報告書番号

V04032

タイトル（和文）

非静力学気象モデルと統計的手法を用いた風速予測手法の評価

タイトル（英文）

Evaluation of wind speed prediction method by using a nonhydrostatic meteorological model and statistical method

概要 （図表や脚注は「報告書全文」に掲載しております）

非静力学気象モデルMM5を用い気象庁GPVを境界条件として, 北海道地域を対象に解像度20kmおよび5kmの予測シミュレーションを行なった。気象官署の観測値との比較を通じた精度の評価の結果,モデルの最寄地点での海面気圧は平均誤差(RMSE値)約1hPaで観測に極めて良く一致した。モデルで得られた10m高度風速は平均誤差2.14m/sであったが,10m高度および850hPa高度の風速・ベクトル風を説明変量,10m高度風速を目的変量として,月毎の重回帰分析に基づく予測を行った結果,平均誤差1.63m/sに改善した。また,20km格子の低解像度のモデル出力を用いた同様の解析結果はほぼ同程度の精度を示したことから,この解像度の段階では,高解像度化よりも統計的手法の方が有効であることがわかった。また,風速スペクトルの検討結果より,5時間より短いスケールの現象の再現性が課題であることがわかった。

概要 （英文）

Weather forecasts have been simulated for the Hokkaido area with a grid size of 5 km by applying the double-nesting method to the nonhydrostatic meteorological model MM5 (fifth-generation Penn State/NCAR Mesoscale Model) using grid point value (GPV) provided by Japan Meteorological Agency (JMA) as the lateral boundary conditions. The output of the model was evaluated by comparing it with the observational data of JMA. The sea level pressure of the model is in a good agreement with observational data; its average RMSE (root-mean-square error) and correlation coefficient are 1 hPa and 0.98, respectively. Although the RMSE of the model wind speed at 10 m height is 2 m/s, a smaller RMSE of 1.6 m/s (the multiple correlation coefficients = 0.62) is achieved by the estimation of model output statistics (MOS) on the basis of regression using wind speed and vector wind at 10 m height and 850 hPa pressure level as predictors. The difference between the RMSE of MOS based on the output of a 5 km grid model and that of a 20 km model is negligible. Therefore, it is concluded that the statistical method MOS of this resolution level is more effective than a higher-resolution model. This, however, does not suggest that a higher-resolution model is generally not better than a lower one; the accuracy of a model using a grid size of less than 5 km should be evaluated. The analysis of wind speed spectrum suggests that it is essential to reproduce well phenomena with a time scale shorter than five hours, for improvement of the prediction accuracy.

報告者

キーワード