全天空短波輻射量為主要的常規氣象觀測參數之一，此資料廣泛應用於氣候變遷與太陽能應用領域。自2002年起，中央氣象局於各局屬氣象站全面更新為Eppley公司出產的Precision Spectral Pyranometer（PSP）進行觀測，屬於具有高準確度符合A級（class A）之輻射計，然而，未搭配統一的校正程序及資料品管流程（Quality Control, QC），以致資料應用上產生疑慮。因此，本研究建置一套適合臺灣氣象局輻射資料之QC流程，檢視2002至2020年共達19年，30個局屬氣象站地面自動氣象觀測系統（ASOS）之全天空短波輻射資料，標註可疑或異常的觀測資料，並給予資料相對應之旗標（Flag 0-7），以提升大氣輻射資料的可用性。長期資料品管流程可分為兩階段檢驗流程，第一階段以單站長期資料進行處理，先濾除異常值後，進行物理可能極限（Physically Possible Limits）與極端最小極限（Extremely Rare Minimum Limits）檢定，用以偵測資料中異常大的誤差值與較大的隨機誤差，此階段30個局屬站之資料品保可用率皆超過90％，接著進行每日氣候最大值（Daily Climate Maximum）檢定，檢視資料於連續6個月期間，掌握是否存在儀器老化、校正係數偏離、儀器毀損故障等情形，扣除有上述疑慮之資料後，各測站之資料品保可用率下降至53－95％之間，其中校正係數偏離是影響資料品質的主要原因，凸顯定期校正輻射儀器之重要性。第二階段為使用第一階段篩選出的高品質資料，進一步加入鄰近測站倆倆進行相關性及時序比較，我們進一步檢查出資料處理器（data logger）所記錄的異常值，並予以濾除；同時在這個階段我們也加入人工判定，註記有疑慮的觀測資料區間。最後，本研究以2020年為基礎，比對QC後高品質地面觀測資料與Himawari-8衛星反演日射量差異，結果顯示兩者的相關係數高達0.96，各月的相關係數介於0.94至0.97。進一步檢驗衛星反演在時空上的表現，結果顯示，在花蓮地區衛星反演比地面觀測高約27.9％，北部地區則有約4.6％的低估，月份上，偏高情形以3月的6.5％最為明顯，整體來說，衛星反演相比地面觀測普遍呈現高估的情形。本研究發展出一套嚴格的輻射資料QC流程，確認過去長達19年的地表太陽輻射資料可靠度，有利於後續科學研究發展。

Solar shortwave irradiance is one of the main conventional meteorological observation parameters which is widely used in the fields of climate change and solar energy. Since 2002, the Central Weather Bureau (CWB) in Taiwan has upgraded its weather stations' pyranometer with the Precision Spectral Pyranometer (PSP) manufactured by Eppley Company. The PSP is a class A pyranometer with high accuracy. However, the lack of complete procedures for calibration and data quality control (QC) has led to doubts about the data usage. Therefore, this study aims to establish a research-based QC procedure for the solar radiation data obtained from 30 weather stations of CWB in 2002 to 2020 (a total of 19 years). Suspicious or abnormal data are given corresponding flags (Flag 0-7) which can serve as a guidance to users who using the data set. The long-term data quality control procedure we developed can be divided into a two-stage inspection process. The first stage is to process long-term data station-by-station. For any station, the procedure starts with filtering out abnormal values, then the Physically Possible Limits and Extremely Rare Minimum Limits are applied, these two limits testing are used to detect abnormally large error values and large random errors in the data. At this stage, the data quality assurance availability rate of the 30 weather stations is all greater than 90%. Finally, the Daily Climate Maximum test (i.e., verifying within 6 consecutive months, whether there exists a circumstance of instrument aging, calibration coefficient drafting, instrument damage, etc.) is applied. After removing the data with the above test, the data quality assurance availability rate of each weather station drops to 53-95%. It is worth to mention that the calibration coefficient drafting is the main reason that affects the quality of the data. It highlights the necessary of a periodic calibration for solar radiation instruments in the network. For the second stage, we used the high-quality data screened in the first stage, and further diagnose the correlation and liner regression between two adjacent stations. We accidentally found the outliers recorded by the data logger. The bad data has been filtered out from the data set. In this stage, we also considered a manual judgment and marked the questionable data. In this study, we further compared the high-quality surface solar radiation data (applied abovementioned QC procedure) with the data retrieved from Himawari-8 satellite in 2020. The results show that the overall correlation coefficient between the two data sets is extremely high, up to 0.96. The correlation coefficient among each month ranges from 0.94 to 0.97. Overall, the satellite retrieved data shows overestimation. A largest overestimation was found for Hualien region (27.9%), while for the northern region tends to underestimation (4.6%) as compared with observation. For monthly comparison, the month of March revealed a larger overestimation of 6.5%. This study has developed a strict QC process for surface solar radiation data and has reviewed the data reliability for the past 19 years. The high-quality data produced from this study is beneficial to the development of subsequent scientific research.