HR: 0800h
AN: H51B-0352    [Abstracts]
TI: How Many Realizations Are Needed to Detect a Significant Change in Simulations of the Global Climate?
AU: * Werth, D
EM: werth@duke.edu
AF: Duke University, Dept. of Civil and Environmental Engineering Box 90287, Durham, NC 27708-0287 United States
AU: Avissar, R
EM: avissar@duke.edu
AF: Duke University, Dept. of Civil and Environmental Engineering Box 90287, Durham, NC 27708-0287 United States
AB: A method to determine if an adequate number of realizations has been created for a numerical experiment of climate change is outlined. After performing a particular number of realizations, how is it possible to know if enough have been done to apply statistical tests with confidence? To determine if the precipitation mean and variance have reached `steady state' values, we plot the relative change as a function of the number of simulations. The changes approach zero for a large number of simulations, and we conclude that the calculated statistics represent the long term behavior of the system. Additionally, we calculate the t-value for various subsets of the data to determine how many are required to see a significant signal. In cases where the signal is detected when a fraction of the simulated data are used, we assume with a higher level of confidence that the signal is real. We also evaluate the benefit of running a single, long simulation as opposed to a sequence of shorter simulations.
DE: 1615 Biogeochemical cycles, processes, and modeling (0412, 0414, 0793, 4805, 4912)
DE: 1616 Climate variability (1635, 3305, 3309, 4215, 4513)
DE: 1626 Global climate models (3337, 4928)
DE: 1630 Impacts of global change (1225)
DE: 1632 Land cover change
SC: Hydrology [H]
MN: Fall Meeting 2005