高解析度光學遙測衛星廣泛應用於環境監測、植生分析及大氣反演等定量遙測領域，其輻射校正精度直接影響後續產品之可靠度。然而，此類衛星多採離天底（off-nadir）觀測且星載校正能力有限，現行 RadCalNet 提供之天底等效地表反射率若直接應用於特定目標校正，易因地表雙向反射分布函數（Bidirectional Reflectance Distribution Function， BRDF）效應造成系統性偏差。本研究以 FORMOSAT-5（FS-5）及 FORMOSAT-8A（FS-8A）為研究對象，建立考量 BRDF 效應之高解析度光學衛星輻射校正架構。首先，以 Railroad Valley Playa 現地多角度量測建立 BRDF 查找表，修正 RadCalNet 地表反射率並完成 FS-5 特定目標校正；其次，評估校正誤差對大氣頂反射率（TOA Reflectance）、植生近紅外反射率（NIRv）及氣膠光學厚度（AOD）反演之影響；進一步比較三種 FS-8A 全色態地表反射率重建方法，建立適用於交互校正之策略；最後整合全球 BRDF 產品建構全球增益圖，提出最佳增益選擇建議。結果顯示，BRDF 校正可將 FS-5 五個波段校正係數 K₀ 中位相對誤差由約13%–17%降低至1%–4%；光譜內插卷積法之地表反射率重建誤差最低，中位絕對誤差約0.24%；建立之全球增益圖則建議單一增益 G = 2.4 可涵蓋約98%之全球（含海洋）觀測情境。本研究建立兼具物理基礎、可追溯性與作業可行性之 BRDF-aware 輻射校正架構，可作為未來高解析度光學衛星校正設計與定量遙測應用之重要參考。;High-resolution optical Earth-observation satellites play an important role in quantitative remote sensing applications， including environmental monitoring， vegetation assessment， and atmospheric retrieval. However， these satellites commonly operate under off-nadir viewing geometries and often have limited onboard radiometric calibration capability. Direct use of the nadir-equivalent surface reflectance products provided by the Radiometric Calibration Network (RadCalNet) may therefore introduce systematic calibration bias due to the bidirectional reflectance distribution function (BRDF) of the Earth′s surface. This dissertation develops a BRDF-aware radiometric calibration framework using FORMOSAT-5 (FS-5) and FORMOSAT-8A (FS-8A) as representative case studies. An empirical BRDF lookup table was first constructed from in situ multi-angular measurements at Railroad Valley Playa to correct RadCalNet surface reflectance for the actual viewing geometry and improve FS-5 vicarious calibration. The influence of calibration accuracy on top-of-atmosphere (TOA) reflectance， near-infrared reflectance of vegetation (NIRv)， and aerosol optical depth (AOD) retrievals was subsequently evaluated. Three broadband panchromatic surface reflectance reconstruction approaches for FS-8A cross-calibration were then compared， followed by the development of a global gain map through integration of global BRDF products， spectral response functions， and saturation analysis. The proposed BRDF correction reduced the median relative error of the absolute calibration coefficient K₀ from approximately 13–17% to 1–4% across the five FS-5 spectral bands. Among the three reconstruction approaches， the spectral interpolation-convolution method achieved the lowest median absolute surface reflectance error of approximately 0.24%. Furthermore， the proposed global gain map suggests that a single gain setting of G = 2.4 can accommodate approximately 98% of global imaging scenarios. Overall， this study establishes a physically consistent and operationally feasible BRDF-aware radiometric calibration framework that integrates BRDF correction， vicarious calibration， cross-calibration， and gain optimization， providing a practical reference for future high-resolution optical satellite calibration and quantitative remote sensing applications.
