Investigating trend detection capabilities on worldwide extreme rainfall occurrences

Abstract:

Theoretical arguments and climate projections suggest that precipitation (P) extremes are expected to increase in a warmer climate. Observational studies have started to confirm these predictions by analyzing rain gauge records with different statistical tests for trend detection. Recent work has shown that trend tests on extremes have serious flaws, which should be properly investigated and taken into account to increase the value of these tools. In this study, we use long-term (100 years) P records from rain gauges of the Global Historical Climate Network (GHCN) and (37 years) gridded P from the ERA-Interim, NCEP-NCAR and MERRA reanalyses to (i) investigate the power of several nonparametric statistical trend tests, and (ii) identify significant trends on extreme precipitation occurrence above the 90^th empirical quantiles by applying the best performing test accounting for field significance. We first conduct Monte Carlo experiments where we (i) generate synthetic timeseries of P occurrences with the same record lengths N_y of the considered databases and introduce a φ-slope trend in a certain number, N_t, of the timeseries; and (ii) apply different trend tests with varying φ, N_t and N_y. Results show that tests based on the Poisson regression are the most powerful and that accounting for field significance improves the interpretation of the results by limiting rejection of the false null hypothesis. We then apply the Poisson regression with field significance to the observed records and find the following. (1) Depending on the quantiles used for the P exceedance, 26%-40% of the GHCN stations exhibit a statistically significant trend, of which 70%-80% are positive and located mainly in the United States and Northern Europe. (2) The significant negative trends are mostly located in Australia. (3) Both positive and negative statistically significant trends emerge from the reanalyses data with consistent results across the products and with the regions identified by the GHCN dataset.