HR: 1340h
AN: A13B-1167    [Abstracts]
TI: The Comparison among different cloud classification schemes using Satellite Imagery
AU: * Hsu, C
EM: Franke@ms14.url.com.tw
AF: National Taiwan University, No. 1, Sec. 4, Roosevelt Road, Taipei, 10617, Taiwan
AU: Tseng, Y
EM: yhtseng@webmail.as.ntu.edu.tw
AF: National Taiwan University, No. 1, Sec. 4, Roosevelt Road, Taipei, 10617, Taiwan
AB: Using GOES-11 satellite imagery data, two different classification techniques, namely Artificial Neural Network (ANN) and Support Vector Machine (SVM), are compared to evaluate the quality of cloud classification. The cloud data is classified into twelve types, stratus (St), Stratocumulus (Sc), Cumulus (Cu), Altocumulus (Ac), Altostratus (As), Cirrus (Ci), Cirrocumulus (Cc), Cirrostratus (Cs), Cumulus Congestus (CuC), Cs associated with convection (CsAn), Cumulonimbus (Cb), Clear (Clr), respectively. Data training and testing are verified based on the cloud classification technique in Naval Research Laboratory. Images are taken from west of the Pacific Ocean area, and training cases are built by randomly extraction of 1000, 5000 and 10000 samples. Limited improvement is achieved when a larger amount of samples is used. The attributes of data samples are extracted via Karhünen-Loöve transform. ANN was developed to mimic the neurophysiology of the human brain so as to detect the complex nonlinear relationship in the data. However, poor performance is observed when irrelevant attributes or small data sets exist. SVM is a newer statistical algorithm in machine learning, particularly suitable for pattern classification and nonlinear regression by minimizing the structural risk. It performs well for the existence of irrelevant attributes data and even small data set. Both classification methods show consistent results with overall accuracy larger than 80%. The accuracy of cloud classification using SVM is generally 3-8% better than that using ANN while the computational cost in prediction using SVM is significantly less than that using ANN.
DE: 0300 ATMOSPHERIC COMPOSITION AND STRUCTURE
DE: 0555 Neural networks, fuzzy logic, machine learning
SC: Atmospheric Sciences [A]
MN: 2007 Fall Meeting