Uncovering the Forced Climate Response from a Single Ensemble Member Using Statistical Learning
- ETH Zurich (Switzerland). Inst. for Atmospheric and Climate Science and Seminar for Statistics
- ETH Zurich (Switzerland). Seminar for Statistics
- ETH Zurich (Switzerland). Inst. for Atmospheric and Climate Science
- National Center for Atmospheric Research, Boulder, CO (United States)
Internal atmospheric variability fundamentally limits predictability of climate and obscures evidence of anthropogenic climate change regionally and on time scales of up to a few decades. Dynamical adjustment techniques estimate and subsequently remove the influence of atmospheric circulation variability on temperature or precipitation. The residual component is expected to contain the thermodynamical signal of the externally forced response but with less circulation-induced noise. Existing techniques have led to important insights into recent trends in regional (hydro-) climate and their drivers, but the variance explained by circulation is often low. Here, we develop a novel dynamical adjustment technique by implementing principles from statistical learning. We demonstrate in an ensemble of Community Earth System Model (CESM) simulations that statistical learning methods, such as regularized linear models, establish a clearer relationship between circulation variability and atmospheric target variables, and need relatively short periods of record for training (around 30 years). The method accounts for, on average, 83% and 78% of European monthly winter temperature and precipitation variability at gridcell level, and around 80% of global mean temperature and hemispheric precipitation variability. We show that the residuals retain forced thermodynamical contributions to temperature and precipitation variability. Accurate estimates of the total forced response can thus be recovered assuming that forced circulation changes are gradual over time. Overall, forced climate response estimates can be extracted at regional or global scales from approximately 3–5 times fewer ensemble members, or even a single realization, using statistical learning techniques. We anticipate the technique will contribute to reducing uncertainties around internal variability and facilitating climate change detection and attribution.
- Research Organization:
- University Corporation for Atmospheric Research, Boulder, CO (United States)
- Sponsoring Organization:
- Swiss Data Science Centre; USDOE Office of Science (SC), Biological and Environmental Research (BER); National Science Foundation (NSF)
- Grant/Contract Number:
- FC02-97ER62402
- OSTI ID:
- 1609295
- Journal Information:
- Journal of Climate, Vol. 32, Issue 17; ISSN 0894-8755
- Publisher:
- American Meteorological SocietyCopyright Statement
- Country of Publication:
- United States
- Language:
- English
Web of Science
Climate change now detectable from any single day of weather at global scale
|
journal | January 2020 |
Similar Records
Timescale for Detecting the Climate Response to Stratospheric Aerosol Geoengineering
Distinct responses of Asian summer monsoon to black carbon aerosols and greenhouse gases