Ecological Research 2
Presenter: Miranda Mortlock
When: Tuesday, July 12, 2016 Time: 9:00 AM - 10:30 AM
Room: Salon B Carson Hall (Level 2)
Session Synopsis:
USING STATISITICAL MODELLING TO IMPROVE THE RESOURCE EFFICIENCY FOR MEASUREMENTS ON LARGE CROP TRIALS FOR PLANT SCIENTISTS.
Large trials for screening leaf area in crop trials currently require the measurement of physical characteristics such as leaf length and width of individual leaves throughout the growing season. This require a large time input by scientists. The use of automation with sensors can only collate those detailed data on relatively small plants. In sorghum trials in SE Queensland there is a large amount of detailed information on individual leaf size for many genotypes and under various environmental conditions. The aim of this work was to investigate the relationships of the leaf length and length with the individual area, in relations to total leaf number and the area per leaf over the growing season. A high quality data set was used to develop a method for estimating total plant leaf area by combining a minimum number of leaf measurements with published relationships on that estimate individual leaf area from the leaf position, total number of leaves, and the tiller number. The model provided good estimates of plant leaf area, across a range of sorghum genotypes and environmental conditions. The methods can easily be extrapolated to other C4 cereals. The models developed can significantly reduce the number of measurements required and this will contribute to the researchers' ability to increase the number of genotypes assessed within a trial and will assist the large throughput of plots for trait monitoring.
Ecological Research 2
Presenter: Ramethaa Pirathiban
When: Tuesday, July 12, 2016 Time: 9:00 AM - 10:30 AM
Room: Salon B Carson Hall (Level 2)
Session Synopsis:
Exploring ecological relevance in a priori selection of variables for species distribution models (SDMs)
Recent reviews of SDM techniques have sought to optimize predictive performance. However, the extent to which such models reflect real-world species distributions also depends heavily on the quality of the input data, such as the bioclimatic indices and habitat descriptors. This confounding effect on model performance is exacerbated by current rapid increases in the number of potential predictor variables available through improved land modelling, remote imagery and acoustic sensing. Attention given to variable selection has potentially great impacts, for instance, when the model is applied for prediction beyond the scope of the training data or for explanation. In general, practitioners using SDMs employ one of three approaches to variable selection. The simplest approach relies on an expert to select the variables, often described as a priori. A second approach explicitly incorporates variable selection into model fitting, which examines a feasible subset of all possible combinations of variables. A third approach uses model averaging, to summarize the overall contribution of a variable, again across a feasible subset of combinations. Typically, SDM users will either consider a small number of variable sets, via the first approach, or else supply all the candidate variables to the second or third approaches. Thus, there is a known tendency for under-fitting in the first approach or over-fitting with the second or third approaches. The latter automated procedures do not necessarily select the best set of explanatory variables. Rather, they select a best subset which is sensitive to the algorithm and the set of criteria used. This is because examining all possible combinations of variables becomes infeasible as the number of variables increase. Bayesian SDMs exist, with several methods previously considered for eliciting and encoding priors on model parameters. However, few methods have been published for informative variable selection; one example is Bayesian trees. Here we develop and refine implementation of an elicitation protocol for variable selection in SDMs that helps makes explicit a priori expert judgements on the ecological relevance and the quality of candidate variables, to a specific species or taxonomic group. We demonstrate how this information can be obtained then define priors and contribute to posterior analysis within Bayesian SDMs.
Ecological Research 2
Presenter: Matthew Schofield
When: Tuesday, July 12, 2016 Time: 9:00 AM - 10:30 AM
Room: Salon B Carson Hall (Level 2)
Session Synopsis:
Efficient estimation of closed population mark-recapture models in continuous-time
A key assumption in standard mark-recapture models is that of instantaneous sampling. This may not be appropriate, e.g. when sampling is conducted over several days, weeks or longer. It can be difficult to determine how many discrete sampling occasions to use and they are often determined arbitrarily. If animals are caught multiple times within a sampling occasion they are typ- ically collapsed to an indicator that denotes that it was caught at least once. An alternative is to consider continuous-time models. Here we use notions of ancillarity to understand continuous-time models. We show that they can be estimated as easily, if not easier than their discrete counterparts. We compare the continuous-time approach to the discrete approach and examine the effect of individual misidentification for the two approaches.
Ecological Research 2
Presenter: Georgy Sofronov
When: Tuesday, July 12, 2016 Time: 9:00 AM - 10:30 AM
Room: Salon B Carson Hall (Level 2)
Session Synopsis:
STATISTICAL ANALYSIS OF TRAIT VARIABILITY IN ONTOGENESIS OF EPIPHYTIC LICHENS
The concept of fitness proposed by Sir Ronald A. Fisher provided the definition of natural selection and made it possible to specify the method of quantitative measurement of selection. Natural selection can be interpreted as the contribution of different genotypes in the next generation with an overall fitness consisted of components that are specific for each species. The challenge is to establish these components and to choose a correct method of their experimental measurement. It is usually very difficult to identify specific genotypes in natural populations and, as a result, studies are often limited to the analysis of phenotypes and traits of specimens in populations located in different ecological conditions. It is very often a case in population biology that analysis of population variability implies the comparison of mean values of a trait in different populations and in different ontogenetic states. In this paper, we study morphometric traits of thalli of Hypogymnia physodes (L.) Nyl. in ontogenesis (in specimens at different ontogenetic states) in different ecological conditions (in different ecotopes) and we take into account not only the mean values (or medians) of traits but also the variability of traits. Ontogenesis of the foliose lichen Hypogymnia physodes is described on the basis of the material obtained from natural populations. Ontogenetic dynamics (diameter of thallus and the number of lobes) and the features of reproductive structures (the number and diameter of labelloid and galeated sorales) are studied in ecologically different pine forests. We show that the use of the standard analysis of variance would not be correct in this case and we propose to use nonparametric methods. We obtain that throughout the ontogenesis the dynamics of the medians and variances of traits may be either similar or different. We also show that the variabilities of traits are different in ecologically different ecotopes. This research is supported by a grant from the Russian Foundation for Basic Research, project 16-04-01198-a.
Ecological Research 2
Presenter: MING ZHOU
When: Tuesday, July 12, 2016 Time: 9:00 AM - 10:30 AM
Room: Salon B Carson Hall (Level 2)
Session Synopsis:
New Hidden Markov Models for Removal Data
Classical removal study the movement of protected animals out of the path of development projects can be used to estimate the abundance of a population within a closed area. The classic removal model (Moran, 1951) assumes a constant capture probability and all animals are available for detection throughout the study, which results in a geometric decline of removed counts of animals over time. However, the real data for some species exhibit unexpected fluctuations. The work is driven by real data on slow worms, Anguis fragilis, common lizards, Zootoca vivipara and great crested newts, Triturus cristatus, where existing methods may give rise to misleading conclusions. Reptiles and amphibians sometimes become undetectable as they may hide underground. This phenomenon can be modelled as a partial hidden process, where the underlying state process describes the movement pattern of animals between survey area and the area outside the study. We have developed an adaptive Hidden Markov Model with Robust Design which allows considerable flexibility in estimating both transitions between underlying states and the size of populations. The model is also extended by the incorporation of climatic covariates to account for time-varying capture probabilities. Comparisons are made with estimates obtained from the classic model, and it demonstrates that the performance of the new model is better under many ecological scenarios. We also consider the effect of sparse data and investigate the use of modelling different sources of data in conjunction with the removal data (Besbes et al, 2002). As the success of these techniques is clearly dependent on the reasonable number of successive primary periods with multiple sampling occasions, we explore the effect of robust design on the precision of parameters using simulation. We are also able to show which combinations of parameters are estimable (Cole et al, 2010) when robust design reduces to a single secondary occasion within primary periods. References Besbeas, P., Freeman, S.N., Morgan, B.J.T. & Catchpole, E.A. (2002) Integrating markrecapturerecovery and census data to estimate animal abundance and demographic parameters. Biometrics 58: 540547. Cole, D. J., Morgan, B.J.T. and Titterington, D. M. (2010) Determining the Parametric Structure of Non-Linear Models. Mathematical Biosciences, 228, 1630. Moran, P.A.P. (1951) A mathematical theory of animal trapping, Biometrika, 38, 307-311.