- 期刊
應用decaying average方法進行地面溫度預報偏差修正之研究
Bias Correction of Surface Temperature Prediction by Using the Decaying Average Algorithm
摘要
數值模式預報無可避免地存在系統性偏差，如何計算模式之系統性偏差並修正之，成為數值天氣預報產品應用的重要課題之一。為此，本研究使用decaying average 方法，針對中央氣象局區域模式之逐時地面溫度預報場，以中央氣象局2.5 公里解析度之地面溫度分析場為真值，計算模式之系統性偏差並進行偏差修正及其效益。首先進行權重之敏感度測試，結果指出，以權重係數0.04 進行decaying average 偏差修正會有最佳的修正效果，當權重係數為0.04 時，約60 天可涵蓋90%的誤差資訊。而在偏差修正效能分析上，模式預報的系統性偏差隨著日夜變化有所不同，大部分之預報時間為冷偏差，只在清晨與傍晚有些微暖偏差的情形，經由decaying average 方法修正後，台灣地區地面溫度預報誤差由–0.43°C 修正至0.02°C、均方根誤差也下降約16%。在預報與觀測差異分布上，不論平地或山區的冷、暖偏差，經由decaying average偏差修正後皆有顯著的改善。因此，使用decaying average 方法除了可計算模式之系統性偏差以瞭解模式預報之誤差行為外，用於偏差修正更能有效提高模式預報能力。
並列摘要
The unavoidable model systematic bias was existed due to the model initial condition, model physics and numerical schemes, etc. So, how to calculate and correct bias becomes an important issue to extensively use the numerical weather prediction products. This paper is to study the performance of using the decaying average algorithm to correct the systematic bias of the hourly surface temperature prediction from the operational regional model in Central Weather Bureau. First of all, the sensitivity of the weighting coefficient in decaying average show that the coefficient of 0.04 resolves the 90% of the whole model errors in 60 days and results in the best correction effect of the systematic bias. The characteristics of the model systematic bias involved with a remarkable diurnal cycle, which is cold bias in most of the time, and is warm bias in early morning and evening. By applying the decaying average bias correction, the bias of the surface temperature is corrected from -0.43°C to 0.02°C, and the root mean square error improved up to 16%. This study shows that the decaying average algorithm is not only able to interpolate the model error behaviors, but also can effectively remove the model errors. In addition, its computing efficiency is especially powerful to apply in an operational purpose.