This paper presents a new test statistic for detecting an abrupt change in environmental data. The proposed test procedure provides more reliable results compared to some existing procedures such as the one proposed by Jaruskova (1997). The analytical properties of the new test statistic are also presented.

MOKHTAR B. ABDULLAH
Department of Statistics
Faculty of Mathematical Sciences
Universiti Kebangsaan Malaysia
43600 Bangi Selangor
Kuala Lumpur, 43600, Malaysia
profmokh@hotmail.com
 
 

A monitoring station at Mace Head, Ireland was opened in January 1987 to serve as a middle-latitude Northern Hemisphere site as part of the Global Atmospheric Gases Experiment (GAGE). The experiment was designed to accurately determine the atmospheric concentrations of CFC-11, CFC-12 and N2O at five globally distributed sites including the one at Mace Head.
We analyze the monthly data on N2O at Mace Head through change-point methods.  The methods applied include those that detect and estimate the presence of unknown change-points in the variability of the N2O data.  The problem of estimating unknown change-points in the variability of a time-series valued process has not been well studied in the literature.

V. K. JANDHYALA
Department of Pure & Applied Mathematics
Washington State University
Pullman  WA 99164-3113, USA
jandhyala@wsu.edu
 
 

The quality of data collected by air pollution monitoring networks is often affected by inaccuracies and missing data problems, mainly due to breakdowns and/or biases of the measurement instruments. In this paper we propose a statistical method to detect, as soon as possible, biases in the measurement devices, in order to improve the quality of collected data on line. The technique is based on the joint use of stochastic modelling and statistical process control algorithms. This methodology is applied to the mean hourly ozone concentrations recorded from one monitoring site of the Bologna urban area network. We set up the monitoring algorithm through Monte Carlo simulations in such a way to detect anomalies in the data within a reasonable delay. The results show several out of control signals that may be caused by problems in the measurement device.

MICHELE SCAGLIARINI
Dipartimento di Scienze Statistiche
Università di Bologna
Via Belle Arti 41
Bologna 40126, Italy
scagliar@stat.unibo.it
 
 

Much of the environmental policy is focused at the restoration of disturbed areas. To see whether the pursued policy has the desired effect one could monitor such areas. In environmental monitoring systems data are sampled successively in time at several sampling locations. To see whether the desired effect is achieved, one would like to start analysing the data as soon as they become available. The statistical framework for such analysis are the sequential methods. In the presented research it is proposed to use the Sequential Probability Ratio Test (SPRT) developed by Wald to test whether there is a trend in the data at hand. It is assumed that the data are multivariate normally distributed. Note that the data are multivariate, because several sampling locations are included in the study. The SPRT focuses on testing a simple null hypothesis, e.g. there is no trend, against a simple alternative hypothesis, e.g. there is a prespecified trend. Furthermore, the underlying model is completely specified, i.e. the mean vector and the covariance matrix are assumed to be known. In former research we proposed to use this SPRT not to test against simple alternative hypothesis but to test against a composite alternative hypothesis, e.g. there is a minimal relevant trend. Obviously, a completely specified model is not very realistic. Therefore, nuisance parameters are introduced into the model. Some further refinements of the SPRT are proposed in order to get a test with satisfactory properties. Before the sequential analysis can start, a fixed number of data have to be gathered to estimate the nuisance parameters. If we assume an unstructured covariance matrix, this fixed number increases with increasing number of sampling locations. We therefore propose to model the covariance matrix. Three different structures are considered and the performance of the SPRT is studied. The use of the SPRT is illustrated by an example on bat data.

MARIA SCHIPPER
Center for Biostatistics
Faculty of Biology
University of Utrecht
Padualaan 14
Utrecht, 3584 CH, The Netherlands
M.Schipper@bio.uu.nl