Rainstorms impacts on water, sediment, and trace elements loads in an urbanized catchment within Moscow city: case study of summer 2020 and 2021

In 2020 and 2021, the city of Moscow, Russia, has experienced two historical rainfall events that had caused major flooding of small rivers. Based on long-term observation datasets from the surrounding weather stations, regional mesoscale COSMO-CLM climate model results, and a detailed hydrological and water quality monitoring data, we performed a pioneer assessment of climate change and urbanization impact on flooding hazard and water quality of the urban Setun River as a case study. Statistically significant rise of some moderate ETCCDI climate change indices (R20mm and R95pTOT) was revealed for the 1966–2020 period, while no significant trends were observed for more extreme indices. The combined impact of climate change and increased urbanization is highly non-linear and results in as much as a fourfold increase in frequency of extreme floods and shift of water regime features which lead to formation of specific seasonal flow patterns. The rainstorm flood wave response time, involving infiltrated and hillslope-routed fraction of rainfall, is accounted as 6 to 11 h, which is more than twice as rapid as compared to the non-urbanized nearby catchments. Based on temporal trends before and after rainfall flood peak, four groups of dissolved chemicals were identified: soluble elements whose concentrations decrease with an increase in water discharge; mostly insoluble and well-sorted elements whose concentrations increase with discharge (Mn, Cs, Cd, Al); elements negatively related to water discharge during flood events (Li, B, Cr, As, Br and Sr); and a wide range of dissolved elements (Cu, Zn, Mo, Sn, Pb, Ba, La, Cs, U) which concentrations remain stable during rainfall floods. Our study identifies that lack of research focused on the combined impacts of climate change and urbanization on flooding and water quality in the Moscow urban area is a key problem in water management advances.


Introduction
The ongoing climate change is leading to complex and diverse changes in different precipitation characteristics. General estimates shows 1-2% per century growth of precipitation total amount from the middle of the twentieth century over the continents (Contractor et al. 2021); with respect to the mid-latitudes, there is a significant positive trend in mean precipitation in the Northern Hemisphere (Trenberth 2011). Moreover, the daily extreme precipitation intensity tends to increase almost everywhere over land (Donat et al. 2016), including the occurrence of heavy rain events over the Northern Eurasia (Semenov and Bengtsson 2002;Mokhov et al. 2005;Groisman et al. 2005;Zolina and Bulygina 2016;Ye et al. 2017). The contribution of heavy convective showers to the total precipitation increases with the statistically significant trend of 1-2% per decade in vast Northern Eurasia regions, reaching 5% per decade at a number of stations, is reported (Chernokulsky et al. 2019;Aleshina et al. 2021). The overall changes in the character of precipitation over the majority of Northern Eurasia regions are characterized by a redistribution of precipitation types toward more heavy showers.
Even more pronouncing effects on precipitation fields are seen under urban surface conditions (Varentsov et al. 2018). Due to specific urban climate which is characterized by deviations of the temperature and humidity from surrounding rural environment, known as urban heat island (UHI) (Lappalainen et al. 2018;Masson et al. 2020), intense precipitation amount growth, increase of turbulence mixing, and more favorable conditions for convective boundary layer uplift are typically observed. Megapolises experiences rainfall structure changes in context of short-live precipitation events intensity growth. Significant increase of extreme hourly rainfall sums is observed in Shanghai during the XX century, its frequency rose up during intense urbanization from 1981 (Liang and Ding 2017). European cities studies indicate also rainfall intensity growth (Faccini et al. 2015;Liang and Ding 2017). These effects have significant environmental and social impacts Vlasov et al. 2020Vlasov et al. , 2021cKosheleva et al. 2022;Nikiforova et al. 2022;Popovicheva et al. 2022). In particular, urban regions are suspected to changes of the timing and magnitude of rainfall events as a result of climate change. Both with proliferation of impervious surfaces, these areas are predicted to significantly alter the flooding experienced in many urban areas of the world (Faccini et al. 2015;Liang and Ding 2017) and become extremely vulnerable to the rainstorm events. Climate change and urban water environment impacts at the global scale (Praskievicz and Chang 2009) with a particular examples from the cities of China (Liang and Ding 2017), UK (Ashley et al. 2005), cities of Europe (Faccini et al. 2015;Liang and Ding 2017).
As far as extreme rainfalls are associated with floods and increase in the amount of contaminants directly introduced into the rivers (Vlasov et al. 2021b), many cities are developing mitigation and adaptation strategies to reduce their vulnerability (Maragno et al. 2018). In this regard, Moscow city remains an exceptional example as the biggest monocentric agglomeration in Europe which has not yet developed green infrastructure (GI) areas and specifically low impact development measures. Due to this, recently large storm overflows, the overfilling of structures in stormwater networks and an increase in the number of the sewer floods have been observed here. Recent flooding across the Moscow city during the summer storms of 2020 and 2021 have highlighted the significant impacts that flooding can have on inundation of the urban areas, erosion with parks and huge sediment transport over urban rivers. In particular, the most significant events were observed during rainstorms on 29-30 May 2020 corresponding to 140% of monthly mean.
Due to lack of hydrological and hydrochemical monitoring by state and municipal agencies in Moscow, studies on flood prediction and adaptation effects are absent. To the best of our knowledge, there is no data regarding impact of extreme precipitation on timing and magnitude of floods, as well as their environmental impacts. Since 2019, Department of Hydrology of Lomonosov Moscow State University has been operating a network of gauging stations in the Setun River watershed (190 km 2 ), which is the largest tributary of the Moskva River located entirely within the bounds of Moscow city. At these stations, autonomous stage-discharge stations are set up, supported by systematic investigations of major and trace element chemistry during flood events Sokolov et al. 2021). Combined with the Meteorological Observatory of Moscow State University, which is located at downstream part of the Setun catchment and provides the atmospheric characteristics and specific weather conditions records including extreme precipitation since 1954 (Chubarova et al. 2011), this area presents an exceptional case to study the impact of rainstorm events on water, sediment and trace elements loads within Moscow city.
This paper summarizes a large dataset of water fluxes, dissolved and particulate element concentrations in the Setun River collected from 2019 to 2021 with a particular focus on comprehensive analyses of flow response to rainstorm-generated events that were observed in Moscow on 29-30 May 2020 and 28-29 June 2021. A high-quality data encompassing detailed flow and turbidity measurements at 3 stations along Setun River and long-term precipitation records from the stations operated in Moscow region, as well as meteorological COSMO-CLM model outputs, were considered in the study to understand hydroclimatic impacts on water and sediment quality during high flow events. Specifically, we aim at statistical analyses of long-term extreme precipitation records (1); reconstructing precipitation patterns during 2020 and 2021 rainstorms over case study Setun River by COSMO-CLM model (2). Additionally, based on monitoring data we aim to assess the magnitude and the transit time of water and sediment flows during short-lived events (3) and to characterize the dynamics of particulate and dissolved element transport in response to precipitation (4). The prepared datasets allow us to show that transported elements can be classified into different groups reflecting their behavior during floods in a small urban river. We also discuss the accuracy and perspectives of the proposed hydroclimatic approach to predict extreme rainfalls and their hydrological impacts, define specific features of urban river floods compared to natural rivers, and outline possible implications for water management system development in Moscow.

Moscow region precipitation data
The Moscow Region has a broad network of weather stations that provide standard precipitation measurements. However, there are only few long-term rainfall observations valuable for robust statistical estimates. We have used the original 6-hourly and daily precipitation sums from the Russian Institute of Hydrometeorological Information -World Data Center (RIHMI-WDC, http:// meteo. ru/ data) database according to the 10 stations within Moscow city and the nearest surrounding Moscow region for a 1966-2021 period, with the longest timeseries belong to Moscow State University Observatory (MSU), Balchug, Nemchinovka, Podmoskovnaya, Tushino, and VDNKh (Fig. 1). Data gaps were filled with http:// pogod aikli mat. ru/ site datasets, where it was possible and available. These stations were used for the long-term climatological estimates. Additional observational datasets were collected for specific periods of interest within the 2020 and 2021 cases (29-30 May 2020 and 28-29 June 2021) from the http:// pogod aikli mat. ru/ Web site, since there are some novel station data available in Moscow region last decades and years. These data included also airport stations Domodedovo, Sheremetyevo, and Vnukovo, as well Strogino, Krasnogorsk, Dolgoprudny, and Butovo, and were used additionally for comparison with modelling results to provide reliable verification. The MSU and Nemchinovka stations are located within the Setun River basin, therefore specific attention was brought to data from these stations. Moreover, the pluviograph datasets from the MSU Observatory were digitized with 10 min time step for periods 29-30 May 2020 and 28-29 June 2021, and used for high-resolution comparison to model precipitation output data.
Long-term precipitation datasets were used to estimate extreme sums frequency, their changes for a considered period, trends estimation based on ETCCDI indices (Karl et al. 1999): Rx1day-monthly maximum 1-day precipitation; R10mm, R20mm, and R30mm-annual count of days when Fig. 1 Long-term precipitation data availability for the Moscow region. Data availability periods presented as colors: period without gaps is presented in green, gaps more than one month are presented in yellow, periods with no data are presented in red precipitation ≥ 10 mm, 20 mm and 30 mm, accordingly; PRCP-TOT-annual total precipitation in wet days (i.e., days with precipitation ≥ 1 mm); R95pTOT and R99pTOT-annual total precipitation when daily precipitation amount > 95 and > 99 percentiles of precipitation on wet days, accordingly. Daily extreme precipitation sums are crucial for short-term flooding cases, especially in urban conditions (Yang et al. 2013;Zhou et al. 2017). Trends were estimated according to the t-and F-tests on the 5% significance level.

COSMO-CLM (CCLM) model setup
The COSMO-CLM (CCLM) model (ver. 5.12) was used for atmospheric simulations. CCLM is the climate version of the non-hydrostatic mesoscale atmospheric COSMO model, including various modifications and extensions adapted to the long-term numerical experiments. It was developed by the German Weather Service (DWD) and CLM-Community (Rockel et al. 2008). The model equations are solved on the rotational Arakawa C-grid (Arakawa and Lamb 1977) in latitude-longitude (φ, λ) coordinates with a pole tilt to minimize the issue of longitude convergence at the pole. The height coordinate is the terrain-following hybrid Gal-Chen coordinate (σ-z system) (Gal-Chen and Somerville 1975; Schär et al. 2002).
The standard configuration of the CCLM model includes the Runge-Kutta integration scheme with the fifth advection order. There is an option to apply the spectral nudging technique (Schubert-Frisius et al. 2017). The Ritter and Geleyn radiation scheme (Ritter and Geleyn 1992) is based on the two-stream version of the radiation transfer equation. The moist and shallow convection is parametrized using Tiedtke mass-flux schemes with equilibrium closure based on moisture convergence (Tiedtke 1989). Turbulence is described by a prognostic turbulence kinetic energy (TKE) based scheme, with a 2.5-order closure (Herzog et al. 2002). A full description of the COSMO model physics, dynamics, and parameterizations is available elsewhere (http:// www. cosmo-model. org/ conte nt/ model/ docum entat ion/ core/ defau lt. htm).
For the land grid cells, the TERRA (Schulz and Vogel 2020) surface active layer model is used. An important part of the COSMO-CLM model v.5.12 and the reason of its use in this work is the TERRA-URB urban canopy parameterization (Wouters et al. 2015(Wouters et al. , 2016Varentsov et al. 2020) allowing to take into account canopy layer energy exchanges between atmosphere and surface and subsurface in detail, to reproduce meso-and microscale dynamics caused by urban canopy impact correctly. These processes consideration reflects the extreme precipitation formation in urban conditions (Varentsov et al. 2018). Since the TERRA-URB urban canopy model was developed taking into account impervious water storage based on a density distribution of water puddles (Wouters et al. 2015), it is well suitable for urban hydrology tasks. Therefore, it is crucial to use the last version of the TERRA-URB parameterization to simulate extreme precipitation cases in Moscow city successfully. The TERRA-URB requires an additional external parameters information about urban surface properties within each modelling domain. This information includes impervious area fraction, anthropogenic heat flux, building fraction and height, canyon height-to-width ratio, facet-level albedo, emissivity, heat capacity, and conductivity data. More detailed information about these data source availability, usability and verification provided in (Varentsov et al. 2020).
Model experiments used the standard nested domains scheme including base and inner domains. ERA5 (Hersbach et al. 2020) global reanalysis data (0.25°, ~ 30 km) used as driving conditions for the base domain with grid size 0.027° (~ 3 km) and 500 grid cells in both directions, covering the most area of the East European Plain. The base domain output data used as driving conditions for inner domain simulations with 0.009° (~ 1 km) grid size focused on the Moscow region only (Fig. 2). Base domain experiments were simulated with 40 s time step, whereas inner domain experiments simulated with 10 s time step. Model used 50 vertical levels; no spectral nudging was applied. COSMO-CLM ~ 1 km grid cells getting inside the Setun River basin contour, 190 grid cells in total. COSMO-CLM experiments were run using the Lomonosov Moscow State University High Performance Computing Center "Lomonosov-2" (Voevodin et al. 2019) resources. Two model experiments were conducted for periods 28.05-02.06.2020 and 27.06-03.07.2021 corresponding to considered 2020 and 2021 summer floods: 29-30.05.2020 and 28-29.06.2021, accordingly. Aims of these experiments were to estimate model capability to reproduce heavy rainfalls in Moscow region during cited periods, to get relevant input data for high-resolution runoff simulations by hydrological model, and to reveal opportunities to tune model parameters for further experiments and other case studies. The model output step was 15 min containing 15-min precipitation sums at each grid cell during the whole experiments period.
The model outputs were summarized over 1 h, 3 h, and 1 day. The 15-min and 1-h model precipitation amounts were compared with 10-min and 1-h summarized pluviogram values for MSU Observatory station; 3-hourly and daily precipitation sums by model compared with standard precipitation measurements according to data and stations listed in Sect. 1. In all cases, we have calculated errors as differences between model and observed values and common verification statistics: mean error, RMSE and correlation between samples.

Hydrological monitoring
Since November 2019, three monitoring stations for discharge, sediment transport, and water and sediment quality were installed in the downstream reach of the Setun River (Fig. 3). Water stage was recorded with the Onset Hobo pressure sensors corrected for barometric pressure with 30-min interval and discharge was measured using the float method at all stages on the monthly basis. Sensors were installed in a stilling well fixed to the stream bank at a stable location. At very low stages, the discharge was measured with conventional impeller current meters and at high stages we used the Sontek RiverSurveyor M9 acoustic doppler current profiler (ADCP). Based on discharge measurements data for each of the stations stage-discharge relationships Q = f(H) were assessed (Table 1). These were further used to recalculate streamflow discharges with 30-min frequency. Integrated

Water quality monitoring datasets
Water samples were collected about 20 cm below the surface, manually from the bank since November 2019 at the downstream station S-3 (Fig. 3). Relatively narrow river channel induces high water velocities (up to 1 m/s) and enhanced turbulence that ensures efficient homogenization of the suspended and dissolved matter across the section. Sampling was performed approximately once a month during the base flow periods. During the spring flood and summer flash flood monitoring frequency increased from monthly to daily. In particular, during 2021 spring freshet and summer rain floods, additional sampling was done. One week before the start of the spring flood (March 19) and 5 days before the start of the flash flood (June 23), samples were taken on the daily basis. In total, more than 50 water samples were collected from the Setun River in 2021; 25 of which were taken during the spring flood and flash rain flood. The freshet datasets were used to compare the effects in water quality driven by rainstorms and snow melting processes on the watershed.
Water temperature, pH, dissolved oxygen (DO) concentrations, and specific conductivity (at 25 °C) were measured in situ by YSI ProSolo, YSI ProODO and YSI Pro30 probes (Yellow Springs, Ohio, USA) at every sampling station. River water samples were taken in PET 5L bottles. Samples of river water were filtered through pre-weighted membrane 0.45 μm filters using a Millipore Vacuum pump station. The filters with suspended particulate matter (SPM) were weighed to determine SPM concentration. Two aliquots of 30 ml of filtered water were collected for cations and trace elements analysis. Processing of samples were carried out at the Krasnovidovo scientific laboratory of Lomonosov Moscow State University.
The amount of easily oxidized organic substances was assessed in unfiltered samples via the five-day biochemical oxygen demand (BOD 5 ) using the iodometric method.
Total organic matter concentration estimated via indirect parameter of chemical oxygen demand (COD) using the dichromate method. The content of total (TN) and total dissolved nitrogen (TDN), total (TP) and total dissolved phosphorus (TDP), total inorganic (TIP) and dissolved inorganic (DIP) phosphorus, and nitrite nitrogen (N-NO 2 ) was determined in filtered and unfiltered samples on a single-beam photometer using standard methods: the Murphy-Riley method for TP, TDP, TIP, and DIP; alkaline persulfate digestion for TN and TDN; and Griess reaction for N-NO 2 . Concentrations of nitrate and ammonia nitrogen (N-NO 3 and N-NH 4 ), as well as major ions (Ca 2+ , Mg 2+ , Na + , K + , SO 4 2− , Cl − ), were determined via ion chromatography on Concise ICSep An2 (USA) and Shodex IC YS-50 columns (Japan). HCO 3 − and CO 3 2− concentrations were determined by titration with hydrochloric acid. The filter with accumulated suspended matter was weighed after filtration to calculate the suspended sediment concentration (SSC) in river waters (mg/L).
Dissolved (filtered water, < 0.45 μm) concentrations of Li, B, Al, Ca, V, Mn, Ni, Cu, Zn, As, Sr, Zr, Mo, Sb, Ba, Pb, Bi, U were determined using the ICP-MS and ICP-AES methods (Elan-6100   Dissolved and particulate major and trace element concentrations and water discharges in the water samples of Setun River are listed in Online Resource 2 which includes dynamics of chemical parameters of the Setun River during the summer storm flood of 2021, as well as the river's chemical profile below and after the flood.

Precipitation patterns
The long-term statistical analysis of daily precipitation over Moscow region indicated extreme values up to 70 to 80 mm during the twentieth century; however, values above 80 and 100 mm appeared 5 times since 2000-2002 (Fig. 4). Positive linear trends are insignificant over all stations on the 5% level with the exception of the Nemchinovka station located at the upstream of case study Setun River catchment. Here, the only significant positive trend has been observed. Therefore, most of the changes in daily precipitation maxima are insignificant with significant exception for 1 station within the Setun River basin.
Case study Setun River catchment stations (MSU and Nemchinovka) show similar extreme daily precipitation indices and patterns. R95TOT at MSU Observatory (Fig. 5) is comparable with daily maxima below 300 mm until 2010. At the same time, in 2013 and after there are values more than 400 mm, and R95TOT values are near 300 mm during 3 years. R99TOT has more striking variability and outliers (in 1973, 2004, 2020), which are larger at Nemchinovka (200-250 mm) than at MSU Observatory (near 200 mm).
Both indices at both stations shows positive linear trends, which are significant for R95TOT at Nemchinovka at 5% significance level and for R99TOT at MSU Observatory at 15% level only.
Significant changes were not detected at the MSU Observatory by the R10mm, R20mm, R30mm indices (Fig. 6). R10mm reaches 20-25 days maxima at MSU Observatory, up to 30 days after 1996. Maximums are distributed more uniform at MSU Observatory, with more outliers at Nemchinovka, where the number of R20mm values more than 5 days increases after mid-1990s. R30mm values demonstrates more variability at MSU Observatory than at Nemchinovka (up to 3-5 days); however, there are no significant trends in this index at both stations. All indices at both stations shows positive trend, except of MSU Observatory R10mm, where an insignificant negative trend is observed. There are only significant positive trends in R10mm and R20mm at Nemchinovka at 5% significance level.
These long-term changing conditions underlie the formation of extreme floods observed on 29-30.05.2020 and 28-29.06.2021 in Moscow. The estimated corresponding daily sums percentiles and the corresponding 100-year days frequencies (Table 2) show extreme character of the observed patterns. Considered stations demonstrated the daily sums over 99% percentile during 29-31.05.2020 and 27-28.06.2021. Moreover, the corresponding number of days with such amounts reached just no more than 100 days, i.e., the recurrence of such events is less than once per year.
In May 2020, the low moved from the Black Sea area leading to the warm front passage in Moscow region. Largescale precipitation enhanced by some meso-scale squall lines formed in warm and moist air mass. Moreover, warm front during this period was transformed into the occlusion front and intensified at evening and night times (Online Resource 3). COSMO-CLM model showed two main rainfall periods on 29 and 31 May (see Fig. 8), the latter case is prevailing by intensity and modelled precipitation sums. Precipitation maxima differed significantly by extreme values (from 2 to 10 mm per 15 min) and time. The first maximum was observed on 29 May 2020 at the morning and early afternoon times (from 00 to 15 h UTC, local time is UTC + 3 h), with short extreme values in early morning near Podmoskovnaya, Vnukovo and Krasnogorsk, i.e., on the west-northwest edge of Moscow region, without significant and continuous maxima over the Setun River basin. The second simulated maximums on 30-31 May 2020 were much more intense and long-lasting (up to 2 or 3 maximums during the whole day).
However, the main rainfall meso-gamma squall-line rainfall system crossed the Setun River basin in the late evening-night (Online Resource 4, a-c) on 30 May from 22.00 to 23.30 UTC from the south to the north. The second rainfall maximum was observed on 31 May from 2.00 to 3.30 UTC which reached smaller values (2-3 mm per 15 min) and concerned less subcatchments over the east and south-west of Fig. 6 R10mm, R20mm, and R30mm values (days) and its trends according to the MSU Observatory weather station the basin. The following rainfalls during the 31 May contributed to 3-4 mm per 15 min. During the end of June 2021 cyclone was formed over the West regions of Russian plain within the low-gradient pressure field, intensified after anticyclonic blocking termination and moved to the Central part of Russian plain leading to enhanced mesoscale convection and squall lines (Online Resource 3) (Alekseeva et al. 2022). COSMO-CLM simulation results indicated three rainfall events on 27, 28 and 30 June with the main maxima on 28 June. The latter event was short-lived (from 13.00 to 16.30 UTC) as compared with the 2020 case. On the contrary, the 2021 rainfall was more intense including grids within area of interest -the Setun basin and developed during the late afternoon time. Maximal intensities reached 7.5 mm per 15 min at Nemchinovka and 11 mm at Vnukovo, there are the largest values among other considered grids. Spatial structure of this rainfall event (Online Resource 4, d-f) model simulated two main precipitation intensity cores crossing the top parts of the Setun River subcatchments during a little more than a half of an hour with maximal intensities exceeding 15 and 20 mm per 15 min. Perhaps, these values and movement of rainfall core contributed significantly to the following flooding.

Streamflow patterns
The event record on 27-29 June 2021 seen from the two weather stations (Table 2) data alongside the streamflow time-series shows an insight on how the rainstorm propagated over the catchment. The rain gauges are located in the central (27,515 -Nemchinovka), controlling the rainfall over the Setun headwaters, and eastern (27,617 -MSU) parts of the catchment, which is mainly covered by the Ramenka River-Setun's right tributary (Fig. 3). The 27,515 rain gauge reported the accumulated 12-h rainfall, the 27,617 rain gauge reported rainfall accumulation at 10-min intervals.
Starting from 17:00 27.06, the 27,617 rain gauge accumulated some 13 mm of rainfall, which almost immediately resulted in a sharp flood peak on the S3 streamflow gauge, which lasted over 5 h. The localization of rainfall was mainly over the eastern part of the catchment, as no significant rainfall or water level rise can be seen at 27,515 rain gauge (Fig. 7) and S1 gauge.
The next pluvial event starting from 13:00 28.06 covered the entire Setun catchment, as both rain gauges reported significant amount of rainfall in several hours. This resulted in a sharp flood emergence with streamflow discharges rising more than tenfold in 10 h, from 2 to 20 m 3 /s on S1 gauge and from around 3 to 40 m 3 /s on S3 gauge. This first floodwave was relatively short-lived-nearly 5 h on S3 and 11 h on S3, but it was then followed by a smaller and longer secondary wave. It lasted from 15:00 to 23:00 28.06, and can be clearly traced on both gauges, but more pronounced on S1.
The concentration of Mg 2+ at the flood peak was 3.3 times lower than during previous low flows; SO 4 2− was 2.7 times lower; HCO 3− , Ca 2+ , and Na + were 2-2.5 times lower; and K + was 1.4 times lower. There was a slow increase in concentrations of these ions immediately after the flood peak, and the higher the concentration of the ion itself, the longer the recovery of the initial low-flow level. Peak concentrations of HCO 3− and Ca 2+ were reached on 28 June, while the remaining ions reached their peak on the evening of 29 June. All ions returned to their previous low flow levels between 30 June and 1 July. However, due to a slight rainfall on 1 July, there was a slight decrease in their concentrations. Mineral silicon, whose concentration at the flood peak was almost two times lower than previously, and orthophosphates (1.5 times lower) followed the same patterns. The concentration of total dissolved organic matter also decreased by 1.8 (according to the COD value in the filtered sample) when the flood passed.

Simulation results verification and accuracy estimates
The COSMO-CLM simulations were verified according to standard approaches including timeseries comparison between stations observations and the nearest neighboring model grid point data. Timeseries compared on daily, 6-h scales for most stations cited in previous Sections and depicted on the Fig. 2. Pluviograph timeseries were compared on 10-min (observations) and 15-min (simulations) scales, as well hourly sums were calculated for both timeseries and compared. In all cases, average mean error (bias), RMSE and correlation were estimated. The simulated rainfall at the MSU Observatory station located within the Setun basin was compared with high-resolution 10-min observations using decoded and digitized pluviograph data (Online Resource 5). Despite the different time steps, we analyzed the overall precipitation intensity course according to model and measurements. The model successfully predicted first rainfall values on 29 May, but with average a half day time lag. The rainfall on 30-31 May was not captured; precipitation was reproduced during other time period and values were absolutely overestimated. At the same time, the main observed precipitation maximum on 29-30 May 2020 was not captured (5 mm per 10 min, up to 15 mm sum per 1 h).
Overall verification according to these data shown just zero correlation (0… − 0.2), however low biases (− 0.15 mm for hourly sums and − 0.3 for 10-15-min values) and significant RMSE (2.7 for hourly sums and 0.88 for 10-15 min values) for both hourly sums and high-resolution intensities. The estimates (Online Resource 5) include comparison between simulated and observed 3-h and daily precipitation sums for 2020 case. This follows the abovementioned results including underestimating daily sums for 29. And 30. May and overestimating it for 31. May by the COSMO-CLM model. Table 4 provides verification statistics for stations data available. Although biases and RMSE are too large reaching 6-8 mm and more than 20 mm for separate stations accordingly, correlation coefficient values are satisfactory just for all stations. At the same time, comparisons on the smaller timescales (3-6 h precipitation sums, Table 3) demonstrates large biases and RMSE in addition to no correlation. These results justify the correct reproduction the large-scale factors of precipitation formation including the warm front intensification and ambient favorable conditions; however it reveals significant mismatches in meso-scale features, such as specific time and regions of squall lines formation, location, development and movement, which are subjects to improve in future investigations (Fig. 8).
The 10-min pluviograph observations at the MSU Observatory station compared with high-resolution simulation data for the nearest COSMO-CLM grid point precipitation amounts (Figs. 9 and 10). In this case model has underestimated the precipitation significantly, reproduced the  Overall verification according to these data shown just zero correlation, however low biases (-0.35 mm for hourly sums and -0.03 for 10-15 min values) and significant RMSE (3.3 for hourly sums and 0.7 for 10-15 min values) for both hourly sums and high-resolution intensities. The key reason for such failed reproduction is that the main intense rainfall area shifted just few km to the west in the model (see Online Resource 4). In this way, the model has reproduced the precipitation maximum of similar intensity, but shifted on few model grids. This is an example of small-scale stochastic features impacting the model behavior and shortcomings of point-by-point comparisons for precipitation amount field, specifically. According to daily sums comparison with simulated values (Tables 3 and 4), there are significant errors; however, RMSE is distinctly smaller than for 2020 case. Correlation coefficients are also less on average, but for some grid points are greater than 0.8 including points corresponding to stations of specific interest Nemchinovka and Vnukovo within the Setun basin. In this case we can conclude the local conditions nearby the Setun River basin were reproduced more satisfactory compared with 2020 case, because the main rainfall events of smaller scales were observed and simulated within narrower time interval and resulted in more similar accumulated daily sums. Despite the model didn't reproduce the correct location of squall lines, it did this for the actual event of comparable intensity. At the same time, more detailed 3-6-hourly sums indicated less biases and RMSE, but poor correlation, which justifies satisfactory course of precipitation values. Therefore, the COSMO-CLM model has captured the mesoscale convective complex in general and formed squall lines correctly, but did not reproduce the time and locations of heavy rainfall events, although the one affected the Setun basin was reproduced satisfactory.
The observed uncertainties require detailed analysis of unreproduced features and causes, as well application of the weather radar data for verification. Spatial verification methods could be useful to estimate mismatches in spatial

Precipitation impact of water runoff and urban conditions
We implemented statistical analyses to verify the observed impacts of extreme rainfalls on the flood events formation. Of the two events considered, the event on 28-29 June 2021 gives a much clearer insight on the propagation of the extreme rainfall into the river network in the Setun catchment. The first floodpeak demonstrated the concentration time of the impermeable part of the catchment, directly contributing the rainfall it received to the river network. For the western part of the catchment, it is estimated around 2 h, and for the whole catchment, around 6 h. This corresponds with the concentration time estimates derived from the long-term time-series rainfall-runoff correlation analysis in the Setun River catchment conducted earlier (Chalov SR, Platonov SV, Moreydo VM, et al. (2022) The small urban river runoff response on 2020 and 2021 extreme rainfalls on the territory of Moscow. Russ Meteorol Hydrol [Manuscript in preparation]). The entire catchment response, involving infiltrated and hillslope-routed fraction of rainfall can account for 6 to 11 h for the western and eastern parts, respectively. Compared to the average concentration time in the Lusyanka nonurbanized catchment, which is 16 h, it is more than twice as rapid. This can be explained solely by larger area of paved impermeable surfaces in the Setun River catchment.
To compare the catchment response to extreme precipitation event in urbanized and non-urbanized areas, we took a case study in the Lusyanka River catchment located in 120 km to the West of Setun catchment (Fig. 10). While the Lusyanka River catchment area is comparable-170 km 2 against 186 km 2 for the Setun River, the landscape cover is highly contrasting, mostly forested or covered with grassland.
To demonstrate the different catchment response, we analysed several rainfall events in July 2020, recorded at the nearby station located in Mozhaysk town (27,509 in Fig. 10) in 25 km to the East of the catchment (Fig. 11). On 7 and 8 July, 2020 the region was hit by a large rainstorm, resulting in over 80 mm rainfall in 6 h. The catchment response was rapid, but not pronounced, as the stage began gradually increase from around 50 cm to around 100 cm in the following hours. Next rainfall event on 12 July with only 20 mm of precipitation resulted in much larger catchment response-in nearly 6 h, the stage increased threefold from 60 to 180 cm. The last 40 mm rainfall on 14-15 July resulted in further ascent of river stage from 100 to 260 cm in 10 h.
In general, the hydrograph shape of the three analyzed events clearly shows the different response to large rainfall events as the catchment is gaining moisture. After the first extreme rainfall event, the response was slow and but gradual, as most of the water infiltrated into the soil and only a fraction was routed into the river network. The next two events show much faster response, as the infiltration capacity was exceeded and most of the rainfall resulted in overland flow. These evidences confirm that the urbanization effects might represent a significant increase in flooding for small catchments, whereas at larger scales, the effects are highly complex and a result of sub-catchment responses (Miller and Hutchins 2017).

Water quality impact of extreme flow events over urban catchment
There are several types of temporal trends in dissolved forms of chemical elements in the river during storm rainfall. Most of the trace element particulate concentrations increase with increasing discharge and reach more or less constant values during the floods, whereas dissolved concentrations decrease. Four main groups of elements might be considered based on their flood-driven behavior. The first group is composed of mostly soluble elements. All major ions (HCO 3− , SO4 2− , Ca 2+ , Mg 2+ , Na + , K + ), mineral silicon, orthophosphates, total dissolved organic matter belong to the group of components whose concentrations decrease with the rise in water discharge (Fig. 12a). Concentrations found in groundwater supplying the river during low-water periods, especially in urban conditions, are always higher than those found in rainwater. However, there is no strong enrichment of rainwater by these ions with slope runoff. The turbidity of the Setun River has a significant impact on the chemical concentrations during high water flow. During the June-July 2021 flood, the SSC was 2 or more times higher than the previous low-flow level (26 mg/l), and at the peak of the flood reached over 500 mg/l. Increased element transfer in suspended form significantly changes the total load and composition of nutrients and organic substances in the river. At the maximum water discharge, concentration of dissolved inorganic phosphorus slightly decreased, while the concentration of its suspended inorganic form increased by 5.6 times, and that of the suspended organic form-by 8.8 times. The percentage of suspended forms of phosphorus thus increased from about 50 to 80% or more during the flood. The percentage of suspended form of total nitrogen also increased during the flood from 10-20% to a maximum of 40% (Fig. 12b). The values of BOD 5 and COD in unfiltered water samples increased by 2 and 2.8 times, respectively, at the peak of the flood compared to low-flow levels. Therefore, flushing of particulate matter from the watershed is an important mechanism of nutrient and organic pollution of urban river water during heavy rainfall and flood formation. Among the nutrients, in contrast to orthophosphates and COD values, the content of dissolved organic phosphorus during the flood significantly increased by 5.7 times compared to the low-flow values, indicating probable flushing of phosphorus-containing organic compounds from the watershed. Similarly, concentrations of total dissolved nitrogen and phosphorus, as well as nitrite nitrogen, almost doubled at the flood peak, which is explained by strong relationship with particulate concentrations (Walling et al. 2008). The increase of sediment concentration induced growth in both particulate and dissolved concentration of wide range of trace elements (Fig. 12c). This group is composed of mostly elements linked to the particulate matter and mostly consists of insoluble and well-sorted elements. Both with increase of SSC, the intensification of sorption lead to significant increase of Mn (20.7-fold), Cs (11.6-fold), Cd (6.8-fold), Al, Nd, and Er (6.1-6.2-fold) dissolved concentrations. Concentrations of some other elements increased 2-3 times. The patterns of temporal variability of concentrations at the falling limb of the flood differed between these elements: for some (Al, Mn, Co, Cd), the rise at flood peak was followed by a sharp decline; by the next day, concentrations returned to almost low-flow level. For others, this decline was extended by several days.
Li, B, Cr, As, Br, and Sr were negatively related to water discharge during flood events-their concentrations (in dissolved form), as well as total dissolved solids concentration, decreased during higher water discharge (Fig. 12d). The decrease in concentrations at flood peak for these elements did not exceed 2.5 times in comparison with low-flow concentrations, and their return to initial values was relatively slow (to June 30-July 1, as was found for major ions).
There was no significant change in concentrations or range of variability for a number of elements, which included Cu, Zn, Mo, Sn, Pb, Ba, La, Cs, U, etc., during the rainfall flood when compared to periods of low flow.
The chemical composition of the river water during floods significantly varied from spring snowmelt floods. During the spring flood, concentrations of major ions were mostly negatively correlated with water discharge, although for Ca 2+ , Mg 2+ , K + , and SO 4 2− a slight increase in the beginning of the rising limb was observed. Concentrations reached their minimum 1 day after the peak discharge. COD in filtered and unfiltered samples increased 2-2.4 times compared to preceding low flow (to 100-120 mgO/L). Dissolved N and P reached 1.3-2 times their baseflow concentrations, their particulate forms increased 4-9 times due to high suspended sediment content (> 300 mg/L), maximum nutrient concentrations were observed before peak water discharge. Minimum concentrations of Li, B, Br, Sr, Cu, Cd, and Ba coincided with peak discharge and were 1.2-3 times lower than their low flow concentrations, for Sb peak concentration was, on the contrary, was 4.3 times higher. Al, V, and Ni increased rapidly 2-4.9 times in the beginning of the flood and then returned to baseflow levels.
Major ions showed similar patterns during both spring flood and flash flood: their concentrations during the snow melt decrease. However, for some of the ions (Ca 2+ , Mg 2+ , K + , SO 4 2− ), as well as for the mineral silicon, concentrations increased slightly along with the river discharge at the beginning of the spring flood. This may be the result of the intensive washout of salts that accumulated on the watershed during winter, along with insufficient dilution by low-mineralized snowmelt water. Due to the greater duration of the spring flood than that in summer one can determine that the minimum concentration of major ions comes with a delay of 1 day from the maximum water discharge. At the same time, the duration of the rising and falling limb for the flood wave itself, and for the ion flow, generally correspond, meaning that the rate of increase and decrease in river runoff is close to the rate of change in the chemical composition of the river.
In contrast to the flash flood, there is a more significant (up to 2.4) increase in the COD value in spring, as well as an increase in dissolved mineral phosphorus content. The total concentrations of dissolved nitrogen and phosphorus during the spring flood are much higher than during low-flows ; showing a non-linear relationship with water discharge: at the beginning of the flood their concentrations increase smoothly, while at the peak of water discharge the content of suspended nitrogen and phosphorus increases during the floods by 4-6 times, increasing the suspended matter content to 300 or more mg/l.
The maximum concentration of nitrites found locally during the spring of 2021 occurred in the middle of the rising limb of the flood, however, during the further rise and fall its concentration was close to that of the low-flows.
Among trace elements, the dynamics of Li, B, Br, and Sr generally mimicked the changes that occurred in the water discharge, while the dynamics of Cr and As content had a more complex form (Cr, monotonic decrease throughout the spring flood; As, increase in concentrations at the peak of the flood). In spring, however, the inverse was observed for Cu, Cd, Zr, Ba, and some other elements, for which such a relationship was not observed in summer.

Implications for water management systems
The observed changes in water regime and high levels of suspended solids and pollutants concentrations during a storm event in the urban Setun River which significantly exceeds the Russian guidelines for surface water are explained by low standards of environmental protection. Moscow is currently operating a storm water drainage system constructed in the middle of the twentieth century. This system accumulates stormwater separately from domestic sewage and treats it primarily from suspended solids and oil products. This situation was quite common among large cities around the world just a few decades ago (Brown et al. 2009;Hale 2016;Goulden et al. 2018). Stormwater facilities, unable to absorb the increasing runoff and pollutants load from the urban watershed (Vorobevskii et al. 2020;Lu et al. 2021), often were unable to provide proper treatment. Therefore, during extreme rainfall and the combined sewage system, a mixture of fecal sewage along with stormwater entered urban rivers (Lee and Bang 2000;Gasperi et al. 2012), contaminating them dramatically.
Despite possible hazards due to hydrological changes and contamination with urban pollutants , small rivers of Moscow are vital for the city's functionality and its economic development. The Setun River catchment area is constantly integrating in the urban landscape with green parks and green infrastructure (Klimanova and Illarionova 2020) and provide, among others, aesthetic and recreation services. At the same time, the observed measures are not integrated into managing of stormwater loads which is widely developed in many cities (Qiao et al. 2019;Rosenberger et al. 2021). The Setun basin seems to have great potential for stormwater management by implementing Low Impact Development (LID) such as stormwater ponds, rain gardens and permeable pavements. The examples of other large cities in Europe, USA, China (Damodaram et al. 2010;Ahammed 2017;Prudencio and Null 2018;Bohman et al. 2020) show that the implementation of retention ponds leads to a 70% reduction of sediment load to the rivers and a 20-40% reduction of dissolved pollutants load to the rivers (Bedan and Clausen 2009;McPhillips et al. 2021). Natural based solutions for urban stormwater control also expand both social infrastructure and are widely supported by citizens (Darnthamrongkul and Mozingo 2021). At the same time, the implementation of green infrastructure requires a complete revision of the existing drainage system throughout the entire catchment (Barbosa et al. 2012;Cettner et al. 2013;Qiao et al. 2020) and designing the spatial organization of LID facilities, which needs a complete understanding of the hydrology and pollutant loads on different parts of the catchment (Rosenberger et al. 2021). Thus, this study is the first step toward designing sustainable stormwater management in the basin of Moscow's urban streams.

Conclusions
The present study is the first attempt to use hydro-meteorological records and models for the largest urban area in Russia to investigate climate change and urbanization impacts on flooding and water quality. The study is based on the most comprehensive overview of the precipitation datasets for the 6 stations within Moscow city and the neighboring Moscow region for a 1966-2021 period. The standard configuration of COSMO-CLM model ver. 5.12 with the last version of the TERRA-URB urban canopy model has reproduced the synoptic-scale and main mesoscale features of 2020 and 2021 summer extreme precipitation events satisfactorily and could be applied and developed in further studies. In some cases, it failed to capture location, time span and intensity of rainfall events on hourly and sub-hourly timescales due to local shifts of precipitation areas in space and time. The COSMO-CLM outputs were combined with urban flow analyses for the Setun River basin which is based on 30-min frequency datasets. Two particular rainstorm events, on 29-31 May 2020, and on 28-29 June 2021 provided a clear insight on the propagation of the intense rainfall into the river network in the Setun catchment.
The key research outcomes of this study are grouped into three categories according to the results of precipitation, flow discharge and water quality data processing: 1. The long-term extreme precipitation data indicates mostly insignificant increase of absolute yearly daily maxima. At the same time, there is a statistically sig-nificant rise of moderate extremeness ETCCDI indices (R95TOT, R20mm) specifically within the Setun basin (at MSU Observatory, and Nemchinovka stations). Meanwhile, indices characterizing the largest precipitation amounts among the ETCCDI classification do not reveal any significant trends for a considered period. This suggests that urban precipitation patterns in Moscow city, emerging in moderate rainfall events frequency, are increased. 2. Whereas the observed increases in frequency and intensity of extreme rainfall are physically consistent with global warming in the twentieth century, the water discharges in small urbanized rivers are entirely driven by urbanization effects, firstly larger area of paved impermeable surfaces. The available evidence has been found for the Setun River catchment concerning twice as rapid rainstorm flood catchment response time compared to the average concentration time in the non-urbanized catchment which is 16 h; significant impacts of values and movement of rainfall core to the small river flooding. 3. The effects of rainstorms in the urbanized catchment of Moscow city are highly complex. It was observed that in spring there various genetic types of waters interact in a complex way within the urban landscape. This alongside the greater duration of the spring flood process itself in comparison to a flash flood, causes some differences in the flash flood dynamics of the chemical composition of water, and a more complex relationship between the chemical parameters and the water discharge. At the same time, many similarities were found between spring floods and flash floods: inverse dynamics of water discharge and concentrations of Li, B, Br, and Sr, attribution of the maximum concentrations of Al and Mn to a raising limb of the flood, and their sharp decrease immediately after that. It was also noted that the level of concentrations post flood returned almost immediately to pre-flood levels for both types of floods.
Due to lacks in wealthy hydro-meteorological records, especially in term of urban rivers flows, further research to combine rainfall intensity, convective storms, urban flooding and first flush events and their changes in the future will help to consider and make adequate decisions for designing sustainable stormwater management in the urban areas.
Author contribution Conceptualization, original draft preparation-Sergey Chalov; numerical experiments conducting and evaluation, precipitation data analysis, writing-Vladimir Platonov; the rainfallrunoff patterns analysis-Vsevolod Moreido; methodology, validation, writing-Oxana Erina, Dmitriy Sokolov, Maria Tereshina, Mikhail Samokhin; precipitation data preparation and visualization-Yulia Yarinich; review, editing-Nikolay Kasimov. All authors have read and agreed to the published version of the manuscript.
Funding Field studies were supported by Russian Science Foundation project 19-77-30004. The analytical experiments were done under Ministry of Science and Higher Education of Russian Federation project 075-15-2021-574. COSMO-CLM model setup is a part of RFBR project 21-55-53039. The methodology of this study is developed under the Interdisciplinary Scientific and Educational School of Lomonosov Moscow State University «Future Planet and Global Environmental Change» and Kazan Federal University Strategic Academic Leadership Program ("PRIORITY-2030"). The research is carried out using the equipment of the shared research facilities of HPC computing resources at Lomonosov Moscow State University. Streamflow patterns analysis was carried out under Governmental Order to Water Problems Institute, Russian Academy of Sciences, subject no. FMWZ-2022-0003, project 3.7.
Data availability The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.
Code availability Not applicable.

Declarations
Ethics approval Not applicable.

Consent to participate Not applicable.
Consent for publication Not applicable.