The APEC Climate Center provides global-scale seasonal climate prediction every month. However, it is problematic to apply the prediction result directly to climate application studies due to its coarse resolution in space and time. In most application studies, climate data is required to capture local basin climate at least in daily time scales. Thus, it is necessary to reproduce detailed climate data from the seasonal prediction. This procedure is called downscaling.

There are two kinds of downscaling methods: dynamical and statistical downscaling. Dynamical downscaling is based on a regional climate model that simulates atmospheric circulation around a target basin better than general circulation models in virtue of the attention to detailed topography. However, it entails high computational cost, and the resulting data has a bias which must be

appropriately corrected for usage in application studies. On the other hand, statistical downscaling can rapidly produce data with low computational burden, and the resulting data is less biased. However, it is based on an empirical relation that is not easy to understand dynamically, and requires reliable observation data collected for a long period.

In this study, we aim to develop APCC seasonal prediction downscaling scheme using a weather generator. A weather generator is a simulator that is based on a statistical model describing local basin climate and has been popularly used in the fields of hydrology and agriculture. In this study, it is employed as a temporal downscaling tool that produces daily weather scenarios from seasonal prediction. However, in general, weather generators are calibrated to climatology and thus do not express inter-annual variability.

Therefore, a new weather generator should be developed. There are two kinds of approaches to construct the statistical models of weather generators: parametric and non-parametric approaches. The parametric approach assumes that weather variables follow a parametric distribution and is adopted by site-specific weather generators. On the other hand, since multisite weather generators have to describe the weather of several sites simultaneously, its statistical model can not avoid being complicated. Thus, the non-parametric approach has been popularly employed due to its flexibility. In this study, we consider a multisite weather generator for covering a local basin rather than a single site, but adopt parametric approach in order to effectively express inter-annual variability, i.e., inter-annually varying parameters are introduced into the statistical model. Moreover, they are linked with seasonal prediction through a prediction model. Thus, the values of the inter-annual varying parameters are determined by a given seasonal prediction and we then produce weather scenarios by running the calibrated weather generator.

The proposed downscaling scheme is applied to the Nakdong river basin during the boreal winter season. First, we estimate the inter-annual variation of the basin climate using observation data. For precipitation, precipitation intensity and mean of dry-spell length are chosen as inter-annually varying parameters. They are estimated by using a weighted average technique for overcoming lack of observation data. For temperature, the intercept terms in temperature regression model are taken, which are interpreted as monthly mean of maximum and minimum daily temperatures adjusted by precipitation effect. They are estimated by the least-square

method as a usual regression analysis.

For linking the basin and global climates, we consider a regression equation as a prediction model. Especially, in order to mitigate overfitting, we employ regularized canonical correlation analysis and variable selection technique as the fitting method. Geopotential height at 500 hPa (Z500) and sea-level pressure (SLP) Z500 and SLP are utilized as predictors that come from the APCC Seasonal Prediction issued in November, a month before the beginning of the season. Z500 appears to have predictability for monthly mean of maximum daily temperature adjusted precipitation effect in December and February, and SLP does for the parameters related to precipitation in December.

1. 서론

2. 연구 자료 및 방법

2.1 연구 자료

2.2 연구 방법

2.2.1 유역기후의 경년변동성 표현을 위한 통계모형

2.2.1.1 유역강수모형

2.2.1.2 유역기온모형

2.2.2 경년변동성 예측모형

2.2.2.1 predictand의 공변동성

2.2.2.2 예측인자의 탐색과 관측장 분석

2.2.2.3 GCM 예측장 분석

2.2.2.4 회귀모형

2.2.2.5 오차항 모형

2.2.2.6 Weather Generator의 모수 조정 및 구동

3. 연구내용

3.1 낙동강 유역에서 경년변동성 통계모형 적합결과

3.1.1 유역강수모형 적합결과

3.1.2 유역기온모형 적합결과

3.2 낙동강 유역의 경년변동에 대한 예측모형

3.2.1 탐색적 사전분석

3.2.2 예측모형

4. 결론 및 토론

REFERENCE

부록