Emulating Indices of Extreme Temperature and Precipitation with Computationally Efficient Algorithms
Earth system models (ESMs) provide detail-rich projections about future climate scenarios. However, these models consume too many computational resources to be efficient for researchers who need hundreds or thousands of possible futures for impact studies. Emulators are algorithms that fill that data gap by producing future climate scenarios in a computationally efficient way. They are traditionally used to project average temperature and precipitation. Now a team of researchers, led by a scientist at the U.S. Department of Energy’s Pacific Northwest National Laboratory, has emulated metrics that represent temperature and precipitation extremes.
Representing indices of extremes when modeling impacts is of critical importance, given that many significant socio-economic or environmental consequences are brought about by these climate hazards. It is necessary to do this in computationally efficient ways to avoid burdening the integrated modeling of human and Earth systems. Generating extreme indices will help future work on modeling human and Earth systems and exploring potential feedbacks. This work shows that computationally efficient emulators can substitute for costly ESM simulations over a wide range of extreme indices, opening the way to a better representation of impacts and their influences.
Several recent studies have shown that average temperature and precipitation, as well as many indices of extremes, appear to change in direct, often linear, relationship to global climate indicators such as global mean temperature. This direct relationship supports the idea of building efficient emulators to predict extremes. The researchers used two known types of emulation methods, simple pattern scaling, and time-shift, to emulate changes in a set of indices for extreme temperature and precipitation under future scenarios. The scenarios spanned global temperature increases by the end of the century of 1.5° Celsius and 2.0° Celsius compared to a pre-industrial baseline, and two higher trajectories, Representative Concentration Pathway 4.5 and Representative Concentration Pathway 8.5.
Then the researchers evaluated the emulators’ accuracy by comparing their output to results from ESMs. To do this, they devised a novel performance metric able to distinguish systematic error from the noise of internal variability, when initial condition ensembles are available to train the emulators. The researchers found that the emulation was accurate for many indices and that internal variability, rather than systematic errors, often had the greatest impact on performance. Incorporating emulations for indices of climate extremes into integrated modeling of human and Earth systems could enhance the representation of impacts and their influences, allow for a tighter coupling between models, and facilitate the exploration of possible feedbacks between the two systems.