本研究使用ERA5再分析資料，針對氣象局一步法(1-tier)全球模式在台灣地區及東亞地區之季節尺度地面（SKT）與海表面（SST）溫度變異預報進行校驗及誤差修正。針對2000-2020年，以ERA5與預報之逐月與逐季平均變異之標準差比值為修正係數，修正模式預報結果。計算模式在修正前後地表溫度之機率密度函數（PDF）、累積機率分佈函數（CDF），與修正前後之連續分級機率評分（CRPS）評估改進後之結果。結果顯示東亞地區逐月尺度下SKT與SST變異，預報誤差各發生在±2°與±1°之間，累進誤差則呈現暖變異時低估，冷變異高估的情況。經過校正程序修正模式輸出後，不論逐月或季節尺度，SKT變異皆以首月預報之改善成果較佳，SST變異則是預報後期，改善效果較好。透過霍夫莫勒圖（Hovmöller diagram）分析SST預報改善前後，在西北太平洋地區隨時間變異的演化情況，結果顯示不但能提升模式掌握此變數時空分布的演變過程，對於其冷暖變異的幅度，亦更接近於真實變化的情況。本結果可提供未來修正大氣模式初始資料場，進行次季節至季節尺度預報之重要依據。

This study aims to improve seasonal scale forecast in surface temperature anomaly from the 1-tier model (1T1) developed by the Central Weather Bureau during (CWB) 2010-2020. To improve the forecasts of skin (SKT) and sea surface (SST) temperature anomalies, the ratio of standard deviations derived from ERA5 and 1T1 is considered as a factor to correct the biases generated by the original forecast. After corrected, the results indicate that this method efficiently reduces the biases in ranges of ±2° and ±1° originally found in the probability forecasts of SKT and SST anomalies, respectively. The systematic bias associated with overestimation (underestimation) occurring in warm (clod) anomaly is also eliminated. It is also found that the model is overall significantly improved in SKT at the first forecast month, as well as in SST at the second and the following forecast months on seasonal time-scale. It should be expected that the proposed method might be deployed to the further CWB operations and improves the seasonal forecast as well.