報告書番号

T98055

タイトル（和文）

統計気候モデルを用いた地域気候変化予測手法の開発（その2） -－東アジアにおける上層気温場からの地上気温の推定－-

タイトル（英文）

Development of Regional Climate Change Prediction Method Using Statistical Downscaling (2)- An Application to the Temperature Field in East Asia -

概要 （図表や脚注は「報告書全文」に掲載しております）

全球気候モデルによって予測される粗い格子点スケールの気候変化情報を地域スケールに翻訳するため、東アジアの気温場について、統計気候モデルを構築し、全球モデルの結果に適用した。特異値分解に基づいた回帰手法を用いて、1月の月平均気温を対象として、東アジア地域の上層データから日本の地上データを推定する統計モデルを構築した。このモデルを、NCAR（米国大気研究センター）が開発した全球気候モデル（CSM）による現状気候再現実験の出力結果に適用し、得られた翻訳結果を物理気候モデル（地域気候モデル）の出力結果と比較した。両手法において高温／低温偏差の傾向は一致したが、多くの地点で物理モデルよりも統計モデルの方が高温に推定した。この差は統計モデルによる北海道の再現性の低さに起因するほか、物理モデルの持つ低温バイアスを反映しているものと考えられる。

概要 （英文）

In this study, a statistical downscaling method, which is developed in order to statistically interpret the output of coarse-resolution global climate models into regional-scale climates, has been evaluated through the application to the winter temperature field in East Asia. A regression model is constructed to relate January mean surface air temperature in Japan and upper-air temperature in East Asia, using singular value decomposition (SVD) of the covariance matrix between the two temperature fields. The surface air temperature anomaly is reproduced well using the three SVD leading modes. The first mode describes the temperature anomaly throughout Japan, whereas the second mode specifically describes the climate in the northernmost island of Hokkaido. However, even when the higher modes are included, the temperature in Hokkaido is not reproduced as well as those in other areas. The fact indicates that the temperature in Hokkaido seems to be influenced significantly by factors other than the upper-air temperature in the region analyzed. The relationship is then applied to the output of present climate run of a global climate model CSM (Climate System Model) developed by NCAR (National Center for Atmospheric Research) in order to reproduce the surface air temperature in Japan. The result is compared with the surface air temperature simulated by RegCM (Regional Climate Model version 2.5 developed by NCAR) nested in the CSM. Both warm and cool climate features are reproduced well by both methods, although the climate simulated by RegCM is cooler at most stations than that estimated using the statistical downscaling method. The difference is especially marked in some peripheral areas, including Hokkaido. The difference appeared in Hokkaido seems to be attributed to the aforementioned factors excluded in the statistical downscaling analysis, whereas the difference in other areas might be attributed to a tendency of RegCM to underestimate the surface air temperature in Japan. It is indicated through the analyses that there is a certain relationship between the surface and upper air temperature fields in East Asia in January; thus the statistical downscaling method might be applied to the region.

報告者

キーワード