透過您的圖書館登入
IP:3.14.15.94
  • 期刊

臺灣地區WRF颱風系集降雨機率預報之評估、校正與經濟價值分析-第二部分:校正

Evaluation, calibration and economic value analysis of the probabilistic forecasts from WRF ensemble prediction system in Taiwan area- part 2: forecast calibration

摘要


中央氣象局WRF系集預報系統(WRF Ensemble Prediction System, WEPS)所提供的颱風降雨機率預報(probabilistic quantitative precipitation forecasts, PQPFs),具有明顯的系統性偏差。本研究嘗試對WEPS所做的降雨機率預報進行校正以修正其偏差,令降雨預報的結果更具有實用價值,讓使用者參考該預報產品做出的決策能得到最大的經濟價值。本研究使用劇烈天氣監測系統(QPESUMS)的雷達估計降雨量做為校正WEPS降雨機率的依據,校正的方法分別為線性迴歸法(Linear Regression, LR)及非線性的類神經網路法(Artificial Neural Networks, ANN),分析其校正後的可信度與區辨能力。在校正實驗中,將訓練樣本區分為全區和陸地,陸地再細分為平地(地形高度低於500公尺)和山區(地形高度高於500公尺),比較不同的訓練樣本對於校正後的結果有何影響。研究結果發現,線性的LR和非線性的ANN方法都能夠成功地修正WEPS的降雨預報。校正後明顯增加了可信度,且對於降雨事件仍具有相當不錯的區辨能力,透過Brier Skill Score (BrSS)的分析則證明校正後提升了預報能力。而將訓練樣本侷限在陸地,能夠將預報結果校正到十分接近完美可信(Perfectly reliable),並對降雨事件的區辨能力良好,預報能力也提升。

並列摘要


It is known that the WRF Ensemble Prediction System (WEPS) of the Central Weather Bureau has large systematic bias in precipitation. This study attempts to correct the systematic forecasting bias in WEPS, and improve the results of rainfall forecast to have more practical value. Users can obtain maximum economic value by making decisions based on this forecast probability product. In this study, the estimate rainfall rate of Quantitative Precipitation Estimation and Segregation Using Multiple Sensors (QPESUMS) was used as the true values for correcting the PQPFs of WEPS. The correction methods were Linear Regression (LR) and Artificial Neural Networks (ANN). Calibration results show that these two techniques successfully correct the wet bias and improve the forecast skill, and calibration effect of these two techniques are quite similar. Verification results from different areas show that there are better reliability and discrimination over land areas after calibration than all area.

延伸閱讀