Dynamic Site Occupancy Models: Statistical Inference of Site Occupancy Dynamics Accounting for State Uncertainty

In various ecological surveys of site occupancy, observation of occupying state is typically uncertain because of classification error. Since ignoring such a state uncertainty in general induces bias in the inference of the occupying state and the ecological processes of occupancy dynamics, classification errors need to be taken into account properly in the processes of data collection and analyses. This paper outlines the dynamic site occupancy model as a statistical model for the inference of site occupancy dynamics that account for classification error and the multistate dynamic site occupancy model as a generalization of the dynamic site occupancy model, in terms of motivations, required census design and model formulation. As an example of the multistate dynamic site occupancy model, which can potentially be applied to various ecological problems, I also present a statistical model for community dynamics that accounts for an observation error that can occur in the observation of sessile organisms and note some merits of using a dynamic site occupancy modeling framework in the inference of sessile community dynamics.

Key words: Ecology, hidden Markov model, hierarchical model, imperfect detection, Pollock's robust design.

Statistical Modeling for Longitudinal Data in Multi-sites Litter-bag Experiment Using Multivariate State-space Model

The decomposition of litter in natural ecosystems is an important process as is primary production that occurs through photosynthesis, which together determine the short- and long-term C budgets in terrestrial ecosystems. In the study of ecosystems, the litter-bag method is commonly used to evaluate the rate of litter decomposition for various sites and species while using different types of experimental manipulation and litter. In this study, we used Long-term Inter-site Decomposition Experiment (LIDET) data from a litter-bag experiment database in North America to estimate decomposition constants of two different species, a deciduous, broadleaf species: sugar maple, and an evergreen: conifer, red cedar. We applied a multivariate state-space model to evaluate litter decomposition constants and the responses of environmental factors for leaf litter of these two species. The LIDET database has time-series data of the remaining mass of litter with four replicates from each of 26 different sites. Bayesian estimation of the state space model revealed the differences of litter decomposition constants and litter precipitation responses between the two species, but no obvious difference in the temperature sensitivity parameter Q₁₀ was observed. In our model, statistical shrinkage using multi-site time series data enabled the estimation of plausible decomposition constants even in sites with data having small sample sizes.

Key words: State-space model, litter decomposition, litter-bag method, carbon cycling.

Statistical Modelling in Fisheries Science

Fisheries science is different from ecology in several ways. Because it is difficult to conduct an experiment in fisheries resource assessment and primary data are from the fishery, there can be a great volume of uncertainty, and estimation results from those data tend to be substantially biased. To cope with such problems, statistical modelling has become one of the main tools in fisheries science and is still an active field of research. Fisheries science is composed of resource assessment and management. We review statistical modellings used in fisheries resource assessment and management. There are many common features between statistical modellings of ecology and fisheries science. We expect that statistical modellings will promote and accelerate cooperation and fusion between ecology and fisheries science.

Key words: Fisheries, statistical models, resource assessment, resource management.

Modeling of Fishery Dynamics with Autoregressive State-space Models for Quantifying Management Effectiveness in the Pacific Chub Mackerel Fishery

Management strategy evaluation (MSE) has increasingly become one of the most important tools for natural resource managements in the applied ecology and fishery sciences. This paper briefly introduces the concept of MSE and then reviews the study in which statistical models describing fishery dynamics work efficiently in MSE. The study quantified management strategy of effort control actually applied to the purse seine fishery catching Pacific chub mackerel, by coupling population dynamics simulation and statistical models based on fishery data. The statistical models, generalized autoregressive state-space models, explicitly describe the relationships among fishing effort, daily catches, and biomass of chub mackerel as stochastic processes. The study revealed two important factors that affect management effectiveness: the autoregressive processes hidden in the daily catch and effort data and the fisher's behavioral change in response to increase of stock biomass. While most MSE applications tend to simplify relationships between fishing effort and catches in a linear manner without consideration of the fisher's behavior, the actual fishery dynamics considering those factors estimated from appropriate statistical models would advance MSE and improve future wildlife and fishery managements.

Key words: Fishery dynamics, management strategy evaluation, chub mackerel, effort control.

Bayesian Isotope Mixing Model for Quantification of Food-web Structure

Quantitative description of food webs provides fundamental information to understand the dynamics of populations, communities, and ecosystems. Recently, stable isotope mixing models have been widely used to quantify dietary proportions of different food resources to a focal consumer. Here we introduce a recently developed mixing model (IsoWeb) that quantifies the structure of a whole food web from stable isotope information of all consumers and resources in the web. Sensitivity analysis using realistic hypothetical food webs suggests that IsoWeb is applicable to a wide variety of food webs differing in the number of species, connectance, sample size, and data variability. Moreover, it is demonstrated that IsoWeb can deal with variation in isotopic fractionation, and can compare the plausibility of different topological candidates for a focal web.

Key words: Community ecology, stable isotope, food web, mixing model, MixSIR, SIAR.

Accuracy Comparison of Machine-learning-based Land-cover Classification Using SPOT5/HRG Data

Land cover mapping provides basic information for advanced science such as ecological management, biodiversity conservation, forest planning and so on. In remote sensing research, the process of creating an accurate land cover map is an important subject. Recently, there has been growing research interest in object-oriented image classification techniques. Object-oriented image classification consists of multi-dimensional features including object features, and thus requires multi-dimensional image classification approaches. For example, a linear model such as the maximum likelihood method of pixel-based classification cannot characterize the patterns or relations of multi-dimensional data. In multi-dimensional image classification, data mining and ensemble learning have been shown to increase accuracy and flexibility. This study examined the use of the object-oriented image classification by multiple machine learning algorithms for land-cover mapping. We applied four classifiers: Classification and regression tree (CART), Decision tree with Boosting, Decision tree with Bagging, Random Forest, and Support Vector Machine (SVM). The study area was Sado Island in Niigata Prefecture, Japan. Pan-sharpened SPOT/HRG imagery (June 2007) was used and classified into the following eight classes: broad-leaved deciduous forest, Japanese cedar, Japanese red pine, bamboo forest, paddy field, urban area, road, and bare land. We prepared four data sets with object and texture features. The number of features increases from data sets I through IV. As a result, CART was unsuitable for multi-dimensional classification. Random Forest, Decision tree with Boosting and SVM showed high classification accuracies. Furthermore, in the data set with the limited features, Decision tree with Boosting was an accurate classifier. Random Forest and SVM are effective for multi-dimensional image classification such as data sets II and III. Decision tree with Boosting is effective for image classification with limited features such as data set I.

Key words: Satellite remote sensing, land cover, image classification, object oriented, ensemble classifier.

Statistical Models for Meta-analysis in Ecology and Evolution

Meta-analysis is now the gold standard for quantitatively summarizing primary studies not only in medical and social sciences, but also in the field of ecology and evolution. Meta-analytic techniques have primarily been developed in medical and social sciences where data sets (i.e. a collection of effect sizes) for meta-analysis are likely more homogeneous, and the data probably have less inter-dependence than those in ecology and evolution. Perhaps, because of this history, two original models of meta-analysis are not actually suitable for modeling data sets from the field of ecology and evolution; the two models are known as the fixed-effect and random effects meta-analysis, both of which assume independence among effect sizes. Meta-analyses in ecology and evolution often need to deal with two types of dependence (or correlated structures) in data: 1) dependent effect sizes within studies, and 2) dependence due to phylogenetic relatedness among species. We review statistical models of meta-analysis, which have been proposed to resolve these two types of dependence. We show that multilevel modeling incorporating phylogenetic comparative methods (termed `phylogenetic multilevel meta-analysis') can appropriately handle typical meta-analytic data in ecology and evolution. We also discuss the concepts of heterogeneity (as I²) in meta-analysis and of R² in meta-regression analysis. Although statistical models suitable for ecological and evolutionary meta-analyses are now available, the use of such models are still limited. Effective educational programs are now required to introduce these suitable meta-analytic models to ecologists and evolutionary biologists.

Key words: Systematic review, quantitative review, data synthesis, hierarchical model, mixed-effects model, phylogeny.

Determinants of Consciousness toward Environmental Conservation among Ordinary Citizens in Japan, South Korea, and China

Building on the promotion of concern about environmental issues in recent years, the ordinary citizen is expected to voluntarily participate in environmental conservation. This study explores factors that influence the development of environmental consciousness among ordinary citizens in Japan, South Korea, and China, based on data collected from ``The East Asian Cross-national Survey on Consciousness toward Culture, Life, and Environment'' in 2010. A logistic regression model was used to examine the relationship between consciousness toward environmental conservation and social, political and institutional factors that facilitate its development among ordinary citizens in the three countries. It was found that the extent of emotional affinity toward nature is related to a higher level of consciousness toward environmental conservation in South Korea, Beijing and Hangzhou. Trust in corporations promotes consciousness of environmental conservation in Japan. In contrast, trust in national government promotes consciousness of environmental conservation in South Korea and Beijing. In addition, social capital enhances environmental consciousness in Hangzhou.

Key words: Probability cross-national comparison, environmental consciousness, social survey, the ordinary citizens, opinion.