| 摘要: | 全天空輻射量(global horizontal irradiance, GHI)為光電能源業者用來監測太陽能系統發電效率的主要參考依據，然而GHI容易會受到天氣條件的影響而有明顯的變化，例如：太陽位置、雲和氣膠…等。目前大部分GHI產品的時間解析度為一小時，這樣的時間解析度不足以反應出GHI短時間內快速變化的性質。在本研究中，我們使用向日葵8號衛星(Himawari-8, H8)十分鐘解析度的資料以及應用單層輻射傳送模式反演高時間解析度的GHI資料。本研究的模式中考慮了氣膠、臭氧、水氣、氣體及雲對輻射量的吸收及散射效應，並利用了地面觀測GHI資料及衛星資料透過經驗公式來建立太陽輻射波段及衛星波段之間的關係。同時也利用晴空判斷法及晴空指數將天空條件分為晴空條件、部分晴空條件及多雲條件，並以此改進模式結果之精確度。
本研究選擇中央氣象局嘉義氣象站2018年1月1日至2018年4月12日之資料作為研究站點及研究時間，利用1月1日至2月28日建立太陽輻射波段及衛星波段之間的線性關係，而後反演3月1日至4月12日之地面輻射量，最後與氣象局地面氣候自動觀測系統(Automatic Climate Observation System, ACOS)之地面輻射量觀測資料進行誤差分析並探討模式之反演表現。結果顯示，晴空下的rMBE(相對平均偏差)、rRMSE(相對均方根誤差)及r2 (判定係數)為 -1.7%、6.8%及0.98，非晴空下的rMBE、rRMSE及r2為 -5.9%、22.7%及0.84，模式在晴空下有極佳的表現。在反演策略中，吾人採用將天空條件分為晴空、部分晴空及多雲分別進行模式反演，整體表現的rMBE、rRMSE及r2為 -5.1%、20.5%及0.86，相較於沒有使用天空條件判斷下的結果(rMBE、rRMSE及r2為 -7.9%、22.7%及0.84)，有較佳的表現。透過不同天空條件下的誤差分析，吾人發現衛星反演地面輻射在計算中最大的誤差由雲所造成，可使反演之輻射量的r2值由0.98下降至0.47。
本研究反演之GHI的時間解析度為十分鐘，相較氣象局所提供的一小時時間解析度GHI反演資料，十分鐘時間解析度較能夠模擬出短時間內雲經過時地面輻射量的變化。為了能夠與氣象局衛星反演產品比對，進一步將ACOS地面觀測GHI與本研究反演之GHI的時間解析度(分別為一分鐘及十分鐘)皆平均至一小時，與平均後的ACOS地面觀測進行誤差分析後，結果顯示本研究反演之GHI平均後的rMBE、rRMSE及r2為 -6.2%、14.3%及0.92，而氣象局所衛星反演之GHI產品的rMBE、rRMSE及r2為-1.5%、11.2%及0.94，可以發現模式結果能與氣象局衛星產品相近。進一步評估本反演法的時空適用性，吾人發現距離建立線性關係式的氣象觀測站越遠，模式所反演之輻射量的誤差也會隨之增加，因此若能定義出台灣各個氣象觀測站代表的影響範圍，便能針對不同區域建立不同的線性回歸式以完成更大範圍的地面輻射量反演。
;Surface solar radiation (or global horizontal irradiance, GHI) data is critical for photovoltaic and electric power companies to monitor power generation efficiency of photovoltaic system. However, GHI highly varied with sky conditions which mainly due to solar position and clouds variations. The typically temporal resolution of GHI product is 1-hour resolution, which might not be enough to represent the GHI variability. In this study, we used Himawari-8 (H8) satellite 10-min resolution data and applied a one-layer radiation transfer model to derived high temporal resolution GHI data. Our model includes the calculations for the scattering and absorption from aerosol, ozone, water vapor, gases and clouds. The observational GHI data has been used to construct relationships between solar spectrum and satellite bands, and further create an empirical function. Furthermore, a clear-sky identification method with sky index (i.e., clear sky, partly clear sky and cloudy sky) is proposed to evaluate the model derived-GHI performance in different sky conditions.
Data from the CWB Chiayi station was used to evaluate the performance of our model deriving results. Statistical results show that the rMBE, rRMSE, and r2 are -1.7% vs -5.9%, 6.8% vs 22.7% and 0.98 vs 0.84 for under clear sky and unclear sky, respectively. The performance of model under clear sky is outstanding, and the performance of model under unclear sky is moderately, and according to the analysis of the results under different sky conditions, we found that the largest error in the model is caused by the clouds, which could reduce the r2 value of derived GHI from 0.98 to 0.47.
We also found that when we consider sky-index-dependent empirical functions using in the model, the r2 value improved from 0.84 to 0.86. Compared to the CWB 1-hr resolution GHI product (rMBE: -1.5%, rRMSE: 11.2% and r2: 0.94), our high-resolution (10-min computation and average to 1-hr for comparison) show the rMBE, rRMSE and r2 value are -6.2%, 14.3% and 0.92, respectively. In order to evaluate the spatial and temporal applicability of our model, we found that the farther away from the station which used to establish the linear relationships, the error of the derived GHI would increase. With the basement of our model, if we could define the effective radius of weather stations in Taiwan, we can establish different linear regression equations at different places, so that GHI can be derived in more region. |
