A single parameter α >>1 is introduced to the Modified Continuity Equation (MCE). The eigenvalues derived from the linearized equations show this modification significantly slows and damps acoustic waves, while neither altering the equation of state nor affecting the amplification or propagation speed of internal gravity waves. The MCE preserves the realistic property of gravity waves while relaxes time-step constraints imposed by acoustic waves, where the speed reduced to 1/ √(α) of that in the unmodified acoustic waves.
The Matsu Archipelago is a coastal marine system heavily influenced by freshwater discharge from the Minjiang River, which is often intensified by episodic typhoons. To understand how such extreme weather event restructure estuarine plankton communities, we conducted high-frequency day–night sampling for nine consecutive days at Nangan Island, Matsu Archipelago, following Typhoon Danas in July 2025. Initial post-typhoon conditions were characterized by low salinity ( 24.5 PSU) followed by subsequent sharp rebound and stabilization ( 32 PSU), thus we defined two distinct phases: a rain-affected “Disturbed period” and a post-rain “Stable period”. During the early Disturbed period, the pioneer taxon Noctiluca scintillans briefly appeared, followed by a transient bloom of the opportunistic dinoflagellate Alexandrium that coincided with the rapid salinity surge between the two periods. After the system transitioned into the Stable period, an exponential diatom bloom dominated by Chaetoceros developed, triggering a rapid increase in copepod abundance. These results demonstrate a rainfall-driven succession from disturbance-associated pioneer and opportunistic assemblages to a stabilized, diatom-dominated community and subsequent grazer response. This pattern reflects recovery toward a typical nutrient-rich and highly productive estuarine ecosystem. By resolving these short-lived transitions, our study highlights the critical role of high-frequency, event-scale observations in capturing transient ecological processes and advancing our understanding of trophic dynamics and ecosystem responses in river-influenced coastal systems.
The TRITON satellite, developed by the Taiwan Space Agency (TASA) and launched in October 2023, is equipped with a global navigation satellite system (GNSS) reflectometry receiver and operates in a high-inclination orbit. The ocean surface wind speed (OSW) is retrieved from GNSS signals reflected off the ocean surface. This study examines the impact of assimilating TRITON OSW in addition to radar observations on convective-scale precipitation forecasts, focusing on a nearshore heavy rainfall event on 23 April 2024 characterized by high position uncertainty. Given the track-based sampling of GNSS-R, the TRITON OSW data available for this event fortuitously cover two critical areas, including weak southwesterly flow over the Taiwan Strait and the Meiyu front, northeast of Taiwan. The TRITON OSW assimilation is conducted using a Weather Research and Forecasting (WRF)-based Radar Ensemble Data Assimilation framework on a 4-km analysis grid with a rapid update cycle of 15 min. Despite the limited temporal availability of OSW data, a positive impact on the wind analysis and very short-term forecast is identified. The assimilation of TRITON OSW data effectively corrects low-level wind, successfully complementing ground-based radar data where near-surface observations are unavailable. Consequently, the weak winds over the Taiwan Strait and offshore southwestern Taiwan, as well as the strong, front-associated winds northeast of Taiwan can be better represented. The OSW assimilation also enhances moisture in the southwestern offshore region. These adjustments yield more accurate predictions for the intensity and location of linear-type nearshore heavy rainfall systems. The improvement stems from enhanced convergence within and above the planetary boundary layer, as well as greater convective instability. Furthermore, the incremental assimilation of TRITON OSW observations with successive cycles provides the possibility of increasing the value of the TRITON OSW, given its dense distribution along a sequence of specular points. This study serves as an initial case-specific evaluation exploring the dynamic mechanisms through which TRITON OSW data can complement the radar assimilation framework and improve nearshore convective representation.
In this study, the features and mechanisms characterizing a heavy rainfall event associated with the Mei-Yu front near the coast of Taiwan on June 7, 2022, were examined through ensemble simulations. Additionally, the correspondence between favorable environmental conditions and heavy precipitation in ensemble simulations was assessed to verify the relationships among the primary factors contributing to precipitation. The study performed ensemble sensitivity analysis (ESA) to capture the relationship between precipitation and dynamic and thermodynamic fields, and it classified ensemble members on the basis of similarities in precipitation or model variables through k-means clustering. ESA revealed wind and water vapor features as crucial drivers of heavy rainfall near the coast. The clustering process effectively separated members with higher rainfall amounts and favorable rainfall conditions from those without such conditions. The large-scale frontal system and mesoscale dynamic processes both contributed to the heavy rainfall along the coast.
In this study, the skill of 24—h quantitative precipitation forecasts (QPFs) for nine verification periods in three Mei-yu events during dry-runs by the cloud-resolving multi-model ensemble (with grid sizes of roughly 1 − 3 km) in the Taiwan Area Heavy-rainfall Prediction Experiment (TAHPEX) is evaluated. Categorical statics of threat score (TS) and bias score (BS) at thresholds of 50 − 350 mm (per 24 h) are employed, for QPFs out to five days at most. Overall, the members show improvements in QPF skills compared to previous forecast verification studies, with TS reaching at least 0.22 at 0 − 24 h, 0.19 at 24 − 48 h, and 0.15 at 48 − 96 h at thresholds of 130 mm and below. The TSs are only slightly lower at 200 mm, but in general ≤ 0.1 at 350 mm. The lead time with some skill is also extended to beyond three days. Most members, however, under-predict heavy rainfall with BS < 1, more serious toward higher thresholds, at ranges close to 48 − 72 h and beyond, and over the plain areas (≤ 300 m in elevation) compared to the mountains (> 300 m). A key result of the present work is that the skill of QPFs and thus the predictability is significantly higher over the mountains in Taiwan than the plains, as a large component of mountain rainfall is phase-locked and stationary. Thus, the mountain regions consistently exhibit higher TSs, BSs closer to 1 (with less under-prediction), and higher probabilities of heavy rainfall derived from the ensemble runs across all members, lead times, and thresholds examined. Among the ensemble products, the exceedance probability from the most-rainy member often has higher TSs than all other products, including probability matching, because it has the least under-forecast and can over-forecast. It is more useful over the mountains. Finally, the finer 1—km members, executed only once per day, show good potential for further improvement to QPFs in heavy-rainfall scenarios.
This study examines the relationship between operational ground-motion parameters and building damage during the 2016 Mw 6.4 Meinong earthquake in southern Taiwan. We integrate strong-motion recordings from 599 Taiwan Strong Motion Instrumentation Program (TSMIP) stations with a manually verified, georeferenced dataset of 626 damaged buildings, including 271 red-tagged and 355 yellow-tagged buildings in Tainan City. The analysis was conducted at the administrative-district level by comparing district-level damage rates with interval-based peak ground acceleration (PGA) and peak ground velocity (PGV) estimates. Non-zero district-level damage rates first appeared in the PGA interval of approximately 150–175 gal and the PGV interval of approximately 10–20 cm/s, corresponding to approximately MMI VI–VII. In this interval-averaged district-level analysis, PGA showed a clearer descriptive correspondence with red- and yellow-tagged building damage rates than PGV. Building-story information indicates that approximately 70
This paper proposes a novel approach to identifying the annual cycle and variability of the East Asia and western Pacific (EA-WP) monsoon. A low-level circulation pattern (LCP) calendar is constructed based on nine LCPs obtained from K-means cluster analysis for 46 years (1979–2024) of daily 850-hPa wind data. The LCP daily occurrence frequency reveals climatological features of the monsoon annual cycle and seasonal progression. The LCP-based summer and winter monsoon indices well represent key monsoon characteristics. A strong East Asian summer monsoon (EASM) corresponds to more frequent occurrences of LCP featuring low-level anticyclonic circulation over the western North Pacific, while a strong East Asian winter monsoon (EAWM) corresponds to more frequent occurrences of LCPs characterized by low-level northerly winds over the South China Sea and the Philippine Sea. The variability of EASM and EAWM is synchronized during the developing phase of ENSO events. Strong (weak) EA-WP monsoon years are marked by strong (weak) EASM followed by strong (weak) EAWM, whereas no clear relationship is found between EASM and its preceding EAWM. Approximately 86
Afternoon thunderstorms (ATs) frequently occur over southwestern Taiwan during the warm season and pose significant challenges to aviation operations owing to their rapid development, intense rainfall, strong winds, and abrupt reductions in visibility. Despite their operational importance, the physical mechanisms governing the propagation and longevity of ATs in the Chiayi region remain poorly documented in the international literature. This study examines two contrasting AT events that occurred near Chiayi Airport on 6 and 7 August 2022, representing a short-lived and a long-lived convective system, respectively, under similar weak synoptic conditions. A comprehensive observational analysis is conducted using dual-polarization radar data, three-dimensional wind field retrievals, surface meteorological observations, and ERA5 reanalysis data. The evolution of convective structure, cold-pool dynamics, vertical wind shear, mid-level moisture, and convective cell interactions is analyzed to identify the mechanisms responsible for the differing storm behaviors. On 6 August, convection initiated in association with terrain-induced local circulations but rapidly weakened before reaching Chiayi Airport. Radar-derived vertical structures reveal a shallow and nearly upright updraft accompanied by strong low-level outflow. The cold-pool propagation speed exceeded the magnitude of the environmental vertical wind shear, allowing the cold pool to undercut the updraft and suppress sustained convective regeneration, consistent with an unfavorable cold-pool–shear balance within the Rotunno–Klemp–Weisman (RKW) framework. Limited mid-level moisture and weak convective cell merging further contributed to the short-lived nature of the system. In contrast, the 7 August event was characterized by a more favorable, though non-ideal, vertical wind structure, enhanced mid-tropospheric moisture, and persistent convective cell merging. The cold-pool propagation speed was comparable to the environmental shear, indicating a near-balanced RKW state. Successive merging of convective cells broadened the updraft region, enhanced system-scale low-level convergence along the gust front, suppressed mid-level dry-air intrusion, and promoted continuous new cell formation. As a result, the convective system evolved into a deep, organized, and long-lived thunderstorm that propagated northeastward and directly impacted Chiayi Airport. The comparison demonstrates that while a favorable cold-pool–shear balance is a necessary condition for maintaining organized convection, it is not sufficient to explain thunderstorm longevity in isolation. Convective cell merging, modulated by mid-level moisture, acts as a critical amplifying mechanism that stabilizes and prolongs the cold-pool–shear interaction under weak synoptic forcing. These findings provide new insights into the dynamics of afternoon thunderstorms over complex coastal–mountain environments and have important implications for short-term convective forecasting and aviation weather operations in southwestern Taiwan.
Abstract Local magnitude (M L ), the earliest magnitude scale developed in southern California, enables rapid earthquake characterization based on observed amplitudes and zero-magnitude reference amplitudes (A 0 ) derived from local events. However, amplitudes are sensitive to path and site effects, motivating the development of the physically robust moment magnitude (M W ). Previous studies have demonstrated an approximately 1:1 relationship between M L and M W for M L < 6.5 in southern California. In contrast, M L in Taiwan systematically overestimates M W , reflecting distinct regional attenuation characteristics. Although prior work has recalibrated the logA 0 attenuation model for shallow earthquakes in Taiwan, deep events for focal depth larger than 35 km exhibit even larger overestimation, with an average bias of 0.5 magnitude units. In this study, we investigate deep earthquakes in Taiwan and establish a new depth-dependent logA 0 attenuation model. A model relying solely on hypocentral distance produces depth-dependent residuals; therefore, a depth term (logD) is incorporated to logA 0 attenuation model. The final regression model is logA 0 = 0.097–1.587logR − 0.0014R + 0.417logD ± 0.273, where R is hypocentral distance and D is focal depth. The results demonstrate that logA 0 attenuation varies systematically with both distance and depth, consistent with Richter’s description that earthquakes at different depths require distinct logA 0 values. The recalibrated M L shows no depth dependence and exhibits a stable 1:1 relationship with M W , with a standard deviation of 0.16. The proposed logA 0 attenuation model enables rapid and reliable M L estimation for deep earthquakes in Taiwan, enhancing real-time hazard assessment and reducing magnitude conversion uncertainty in combined earthquake catalogs.
Abstract Reliable flood susceptibility assessment under current and future climate-change scenarios is critical for regional disaster preparedness but is often limited by the scarcity of historical inundation records and field-surveyed data. To overcome this limitation, this study introduces a hybrid modeling framework that integrates high-performance hydrodynamics with machine learning (ML) to synthesize robust training datasets. Utilizing the GPU-accelerated Two-dimensional Runoff Inundation Toolkit for Operational Needs (TRITON), we simulated 275 maximum precipitation events derived from historical, Shared Socioeconomic Pathways (SSP), and Global Warming Level (GWL) scenarios. These physics-based simulations served as ground truth for training three ML classifiers: Convolutional Neural Networks (CNN), Random Forest (RF), and Support Vector Machine (SVM). Applied to a flood-prone region in central Taiwan, the framework achieved high predictive accuracy (approx. 0.80) across all models, with RF demonstrating superior stability. The assessment reveals that under climate change scenarios, areas classified as “very-high susceptibility” will expand significantly—from 7% in the historical baseline to 15% under GWL 4.0. This study demonstrates that coupling GPU-based physical models with data-driven algorithms effectively overcomes data scarcity, providing a scalable and scientifically rigorous tool for flood risk management in a changing climate.
Abstract This study investigates the interdecadal variability of intraseasonal oscillation (ISO) activity in the western North Pacific (WNP) during July-September using a dataset from 1979 to 2021. Analysis of the regime shift index identifies three distinct epochs: 1979–1993, 1994–2004, and 2005–2021. The middle epoch stands out for its pronounced westward extension of the subtropical anticyclone and peak activity of the westerly northward-propagating ISO. During this period, the interaction between the enhanced subtropical anticyclone and the strengthened ISO cyclonic anomalies intensifies southeasterly winds south of Japan. These winds act as a barrier to the formation and development of tropical cyclones (TCs) in the region east of the strong wind zone. Concurrently, the anticyclonic anomalies over the South China Sea hinder TC movement into this area. These combined effects lead to more clustered TC tracks between Taiwan and Japan, with increased TC frequency, intensity, and duration, supported by an active ISO background. The findings highlight a strong connection between interdecadal ISO variability and TC characteristics, underscoring the relevance of ISO behavior in long-term climate projections and TC forecasting. These insights are essential for improving prediction accuracy in the face of shifting climate patterns.
Magnetic field data from Swarm’s magnetometers were investigated during the 2025 Mw 8.8 Kamchatka earthquake. The examination focuses on the spatiotemporal distribution of magnetic data with Flag 8, which indicates discrepancies between the scalar and vector magnetometers. Flag 8 was observed along Swarm satellite orbits near the epicenter prior to the earthquake, initially appearing approximately 16 days before the mainshock and persisting until 9 days after. The results indicate that the spatial distribution of Flag 8 exhibits high density and extensive coverage around the epicentral region, showing a close relationship with the earthquake. Further examination suggests that neither polar geomagnetic activity nor the South Atlantic Anomaly can adequately explain the observed pattern. This implies that the observed Flag 8 is more likely associated with the seismic activity in the Kamchatka Peninsula. This study highlights Flag 8 as a complementary tool for detecting pre-earthquake magnetic anomalous phenomenon and advancing earthquake precursor research.
Abstract In this study, we compiled historical and instrumentally-recorded larger-sized earthquakes with M s ≥5 in the Taichung-Changhua-Nantou area, Taiwan. Totally, thirty-one M s ≥5 earthquakes, including mainshocks of earthquake sequences and the larger-sized events of two swarms, occurred in the area from 1795 to 2025 are taken into account. The epicentral distribution demonstrates that the earthquakes happened mainly in Nantou. The temporal variation in earthquakes displays irregular recurrence behavior with low periodicity. For the whole area, the time series may be separated into three time intervals. Considering the inter-occurrence time between two sequent events, the largest value was 22,242 days (60.94 years) and the smallest one was 7 days (0.0193 years). The average inter-occurrence times for the whole time interval, the first, second, and the third time intervals are, respectively, 2656.70 days (7.28 years), 3604.64 days (9.88 years), 853.73 days (2.34 years), and 6811.00 days (18.66 years). In terms of inter-occurrence time, observed probabilities only slightly deviate from the theoretical Poisson probability function. The activity of earthquakes is slightly high in September, November, and December, and low in May and August. Seismic activity is about two times higher in the local night-time than in the local daytime, thus indicating that the effects on generating M s ≥5 earthquakes could be higher from lunar tides than from solar tides.
Abstract Shallow S-wave velocity (VS) structures in Tainan City, Taiwan, were characterized using microtremor array data for 18 sites compiled by Huang et al. (Terr Atmos Ocean Sci 35:20, 2024). Theoretical transfer functions were calculated at the surface relative to six depth formations (50, 100, 200, 400, 700, and 1,000 m) and five VS formations (350, 550, 750, 1,000, and 1,500 m/s) by using the Haskell method. Higher predominant frequencies were observed in the Tainan tableland (central part of the study area), whereas lower predominant frequencies were observed in the Anping plain (western part of the study area). If the S-wave velocity in the Tertiary basement is 1,500 m/s, the predominant frequencies of Quaternary sediments range from approximately 0.2 Hz (in the Anping plain) to 0.65 Hz (in the Tainan tableland), decreasing radially from the central to surrounding sites. We compared the horizontal-to-vertical (H/V) ratios of microtremors with the predominant frequencies calculated using theoretical transfer functions. Our findings suggest that site amplifications derived from H/V ratios are primarily attributable to alluvium from the surface to depths of 50–100 m at most sites. For sites in the Tainan tableland, amplifications derived from H/V ratios are primarily attributable to alluvium between the surface and the formation of VS = 550–750 m/s. For sites in the Anping plain and the Dawan lowland, amplifications are primarily attributable to alluvium between the surface and the formation of VS = 350–550 m/s.
Accurate reconstruction of rock thermal history is prerequisite for understanding fundamental geological processes and assessing resource potential. Geothermometers provide critical constraints on the peak thermal exposure experienced by rocks or minerals during burial, diagenesis, or metamorphism. Conventional methods, such as vitrinite reflectance (VR) and fission-track (FT) thermochronology, yield valuable tem- perature data but are often constrained by time-intensive procedures, high resource consumption, and limitations in sample quantity or analytical spatial resolution. Micro-Fourier Transform Infrared (Micro-FTIR) Spectroscopy presents a viable alter- native for analyzing thermal alteration in geological media. This technique monitors infrared absorption changes in molecular bonds, revealing functional group transfor- mations that are acutely sensitive to thermal maturation and are effective proxies for reconstructing thermal history. This research optimizes an IR Geothermometer proto- col to address the intrinsic limitations of established techniques, thereby enhancing the precision of thermal event reconstruction. The methodology encompasses acid treatment for the isolation of organic components, Micro-FTIR analysis, and continuous heating experiments, resulting in broader sample compatibility and refined tempera- ture constraints. Methodological validation was performed by comparing IR-derived temperature estimates against VR data from coal samples and FT data from sedi- mentary and metamorphic rocks in Taiwan.The findings establish a robust empirical correlation between the observed IR thermal signatures and independent geothermo- metric controls, thereby confirming the technique’s efficacy for geological temperature history reconstruction. Utilizing the rapid spectral acquisition, high spatial resolu- tion, and capacity to register multiple thermal events inherent to IR spectroscopy, this technique offers an efficient, high-resolution methodology for thermal assessment in geological samples, with broad utility across resource exploration and petroleum geology.
Weathering processes occur on terrestrial surfaces in combination with atmospheric and oceanic environments. Salt weathering often results in the formation of a honeycomb structure (tafoni) and a concave microrelief. Tafoni has been investigated extensively in geomorphological studies, whereas causative factors, including microbial effects, have rarely been investigated in previous studies. This study discusses the distribution of microbial communities as an environmental factor in honeycomb weathering on sandstone surfaces. The study sites are located on Yonaguni Island in the Ryukyu Arc, where Miocene sandstone (Yaeyama Group) is extensively exposed on the island and forms numerous cliffs with a honeycomb structure resulting from physical disintegration due to salt weathering. Microbial community analysis identified potential candidates that could contribute to honeycomb weathering. Acidobacteria can dissolve iron-rich minerals through the production of organic acids, whereas other bacteria, such as Actinobacteria, utilize iron ions via siderophores. Microbial activity, particularly associated with abundant Cyanobacteria, may contribute to chemical and physical weathering by increasing decomposition and disintegration.
This study investigates the decadal variation in the modulation of the El Niño–Southern Oscillation (ENSO) on the summer Indian Ocean Basin Mode (IOBM), revealing a weak–strong–weak pattern since the 1950s. Since the 1950s, the ENSO–summer IOBM relationship has experienced two decadal shifts, with the 1980s and 2000s as transition points. Notably, even during the two weak-correlation subperiods, the spatiotemporal evolution of Indian Ocean SST shows distinct differences, and ENSO’s modulation processes vary significantly. A strong ENSO–summer IOBM relationship occurs only during 1982–2003, maintained by slower ENSO decay, prolonged atmospheric bridge effects, effective propagation of oceanic Rossby waves in the southern Indian Ocean, and a well-established wind–evaporation–SST (WES) feedback sustaining the IOBM through summer. Weak relationships are observed before 1981 and after 2004, as the IOBM decays in early spring (1958–1981) or late spring (2004–2022). The rapid decay of the IOBM in the first subperiod is linked to fast ENSO decay, which terminates both atmospheric teleconnections and oceanic processes early, causing the warming to vanish by early spring. In the third subperiod, although slower ENSO decay allowed longer persistence, anomalous local winds disrupted the WES feedback, preventing basin-wide warming after late spring. These findings emphasize the importance of process-based analyses in understanding and predicting ENSO’s delayed impacts on the Indian Ocean.
Global Navigation Satellite Systems-Reflectometry (GNSS-R) technique is used to explore the Earth environment by using the Earth surface reflected GNSS signal. The Earth surface reflected GNSS signal can be used to retrieve the Earth surface parameters. The space based GNSS-R, which set receiver on the satellite in space to receive the Earth reflected GNSS signal, is developed from the early 21st century. Triton, a Taiwan designed and manufactured experimental micro-satellite, is one of the satellites for GNSS-R mission and was launched in October 9th, 2023. The mission payload of Triton is Taiwan Space Agency (TASA) self-developed GNSS-R receiver and used to process ocean surface reflected Global Positioning System (GPS) signal. The product of mission payload is delay-Doppler map (DDM) for ocean surface wind speed retrieving. The GNSS-R retrieval system of Triton is developed in Taiwan R/RO process system (TROPS) to retrieve ocean surface wind speed by using DDM. In the retrieval process, the first step is DDM calibration, which is used to remove the influence of payload hardware in the signal strength. Then the calibrated DDM is used to calculate supporting data, such like normalized bistatic radar cross section (NBRCS). After that, the calibrated DDM and supporting data can be used to retrieve ocean surface wind speed. Before retrieving ocean surface wind speed by using GNSS-R function in TROPS, the geophysical model function (GMF) needs to be developed. The ocean surface wind speed product of Triton has been released freely in Taiwan Analysis Center for COSMIC (TACC). In this paper, the detail of retrieval process is introduced. The retrieval ocean surface wind speed is compared with those obtained from European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis v5 (EAR5) for validation. Furthermore, some supporting data is also be demonstrated.
Accurate and rapid earthquake magnitude estimation is essential for effective Earthquake Early Warning Systems (EEWS), particularly in seismically active regions such as Taiwan. In this study, we propose a frequency-domain multi-station deep learning approach that integrates a Fourier Neural Operator (FNO) and a Long Short-Term Memory (LSTM) network to estimate earthquake magnitudes using near real-time waveform data. The model is trained on 7,025 events recorded by Taiwan’s networks between 2012 and 2024. By jointly analyzing three-component waveforms from the ten nearest stations and incorporating hypocentral distance corrections, the proposed method achieves high accuracy within seconds after the fourth triggered P-wave arrival. Evaluation on a held-out test set demonstrates significant improvements over the current Pd-based single-station method currently used by the Central Weather Administration (CWA), with lower mean absolute error, root mean squared error, and prediction variance. In real EEWS scenarios from January to June 2025, the model consistently outperformed the operational method, even when using early-stage location estimates, and showed further gains when provided with precise hypocenters from the catalog. The proposed approach combines robustness, and computational efficiency, enabling practical deployment in EEWS of Taiwan.
In Taiwan, the Long-Term Weather Outlook, which is provided by the Central Weather Administration (CWA) and indicates the likelihood of different weather categories in the future, has been applied to assess streamflow. However, the previous approaches had not addressed the uncertainty arising from both intrinsic uncertainty and the mismatch between the stations targeted by the weather outlook and the locations where it is applied. In order to address this issue, this study used the monthly weather outlook by the CWA for Northern Taiwan, along with its past prediction performance, to estimate the future exceedance probability of streamflow at the Shimen Reservoir. Logistic regression was employed to calculate the posterior probabilities of future weather categories based on the CWA’s weather outlook. Results suggest that the posterior probabilities of weather classes—the modified outlook—differ from those indicated by the original outlook and enhance inflow prediction performance, highlighting the importance of estimating these probabilities. Using the streamflow exceedance probability based on this modified outlook, water resource decisions can be made more appropriately compared with traditional approaches. Our study presents a novel approach for enhancing the utility of existing public weather forecast products in Taiwan, supporting decision-making under uncertainty.
