Environmental Research 1
Presenter: Juha Heikkinen
When: Monday, July 11, 2016 Time: 2:00 PM - 3:30 PM
Room: Salon B Carson Hall (Level 2)
Session Synopsis:
Utilizing prior information in environmental inventory design - experiences from forest inventories
Currently, some prior information is available for any large-scale assessment of the environment. Typically, it comes from a satellite image or a thematic map based on an earlier inventory. There are already several examples of national or regional forest inventories, which demonstrate significant improvements in efficiency, when prior information is utilized in the field sampling design. This presentation focuses on two-phase sampling, where systematic cluster sampling is applied in the first phase. We begin by explaining, why systematic cluster sampling is so common in forest inventories. The first-phase clusters are then stratified on the basis of the prior information. We explain why simple stratified sampling is not practical. The efficiency of the presented two-phase approach is then demonstrated in several cases including national forest inventories of Finland and a pilot inventory in Kenya. We argue that a similar approach could result in great improvements in many other kinds of assessments of the environment. Finally, we shall present results of a study on the effects of the size of the first-phase sample. It should be possible to find an optimal size. On one hand, the first-phase sample should be large enough to allow for flexible allocation of second-phase sample between the strata. On the other hand, the larger the first-phase sample is the further away from systematic and the closer to random the second-phase sample is.
Environmental Research 1
Presenter: Ernst Linder
When: Monday, July 11, 2016 Time: 2:00 PM - 3:30 PM
Room: Salon B Carson Hall (Level 2)
Session Synopsis:
Statistical Downscaling of Climate Model Projections of Variables that have Temporal Dependence Structure
A popular method in climate model downscaling is the quantile translation method based on the quantile-quantile relationship between a variable of a climate model output and the corresponding weather variable at a monitoring station. The method is based on the assumption of asynchronicity of climate and weather and boils down to translating the cumulative distribution functions (cdf). In the case of purely random series, such as for extremes, the method is based on applying the relationship between the quantile pairs to future model outputs, and obtaining prediction intervals. We extend the method to variables that are dependent on covariates, have seasonality, and have temporal dependence using time series model fitting. For daily temperature averages, we propose low-order regression splines for trend and cycle, and low-order autoregressive model for dependence. A cdf transfer function is then defined for trend and cycle, and a non-parametric quantile matching is applied for the remainder. For uncertainty quantification, we propose a parametric bootstrap for the parametric part. We present an application of infrastructure design and adaptation to climate change that examines the timing of freezing and thawing of soils and road beds. There are load (weight) restrictions for trucks during early spring thawing. Our model predicts desired endpoints of freezing and thawing and provides prediction intervals. It is of interest for future planning to determine if and how climate change will affect the date and depth of spring thaw.
Environmental Research 1
Presenter: Kimihiro Noguchi
When: Monday, July 11, 2016 Time: 2:00 PM - 3:30 PM
Room: Salon B Carson Hall (Level 2)
Session Synopsis:
Short-Term Forecasting of Atmospheric Data Using Singular Spectrum Analysis
Singular spectrum analysis (SSA) has recently emerged as a powerful tool for modeling and forecasting time series data. SSA allows us to flexibly capture trend and seasonality in such data, providing accurate short-term forecasts. In this talk, we present how SSA combined with an autoregressive (AR) model can be employed to produce accurate short-term point and interval forecasts of daily maximum ozone level data. Specifically, we suggest joint modeling of mean and volatility based on data with symmetrizing range-preserving transformations. The proposed model is compared to the benchmark model without any consideration of data transformation and conditional heteroscedasticity. Moreover, we describe applications of the proposed model to other atmospheric data.
Environmental Research 1
Presenter: Nadarajah Ramesh
When: Monday, July 11, 2016 Time: 2:00 PM - 3:30 PM
Room: Salon B Carson Hall (Level 2)
Session Synopsis:
MODELLING SEASONAL RAINFALL AT FINE TIME-SCALES USING DOUBLY STOCHASTIC POISSON PROCESSES
Point process theory lends itself to the modelling of rainfall data and has been widely used for this purpose. The doubly stochastic Poisson process or Cox process, introduced in a seminal paper by Cox (1955), is a point process whose rate of occurrence is determined by a stochastic process. Models based on the doubly stochastic Poisson process provide a solid framework for analysing fine time-scale rainfall data. One form of the model arises when the underlying stochastic process becomes a continuous-time irreducible Markov process X(t) on a finite state space. Models of this form have been used to analyse rainfall data by several authors, since their likelihood can be calculated and maximized numerically. Ramesh et al. (2012) explored this class of models for analysing tipping-bucket rainfall data at a single-site. The purpose of this paper is to extend the univariate class of models for fine time-scale rainfall to accommodate seasonality in the analysis of winter season rainfall data. Seasonal doubly stochastic Poisson process models are developed and their application is illustrated in an analysis of tipping-bucket rain gauge data from Bracknell, England. One advantage of using this class of models is that their likelihood can be calculated by conditioning on the underlying Markov process. As a result, the maximum likelihood method is utilised to estimate the parameters of the proposed model. Second-order properties of the sub-hourly rainfall aggregations in discrete time intervals are used for model assessment. Ramesh, N.I., Onof, C. and Xie, D. (2012). Doubly Stochastic Poisson Process models for precipitation at fine time-scales, Advances in Water Resources, 2012; 45: 58-64.
Environmental Research 1
Presenter: Dale Zimmerman
When: Monday, July 11, 2016 Time: 2:00 PM - 3:30 PM
Room: Salon B Carson Hall (Level 2)
Session Synopsis:
A Test for Tail-Up Spatial Dependence on a Stream Network
Recently, several models have been proposed for spatial dependence among observations taken at point locations on a stream network. Among these is the so-called "tail-up" model, under which only those observations from sites that are flow-connected (i.e., one is downstream of the other) are correlated; observations taken on flow-unconnected branches of the network are uncorrelated. We adapt the Diblasi-Bowman test for spatial independence in Euclidean geostatistics (Diblasi and Bowman, Biometrics, 2001) to test for this type of dependence. The test statistic is a scaled measure of ``flatness" of the stream-distance semivariogram among flow-unconnected sites. Distributional properties of the statistic are investigated under the assumption of an independent Gaussian process on the stream network. The size and power of the test are investigated via simulation. An example illustrates the use of the test.
