Parameter Perturbations with the GFDL Model: Smoothness and Uncertainty
Abstract
We found that smoothness characterizes the response of global precipitation to perturbations of 6 parameters related to cloud physics and circulation in 50-year AMIP simulations performed with the GFDL model at 1x1 degree resolution. Specifically, the AGCM depends quadratically to parameters (Fig.1a). Linearization of the derivative of a cost function (the globally averaged squared difference between model and observations; here illustrated for the entrainment rate) up to at least the 2nd order around the standard case (eo=10) proofs necessary for optimization purposes to correctly predict where the optimum value lies (Fig.1b), and reflects the relevance of the non linearity of the response. The linearization also provides indications about desirable changes in the parameters' values for regional optimization, which may be locally different from that of the global average. Uncertainty of precipitation varies from -9 to 6% of the model's standard version and is highest for the ice-fall-speed in stratiform clouds and the entrainment in convective clouds, which are the parameters with the widest range of possible values (Fig.2). The smooth behavior and a quantified measure of the sensitivity we report here are the backbones for the design of computationally effective multi-parameter perturbations and model optimization, which ultimately improve the reliability of AGCMs simulations Smoothness and optimum parameter value for the entrainment rate. a) Root mean squared error and fits based on values eo=[8,16] and extrapolated over eo=[4,6]; b) derivative of the cost function computed at different levels of precision in the linearization (blue, green and black lines) and numerically using 1) the quadratic fit n the expression of the cost function (red line) and 2) only AGCM output (pink line). Note that the linearization determines the correct value of the minimum without using any information about model's output in that point: the quadratic fit is based on data corresponding to eo=[8,16]
Uncertainty of global mean precipitation to model parameters with respect to the model standard version- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2013
- Bibcode:
- 2013AGUFM.H31I..07Z
- Keywords:
-
- 1626 GLOBAL CHANGE Global climate models;
- 1854 HYDROLOGY Precipitation;
- 3275 MATHEMATICAL GEOPHYSICS Uncertainty quantification;
- 0320 ATMOSPHERIC COMPOSITION AND STRUCTURE Cloud physics and chemistry