粒徑分布融合技術與反演化學組成

Abstract

微粒粒徑分佈（PSD）對於理解氣膠對環境以及人體健康的影響非常重要。有鑑於此，本研究在臺灣臺中地區架設IMPACT移動監測站以掃描電移動度分徑儀（SMPS）和氣動粒徑分析儀（APS）測量奈米至微米大小氣膠的PSD。然而全面理解氣膠各項性質涉及完整且廣泛的微粒粒徑分佈，故本研究建立融合技術將不同測量原理所量測的微粒粒徑分佈結合。此外，雖然氣膠的分徑化學組成也非常重要，但分徑成分的測量方法多為繁瑣且具有挑戰性。即使不進行分徑，而只測量整體的氣膠化學成分也仍有一定的限制。因此，本研究提出並測試數種微粒粒徑分佈融合方法後，以長時間觀測數據推定分徑氣膠化學組成的狀態，同時提出改善以光學量測反演氣膠化學成分的演算法。 本研究對於融合微粒粒徑分佈的數值方法進行相關討論並探討各方法的限制。在考慮到兩種不同的融合技術相關數值方法以及修正APS計數效率的影響後，最終有四種不同的融合結果。此外，將整合質量濃度量測結果至資料融合過程中，並使用適應函數進行評估，開發出優化版本的流程，且該版本也產出四種結果。融合結果顯示，在考慮到數量、表面積和體積濃度並在優化流程中套用APS修正後的演算法有最佳的表現。 本研究整合微粒粒徑分佈並經由化學物種分類所定義的物種主導時期，可以獲取分徑化學成分的概況。雖然結果顯示各化學成分的粒徑分佈與過去觀測研究類似，但此方法仍無法提供高時間解析度的分徑化學資訊，且也無法進行正確的量化分析，僅能提供定性描述。儘管有以上的限制，但其仍然具有反演分徑化學成分的潛力，且可避免複雜且繁瑣的分徑採樣過程。 此外，為反演高時間解析度之氣膠化學成分，本研究遂犧牲粒徑解析，另以光學量測訊號搭配融合後之粒徑分布進行反演。此方法涉及多種技術的結合：以測量數據模擬氣膠折射率、以融合技術計算有效密度、以體積平均混合（VAM）模型製作參數查找表以及建構不同目的之參數組合。當假設各成分質量濃度總和相等於所測得PM2.5質量濃度時，使用不同最佳化條件則得出四種不同結果，而另外四種結果則是另假設各成分質量濃度總和低於所測得PM2.5質量濃度。雖然所有反演結果對於黑碳（BC）質量濃度的估算都十分準確，但其他氣膠化學組成則不然。導致這些差異的因素可能源於1.光學折射率與微粒密度的不確定性；2.計算效能有限；3.模擬與實際折射率的差異；4.有效密度與物質密度不一致的趨勢。未來的研究應著重於改善各化學成分參數的估算、增加計算資源並以更廣泛的實驗數據驗證，或可進一步優化此演算法。

The importance of Particle Size Distribution (PSD) is emphasized due to its significant role in understanding aerosol impacts on the environment and health. The IMPACT mobile monitoring station was established to measure the submicron and supermicron PSDs using a scanning mobility particle sizer (SMPS) and an aerodynamic particle sizer (APS), respectively. Merging techniques of PSDs measured by different aerosol sizers based on different principles are crucial for comprehending aerosol properties. On the other hand, while chemical composition in aerosols is important, size-resolved measurement is challenging and cumbersome. Even without considering size-resolved measurements, there are still inherent limitations. This study proposes and tests several PSD data merging schemes, captures the general profile of size-resolved chemical composition utilizing the long-term measurement data, and refines method for retrieving chemical composition. The research compares numerical methods for the merging process, discussing the limitations of various approaches. Four distinct merged results were obtained by employing two different numerical algorithms and considering the correction of low counting efficiency for the overlapping sizing range by SMPS and APS. Additionally, mass concentrations were integrated into this merging process using a fitness function, leading to an enhanced procedure and the production of another four refined results. The merging results demonstrated that the algorithm considering number, surface, and volume concentration with an APS correction curve in the enhanced workflow exhibited the best performance. The profile of size-resolved chemical composition is created by integrating PSD data with the dominant periods, which are determined by classifying chemical species. While it roughly describes the PSD of various species and aligns with previous measurement research, it still fails to provide high temporal resolution data and cannot quantify the results. Despite these limitations, the proposed approach demonstrates the potential for retrieving size-resolved chemical composition and avoids the complex and cumbersome process of size-resolved sampling. Additionally, to retrieve high temporal resolution aerosol chemical compositions, this study sacrifices size resolution, instead using optical measurement data combined with merged PSD for retrieval. The method involves the combination of techniques: simulating aerosol refractive index with measured data, calculating effective density through a merging process, creating look-up table with the Volume Average Mixing (VAM) model, and constructing parameter combinations for various purposes. Under the assumption that the sum of mass concentration equals the PM2.5 mass concentration, four results were obtained using different optimal conditions. Another four results were derived under a contrasting condition where the sum of mass concentration was lower than the PM2.5 mass concentration. While all results provided accurate estimations for Black Carbon (BC) mass concentration, the results of other aerosol composition were less convincing. Several factors may contribute to these discrepancies, including inaccuracies in species parameters (refractive index and density), limited computing power, discrepancies between simulated and actual refractive index values, or unexpected trends between effective density and material density. Future research should focus on addressing these challenges by improving parameter estimations of each chemical composition, increasing computational resources, and further optimizing this algorithm.