PLENARY SPEAKERS

Topic: Information and complexity of ecosystems

   Professor Sven Erik Jørgensen
   Afflilation: Royal Danish School for Pharmacy, Denmark

Abstract: For estimtion of eco-exergy, for qunatificatin of the evolution, for comparisons of the complexity of two organisms or two ecosystems, and for evaluation of ecosystem health, it is important to have a measure of organism complexity. Exergy estimations based on biomass and information for organisms can with good approximation be found as: Ex = ß c, where c is the concentration of biomass and ß is a weighting factor, that accounts for the information that the organisms carry (Jørgensen, 2002). The determination of ß for various organisms has been based on the number of coding genes, but recent research has shown that some of the non-coding genes are crucial for the control, maintenance, and development of the organisms. The results (Eichler and Sankoff, 2003) of ongoing whole-genome projects have therefore be applied in order to obtain more accurate ß-values. These new ß-values are several times bigger than the previously applied values. The number of amino acids coding per gene has probably been underestimated in the previous calculations. However, applications of the former values, for instance in ecosystem health assessment, where exergy is used as ecological indicator (referred as Exergy index) and in the development of structurally dynamic models, are still valid. Because the exergy calculations were applied only as relative measures.

Several indirct methods to determine the complexity of organims are presented. It is shown that the ß-value, which can be considered a measure of organism complexity, are well correlated to the age of the organisms (mya), to the number of cell types, to the minimum DNA-content, to the ratio non-coding genes vs. total number of genes (Mattick, 2003) and to the ß-values, determined by Fonseca et al. on basis of the total amount of DNA. Indirect determinations were therefore able to expand and improve the previous list of ß-values. The previous list had only 19 values, while the list based on the whole-genome project has 16 ß-values. The expanded list presented in this paper contains 56 ß-values. To reduce the uncertainty of the values, although assuming an apparent loss of discriminating power, it was decided to lump some organisms together in one group when it was know from the evolutionary tree that the organisms were closely related. It implies that the averages of ß-values determined by different methods were applied, which should give a higher certainty. The result is a list with 45 ß-values, that hopefully will improve the use of ß-values to calculate the exergy for assessment of ecosystem health and for the development of structurally dynamic models.

 (References in the abstract will be listed soon.)

 Topic: Lattice models for forest dynamics: mathematical analysis

   Professor Yoh Iwasa
   Afflilation: Kyushu University, Japan

Abstract: Recent years, spatial data of forest became available from a long-term permanent plots and remote-sensing studies. These together with rapid development of mathematical tools of spatial ecology have made forest ecosystem as an important focus of theoretical ecology. I will talk several examples of theoretical analyses of spatial processes motivated by forests ecosystem studies. [1] Wave regeneration in subalpine Abies forest is an example of large scaled pattern formation in forest ecosystem, in which tall trees exposed to wind experience enhanced mortality, which would spontaneously produce a large scale traveling wave of tree regeneration. We can show that two-dimensional models generate more regular patterns than one-dimensional models, and that the effect of stochasticity tend to generate more regular wave pattern. [2] Spatial data of vegetation height of Barro Colorado Island (neotropical seasonal forest) and Ogawa (cool temperate forest) show that the rate of tree falls increases with the number of neighbors of short vegetation height (gap sites), indicating interaction between neighbors. We can prove that the spatial patterns generated by this spatial Markov chain is mathematically equivalent to the Ising model in physics. The dynamics can be usefully analyzed by pair-approximation and other moment closure methods, showing the importance of considering spatial clumping of gaps. [3] The analysis of spatial patterns by the variance-quadrat size plot can discriminate the two-state spatial Markov chain, three-state model (developed for mussel beds in the rocky intertidal), and the wave-regeneration model. According to statistical analysis, spatial data of BCI and Ogawa forests are more consistent with wave-regeneration model than two-state model. [4] We discuss the masting or intermittent reproduction of forests trees synchronized over hundreds of kilometers, observed for beech, oak, and many other trees. We introduce a coupled map lattice model for the evergy reserve of individuals. The need of receiving outcross pollen by other individual makes trees synchronized (pollen coupling). However for the trees to syhchronize over a long distance as observed, both pollen coupling and environmental fluctuation are needed.

 Topic: Feature Extraction and Computational Intelligence

   Professor Evangelia Micheli-Tzanakou
   Afflilation: Rutgers University, USA

Abstract: One of the major problems a researcher faces is what is learned from data obtained by various methods and different techniques. This presentation will discuss and compare topics such as: Statistical Advances and Challenges, as well as Feature Extraction in Computational Intelligence methodologies.

Often a simple model describes the data well, simply because the S/N ratio is too small for detection of more complex structures-which for example is the case with medical data involving human subjects. One has a lot of variability both in intra- and inter-sets of data. Some important simple tools used for a long time are: Linear Regression, Discriminant analysis, Principal Component Analysis etc. In all of these, the size of the data set matters. Huge data sets create memory problems. The question is how do we handle different data types and how do we handle them? What if the data are correlated? What if we have complex data structures?

Some examples of “features” will be given and different feature extraction methods will be discussed in combination with computational intelligence algorithms. One algorithm in particular, ALOPEX, will be presented as used in many applications. The main characteristic of this algorithm is that it is biologically inspired, it adaptive and it updates all parameters simultaneously. At every iteration, it also checks for its performance and automatically adjusts itself to avoid local minima or maxima. Its importance can be seen from the fact that changes the flow of information and from a feature extractor it becomes a feature generator with minimum cost. In combination with modular neural networks and fuzzy logic, this algorithm acts as an integrator of available data in generating new patterns of information.

 KEYNOTE SPEAKERS

Topic: Using ecoinformatics tools to model hierarchically-structured aquatic ecosystems with implications for conservation

   Professor LeRoy Poff
   Afflilation: Colorado State University, USA

Abstract: With the recent increase in the collection, synthesis, and availability of ecological data, ecologists are now faced with the difficult and exciting challenge of developing robust predictive models that relate various ecosystem states to key descriptors of the environment. Often, such datasets are assembled where some ecological response and set of environmental variables are measured across many sites at different scales, yet the hierarchical nature of the data is rarely appreciated in the modelling process. The complex and non-linear relationships between environmental variables and ecosystem responses presents an opportunity to better employ ecoinformatics tools to develop more robust predictive models. However, unless these tools are applied with consideration of the natural structure in the environmental variables in mind, the success and applicability of the resulting predictive models may ultimately be limited.

In this paper, we develop a hierarchical artificial neural network (ANN) and apply it to a dataset from almost 300 stream sites collected as part of the US Environmental Protection Agency's program to monitor ecological health of the western United States. The goal of the analysis is to model the "ecological health" (measured as benthic macroinvertebrate community composition) across the sites in terms of multi-scale environmental variables, collected at the local (habitat) and whole catchment scales, both of which are known to be important predictors of stream health. We develop a hierarchical ANN to distinguish between local and regional influences on the stream health response variable. To test the utility of this approach, we compare the predictive ability of this hierarchical ANN against the performance of an unconstrained ANN (null model).

Our application illustrates the importance of constructing predictive models that incorporate knowledge about the underlying processes operating to structure aquatic ecosystems at multiple spatial scales. Integrating complex machine-learning techniques with hierarchically-structured data has implications for conservation at the landscape scale, where ecological restoration of localities often requires both local scale remediation, but within a catchment context.

 Topic: Habitat monitoring using sensor networks: a review

   Professor S. S. Iyengar
   Afflilation: Louisiana State University
   Professor E. C. Cho
   Afflilation: USA & Kentucky State University, USA

Abstract: A Distributed Sensor Network is a set of scattered intelligent sensors designed to obtain measurments from the environment., abstract relevant information ,and derive appropriate inferences.Interst in these systems stems from a realization of the application to data driven problems of habitat monitoring. The search for efficient DSN structures for data collection in unsructured environment has become an important research problems in Ecological informatics. In this seminar we address the following questions.

1. What environmental factors are important in monitoring the habitat population?
2. What patterns can be formulated based on Data driven analysis?
3. How do we characterize the behaviral pattern of the habitat based on the spatial sampling of the habitat?

The goal of this seminar is to provide a forum for discussing a model based solution to problems arising from the data complexities of these systems.An overview of both theoretical and application of sensor networks to these problems arising in Ecosystems.

 Topic: Sensitivity analysis in practice

   Dr. Andrea Saltelli    
   Afflilation: IPSC, European Commission, Italy

Abstract: Quite often sensitivity analysis (SA) is identified, almost as if it were a mathematical definition, with a differentiation of the output with respect to the input. This definition is coherent with a vast set of applications of SA to, e.g., the solution of inverse problems, the estimation of expensive models in the neighbourhood of a given set of boundary conditions and others. There is a vast literature on efficient ways of computing, directly or indirectly, matrices of variously normalised system derivatives. This approach to sensitivity has prevailed in the modelling community, also when the objective of the analysis was to ascertain the relative importance of input factors in the presence of finite ranges of factors uncertainties. In order to show how practices are at present, Science Online has been searched to identify and then review all recent articles having "sensitivity analysis" as a keyword. We contrast the present practices, mostly based on derivatives or one-factor-at a time (OAT) approaches, with more recent available good practices, such as variance based measures. These are able to overcome OAT shortcomings, are easy to implement, and allow the concept of factors importance to be defined rigorously. The role of sensitivity analysis in the scientific method is also discussed.

 Topic: Computational approaches to mate choice: insights into the evolution of brain and behavior

   Professor Steven Phelps    
   Afflilation: University of Florida, USA

Abstract: All of animal behavior can be regarded as a series decisions, and perhaps none of these decisions are more thoroughly studied than mate choice. One clear theme to emerge from the study of mate choice is that the mechanisms of behaviors have important evolutionary consequences. I review series of studies in which we employ simple computational methods, including neural network models and psychophysics, to model behavioral mechanisms. This approach enables us to derive and test empirical predictions in a well studied species, the tungara frog (Physalaemus pustulosus). We find that genetic algorithms allow find reasonable solutions to the problem of call recognition, make informative predictions regarding the influence of evolutionary history on current decisions, and are remarkably good predictors of the behavior of female frogs. In the second set of studies, we show how basic principles of psychophysics and economics can be applied to mating decisions. Doing so leads to a statistical framework that allows us to estimate and compare female preferences from a variety of tasks, and clarifies relations between mating judgments made in the context of mate choice and species recognition. I conclude with current work that aims to integrate the evolution of nervous systems and gene expression into a broader framework of animal decision mechanisms.

 Topic: Ecological informatics and grappling with the complexity of microorganisms in ecological system

   Professor Peter Noble & David Stahl    
   Afflilation: University of Washington, USA

Abstract: Unraveling the complexity of microbial communities is a major challenge because: (i) most microbes in the environment remain uncharacterized, (ii) the influence of environmental gradients on the spatiotemporal patterns of microbial communities is essentially unknown, (iii) statistical approaches to analyze complex, nonlinear microbial community and short environmental time series data sets are not widely available, and (iv) cost-effective molecular approaches to rapidly characterize microbial communities have not been developed. DNA microarray technology offers an approach to characterize environmental microbes because it provides parallel nucleic acid hybridizations for a large number of immobilized oligonucleotide probes targeting microbes at different levels of taxonomic resolution. Although DNA microarrays are reasonably well established for studies of model organisms in well-defined laboratory settings, the application of this technology to uncharacterized microbial diversity imposes additional demands on implementation; particularly for the requirement for adequately discriminating between target and non-target nucleic acids in undefined mixtures. Our approach to ensure adequate discrimination between probes and target sequences is to record and analyze thermal dissociation (melt profiles) of probe-target duplexes using artificial neural networks (NNs). NNs were used to discriminate nonlinear melting profiles of target and non-target populations that differ by a single or multiple internally mismatched base-pairs. This level of specificity is needed to resolve variants of highly conserved rRNA genes and to distinguish between closely related target and non-target microorganisms. My talk focuses on NNs, microarrays, and microbial communities.