Time Series Of Variables Research Articles

Nuclear accident-derived 129I in several river water, eastern Fukushima, Japan, 2016-2020.

Radionuclides, including 129I, were released into the atmosphere by the Fukushima nuclear power plant accident. We measured the dissolved 129I concentration in 11 rivers in eastern Fukushima from 2016 to 2020 to clarify the 129I concentration level in river water under base-flow conditions. During the study period, the maximum 129I concentration in the river water was 1.6×10-6 Bq L-1 in the Ukedo River in 2016, and the minimum was 1.2×10-7 Bq L-1 in the Uda River in 2017, with a difference of over an order of magnitude between the two. Water samples from the Ukedo, Niida-U, and Nagadoro Rivers in eastern Fukushima were strongly affected 129I from the Fukushima accident. In addition, we found a significant correlation between concentration of 129I in river water and mean 129I contamination density in the catchment. The time-series variation of 129I in river water from 2016 to 2020 was small and showed no year-to-year decrease. The 129I concentration in each river hardly decreased compared to the 137Cs decay-corrected concentration during the observation period, and it was expected that a certain concentration of dissolved 129I would continue to flow out over a long period of time.

Sub-hourly forecasting of fire potential using machine learning on time series of surface weather variables

Background Rapidly developing pre-fire weather conditions contributing to sudden fire outbreaks can have devastating consequences. Accurate short-term forecasting is important for timely evacuations and effective fire suppression measures. Aims This study aims to introduce a novel machine learning-based approach for forecasting fire potential and to test its performance in the Sunshine Coast region of Queensland, Australia, over a period of 15 years from 2002 to 2017. Methods By analysing real-time data from local weather stations at a sub-hourly temporal resolution, we aimed to identify distinct weather patterns occurring hours to days before fires. We trained random forest machine learning models to classify pre-fire conditions. Key results The models achieved high out-of-sample accuracy, with a 47% higher accuracy than the standard fire danger index for the region. When simulating real forecasting conditions, the model anticipated 75% of the fires (11 out of 15). Conclusions This method provides objective, quantifiable information, enhancing the precision and effectiveness of fire warning systems. Implications The proposed forecasting approach supports decision-makers in implementing timely evacuations and effective fire suppression measures, ultimately reducing the impact of fires.

Detecting and Explaining Long-Term Trends in a Weed Community in a Biennial Cereal–Legume Rotation

The technique of Dynamic Factor Analysis (DFA), which aims to reduce the dimensionality of time-series data, was utilized in order to model the changes over time in eight different long-time-series weeds (26 years) growing in a biennial cereal–legume rotation. The aim of the present study was to determine the existence of long-term trends in a weed community and to identify the factors that determine them. A common trend was extracted that captured the main signal of abundance over time, indicating latent influences affecting all species. Canonical correlation analysis showed strong associations between the common trend and specific weed species, suggesting differential responses to this latent factor. Local (temperature and precipitation) and global weather factors (North Atlantic Oscillation (NAO)) were considered as explanatory variables to explain the common trend. The local weather variables considered did not play a significant role in explaining the commonly observed trend. Conversely, NAO showed a significant relationship with the weed community, indicating its potential role in shaping long-term weed dynamics. DFA was found to be useful for studying the variability in multivariate weed time-series without the need for detailed a priori information on the underlying mechanisms governing weed population dynamics. Overall, this study provided valuable insights into the long-term drivers of weed dynamics and set the stage for future research in this area.

Machinery fault diagnosis based on a transfer learning model with local sparse structure of PMDCNN-LSTM

Aiming at the fault diagnosis problem in practical applications where the performance of the model degrades when applied to new domains or tasks due to limited and approximate training data, a transfer learning (TL) model with a local sparse structure based on a parallel multichannel convolutional neural network (CNN) and a long-short-term memory network (LSTM) (PMDCNN-LSTM) is proposed. The model is combined with CNN, LSTM, and TL to achieve parameter sharing by training learning parameters in the source domain and transferring them to the target domain. At the same time, the model is transformed from training data to real-world applications by integrating samples from the target domain, which improves generalization while maintaining the ability to identify faults in the training set. The advantages of CNN and LSTM are combined in this model compared with the traditional TL model, which can efficiently process data containing spatio-temporal features while maintaining sensitivity to time-series variations, exhibiting higher performance and advantages in complex data processing. It is shown by the experimental results that excellent performance and generalization ability can still be maintained by the model in high-noise environments, which is considered to be of great significance for research on bearing and gear fault diagnosis.

Atmospheric River Frequency‐Category Characteristics Shape U.S. West Coast Runoff

AbstractThis study investigates the factors influencing runoff response to atmospheric rivers (ARs) over the U.S. West Coast. We focused on runoff time series variations impacted by AR characteristics (e.g., category and frequency) and land preconditions during Northern Hemisphere cool seasons in the period of 1940–2023. Results show that high‐category ARs significantly increase local runoff with higher hourly precipitation rates leading to a greater incremental rate and peak runoff. Extreme runoff increases greatly with the AR category with an increase rate up to 12.5 times stronger than non‐extreme runoff. Besides the AR category, land preconditions such as soil moisture and snowpack also play crucial roles in modulating runoff response. We found that runoff induced by weak‐category ARs is more sensitive to land preconditions than high‐category ARs, with high peak runoff occurring when soil is nearly saturated. Additionally, more than 50% of high‐peak‐runoff events in snow‐covered grid cells are associated with rain‐on‐snow events particularly for the events associated with weaker ARs. Regression analysis reveals that AR precipitation and land preconditions jointly influence runoff, emphasizing the importance of including soil moisture and snowpack levels in AR impact assessments. The study also highlights the intensified runoff response to back‐to‐back ARs with short intervals, which may become more frequent with climate warming, posing increased flood risks via facilitating wet soil conditions. Our findings have significant implications for AR risk predictions and the development of prediction models for AR‐induced runoff.

A two-layer optimal scheduling method for microgrids based on adaptive stochastic model predictive control

Abstract Renewable energy is highly susceptible to weather and environmental factors, and changes dramatically in a short period, which makes it difficult for traditional fixed-period scheduling of microgrids to capture the time-series variations of source and load, exacerbating the power imbalances of systems. To address this problem, a novel two-layer rolling optimization framework for microgrids based on adaptive stochastic model predictive control is proposed in this paper. Firstly, a two-layer microgrid optimization model based on stochastic model predictive control is established, in which measurement technology plays an indispensable role in the quality of uncertain scenarios and optimal decision results of the microgrid, as well as providing impetus for the development of the microgrid. In the upper-layer prescheduling stage, the optimal control sequence under multiple uncertainties is obtained by solving a scenario-based chance constraint programming model. In the lower-layer power compensation stage, the measured scenario errors are regarded as fluctuations, and the energy storage devices are preferentially used for smoothing processing. Secondly, an adaptive period division method using Wasserstein distance-based hierarchical clustering is developed to guide intraday online scheduling. The concepts of distribution similarity and distribution loss are presented to adaptively divide the periods under uncertain conditions, so as to overcome the disturbance of uncertainty and improve the flexibility of microgrid scheduling. Finally, the simulation results show that the proposed method can flexibly deal with the uncertainty at different time scales, and achieve 53.63 kWh less compensation power and 19.12% lower operating cost than traditional fixed-period scheduling methods, and thus effectively improve the economy and reliability of microgrid operation.

Comparative study on lake ice phenology changes and driving factors in large lakes of mid-latitude Xinjiang, China.

The changes in lake ice phenology (LIP) can intuitively reflect the climate evolution in the regions where lakes are located, serving as an important indicator of climate change. The Tianshan Mountains, situated at the southern edge of freezing lakes in the Northern Hemisphere, are a crucial water resource base in Xinjiang and support significant ecosystems closely related to human activities. In the context of intensified climate change, this study focuses on the geographical location, altitude, and water quality differences among large lake groups in the mid-latitude region of Xinjiang, aiming to explore the characteristics of LIP changes in these lakes and their responses to driving factors, thereby providing a basis for effective environmental management and protection. This research conducts a comparative analysis of the LIP changes and driving factors of three large lakes-Sayram Lake (SL), Bosten Lake (BL), and Ebnur Lake (EL)-using multi-source remote sensing data to reveal the response and adaptation mechanisms of lakes under global warming. It effectively captures the time series variations of ice formation and melting, as well as the common responses to environmental and climatic factors. The results indicate that SL has experienced significant climate change effects, with earlier freezing times and accelerated melting speeds; In contrast, EL and BL have shown relatively minor changes, suggesting that geographical and hydrological factors may buffer the impacts of climate. The study finds that all three lakes are jointly influenced by environmental factors such as temperature, wind speed, and precipitation; however, due to differences in altitude, lake surface area, and water transparency, their responses to these climatic factors vary significantly. For instance, SL's high altitude gives water transparency a dominant role in LIP, while BL's larger surface area enhances the impact of precipitation and thermal capacity on the melting process. This indicates that, despite facing similar climate pressures, local environmental conditions can lead to different trends in ice phenology changes. This study offers a novel and efficient monitoring method for LIP, providing valuable insights for future LIP research and water resource management.

Test for Change in Error Variance of Multiple Time Series

Changes in levels of multiple time series based on their common components helps characterize the shared behavior of the data generating process with known events causing perturbations in their movements. However, changes in variance of the error structure leads to misspecification of models that complicates statistical inference. Volatility structures may incorporate variance into the time series models, but these can easily lead to overparameterization in multiple time series. A model with inherent changes in variance structure is used to develop a bootstrap-based test for presence of changes in variance of multiple time series. Simulation studies shows that the test is correctly-sized and powerful compared to CUSUM-based test in a wide range of scenarios. The test is advantageous specially in the presence of strong autocorrelations and even in unbalanced data. Using global stock prices with the onset of lockdowns against the COVID19 pandemic as the stimulus of the changepoint, the test was able to detect significant changes in error variance before and during the pandemic. While CUSUM-based test also recognized the significant changes in multiple time series, linear trend was misconstrued as evidence of the presence of change in variance in stock prices.

Adaptive parameters identification for nonlinear dynamics using deep permutation invariant networks

The promising outcomes of dynamical system identification techniques, such as SINDy (Brunton et al. in Proc Natl Acad Sci 113(15):3932–3937, 2016), highlight their advantages in providing qualitative interpretability and extrapolation compared to non-interpretable deep neural networks (Rudin in Nat Mach Intell 1(5):206–215, 2019). These techniques suffer from parameter updating in real-time use cases, especially when the system parameters are likely to change during or between processes. Recently, the OASIS (Bhadriraju et al. in AIChE J 66(11):16980, 2020) framework introduced a data-driven technique to address the limitations of real-time dynamical system parameters updating, yielding interesting results. Nevertheless, we show in this work that superior performance can be achieved using more advanced model architectures. We present an innovative encoding approach, based mainly on the use of Set Encoding methods of sequence data, which give accurate adaptive model identification for complex dynamic systems, with variable input time series length. Two Set Encoding methods are used: the first is Deep Set (Zaheer et al. in Adv Neural Inf Process Syst 30, 2017), and the second is Set Transformer (Lee et al. in: International conference on machine learning, PMLR, pp 3744–3753 2019). Comparing Set Transformer to OASIS framework on Lotka–Volterra for real-time local dynamical system identification and time series forecasting, we find that the Set Transformer architecture is well adapted to learning relationships within data sets. We then compare the two Set Encoding methods based on the Lorenz system for online global dynamical system identification. Finally, we trained a Deep Set model to perform identification and characterization of abnormalities for 1D heat-transfer problem.

Nexus of crude oil and clean energy stock indices: Evidence from time-vector-auto-regression in conjunction with conditional-autoregressive-value-at-risk.

The current study aims to elicit information regarding the tail risk transmission mechanism between crude oil (CO) and selected clean energy (CE) stock indices across time and during certain economic events. A Time-Varying Parameter Vector Auto-Regressive model (TVP-VAR) paired with the conditional autoregressive value-at-risk (CAViaR) approach was used to investigate data from January 1, 2015 to December 29, 2022. Overall, we show that an increased vulnerability to tail risk and deficits might be linked to dynamic spillover over examined markets. We also provide evidence that connectedness rises during significant crisis situations, and the last epidemic has the potential to make a lasting impact on the various marketplaces of concern. According to the return and conditional variance time-series, CE stock indices are the most important source of return shocks to CO. However, the CO is the primary cause of volatility in CE stock indices. During the recent virus pandemic, the most significant volatility shock transmissions from CO to CE stock indices occurred. During the Russia-Ukraine war, volatility shocks to CO were mostly caused by CE stock indices. The results of our study offer concrete consequences and new perspectives to various market players in order to improve the management and understanding of risks.

Autocorrelation and Cross-Correlation of Climatological Time Series: A Bibliometric Analysis

Objective: This article aims to contribute to the investigation of the evolution of scientific knowledge related to autocorrelation and cross-correlation analysis techniques in time series of climatological variables, wind speed and solar radiation, in order to understand what is discussed on the subject and, in addition, to support future productions. Theoretical Framework: Conducting a study to investigate trends in the evolution of publications related to autocorrelation and cross-correlation measures, used to analyze climatological time series, is relevant to consolidate and deepen knowledge in this scientific field. Understanding the behavior of climatic elements is relevant in several areas of application: the spread of respiratory diseases; agricultural and urban planning tools; the relationship between climatological factors and pollutants; and renewable energy sources. Method: To achieve this objective, the bibliometric review technique was used to map the literature on the research topic, assessing the impact of production by country, identifying the authors, institutions and journals that stand out, trends in the number of publications and microanalysis. Results and Discussion: A total of 173 scientific articles were recorded during the analysis period, with the highest productivity peaks during the years 2013 and 2019 with 12 and 13 publications, respectively, representing together 18.25% of the total publications and from these years onwards a significant growth in the number of publications was evidenced. The countries that stood out the most in number of publications were China (21.96%), followed by the United States (4.05%) and Japan (4.05%). The most relevant authors were Li, Y. in the lead with 5 articles and Dinpashoh, Y. with 4, the same authors maintained productions on the subject over time. Conclusion: Our findings highlighted the relevance of mapping scientific advances related to autocorrelation and cross-correlation of climatological time series. It is noted that China is the country with the highest number of citations received and its significant number of articles and citations may indicate that this country has relevant scientific publications on the subject. These results will be useful as a constructive reference for future research on the subject.

Multi-modal prediction of extracorporeal support-a resource intensive therapy, utilizing a large national database.

Extracorporeal membrane oxygenation (ECMO) is among the most resource-intensive therapies in critical care. The COVID-19 pandemic highlighted the lack of ECMO resource allocation tools. We aimed to develop a continuous ECMO risk prediction model to enhance patient triage and resource allocation. We leveraged multimodal data from the National COVID Cohort Collaborative (N3C) to develop a hierarchical deep learning model, labeled "PreEMPT-ECMO" (Prediction, Early Monitoring, and Proactive Triage for ECMO) which integrates static and multi-granularity time series features to generate continuous predictions of ECMO utilization. Model performance was assessed across time points ranging from 0 to 96hours prior to ECMO initiation, using both accuracy and precision metrics. Between January 2020 and May 2023, 101400 patients were included, with 1298 (1.28%) supported on ECMO. PreEMPT-ECMO outperformed established predictive models, including Logistic Regression, Support Vector Machine, Random Forest, and Extreme Gradient Boosting Tree, in both accuracy and precision at all time points. Model interpretation analysis also highlighted variations in feature contributions through each patient's clinical course. We developed a hierarchical model for continuous ECMO use prediction, utilizing a large multicenter dataset incorporating both static and time series variables of various granularities. This novel approach reflects the nuanced decision-making process inherent in ECMO initiation and has the potential to be used as an early alert tool to guide patient triage and ECMO resource allocation. Future directions include prospective validation and generalizability on non-COVID-19 refractory respiratory failure, aiming to improve patient outcomes.

Estimation of turbulent transport coefficients by the conditional variance method

Abstract A novel approach for estimating turbulent transport coefficients in fusion devices is presented. The diagnostic method is established on the analysis of the conditional variance of one-point time series of density or temperature fluctuations. It is tested on data obtained from probe measurements in the edge of the tokamak ASDEX Upgrade and the stellarator Wendelstein 7-X, and on synthetic data from the gyrofluid transport model GEMR. The approach demonstrates a remarkable degree of accuracy, typically within a factor of two of the actual transport measured by more difficult means. It is a simple and accurate way of evaluating turbulent particle and heat transport coefficients that does not require measurements of the velocity fluctuations.

Synchronous and anti-phase drumming elicit similar prosocial behavior ratings

PurposeMusic performance facilitates prosociality across many cultures and contexts. Interestingly, the relationship between prosociality and sensorimotor synchronization (SMS) has so far primarily been demonstrated in the context of in-phase synchrony with only a few mixed results for anti-phase coordination. In anti-phase coordination, participants move at the same rate, at opposite phases, which also requires high levels of coordination and attention. This case is particularly relevant for music and prosociality, as music regularly involves naturalistic anti-phase coordination. We thus tested whether anti-phase synchronization is as effective as in-phase synchronization at eliciting prosocial behavior.MethodsDyads (N = 50 dyads) were randomly assigned to complete four trials of a drumming sensorimotor synchronization-continuation task (SCT) either alone, synchronously or in anti-phase. Before and after the drumming task, dyads completed a behavioral economics game involving trust. Additionally, a questionnaire about trust, cooperation, affect, and similarity was given after the drumming task.ResultsCooperation rates in the stag-hunt game were near ceiling (~87%) across all conditions pre-SCT, with negligible change after the drumming task. Questionnaire items were analyzed using Bayesian probit mixed effects models to account for dyadic sampling and ordinal data, and to provide evidence in favor of the null hypothesis. Models provided moderate to extremely strong evidence that the anti-phase and in-phase coordination conditions rated their affect, trust, similarity, and cooperation more strongly than dyads in the alone condition (all BF10 > 3). When only comparing the anti-phase and in-phase conditions, moderate evidence in favor of the null (i.e., that phase does not affect ratings) was found for all questions (all BF10 < 0.3). Descriptions of the posterior, as well as leave-one-out cross validation (LOO) results, were in general accordance with the Bayes Factor results.ConclusionEvidence indicates anti-phase drumming coordination is as effective as in-phase in increasing perceived trust, cooperation, affect, and similarity. Future analyses will examine how other characteristics of the drumming coordination, such as the lag-1 autocorrelation and variability of the inter-tap interval time-series, relate to prosocial behavior and ratings of trust and cooperation.

Mapping multi-modal dynamic network activity during naturalistic music listening

Abstract The human brain is a complex, adaptive system capable of parsing complex stimuli and generating complex behaviour. Understanding how to model and interpret the dynamic relationship between brain, behaviour, and the environment will provide vital information on how the brain responds to real-world stimuli, develops and ages, and adapts to pathology. Modelling together numerous streams of dynamic data, however, presents sizable methodological challenges. In this paper, we present a novel workflow and sample interpretation of a data set incorporating brain, behavioural, and stimulus data from a music listening study. We use hidden Markov modelling (HMM) to extract state timeseries from continuous high-dimensional EEG and stimulus data, estimate timeseries variables consistent with HMM from continuous low-dimensional behavioural data, and model the multi-modal data together using partial least squares (PLS). We offer a sample interpretation of the results, including a discussion on the limitations of the currently available tools, and discuss future directions for dynamic multi-modal analysis focusing on naturalistic behaviours.

Tipping detection using climate networks.

The development of robust Early Warning Signals (EWSs) is necessary to quantify the risk of crossing tipping points in the present-day climate change. Classically, EWSs are statistical measures based on time series of climate state variables, without exploiting their spatial distribution. However, spatial information is crucial to identify the starting location of a transition process and can be directly inferred by satellite observations. By using complex networks constructed from several climate variables on the numerical grid of climate simulations, we seek for network properties that can serve as EWSs when approaching a state transition. We show that network indicators such as the normalized degree, the average length distance, and the betweenness centrality are capable of detecting tipping points at the global scale, as obtained by the MIT general circulation model in a coupled-aquaplanet configuration for CO2 concentration-driven simulations. The applicability of such indicators as EWSs is assessed and compared to traditional methods. We also analyze the ability of climate networks to identify nonlinear dynamical patterns.

Multi-domain analysis of ultra-short-term HRV for breathing pattern classification in wearable health devices

This research paper introduces an innovative approach to classify heart rate variability (HRV) time series into paced and spontaneous breathing patterns to reflect changes in the autonomic nervous system. This type of classification is beneficial in wearable devices for stress/relaxation level detection and in deciding therapeutic interventions. The “Multi-Domain Approach” methodology integrates three different techniques: standard HRV features, fuzzy recurrence plot (FRP)-based FRP_GLCM, and empirical mode decomposition-based IMF_FRP_GLCM. The study concentrates on analyzing HRV time series within shorter data segments, aligning with the requirements of contemporary wearable health devices and biofeedback systems. HRV data collected during spontaneous and slow-paced breathing were analyzed across data segments of 5, 4, 3, 2, and 1 min, incorporating feature selection and reduction methods. Results demonstrated that standard HRV features yielded optimal performance for 5-min segments, achieving an average accuracy of 90%. Interestingly, IMF_FRP features achieved comparable accuracy even for 1-min segments. As segment duration decreased, standard HRV feature accuracy declined while IMF_FRP accuracy stayed intact, eventually matching 5-min segment accuracy levels. The study underscores the surging demand for shorter data segment HRV analysis, driven by advancements in wearable smart watches technology and mobile applications for monitoring health and managing stress.

Predicting early mortality in hemodialysis patients: a deep learning approach using a nationwide prospective cohort in South Korea

Early mortality after hemodialysis (HD) initiation significantly impacts the longevity of HD patients. This study aimed to quantify the effect sizes of risk factors on mortality using various machine learning approaches. A cohort of 3284 HD patients from the CRC-ESRD (2008–2014) was analyzed. Mortality risk models were validated using logistic regression, ridge regression, lasso regression, and decision trees, as well as ensemble methods like bagging and random forest. To better handle missing data and time-series variables, a recurrent neural network (RNN) with an autoencoder was also developed. Additionally, survival models predicting hazard ratios were employed using survival analysis techniques. The analysis included 1750 prevalent and 1534 incident HD patients (mean age 58.4 ± 13.6 years, 59.3% male). Over a median follow-up of 66.2 months, the overall mortality rate was 19.3%. Random forest models achieved an AUC of 0.8321 for first-year mortality prediction, which was further improved by the RNN with autoencoder (AUC 0.8357). The survival bagging model had the highest hazard ratio predictability (C-index 0.7756). A shorter dialysis duration (< 14.9 months) and high modified Charlson comorbidity index scores (7–9) were associated with hazard ratios up to 7.76 (C-index 0.7693). Comorbidities were more influential than age in predicting early mortality. Monitoring dialysis adequacy (KT/V), RAAS inhibitor use, and urine output is crucial for assessing early prognosis.

Deep learning-based time series forecasting

With the advancement of deep learning algorithms and the growing availability of computational power, deep learning-based forecasting methods have gained significant importance in the domain of time series forecasting. In the past decade, there has been a rapid rise in time series forecasting approaches. This paper comprehensively reviews the advancements in deep learning-based forecasting models spanning 2014 to 2024. We provide a comprehensive examination of the capabilities of these models in capturing correlations among time steps and time series variables. Additionally, we explore methods to enhance the efficiency of long-term time series forecasting and summarize the diverse loss functions employed in these models. Moreover, this study systematically evaluates the effectiveness of these approaches in both univariate and multivariate time series forecasting tasks across diverse domains. We comprehensively discuss the strengths and limitations of various algorithms from multiple perspectives, analyze their capacity to capture different types of time series information, including trend and season patterns, and compare methods for enhancing the computational efficiency of these models. Finally, we summarize the experimental results and discuss the future directions in time series forecasting. Codes and datasets are available at https://github.com/TCCofWANG/Deep-Learning-based-Time-Series-Forecasting.

Detecting infrastructure hazard potential change by SAR techniques on postseismic surface deformation: A case study of 2016 Meinong earthquake in southwestern Taiwan

Over the past two decades, multi-temporal InSAR techniques have been successfully utilized to monitor surface deformation throughout earthquake cycles. As sensor resolution and observation frequency of radar satellites have significantly improved, the spatial monitoring capability has expanded from tectonic to regional and local scales. Although regions with higher deformation rates are generally considered to have greater hazard potential, from an engineering perspective, locations with abrupt displacement rate changes may pose greater risks. This paper presents a robust method to transform tectonic-scale geodetic data into engineering-scale deformation tolerance ratios for assessing infrastructure safety during the postseismic period. Using the persistent scatterer InSAR (PSI) method, we measured line-of-sight (LOS) displacements from Sentinel-1 satellite images. By correcting LOS velocities with GNSS data and constraining the velocity in the N-S direction, we inverted the 2D (E-W and UD) postseismic deformation rates for the three years following the 2016 Meinong earthquake. The 2D rates were then converted into annual deformation tolerance ratios (ADTR), with a statistically established threshold to identify infrastructure segments with high hazard potential. Additionally, we conducted time-series variation analysis on the inherent pixel unit to evaluate the state of hazard potential and characterize the spatiotemporal behavior of fault systems during the postseismic period. Our results indicated that the postseismic deformation rates in the E-W and UD directions are 1.5 and 2–3 times higher, respectively, than those in the interseismic period. The horizontal ADTR index of high-speed railway infrastructures identified a high-hazard potential segment located on the Chegualin fault in Kaohsiung, while the vertical ADTR index highlighted a high-hazard potential segment at a junction between multiple fault extensions in Tainan. Moreover, statistical time-series variation analysis revealed a gradual decrease in hazard potential after 2017, which is related to a slow earthquake that occurred in early to mid-2018, modulating regional stress during the postseismic period. Our novel method not only highlights the impact of postseismic deformation on infrastructure but also provides a crucial basis for future earthquake hazard prevention and mitigation.