*********************************
There is now a CONTENT FREEZE for Mercury while we switch to a new platform. It began on Friday, March 10 at 6pm and will end on Wednesday, March 15 at noon. No new content can be created during this time, but all material in the system as of the beginning of the freeze will be migrated to the new platform, including users and groups. Functionally the new site is identical to the old one. webteam@gatech.edu
*********************************
School of Civil and Environmental Engineering
Ph.D. Thesis Defense Announcement
Bias Correction of Outputs from Global Circulation Models: Focus on the Use of Artificial Neural Networks
by:
Sanaz Moghim
Advisor:
Dr. Rafael L. Bras
Committee Members:
Dr. Aris P. Georgakakos – CEE; Dr. Jingfeng Wang - CEE; Dr. Kou-Lin Hsu – UC Irvine
Dr. Yi Deng - EAS
Date & Time: April 1 @ 5pm
Location: Mason 2119
ABSTRACT
Climate studies and effective management plans require unbiased climate datasets. This study develops a new bias correction approach using a three layer feedforward neural network to reduce the biases of climate variables (temperature and precipitation) over the Northern South America. Meteorological variables including skin temperature, specific humidity, net longwave and shortwave radiation with raw temperature (before bias correction) for temperature, lag zero, one, two, three precipitation, and the standard deviation from 3 by 3 neighbors around the pixel of interest for precipitation are selected as proper sets of inputs to the network. The modeled data are provided by the Community Climate System Model (CCSM3). Results show that the trained ANN has generalization ability in time and space to improve the estimation error, correlation, and probabilistic structure of the estimated field. In addition, this study assesses the regionalization ability of the regression models to delineate the study domain by defining the minimum number of training pixels. The delineation of the region based on the systematic errors of CCSM3 outputs is consistent with the physical features of the domain such as land cover, topography, and atmospheric circulation patterns over the region. Results also confirm that a small number of training pixels suffices to achieve a desired response. This proposed model is an effective tool to improve the quality of the data and prediction due to two main features: first the trained model can perform well when observations are not available in time or space, and second it can save significant computational requirements, time and memory usage.
