研究與教學
國際交流
榮耀分享
探索理院
活動報導
人事動態
第67期出刊日:2026.10.01

Can a sounding predict afternoon thunderstorms? A TaiwanVVM-based framework for uncovering the role of boundary layer dynamics

探空能否預測午後雷雨?——以 TaiwanVVM 為基礎的框架揭示邊界層動力的關鍵角色


Department of Atmospheric Science
Prof. Chien-Ming Wu
大氣科學系 吳健銘 教授
 

A sounding is typically used to diagnose the thermodynamic properties of the atmosphere. A classical parcel model is generally used to assess the potential for convective development (i.e., through CAPE/CIN diagnoses). However, it is well known that under the diurnal cycle over land, the development of shallow convection within the boundary layer can strongly influence the onset of deep convection (Wu et al. 2009). In addition, the horizontal heterogeneity of surface sensible and latent heat fluxes, shaped by previous convective events, can significantly modify both the mean precipitation and the timing of peak rainfall during the diurnal cycle (Wu et al. 2015). These findings highlight the critical role of boundary layer processes in predicting afternoon thunderstorms. Over a mountainous island such as Taiwan, this environmental information, combined with strong orographic forcing and locally driven circulations, provides an even stronger constraint on the formation of precipitation hotspots associated with orographically locked convection. This provides an opportunity to predict afternoon thunderstorms in numerical simulations using only soundings that represent the upstream large-scale environment.

In TaiwanVVM, orographic effects are embedded in the vorticity equation, resulting in a non-local flow structure that is essential for capturing these precipitation hotspots. The flow-dependent orographic effects (Wu and Arakawa 2011; Wu et al. 2019) enable the simulation of local circulations generated by buoyancy gradients while minimizing contamination by numerical diffusion. We demonstrate that the formation and intensity of these hotspots strongly depend on interactions among the physical processes governing local energetics and cloud dynamics. Accurately estimating future changes in these hotspots therefore requires a model with sufficiently high spatial resolution as well as an appropriate representation of the key physical processes.

Ensembles of TaiwanVVM large-eddy simulations (Δx = 500 m) are designed to capture summertime diurnal convection over Taiwan under conditions dominated by local circulation (Chen et al. 2024). Precipitation hotspots identified from long-term observations are well reproduced by the present-day ensemble simulations when realistic environmental variability is included. A pseudo-global warming experiment is then conducted to examine changes in convective structures and the resulting local rainfall responses. Under a uniform warming scenario of 3 K with conserved relative humidity, the thermodynamic environment exhibits increased convective available potential energy and a slight reduction in convective inhibition (CIN), primarily due to enhanced low-level water vapor in the marine boundary layer. These high-resolution simulations, conducted under a well-defined weather regime, provide critical information for assessing future changes in extreme rainfall associated with orographically locked diurnal convection, with implications for natural hazards and water resources.

Furthermore, the robustness of the TaiwanVVM ensemble dataset can be leveraged to train an AI model that can substantially accelerate prediction while preserving physical explainability through a variational autoencoder (VAE; e.g., Hsieh and Wu 2024). By compressing the TaiwanVVM results into a small number of latent dimensions, we can project and interpret key physical processes within this reduced space. With the aid of a large number of TaiwanVVM ensemble simulations, we can develop an AI-TaiwanVVM framework that generates local flow and precipitation patterns using only upstream soundings as input.

 

Wu, C. M., Stevens, B., & Arakawa, A. (2009). What controls the transition from shallow to deep con-vection? Journal of the Atmospheric Sciences, 66(6), 1793-1806.

Wu, C. M.*, & Arakawa, A. (2011). Inclusion of surface topography into the vector vorticity equation model (VVM). Journal of Advances in Modeling Earth Systems, 3(2).

Wu, C. M.*, Lo, M. H., Chen, W. T., & Lu, C. T. (2015). The impacts of heterogeneous land surface fluxes on the diurnal cycle precipitation: A framework for improving the GCM representation of land‐atmosphere interactions. Journal of Geophysical Research: Atmospheres, 120(9), 3714-3727.

Wu, C. M.*, Lin, H. C., Cheng, F. Y., & Chien, M. H. (2019). Implementation of the land surface pro-cesses into a vector vorticity equation model (VVM) to study its impact on afternoon thunderstorms over complex topography in Taiwan. Asia-Pacific Journal of Atmospheric Sciences, 55(4), 701-717.

Chen, W.-T., Y.-H. Chang, C.-M. Wu* and H.-Y. Huang (2024) The future extreme precipitation systems of orographically locked diurnal convection: the benefits of using large-eddy simulation ensembles. Environmental Research: Climate

Hsieh, M.-K., and C.-M. Wu* (2024) Developing an Explainable Variational Autoencoder (VAE) Framework for Accurate Representation of Local Circulation in Taiwan. JGR Atmosphere
 

探空資料通常用於診斷大氣的熱力學性質。經典的氣塊模型一般用來評估對流發展的潛勢(即透過 CAPE/CIN 的診斷)。然而,眾所周知,在陸地的日變化循環下,邊界層內淺對流的發展能夠顯著影響深對流的觸發(Wu et al. 2009)。此外,由先前對流事件所形塑的地表感熱與潛熱通量之水平異質性,也能在日變化循環中顯著改變平均降水量以及尖峰降雨出現的時間(Wu et al. 2015)。這些研究結果突顯了邊界層過程在預測午後雷雨中的關鍵角色。對於像臺灣這樣的多山島嶼而言,這些環境資訊若再結合強烈的地形強迫與局地驅動的環流,將對與地形鎖定對流相關的降水熱點形成提供更強的約束。這也為在數值模擬中僅使用代表上游大尺度環境的探空資料來預測午後雷雨提供了可能性。
在 TaiwanVVM 中,地形效應被嵌入渦度方程式中,從而產生一種非局域的流場結構,而這對於捕捉這些降水熱點至關重要。流場依賴的地形效應(Wu and Arakawa 2011;Wu et al. 2019)使模式能夠模擬由浮力梯度所驅動的局地環流,同時將數值誤差的影響降至最低。我們證明,這些降水熱點的形成與強度高度依賴於支配局地能量學與雲動力學之物理過程間的交互作用。因此,若要準確估計這些熱點在未來的變化,必須使用具有足夠高空間解析度且能適當表現關鍵物理過程的模式。
我們設計了一系列 TaiwanVVM 大渦模擬(Δx = 500 m)的集合,以捕捉在局地環流主導條件下,臺灣夏季的日變化對流(Chen et al. 2024)。長期觀測所識別的降水熱點,在納入真實環境變異性後,能被現今氣候條件下的集合模擬良好再現。隨後,我們進行一項假想全球暖化(pseudo-global warming)實驗,以檢視對流結構的變化及其對局地降雨的影響。在相對濕度維持不變、均勻升溫 3 K 的情境下,熱力環境呈現出對流可用位能(CAPE)的增加與對流抑制(CIN)的輕微減弱,主要源於海洋邊界層低層水汽的增加。這些在明確氣象型態下進行的高解析度模擬,為評估與地形鎖定日變化對流相關的極端降雨未來變化提供了關鍵資訊,並對自然災害與水資源管理具有重要意涵。

此外,TaiwanVVM 集合資料的穩健性可用於訓練一個 AI 模型,透過變分自編碼器(VAE;例如 Hsieh and Wu 2024)在大幅加速預測的同時保留物理可解釋性。透過將 TaiwanVVM 的結果壓縮至少數潛在維度,我們可以在此降維空間中投射並詮釋關鍵物理過程。藉由大量 TaiwanVVM 集合模擬的協助,我們可發展出 AI-TaiwanVVM 框架,僅以上游探空資料作為輸入,即可生成局地流場與降水型態。