Rapidly Exploring Earth System Model Temperature and Precipitation Uncertainty
Earth system models (ESMs) are high-quality, detailed models of complex, interacting physical processes in the atmosphere, land, and ocean. Extreme events, like drought and heat waves, may be represented in ESM runs. However, running simulations thousands of times is necessary to sample these extremes for a comprehensive understanding of their impact on and interaction with human systems. This limits what researchers can study and restricts their understanding of how future extreme events might affect humans. Now, researchers at the U.S. Department of Energy’s Pacific Northwest National Laboratory (PNNL) have developed an emulator called fldgen v2.0 that generates global gridded time series with internal variability and space-time and cross-variable correlation. The software is able to rapidly explore ESM factors that a more computationally expensive ESM could only have produced if run thousands of times.
This research provides a way to study temperature and precipitation extremes more quickly and on more comprehensive scales than was previously available. Researchers who explore the interactions between human and natural systems will now have the ability to generate thousands of scenarios that can include different kinds of extreme events to study. This will inform studies that focus on how humans prepare for and respond to these extreme events.
Extreme events such as droughts are both infrequent and hugely impactful to humans. Process-based ESMs have been used to simulate these events and their associated impacts but are too computationally expensive to generate a robust picture of these events and their associated uncertainties. While these models are valuable, they may not include the large numbers of extreme events that researchers studying interactions between human and natural systems may be interested in exploring. In order to create so many simulations at once, fldgen v2.0 jointly produces temperature and precipitation realizations. Fldgen v2.0 is designed in a way that allows resarchers to generate thousands of two-variable fields for their simulations, which not only retain each variable’s spatial and temporal variance and covariance, but also variation between the variables themselves. The emulation procedure in fldgen v2.0 allows for the rapid generation of thousands of temperature precipitation time series that share the same statistical properties in space, time, and across variables as the ESM would have produced if it could have been run thousands of times.