Statistical Analysis in Climate Research

  • Hans von Storch &
  • Francis W. Zwiers
Cambridge University Press: 1999. 494 pp. £65, $110

An article published last year called “Statistics education in the atmospheric sciences” by Timothy Brown and four other leading meteorological and climatological statisticians lists 16 statistical books that emphasize applications to atmospheric science. Of these, only four have the breadth and internal cohesion to qualify as textbooks for classroom use. Three of these have appeared since 1994, more than 35 years after the publication of the first, the venerable Some Applications of Statistics to Meteorology by Hans A. Panofsky and Glenn W. Brier. The most recent on the list, Statistical Analysis in Climate Research by Hans von Storch and Francis W. Zwiers, is easily the most ambitious and, from the standpoint of a practising dynamic climatologist (from graduate student to senior level), the most valuable. The book is substantial in scope, rigour and its oversized format, and extraordinarily thorough in content. Very little of importance to the subject area is left untreated, which is hardly surprising considering the stature of the authors, two of the leading academicians in the science of climate variability.

The text is advanced, serving equally well as a reference, but adopts a tone and philosophical stance absolutely crucial in this era of almost unlimited access to computer power, data and sophisticated analytical software. The book sets out “… to provide … the background needed to apply statistical methodology correctly and usefully”. The authors do not serve up cook-book recipes, “because they are dangerous for anyone who does not understand the basic concepts of statistics”. Rather, they set out to establish the base of understanding necessary to prevent “… falling into the many pitfalls specific to our field, such as multiplicity in statistical tests, the serial dependence within samples, or the enormous size of the climate phase space”. Developing respect for these difficulties for the climate scientist is the most important goal of the book.

The text is organized logically into an Introduction and six parts, entitled respectively “Fundamentals”, “Confirmation and analysis”, “Fitting statistical models”, “Time series”, “Eigen techniques” and “Other topics”. Each part (except the first) is preceded by a useful overview. All but part VI (intentionally) are tightly organized, and practically constitute mini-courses, all progressing to advanced levels. The appendices are followed by a comprehensive, large reference list. The book's figures, equations and layout are attractively presented, and examples, references and cross-references are plentiful. Traditional textbook problems are not included, but examples are sufficiently frequent and well developed to compensate for this.

The book can definitely be the basis for a series of courses covering almost all of the topics suggested by Brown et al. for an atmospheric scientist's statistical education, lacking only material on the major topics of Bayesian inference, hierarchical Bayesian analysis, and decision theory. Among the minor omissions are cluster and discriminant analysis, whose application to climate problems has been limited recently, and, surprisingly, the skewness of precipitation distributions and how they are described. Otherwise, except for topics specific to meteorological rather than climatological applications, the book is quite complete. In the context of the curriculum proposed by Brown et al., this book, complemented by both Daniel Wilk's more diverse and basic Statistical Methods in the Atmospheric Sciences (Academic, 1995) and Edward Epstein's monograph Statistical Inference and Prediction in Climatology: A Bayesian Approach (American Meteorological Society, 1985), should form the centrepiece of a climate analyst's reference shelf.