Decoding urbanization effects on the water cycle: a multi-faceted analysis of balance, magnitude, and memory Open Access

Water cycle alteration is a critical focus of current hydrological research, driven by factors such as climate change, urbanization, land use change, and agricultural practices (e.g., Vörösmarty & Sahagian 2000; Gordon et al. 2008; Sterling et al. 2013; d’Odorico et al. 2019; Douville et al. 2021). Among these, urbanization stands out as a major contributor to the alteration in the water cycle (Brilly et al. 2006; Mcgrane 2016; Khadka et al. 2020). With urbanization intensifying globally and the urban population projected to increase by ∼1.5 times, reaching ∼6 billion by 2045 (World Bank 2023), its impacts on water cycle alteration are getting increasingly significant. It should be noted that urbanization-induced changes in the water cycle vary not only across spatial scales, but also among different components, such as precipitation, runoff, and evapotranspiration, each affected at different rates (Fang et al. 2020; Stephens et al. 2021; Hobeichi et al. 2022). These alterations inevitably lead to adverse impacts on water quantity, quality, and the associated ecosystems (Bao & Fang 2012; Gu 2019; United Nations 2019). Consequently, urbanization impacts on altering the water cycle have been extensively studied in hydrological fields (e.g., Konrad & Booth 2005; Misra 2011; Braud et al. 2013; Li et al. 2018; Ren et al. 2022). A representative global trend indicates that urban areas are likely to experience increased runoff along with a decrease in infiltration and evaporation (e.g., Arnold & Gibbons 1996; Cheng & Wang 2002; Ramier et al. 2006; O'driscoll et al. 2010; Astuti et al. 2019).

While urbanization-induced hydrological changes have been extensively studied, unknown gaps persist. These include not only uncovering memory dynamics across the full scale of the water cycle but also explaining water cycle alteration through an integrated perspective. In fact, temporal memory for precipitation and/or discharge has been primarily explored over a specific time period within the recent few decades using power spectral analysis (PSA) (Gall et al. 2013; Kim et al. 2016). Nonetheless, a comprehensive understanding remains limited due to the narrow scope of variables and timeframes typically analyzed. This study addresses this gap by applying spectral analysis to simultaneously identify temporal memories across core components of the water cycle, i.e., precipitation, runoff, and evaporation. Furthermore, this study seeks to capture how memory characteristics change under anthropogenic influences by covering both pre- and post-urbanization periods. This temporal framing enables us to discern whether and how urbanization modifies the persistence or variability of hydrological signals over time. Importantly, our study is among the first to concurrently assess temporal memory across the core water cycle components within an urban context.

Particularly for urban catchments, understanding hydrological responses to a given storm event requires spatially explicit information about surface and subsurface attributes of the urbanized landscape, including, but not limited to, impervious areas and drainage network configurations. Such information is instrumental in estimating surface runoff and stormwater discharges from drainage systems, which ultimately contribute to the total runoff at the catchment outlet. Among numerous rainfall–runoff models available for urban catchments (Fletcher et al. 2013; Szeląg et al. 2022), the stormwater management model (SWMM) (Rossman 2010) is widely used to simulate hydrological processes at the scale of small urban catchments, providing reliable results, especially in terms of runoff magnitude (e.g., Sillanpää & Koivusalo 2015; Bisht et al. 2016; Yao et al. 2016). However, the measures implemented in the SWMM to simulate evaporation rates have been questioned (e.g., Feng & Burian 2016; Peng & Stovin 2017), primarily due to the SWMM's assumption that the simulated evaporation rate is adjusted to match the surface moisture content when the available surface moisture is less than the evaporation rate (Kim & Kim 2022). This limitation has been addressed by incorporating additional modules into the original SWMM, such as the SWMM-PP (Randall et al. 2020) and the SWMM-UrbanEVA (Hörnschemeyer et al. 2021). Nonetheless, the enhanced performances of these models in simulating reliable evaporation rates are only demonstrated for short-term, intra-annual periods, leaving a significant gap in their applicability for long-term hydrological simulations.

To overcome this gap, a promising approach is to utilize the Budyko framework, a conceptual lumped model that characterizes the long-term water balance using aridity and evaporative indices (Budyko 1974), facilitating the estimation of dependable long-term evaporative rates. This framework is founded on theoretical principles of water and energy balance, providing an analytical method to relate precipitation, evaporation, and potential evapotranspiration (PET), ultimately enabling a comprehensive understanding of hydrological processes (Roderick & Farquhar 2011; Xu et al. 2013). The Budyko curve has been both theoretically and empirically proved to be a reliable, effective and stand-alone method for analyzing water and energy balance in specific catchments, primarily due to its simplicity and minimal parameter requirements (Reaver et al. 2022). The Budyko framework has demonstrated its value in offering insights into water cycle dynamics influenced by changes in climate and land use/land cover (Carmona et al. 2014; Gao et al. 2016; Wu et al. 2017; Ning et al. 2019; Vora & Singh 2021). For example, Zhao et al. (2014) examined Budyko curves for 64 watersheds in China to investigate trends in runoff changes induced by urbanization and precipitation variation. S. Kim et al. (2021) analyzed changes in the aridity and evaporative indices for the Han River basin in South Korea across past, present, and future periods, presenting the magnitude of influence under future climate and land use scenarios. In addition, shifts among Budyko curves have been used to capture the effects of changing climate conditions and human activities on long-term water balance in specific catchments (e.g., Dey & Mishra 2017; Yang et al. 2022). Movements within a given Budyko curve have also been instrumental in investigating the intra- and inter-annual variability of water cycle components over long-term periods (e.g., Yang et al. 2007; Chen et al. 2013; Mo et al. 2024). Notably, the Budyko framework has been employed to validate remote sensing data (Mianabadi et al. 2020) and to assess the performance of rainfall–runoff model simulations (Nijzink et al. 2018), corroborating its broader applicability beyond traditional water balance analyses.

Building on these conceptual and methodological foundations, we pursue a more integrated approach to advance the understanding of the long-term hydrological alterations driven by urbanization. While previous studies have applied either the Budyko framework or magnitude-based analysis to assess water cycle alterations (Rysman et al. 2013; Bai et al. 2020), relatively few have examined these methods in combination, and even fewer have incorporated temporal memory analysis. In this study, we bring together three complementary approaches: (1) the Budyko framework, (2) the magnitude analysis of core water cycle components, and (3) PSA. This integration enables us to evaluate urbanization impacts in terms of water–energy balance, magnitude shifts, and long-term memory characteristics. Ultimately, we aim to demonstrate the utility of this novel multi-faced approach in characterizing hydrological responses to urbanization in a real-world urban setting.

To this end, we selected the portion of Seoul National University (SNU)'s Gwanak campus, which is the largest single-campus university in South Korea, that lies within the Dorim Stream catchment in southwestern Seoul. As the capital of South Korea, Seoul has undergone remarkable urbanization over the last few decades, reaching a population density of ∼16,000 people/km² across ∼605 km² (Seoul City Hall 2023). Since 1988, it has been recognized as a globally renowned megacity with more than 10 million residents, accounting for ∼24% of the nation's total population (Child Hill & Kim 2000). Despite its socio-cultural and demographic importance (Kim et al. 2001), understanding the water cycle at the entire city scale has been challenging, mainly due to the mismatch between administrative boundaries and hydrological units (e.g., Herrfahrdt-Pähle 2010; OECD 2012). Consequently, hydrological research has largely focused on smaller urban catchments within the city. The Gwanak campus, spanning 4.1 km² with over 250 buildings and ∼54,000 residents as of 2021 (SNU Diversity Council 2022), has experienced rapid urban development mirroring the broader urbanization of Seoul. Short-term studies have linked these changes to increased flooding and drying events downstream (Yi et al. 2010; Park 2015), as well as altered hydrological flow connectivity (Lee 2010). However, a comprehensive understanding of the catchment's long-term water cycle dynamics across its core components remains elusive. This site provides a valuable context for applying our integrated approach to investigate how urbanization affects the long-term behavior of core water cycle components.

The Gwanak campus of SNU is located in Gwanak-gu, Seoul, South Korea (37° 27′ 33″ N/126° 57′ 11″ E), covering a total area of 4.1 km² (SNU FMO 2024). It is the largest single campus in South Korea by area size (Kim 2020). Within this campus, the Dorim Stream catchment, covering 2.9 km², has been selected as the study area. This catchment includes all the main buildings and facilities for education, research, and campus services, making it representative of the primary urbanized area of the campus. Meanwhile, the Bongcheon Stream catchment (1.2 km²) consists of student dormitories and a few supporting facilities (Seoul National University 2024). The study area is positioned halfway up Gwanak Mountain and features an average slope of 12.6°, which is steeper compared to the national average of 5.7° (NGII 2018). This area is influenced by the Asian monsoon climate, characterized by strong seasonal variation of precipitation, with particularly extreme magnitudes during summer (Shin et al. 2016; H. R. Kim et al. 2023). It is known that the Dorim Stream has encountered numerous flood and inundation incidents since the 1960s (Park 2015; Won et al. 2022).

Each component of the water cycle for the study catchment was characterized for both pre- and post-urbanization periods (with subscripts i and j hereafter), defined as for the years of 1960–1964 and 2008–2012, respectively. These specific periods were selected to represent distinct phases of urbanization in the study area, providing a clear contrast between conditions before and after urban development. Table 1 summarizes the studied hydrological variables, data collection methods, corresponding stations for each period of interest, and the analyzed features of these variables.

Precipitation (P_obs): For period i, we used daily precipitation data observed at the Jongno-gu station in Seoul, referred to as the Seoul station operated by the Korea Meteorological Administration (KMA). This station is the only one in Seoul with available observation records for period i, among a total of 31 meteorological observatories accessible through the Automatic Weather Stations and Automated Surface Observing Systems (http://data.kma.go.kr/). For period j, we analyzed daily precipitation data measured at the Gwanak station located in the study catchment, which is also operated by the KMA. The locations of the two stations are depicted as S_R (latitude: 37.57142, longitude: 126.9658) and S_S (latitude: 37.45284, longitude: 126.95015), respectively, in the nested Seoul map in Figure 1. It should be noted that the fluctuations of precipitation measured at S_R are consistent with that of S_S, despite the ∼13 km distance between the two stations. When comparing the annual precipitation data observed at S_R and S_S for the period from 1997 to 2022, we achieved a correlation coefficient value of 0.83 (Figure S1 in Supplementary Information).

PET: We calculated PET, required for depicting the Budyko curve, by applying the Hargreaves method expressed as a function of extraterrestrial radiation and daily minimum, mean, and maximum air temperatures (Hargreaves & Samani 1985, Table S2 in Supplementary Information). We used temperature data measured at the S_R and S_S stations for periods i and j, respectively. The extraterrestrial radiation values were obtained from latitude-specific data provided by Allen et al. (1998), using monthly average values corresponding to the 15th day of each month. Although other empirical models, such as the FAO-56 Penman–Monteith model (Allen et al. 1998), the Makkink model (Makkink 1957), and the Valiantzas 2 model (Valiantzas 2013), could have been applied, the input variables required for these models were only available for period j at the reference site S_R. In contrast, temperature data for the Hargreaves method was consistently available at S_R station for both periods and at S_S station for period j. It is also noteworthy that this method is ranked as the best among solely temperature-based models studied in Lee et al. (2023). The PET_H estimates from the Hargreaves method were validated for period j by comparing them with those calculated using the Makkink and Valiantzas 2 models (see Section 3.1).

Runoff (Q): Due to the lack of observational data for Q in our study catchment, we simulated runoff (Q_S) using the Storm Water Management Model, SWMM (see Section 2.4). The simulated Q_S values were further integrated with the Budyko framework to derive seasonal and annual runoff values (Q_BZ and Q_B, respectively). This integration ensured that the water cycle components were consistent across temporal scales.

Evaporation (E): E was estimated from the Budyko framework, which provided seasonal and annual values (E_BZ and E_B, respectively). However, since daily evaporation were unavailable from this approach, we utilized daily evaporation records measured by a small pan evaporimeter at the S_R station. The measured values were converted to actual evaporation rates (E_A) by applying a pan coefficient of 0.7 (Chow et al. 1988; Dingman 2015), providing the E_A availability for both pre- and post-urbanization periods.

The temporal resolutions of the analyzed data inevitably varied depending on the specific water cycle aspects. For evaluating the water and energy balance, daily data (P_obs, PET_H, Q_S) were upscaled to annual resolutions. To quantify the magnitude of the water cycle components, we used the upscaled dataset at both seasonal data (P_obsZ, Q_BZ and E_BZ) and their annual representation (i.e., the sum of seasonal values; denoted as P_obs, Q_B, and E_B). These datasets allowed for the characterization of the temporal variabilities in individual water cycle components. To identify the temporal memory of the water cycle components, daily data (P_obs, Q_S, E_A) were subjected to PSA (see Section 2.5).

It is noteworthy that the parameter ω can be analytically derived using representative values of PET, P, and E, all described at the same temporal resolution. Here, we derived the ω values for the pre- and post-urbanization periods employing a 5-year annual average of PET calculated by the Hargreaves method (PET_H), observed precipitation (P_obs), SWMM-estimated runoff (Q_S), and the resultant evaporation (E_S = P_obs – Q_S) based on the water mass balance. The E_S values were verified by comparing with the actual evaporation (E_A) observed at the S_R station for the pre-urbanization period (see Section 3.1).

The SWMM, a comprehensive urban rainfall–runoff model, is capable of simulating runoff based on the input data of precipitation over urban catchments (Rossman 2010). The model for the post-urbanization period j configured our study catchment with 62 sub-catchments, 98 conduits, 89 junctions, 2 weirs, 4 storages, and 10 outfalls (Figure S2B in Supplementary Information). For the pre-urbanization period (i), the same sub-catchment boundaries were applied, but the catchment had 0% imperviousness and only one storage pond (Figure S2A in Supplementary Information). Detailed setting options, including the infiltration method, hydrologic and hydraulic parameters for both periods, as well as the SWMM validation process, are described in Text S1 and Table S3 in Supplementary Information. To ensure the reliability of model setups, we compared our simulated runoff with the simulation results reported by Seo & Choi (2022).

The power-law exponent α was calculated based on the binned data to reduce inevitable effects of the data scattering. In this study, when α values were statistically significantly divided into positive and negative values (R² > 0.7, p-value < 0.05), we delineated the extent of f to derive a non-negative value of α, defining the frequency threshold (f*) as the minimum f within the extent. The identified f* enabled us to understand the scale of periodicity that holds the consistent degree of memory in a given time-series, by calculating the periodicity threshold (T* = 1/f*). Besides the magnitude of α, the threshold values of f* and T* were used to investigate the differences between the pre- and post-urbanization periods for the observed precipitation (P_obs), actual evaporation (E_A), and the SWMM-simulated runoff (Q_s).

The seasonal aridity index (∅_Z) increased after urbanization in both spring and winter, while there was no remarkable change in autumn and a decrease was observed in summer (Figure 2). Given a threshold of the aridity index ∅ = 1 as the division between wet and dry regimes, ∅ values in the post-urbanization period for spring and winter exceeded the threshold at 1.33 and 1.76, respectively. In contrast, those for summer and autumn were below the threshold, at 0.37 and 0.79, respectively. It is noteworthy that the regime shift from wet to dry occurred only in spring, whereas the already dry winter turned drier, and the wet summer became wetter as urbanization intensified in our study area. Moreover, it should be noted that the aridity index reflects the interplay between P and PET. Despite being the lowest PET in winter, it still exceeded the corresponding winter P, leading to a higher ∅. Conversely, the 44.2% increase in summer P caused a decrease in ∅, as the increase in PET did not keep pace with the rise in P (Figures S3A-B in Supplementary Information). Particularly in summer, the forward shift along the Budyko curve after urbanization aligned well with the findings of Potter et al. (2005), which present that Australian catchments with summer-dominated rainfall are positioned in the frontal part of the Budyko curve defined by the Fu method.

Regarding the evaporative index ((E/P)_Z), all four seasons showed a decrease of 0.17–0.29 from the pre-urbanization to the post-urbanization period. The finding underpinned that urbanization consistently contributed an increase in the ratio of surface runoff to precipitation for all seasons (Figure S3C in Supplementary Information). Thus, our analysis of the aridity and evaporative indexes derived from the ω values provides a conceptual understanding of the hydrological processes in the urbanized study area. However, as Reaver et al. (2022) highlighted, it is essential to go beyond the parametric Budyko framework and employ process-based models to further justify and advance the Budyko framework's application. By integrating both frameworks, future studies may better capture the complexities of hydrological responses, ensuring deeper understanding in various climatic and urban contexts.

Our finding on the ω magnitudes was well aligned with the range of 1.3–4.9 found by Bai et al. (2020) analyzing 206 watersheds in China, as well as the range of 1.4–2.9 presented by K. Kim et al. (2021) investigating 183 catchments constituting the Han River basin in South Korea. In particular, K. Kim et al. (2021) report lower ω values for catchments with a greater fraction of built-up area, which supports our result on the decrease of ω values with urbanization. They also propose an empirical model to estimate ω based on the fraction of each land cover category. Applying this model to our study area yielded ω values of 2.74 and 1.99 for the pre- and post-urbanization periods, respectively (ω’_i and ω’_j in Figure S4 in Supplementary Information). The ω value for the pre-urbanization period was similar to our calculated ω value, differing by only 0.11. However, the ω value for the post-urbanization period was 0.48 lower than their model's estimate. This difference is highly attributable to deviations in land cover ratios caused by the scale difference. The empirical model was developed on the basis of much larger catchments (39.2 to 449.8 km²) than our studied catchment of 2.9 km². To address this discrepancy, we recalculated the ω value using the land cover ratios from their catchment boundary (63.6 km²) that includes our study area (Figure S5 in Supplementary Information). This adjustment increased the built-up area ratio to 75.7% for the post-urbanization period, almost double our study area's 37.9%, resulting in a recalculated ω value of 1.23. The overall pattern of decreasing ω values along with urbanization is consistent across both methods, which supports the reliability of our results.

It should be noted that we validated the E_S values calculated from P_obs – Q_S (Figure S6 in Supplementary Information), as well as PET values derived using the Hargreaves method (PET_H, Figure S7 in Supplementary Information), both of which are intimately related to the reliability of the ω values presented. The estimated evaporation (E_{S, i}) for the pre-urbanization period i was 775.6 mm, showing only ∼1% difference from the actual evaporation (E_{A, i}) of 784.8 mm (Figure S8 in Supplementary Information). For the post-urbanization period j, we additionally compared the PET_H values to PET estimates from two independent approaches: the Makkink and the Valiantzas 2 models. These models were selected as they provide the most accurate PET results among purely radiation-based and combined temperature-and-radiation-based models, respectively, across South Korea for the period of 1972–2021 (Lee et al. 2023). Comparative analyses showed that statistically satisfactory correlations between the PET_H values and each of PET results from the two independent models for period j (Figures S9A-B and Text S2 in Supplementary Information), supporting the reliable interpretation of the ω values in the aforementioned Budyko domain.

Regarding the urbanization impact on decreasing total evaporation, it is globally found that the average annual evaporation from remote sensing data in urban areas is lower than that in non-urban areas, excluding arid regions (Mazrooei et al. 2021). Dow & Dewalle (2000) also demonstrate the decrease of evaporation calculated by the water mass balance across 51 urbanized watersheds in the eastern United States. Moreover, the reduction of vegetated areas has been shown to decrease the evaporation magnitude, as evidenced by satellite data, observation records, and hydrologic model simulations (e.g., Carlson & Arthur 2000; Feldman et al. 2023). Indeed, these findings are consistent with our result that urbanization procedures over our study area in a monsoon climate zone resulted in similar trends of land cover changes, i.e., built-up land increased by ∼38% while forest land decreased by a similar amount (see Section 2.1).

For the individual four seasons, the changes in P_obs, Q_B, and E_B between the periods i and j showed different characteristics (Figure 3). After urbanization, the magnitude of P_obs increased only in summer by 44.2% and decreased most in spring by 32.5%, followed by winter (10.9%) and autumn (7.2%) (Figure S3B in Supplementary Information). This led to an increase in total precipitation throughout the studied periods. The Q_BZ values increased in all seasons: 134.6% in winter, 105.1% in summer, 37.1% in autumn, and 1.2% in spring (Figure S3C in Supplementary Information). In contrast, the E_BZ magnitude declined in all seasons by 28.1–50.5%, resulting in an overall decrease between the two periods (Figure S3D in Supplementary Information). These seasonal-scale dynamics of water cycle components have also been highlighted in former studies (Branger et al. 2013; Bian et al. 2021), underlining the importance of examining seasonal variability. Given the uneven changes in the magnitude of water cycle components across seasons after urbanization, it is crucial to consider the seasonal dynamics of these components when evaluating urban hydrological impacts and formulating effective water resources management.

In particular, the most significant magnitude changes were observed in summer among four seasons in the case of P_obs and Q_B for the post-urbanization period, with a 335.6 mm increase in P_obs (44.2% increased) and 433.3 mm increase in Q_B (105.1% increased). This implies the need for comprehensive flood control measures and policies especially for summer in this study area, as extreme rainfall events are expected to occur more frequently in the future. For example, a major flood event in 2011, during the post-urbanization period, served as a crucial trigger for subsequent flood control policy initiatives in South Korea. Indeed, the capital city of Seoul has conducted flood control policy studies including the installation of underground storage tanks (Seoul City Safety Office 2013). Specifically for the studied campus catchment, flood control measures have been continuously implemented for the last decades (Seo et al. 2012; Seo & Choi 2022; Y. O. Kim et al. 2023).

The Q_B and E_B values estimated by integrating the SWMM and Budyko framework were indeed reliable in quantifying the total magnitude and seasonal variabilities of water cycle components in our campus catchment for both pre- and post-urbanization periods. For SWMM-simulated runoff, we validated our post-urbanization period runoff results using the simulation results from Seo & Choi (2022), as there were no runoff data exactly observed at the outlets of the study area. Seo & Choi (2022) present runoff for the three sub-catchments (Sub₁, Sub₂, Sub₃) within our study area (Figure S2B in Supplementary Information). We obtained R² of 0.74–0.75 and RMSE of 1.03–1.66 mm, validating our model setup and simulation results (Figure S10 in Supplementary Information). Note that the referred simulation was conducted by the HEC-HMS model for 8–9 August 2022 with the same CN infiltration method as our study applied to the SWMM setup. They employed the corresponding curve number values for post-urbanization and conducted the simulations accordingly. Our validation was limited to a comparison with another model due to the lack of measured runoff data within the campus. Although future calibration with observational equipment is necessary to increase accuracy, the reasonably high R² values and low RMSE values presented earlier indicate that the SWMM model parameters were appropriately applied. A few studies (e.g., Güntner 2008; Clark et al. 2017; Janicka et al. 2023) show that comparing simulation results from various models is an effective analysis method for ungauged catchments like ours.

We found that the estimated evaporation (E_B) in this study closely matched the reference evaporation values (E_ref) (Figure S11 in Supplementary Information), which were calculated using the empirical model proposed by K. Kim et al. (2021) to derive for the Fu parameter ω based on land cover area ratios. Similar to E_B, the total magnitude of E_ref consistently decreased after urbanization. The seasonal distribution of evaporation was almost identical between E_BZ and E_ref during both periods. The cross-checking of these evaporation estimates suggests that this study made significant advancement in understanding evaporation dynamics within the studied campus catchment, a topic that was largely overlooked in previous research. For example, Park (2015) analyzed, without considering the evaporation component, the impacts of constructing the SNU campus on the water cycle. Simiarly, Lee (2010) investigates the influences of urbanization on the Dorim Stream catchment, which includes our study area, focusing on the stream's hydrological connectivity but without addressing evaporation dynamics.

Between the pre- and post-urbanization periods, only precipitation P_obs showed a significant increase in α (from 0.3 to 0.5, p-value< 0.05). This suggests a shift toward stronger long-term memory such as seasonality (Figure 4(a)). The result was aligned with the finding of Kim et al. (2016) on the α range of 0.37 ± 0.06 for observed rainfall data from 1980 to 2009 over 78 unit-watersheds in South Korea. Even for four U.S. watersheds, the PSA of rainfall data also reports some memory as α = 0.1–0.32 (Gall et al. 2013). Indeed, our findings were in good agreement with Rysman et al. (2013) and Martinez-Villalobos & Neelin (2023) which identify steeper power-law slopes as extreme rainfall magnitudes get more concentrated in a certain season. For Q_S, while the α value also increased slightly from 0.4 to 0.5 after urbanization (Figure 4(b)), the change was not statistically significant (p-value > 0.05). Nevertheless, it was noteworthy that the memory of Q_S was identical to that of P_obs in the post-urbanization period, suggesting almost no landscape filtering. That is, expanding impervious areas at more urbanized status induced quicker routing as overland flows and subsurface flows with minimal infiltration, resulting in the synchronization of surface runoff and precipitation (Gall et al. 2013).

Regarding the memory of E_A, the α values remained stable around 0.8 to 0.9 throughout both pre- and post-urbanization periods (Figure 4(c)), indicating relatively long-term memory. The α values for E_A showed no statistically significant shift from positive to negative, resulting in no identifiable f* value in the post-urbanization period. Given that the minimum f in the post-urbanization period is lower than the pre-urbanization f* value, this indicates that the periodic patterns in the post-urbanization period have become more extended than the pre-urbanization period. These findings were aligned with the findings of Blanco-Macías et al. (2011), which demonstrate long-term memory when examining monthly pan evaporation over longer periods encompassing the pre- and post-urbanization periods of our study. We acknowledge that the hydrological response of our study catchment was governed by complex interplay among many factors such as the spatial distribution of land covers, the precipitation patterns, and soil properties besides the fraction of urbanization. A follow-up study needs to unravel the relative contribution of governing factors over diverse small urbanized catchments.

This study presented a novel, multi-faceted approach – combining the Budyko framework, magnitude analysis, and PSA – to investigate the long-term impacts of urbanization on water cycle dynamics in a representative urban catchment, using the nation's largest campus catchment in Seoul, South Korea. By capturing pre-urbanization (1960s) and post-urbanization (2010s) conditions, this study uniquely advanced hydrological research by concurrently evaluating changes in water–energy balance, magnitude, and temporal memory across key water cycle components (i.e., precipitation, runoff, and evaporation) within a single urban system. In doing so, it directly addressed two critical knowledge gaps identified in the literature: the lack of integrated assessments and the limited exploration of temporal memory under urbanization. The Budyko-based analysis showed a marked shift of the Budyko curve toward a more runoff-dominated regime following ∼38% increase in built-up area, indicating reduced evaporative capacity and enhanced runoff generation. Seasonal analysis exhibited intensifying dryness in spring and winter, in contrast to increasing humidity in summer. Magnitude comparisons identified ∼15% increase in observed precipitation P_obs, ∼75% increase in estimated runoff Q_B, and ∼38% decrease in calculated evaporation E_B, underscoring the hydrological imbalance induced by urban growth. These findings reinforce the urgent need for comprehensive stormwater management, particularly as extreme summer rainfall events are highly expected to intensify. Notably, spectral analysis uncovered significant memory across all components for both pre- and post-urbanization periods. In both cases, P_obs and SWMM-simulated runoff Q_S exhibited relatively weak long-term memory, whereas actual evaporation E_A showed comparatively strong long-term memory. A remarkable urbanization-induced change in memory was found only for P_obs. It was particularly noteworthy that the memory of P_obs and Q_S synchronized to the same periodicity after urbanization, suggesting a tighter coupling of atmospheric input and catchment response in urbanized settings. This result was well aligned with the greater dominance of summer as a key season for both P_obs and Q_S in the post-urbanization period. By integrating multiple analytical perspectives, this study offers a more comprehensive and nuanced understanding of how urbanization reshapes the temporal dynamics of the water cycle, contributing a significant advancement to urban hydrological research.

Leveraging these integrated understandings of water cycle alterations, this study further underscores the necessity of expanding urban water research beyond quantity-focused assessments to encompass broader spatio-temporal implications including urban-stream water quality and aquatic ecosystem integrity. Given the intimate connections between urban areas and their neighboring streams, urbanization effects need to be examined not only from a hydrological perspective but also in terms of their ecological and biogeochemical consequences. The expansion of impervious surfaces in urban catchments promotes surface runoff, which in turn accelerates the degradation of stream water quality, such as by reducing pH levels and increasing nutrient concentrations. Furthermore, land cover change toward urbanization is likely to disturb or destroy suitable habitat conditions for aquatic and riverine ecosystems, not only by altering flow regimes and in-stream chemical composition, but also by increasing water temperature, nutrient loading, and sediment accumulation. This study provides foundational insights into these urbanization-induced challenges and highlights the need for future research to comprehensively evaluate impacts on water quantity, quality, and ecological integrity, ultimately supporting the development of integrated urban-stream management strategies.

This work was supported by the Creative-Pioneering Researchers Program through Seoul National University and by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (No. RS-2025-00523350).

This work was supported by the Creative-Pioneering Researchers Program through Seoul National University and by the National Research Foundation of Korea (No. RS-2025-00523350).

H.K. contributed to data curation, methodology, formal analysis, investigation, writing – original draft, writing – review & editing, visualization. Y.-O.K. contributed to conceptualization, writing – review & editing, project administration. S.Y. contributed to conceptualization, methodology, writing – review & editing, supervision, funding acquisition.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.