dc.description.abstract | ABSTRACT
The main purpose of this research is to develop a statistic forecast system for pentad to monthly scales prediction. The basic structure of this system was built by the persistence neutralization method and the linear regressive model. The persistence neutralization method filtered out the persistence of variables to distinguish the relationship between lead time and lag time. It had better performance than the persistence forecast. At first, the persistence neutralization method was used to transform the variables of predictand for neutralizing the persistence effect in climate data. Then, the predictive predictors were picked out by using the linear regressive model to develop a statistic forecast system for pentad to monthly scales prediction.
60 climate variables were used, including the outgoing longwave radiation (OLR), sea surface temperature (SST), estimated precipitation version1 (Precip), and mean sea level pressure (mslp), etc. Because each variable had different seasonal influence, the annual data were divided into six periods to construct the prediction system. First, we used the persistence neutralization method and the linear regressive model to neutralize and filter out of the persistence effect in 60 kinds of climate variables. The OLR field was used to be predictand and all 60 climate variables were used to be predictors. Each predictors had different predictive skill in different periods. We calculated the correlation coefficient and root mean square errors between OLR (predictand) and all climate variables. The spacial distribution of correlation coefficient between 40oS and 40oN was exhibited the relationship between predictand and predictors.
11 variables were selected in January and February. The correlation coefficient was more than 0.8 over the tropical Eastern Pacific and exceeded 0.6 in the north of Australia, Indonesia, Philippine, and South China Sea. In March and April, the correlation coefficient was more than 0.8 from the date line to 70oW on tropical Eastern Pacific and was about 0.6 near 120oE from 10oN to the Equator. In May and June, the correlation coefficient was 0.7 near 120oW on tropical Pacific Ocean, from 160oE to 70oW in Pacific Ocean, South America, and Australia. There was more than 0.8 in South Africa. High correlation exited from 0oE to 60oE and 40oN to 20oS in July and August. In September and October, the correlation coefficient was more than 0.7 from 120oE to 0oE and 40oN to 20oS and was 0.6 near 20oS in South America. The correlation coefficient in November and December were similar to September and October. But the atmos column precipitation water and absolute vorticity on 850hPa showed the best predictive skill to predict OLR. The high correlation areas between predictand and each predictor were dissimilar in different periods, but displayed consistency in same period.
| en_US |