On the efficiency of the indicator approach in geostatistics

A common problem in geostatistics is to determine whether or not the value of a random field at an unsampled location exceeds a specified threshold using observed values of the random field at sampled locations. Under the indicator approach, the only information used to classify the unobserved value is whether or not the observed values exceed the threshold. This note shows that the loss of information from applying the indicator approach may be modest in the case where the underlying random field is Gaussian.     

Identifying regional seismic relationships by cross-correlation

The identification of relationships between seismicity in different regions is commonly based on establishing the statistical significance of the largest absolute value in the cross-correlogram between time series of seismicity. The assessment of significance typically ignores the selection of the largest absolute value in the cross-correlogram and consequently exaggerates significance. In this note, a simple method for correcting the significance level for selection is presented. The method is valid as long as the two series are each serially independent. An example concerning seismicity in Mexico and Kamchatka is presented. In this case, correcting for selection has a dramatic effect.     

On Testing for Change in Extreme Events

A common problem in climatology is detecting a change in the frequency or magnitude of extreme events in an historical time series. This paper compares the performance of two general approaches to this problem in a simple situation. The first approach is based on modelling the entire distribution, while the second approach focuses on the tail of the distribution. Although the second approach has clear advantages, the results of this paper suggest that it can also involve a substantial loss of information.     

The Value of Improved ENSO Prediction to U.S. Agriculture

The economic value of long-range weather prediction is measured by the increase in social welfare arising from the use of the prediction in economic decisionmaking. This paper describes a study of the economic value of ENSO prediction to U.S. agriculture. The interdisciplinary study involved the analysis of data and models from meteorology, plant science, and economics under a framework based on Bayesian decision analysis. The estimated annual value of perfect ENSO prediction to U.S. agriculture is $323 million.     

A note on the statistics of the largest geomagnetic storms per solar cycle

This note points out a problem with the way in which extreme value distributions have been fit to the intensities of the largest geomagnetic storms per solar cycle. An alternative method is described. This method is applied to observations of the three largest geomagnetic storms in solar cycles 11–22.     

On the detection of greenhouse warming

Two facts that have been cited as evidence of the onset of greenhouse warming are the extent to which the most recent value in a global temperature series is unusually warm and the observation that the four warmest years on record occurred in the 1980's. We examine these results in more detail, and we address the question as to whether these facts constitute evidence in favor of the detection of greenhouse warming. We conclude that they do not support detection unless we are prepared to attribute all warming in the data to the greenhouse effect.     

Estimating the size of the source population from a matched sample of parts

This paper discusses the problem of estimating the size of a source population from a sample of matched, disaritculated body parts. Maximum likelihood estimation and the construction of a confidence interval are described. This problem is connected to the ecological problem of estimating the number of spectes in a sampled community.     

Mapping by simple indicator kriging

The problem of predicting the type of an unsampled point in a two-type map is addressed using a procedure called simple indicator kriging. This procedure estimates the conditional probability that the point is of one type or the other given the types of sample points. Simple indicator kriging is applied to examples of certain map models. It is shown to perform well, and in some cases to be exact in a specified sense. Use of simple indicator kriging to estimate multivariate normal quandrant probabilities is explored.