-69.
Zadeh, L.A., 1965. Fuzzy Sets. Information and Control, v.8, p.338-353.
Return to the publications index
Top of page