Clarifying the importance of serial correlation and field significance in detection of trends in extreme rainfall

Rainfall extremes are expected to intensify in a warmer environment according to theoretical arguments and climate model projections. Inferential analysis involving statistical trend testing procedures are frequently used to validate this scenario by investigating whether significant changes in precipitation measurements can be detected. Recent studies have shown that statistical trend tests applied to hydrological data might be misinterpreted if (1) the analyzed time series exhibit autocorrelation, and (2) field significance is not considered when tests are applied multiple times. In this study, these aspects have been investigated using time series of frequencies (or counts) of rainfall extremes derived from long-term (100 years) daily rainfall records of 1087 gauges of the Global Historical Climate Network (GHCN) database. Monte Carlo experiments are carried out by generating random synthetic count time series with the Poisson first-order Integer-valued AutoRegressive model (Poisson-INAR(1)) characterized by different sample size, level of autocorrelation, and trend magnitude. The main results are as follows. (1) Empirical autocorrelations are highly consistent with those exhibited by uncorrelated and non-stationary count time series, while empirical trends cannot be explained as the exclusive effect of autocorrelation; moreover, accounting for the impact of serial correlation has a limited impact on tests’ performance. (2) Accounting for field significance prevents wrong interpretations of results of multiple tests by limiting type-I errors, but it may reduce test power; a careful use of local test outcomes could help identify regions with potentially significant changes where clusters of multiple trends with coherent signs are detected. (3) Statistical trend tests based on linear and Poisson regressions are more powerful than nonparametric tests (e.g., Mann-Kendall) when applied to count time series. Finally, using these methodological insights, spatial patterns of statistically significant increasing (decreasing) trends emerge in central and eastern North America, northern Europe, part of northern Asia, and central regions of Australia (southwestern North America, part of southern Europe, and southwestern and southeastern regions of Australia).