Journal of Hydrology, 122 (1991) 129 140 129 Elsevier Science Publishers B.V., Amsterdam fie AN EVALUATION OF PROCEDURES TO ESTIMATE MONTHLY PRECIPITATION PROBABILITIES DAVID R. LEGATES Department of Geography, Collegeof Geosciences, University of Oklahoma, Norman, OK 73019 (U.S.A.) (Received March 8, 1990; accepted after revision May 1, 1990) ABSTRACT Legates, D.R., 1991. An evaluation of procedures to estimate monthly precipitation probabilities. J. Hydrol., 122: 129140. Many frequency distributions have been used to evaluate monthly precipitation probabilities. Eight of these distributions (including Pearson type III, extreme value, and transform normal probability density functions) are comparatively examined to determine their ability to represent accurately variations in monthly precipitation totals for global hydroclimatological analyses. Results indicate that a modified version of the Box Cox transform normal distribution more adequately describes the 'true' precipitation distribution than does any of the other methods. This assessment was made using a cross-validation procedure for a global network of 253 stations f'or which at least 100 years of monthly precipitation totals were available. INTRODUCTION The frequency of precipitation events often is required in meteorological, climatological, and hydrological analyses as well as in engineering, agricul- tural, and water management applications. This information is needed, for example, to account for the non-normal distribution of precipitation in large- scale intercomparison studies (Bradley et al., 1987; Diaz et al., 1989) and to assess the impact of climate change not only on the mean field (cf. Mearns et al., 1984) but also on the variance and distribution of extreme events (cf. Katz and Brown, 1989). Construction of flood regulation and storm drainage facilities as well as the assessment of the potential for crop failure from both a deficiency and an overabundance of rainfall also use precipitation probabili- ties. Many situations arise, however, where the precipitation frequency distri- bution and even the probability of extreme precipitation events (e.g. 50- or 100-year return periods) must be estimated from a fairly short (i.e. ten- to twenty-year) period of record (Dickinson, 1977). This problem is particularly accentuated when monthly precipitation totals are evaluated and by the fact that outliers greatly influence the shape of the distribution, thereby decreasing the confidence with which the frequency of extremes can be estimated (Hersh- field, 1962). Consequently, the precipitation frequency distribution is often 0022-1694/91/$03.50 (~:~ 1991 Elsevier Science Publishers B.V.