Examining resilience of taxi ridership during rainstorms and its non-linear associations with built environment

Examining resilience of taxi ridership during rainstorms and its non-linear associations with built environment | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Examining resilience of taxi ridership during rainstorms and its non-linear associations with built environment Jixiang Liu, Hongyu Wu, Longzhu Xiao, Bo Wang This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8938756/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract In the era of climate change, increasingly intense and frequent extreme weather events pose grave threats to urban life. Being a critical part of urban transportation system, taxis can facilitate understanding of human mobility and its variations during extreme weather events. Nevertheless, limited studies have delved into how taxi ridership responds to and recovers from the shocks induced by such events. Using Xiamen, China, as the case and employing random forest method, this study delves into resilience of taxi ridership and its non-linear relationships with built environment factors in the face of a rainstorm. Results show that (1) distance to city center, building density, and property price (as a proxy for socioeconomic features) are the most important contributors; (2) all the independent variables have salient non-linear effects on taxi ridership resilience, and obvious threshold effects exist; and (3) synergistic effects exist between certain independent variables, such as population density and land use mix. These findings can provide a solid knowledge base for formulating and executing nuanced intervention strategies for resilience promotion. Travel behavior resilience taxi ridership built environment non-linearity machine learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 1. Introduction The recent years have witnessed the increasing frequency and intensity of diverse extreme weather events against the background of global climate change (Utsumi and Kim 2022 ). Among those events, rainstorms have particularly adverse impacts on a wide variety of aspects of cities, including infrastructures, energy supply, built environment, economic vibrancy, human behavior, and safety, health, and well-being of people (Liu et al. 2025 , Wang et al. 2020 ). For example, a phenomenal rainstorm brought about by the Hurricane Harvey attacked north Texas coast in late August 2017, inundating almost one third of Houston at one point (LeComte 2020 ). The rainstorm that happened on July 20th, 2021 in Zhengzhou, China, led to a serious flood and caused 380 deaths and a direct 40-billion-yuan economic loss (Tang et al. 2023 ). In the end of July 2023, a 1-in-100-year record-breaking rainstorm hit Beijing, China (Huang et al. 2025 ), and caused more than 33 people to be dead and 18 to be missing. Thereby, it is both essential and meaningful to uncover how cities are impacted by and recover from rainstorms, thus offering knowledge base for targeted policy interventions. Taxis are an integral component of the urban transportation system. As a transport mode being capable of providing 24-hour available and spatially flexible services, taxis have become an ideal solution for the first/last mile transportation problem and a reliable complement for the public transport system (Zhu et al. 2024 , Zhu et al. 2017 ). On the other hand, as a typical type of surface transportation, taxis are notably susceptible to extreme weather events like rainstorms. For example, Chen, Liu and Gao ( 2024 ) revealed that in Wuhan, China, the taxi volume dropped averagely by 4.16% weekly due to rainfall. Taxi ridership in New York City was found to decline during adverse weather conditions (including heavy rain and strong wind), particularly in the weekend and at night (Bian, Wilmot and Wang 2019 ). In the case of Nanjing, China, Zhang et al. ( 2019 ) uncovered that taxis are more sensitive to rainstorms than other non-private transportation modes like buses. Nevertheless, some researchers have gained some counter-intuitive insights into the effects of rainstorms on taxi travel behavior. For instance, Lepage and Morency ( 2021 ) discovered the positive effects of rainfall on taxi ridership; in their study in Montreal, Canada, four consecutive hours of rains had been found to increase taxi ridership by 13.9%. Likewise, Qiao, Haraguchi and Lall ( 2025 ) found in New York City that taxi ridership increased throughout the extreme rainfall events, demonstrating the role of taxis for urban resilience. Hence, it is essential to take a deeper look into how taxi ridership responds to rainstorms. Resilience, as mentioned above, has become an increasingly popular approach to delving into the variations and recovery of travel behavior in the face of extreme weather events. In the field of transportation, resilience can be defined as the capacity of a transportation system to stand a disruptive event and rebound back to normal operational condition in an acceptable amount of time (Chan and Schofer 2016 , da Mata Martins, da Silva and Pinto 2019 ). Since transportation system can be divided into the supply part (e.g., facilities, infrastructures, networks) and demand part (transportation users), transportation resilience has also been interpreted from the perspectives of supply and demand (Liu et al. 2025 ). Overall, research regarding transportation resilience from the supply perspective has predominates, while academic explorations on transportation resilience from the demand perspective (i.e., resilience of people’s travel behavior patterns during disruptions, referred to as travel behavior resilience hereafter) are still limited. For example, in their case study in Kunming, China, Wang et al. ( 2022 ) first defined travel behavior resilience and measured it with variations of public transit trips during COVID-19. Using Chengdu, China as an example, Peng et al. ( 2024 ) evaluated resilience of ride-hailing behavior and examined its relationships with the built environment. Moreover, despite the irreplaceable role that taxis play in the transportation system and the ability of taxi ridership of reflecting human mobility, seldom has resilience of taxi ridership and its influencing factors been examined, particularly against the backdrop of extreme weather events like rainstorms. Built environment has widely been confirmed as a key influencing factor of taxi travel behavior. Since the introduction of D’s framework by Ewing and Cervero ( 2001 ) and Ewing and Cervero ( 2010 ), researchers have established strong connections between taxi ridership and multiple dimensions of built environment, such as density, diversity, design, destination accessibility, and distance to transit, among others (Chen et al. 2021 , Li et al. 2024 , Zhu et al. 2022 , Zhu et al. 2024 , Yang et al. 2022 ). Meanwhile, recent years have also witnessed increasing academic attention towards the role of built environment in affecting urban resilience. As revealed, built environment with certain features (e.g., multi-centricity from a macro perspective and higher land use diversity and medium-level population density from a meso or micro perspective) are indeed beneficial for urban resilience (Hao and Wang 2022 , Irajifar, Sipe and Alizadeh 2016 , Sharifi 2019a , Sharifi 2019b , Wang et al. 2024 ). Xiao, Wei and Wu ( 2022 ) and Peng et al. ( 2024 ) are among the pioneering studies that examined the effects of built environment on resilience of transportation from the demand perspective (i.e., travel behavior resilience), and the latter even corroborated the non-linearity in the relationships. However, knowledge regarding how built environment influences resilience of taxi ridership is still scarce. Given the widely confirmed non-linear effects of built environment on travel behavior and the potential synergistic effects among built-environment variables (Xiao and Wei 2023 , Yang et al. 2024 , Liu et al. 2023 ), disentangling the complicated relationships between built environment and resilience of taxi ridership is essential for providing evidence base to promote travel behavior resilience. Therefore, employing Xiamen Island as the study area, this study attempts to examine the non-linear relationships between built environment and resilience of taxi ridership, using an advanced machine learning method, i.e., random forest. It aims to answering three research questions: (1) Which are the most important built-environment variables for resilience of taxi ridership? (2) Do the built-environment variables have non-linear effects on taxi ridership resilience? (3) Are there any synergistic effects among the built-environment variables? This study contributes to the literature in the following three aspects. First, utilizing the taxi trip dataset in Xiamen Island, it scientifically quantifies resilience of taxi ridership facing a typical type of extreme weather events, i.e., rainstorm. Second, through revealing the hierarchy of relative contributions of built-environment variables, it clearly demonstrates the most important influencing factors of resilience of taxi ridership. Third, by visualizing the complicated relationships between built environment and resilience of taxi ridership without pre-defining the relationship shapes, it offers refined recommendations for policymakers to formulate and implement tailored measures to enhance transport resilience. The remainder of the paper unfolds as follows. Section 2 reviews the relevant literature. Section 3 introduces the study area, data, variables, and modeling methods. Section 4 presents and analyzes the research results. Section 5 discusses the findings and Section 6 concludes the study. 2. Literature review 2.1 Relationships between built environment and taxi ridership The built environment has long been established as a fundamental determinant of travel behavior (Ewing and Cervero 2010 ). Among the array of travel modes, taxi constitutes a vital segment of urban transport systems, exhibiting notable spatiotemporal variation closely tied to the characteristics of the built environment (Cervero 2013 , Liu, Ding and Chen 2020b , Qian and Ukkusuri 2015 , Chen et al. 2021 ). To systematically capture these attributes influencing taxi ridership, the dominant analytical framework employs the widely accepted “5Ds” framework, which encompasses density, design, diversity, destination accessibility, and distance to transit (Ewing and Cervero 2010 , Ewing and Cervero 2001 ). Technological advances, especially the widespread application of high-resolution positioning tools such as GPS, have enabled the large-scale and fine-grained collection of taxi trajectory data, including pick-up and drop-off points, travel durations, and distances (Sun, Zhang and Shen 2018 , Zhang et al. 2020 ), thereby laying a solid empirical foundation for analyzing built environment–taxi mobility interactions. A growing body of empirical evidence has revealed considerable heterogeneity in the effects of built environment features on taxi ridership, both in direction and magnitude (Zhu et al. 2024 ). Among these features, land-use mix consistently emerges as a key explanatory factor, with higher levels generally associated with increased taxi ridership. Land-use mix is commonly measured using entropy-based indicators that incorporate land-use categories, such as residential, commercial, office, and green spaces (Qian and Ukkusuri 2015 , Yu and Peng 2019 , Liu et al. 2020b ). In parallel, recent studies highlight the significant role of specific POI types—particularly financial and entertainment venues—in shaping spatial variation in taxi activity (Li et al. 2024 ). Beyond land-use mix, other built environment variables, such as population and employment densities, street network design, and transit infrastructure, have also exerted meaningful influence on taxi ridership. Empirical findings generally support positive associations with population density, employment concentration, and intersection density (Yang and Gonzales 2014 , Liu et al. 2020b , Zhu et al. 2022 ), while higher bus stop density tends to exhibit substitution effects, negatively correlating with taxi ridership (Liu et al. 2020b ). Furthermore, increased taxi ridership has been linked to prolonged bus headways, implying a modal shift under diminished transit service quality (Yang and Gonzales 2014 ). Interestingly, metro accessibility has been positively associated to taxi ridership (Qian and Ukkusuri 2015 ). Importantly, the impacts of these built environment variables are not strictly linear. A range of studies have identified threshold effects—wherein the marginal influence of a variable plateaus or shifts upon reaching critical values (Chen et al. 2021 ). Moreover, these effects are marked by considerable spatiotemporal heterogeneity, with varying influences across different temporal, spatial, and seasonal dimensions (Zhu et al. 2024 , Zhu et al. 2022 , Chen et al. 2021 ). For instance, road network density exerts a more pronounced influence in southern urban regions with less-developed infrastructure, and taxi ridership tends to spike during weekdays and peak hours (Liu et al. 2020b ). Seasonal variations has also been documented, with warmer months (e.g., autumn) witnessing elevated taxi demand and rainfall demonstrating a positive but non-linear relationship with taxi ridership (Liu et al. 2020b ). Such findings underscore the multifaceted and context-dependent nature of taxi behavior in urban settings (Zhu et al. 2022 ). Nevertheless, current scholarship has predominantly examined these relationships under typical, undisturbed conditions. Much less attention has been devoted to how the built environment affects the variations of travel behavior during disruptive events, such as extreme weather events. Incidents like heavy rainfall can severely compromise transport system functionality, prompting deviations from routine travel patterns. Under such conditions, analyzing the role of built environment features in facilitating mobility recovery offers critical insights into their capacity to foster transport resilience. 2.2 Travel behavior resilience and its built-environment influencing factors The concept of “resilience” was originally introduced by Holling ( 1973 ), denoting the capacity of an ecosystem to absorb shocks while retaining their essential structure and function. As scholarly paradigms evolved, the notion of resilience expanded into urban contexts, culminating in the widely discussed framework of “urban resilience” (Tongyue, Plnyi and Chaolin 2014 ). In the face of mounting uncertainties, including climate change, economic instability, and infrastructural strain, cities are increasingly viewed as complex adaptive systems whose resilience encompasses both physical robustness and social adaptability (Huang and Wang 2024 , Meerow, Newell and Stults 2016 ). Recent advances in resilience research advocate a dual-perspective approach, encompassing both supply-side and demand-side dimensions, aiming to systematically capture the diverse sources of recovery capacity embedded within the built environment (Huang and Wang 2024 ). On the supply side, resilience emphasizes the structural integrity and recovery potential of physical infrastructure. Numerous studies have demonstrated that urban resilience is significantly influenced by macro-scale factors (e.g., landscape connectivity) and meso-scale factors (e.g., land-use mix). Specifically, compact and polycentric urban forms, moderate density, higher land-use diversity, and fine-grained block design are frequently associated with enhanced resilience, with the interplay between land-use mix and density proving particularly influential (Sharifi 2019a , Sharifi 2019b ). In the context of Chinese cities, multi-source data assessments reveal that core urban zones in Guangzhou exhibit higher recovery capacities than peripheral new districts (Ruan, Chen and Yang 2021 ). Similarly, an evaluation of flood resilience in Nanjing indicates that since 2006, the relative influence of economic, institutional, cultural, and physical resilience components has markedly increased (Wang, Li and Zhang 2021 ). In addition, findings from Australia suggest that population density significantly enhances post-disaster recovery, with the effect displaying clear non-linear characteristics during flood events (Irajifar et al. 2016 ). Meanwhile, triggered by the emergence of disruptive events such as the COVID-19 pandemic, increasing scholarly attention has shifted toward the demand side of urban resilience. Here, the focus is on people’s behavioral adaptability and the heterogeneity of response across social groups (Olsson et al. 2015 ). Travel behavior, as a core component of daily human activities, is characterized by spatial dependence, spatiotemporal dynamics, and group heterogeneity. These features render it a critical lens for understanding the interaction and evolving mechanisms between travel demand and the built environment, as well as a pivotal entry point for research on transport resilience (Huang and Wang 2024 ). Emerging studies have begun to examine the resilience of travel behavior under various stressors. For example, during the COVID-19, the decline in public transit ridership has been attributed not only to physical attributes of the built environment (Hu and Chen 2021 , Khavarian-Garmsir, Sharifi and Moradpour 2021 ), but more fundamentally to the long-term influence of the built environment in shaping individual travel preferences, which in turn affects behavioral responses across different social groups (Gan et al. 2020 , Vergel-Tovar and Rodriguez 2018 ). Although land-use mix use theoretically enhances accessibility to destinations such as shops and schools, its practical benefit was reduced under pandemic due to widespread temporary closures (He et al. 2022 ). Furthermore, overly compact urban design may reduce building coverage and traffic infrastructure redundancy, exacerbating system vulnerability and negatively impacting travel resilience (Xiao et al. 2022 ). Social vulnerability compounds these effects, transit ridership dropped more sharply in neighborhoods with higher proportions of marginalized populations, illustrating disparities in resilience capacity (Xiao et al., 2022 ). Beyond pandemics, environmental disruptions like air pollution have also been investigated, a recent study employed the Gradient Boosting Decision Tree (GBDT) model and identified significant non-linear effects of built environment attributes on ride-hailing travel resilience, including population density, densities of bus and rail stations, and land-use mix (Peng et al. 2024 ). However, the bulk of existing research remains centered on social disruptions. Comparatively fewer studies have addressed resilience in the context of extreme weather events, which differ fundamentally from pandemics in their physical intensity, temporal immediacy, and impact mechanisms (Solomon 2007 ). These differences suggest that built environment variables may interact with travel behavior resilience in unique ways during weather-induced disruptions. As climate-related hazards become increasingly frequent and severe, there is a pressing need to understand how built environment can either mitigate or exacerbate their impacts. Investigating this nexus offers not only conceptual expansion of demand-side resilience theory, but also provides actionable insights for climate-adaptive urban mobility planning, which is a task of urgent theoretical and practical significance in the era of climate change. 3. Methodology 3.1 Study area and case of extreme weather event This study selects Xiamen Island as the case study area, i.e., the core urban district of Xiamen City. Covering approximately 158 km², Xiamen Island had a population of 2.067 million in 2023 and functions as the city’s political, economic, cultural, and transportation hub. Siming and Huli districts form Xiamen Island, being separated from the other four districts of Xiamen City (i.e., Haicang, Jimei, Tong’ an, and Xiang’ an Districts) by the sea (Fig. 1 ). Xiamen Island was chosen for this study for several reasons. First, its high population density and thriving tourism industry generate intense transportation demand. The city's scenic landscapes and cultural heritage attract large numbers of tourists, leading to concentrated travel needs from both residents and visitors. Second, Xiamen Island has a highly compact and high-density built environment with a multifunctional urban structure. Major commercial centers (e.g., Zhongshan Road), residential areas, transportation hubs (e.g., Xiamen Railway Station), and key tourist attractions (e.g., Xiamen Botanical Garden) are all located within the island. Third, metro development in Xiamen has been relatively slow due to geographic constraints and population growth rate. With only a few operational metro lines, the city lacks a fully developed rail transit network. Consequently, surface transportation remains a dominant mode of travel, and taxis play a crucial role in meeting mobility demands. Lastly, Xiamen’s subtropical maritime climate brings an average annual rainfall of about 1,200 mm, peaking from May to August. Frequent heavy rainfall and extreme weather events further challenge the stability of the urban transportation system. These factors make Xiamen Island an ideal case for examining the relationship between the built environment and the resilience of taxi ridership. The findings of this study can offer insights for similar medium-sized cities with underdeveloped metro systems and frequent extreme weather events, supporting transportation planning and climate adaptation strategies. This study collected records of extreme rainfall events in Xiamen in recent years and selected the extreme rainfall event on May 7, 2018, as the case study. On this day, Xiamen experienced an exceptionally heavy rainstorm, with hourly precipitation exceeding 100 mm in multiple locations on Xiamen Island. 3.2 Data This study utilizes two types of data: taxi trip data and built environment data. The taxi trip data, provided by the Xiamen Transportation Bureau, includes key trip details such as the latitude and longitude of trip origins and destinations, trip distance, trip duration, pick-up time, and drop-off time, with an average of 130,000 taxi trips recorded daily. The built environment data, sourced from Xiamen University, encompasses land-use planning, road network data, building information, points of interest (POI), the locations of important public facilities and so on. All datasets were processed using ArcGIS 10.8 to ensure data standardization and usability. 3.3 Variables 3.3.1 Resilience of taxi ridership Resilience of taxi ridership serves as the dependent variable in this study, and the Resilience Triangle framework is employed to quantify the capacity of taxi ridership to withstand and recover from extreme weather events. The Resilience Triangle has widely been utilized in previous studies (Wang et al. 2022 ) as a framework for describing system performance changes under external shocks (e.g., COVID-19) and assessing its response and recovery capacity. As illustrated in Fig. 2 , the performance curve (denoted as P(t)) represents how system performance (Y-axis) changes over time (X-axis) during a disruption event. Specifically, the Resilience Triangle consists of three key phases: 1) Pre-disruption Performance, indicating the system's stable operational level before the disruption occurs. 2) Decline Performance, denoting the degree of system deterioration and its vulnerability during an external shock. 3) Recovery Phase, i.e., the process by which the system restores itself from the post-disruption low point to normal conditions. The area of the Resilience Triangle serves as an indicator of the resilience of the system—a larger area signifies a longer recovery period and lower resilience, whereas a smaller area indicates a stronger recovery capability and higher resilience. This study assesses the resilience of taxi ridership by examining residents' travel behavior during the extreme rainstorm event on May 7, 2018, in Xiamen. To ensure spatial consistency in the analysis, the entire Xiamen Island was divided into 500 m × 500 m grid cells. The calculation of taxi ridership resilience follows a three-step process. First, the baseline (𝑃_baseline) was established by calculating the hourly average taxi pick-up and drop-off counts for each grid based on the five Mondays preceding the rainstorm (i.e., April 2, 9, 16, 23, and 30, 2018). This ensured a reliable reference for assessing disruptions. Second, the maximum loss (impact and performance decline) was determined by identifying the minimum taxi ridership (𝑃_min) under the rainstorm, comparing hourly taxi demand with the baseline. To account for spatial heterogeneity, the study measured the relative change proportions instead of absolute values. Third, the recovery time was identified using the Plateau Detection Algorithm, which detected the stabilization point (𝑡3) based on a rolling variance tolerance of 0.15 over 12 consecutive hours. Finally, based on the Resilience Triangle framework, the resilience value of taxi ridership for each grid was calculated using Eq. ( 1 ): $$\:Resilience=\:\frac{1}{{\int\:}_{t1}^{t3}{P}_{baseline}-{\int\:}_{t1}^{t3}{P}_{min}}\approx\:\frac{2}{\left({P}_{baseline}-{P}_{\text{m}\text{i}\text{n}}\right)*\left(t3-t1\right)}$$ 1 ; 3.3.2 Independent variables This study derives independent variables from two dimensions: socioeconomic attributes and built environment characteristics. To ensure spatial consistency, a 500 m × 500 m grid unit is used for calculation. Table 1 provides detailed definitions and statistical information for each independent and dependent variable. The built environment variables are measured based on the classic “5Ds” model (Ewing and Cervero 2001 , Ewing and Cervero 2010 ), including density (population density, job density), diversity (land use mix index), design (intersection density), public transport accessibility (bus route density), and destination accessibility (distance to city center). Additionally, socioeconomic variables are represented by property prices in 2018, serving as an indicator of residents’ economic status. Table 1 Descriptive statistics of dependent and independent variables Variables Descriptions Mean (S.D.) Dependent variable Resilience Resilience of taxi ridership measured using Resilience Triangle. 2.17 (1.43) Independent variables Socioeconomics Property price Property price in 2018 (Yuan/m 2 ). 48944.10 (9200.16) Built environment Distance to CBD Network distance to Zhongshan Road (km). 13.89 (14.11) Population density Population density index based on Baidu Heat Map. 263.67 (185.45) Building density Total area of buildings per km 2 (m 2 ). 1670.25 (1254.54) Job density No. of employment positions per km 2 . 8237.98 (16100.95) Road density Road length per km 2 (km). 12.93 (16.82) Intersection density No. of three-or-more-way intersections per km 2 . 16.70 (20.21) Bus route density No. of bus routes per km 2 . 48.76 (84.65) Land-use mix The degree to which different land-use types are integrated within the grid. In this study, we identified twelve land-use types, i.e., industry, residence, business and commerce, urban village, green land and public open space, transportation and logistics, education and research, spatially designated land, culture and sports, municipal administration, medicine, and others. 0.58 (0.21) 3.3.3 Methodology: random forest This study employs the random forest method to investigate the relationships between built environment and taxi ridership resilience. Originally proposed by (Breiman 2001 ), random forest is a frequently used ensemble learning method applicable to both classification and regression tasks (Yan, Liu and Zhao 2020 ). It consists of three main components: data sampling, decision tree training, and ensemble learning. First, given a training dataset \(\:D=\{\left({X}_{1},{Y}_{1}\right),\left({X}_{2},{Y}_{2}\right),\dots\:,\left({X}_{N},{Y}_{N}\right)\) }, \(\:k\) different training subsets \(\:{D}_{1},{D}_{2},\dots\:,{D}_{\text{k}}\) are generated using bootstrap sampling, where each subset \(\:{D}_{i}\) consists of \(\:N\) samples with replacement. The randomness in sample selection helps mitigate the overfitting issue. Second, \(\:\:k\) decision trees are trained, with each decision tree \(\:{T}_{i}\) using only its corresponding training subset \(\:{D}_{i}\) . At each one, a random subset of \(\:m\) features (typically \(\:m=\sqrt{M}\) , where \(\:M\) is the total number of features) is selected for splitting. Training continues until a stopping criterion, such as the maximum tree depth, is reached. The model’s performance is influenced by three key parameters: (1) tree depth, which affects individual tree performance; (2) the number of selected features, which controls correlations among the trees; and (3) the number of trees, which determines the forest size. Finally, during prediction, each tree produces an independent output, and the final result is obtained by averaging (for regression) or majority voting (for classification). In this study, as a regression task, the final prediction is calculated as: $$\:\widehat{Y}=\frac{1}{k}\sum\:_{i=1}^{k}{y}_{i}$$ 2 ; The random forest method offers several advantages. It effectively reduces variance compared to single decision trees, performs well on high-dimensional data, and provides feature importance evaluation (Menze et al. 2009 , Hastie et al. 2009 ). It is robust to missing values and outliers, does not require data standardization (Liaw and Wiener 2002 ) and does not assume a predefined relationship between independent and dependent variables, making it suitable for modeling complex non-linear patterns (Ho 1998 ). Additionally, its parallel processing capability enhances computational efficiency (Geurts, Ernst and Wehenkel 2006 ). However, random forest has some limitations. Its complexity makes interpretation challenging compared to single decision trees (Louppe 2014 ). It is also sensitive to highly correlated features, as redundant variables may be repeatedly selected (Strobl, Malley and Tutz 2009 ). Moreover, it lacks statistical inference capabilities, such as significance tests or confidence intervals, which are common in traditional statistical models (Wright, Ziegler and König 2016 ). Despite these limitations, random forest remains a powerful tool for analyzing complex relationships and is increasingly used in spatial and transportation research. 4. Results 4.1 Model fit specifics For examining and enhancing the generalizability of the random forest models, 70% of the grids were randomly selected to act as the training set, while the remaining 30% as the test set. Moreover, we tuned three key parameters, i.e., number of trees, number of splitting variables, and maximum depth of tree, to obtain optimal model performance and mitigate the overfitting issue. After tuning, the best number of trees turned out to be 500, and number of splitting variables and maximum depth of tree were 6 and 1000, respectively. The root mean square error (RMSE) and pseudo-R 2 of the final model are 11.343 and 0.365, indicating a satisfactory performance. The final model provides a basis for quantifying the relative importance of independent variables and drawing the partial dependence plots. 4.2 Relative importance of independent variables In machine learning, the relative importance quantifies the contribution of individual variable to the model's predictions, facilitating feature selection, model interpretation, and optimization. As shown in Table 1 , the relative importance of independent variables in predicting resilience of taxi ridership is ranked by contribution magnitude, with all values normalized to sum to 100%. Overall, the built environment plays a significantly greater role in the resilience of taxi ridership than the single socio-demographic factor, i.e., property price. Among built environment factors, distance to the city center is the most influential predictor of taxi ridership resilience, with a relative importance share of 41.05%, far surpassing other variables. This finding is expectable, as it reflects regional differences in travel demand, influences transportation accessibility, and shapes travel mode choices, all of which affect resilience of taxi ridership. Similarly, building density plays a crucial role (with a relative importance of 23.43%). Building density closely correlates with infrastructure completeness, traffic demand fluctuations, and mobility conditions, and hence is a key determinant of taxi ridership resilience. Other built environment variables, such as land use mix (3.61%), road density (2.75%), bus route density (2.47%), and intersection density (2.00%), only have marginal contributions to the model. Property price ranks third in relative importance. This is understandable given that property price serves as a vital indicator of residents’ income levels and directly influences their flexibility in choosing travel modes. Particularly under extreme weather conditions, higher-income residents are more likely to opt for taxis as a convenient mode of transport, thereby affecting the resilience of taxi ridership (Pan, Shen and Zhao 2013 ). Table 1 Relative importance of independent variables. Independent variables Relative importance (%) Ranking Socio-demographics Property price 10.63% 3 Built environment Distance to city center 41.05% 1 Building density 23.43% 2 Job density 7.78% 4 Population density 6.27% 5 Land use mix 3.61% 6 Road density 2.75% 7 Bus route density 2.47% 8 Intersection density 2.00% 9 Sum 89.37% 4.3 Non-linear relationships between built environment and resilience of taxi ridership Partial Dependence Plot (PDP) is a widely used visualization tool in machine learning that illustrates the marginal effects of individual features on model predictions. By holding all other variables constant, PDP reveals the unconstrained non-linear relationship between independent and dependent variables. Figure 3 to Fig. 7 display the partial dependence plots of key independent variables in this study, where the Y-axes represent resilience of taxi ridership, and the X-axes denotes the values of a specific dependent variable. Overall, all variables exhibit non-linear relationships with resilience of taxi ridership, characterized by evident threshold effects. Figure 3 illustrates the non-linear relationship between property price and resilience of taxi ridership, revealing a distinct threshold effect in an “N”-shaped pattern. Specifically, when property prices range from 30,000 to 40,000 (RMB/m²), the resilience of taxi ridership exhibits a sharp increase. Growth then slows, followed by a decline that reaches its lowest point at 60,000 (RMB/m²), before gradually rebounding. Property prices can generally indicate income and affordability of people, thus indirectly affecting their transport resources and life patterns. Thereby, people reside or work in areas with relatively low property prices (in our case, areas with property prices lower than 35,000) are probably more reliable on cheaper transport modes, such as walking, electric bikes, and public transit, and their tendency of taking taxis can be quite susceptible to extreme events like rainstorms. This finding aligns with our expectation and is consistent with Hong et al. ( 2021 ), which suggests that socio-economic factors significantly influence resilience of taxi ridership, with lower-income groups facing greater challenges during disasters such as hurricanes due to limited resources. In terms of residents of areas with high property prices (higher than 55,000), they may have more flexible schedules and daily routines than others, and thus they can alter their travel behaviors more freely in accordance to disturbing events. These can explain the relatively low resilience of taxi ridership in the areas with low and high property prices. Figure 4 visualizes the correlations between resilience of taxi ridership and two built environment variables, i.e., distance to the city center and building density. Both the correlations are generally positive, albeit to distinct thresholds. When distance to the city center is between 0 and 5 km, resilience of taxi ridership remains low and stable with no significant fluctuations. This finding is interesting and yet understandable. This study identifies Zhongshan Road as the city center of Xiamen. Zhongshan Road is indeed the birthplace of Xiamen City. However, in recent years, Zhongshan Road has increasingly evolved into a tourism and commerce center of Xiamen. Since touring and shopping are two typical types of non-compulsory travel purposes, during extreme weather events, tourists and shoppers can easily modify their travel behaviors, which can partly explain the lower resilience of taxi ridership in areas near Zhongshan Road. Meanwhile, urban central areas are oftentimes characterized with more easily accessible transportation modes, such as public transit, which can effectively substitute taxis in the face of extreme events. Moreover, these areas may be more congestion prone. These can also contribute to the relatively low resilience of areas with proximity to city center. Yet, once the distance exceeds 5 km, the resilience of taxi ridership sharply increases, stabilizing around 10 km. At approximately 14 km from the city center, resilience of taxi ridership begins to decline gradually. This is consistent with the finding (Zhang and Li 2020 ), which show that regions farther from the city center are less affected by heavy rainfall across different types of functional areas. The relationship between building density and resilience of taxi ridership is relatively straightforward, presenting a nearly linear shape when building density is below 2000 (m²/km²) and stabilizing with minimal fluctuations afterwards. Low building density areas typically feature low-level development and more dispersed urban functions, thus resulting in fewer transportation demand. In contrast, areas with higher building density can usually generate and attract more stable taxi demands. Figure 5 shows the non-linear relationship between population density, job density, and the resilience of taxi ridership. Both variables exhibit a positive correlation with resilience of taxi ridership. Specifically, when population density is below 390 per m², resilience of taxi ridership increases almost monotonically. Afterwards, taxi ridership stays stable until around 600 per m² where it experiences a slight increase and keeps steady. It is worth noting that data is sparse when population density exceeds around 400, and thus the results obtained therein can be less reliable. When job density is below approximately 4,000, resilience of taxi ridership undergoes a sharp growth; afterwards, the effects of job density on taxi ridership density seem to be unimportant. The positive effects of population and job density on taxi ridership density align with our expectation and prior studies (Hao and Wang 2022 , Irajifar et al. 2016 ), since high population or job density mean high and stable transportation demand. However, areas with high population and/or job density are usually characterized with high traffic flow and thus are easily congested. When population and/or job density arrive at certain thresholds, the negative effects of resulting congestion may outrun the positive effects of compacity, particularly during rainstorms. Figure 6 depicts the associations between land use mix, road network density, and taxi ridership resilience. Generally, the former has positive impacts, while the latter negative ones. In the range from 0 to around 0.5, despite some fluctuations, land use mix has a nearly linear relationship with taxi ridership resilience. This finding is reasonable because higher land use mix indicates that diverse urban functions are more closely integrated, which can generate lasting transportation needs and thus higher taxi ridership resilience. After that, the relationship generally stays positive, but the increasing rate gets smaller. When road density is within the range from 0 to 17 (km/km²), resilience of taxi ridership fluctuates slightly at the high level. However, once road density drops below 17, resilience of taxi ridership declines sharply and then stabilizes at around 21 (km/km²). In areas with low road network density, due to the lack of transportation options, the demands for taxis can be relatively high and stable, and those demands will probably last during extreme weather events, such as rainstorms, resulting in higher resilience of taxi ridership. By contrast, high-road-density districts can be characterized with diverse transportation alternatives on the one hand, and more susceptible to congestions, especially in the face of extreme weather events, on the other hand. Hence, the taxi demands therein can be lower and more unstable, aligning with prior findings (Liu et al., 2020). As shown in Fig. 7 , bus route density and road intersection density have similar increase-then-decrease associations with resilience of taxi ridership. In the range of 0 to 120 (routes/km 2 ), bus route density has nearly linear positive effects of taxi ridership; after that, resilience of taxi ridership declines slowly as bus route density grows. This is an interesting finding, which shows that within certain threshold, increasing public transit coverage can also lead to more resilient taxi demands. Higher public transit coverage can enhance the transportation accessibility of a certain area, and hence bring about more human activities and thus higher and more steady taxi demand, i.e., higher taxi ridership resilience. However, exceeding certain threshold, the substitution effects of public transit together with the resulting congestion may prevail. Similarly, when intersection density is in the range of 0 to 25 (per km 2 ), taxi ridership climbs dramatically; afterwards, it declines with some fluctuations. Intersection density indicates the connectedness of an area. Higher the intersection density, more connected the area. Therefore, the positive relationship between intersection density and taxi ridership resilience is understandable. Yet, excessively high intersection density may lessen the traffic efficiency and probably lead to congestions. 4.4 Synergistic effects between built environment and resilience of taxi ridership A partial dependence plot can clearly demonstrate the non-linear association between dependent variable and a certain independent variable with other variables are controlled. However, it is widely confirmed that independent variables can interact with one another to strengthen each other’s effects, resulting synergistic effects. That is to say, when certain variables change together, their combined effects can be larger than the mere sum of their individual effect. We use two-dimensional partial dependence plots (2D-PDPs) to present the synergistic effects between several representative built-environment variables, as shown in Fig. 8 . In the figure, the X-axis and Y-axis represent the variables that interact with each other, while the color gradient illustrates taxi ridership resilience, with yellow indicating higher resilience and dark blue indicating lower resilience. As shown, most synergistic effects are achieved when the two variables are in their high-value ranges at the same time. For example, when distance to the city center is within the range from 10 to 15 km, the higher building density (or population density), the higher resilience of taxi ridership. Similarly, when population density and land use mix are in their higher range (above 600 per hm 2 and 0.4 respectively) at the same time, the areas are characterized with higher taxi ridership resilience. However, inconsistencies exist. When property price is in its low to medium range (35,000 to 52,000 Yuan/m 2 ), the higher building density, the higher taxi ridership resilience. When land use mix is at a higher level and yet intersection density at a low level, resilience of taxi ridership is high. 5. Discussion Under the backdrop of increasing intensity and frequency of extreme weather events brought about by climate change, policymakers, researchers, and practitioners have been striving to formulate countermeasures, among which mitigation and adaptation are two types of widely adopted strategies (Wang et al. 2023 ). Comparatively, adaptation strategies, i.e., learning to cope with the impacts of climate change and the consequent extreme weather events (like rainstorms), decreasing vulnerability, and enhancing resilience, are more urgent and time- and resource-saving. Taxis being an important part of urban transportation system, resilience of taxi ridership can effectively epitomize the variations of travel behavior of urban residents and resisting and recovering capacity of transportation facilities and infrastructures. Taking a deeper look into taxi ridership resilience and its influencing factors can provide evidence base for promoting transportation resilience and better satisfying urban residents’ transportation demand during disruptive weather events. Prior studies have established the associations between taxi ridership and built environment features and socioeconomic factors (Lyu et al. 2023 , Zhu et al. 2024 , Chen et al. 2021 ). However, how taxi ridership is affected by and recover from extreme weather events has been largely ignored. This study, to the extent of our knowledge, is among the first ones to examine taxi ridership resilience in the face of such disruptive events and its relationships with built environment. We obtain some interesting findings in this study. First, we find that among the influencing factors, distance to city center, building density, and property prices are the most important contributors of resilience of taxi ridership. Distance to city center, as a frequently used indicator of location, has been confirmed as an all-important determinant of diverse travel behavior dimensions, such as transport mode choice, commuting duration, and driving distance (Liu and Xiao 2024 , Ding, Cao and Næss 2018 , Liu and Xiao 2023 ). Our study further corroborates the vital contribution of location for travel behavior resilience, implying that we can promote resilience during extreme weather events through adjusting regional spatial structure and enhancing regional connectivity. Interestingly, building density is found to be a much more important variable for taxi ridership resilience than population density, which can be due to that there may be neither high nor stable taxi demands in some areas characterized with high population density, e.g., factory dormitory and urban village (Liu et al. 2020a ). This finding suggests that development intensity is also a feasible intervention target for promotion of taxi ridership resilience. Property price plays a critical role, which is expectable since property price could not only suggest the affordability and travel resources and patterns of residents in certain areas, but also closely relate with the location, land use compositions, and physical environment quality. The hierarchy of relative importance of the independent variables remind the policy makers that intervention measures on such built environment dimensions as regional accessibility/connectivity, development intensity, and urban functional diversity can be more efficient. Second, we observe that all the influencing factors selected in this study are associated with resilience of taxi ridership in a non-linear pattern, with obvious thresholds present, which provides fruitful theoretical and practical implications. Roughly, the shapes of the abovementioned associations can be divided into three categories, including increase-fluctuate-decrease, increase-fluctuate-increase, and fluctuate-decrease. Property price, distance to city center, bus route density, and intersection density belong to the first category, i.e., increase-fluctuate-decrease. Building density, population density, job density, and land use correlate with taxi ridership resilience mix in an increase-fluctuate-increase pattern, while road density belongs to the third category. Most of the relationships uncovered are in accordance with our expectations or findings of prior studies, in spite of some counterintuitive findings as we further discuss as follows. As revealed, areas with medium-to-high-level property price tend to be characterized with the highest resilience of taxi ridership. This finding can be understood from the perspectives of both people and space. Property price being a reliable indicator of income, people living in areas with medium-to-high property prices may have more fixed schedule than people living in high-property-price areas, while higher travel budgets than people in low-price areas. Meanwhile, as discovered by prior studies (Yang et al. 2020 , Yang, Zhou and Shyr 2019 , Xiao, Orford and Webster 2016 ), property prices are significantly associated with transportation accessibility and physical environment features; medium-to-high-property-price areas may have more agreeable accessibility and built-environment features for taxis. Also, areas with a distance of 10–12 km to the city center, rather than the city center itself, have the highest resilience of taxi ridership, suggesting that these areas are evolving into sub-centers with different functions from those of the city center we selected in this study (i.e., Zhongshan Road), and these functions have generated and attracted more stable taxi trips during rainstorms. These salient non-linear relationships, together with the thresholds, provide plentiful policy implications. For instance, the most effective range for population density is below around 380 (/hm 2 ), while those for building density and land use mix are below 2,000 (/km 2 ) and below 0.5, individually. It is best to control road density under 17.5 (km/km 2 ). Beyond these ranges, the interventions may be inefficient or even detrimental for resilience promotion. It is worth noting that these ranges may be highly dependent on the contexts, which, however, can enlighten the decision makers that caution should be taken when intervening with the built environment and that over-diversification and excessively high density should be avoided. Moreover, these thresholds and ranges provide transportation operators and urban residents insights regarding transportation resources allocation and individual travel decision-making. Specifically, during extreme weather events, taxi or ride-hailing companies can dispatch more vehicles to areas with certain built environment features (e.g., concentrating urban functions like business, high building density, medium road density) to satisfy the lasting demand therein. Urban residents can also select their travel destinations or making travel plans according to those features. Third, the synergistic effects uncovered among certain influencing factors echo the views of Ewing and Cervero ( 2017 ) and Handy ( 2017 ) that by working synergistically, the combined effects of built environment variables can be quite large. These effects imply that modifying some variables simultaneously (e.g., population density & land use mix, population density & job density, and land use mix & intersection density) could be more efficient than modifying individual variables for resilience promotion. They also warn the policy makers against the collateral effects of their actions in intervening with the built environment for enhancing travel behavior resilience. 6. Conclusions Employing random forest method, this study examines resilience of taxi ridership in the face of a typical extreme weather event, i.e., rainstorm, and its socioeconomical and built environment influencing factors. Our findings disclose the relative importance and complex effects of those influencing factors. As revealed, distance to city center, building density, and property price are the most important contributors of taxi ridership resilience, followed by job and population density, and other variables seem less critical. Meanwhile, all the independent variables are found to have salient non-linear relationships with resilience of taxi ridership, with obvious threshold effects. Distance to city center, land use mix, and density of building, population, job, bus route, and intersection have generally positive associations with taxi ridership resilience; road density has generally negative effects; and relationship between taxi ridership resilience and property price presents a complicated increase-fluctuate-decrease pattern. Moreover, synergistic effects exist between certain independent variables, e.g., population density & land use mix. These findings build a holistic and thorough view of resilience of taxi ridership, providing abundant theoretical and practical insights. This study still has some room for further enhancement. First, due to the lack of data sources, we did not involve some key socioeconomic variables, such as (average) personal or household income, into our analyses. Instead, we used property price as a proxy. Although property price is widely proven to be capable of representing people’s socioeconomic status feature, future studies are recommended to investigate the effects of such variables as income and transport expense budget directly. Second, we employed a rainstorm as the example for extreme weather events. Although we believe that rainstorm is representative enough, we acknowledge the differences between rainstorm and other disruptive events, e.g., heatwave and public health crisis. Hence, comparative studies are necessary for drawing generalized conclusions, which also points to the future research direction. Declarations Author Contribution J.L. and B.W. conceptualized the study. H.W. and L.X. developed the methodology. H.W. wrote the original draft of the manuscript. J.L. and B.W. reviewed and edited the manuscript. J.L. conducted the investigation. H.W. conducted the formal analysis. J.L. and B.W. were responsible for supervision, project administration, and funding acquisition. All authors have read and agreed to the final version of the manuscript. Data Availability Data are available on request. References Bian, R. R., C. G. Wilmot & L. Wang (2019) Estimating spatio-temporal variations of taxi ridership caused by Hurricanes Irene and Sandy: A case study of New York City. Transportation Research Part D: Transport and Environment, 77 , 627-638. Breiman, L. (2001) Random forests. Machine learning, 45 , 5-32. Cervero, R. (2013) Linking urban transport and land use in developing countries. Journal of transport and land use, 6 , 7-24. Chan, R. & J. L. Schofer (2016) Measuring transportation system resilience: Response of rail transit to weather disruptions. Natural Hazards Review, 17 , 05015004. Chen, C., T. Feng, C. Ding, B. Yu & B. Yao (2021) Examining the spatial-temporal relationship between urban built environment and taxi ridership: Results of a semi-parametric GWPR model. Journal of Transport Geography, 96 , 103172. Chen, R., L. Liu & Y. Gao (2024) The Association between Rainfall and Taxi Travel Activities: A Case Study from Wuhan, China. Journal of Advanced Transportation, 2024 , 1404168. da Mata Martins, M. C., A. N. R. da Silva & N. Pinto (2019) An indicator-based methodology for assessing resilience in urban mobility. Transportation Research Part D: Transport and Environment, 77 , 352-363. Ding, C., X. J. Cao & P. Næss (2018) Applying gradient boosting decision trees to examine non-linear effects of the built environment on driving distance in Oslo. Transportation Research Part A: Policy and Practice, 110 , 107-117. Ewing, R. & R. Cervero (2001) Travel and the built environment: a synthesis. Transportation research record, 1780 , 87-114. -- (2010) Travel and the built environment: A meta-analysis. Journal of the American planning association, 76 , 265-294. -- (2017) “Does compact development make people drive less?” The answer is yes. Journal of the American Planning Association, 83 , 19-25. Gan, Z., M. Yang, T. Feng & H. J. Timmermans (2020) Examining the relationship between built environment and metro ridership at station-to-station level. Transportation Research Part D: Transport and Environment, 82 , 102332. Geurts, P., D. Ernst & L. Wehenkel (2006) Extremely randomized trees. Machine learning, 63 , 3-42. Handy, S. (2017) Thoughts on the meaning of Mark Stevens’s meta-analysis. Journal of the American Planning Association, 83 , 26-28. Hao, H. & Y. Wang (2022) Disentangling relations between urban form and urban accessibility for resilience to extreme weather and climate events. Landscape and urban planning, 220 , 104352. Hastie, T., R. Tibshirani, J. H. Friedman & J. H. Friedman. 2009. The elements of statistical learning: data mining, inference, and prediction . Springer. He, Q., D. Rowangould, A. Karner, M. Palm & S. LaRue (2022) Covid-19 pandemic impacts on essential transit riders: findings from a US Survey. Transportation Research Part D: Transport and Environment, 105 , 103217. Ho, T. K. (1998) The random subspace method for constructing decision forests. IEEE transactions on pattern analysis and machine intelligence, 20 , 832-844. Holling, C. S. 1973. Resilience and stability of ecological systems. International Institute for Applied Systems Analysis Laxenburg. Hong, B., B. J. Bonczak, A. Gupta & C. E. Kontokosta (2021) Measuring inequality in community resilience to natural disasters using large-scale mobility data. NATURE COMMUNICATIONS, 12. Hu, S. & P. Chen (2021) Who left riding transit? Examining socioeconomic disparities in the impact of COVID-19 on ridership. Transportation Research Part D: Transport and Environment, 90 , 102654. Huang, J. & J. Wang (2024) Travel behavior resilience: Theory, method, and empirical studies. Journal of Geographical Sciences, 34 , 1657-1674. Huang, J., T. Zhang, D. Sun & H. Wang (2025) Urban human mobility changes based on functional areas during extreme rainstorm event: A case of Beijing “23· 7” rainstorm event. Cities, 163 , 106003. Irajifar, L., N. Sipe & T. Alizadeh (2016) The impact of urban form on disaster resiliency: A case study of Brisbane and Ipswich, Australia. International Journal of Disaster Resilience in the Built Environment, 7 , 259-275. Khavarian-Garmsir, A. R., A. Sharifi & N. Moradpour (2021) Are high-density districts more vulnerable to the COVID-19 pandemic? Sustainable Cities and Society, 70 , 102911. LeComte, D. (2020) A decade of extremes: The top-10 record-breaking events of recent years. Weatherwise, 73 , 14-23. Lepage, S. & C. Morency (2021) Impact of weather, activities, and service disruptions on transportation demand. Transportation research record, 2675 , 294-304. Li, Z., X. Wang, F. Gao, J. Tang & H. Xu (2024) Analysis of mobility patterns for urban taxi ridership: the role of the built environment. Transportation, 51 , 1409-1431. Liaw, A. & M. Wiener (2002) Classification and regression by randomForest. R news, 2 , 18-22. Liu, J., J. Cui, L. Xiao, D. Lin & L. Yang (2025) Non-linear built environment effects on travel behavior resilience under extreme weather events. Transportation Research Part D: Transport and Environment, 143 , 104753. Liu, J. & L. Xiao (2023) Non-linear relationships between built environment and commuting duration of migrants and locals. Journal of Transport Geography, 106 , 103517. -- (2024) Socioeconomic differences in effect size: predicting commuting mode choice of migrants and locals using a light gradient boosting approach. Transportation, 51 , 1-24. Liu, J., L. Xiao, L. Yang & J. Zhou (2020a) A tale of two social groups in Xiamen, China: Trip frequency of migrants and locals and its determinants. Travel Behaviour and Society, 20 , 213-224. Liu, Q., C. Ding & P. Chen (2020b) A panel analysis of the effect of the urban environment on the spatiotemporal pattern of taxi demand. Travel Behaviour and Society, 18 , 29-36. Liu, Y., Y. Li, W. Yang & J. Hu (2023) Exploring nonlinear effects of built environment on jogging behavior using random forest. Applied geography, 156 , 102990. Louppe, G. 2014. Understanding random forests: From theory to practice. Universite de Liege (Belgium). Lyu, T., Y. Wang, S. Ji, T. Feng & Z. Wu (2023) A multiscale spatial analysis of taxi ridership. Journal of Transport Geography, 113 , 103718. Meerow, S., J. P. Newell & M. Stults (2016) Defining urban resilience: A review. Landscape and urban planning, 147 , 38-49. Menze, B. H., B. M. Kelm, R. Masuch, U. Himmelreich, P. Bachert, W. Petrich & F. A. Hamprecht (2009) A comparison of random forest and its Gini importance with standard chemometric methods for the feature selection and classification of spectral data. BMC bioinformatics, 10 , 1-16. Olsson, L., A. Jerneck, H. Thoren, J. Persson & D. O’Byrne (2015) Why resilience is unappealing to social science: Theoretical and empirical investigations of the scientific use of resilience. Science advances, 1 , e1400217. Pan, H., Q. Shen & T. Zhao (2013) Travel and Car Ownership of Residents near New Suburban Metro Stations in Shanghai, China. Transportation Research Record Journal of the Transportation Research Board, 2394 , 63-69. Peng, Y., J. Liu, F. Li, J. Cui, Y. Lu & L. Yang (2024) Resilience of ride-hailing services in response to air pollution and its association with built-environment and socioeconomic characteristics. Journal of Transport Geography, 120 , 103971. Qian, X. & S. V. Ukkusuri (2015) Spatial variation of the urban taxi ridership using GPS data. Applied geography, 59 , 31-42. Qiao, M., M. Haraguchi & U. Lall (2025) Enhancing urban resilience to extreme weather: The roles of human transition paths among multiple transportation modes. International Journal of Geographical Information Science, 39 , 2408-2427. Ruan, J., Y. Chen & Z. Yang (2021) Assessment of temporal and spatial progress of urban resilience in Guangzhou under rainstorm scenarios. International Journal of Disaster Risk Reduction, 66 , 102578. Sharifi, A. (2019a) Resilient urban forms: A macro-scale analysis. Cities, 85 , 1-14. -- (2019b) Urban form resilience: A meso-scale analysis. Cities, 93 , 238-252. Solomon, S. (2007) Contribution of Working Group I to the fourth assessment report of the Intergovernmental Panel on Climate Change 2007. (No Title) . Strobl, C., J. Malley & G. Tutz (2009) An introduction to recursive partitioning: rationale, application, and characteristics of classification and regression trees, bagging, and random forests. Psychological methods, 14 , 323. Sun, D. J., K. Zhang & S. Shen (2018) Analyzing spatiotemporal traffic line source emissions based on massive didi online car-hailing service data. Transportation Research Part D: Transport and Environment, 62 , 699-714. Tang, J., P. Zhao, Z. Gong, H. Zhao, F. Huang, J. Li, Z. Chen, L. Yu & J. Chen (2023) Resilience patterns of human mobility in response to extreme urban floods. National science review, 10 , nwad097. Tongyue, L., N. Plnyi & G. U. Chaolin (2014) A Review on Research Frameworks of Resilient Cities. Urban Planning Forum . Utsumi, N. & H. Kim (2022) Observed influence of anthropogenic climate change on tropical cyclone heavy rainfall. Nature Climate Change, 12 , 436-440. Vergel-Tovar, C. E. & D. A. Rodriguez (2018) The ridership performance of the built environment for BRT systems: Evidence from Latin America. Journal of transport geography, 73 , 172-184. Wang, B., B. P. Loo, F. Zhen & G. Xi (2020) Urban resilience from the lens of social media data: Responses to urban flooding in Nanjing, China. Cities, 106 , 102884. Wang, B., F. Zhang, J. Liu & Z. Tan (2024) The impacts of extreme hot weather on metro ridership: A case study of Shenzhen, China. Journal of Transport Geography, 117 , 103899. Wang, F., J. D. Harindintwali, K. Wei, Y. Shan, Z. Mi, M. J. Costello, S. Grunwald, Z. Feng, F. Wang & Y. Guo (2023) Climate change: Strategies for mitigation and adaptation. The Innovation Geoscience, 1 , 100015-1-100015-37. Wang, J., J. Huang, H. Yang & D. Levinson (2022) Resilience and recovery of public transport use during COVID-19. Npj Urban Sustainability, 2 , 18. Wang, P., Y. Li & Y. Zhang (2021) An urban system perspective on urban flood resilience using SEM: evidence from Nanjing city, China. Natural hazards, 109 , 2575-2599. Wright, M. N., A. Ziegler & I. R. König (2016) Do little interactions get lost in dark random forests? BMC bioinformatics, 17 , 1-10. Xiao, W. & Y. D. Wei (2023) Assess the non-linear relationship between built environment and active travel around light-rail transit stations. Applied geography, 151 , 102862. Xiao, W., Y. D. Wei & Y. Wu (2022) Neighborhood, built environment and resilience in transportation during the COVID-19 pandemic. Transportation Research Part D: Transport and Environment, 110 , 103428. Xiao, Y., S. Orford & C. J. Webster (2016) Urban configuration, accessibility, and property prices: A case study of Cardiff, Wales. Environment and Planning B: Planning and Design, 43 , 108-129. Yan, X., X. Liu & X. Zhao (2020) Using machine learning for direct demand modeling of ridesourcing services in Chicago. Journal of Transport Geography, 83 , 102661. Yang, C. & E. J. Gonzales (2014) Modeling taxi trip demand by time of day in New York City. Transportation Research Record, 2429 , 110-120. Yang, H., J. Huo, R. Pan, K. Xie, W. Zhang & X. Luo (2022) Exploring built environment factors that influence the market share of ridesourcing service. Applied Geography, 142 , 102699. Yang, H., J. Peng, Y. Lu, J. Wang & X. Yan (2024) Nonlinear impact of built environment on people with disabilities’ metro use behavior. Applied Geography, 169 , 103323. Yang, L., X. Chu, Z. Gou, H. Yang, Y. Lu & W. Huang (2020) Accessibility and proximity effects of bus rapid transit on housing prices: Heterogeneity across price quantiles and space. Journal of Transport Geography, 88 , 102850. Yang, L., J. Zhou & O. F. Shyr (2019) Does bus accessibility affect property prices? Cities, 84 , 56-65. Yu, H. & Z.-R. Peng (2019) Exploring the spatial variation of ridesourcing demand and its relationship to built environment and socioeconomic factors with the geographically weighted Poisson regression. Journal of Transport Geography, 75 , 147-163. Zhang, B., S. Chen, Y. Ma, T. Li & K. Tang (2020) Analysis on spatiotemporal urban mobility based on online car-hailing data. Journal of Transport Geography, 82 , 102568. Zhang, F., Z. Li, N. Li & D. Fang (2019) Assessment of urban human mobility perturbation under extreme weather events: A case study in Nanjing, China. Sustainable Cities and Society, 50 , 101671. Zhang, X. & N. Li (2020) Assessment of the Correlation between Extreme Weather Event-Induced Human Mobility Perturbation in Urban Areas and their Spatial Characteristics based on Taxi Trajectories. Zhu, D., N. Wang, L. Wu & Y. Liu (2017) Street as a big geo-data assembly and analysis unit in urban studies: A case study using Beijing taxi data. Applied Geography, 86 , 152-164. Zhu, P., J. Huang, J. Wang, Y. Liu, J. Li, M. Wang & W. Qiang (2022) Understanding taxi ridership with spatial spillover effects and temporal dynamics. Cities, 125 , 103637. Zhu, P., J. Li, K. Wang & J. Huang (2024) Exploring spatial heterogeneity in the impact of built environment on taxi ridership using multiscale geographically weighted regression. Transportation, 51 , 1963-1997. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8938756","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":596337423,"identity":"81ae81cf-73f1-4d37-bd68-70c2fb5b159a","order_by":0,"name":"Jixiang Liu","email":"","orcid":"","institution":"Xiamen University","correspondingAuthor":false,"prefix":"","firstName":"Jixiang","middleName":"","lastName":"Liu","suffix":""},{"id":596337426,"identity":"b53cad1d-0f36-481b-8602-38aadbbd5465","order_by":1,"name":"Hongyu Wu","email":"","orcid":"","institution":"Xiamen University","correspondingAuthor":false,"prefix":"","firstName":"Hongyu","middleName":"","lastName":"Wu","suffix":""},{"id":596337430,"identity":"6fa37efe-997d-4a7d-9351-410a9273e7ab","order_by":2,"name":"Longzhu Xiao","email":"","orcid":"","institution":"Xiamen University","correspondingAuthor":false,"prefix":"","firstName":"Longzhu","middleName":"","lastName":"Xiao","suffix":""},{"id":596337432,"identity":"bc8dbb8a-62fb-4c52-b549-003953d6ff69","order_by":3,"name":"Bo Wang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABEElEQVRIiWNgGAWjYJACZgYGGzl+IOPAA5gQD2EtacaSDUAtCSRoOZS44QCQRZQWg+NnD78uqDhgbHzt8EOgLXWJ82ckMD5428Ygb45Ly5m8NOsZZ+7Imd1OMwBqOZy44UYCs+HcNgbDnQ3YtZgdyDEz5m17Zmx2OwGk5UDiBokENmneNgYgF4eW829AWg4nbp6d/gHmMPbfeLXcyDF+DNKyQToHZAtzYsONBDZmfFrsb7wxY55xJs1Y4nZOwYEEg8PGG848bJacc07CcAMOLZL9OcafCyqAUTk7ffOHDxV1svPbkw9+eFNmI4/LFiBgk0CwDRgcGxgYG4AsCVzKQYD5A4pL8SkdBaNgFIyCkQkAwL9kndYFlzUAAAAASUVORK5CYII=","orcid":"","institution":"Sun Yat-sen University","correspondingAuthor":true,"prefix":"","firstName":"Bo","middleName":"","lastName":"Wang","suffix":""}],"badges":[],"createdAt":"2026-02-22 11:24:05","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8938756/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8938756/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":104779224,"identity":"04832266-d800-4c36-897c-948efb9ea23d","added_by":"auto","created_at":"2026-03-17 07:36:52","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":222246,"visible":true,"origin":"","legend":"\u003cp\u003eStudy area: Xiamen Island\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-8938756/v1/1f0afdfb2127d4c6bfc6c01f.png"},{"id":103724439,"identity":"ae938a93-f61d-4f6d-907e-4af25a0db071","added_by":"auto","created_at":"2026-03-02 07:57:19","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":98219,"visible":true,"origin":"","legend":"\u003cp\u003eSchema of the resilience triangle\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-8938756/v1/254f66c4a37e6fc19343bd57.png"},{"id":103724444,"identity":"b0f99fc9-cb6e-4200-a96d-1a2f9aac761f","added_by":"auto","created_at":"2026-03-02 07:57:19","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":42454,"visible":true,"origin":"","legend":"\u003cp\u003eNon-linear effects of property price on resilience of taxi ridership\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-8938756/v1/431bbcfc5ac7784d0f211c1f.png"},{"id":103724446,"identity":"c70b09c9-1b7e-4c96-a173-ac530169c99c","added_by":"auto","created_at":"2026-03-02 07:57:20","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":54681,"visible":true,"origin":"","legend":"\u003cp\u003eNon-linear effects of distance to city center and building density on resilience of taxi ridership\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-8938756/v1/f479eea6058e18d015826852.png"},{"id":103724441,"identity":"ee11ad06-94d0-4e17-b918-bf9e904d490f","added_by":"auto","created_at":"2026-03-02 07:57:19","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":49039,"visible":true,"origin":"","legend":"\u003cp\u003eNon-linear effects of population density and job density on resilience of taxi ridership\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-8938756/v1/6cb24ee28d2b1e5173a20318.png"},{"id":103724449,"identity":"30b159b4-31b7-4b8f-842b-0654a8254de8","added_by":"auto","created_at":"2026-03-02 07:57:21","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":55866,"visible":true,"origin":"","legend":"\u003cp\u003eNon-linear effects of land use mix and road density on resilience of taxi ridership\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-8938756/v1/4ef8fa896419e5091eedb2bf.png"},{"id":104399919,"identity":"e8dbc19a-c750-49e3-a06f-a574ecb84e84","added_by":"auto","created_at":"2026-03-11 12:08:09","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":56218,"visible":true,"origin":"","legend":"\u003cp\u003eNon-linear effects of bus route density and intersection density on resilience of taxi ridership\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-8938756/v1/85ca2ce391865b9f814af899.png"},{"id":103724443,"identity":"d3435b2a-042e-456c-89d3-e7d9327d582b","added_by":"auto","created_at":"2026-03-02 07:57:19","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":289946,"visible":true,"origin":"","legend":"\u003cp\u003eSynergistic effects of independent variables on resilience of taxi ridership\u003c/p\u003e","description":"","filename":"8.png","url":"https://assets-eu.researchsquare.com/files/rs-8938756/v1/fd334c171de5f78af9cdfbeb.png"},{"id":104783682,"identity":"4196d098-5d82-477d-97fe-0cbeb94cc592","added_by":"auto","created_at":"2026-03-17 08:03:18","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1784453,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8938756/v1/62e0eb9b-3c07-4ad0-9f4a-994154c5a928.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Examining resilience of taxi ridership during rainstorms and its non-linear associations with built environment","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eThe recent years have witnessed the increasing frequency and intensity of diverse extreme weather events against the background of global climate change (Utsumi and Kim \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Among those events, rainstorms have particularly adverse impacts on a wide variety of aspects of cities, including infrastructures, energy supply, built environment, economic vibrancy, human behavior, and safety, health, and well-being of people (Liu et al. \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2025\u003c/span\u003e, Wang et al. \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). For example, a phenomenal rainstorm brought about by the Hurricane Harvey attacked north Texas coast in late August 2017, inundating almost one third of Houston at one point (LeComte \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). The rainstorm that happened on July 20th, 2021 in Zhengzhou, China, led to a serious flood and caused 380 deaths and a direct 40-billion-yuan economic loss (Tang et al. \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). In the end of July 2023, a 1-in-100-year record-breaking rainstorm hit Beijing, China (Huang et al. \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), and caused more than 33 people to be dead and 18 to be missing. Thereby, it is both essential and meaningful to uncover how cities are impacted by and recover from rainstorms, thus offering knowledge base for targeted policy interventions.\u003c/p\u003e \u003cp\u003eTaxis are an integral component of the urban transportation system. As a transport mode being capable of providing 24-hour available and spatially flexible services, taxis have become an ideal solution for the first/last mile transportation problem and a reliable complement for the public transport system (Zhu et al. \u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e2024\u003c/span\u003e, Zhu et al. \u003cspan citationid=\"CR74\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). On the other hand, as a typical type of surface transportation, taxis are notably susceptible to extreme weather events like rainstorms. For example, Chen, Liu and Gao (\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) revealed that in Wuhan, China, the taxi volume dropped averagely by 4.16% weekly due to rainfall. Taxi ridership in New York City was found to decline during adverse weather conditions (including heavy rain and strong wind), particularly in the weekend and at night (Bian, Wilmot and Wang \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). In the case of Nanjing, China, Zhang et al. (\u003cspan citationid=\"CR72\" class=\"CitationRef\"\u003e2019\u003c/span\u003e) uncovered that taxis are more sensitive to rainstorms than other non-private transportation modes like buses. Nevertheless, some researchers have gained some counter-intuitive insights into the effects of rainstorms on taxi travel behavior. For instance, Lepage and Morency (\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) discovered the positive effects of rainfall on taxi ridership; in their study in Montreal, Canada, four consecutive hours of rains had been found to increase taxi ridership by 13.9%. Likewise, Qiao, Haraguchi and Lall (\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) found in New York City that taxi ridership increased throughout the extreme rainfall events, demonstrating the role of taxis for urban resilience. Hence, it is essential to take a deeper look into how taxi ridership responds to rainstorms.\u003c/p\u003e \u003cp\u003eResilience, as mentioned above, has become an increasingly popular approach to delving into the variations and recovery of travel behavior in the face of extreme weather events. In the field of transportation, resilience can be defined as the capacity of a transportation system to stand a disruptive event and rebound back to normal operational condition in an acceptable amount of time (Chan and Schofer \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2016\u003c/span\u003e, da Mata Martins, da Silva and Pinto \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Since transportation system can be divided into the supply part (e.g., facilities, infrastructures, networks) and demand part (transportation users), transportation resilience has also been interpreted from the perspectives of supply and demand (Liu et al. \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Overall, research regarding transportation resilience from the supply perspective has predominates, while academic explorations on transportation resilience from the demand perspective (i.e., resilience of people\u0026rsquo;s travel behavior patterns during disruptions, referred to as travel behavior resilience hereafter) are still limited. For example, in their case study in Kunming, China, Wang et al. (\u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) first defined travel behavior resilience and measured it with variations of public transit trips during COVID-19. Using Chengdu, China as an example, Peng et al. (\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) evaluated resilience of ride-hailing behavior and examined its relationships with the built environment. Moreover, despite the irreplaceable role that taxis play in the transportation system and the ability of taxi ridership of reflecting human mobility, seldom has resilience of taxi ridership and its influencing factors been examined, particularly against the backdrop of extreme weather events like rainstorms.\u003c/p\u003e \u003cp\u003eBuilt environment has widely been confirmed as a key influencing factor of taxi travel behavior. Since the introduction of D\u0026rsquo;s framework by Ewing and Cervero (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2001\u003c/span\u003e) and Ewing and Cervero (\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2010\u003c/span\u003e), researchers have established strong connections between taxi ridership and multiple dimensions of built environment, such as density, diversity, design, destination accessibility, and distance to transit, among others (Chen et al. \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2021\u003c/span\u003e, Li et al. \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2024\u003c/span\u003e, Zhu et al. \u003cspan citationid=\"CR75\" class=\"CitationRef\"\u003e2022\u003c/span\u003e, Zhu et al. \u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e2024\u003c/span\u003e, Yang et al. \u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Meanwhile, recent years have also witnessed increasing academic attention towards the role of built environment in affecting urban resilience. As revealed, built environment with certain features (e.g., multi-centricity from a macro perspective and higher land use diversity and medium-level population density from a meso or micro perspective) are indeed beneficial for urban resilience (Hao and Wang \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2022\u003c/span\u003e, Irajifar, Sipe and Alizadeh \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2016\u003c/span\u003e, Sharifi \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2019a\u003c/span\u003e, Sharifi \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2019b\u003c/span\u003e, Wang et al. \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Xiao, Wei and Wu (\u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) and Peng et al. (\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) are among the pioneering studies that examined the effects of built environment on resilience of transportation from the demand perspective (i.e., travel behavior resilience), and the latter even corroborated the non-linearity in the relationships. However, knowledge regarding how built environment influences resilience of taxi ridership is still scarce. Given the widely confirmed non-linear effects of built environment on travel behavior and the potential synergistic effects among built-environment variables (Xiao and Wei \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2023\u003c/span\u003e, Yang et al. \u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e2024\u003c/span\u003e, Liu et al. \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), disentangling the complicated relationships between built environment and resilience of taxi ridership is essential for providing evidence base to promote travel behavior resilience.\u003c/p\u003e \u003cp\u003eTherefore, employing Xiamen Island as the study area, this study attempts to examine the non-linear relationships between built environment and resilience of taxi ridership, using an advanced machine learning method, i.e., random forest. It aims to answering three research questions: (1) Which are the most important built-environment variables for resilience of taxi ridership? (2) Do the built-environment variables have non-linear effects on taxi ridership resilience? (3) Are there any synergistic effects among the built-environment variables? This study contributes to the literature in the following three aspects. First, utilizing the taxi trip dataset in Xiamen Island, it scientifically quantifies resilience of taxi ridership facing a typical type of extreme weather events, i.e., rainstorm. Second, through revealing the hierarchy of relative contributions of built-environment variables, it clearly demonstrates the most important influencing factors of resilience of taxi ridership. Third, by visualizing the complicated relationships between built environment and resilience of taxi ridership without pre-defining the relationship shapes, it offers refined recommendations for policymakers to formulate and implement tailored measures to enhance transport resilience.\u003c/p\u003e \u003cp\u003eThe remainder of the paper unfolds as follows. Section \u003cspan refid=\"Sec2\" class=\"InternalRef\"\u003e2\u003c/span\u003e reviews the relevant literature. Section \u003cspan refid=\"Sec5\" class=\"InternalRef\"\u003e3\u003c/span\u003e introduces the study area, data, variables, and modeling methods. Section \u003cspan refid=\"Sec12\" class=\"InternalRef\"\u003e4\u003c/span\u003e presents and analyzes the research results. Section \u003cspan refid=\"Sec17\" class=\"InternalRef\"\u003e5\u003c/span\u003e discusses the findings and Section \u003cspan refid=\"Sec18\" class=\"InternalRef\"\u003e6\u003c/span\u003e concludes the study.\u003c/p\u003e"},{"header":"2. Literature review","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Relationships between built environment and taxi ridership\u003c/h2\u003e \u003cp\u003eThe built environment has long been established as a fundamental determinant of travel behavior (Ewing and Cervero \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2010\u003c/span\u003e). Among the array of travel modes, taxi constitutes a vital segment of urban transport systems, exhibiting notable spatiotemporal variation closely tied to the characteristics of the built environment (Cervero \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2013\u003c/span\u003e, Liu, Ding and Chen \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2020b\u003c/span\u003e, Qian and Ukkusuri \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2015\u003c/span\u003e, Chen et al. \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). To systematically capture these attributes influencing taxi ridership, the dominant analytical framework employs the widely accepted \u0026ldquo;5Ds\u0026rdquo; framework, which encompasses density, design, diversity, destination accessibility, and distance to transit (Ewing and Cervero \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2010\u003c/span\u003e, Ewing and Cervero \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2001\u003c/span\u003e). Technological advances, especially the widespread application of high-resolution positioning tools such as GPS, have enabled the large-scale and fine-grained collection of taxi trajectory data, including pick-up and drop-off points, travel durations, and distances (Sun, Zhang and Shen \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2018\u003c/span\u003e, Zhang et al. \u003cspan citationid=\"CR71\" class=\"CitationRef\"\u003e2020\u003c/span\u003e), thereby laying a solid empirical foundation for analyzing built environment\u0026ndash;taxi mobility interactions.\u003c/p\u003e \u003cp\u003eA growing body of empirical evidence has revealed considerable heterogeneity in the effects of built environment features on taxi ridership, both in direction and magnitude (Zhu et al. \u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Among these features, land-use mix consistently emerges as a key explanatory factor, with higher levels generally associated with increased taxi ridership. Land-use mix is commonly measured using entropy-based indicators that incorporate land-use categories, such as residential, commercial, office, and green spaces (Qian and Ukkusuri \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2015\u003c/span\u003e, Yu and Peng \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2019\u003c/span\u003e, Liu et al. \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2020b\u003c/span\u003e). In parallel, recent studies highlight the significant role of specific POI types\u0026mdash;particularly financial and entertainment venues\u0026mdash;in shaping spatial variation in taxi activity (Li et al. \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Beyond land-use mix, other built environment variables, such as population and employment densities, street network design, and transit infrastructure, have also exerted meaningful influence on taxi ridership. Empirical findings generally support positive associations with population density, employment concentration, and intersection density (Yang and Gonzales \u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e2014\u003c/span\u003e, Liu et al. \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2020b\u003c/span\u003e, Zhu et al. \u003cspan citationid=\"CR75\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), while higher bus stop density tends to exhibit substitution effects, negatively correlating with taxi ridership (Liu et al. \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2020b\u003c/span\u003e). Furthermore, increased taxi ridership has been linked to prolonged bus headways, implying a modal shift under diminished transit service quality (Yang and Gonzales \u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). Interestingly, metro accessibility has been positively associated to taxi ridership (Qian and Ukkusuri \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2015\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eImportantly, the impacts of these built environment variables are not strictly linear. A range of studies have identified threshold effects\u0026mdash;wherein the marginal influence of a variable plateaus or shifts upon reaching critical values (Chen et al. \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Moreover, these effects are marked by considerable spatiotemporal heterogeneity, with varying influences across different temporal, spatial, and seasonal dimensions (Zhu et al. \u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e2024\u003c/span\u003e, Zhu et al. \u003cspan citationid=\"CR75\" class=\"CitationRef\"\u003e2022\u003c/span\u003e, Chen et al. \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). For instance, road network density exerts a more pronounced influence in southern urban regions with less-developed infrastructure, and taxi ridership tends to spike during weekdays and peak hours (Liu et al. \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2020b\u003c/span\u003e). Seasonal variations has also been documented, with warmer months (e.g., autumn) witnessing elevated taxi demand and rainfall demonstrating a positive but non-linear relationship with taxi ridership (Liu et al. \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2020b\u003c/span\u003e). Such findings underscore the multifaceted and context-dependent nature of taxi behavior in urban settings (Zhu et al. \u003cspan citationid=\"CR75\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eNevertheless, current scholarship has predominantly examined these relationships under typical, undisturbed conditions. Much less attention has been devoted to how the built environment affects the variations of travel behavior during disruptive events, such as extreme weather events. Incidents like heavy rainfall can severely compromise transport system functionality, prompting deviations from routine travel patterns. Under such conditions, analyzing the role of built environment features in facilitating mobility recovery offers critical insights into their capacity to foster transport resilience.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Travel behavior resilience and its built-environment influencing factors\u003c/h2\u003e \u003cp\u003eThe concept of \u0026ldquo;resilience\u0026rdquo; was originally introduced by Holling (\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e1973\u003c/span\u003e), denoting the capacity of an ecosystem to absorb shocks while retaining their essential structure and function. As scholarly paradigms evolved, the notion of resilience expanded into urban contexts, culminating in the widely discussed framework of \u0026ldquo;urban resilience\u0026rdquo; (Tongyue, Plnyi and Chaolin \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). In the face of mounting uncertainties, including climate change, economic instability, and infrastructural strain, cities are increasingly viewed as complex adaptive systems whose resilience encompasses both physical robustness and social adaptability (Huang and Wang \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2024\u003c/span\u003e, Meerow, Newell and Stults \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2016\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eRecent advances in resilience research advocate a dual-perspective approach, encompassing both supply-side and demand-side dimensions, aiming to systematically capture the diverse sources of recovery capacity embedded within the built environment (Huang and Wang \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). On the supply side, resilience emphasizes the structural integrity and recovery potential of physical infrastructure. Numerous studies have demonstrated that urban resilience is significantly influenced by macro-scale factors (e.g., landscape connectivity) and meso-scale factors (e.g., land-use mix). Specifically, compact and polycentric urban forms, moderate density, higher land-use diversity, and fine-grained block design are frequently associated with enhanced resilience, with the interplay between land-use mix and density proving particularly influential (Sharifi \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2019a\u003c/span\u003e, Sharifi \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2019b\u003c/span\u003e). In the context of Chinese cities, multi-source data assessments reveal that core urban zones in Guangzhou exhibit higher recovery capacities than peripheral new districts (Ruan, Chen and Yang \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Similarly, an evaluation of flood resilience in Nanjing indicates that since 2006, the relative influence of economic, institutional, cultural, and physical resilience components has markedly increased (Wang, Li and Zhang \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). In addition, findings from Australia suggest that population density significantly enhances post-disaster recovery, with the effect displaying clear non-linear characteristics during flood events (Irajifar et al. \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2016\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eMeanwhile, triggered by the emergence of disruptive events such as the COVID-19 pandemic, increasing scholarly attention has shifted toward the demand side of urban resilience. Here, the focus is on people\u0026rsquo;s behavioral adaptability and the heterogeneity of response across social groups (Olsson et al. \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). Travel behavior, as a core component of daily human activities, is characterized by spatial dependence, spatiotemporal dynamics, and group heterogeneity. These features render it a critical lens for understanding the interaction and evolving mechanisms between travel demand and the built environment, as well as a pivotal entry point for research on transport resilience (Huang and Wang \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Emerging studies have begun to examine the resilience of travel behavior under various stressors. For example, during the COVID-19, the decline in public transit ridership has been attributed not only to physical attributes of the built environment (Hu and Chen \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2021\u003c/span\u003e, Khavarian-Garmsir, Sharifi and Moradpour \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), but more fundamentally to the long-term influence of the built environment in shaping individual travel preferences, which in turn affects behavioral responses across different social groups (Gan et al. \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2020\u003c/span\u003e, Vergel-Tovar and Rodriguez \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Although land-use mix use theoretically enhances accessibility to destinations such as shops and schools, its practical benefit was reduced under pandemic due to widespread temporary closures (He et al. \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Furthermore, overly compact urban design may reduce building coverage and traffic infrastructure redundancy, exacerbating system vulnerability and negatively impacting travel resilience (Xiao et al. \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Social vulnerability compounds these effects, transit ridership dropped more sharply in neighborhoods with higher proportions of marginalized populations, illustrating disparities in resilience capacity (Xiao et al., \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Beyond pandemics, environmental disruptions like air pollution have also been investigated, a recent study employed the Gradient Boosting Decision Tree (GBDT) model and identified significant non-linear effects of built environment attributes on ride-hailing travel resilience, including population density, densities of bus and rail stations, and land-use mix (Peng et al. \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eHowever, the bulk of existing research remains centered on social disruptions. Comparatively fewer studies have addressed resilience in the context of extreme weather events, which differ fundamentally from pandemics in their physical intensity, temporal immediacy, and impact mechanisms (Solomon \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2007\u003c/span\u003e). These differences suggest that built environment variables may interact with travel behavior resilience in unique ways during weather-induced disruptions. As climate-related hazards become increasingly frequent and severe, there is a pressing need to understand how built environment can either mitigate or exacerbate their impacts. Investigating this nexus offers not only conceptual expansion of demand-side resilience theory, but also provides actionable insights for climate-adaptive urban mobility planning, which is a task of urgent theoretical and practical significance in the era of climate change.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Methodology","content":"\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Study area and case of extreme weather event\u003c/h2\u003e \u003cp\u003eThis study selects Xiamen Island as the case study area, i.e., the core urban district of Xiamen City. Covering approximately 158 km\u0026sup2;, Xiamen Island had a population of 2.067\u0026nbsp;million in 2023 and functions as the city\u0026rsquo;s political, economic, cultural, and transportation hub. Siming and Huli districts form Xiamen Island, being separated from the other four districts of Xiamen City (i.e., Haicang, Jimei, Tong\u0026rsquo; an, and Xiang\u0026rsquo; an Districts) by the sea (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eXiamen Island was chosen for this study for several reasons. First, its high population density and thriving tourism industry generate intense transportation demand. The city's scenic landscapes and cultural heritage attract large numbers of tourists, leading to concentrated travel needs from both residents and visitors. Second, Xiamen Island has a highly compact and high-density built environment with a multifunctional urban structure. Major commercial centers (e.g., Zhongshan Road), residential areas, transportation hubs (e.g., Xiamen Railway Station), and key tourist attractions (e.g., Xiamen Botanical Garden) are all located within the island. Third, metro development in Xiamen has been relatively slow due to geographic constraints and population growth rate. With only a few operational metro lines, the city lacks a fully developed rail transit network. Consequently, surface transportation remains a dominant mode of travel, and taxis play a crucial role in meeting mobility demands. Lastly, Xiamen\u0026rsquo;s subtropical maritime climate brings an average annual rainfall of about 1,200 mm, peaking from May to August. Frequent heavy rainfall and extreme weather events further challenge the stability of the urban transportation system.\u003c/p\u003e \u003cp\u003eThese factors make Xiamen Island an ideal case for examining the relationship between the built environment and the resilience of taxi ridership. The findings of this study can offer insights for similar medium-sized cities with underdeveloped metro systems and frequent extreme weather events, supporting transportation planning and climate adaptation strategies.\u003c/p\u003e \u003cp\u003eThis study collected records of extreme rainfall events in Xiamen in recent years and selected the extreme rainfall event on May 7, 2018, as the case study. On this day, Xiamen experienced an exceptionally heavy rainstorm, with hourly precipitation exceeding 100 mm in multiple locations on Xiamen Island.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Data\u003c/h2\u003e \u003cp\u003eThis study utilizes two types of data: taxi trip data and built environment data. The taxi trip data, provided by the Xiamen Transportation Bureau, includes key trip details such as the latitude and longitude of trip origins and destinations, trip distance, trip duration, pick-up time, and drop-off time, with an average of 130,000 taxi trips recorded daily. The built environment data, sourced from Xiamen University, encompasses land-use planning, road network data, building information, points of interest (POI), the locations of important public facilities and so on. All datasets were processed using ArcGIS 10.8 to ensure data standardization and usability.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Variables\u003c/h2\u003e \u003cdiv id=\"Sec9\" class=\"Section3\"\u003e \u003ch2\u003e3.3.1 Resilience of taxi ridership\u003c/h2\u003e \u003cp\u003eResilience of taxi ridership serves as the dependent variable in this study, and the Resilience Triangle framework is employed to quantify the capacity of taxi ridership to withstand and recover from extreme weather events. The Resilience Triangle has widely been utilized in previous studies (Wang et al. \u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) as a framework for describing system performance changes under external shocks (e.g., COVID-19) and assessing its response and recovery capacity.\u003c/p\u003e \u003cp\u003eAs illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, the performance curve (denoted as P(t)) represents how system performance (Y-axis) changes over time (X-axis) during a disruption event. Specifically, the Resilience Triangle consists of three key phases: 1) Pre-disruption Performance, indicating the system's stable operational level before the disruption occurs. 2) Decline Performance, denoting the degree of system deterioration and its vulnerability during an external shock. 3) Recovery Phase, i.e., the process by which the system restores itself from the post-disruption low point to normal conditions. The area of the Resilience Triangle serves as an indicator of the resilience of the system\u0026mdash;a larger area signifies a longer recovery period and lower resilience, whereas a smaller area indicates a stronger recovery capability and higher resilience.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThis study assesses the resilience of taxi ridership by examining residents' travel behavior during the extreme rainstorm event on May 7, 2018, in Xiamen. To ensure spatial consistency in the analysis, the entire Xiamen Island was divided into 500 m \u0026times; 500 m grid cells. The calculation of taxi ridership resilience follows a three-step process. First, the baseline (\u0026#119875;_baseline) was established by calculating the hourly average taxi pick-up and drop-off counts for each grid based on the five Mondays preceding the rainstorm (i.e., April 2, 9, 16, 23, and 30, 2018). This ensured a reliable reference for assessing disruptions. Second, the maximum loss (impact and performance decline) was determined by identifying the minimum taxi ridership (\u0026#119875;_min) under the rainstorm, comparing hourly taxi demand with the baseline. To account for spatial heterogeneity, the study measured the relative change proportions instead of absolute values. Third, the recovery time was identified using the Plateau Detection Algorithm, which detected the stabilization point (\u0026#119905;3) based on a rolling variance tolerance of 0.15 over 12 consecutive hours. Finally, based on the Resilience Triangle framework, the resilience value of taxi ridership for each grid was calculated using Eq.\u0026nbsp;(\u003cspan refid=\"Equ1\" class=\"InternalRef\"\u003e1\u003c/span\u003e):\u003cdiv id=\"Equ1\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ1\" name=\"EquationSource\"\u003e\n$$\\:Resilience=\\:\\frac{1}{{\\int\\:}_{t1}^{t3}{P}_{baseline}-{\\int\\:}_{t1}^{t3}{P}_{min}}\\approx\\:\\frac{2}{\\left({P}_{baseline}-{P}_{\\text{m}\\text{i}\\text{n}}\\right)*\\left(t3-t1\\right)}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e1\u003c/div\u003e\u003c/div\u003e;\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section3\"\u003e \u003ch2\u003e3.3.2 Independent variables\u003c/h2\u003e \u003cp\u003eThis study derives independent variables from two dimensions: socioeconomic attributes and built environment characteristics. To ensure spatial consistency, a 500 m \u0026times; 500 m grid unit is used for calculation. Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e1\u003c/span\u003e provides detailed definitions and statistical information for each independent and dependent variable. The built environment variables are measured based on the classic \u0026ldquo;5Ds\u0026rdquo; model (Ewing and Cervero \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2001\u003c/span\u003e, Ewing and Cervero \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2010\u003c/span\u003e), including density (population density, job density), diversity (land use mix index), design (intersection density), public transport accessibility (bus route density), and destination accessibility (distance to city center). Additionally, socioeconomic variables are represented by property prices in 2018, serving as an indicator of residents\u0026rsquo; economic status.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eDescriptive statistics of dependent and independent variables\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariables\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDescriptions\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMean (S.D.)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eDependent variable\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eResilience\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eResilience of taxi ridership measured using Resilience Triangle.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2.17 (1.43)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eIndependent variables\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eSocioeconomics\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eProperty price\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eProperty price in 2018 (Yuan/m\u003csup\u003e2\u003c/sup\u003e).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e48944.10 (9200.16)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cem\u003eBuilt environment\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDistance to CBD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNetwork distance to Zhongshan Road (km).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e13.89 (14.11)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePopulation density\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePopulation density index based on Baidu Heat Map.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e263.67 (185.45)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBuilding density\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTotal area of buildings per km\u003csup\u003e2\u003c/sup\u003e (m\u003csup\u003e2\u003c/sup\u003e).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1670.25 (1254.54)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJob density\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNo. of employment positions per km\u003csup\u003e2\u003c/sup\u003e.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e8237.98 (16100.95)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRoad density\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRoad length per km\u003csup\u003e2\u003c/sup\u003e (km).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e12.93 (16.82)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIntersection density\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNo. of three-or-more-way intersections per km\u003csup\u003e2\u003c/sup\u003e.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e16.70 (20.21)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBus route density\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNo. of bus routes per km\u003csup\u003e2\u003c/sup\u003e.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e48.76 (84.65)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLand-use mix\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThe degree to which different land-use types are integrated within the grid. In this study, we identified twelve land-use types, i.e., industry, residence, business and commerce, urban village, green land and public open space, transportation and logistics, education and research, spatially designated land, culture and sports, municipal administration, medicine, and others.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.58 (0.21)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section3\"\u003e \u003ch2\u003e3.3.3 Methodology: random forest\u003c/h2\u003e \u003cp\u003eThis study employs the random forest method to investigate the relationships between built environment and taxi ridership resilience. Originally proposed by (Breiman \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2001\u003c/span\u003e), random forest is a frequently used ensemble learning method applicable to both classification and regression tasks (Yan, Liu and Zhao \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). It consists of three main components: data sampling, decision tree training, and ensemble learning.\u003c/p\u003e \u003cp\u003eFirst, given a training dataset \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:D=\\{\\left({X}_{1},{Y}_{1}\\right),\\left({X}_{2},{Y}_{2}\\right),\\dots\\:,\\left({X}_{N},{Y}_{N}\\right)\\)\u003c/span\u003e\u003c/span\u003e}, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:k\\)\u003c/span\u003e\u003c/span\u003e different training subsets \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{D}_{1},{D}_{2},\\dots\\:,{D}_{\\text{k}}\\)\u003c/span\u003e\u003c/span\u003e are generated using bootstrap sampling, where each subset \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{D}_{i}\\)\u003c/span\u003e\u003c/span\u003e consists of \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:N\\)\u003c/span\u003e\u003c/span\u003e samples with replacement. The randomness in sample selection helps mitigate the overfitting issue.\u003c/p\u003e \u003cp\u003eSecond,\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\:k\\)\u003c/span\u003e\u003c/span\u003e decision trees are trained, with each decision tree \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{T}_{i}\\)\u003c/span\u003e\u003c/span\u003e using only its corresponding training subset \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{D}_{i}\\)\u003c/span\u003e\u003c/span\u003e. At each one, a random subset of \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:m\\)\u003c/span\u003e\u003c/span\u003e features (typically \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:m=\\sqrt{M}\\)\u003c/span\u003e\u003c/span\u003e, where \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:M\\)\u003c/span\u003e\u003c/span\u003e is the total number of features) is selected for splitting. Training continues until a stopping criterion, such as the maximum tree depth, is reached. The model\u0026rsquo;s performance is influenced by three key parameters: (1) tree depth, which affects individual tree performance; (2) the number of selected features, which controls correlations among the trees; and (3) the number of trees, which determines the forest size.\u003c/p\u003e \u003cp\u003eFinally, during prediction, each tree produces an independent output, and the final result is obtained by averaging (for regression) or majority voting (for classification). In this study, as a regression task, the final prediction is calculated as:\u003cdiv id=\"Equ2\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ2\" name=\"EquationSource\"\u003e\n$$\\:\\widehat{Y}=\\frac{1}{k}\\sum\\:_{i=1}^{k}{y}_{i}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e2\u003c/div\u003e\u003c/div\u003e;\u003c/p\u003e \u003cp\u003eThe random forest method offers several advantages. It effectively reduces variance compared to single decision trees, performs well on high-dimensional data, and provides feature importance evaluation (Menze et al. \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2009\u003c/span\u003e, Hastie et al. \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2009\u003c/span\u003e). It is robust to missing values and outliers, does not require data standardization (Liaw and Wiener \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2002\u003c/span\u003e) and does not assume a predefined relationship between independent and dependent variables, making it suitable for modeling complex non-linear patterns (Ho \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e1998\u003c/span\u003e). Additionally, its parallel processing capability enhances computational efficiency (Geurts, Ernst and Wehenkel \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2006\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eHowever, random forest has some limitations. Its complexity makes interpretation challenging compared to single decision trees (Louppe \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). It is also sensitive to highly correlated features, as redundant variables may be repeatedly selected (Strobl, Malley and Tutz \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2009\u003c/span\u003e). Moreover, it lacks statistical inference capabilities, such as significance tests or confidence intervals, which are common in traditional statistical models (Wright, Ziegler and K\u0026ouml;nig \u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). Despite these limitations, random forest remains a powerful tool for analyzing complex relationships and is increasingly used in spatial and transportation research.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"4. Results","content":"\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Model fit specifics\u003c/h2\u003e \u003cp\u003eFor examining and enhancing the generalizability of the random forest models, 70% of the grids were randomly selected to act as the training set, while the remaining 30% as the test set. Moreover, we tuned three key parameters, i.e., number of trees, number of splitting variables, and maximum depth of tree, to obtain optimal model performance and mitigate the overfitting issue. After tuning, the best number of trees turned out to be 500, and number of splitting variables and maximum depth of tree were 6 and 1000, respectively. The root mean square error (RMSE) and pseudo-R\u003csup\u003e2\u003c/sup\u003e of the final model are 11.343 and 0.365, indicating a satisfactory performance. The final model provides a basis for quantifying the relative importance of independent variables and drawing the partial dependence plots.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e4.2 Relative importance of independent variables\u003c/h2\u003e \u003cp\u003eIn machine learning, the relative importance quantifies the contribution of individual variable to the model's predictions, facilitating feature selection, model interpretation, and optimization. As shown in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e1\u003c/span\u003e, the relative importance of independent variables in predicting resilience of taxi ridership is ranked by contribution magnitude, with all values normalized to sum to 100%. Overall, the built environment plays a significantly greater role in the resilience of taxi ridership than the single socio-demographic factor, i.e., property price.\u003c/p\u003e \u003cp\u003eAmong built environment factors, distance to the city center is the most influential predictor of taxi ridership resilience, with a relative importance share of 41.05%, far surpassing other variables. This finding is expectable, as it reflects regional differences in travel demand, influences transportation accessibility, and shapes travel mode choices, all of which affect resilience of taxi ridership. Similarly, building density plays a crucial role (with a relative importance of 23.43%). Building density closely correlates with infrastructure completeness, traffic demand fluctuations, and mobility conditions, and hence is a key determinant of taxi ridership resilience. Other built environment variables, such as land use mix (3.61%), road density (2.75%), bus route density (2.47%), and intersection density (2.00%), only have marginal contributions to the model. Property price ranks third in relative importance. This is understandable given that property price serves as a vital indicator of residents\u0026rsquo; income levels and directly influences their flexibility in choosing travel modes. Particularly under extreme weather conditions, higher-income residents are more likely to opt for taxis as a convenient mode of transport, thereby affecting the resilience of taxi ridership (Pan, Shen and Zhao \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2013\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eRelative importance of independent variables.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIndependent variables\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRelative importance (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRanking\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSocio-demographics\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eProperty price\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e10.63%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eBuilt environment\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDistance to city center\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e41.05%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBuilding density\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e23.43%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJob density\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e7.78%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePopulation density\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e6.27%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLand use mix\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.61%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRoad density\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.75%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBus route density\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.47%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e8\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIntersection density\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.00%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e9\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSum\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e89.37%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e4.3 Non-linear relationships between built environment and resilience of taxi ridership\u003c/h2\u003e \u003cp\u003ePartial Dependence Plot (PDP) is a widely used visualization tool in machine learning that illustrates the marginal effects of individual features on model predictions. By holding all other variables constant, PDP reveals the unconstrained non-linear relationship between independent and dependent variables. Figure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e to Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e display the partial dependence plots of key independent variables in this study, where the Y-axes represent resilience of taxi ridership, and the X-axes denotes the values of a specific dependent variable. Overall, all variables exhibit non-linear relationships with resilience of taxi ridership, characterized by evident threshold effects.\u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e illustrates the non-linear relationship between property price and resilience of taxi ridership, revealing a distinct threshold effect in an \u0026ldquo;N\u0026rdquo;-shaped pattern. Specifically, when property prices range from 30,000 to 40,000 (RMB/m\u0026sup2;), the resilience of taxi ridership exhibits a sharp increase. Growth then slows, followed by a decline that reaches its lowest point at 60,000 (RMB/m\u0026sup2;), before gradually rebounding. Property prices can generally indicate income and affordability of people, thus indirectly affecting their transport resources and life patterns. Thereby, people reside or work in areas with relatively low property prices (in our case, areas with property prices lower than 35,000) are probably more reliable on cheaper transport modes, such as walking, electric bikes, and public transit, and their tendency of taking taxis can be quite susceptible to extreme events like rainstorms. This finding aligns with our expectation and is consistent with Hong et al. (\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), which suggests that socio-economic factors significantly influence resilience of taxi ridership, with lower-income groups facing greater challenges during disasters such as hurricanes due to limited resources. In terms of residents of areas with high property prices (higher than 55,000), they may have more flexible schedules and daily routines than others, and thus they can alter their travel behaviors more freely in accordance to disturbing events. These can explain the relatively low resilience of taxi ridership in the areas with low and high property prices.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e visualizes the correlations between resilience of taxi ridership and two built environment variables, i.e., distance to the city center and building density. Both the correlations are generally positive, albeit to distinct thresholds. When distance to the city center is between 0 and 5 km, resilience of taxi ridership remains low and stable with no significant fluctuations. This finding is interesting and yet understandable. This study identifies Zhongshan Road as the city center of Xiamen. Zhongshan Road is indeed the birthplace of Xiamen City. However, in recent years, Zhongshan Road has increasingly evolved into a tourism and commerce center of Xiamen. Since touring and shopping are two typical types of non-compulsory travel purposes, during extreme weather events, tourists and shoppers can easily modify their travel behaviors, which can partly explain the lower resilience of taxi ridership in areas near Zhongshan Road. Meanwhile, urban central areas are oftentimes characterized with more easily accessible transportation modes, such as public transit, which can effectively substitute taxis in the face of extreme events. Moreover, these areas may be more congestion prone. These can also contribute to the relatively low resilience of areas with proximity to city center. Yet, once the distance exceeds 5 km, the resilience of taxi ridership sharply increases, stabilizing around 10 km. At approximately 14 km from the city center, resilience of taxi ridership begins to decline gradually. This is consistent with the finding (Zhang and Li \u003cspan citationid=\"CR73\" class=\"CitationRef\"\u003e2020\u003c/span\u003e), which show that regions farther from the city center are less affected by heavy rainfall across different types of functional areas.\u003c/p\u003e \u003cp\u003eThe relationship between building density and resilience of taxi ridership is relatively straightforward, presenting a nearly linear shape when building density is below 2000 (m\u0026sup2;/km\u0026sup2;) and stabilizing with minimal fluctuations afterwards. Low building density areas typically feature low-level development and more dispersed urban functions, thus resulting in fewer transportation demand. In contrast, areas with higher building density can usually generate and attract more stable taxi demands.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e shows the non-linear relationship between population density, job density, and the resilience of taxi ridership. Both variables exhibit a positive correlation with resilience of taxi ridership. Specifically, when population density is below 390 per m\u0026sup2;, resilience of taxi ridership increases almost monotonically. Afterwards, taxi ridership stays stable until around 600 per m\u0026sup2; where it experiences a slight increase and keeps steady. It is worth noting that data is sparse when population density exceeds around 400, and thus the results obtained therein can be less reliable. When job density is below approximately 4,000, resilience of taxi ridership undergoes a sharp growth; afterwards, the effects of job density on taxi ridership density seem to be unimportant. The positive effects of population and job density on taxi ridership density align with our expectation and prior studies (Hao and Wang \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2022\u003c/span\u003e, Irajifar et al. \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2016\u003c/span\u003e), since high population or job density mean high and stable transportation demand. However, areas with high population and/or job density are usually characterized with high traffic flow and thus are easily congested. When population and/or job density arrive at certain thresholds, the negative effects of resulting congestion may outrun the positive effects of compacity, particularly during rainstorms.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigure\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e depicts the associations between land use mix, road network density, and taxi ridership resilience. Generally, the former has positive impacts, while the latter negative ones. In the range from 0 to around 0.5, despite some fluctuations, land use mix has a nearly linear relationship with taxi ridership resilience. This finding is reasonable because higher land use mix indicates that diverse urban functions are more closely integrated, which can generate lasting transportation needs and thus higher taxi ridership resilience. After that, the relationship generally stays positive, but the increasing rate gets smaller.\u003c/p\u003e \u003cp\u003eWhen road density is within the range from 0 to 17 (km/km\u0026sup2;), resilience of taxi ridership fluctuates slightly at the high level. However, once road density drops below 17, resilience of taxi ridership declines sharply and then stabilizes at around 21 (km/km\u0026sup2;). In areas with low road network density, due to the lack of transportation options, the demands for taxis can be relatively high and stable, and those demands will probably last during extreme weather events, such as rainstorms, resulting in higher resilience of taxi ridership. By contrast, high-road-density districts can be characterized with diverse transportation alternatives on the one hand, and more susceptible to congestions, especially in the face of extreme weather events, on the other hand. Hence, the taxi demands therein can be lower and more unstable, aligning with prior findings (Liu et al., 2020).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAs shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e, bus route density and road intersection density have similar increase-then-decrease associations with resilience of taxi ridership. In the range of 0 to 120 (routes/km\u003csup\u003e2\u003c/sup\u003e), bus route density has nearly linear positive effects of taxi ridership; after that, resilience of taxi ridership declines slowly as bus route density grows. This is an interesting finding, which shows that within certain threshold, increasing public transit coverage can also lead to more resilient taxi demands. Higher public transit coverage can enhance the transportation accessibility of a certain area, and hence bring about more human activities and thus higher and more steady taxi demand, i.e., higher taxi ridership resilience. However, exceeding certain threshold, the substitution effects of public transit together with the resulting congestion may prevail. Similarly, when intersection density is in the range of 0 to 25 (per km\u003csup\u003e2\u003c/sup\u003e), taxi ridership climbs dramatically; afterwards, it declines with some fluctuations. Intersection density indicates the connectedness of an area. Higher the intersection density, more connected the area. Therefore, the positive relationship between intersection density and taxi ridership resilience is understandable. Yet, excessively high intersection density may lessen the traffic efficiency and probably lead to congestions.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e4.4 Synergistic effects between built environment and resilience of taxi ridership\u003c/h2\u003e \u003cp\u003eA partial dependence plot can clearly demonstrate the non-linear association between dependent variable and a certain independent variable with other variables are controlled. However, it is widely confirmed that independent variables can interact with one another to strengthen each other\u0026rsquo;s effects, resulting synergistic effects. That is to say, when certain variables change together, their combined effects can be larger than the mere sum of their individual effect. We use two-dimensional partial dependence plots (2D-PDPs) to present the synergistic effects between several representative built-environment variables, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e. In the figure, the X-axis and Y-axis represent the variables that interact with each other, while the color gradient illustrates taxi ridership resilience, with yellow indicating higher resilience and dark blue indicating lower resilience.\u003c/p\u003e \u003cp\u003eAs shown, most synergistic effects are achieved when the two variables are in their high-value ranges at the same time. For example, when distance to the city center is within the range from 10 to 15 km, the higher building density (or population density), the higher resilience of taxi ridership. Similarly, when population density and land use mix are in their higher range (above 600 per hm\u003csup\u003e2\u003c/sup\u003e and 0.4 respectively) at the same time, the areas are characterized with higher taxi ridership resilience. However, inconsistencies exist. When property price is in its low to medium range (35,000 to 52,000 Yuan/m\u003csup\u003e2\u003c/sup\u003e), the higher building density, the higher taxi ridership resilience. When land use mix is at a higher level and yet intersection density at a low level, resilience of taxi ridership is high.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"5. Discussion","content":"\u003cp\u003eUnder the backdrop of increasing intensity and frequency of extreme weather events brought about by climate change, policymakers, researchers, and practitioners have been striving to formulate countermeasures, among which mitigation and adaptation are two types of widely adopted strategies (Wang et al. \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Comparatively, adaptation strategies, i.e., learning to cope with the impacts of climate change and the consequent extreme weather events (like rainstorms), decreasing vulnerability, and enhancing resilience, are more urgent and time- and resource-saving. Taxis being an important part of urban transportation system, resilience of taxi ridership can effectively epitomize the variations of travel behavior of urban residents and resisting and recovering capacity of transportation facilities and infrastructures. Taking a deeper look into taxi ridership resilience and its influencing factors can provide evidence base for promoting transportation resilience and better satisfying urban residents\u0026rsquo; transportation demand during disruptive weather events.\u003c/p\u003e \u003cp\u003ePrior studies have established the associations between taxi ridership and built environment features and socioeconomic factors (Lyu et al. \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2023\u003c/span\u003e, Zhu et al. \u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e2024\u003c/span\u003e, Chen et al. \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). However, how taxi ridership is affected by and recover from extreme weather events has been largely ignored. This study, to the extent of our knowledge, is among the first ones to examine taxi ridership resilience in the face of such disruptive events and its relationships with built environment. We obtain some interesting findings in this study.\u003c/p\u003e \u003cp\u003eFirst, we find that among the influencing factors, distance to city center, building density, and property prices are the most important contributors of resilience of taxi ridership. Distance to city center, as a frequently used indicator of location, has been confirmed as an all-important determinant of diverse travel behavior dimensions, such as transport mode choice, commuting duration, and driving distance (Liu and Xiao \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2024\u003c/span\u003e, Ding, Cao and N\u0026aelig;ss \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2018\u003c/span\u003e, Liu and Xiao \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Our study further corroborates the vital contribution of location for travel behavior resilience, implying that we can promote resilience during extreme weather events through adjusting regional spatial structure and enhancing regional connectivity. Interestingly, building density is found to be a much more important variable for taxi ridership resilience than population density, which can be due to that there may be neither high nor stable taxi demands in some areas characterized with high population density, e.g., factory dormitory and urban village (Liu et al. \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2020a\u003c/span\u003e). This finding suggests that development intensity is also a feasible intervention target for promotion of taxi ridership resilience. Property price plays a critical role, which is expectable since property price could not only suggest the affordability and travel resources and patterns of residents in certain areas, but also closely relate with the location, land use compositions, and physical environment quality. The hierarchy of relative importance of the independent variables remind the policy makers that intervention measures on such built environment dimensions as regional accessibility/connectivity, development intensity, and urban functional diversity can be more efficient.\u003c/p\u003e \u003cp\u003eSecond, we observe that all the influencing factors selected in this study are associated with resilience of taxi ridership in a non-linear pattern, with obvious thresholds present, which provides fruitful theoretical and practical implications. Roughly, the shapes of the abovementioned associations can be divided into three categories, including increase-fluctuate-decrease, increase-fluctuate-increase, and fluctuate-decrease. Property price, distance to city center, bus route density, and intersection density belong to the first category, i.e., increase-fluctuate-decrease. Building density, population density, job density, and land use correlate with taxi ridership resilience mix in an increase-fluctuate-increase pattern, while road density belongs to the third category. Most of the relationships uncovered are in accordance with our expectations or findings of prior studies, in spite of some counterintuitive findings as we further discuss as follows. As revealed, areas with medium-to-high-level property price tend to be characterized with the highest resilience of taxi ridership. This finding can be understood from the perspectives of both people and space. Property price being a reliable indicator of income, people living in areas with medium-to-high property prices may have more fixed schedule than people living in high-property-price areas, while higher travel budgets than people in low-price areas. Meanwhile, as discovered by prior studies (Yang et al. \u003cspan citationid=\"CR68\" class=\"CitationRef\"\u003e2020\u003c/span\u003e, Yang, Zhou and Shyr \u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e2019\u003c/span\u003e, Xiao, Orford and Webster \u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e2016\u003c/span\u003e), property prices are significantly associated with transportation accessibility and physical environment features; medium-to-high-property-price areas may have more agreeable accessibility and built-environment features for taxis. Also, areas with a distance of 10\u0026ndash;12 km to the city center, rather than the city center itself, have the highest resilience of taxi ridership, suggesting that these areas are evolving into sub-centers with different functions from those of the city center we selected in this study (i.e., Zhongshan Road), and these functions have generated and attracted more stable taxi trips during rainstorms.\u003c/p\u003e \u003cp\u003eThese salient non-linear relationships, together with the thresholds, provide plentiful policy implications. For instance, the most effective range for population density is below around 380 (/hm\u003csup\u003e2\u003c/sup\u003e), while those for building density and land use mix are below 2,000 (/km\u003csup\u003e2\u003c/sup\u003e) and below 0.5, individually. It is best to control road density under 17.5 (km/km\u003csup\u003e2\u003c/sup\u003e). Beyond these ranges, the interventions may be inefficient or even detrimental for resilience promotion. It is worth noting that these ranges may be highly dependent on the contexts, which, however, can enlighten the decision makers that caution should be taken when intervening with the built environment and that over-diversification and excessively high density should be avoided. Moreover, these thresholds and ranges provide transportation operators and urban residents insights regarding transportation resources allocation and individual travel decision-making. Specifically, during extreme weather events, taxi or ride-hailing companies can dispatch more vehicles to areas with certain built environment features (e.g., concentrating urban functions like business, high building density, medium road density) to satisfy the lasting demand therein. Urban residents can also select their travel destinations or making travel plans according to those features.\u003c/p\u003e \u003cp\u003eThird, the synergistic effects uncovered among certain influencing factors echo the views of Ewing and Cervero (\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2017\u003c/span\u003e) and Handy (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2017\u003c/span\u003e) that by working synergistically, the combined effects of built environment variables can be quite large. These effects imply that modifying some variables simultaneously (e.g., population density \u0026amp; land use mix, population density \u0026amp; job density, and land use mix \u0026amp; intersection density) could be more efficient than modifying individual variables for resilience promotion. They also warn the policy makers against the collateral effects of their actions in intervening with the built environment for enhancing travel behavior resilience.\u003c/p\u003e"},{"header":"6. Conclusions","content":"\u003cp\u003eEmploying random forest method, this study examines resilience of taxi ridership in the face of a typical extreme weather event, i.e., rainstorm, and its socioeconomical and built environment influencing factors. Our findings disclose the relative importance and complex effects of those influencing factors. As revealed, distance to city center, building density, and property price are the most important contributors of taxi ridership resilience, followed by job and population density, and other variables seem less critical. Meanwhile, all the independent variables are found to have salient non-linear relationships with resilience of taxi ridership, with obvious threshold effects. Distance to city center, land use mix, and density of building, population, job, bus route, and intersection have generally positive associations with taxi ridership resilience; road density has generally negative effects; and relationship between taxi ridership resilience and property price presents a complicated increase-fluctuate-decrease pattern. Moreover, synergistic effects exist between certain independent variables, e.g., population density \u0026amp; land use mix. These findings build a holistic and thorough view of resilience of taxi ridership, providing abundant theoretical and practical insights.\u003c/p\u003e \u003cp\u003eThis study still has some room for further enhancement. First, due to the lack of data sources, we did not involve some key socioeconomic variables, such as (average) personal or household income, into our analyses. Instead, we used property price as a proxy. Although property price is widely proven to be capable of representing people\u0026rsquo;s socioeconomic status feature, future studies are recommended to investigate the effects of such variables as income and transport expense budget directly. Second, we employed a rainstorm as the example for extreme weather events. Although we believe that rainstorm is representative enough, we acknowledge the differences between rainstorm and other disruptive events, e.g., heatwave and public health crisis. Hence, comparative studies are necessary for drawing generalized conclusions, which also points to the future research direction.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eJ.L. and B.W. conceptualized the study. H.W. and L.X. developed the methodology. H.W. wrote the original draft of the manuscript. J.L. and B.W. reviewed and edited the manuscript. J.L. conducted the investigation. H.W. conducted the formal analysis. J.L. and B.W. were responsible for supervision, project administration, and funding acquisition. All authors have read and agreed to the final version of the manuscript.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eData are available on request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eBian, R. R., C. G. Wilmot \u0026amp; L. Wang (2019) Estimating spatio-temporal variations of taxi ridership caused by Hurricanes Irene and Sandy: A case study of New York City. \u003cem\u003eTransportation Research Part D: Transport and Environment,\u003c/em\u003e 77\u003cstrong\u003e,\u003c/strong\u003e 627-638.\u003c/li\u003e\n\u003cli\u003eBreiman, L. (2001) Random forests. \u003cem\u003eMachine learning,\u003c/em\u003e 45\u003cstrong\u003e,\u003c/strong\u003e 5-32.\u003c/li\u003e\n\u003cli\u003eCervero, R. (2013) Linking urban transport and land use in developing countries. \u003cem\u003eJournal of transport and land use,\u003c/em\u003e 6\u003cstrong\u003e,\u003c/strong\u003e 7-24.\u003c/li\u003e\n\u003cli\u003eChan, R. \u0026amp; J. L. Schofer (2016) Measuring transportation system resilience: Response of rail transit to weather disruptions. \u003cem\u003eNatural Hazards Review,\u003c/em\u003e 17\u003cstrong\u003e,\u003c/strong\u003e 05015004.\u003c/li\u003e\n\u003cli\u003eChen, C., T. Feng, C. Ding, B. Yu \u0026amp; B. Yao (2021) Examining the spatial-temporal relationship between urban built environment and taxi ridership: Results of a semi-parametric GWPR model. \u003cem\u003eJournal of Transport Geography,\u003c/em\u003e 96\u003cstrong\u003e,\u003c/strong\u003e 103172.\u003c/li\u003e\n\u003cli\u003eChen, R., L. Liu \u0026amp; Y. Gao (2024) The Association between Rainfall and Taxi Travel Activities: A Case Study from Wuhan, China. \u003cem\u003eJournal of Advanced Transportation,\u003c/em\u003e 2024\u003cstrong\u003e,\u003c/strong\u003e 1404168.\u003c/li\u003e\n\u003cli\u003eda Mata Martins, M. C., A. N. R. da Silva \u0026amp; N. Pinto (2019) An indicator-based methodology for assessing resilience in urban mobility. \u003cem\u003eTransportation Research Part D: Transport and Environment,\u003c/em\u003e 77\u003cstrong\u003e,\u003c/strong\u003e 352-363.\u003c/li\u003e\n\u003cli\u003eDing, C., X. J. Cao \u0026amp; P. N\u0026aelig;ss (2018) Applying gradient boosting decision trees to examine non-linear effects of the built environment on driving distance in Oslo. \u003cem\u003eTransportation Research Part A: Policy and Practice,\u003c/em\u003e 110\u003cstrong\u003e,\u003c/strong\u003e 107-117.\u003c/li\u003e\n\u003cli\u003eEwing, R. \u0026amp; R. Cervero (2001) Travel and the built environment: a synthesis. \u003cem\u003eTransportation research record,\u003c/em\u003e 1780\u003cstrong\u003e,\u003c/strong\u003e 87-114.\u003c/li\u003e\n\u003cli\u003e-- (2010) Travel and the built environment: A meta-analysis. \u003cem\u003eJournal of the American planning association,\u003c/em\u003e 76\u003cstrong\u003e,\u003c/strong\u003e 265-294.\u003c/li\u003e\n\u003cli\u003e-- (2017) \u0026ldquo;Does compact development make people drive less?\u0026rdquo; The answer is yes. \u003cem\u003eJournal of the American Planning Association,\u003c/em\u003e 83\u003cstrong\u003e,\u003c/strong\u003e 19-25.\u003c/li\u003e\n\u003cli\u003eGan, Z., M. Yang, T. Feng \u0026amp; H. J. Timmermans (2020) Examining the relationship between built environment and metro ridership at station-to-station level. \u003cem\u003eTransportation Research Part D: Transport and Environment,\u003c/em\u003e 82\u003cstrong\u003e,\u003c/strong\u003e 102332.\u003c/li\u003e\n\u003cli\u003eGeurts, P., D. Ernst \u0026amp; L. Wehenkel (2006) Extremely randomized trees. \u003cem\u003eMachine learning,\u003c/em\u003e 63\u003cstrong\u003e,\u003c/strong\u003e 3-42.\u003c/li\u003e\n\u003cli\u003eHandy, S. (2017) Thoughts on the meaning of Mark Stevens\u0026rsquo;s meta-analysis. \u003cem\u003eJournal of the American Planning Association,\u003c/em\u003e 83\u003cstrong\u003e,\u003c/strong\u003e 26-28.\u003c/li\u003e\n\u003cli\u003eHao, H. \u0026amp; Y. Wang (2022) Disentangling relations between urban form and urban accessibility for resilience to extreme weather and climate events. \u003cem\u003eLandscape and urban planning,\u003c/em\u003e 220\u003cstrong\u003e,\u003c/strong\u003e 104352.\u003c/li\u003e\n\u003cli\u003eHastie, T., R. Tibshirani, J. H. Friedman \u0026amp; J. H. Friedman. 2009. \u003cem\u003eThe elements of statistical learning: data mining, inference, and prediction\u003c/em\u003e. Springer.\u003c/li\u003e\n\u003cli\u003eHe, Q., D. Rowangould, A. Karner, M. Palm \u0026amp; S. LaRue (2022) Covid-19 pandemic impacts on essential transit riders: findings from a US Survey. \u003cem\u003eTransportation Research Part D: Transport and Environment,\u003c/em\u003e 105\u003cstrong\u003e,\u003c/strong\u003e 103217.\u003c/li\u003e\n\u003cli\u003eHo, T. K. (1998) The random subspace method for constructing decision forests. \u003cem\u003eIEEE transactions on pattern analysis and machine intelligence,\u003c/em\u003e 20\u003cstrong\u003e,\u003c/strong\u003e 832-844.\u003c/li\u003e\n\u003cli\u003eHolling, C. S. 1973. Resilience and stability of ecological systems. International Institute for Applied Systems Analysis Laxenburg.\u003c/li\u003e\n\u003cli\u003eHong, B., B. J. Bonczak, A. Gupta \u0026amp; C. E. Kontokosta (2021) Measuring inequality in community resilience to natural disasters using large-scale mobility data. \u003cem\u003eNATURE COMMUNICATIONS,\u003c/em\u003e 12.\u003c/li\u003e\n\u003cli\u003eHu, S. \u0026amp; P. Chen (2021) Who left riding transit? Examining socioeconomic disparities in the impact of COVID-19 on ridership. \u003cem\u003eTransportation Research Part D: Transport and Environment,\u003c/em\u003e 90\u003cstrong\u003e,\u003c/strong\u003e 102654.\u003c/li\u003e\n\u003cli\u003eHuang, J. \u0026amp; J. Wang (2024) Travel behavior resilience: Theory, method, and empirical studies. \u003cem\u003eJournal of Geographical Sciences,\u003c/em\u003e 34\u003cstrong\u003e,\u003c/strong\u003e 1657-1674.\u003c/li\u003e\n\u003cli\u003eHuang, J., T. Zhang, D. Sun \u0026amp; H. Wang (2025) Urban human mobility changes based on functional areas during extreme rainstorm event: A case of Beijing \u0026ldquo;23\u0026middot; 7\u0026rdquo; rainstorm event. \u003cem\u003eCities,\u003c/em\u003e 163\u003cstrong\u003e,\u003c/strong\u003e 106003.\u003c/li\u003e\n\u003cli\u003eIrajifar, L., N. Sipe \u0026amp; T. Alizadeh (2016) The impact of urban form on disaster resiliency: A case study of Brisbane and Ipswich, Australia. \u003cem\u003eInternational Journal of Disaster Resilience in the Built Environment,\u003c/em\u003e 7\u003cstrong\u003e,\u003c/strong\u003e 259-275.\u003c/li\u003e\n\u003cli\u003eKhavarian-Garmsir, A. R., A. Sharifi \u0026amp; N. Moradpour (2021) Are high-density districts more vulnerable to the COVID-19 pandemic? \u003cem\u003eSustainable Cities and Society,\u003c/em\u003e 70\u003cstrong\u003e,\u003c/strong\u003e 102911.\u003c/li\u003e\n\u003cli\u003eLeComte, D. (2020) A decade of extremes: The top-10 record-breaking events of recent years. \u003cem\u003eWeatherwise,\u003c/em\u003e 73\u003cstrong\u003e,\u003c/strong\u003e 14-23.\u003c/li\u003e\n\u003cli\u003eLepage, S. \u0026amp; C. Morency (2021) Impact of weather, activities, and service disruptions on transportation demand. \u003cem\u003eTransportation research record,\u003c/em\u003e 2675\u003cstrong\u003e,\u003c/strong\u003e 294-304.\u003c/li\u003e\n\u003cli\u003eLi, Z., X. Wang, F. Gao, J. Tang \u0026amp; H. Xu (2024) Analysis of mobility patterns for urban taxi ridership: the role of the built environment. \u003cem\u003eTransportation,\u003c/em\u003e 51\u003cstrong\u003e,\u003c/strong\u003e 1409-1431.\u003c/li\u003e\n\u003cli\u003eLiaw, A. \u0026amp; M. Wiener (2002) Classification and regression by randomForest. \u003cem\u003eR news,\u003c/em\u003e 2\u003cstrong\u003e,\u003c/strong\u003e 18-22.\u003c/li\u003e\n\u003cli\u003eLiu, J., J. Cui, L. Xiao, D. Lin \u0026amp; L. Yang (2025) Non-linear built environment effects on travel behavior resilience under extreme weather events. \u003cem\u003eTransportation Research Part D: Transport and Environment,\u003c/em\u003e 143\u003cstrong\u003e,\u003c/strong\u003e 104753.\u003c/li\u003e\n\u003cli\u003eLiu, J. \u0026amp; L. Xiao (2023) Non-linear relationships between built environment and commuting duration of migrants and locals. \u003cem\u003eJournal of Transport Geography,\u003c/em\u003e 106\u003cstrong\u003e,\u003c/strong\u003e 103517.\u003c/li\u003e\n\u003cli\u003e-- (2024) Socioeconomic differences in effect size: predicting commuting mode choice of migrants and locals using a light gradient boosting approach. \u003cem\u003eTransportation,\u003c/em\u003e 51\u003cstrong\u003e,\u003c/strong\u003e 1-24.\u003c/li\u003e\n\u003cli\u003eLiu, J., L. Xiao, L. Yang \u0026amp; J. Zhou (2020a) A tale of two social groups in Xiamen, China: Trip frequency of migrants and locals and its determinants. \u003cem\u003eTravel Behaviour and Society,\u003c/em\u003e 20\u003cstrong\u003e,\u003c/strong\u003e 213-224.\u003c/li\u003e\n\u003cli\u003eLiu, Q., C. Ding \u0026amp; P. Chen (2020b) A panel analysis of the effect of the urban environment on the spatiotemporal pattern of taxi demand. \u003cem\u003eTravel Behaviour and Society,\u003c/em\u003e 18\u003cstrong\u003e,\u003c/strong\u003e 29-36.\u003c/li\u003e\n\u003cli\u003eLiu, Y., Y. Li, W. Yang \u0026amp; J. Hu (2023) Exploring nonlinear effects of built environment on jogging behavior using random forest. \u003cem\u003eApplied geography,\u003c/em\u003e 156\u003cstrong\u003e,\u003c/strong\u003e 102990.\u003c/li\u003e\n\u003cli\u003eLouppe, G. 2014. Understanding random forests: From theory to practice. Universite de Liege (Belgium).\u003c/li\u003e\n\u003cli\u003eLyu, T., Y. Wang, S. Ji, T. Feng \u0026amp; Z. Wu (2023) A multiscale spatial analysis of taxi ridership. \u003cem\u003eJournal of Transport Geography,\u003c/em\u003e 113\u003cstrong\u003e,\u003c/strong\u003e 103718.\u003c/li\u003e\n\u003cli\u003eMeerow, S., J. P. Newell \u0026amp; M. Stults (2016) Defining urban resilience: A review. \u003cem\u003eLandscape and urban planning,\u003c/em\u003e 147\u003cstrong\u003e,\u003c/strong\u003e 38-49.\u003c/li\u003e\n\u003cli\u003eMenze, B. H., B. M. Kelm, R. Masuch, U. Himmelreich, P. Bachert, W. Petrich \u0026amp; F. A. Hamprecht (2009) A comparison of random forest and its Gini importance with standard chemometric methods for the feature selection and classification of spectral data. \u003cem\u003eBMC bioinformatics,\u003c/em\u003e 10\u003cstrong\u003e,\u003c/strong\u003e 1-16.\u003c/li\u003e\n\u003cli\u003eOlsson, L., A. Jerneck, H. Thoren, J. Persson \u0026amp; D. O\u0026rsquo;Byrne (2015) Why resilience is unappealing to social science: Theoretical and empirical investigations of the scientific use of resilience. \u003cem\u003eScience advances,\u003c/em\u003e 1\u003cstrong\u003e,\u003c/strong\u003e e1400217.\u003c/li\u003e\n\u003cli\u003ePan, H., Q. Shen \u0026amp; T. Zhao (2013) Travel and Car Ownership of Residents near New Suburban Metro Stations in Shanghai, China. \u003cem\u003eTransportation Research Record Journal of the Transportation Research Board,\u003c/em\u003e 2394\u003cstrong\u003e,\u003c/strong\u003e 63-69.\u003c/li\u003e\n\u003cli\u003ePeng, Y., J. Liu, F. Li, J. Cui, Y. Lu \u0026amp; L. Yang (2024) Resilience of ride-hailing services in response to air pollution and its association with built-environment and socioeconomic characteristics. \u003cem\u003eJournal of Transport Geography,\u003c/em\u003e 120\u003cstrong\u003e,\u003c/strong\u003e 103971.\u003c/li\u003e\n\u003cli\u003eQian, X. \u0026amp; S. V. Ukkusuri (2015) Spatial variation of the urban taxi ridership using GPS data. \u003cem\u003eApplied geography,\u003c/em\u003e 59\u003cstrong\u003e,\u003c/strong\u003e 31-42.\u003c/li\u003e\n\u003cli\u003eQiao, M., M. Haraguchi \u0026amp; U. Lall (2025) Enhancing urban resilience to extreme weather: The roles of human transition paths among multiple transportation modes. \u003cem\u003eInternational Journal of Geographical Information Science,\u003c/em\u003e 39\u003cstrong\u003e,\u003c/strong\u003e 2408-2427.\u003c/li\u003e\n\u003cli\u003eRuan, J., Y. Chen \u0026amp; Z. Yang (2021) Assessment of temporal and spatial progress of urban resilience in Guangzhou under rainstorm scenarios. \u003cem\u003eInternational Journal of Disaster Risk Reduction,\u003c/em\u003e 66\u003cstrong\u003e,\u003c/strong\u003e 102578.\u003c/li\u003e\n\u003cli\u003eSharifi, A. (2019a) Resilient urban forms: A macro-scale analysis. \u003cem\u003eCities,\u003c/em\u003e 85\u003cstrong\u003e,\u003c/strong\u003e 1-14.\u003c/li\u003e\n\u003cli\u003e-- (2019b) Urban form resilience: A meso-scale analysis. \u003cem\u003eCities,\u003c/em\u003e 93\u003cstrong\u003e,\u003c/strong\u003e 238-252.\u003c/li\u003e\n\u003cli\u003eSolomon, S. (2007) Contribution of Working Group I to the fourth assessment report of the Intergovernmental Panel on Climate Change 2007. \u003cem\u003e(No Title)\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eStrobl, C., J. Malley \u0026amp; G. Tutz (2009) An introduction to recursive partitioning: rationale, application, and characteristics of classification and regression trees, bagging, and random forests. \u003cem\u003ePsychological methods,\u003c/em\u003e 14\u003cstrong\u003e,\u003c/strong\u003e 323.\u003c/li\u003e\n\u003cli\u003eSun, D. J., K. Zhang \u0026amp; S. Shen (2018) Analyzing spatiotemporal traffic line source emissions based on massive didi online car-hailing service data. \u003cem\u003eTransportation Research Part D: Transport and Environment,\u003c/em\u003e 62\u003cstrong\u003e,\u003c/strong\u003e 699-714.\u003c/li\u003e\n\u003cli\u003eTang, J., P. Zhao, Z. Gong, H. Zhao, F. Huang, J. Li, Z. Chen, L. Yu \u0026amp; J. Chen (2023) Resilience patterns of human mobility in response to extreme urban floods. \u003cem\u003eNational science review,\u003c/em\u003e 10\u003cstrong\u003e,\u003c/strong\u003e nwad097.\u003c/li\u003e\n\u003cli\u003eTongyue, L., N. Plnyi \u0026amp; G. U. Chaolin (2014) A Review on Research Frameworks of Resilient Cities. \u003cem\u003eUrban Planning Forum\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eUtsumi, N. \u0026amp; H. Kim (2022) Observed influence of anthropogenic climate change on tropical cyclone heavy rainfall. \u003cem\u003eNature Climate Change,\u003c/em\u003e 12\u003cstrong\u003e,\u003c/strong\u003e 436-440.\u003c/li\u003e\n\u003cli\u003eVergel-Tovar, C. E. \u0026amp; D. A. Rodriguez (2018) The ridership performance of the built environment for BRT systems: Evidence from Latin America. \u003cem\u003eJournal of transport geography,\u003c/em\u003e 73\u003cstrong\u003e,\u003c/strong\u003e 172-184.\u003c/li\u003e\n\u003cli\u003eWang, B., B. P. Loo, F. Zhen \u0026amp; G. Xi (2020) Urban resilience from the lens of social media data: Responses to urban flooding in Nanjing, China. \u003cem\u003eCities,\u003c/em\u003e 106\u003cstrong\u003e,\u003c/strong\u003e 102884.\u003c/li\u003e\n\u003cli\u003eWang, B., F. Zhang, J. Liu \u0026amp; Z. Tan (2024) The impacts of extreme hot weather on metro ridership: A case study of Shenzhen, China. \u003cem\u003eJournal of Transport Geography,\u003c/em\u003e 117\u003cstrong\u003e,\u003c/strong\u003e 103899.\u003c/li\u003e\n\u003cli\u003eWang, F., J. D. Harindintwali, K. Wei, Y. Shan, Z. Mi, M. J. Costello, S. Grunwald, Z. Feng, F. Wang \u0026amp; Y. Guo (2023) Climate change: Strategies for mitigation and adaptation. \u003cem\u003eThe Innovation Geoscience,\u003c/em\u003e 1\u003cstrong\u003e,\u003c/strong\u003e 100015-1-100015-37.\u003c/li\u003e\n\u003cli\u003eWang, J., J. Huang, H. Yang \u0026amp; D. Levinson (2022) Resilience and recovery of public transport use during COVID-19. \u003cem\u003eNpj Urban Sustainability,\u003c/em\u003e 2\u003cstrong\u003e,\u003c/strong\u003e 18.\u003c/li\u003e\n\u003cli\u003eWang, P., Y. Li \u0026amp; Y. Zhang (2021) An urban system perspective on urban flood resilience using SEM: evidence from Nanjing city, China. \u003cem\u003eNatural hazards,\u003c/em\u003e 109\u003cstrong\u003e,\u003c/strong\u003e 2575-2599.\u003c/li\u003e\n\u003cli\u003eWright, M. N., A. Ziegler \u0026amp; I. R. K\u0026ouml;nig (2016) Do little interactions get lost in dark random forests? \u003cem\u003eBMC bioinformatics,\u003c/em\u003e 17\u003cstrong\u003e,\u003c/strong\u003e 1-10.\u003c/li\u003e\n\u003cli\u003eXiao, W. \u0026amp; Y. D. Wei (2023) Assess the non-linear relationship between built environment and active travel around light-rail transit stations. \u003cem\u003eApplied geography,\u003c/em\u003e 151\u003cstrong\u003e,\u003c/strong\u003e 102862.\u003c/li\u003e\n\u003cli\u003eXiao, W., Y. D. Wei \u0026amp; Y. Wu (2022) Neighborhood, built environment and resilience in transportation during the COVID-19 pandemic. \u003cem\u003eTransportation Research Part D: Transport and Environment,\u003c/em\u003e 110\u003cstrong\u003e,\u003c/strong\u003e 103428.\u003c/li\u003e\n\u003cli\u003eXiao, Y., S. Orford \u0026amp; C. J. Webster (2016) Urban configuration, accessibility, and property prices: A case study of Cardiff, Wales. \u003cem\u003eEnvironment and Planning B: Planning and Design,\u003c/em\u003e 43\u003cstrong\u003e,\u003c/strong\u003e 108-129.\u003c/li\u003e\n\u003cli\u003eYan, X., X. Liu \u0026amp; X. Zhao (2020) Using machine learning for direct demand modeling of ridesourcing services in Chicago. \u003cem\u003eJournal of Transport Geography,\u003c/em\u003e 83\u003cstrong\u003e,\u003c/strong\u003e 102661.\u003c/li\u003e\n\u003cli\u003eYang, C. \u0026amp; E. J. Gonzales (2014) Modeling taxi trip demand by time of day in New York City. \u003cem\u003eTransportation Research Record,\u003c/em\u003e 2429\u003cstrong\u003e,\u003c/strong\u003e 110-120.\u003c/li\u003e\n\u003cli\u003eYang, H., J. Huo, R. Pan, K. Xie, W. Zhang \u0026amp; X. Luo (2022) Exploring built environment factors that influence the market share of ridesourcing service. \u003cem\u003eApplied Geography,\u003c/em\u003e 142\u003cstrong\u003e,\u003c/strong\u003e 102699.\u003c/li\u003e\n\u003cli\u003eYang, H., J. Peng, Y. Lu, J. Wang \u0026amp; X. Yan (2024) Nonlinear impact of built environment on people with disabilities\u0026rsquo; metro use behavior. \u003cem\u003eApplied Geography,\u003c/em\u003e 169\u003cstrong\u003e,\u003c/strong\u003e 103323.\u003c/li\u003e\n\u003cli\u003eYang, L., X. Chu, Z. Gou, H. Yang, Y. Lu \u0026amp; W. Huang (2020) Accessibility and proximity effects of bus rapid transit on housing prices: Heterogeneity across price quantiles and space. \u003cem\u003eJournal of Transport Geography,\u003c/em\u003e 88\u003cstrong\u003e,\u003c/strong\u003e 102850.\u003c/li\u003e\n\u003cli\u003eYang, L., J. Zhou \u0026amp; O. F. Shyr (2019) Does bus accessibility affect property prices? \u003cem\u003eCities,\u003c/em\u003e 84\u003cstrong\u003e,\u003c/strong\u003e 56-65.\u003c/li\u003e\n\u003cli\u003eYu, H. \u0026amp; Z.-R. Peng (2019) Exploring the spatial variation of ridesourcing demand and its relationship to built environment and socioeconomic factors with the geographically weighted Poisson regression. \u003cem\u003eJournal of Transport Geography,\u003c/em\u003e 75\u003cstrong\u003e,\u003c/strong\u003e 147-163.\u003c/li\u003e\n\u003cli\u003eZhang, B., S. Chen, Y. Ma, T. Li \u0026amp; K. Tang (2020) Analysis on spatiotemporal urban mobility based on online car-hailing data. \u003cem\u003eJournal of Transport Geography,\u003c/em\u003e 82\u003cstrong\u003e,\u003c/strong\u003e 102568.\u003c/li\u003e\n\u003cli\u003eZhang, F., Z. Li, N. Li \u0026amp; D. Fang (2019) Assessment of urban human mobility perturbation under extreme weather events: A case study in Nanjing, China. \u003cem\u003eSustainable Cities and Society,\u003c/em\u003e 50\u003cstrong\u003e,\u003c/strong\u003e 101671.\u003c/li\u003e\n\u003cli\u003eZhang, X. \u0026amp; N. Li (2020) Assessment of the Correlation between Extreme Weather Event-Induced Human Mobility Perturbation in Urban Areas and their Spatial Characteristics based on Taxi Trajectories.\u003c/li\u003e\n\u003cli\u003eZhu, D., N. Wang, L. Wu \u0026amp; Y. Liu (2017) Street as a big geo-data assembly and analysis unit in urban studies: A case study using Beijing taxi data. \u003cem\u003eApplied Geography,\u003c/em\u003e 86\u003cstrong\u003e,\u003c/strong\u003e 152-164.\u003c/li\u003e\n\u003cli\u003eZhu, P., J. Huang, J. Wang, Y. Liu, J. Li, M. Wang \u0026amp; W. Qiang (2022) Understanding taxi ridership with spatial spillover effects and temporal dynamics. \u003cem\u003eCities,\u003c/em\u003e 125\u003cstrong\u003e,\u003c/strong\u003e 103637.\u003c/li\u003e\n\u003cli\u003eZhu, P., J. Li, K. Wang \u0026amp; J. Huang (2024) Exploring spatial heterogeneity in the impact of built environment on taxi ridership using multiscale geographically weighted regression. \u003cem\u003eTransportation,\u003c/em\u003e 51\u003cstrong\u003e,\u003c/strong\u003e 1963-1997.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Travel behavior resilience, taxi ridership, built environment, non-linearity, machine learning","lastPublishedDoi":"10.21203/rs.3.rs-8938756/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8938756/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eIn the era of climate change, increasingly intense and frequent extreme weather events pose grave threats to urban life. Being a critical part of urban transportation system, taxis can facilitate understanding of human mobility and its variations during extreme weather events. Nevertheless, limited studies have delved into how taxi ridership responds to and recovers from the shocks induced by such events. Using Xiamen, China, as the case and employing random forest method, this study delves into resilience of taxi ridership and its non-linear relationships with built environment factors in the face of a rainstorm. Results show that (1) distance to city center, building density, and property price (as a proxy for socioeconomic features) are the most important contributors; (2) all the independent variables have salient non-linear effects on taxi ridership resilience, and obvious threshold effects exist; and (3) synergistic effects exist between certain independent variables, such as population density and land use mix. These findings can provide a solid knowledge base for formulating and executing nuanced intervention strategies for resilience promotion.\u003c/p\u003e","manuscriptTitle":"Examining resilience of taxi ridership during rainstorms and its non-linear associations with built environment","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-02 07:56:39","doi":"10.21203/rs.3.rs-8938756/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"333eae13-ed21-40ad-bbb7-8ab227a21bdc","owner":[],"postedDate":"March 2nd, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-04-24T08:09:49+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-02 07:56:39","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8938756","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8938756","identity":"rs-8938756","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}