A validation study is conducted regarding aerosol optical size property retrievals from measurements of the direct sun beam only (without the aid of diffuse radiation). The study focuses on using real data to test the new GRASP-AOD application, which uses only spectral optical depth measurements to retrieve the total column aerosol size distributions, assumed to be bimodal lognormal. In addition, a set of secondary integral parameters of aerosol size distribution and optical properties are provided: effective radius, total volume concentration and fine-mode fraction of aerosol optical depth (AOD). The GRASP-AOD code is applied to almost 3 million observations acquired over 20 years (1997–2016) at 30 AERONET (Aerosol Robotic Network) sites. These validation sites have been selected based on known availability of an extensive data record, significant aerosol load variability throughout the year, wide worldwide coverage and diverse aerosol types and source regions. The output parameters are compared to those coming from the operational AERONET retrievals. The retrieved fine-mode fractions at 500 nm (τ_f(500)) obtained by the GRASP-AOD application are compared to those retrieved by the spectral deconvolution algorithm (SDA) and by the AERONET aerosol retrieval algorithm. The size distribution properties obtained by the GRASP-AOD are compared to their equivalent values from the AERONET aerosol retrieval algorithm. The analysis showed the convincing capacity of the GRASP-AOD approach to successfully discriminate between fine- and coarse-mode extinction to robustly retrieve τ_f(500). The comparisons of 2 million results of τ_f(500) retrieval by the GRASP-AOD and SDA showed high correlation with a root mean square error (RMSE) of 0.015. Also, the analysis showed that the τ_f(500) values computed by the AERONET aerosol retrieval algorithm agree slightly better with the GRASP-AOD (RMSE = 0.018, from 148 526 comparisons) than with the SDA (RMSE = 0.022, from 127 203 comparisons). The comparisons of the size distribution retrieval showed agreement for the fine-mode median radius between the GRASP-AOD and AERONET aerosol retrieval algorithm results with an RMSE of 0.032 µm (or 18.7 % in relative terms) for the situations when τ(440)>0.2 occur for more than 80 000 pairs of the study. For the cases where the fine mode is dominant (i.e., α>1.2), the RMSE is only of 0.023 µm (or 13.9 % in relative terms). Major limitations in the retrieval were found for the characterization of the coarse-mode details. For example, the analysis revealed that the GRASP-AOD retrieval is not sensitive to the small variations of the coarse-mode volume median radius for different aerosol types observed at different locations. Nonetheless the GRASP-AOD retrieval provides reasonable agreement with the AERONET aerosol retrieval algorithm for overall coarse-mode properties with with RMSE = 0.500 µm (RMSRE = 20 %) when τ(440)>0.2. The values of effective radius and total volume concentration computed from the GRASP-AOD retrieval have been compared to those estimated by the AERONET aerosol retrieval algorithm. The RMSE values of the correlations were 30 % for the effective radius and 25 % for the total volume concentration when τ(440)>0.2. Finally, the study discusses the importance of employing the assumption of bimodal lognormal size distribution. It also evaluates the potential of using ancillary data, in particular aureole measurements, for improving the characterization of the aerosol coarse-mode properties.

中文翻译：

---

通过光谱光学深度测量表征气溶胶尺寸特性：使用长期 AERONET 数据对 GRASP-AOD 代码进行全局验证

对仅从直接太阳光束的测量（没有漫射辐射的帮助）中检索气溶胶光学尺寸特性进行了验证研究。该研究的重点是使用真实数据来测试新的 GRASP-AOD 应用程序，该应用程序仅使用光谱光学深度测量来检索总柱状气溶胶粒径分布，假定为双峰对数正态分布。此外，还提供了一组气溶胶尺寸分布和光学特性的二次积分参数：有效半径、总体积浓度和气溶胶光学深度 (AOD) 的细模分数。GRASP-AOD 代码应用于 20 年（1997-2016 年）在 30 个 AERONET（气溶胶机器人网络）站点获得的近 300 万次观测。这些验证地点是根据广泛数据记录的已知可用性选择的，全年显着的气溶胶负荷变化、广泛的全球覆盖范围以及不同的气溶胶类型和来源区域。输出参数与来自操作 AERONET 检索的参数进行比较。在 500 nm (将通过 GRASP-AOD 应用程序获得的τ _f (500) ) 与通过光谱解卷积算法 (SDA) 和 AERONET 气溶胶检索算法检索到的值进行比较。将 GRASP-AOD 获得的粒径分布特性与来自 AERONET 气溶胶反演算法的等效值进行比较。分析表明 GRASP-AOD 方法具有令人信服的能力，可以成功区分细模式和粗模式消光以稳健地检索τ _f (500)。GRASP-AOD 和 SDA对τ _f (500)检索的 200 万个结果的比较表明，均方根误差 (RMSE) 为0.015具有高度相关性。此外，分析表明，由 AERONET 气溶胶检索算法计算的τ _f (500)值与 GRASP-AOD（RMSE =  0.018，来自 148 526 次比较）的一致性略好 于 SDA（RMSE  =  0.022，来自 127 203 次比较）。尺寸分布检索比较表明对于GRASP-AOD和AERONET气溶胶检索算法的结果之间具有0.032的RMSE精细模式中值半径协议 μ为情况米（或相对而言18.7％）时τ（440） >0.2出现在超过 80 000 对研究中。对于精细模式占主导地位的情况（即α >1.2），RMSE 仅为 0.023  µm（或相对而言的 13.9%）。对于粗模式细节的表征，发现了检索中的主要限制。例如，分析表明 GRASP-AOD 反演对在不同位置观察到的不同气溶胶类型的粗模式体积中值半径的微小变化不敏感。尽管如此，GRASP-AOD检索提供与AERONET气溶胶检索算法整体粗模式属性合理协议与RMSE  =  0.500  μ M（RMSRE  =  20％）时τ（440）> 0.2. 将 GRASP-AOD 反演计算的有效半径和总体积浓度值与 AERONET 气溶胶反演算法估计的值进行了比较。当τ (440)>0.2时，相关性的 RMSE 值为有效半径的 30% 和总体积浓度的 25% 。最后，该研究讨论了采用双峰对数正态分布假设的重要性。它还评估了使用辅助数据的潜力，特别是光环 测量，用于改进气溶胶粗模式特性的表征。