| 摘要: | 本研究使用2011-2021年七至九月之中央氣象局全臺地面氣象站觀測資料，研究弱綜觀條件下之午後對流肇始(convection initiation，CI)事件環境特徵，並建立模糊邏輯法午後對流肇始事件(cluster of convection initiation，CCI)之預報。一般而言，對流肇始(CI)表示單一對流被雷達偵測到的時空間資訊，但臺灣地區的CI常以成群的方式在相似的時空間發生，故本研究將時空間相近之CI是為群體事件，即對流肇始事件(CCI)。
研究中使用Storm Motion Analysis by Radar Tracking (SMART)系統找出午後對流肇始資訊，以群體事件之角度審視當日午後對流之肇始，若有群體的午後對流發生，則該日視為午後對流肇始事件(CCI)，反之為空事件(Null)。統計結果顯示CCI事件發生時間落在午後時段(13-17 LST)，並以14-16 LST為主要發生時段，並以中臺灣有最高之發生頻率。地面環境分析顯示水氣相關的變數與東西風分量(U風)有較明顯的差異。以北部區域為例，午後對流肇始事件中，水氣較多且隨海風向內陸傳播，使相當位溫較高。U風則反映海風特徵，不論在CCI事件或Null事件中，皆有明顯的邊界存在。該邊界在CCI事件中較為北邊，Null事件則較為南邊，顯示海風在Null事件被侷限南邊的區域。探空資料則顯示CCI事件中，中低層風向為西南風與較高的水氣混合比(簡稱混合比)約0.5-1.5 g/kg與相當位溫約1.5-3.4 K，顯示環境較為潮濕並具有較多能量，有利於對流發展。
將模糊邏輯法與全臺地面氣象站及地面分析所看到之特徵差異結合，計算機率密度函數(probability of density function，PDF)與標準化機率密度函數(normalized probability of density function，NPDF)，並以PDF重疊面積之倒數為權重進行午後對流肇始事件的預報。結果具有較高的臨界成功指數(critical success index，CSI)約0.63-0.83、偵測率(probability of detection，POD)約0.73-0.96與公正預兆得分(equitable threat score，ETS)約0.16-0.53，以及較低的誤報率(false alarm ratio，FAR)約0.14-0.25與接近1的偏倚得分(bias score，BS)，顯示透過模糊邏輯法簡單地結合大量地面觀測資料進行預報有其實用價值與發展潛力。
取用頻率較高的測站與變數顯示中臺灣與北臺灣之水氣相關變數為模糊邏輯法判斷是否CCI事件的主要依據，代表中臺灣與北臺灣之水氣相關變數對於CCI事件有重大的影響力。而其他變數則扮演輔助的角色，其中又以與風有關之變數如U風分量、V風分量、風速與風向等較為重要。
;A fuzzy logic algorithm for convection initiation (CI) forecast utilizing 11-year surface observation data in July, August, and September in Taiwan is developed and investigated. The location and time of 10-year (2011-2020) weak-synoptic CI and non-CI events were identified by the tracking results of convective thunderstorms by Storm Motion Analysis by Radar Tracking (SMART). The surface characteristics from 191-232 automatic weather stations, namely temperature, relative humidity, water vapor mixing ratio (mixing ratio), equivalent potential temperature (θ_e), U-wind, V-wind, wind speed and wind direction, for CI and non-CI events were investigated.
The statistical results show that the CI event, under the weak-synoptic condition, mainly occurred during the afternoon hours (13-17 LST) with a peak occurrence between 14-16 LST with the highest occurrence frequency in the central Taiwan regions in Chiayi, Yunlin, and Changhua. The surface environment analysis revealed significant differences in moisture-related variables and the zonal wind component (U-wind). In north Taiwan, CI events have higher moisture propagated inland with the sea breeze, resulting in higher equivalent potential temperature (θ_e). Both CI and non-CI events have evident boundaries of the U-wind, indicating the confinement of sea breeze to the southern region during null events. Sounding data show that during CI events, the lower to mid-level winds were southwest winds, the water vapor mixing ratio was ~0.5-1.5 g/kg higher and the equivalent potential temperature was ~1.5-3.4 K higher. These indicate a more humid environment with higher energy that favors convection initiations.
The probability density functions (PDFs) of all variables were calculated for CI and non-CI events. The normalized PDFs of each variable are derived as membership functions. The weighting of each membership function is determined by the reciprocal of the overlapping area of PDF. The self-check results from the fuzzy logic algorithm show high value of probability of detection (POD), success ratio (SR), and critical success ratio (CSI), as well as low probability of false detection (POFD). An independent year (2021) of CI and non-CI events was applied to examine the applicability of the fuzzy logic algorithm. The results indicate that the scores gradually increase as the time approaching to afternoon (CSI: 0.60~0.70, POD: 0.76~0.84, FAR: 0.25~0.18). The results suggest the potential of utilizing the fuzzy logic algorithm for the CI forecast.
Additionally, the usage frequency of each station of each variable demonstrated that water vapor the mixing ratio and the θ_e in central and northern Taiwan were the primary variables for the fuzzy logic algorithm to determine CI events. This suggests that moisture-related variables in central and northern Taiwan have a significant influence on CI events. Other variables played auxiliary roles, with wind-related variables such as the U-wind component, V-wind component, wind speed, and wind direction. |
