Skip to main content
U.S. flag

An official website of the United States government

Observation-based metrics for evaluating typical precipitation across generations of CMIP

Presentation Date
Tuesday, December 11, 2018 at 4:45pm
Walter E Washington Convention Center 152B



Some of the most persistent systematic errors in climate models are found in precipitation, including its frequency and intensity. Three metrics that quantify characteristics of typical precipitation are the rain amount peak, rain amount width, and rain frequency peak; these were developed and applied to two observational datasets in Pendergrass and Deser (2017). Using software incorporated into the PCMDI Metrics Package (PMP, Gleckler et al 2016), we apply these metrics to daily precipitation from CMIP3 and CMIP5 generations of climate model simulations to evaluate these simulations against observations, in preparation for application to CMIP6 simulations as they become available. In this presentation we discuss a suite of tests to establish the robustness of our metrics. These tests include quantifying how our statistics are impacted by the spatial scale at which they are accumulated, the consistency or lack thereof across in the observational products used (TRMM, GPCP, and CMORPH), and sampling uncertainty as evidenced from multiple realizations of the same model.

This work is part of a broader larger effort to facilitate targeted improvement of precipitation in model development. This effort includes developing a set of observational benchmarks for various aspects of precipitation, including the metrics for typical precipitation, which will be implemented into PMP. These benchmarks will be used to evaluate previous and current generations of CMIP simulations to document the evolution of systematic biases in precipitation over the history of CMIP, and to establish a baseline skill for many characteristics of precipitation. These benchmarks and software for efficiently and systematically applying them to model simulations will be made available to model developers to enable them to gauge performance changes during model development. A future follow-on evaluation will document progress toward improvement in precipitation in the next generation of climate model simulations.

  • Gleckler, Doutriaux, Durack, Taylor, Zhang, Williams, Mason, and Servonnat (2016), A More Powerful Reality Test for Climate Models, EOS, doi:10.1029/2016EO051663.
  • Pendergrass, and Deser (2017), Climatological characteristics of typical daily precipitation, J Clim, doi:10.1175/JCLI-D-16-0684.1.
Funding Program Area(s)