Standards in Climate Modeling: What worked, what didn't and why?

Gavin Schmidt NASA Goddard Institute for Space Science Standards in Climate Modeling: What worked, what didn’t and why? Climate Models have a long history • Multiple efforts in the 1970s to build general circulation models • Combining atmospheric dynamics, radiative transfer, and convection Photo: Josh Wolfe Climate Models are a product of reduction and synthesis • Previously separate communities w/distinct questions/ methods • Assimilation within the GCM paradigm which has expanded to include more Earth System Model (ESM) elements • New models able to answer a much broader range of questions Photo: Josh Wolfe CMIP1/2 CMIP3 CMIP5 CMIP6 AMIP CMIP7 2020’s MIP = Model Intercomparison Project AMIP = Atmospheric MIP CMIP = Coupled MIP AMIP (1992) • Though the 1980s, it was clear that different models gave different answers for key results (e.g. the temperature rise from 2xCO2) • Realization that community coordination was needed to understand source(s) of the difference • AMIP was proposed to help (Gates et al. 1992) • Input into IPCC SAR (1995) Initial AMIP conception X Unorganized comparisons A single experiment design: Atmospheric-only simulations driven by ocean surface conditions 1979-1988 (1979-2001 in phase 2) Common set of requested diagnostics, stored in a common format in a central server (at DOE). Approved diagnostics teams were set up to analyze specific questions. “The availability of observational data with which to validate the models' performance is essential to the success of AMIP” ~32 variables (expanded to 114 in phase 2) ~29 models (32 in phase 2) CMIP1+2 • By mid 1990s, sufficient ‘coupled’ models existed (with interactive oceans) to compare across groups • CMIP analogous to AMIP (Meehl, 1995; 1997) • Examination of “present day” control run, and 1% increasing CO2. • Input used in IPCC TAR (2001) IPCC TAR (2001) Introduction of the “multi-model ensemble” Realization that participation in the collective effort was the key to being referenced/influential Synchronization of model center development cycles with IPCC CMIP3 (2003/4) • Massive expansion of experiments and diagnostics (Meehl et al., 2007) • Merge of AMIP/CMIP • More federated subprojects incl. more ‘idealized’ configurations for which no observations are available, and paleoclimate • Move to open data! • Very influential in IPCC AR4 (2007) Introduction of the “historical” experiment and attribution runs Much greater downloads and use by broader community IPCC AR4 (2007) CMIP5 • Further expansion: • 37 experiments (tiered) • 59 models • Controlled vocabulary for models/experiments/variables • Taylor et al (2011) • Synchronized with IPCC AR5 (2016) • Belated (and ineffective) DOI assignment to output CMIP6 (2016) • Eyring et al (2016) • Immediate DOI as data published. (Not widely used tho) • “Hot model problem” (Hausfather et al, 2023) • End of “model democracy”! Used as input into IPCC AR6 projections CMIP7 (2025+) • “expansion of the DECK to include historical, effective radiative forcing, and focus on CO2-emissions-driven experiments; sustained support for community MIPs; periodic updating of historical forcings and diagnostics requests” • Dunne et al (2025) In prep for IPCC AR7…