P. A. Aguilera

Review: Bayesian networks in environmental modelling

Bayesian networks (BNs), also known as Bayesian belief networks or Bayes nets, are a kind of probabilistic graphical model that has become very popular to practitioners mainly due to the powerful probability theory involved, which makes them able to deal with a wide range of problems. The goal of this review is to show how BNs are being used in environmental modelling. We are interested in the application of BNs, from January 1990 to December 2010, in the areas of the ISI Web of Knowledge related to Environmental Sciences. It is noted that only the 4.2% of the papers have been published under this item. The different steps that configure modelling via BNs have been revised: aim of the model, data pre-processing, model learning, validation and software. Our literature review indicates that BNs have barely been used for Environmental Science and their potential is, as yet, largely unexploited.

Application of the Kohonen neural network in coastal water management: methodological development for the assessment and prediction of water quality.

Kohonen neural network (KNN) was applied to nutrient data (ammonia, nitrite, nitrate and phosphate) taken from coastal waters in a Spanish tourist area. The activation maps obtained were not sufficient to evaluate and predict the trophic status of coastal waters. To achieve this aim, a new methodology is proposed which uses as its starting point the activation maps obtained from KNN. Firstly, to evaluate the trophic status of the coastal waters, it consists of the development of a quadrat system which enables a better classification than the traditional classification based simply on standardized data. The new classification allows clear differentiation of water quality within the mesotrophic band. Secondly, and in order to use the activation maps as predictive tools, the trophic classification, obtained from activation maps, was transposed onto new activation maps. To do this, the activation maps of the sampling points which defined each trophic group were superimposed. To avoid unnecessary complexity and to facilitate the process, this superimposition was undertaken only where the frequency exceeded 0.05. In this way, four frequency maps related to the trophic status of coastal waters (potentially eutrophic, high mesotrophic, low mesotrophic and oligotrophic) were obtained. There was no loss of relevant information in the new maps thus obtained. These frequency maps served as the basis for the successful prediction of the trophic status of random samples of coastal waters. This methodology, based on KNN, is proposed as a tool to aid the decision-making in coastal water quality management.

Hybrid Bayesian network classifiers: Application to species distribution models

Bayesian networks are one of the most powerful tools in the design of expert systems located in an uncertainty framework. However, normally their application is determined by the discretization of the continuous variables. In this paper the naive Bayes (NB) and tree augmented naive Bayes (TAN) models are developed. They are based on Mixtures of Truncated Exponentials (MTE) designed to deal with discrete and continuous variables in the same network simultaneously without any restriction. The aim is to characterize the habitat of the spur-thighed tortoise (Testudo graeca graeca), using several continuous environmental variables, and one discrete (binary) variable representing the presence or absence of the tortoise. These models are compared with the full discrete models and the results show a better classification rate for the continuous one. Therefore, the application of continuous models instead of discrete ones avoids loss of statistical information due to the discretization. Moreover, the results of the TAN continuous model show a more spatially accurate distribution of the tortoise. The species is located in the Donana Natural Park, and in semiarid habitats. The proposed continuous models based on MTEs are valid for the study of species predictive distribution modelling.

Groundwater quality assessment using data clustering based on hybrid Bayesian networks

Bayesian networks (BNs) have become a standard in the field of Artificial Intelligence as a means of dealing with uncertainty and risk modelling. In recent years, there has been particular interest in the simultaneous use of continuous and discrete domains, obviating the need for discretization, using so-called hybrid BNs. In these hybrid environments, Mixtures of Truncated Exponentials (MTEs) provide a suitable solution for working without any restriction. The objective of this study is the assessment of groundwater quality through the design and application of a probabilistic clustering, based on hybrid Bayesian networks with MTEs. Firstly, the results obtained allows the differentiation of three groups of sampling points, indicating three different classes of groundwater quality. Secondly, the probability that a sampling point belongs to each cluster allows the uncertainty in the clusters to be assessed, as well as the risks associated in terms of water quality management. The methodology developed could be applied to other fields in environmental sciences.

Differentiation of Clams from Fishing Areas as an Approximation to Coastal Quality Assessment

The quality assessment of fishing areas on the basis of the levels of heavy metals in clams ( Chamelea gallina) was attempted by using discriminant analysis. Five metals, Hg, Pb, Cd, Cu and Zn, were analyzed in the soft edible parts of clams from three fishing areas. The descriptive methods applied to data obtained do not show enough differences between sampling stations for management purposes. Only discriminant analysis is successful in the differentiation between all fishing areas. Through the first discriminant function, the group centroids are proposed as index of different source of clams. These values standardized are proposed as coastal quality index.

Dermal exposure to pesticides in greenhouses workers: Discrimination and selection of variables for the design of monitoring programs

Dermal exposure to pesticides is one of the main sanitaryproblems which greenhouses workers face. With the dual aimsof establishing both the body part that receives the greatestexposure and the variable that has greatest influence on this exposure level, 22 pesticide application trials were performed. Trials were carried out in different greenhouse vegetable crops,using different pesticides and different spray diameters from the spray gun. In order to determine dermal exposure, the wholebody method was used. Pieces of the applicator suit were subject to an extraction procedure and their pesticide contentdetermined using GC-NPD analysis. Multivariate analysis were applied to the data obtained. Principal component analysis showed that all trials produced a high exposure level on lowerleft leg and lower right leg. Cluster analysis distinguished between three sample groups. The most and the least affectedparts were clearly distinguished. Discriminant analysis indicated that the thin drop size of the spray gun is responsible for both the differences between groups and the minimum or maximum exposure level measured on the applicatorsuit. Therefore, selecting the variables, lower legs and thindrop size, is considered fundamental in designing programs formonitoring pesticide exposure.

Selection of ecological indicators for the conservation, management and monitoring of Mediterranean coastal salinas

Salinas systems are artificial wetlands which are interesting from the viewpoint of nature conservation. They play an important role both as habitats for migratory waterbird species and as nodes of biotic connectivity networks. In the Mediterranean basin, where the coastal salinas are highly significant as alternative and complementary habitats for waterbirds, a process of abandonment occurs, and many seminatural systems of this kind are disappearing. This abandonment is having serious consequences for migratory bird populations and for the ecological role these play. In the present paper, this group of waterbird species has been used to evaluate these wetlands for conservation purposes. We have developed a methodological approach for the selection of ecological indicators for the conservation and management of these Mediterranean habitats and waterbird assemblages, the main consumers therein. The stepwise procedure developed constitutes a practical tool for this task. Application thereof enabled us to differentiate the habitats available for the waterbirds and to identify the biotic and abiotic indicators for the maintenance and management of the salina ecosystems. These variables can then be incorporated into monitoring programs.

Regression using hybrid Bayesian networks: Modelling landscape–socioeconomy relationships

Abstract Modelling environmental systems becomes a challenge when dealing directly with continuous and discrete data simultaneously. The aim in regression is to give a prediction of a response variable given the value of some feature variables. Multiple linear regression models, commonly used in environmental science, have a number of limitations: (1) all feature variables must be instantiated to obtain a prediction, and (2) the inclusion of categorical variables usually yields more complicated models. Hybrid Bayesian networks are an appropriate approach to solve regression problems without such limitations, and they also provide additional advantages. This methodology is applied to modelling landscape–socioeconomy relationships for different types of data (continuous, discrete or hybrid). Three models relating socioeconomy and landscape are proposed, and two scenarios of socioeconomic change are introduced in each one to obtain a prediction. This proposal can be easily applied to other areas in environmental modelling.

Modelling uncertainty in social-natural interactions

Socio-ecological systems can be represented as a complex network of causal interactions. Modelling such systems requires methodologies that are able to take uncertainty into account. Due to their probabilistic nature, Bayesian networks are a powerful tool for representing complex systems where interactions between variables are subject to uncertainty. In this paper, we study the interactions between social and natural subsystems (land use and water flow components) using hybrid Bayesian networks based on the Mixture of Truncated Exponentials model. This study aims to provide a new methodology to model systemic change in a socio-ecological context. Two endogenous changes - agricultural intensification and the maintenance of traditional cropland - are proposed. Intensification of the agricultural practices leads to a rise in the rate of immigration to the area, as well as to greater water losses through evaporation. By contrast, maintenance of traditional cropland hardly changes the social structure, while increasing evapotranspiration rates and improving the control over runoff water. These results indicate that hybrid Bayesian networks are an excellent tool for modelling social-natural interactions. Uncertainty has to be taken into account in Socio-ecological system modelling.Socio-ecological system is modelled by hybrid BNs.Extreme-values probabilities are provided as a new tool to assess systemic change.Hybrid BNs can represent complex systems under conditions of uncertainty.

Continuous Bayesian networks for probabilistic environmental risk mapping

Bayesian networks (BNs) are being increasingly applied to environmental research. Nonetheless, most of the literature related to environmental sciences use discrete or discretized data, which entails a loss of information. We propose a novel methodology based on continuous BNs to predict the probability that surface waters do not meet the standards, in relation to nitrate concentration, established by the European Water Framework Directive. In order to achieve our purpose, a Tree Augmented Naive Bayes (TAN), was developed and applied to estimate and map the risk of failing to meet the European standards established. The TAN models were tested by means of the k-fold cross validation method. The results revealed that the TAN model performed proper risk maps and suggested that poor water quality is highly probable in watersheds dominated by irrigated herbaceous crops. On the contrary, “good surface water status” is more likely to occur in areas where forest is notably present.