Department of Climate and Space Sciences and Engineering in the College of Engineering at the University of Michigan

CLaSP Seminar Series - Prof. Simon Tett

Date: September 6, 2018
Time: 3:30 pm - 5:00 pm
Location: CSRB Auditorium, room 2246

Our guest for this week's CLaSP Seminar Series will be Prof. Simon Tett, of the University of University of Edinburgh, UK. Please join us! 

Title: "Does Model Calibration reduce Uncertainty in Climate Projections?"

Abstract: Uncertainty in climate projections in response to increases in greenhouse gases and other forcings is very large as exemplified by the long standing uncertainty range in Equilibrium Climate Sensitivity (ECS) of 1.5-4.5K and the more recent uncertainty range in Transient Climate Response (TCR) of 1-2.5K. TCR is related to carbon budgets that keep warming below target amounts and is more relevant to policy than ECS. Uncertainty in climate projections has been estimated by ensembles of different models (multi-model ensemble) which share many common characteristics and so are not truly independent. An alternative approach, to using the multi-model ensemble, is to perturb the parameters within a model to generate a broad range of models. Such a Perturbed Physics Ensemble (PPE) suggest a broad range of possible feedbacks at least as wide as the multi-model ensemble.

We have applied a simple optimisation technique to calibrate the old, but still well performing, HadAM3 atmospheric model against multiple observations. Order 100 simulations each of 6 years were needed to optimise the model parameters. The atmospheric model was then coupled to the ocean model, to form a perturbed HadCM3 model, and ran for twenty years. The coupled atmosphere/ocean was considered successful if over the last decade global-average temperatures were consistent with pre-industrial temperatures. Studies in which 7 and 14 parameters were optimised, with initial parameter sets set to different combinations of extreme values for each parameter. For the 7-parameter optimisations ten out of twelve cases succeeded while for the 14-parameter cases only five out of fifteen cases succeeded. A possible reason for the poorer performance of the 14-parameter cases is that some of the parameter perturbations have only a small impact on the cost function.

For each of the successful parameter studies a set of coupled model simulations were done to determine the ECS and TCR of the model. The range in ECS and TCR is comparable to internal variability for the 7-parameter parameters sets and about 10% for the 14-parameter sets likely due to changes in sea-ice parameters. The range of normalised changes in surface temperature, precipitation and annual extremes are also small. Larger differences, compared to two perturbed and calibrated ensembles, in TCR and ECS occur when the model is modified to include the indirect effect of aerosols and recalibrated suggesting that uncertainty in models arises from the structural way in which unresolved process are parametrised.


Upcoming Events

December 12th
6:00 pm - 9:00 pm
Michigan Reception at AGU
January 8th
6:00 pm - 8:00 pm
Michigan Reception at AMS
January 10th
3:30 pm - 5:00 pm
CLaSP Seminar Series - Dr. Arnaud Mahieux
January 17th
3:30 pm - 5:00 pm
CLaSP Seminar Series - Prof. Marco Velli
More Events