Shih-Hao Su

This study investigates the climatological characteristics of remote rainfall in northeastern Taiwan during the cold half‐year of 1980–2023. We identify critical factors driving heavy precipitation when tropical cyclones (TCs) are located south of Taiwan. Spatial analysis reveals distinct rainfall ‘hotspots’ that shift southward from Yilan depending on the TC's specific position. Statistical results indicate that heavy rainfall is primarily governed by extrinsic environmental thresholds, specifically moisture convergence and specific humidity. Validation using 2024–2025 cases confirm the robustness of these thresholds. We conclude that the interaction between background northeasterly flow and TC positioning acts as the dominant control. This relationship suggests that future climate‐induced changes in TC tracks may displace established rainfall hotspots, potentially altering regional hydroclimatic risks.

The climate characteristics of remote rainfall events in Taiwan from September to February over 41 years (1980–2020) are studied. These events are induced by the interaction between the northeasterly flow and the typhoon's outer circulation. Our findings reveal that rainfall in northeastern Taiwan becomes more prominent when tropical cyclones move to the remote rainfall‐prone area, located in the north Philippine area to the northern South China Sea, and when the background northeasterly wind speed exceeds 7 m s⁻¹. Under these criteria, the confluence of the typhoon's outer circulation and the northeasterly flow creates a convergence area that enhances rainfall in northeastern Taiwan, increasing the occurrence of moderate to extreme rainfall (ER) events. This leads to an average enhancement in rainfall amount of 80–220 mm per day. Additionally, when typhoons are in the remote rainfall‐prone area, there is a greater than 20% chance for events with maximum rainfall over 200 mm day⁻¹ to occur, particularly in the region of 20°–22°N, 116°−124°E, and north to Luzon Island. In this area, the occurrence rate can exceed more than a 45% chance. The highest risk of ER events occurs between 20°–22°N and 118°−120°E, with a probability of over 90%. Notably, the convergence area for the Taiwan cases does not necessarily coincide with the baroclinic forcing as that associated with remote rainfall events observed in Japan, Korea, and North America.

This study applied representation learning algorithms to satellite images and evaluated the learned latent spaces with classifications of various weather events. The algorithms investigated include the classical linear transformation, i.e., principal component analysis (PCA), state-of-the-art deep learning method, i.e., convolutional autoencoder (CAE), and a residual network pre-trained with large image datasets (PT). The experiment results indicated that the latent space learned by CAE consistently showed higher threat scores for all classification tasks. The classifications with PCA yielded high hit rates but also high false-alarm rates. In addition, the PT performed exceptionally well at recognizing tropical cyclones but was inferior in other tasks. Further experiments suggested that representations learned from higher-resolution datasets are superior in all classification tasks for deep-learning algorithms, i.e., CAE and PT. We also found that smaller latent space sizes had little impact on the classification task’s hit rate. Still, a latent space dimension smaller than 128 caused a significantly higher false-alarm rate. Though the CAE can learn latent spaces effectively and efficiently, the interpretation of the learned representation lacks direct connections to physical attributions. Therefore, developing a physics-informed version of CAE can be a promising outlook for the current work.

This study explains why observed storms with long concentric eyewall (CE) duration often have a large moat size. A series of slab boundary layer model experiments are conducted with the free atmosphere forcing from different CE structures. The experimental results indicate that the inner eyewall pumping (IEP), the maximum vertical motion at the inner eyewall region, increases with the increase in moat size d and vortex maximum wind v_m, the decrease in the inner eyewall radius r_m, and increase in inner eyewall vorticity. A scaling law is derived from hundreds of SBL experiments with four variables: IEP w_in, vortex inner radius r_m, vortex maximum wind v_m, and moat size d. The dimensionless scaling law is where w^∗ = w_in/v_m, d^∗ = ζd/c, ζ is the vorticity 2v_m/r_m, and c is a constant gravity wave speed. The moat size divided by the Rossby length c/ζ, the dimensionless moat size d^∗, combines the effect of the moat size and the vortex pressure gradient force to accelerate the inflow and to enlarge the IEP for inner eyewall convection maintenance. A phase diagram of IEP with vortex structure dimensionless moat ζd/c and vortex intensity v_m is constructed based on the scaling law for the aircraft and satellite observations. The eyewall replacement cycle may be viewed as the process to reduce the IEP from the decrease of the vortex intensity and the vortex size of the dimensionless moat. This leads to the ultimate disappearance of the inner eyewall.

This study aims to propose a strategy to optimize the performance of the Support Vector Machine (SVM) scheme for extreme Meiyu rainfall prediction over southern Taiwan. Variables derived from Climate Forecast System Reanalysis (CFSR) dataset are the candidates for predictor selection. A series of experiments with different combinations of predictors and domains are designed to obtain the optimal strategy for constructing the SVM scheme. The results reveal that the accuracy (ACC), positive predictive values (PPV), probability of detection (POD), and F1-score can exceed 0.6 on average. Choosing the predictors associated with the Meiyu system and determine the domain associated with the correlations between selected predictors and predictand can improve the forecast performance. Our strategy shows the potential to predict extreme Meiyu rainfall in southern Taiwan with lead times from 16 h to 64 h. The F1-score analysis further demonstrates that the forecast performance of our scheme is stable, with slight inter-annual fluctuations from 1990 to 2019. Higher performance would be expected when the north of the South China Sea is characterized by stronger southwesterly flow and abundant low-level moisture for a given year.

Typhoon Saola affected Taiwan from July 30, 2012, to August 03, 2012. It started in a northwestward track, and it was weakening to Category 1 storm and had a sharp cyclonic deflection track when it was about to made landfall in Taiwan on August 02. After the deflection, the typhoon moved northward and then northwestward across northern Taiwan. The cyclonic southward deflection and the northward motion constituted 12 h of looping motion which increase both the typhoon duration time and the rainfall in northeastern Taiwan. This research is to study the looping motion by the Weather Research and Forecasting model for this very weak typhoon. The model is able to reproduce the looping motion similar to that of observation. The model shows a strong asymmetry of precipitation and cross‐mountain downslope wind near the typhoon center while the typhoon track is looping toward the southwest. The potential vorticity (PV) tendency diagnosis confirms that diabatic heating and downslope wind vertical stretching are the major contributors to tropical cyclone southwestward motion, rather than the commonly referred channeling effect. The asymmetric rainfall disappears when the tropical cyclone starts moving northward in its end‐stage of looping. The northward motion is dominated by horizontal advection of PV tendency. The results highlight the influence of convection and downslope wind near steep terrain on the initial southward motion in the looping track. The model sensitivity study indicates that a large part of track uncertainty resides in the interaction of model internal dynamics and model cloud microphysics.

In this study, we propose a volume‐to‐point framework for quantitative precipitation estimation (QPE) based on the Quantitative Precipitation Estimation and Segregation Using Multiple Sensor (QPESUMS) Mosaic Radar data set. With a data volume consisting of the time series of gridded radar reflectivities over the Taiwan area, we used machine learning algorithms to establish a statistical model for QPE in weather stations. The model extracts spatial and temporal features from the input data volume and then associates these features with the location‐specific precipitations. In contrast to QPE methods based on the Z–R relation, we leverage the machine learning algorithms to automatically detect the evolution and movement of weather systems and associate these patterns to a location with specific topographic attributes. Specifically, we evaluated this framework with the hourly precipitation data of 45 weather stations in Taipei during 2013–2016. In comparison to the operational QPE scheme used by the Central Weather Bureau, the volume‐to‐point framework performed comparably well in general cases and excelled in detecting heavy‐rainfall events. By using the current results as the reference benchmark, the proposed method can integrate the heterogeneous data sources and potentially improve the forecast in extreme precipitation scenarios.

This study conducted a cluster analysis on the fine particulate matter (PM_2.5) data over Taiwan from 2006 to 2015 and diagnosed their association with the synoptic weather patterns. Five clusters are identified via a hierarchical clustering algorithm; three of them correspond to severe events, each with a distinct pattern of temporal evolution within the 240-h window. The occurrence of the different clusters exhibits strong seasonal variation. Two of the polluted clusters are more frequently associated with weak synoptic weather, while the other one is related to northeasterly winds and fronts. Detailed case studies show that the weather patterns’ temporal evolutions clearly modulate the transition among various pollution clusters by influencing the changes in local circulation and atmospheric stability. In winter, the clusters characterizing severe PM_2.5 pollution events occur when Taiwan is influenced by persistent weak synoptic condition, while in autumn, the long-range transport by strong northerly winds leads to the occurrence of severe PM_2.5 pollution. The current results shed light on the potential of combining the data-driven approach and the numerical weather forecasting model to provide extended range forecasts of local air pollution forecasts.

In this study, we proposed a volume-to-point framework for quantitative precipitation estimation (QPE) based on the QPESUMS (Quantitative Precipitation Estimation and Segregation Using Multiple Sensor) Mosaic Radar dataset. With a data volume consisting of the time series of gridded radar reflectivities over the Taiwan area, we used machine learning algorithms to establish a statistical model for QPE in weather stations. The model extracts spatial and temporal features from the input data volume and then associates these features with the location-specific precipitations. In contrast to QPE methods based on the Z-R relation, we leverage the machine learning algorithms to automatically detect the evolution and movement of weather systems and associate these patterns to a location with specific topographic attributes. Specifically, we evaluated this framework with the hourly precipitation data of 45 weather stations in Taipei during 2013 ~ 2016. In comparison to the operational QPE scheme used by the Central Weather Bureau (CWB), the volume-to-point framework performed comparably well in general cases and excelled in detecting heavy-rainfall events. By using the current results as the reference benchmark, the proposed method can integrate the heterogeneous data sources and potentially improve the QPF in extreme precipitation scenarios.