Atmospheric Sciences & Global Change Division
Seeking and Destroying Algorithm Errors
Scientists at PNNL found and corrected previously unrecognized errors of up to 30% in commonly used atmospheric computer models.
Results: Two scientists at Pacific Northwest National Laboratory have identified and reduced previously unrecognized errors in atmospheric computer models, enabling detailed cloud-model simulations that are much more accurate. The team found the errors in complex mathematical formulas known as advection algorithms, which are important parts of all atmospheric models that analyze aerosol and cloud interactions. Until now, the magnitude of this problem had not been realized and documented. Detailed cloud models are in important tool for increasing our understanding of clouds and improving their representation in climate models.
Why it Matters: The atmosphere contains many gases and tiny aerosol particles—also called tracers—that move with the air. This advection, or transport, of these tracers is the most important process represented in atmospheric climate models. It also is extremely difficult to represent. Climate scientists use advection algorithms to simulate the transport of the tracers in climate models. Errors in these algorithms severely limit the model's ability to accurately simulate the effect of pollution on clouds, for example, which is one of the most uncertain processes in predicting climate change.
Methods: The scientists discovered the errors while applying a detailed simulated representation of aerosol particles and cloud droplets in a three-dimensional cloud model. Because these errors do not show up in algorithm tests that simulate a single tracer, the scientists developed and conducted new tests that involved multiple tracers. They then applied these tests to four algorithms that are commonly used in atmospheric climate models.
(Left to right) Drs. Richard Easter and Mikhail Ovtchinnikov are applying the new algorithms to more realistic climate models.
The scientists were startled to discover that when the algorithms simulated the advection of more than one tracer—a more realistic approach—errors in the algorithm were on the order of 10 to 30%. They also found that the magnitude of the errors increased as more tracers and processes were added to the atmospheric models. The good news is that after identifying and quantifying the errors, the team corrected them.
What Next: The team is putting the new algorithms to use in three-dimensional aerosol-cloud interactions in more realistic models.
Research Team: Mikhail Ovtchinnikov and Richard C. Easter.
Acknowledgments: PNNL is transforming the Nation's ability to predict climate change and its impacts. This research was supported by PNNL's Laboratory Directed Research and Development program as part of the Aerosol Climate Initiative.
Reference: Ovtchinnikov, M., and R. C. Easter: Accepted. "Nonlinear Advection Algorithms Applied to Interrelated Tracers: Errors and Implications for Modeling Aerosol-Cloud Interactions." Monthly Weather Review, doi: 10.1175/2008MWR2626.1.