Large uncertainties in observed daily precipitation extremes over land

We explore uncertainties in observed daily precipitation extremes over the terrestrial tropics and subtropics (50°S–50°N) based on five commonly used products: the Climate Hazards Group InfraRed Precipitation with Stations (CHIRPS) dataset, the Global Precipitation Climatology Centre-Full Data Daily (GPCC-FDD) dataset, the Tropical Rainfall Measuring Mission (TRMM) multi-satellite research product (T3B42 v7), the Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks–Climate Data Record (PERSIANN-CDR), and the Global Precipitation Climatology Project’s One-Degree Daily (GPCP-1DD) dataset. We use the precipitation indices R10mm and Rx1day, developed by the Expert Team on Climate Change Detection and Indices, to explore the behavior of “moderate” and “extreme” extremes, respectively. In order to assess the sensitivity of extreme precipitation to different grid sizes we perform our calculations on four common spatial resolutions (0.25° × 0.25°, 1° × 1°, 2.5° × 2.5°, and 3.75° × 2.5°). The impact of the chosen “order of operation” in calculating these indices is also determined. Our results show thatmoderate extremes are relatively insensitive to product and resolution choice, while extreme extremes can be very sensitive. For example, at 0.25° × 0.25° quasi-global mean Rx1day values vary from 37mmin PERSIANN-CDR to 62mm in T3B42. We find that the interproduct spread becomes prominent at resolutions of 1° × 1° and finer, thus establishing a minimum effective resolution at which observational products agree. Without improvements in interproduct spread, these exceedingly large observational uncertainties at high spatial resolution may limit the usefulness of model evaluations. As has been found previously, resolution sensitivity can be largely eliminated by applying an order of operation where indices are calculated prior to regridding. However, this approach is not appropriate when true area averages are desired (e.g., for model evaluations).