Impact of river reconstruction on groundwater flow during bank filtration assessed by transient three-dimensional modelling of flow and heat transport
- 142 Downloads
Bank filtration (BF) is an established indirect water-treatment technology. The quality of water gained via BF depends on the subsurface capture zone, the mixing ratio (river water versus ambient groundwater), spatial and temporal distribution of subsurface travel times, and subsurface temperature patterns. Surface-water infiltration into the adjacent aquifer is determined by the local hydraulic gradient and riverbed permeability, which could be altered by natural clogging, scouring and artificial decolmation processes. The seasonal behaviour of a BF system in Germany, and its development during and about 6 months after decolmation (canal reconstruction), was observed with a long-term monitoring programme. To quantify the spatial and temporal variation in the BF system, a transient flow and heat transport model was implemented and two model scenarios, ‘with’ and ‘without’ canal reconstruction, were generated. Overall, the simulated water heads and temperatures matched those observed. Increased hydraulic connection between the canal and aquifer caused by the canal reconstruction led to an increase of ~23% in the already high share of BF water abstracted by the nearby waterworks. Subsurface travel-time distribution substantially shifted towards shorter travel times. Flow paths with travel times <200 days increased by ~10% and those with <300 days by 15%. Generally, the periodic temperature signal, and the summer and winter temperature extrema, increased and penetrated deeper into the aquifer. The joint hydrological and thermal effects caused by the canal reconstruction might increase the potential of biodegradable compounds to further penetrate into the aquifer, also by potentially affecting the redox zonation in the aquifer.
KeywordsBank filtration Groundwater/surface-water relations Groundwater modelling
Impact de la reconstruction d’une rivière sur l’écoulement des eaux souterraines via la filtration sur berge évalué par un modèle tridimensionnel en régime transitoire de l’écoulement et du transport de chaleur
La filtration par les berges (FB) est une technologie éprouvée de traitement indirect de l’eau. Les gains de qualité par FB dépendent de la zone d’appel en sub-surface, des proportions de mélange (eau de rivière versus eaux souterraines environnantes), des distributions spatiale et temporelle des temps de transit souterrains, et des modalités de température souterraine. L’infiltration des eaux superficielles vers l’aquifère adjacent est déterminée par le gradient hydraulique local et par la conductivité hydraulique du lit de la rivière, qui peut être altérée par les processus de colmatage naturel, d’affouillement et de décolmatage artificiel. Un programme de suivi à long terme a permis d’observer le comportement saisonnier d’un système de FB en Allemagne, et son développement durant le décolmatage (reconstruction du canal) puis six mois après. Une modélisation transitoire des écoulements et du transport de chaleur a été mise en œuvre pour quantifier les variations temporelles et spatiales du système de FB, puis deux scénarios ont été générés, avec et sans la reconstruction du canal. Dans l’ensemble, les charges hydrauliques et les températures simulées sont concordantes avec les observations. L’amélioration de la connexion hydraulique entre le canal et l’aquifère due à la reconstruction du canal a pour conséquence une augmentation de 23% de la proportion déjà élevée d’eau extraite par FB dans les installations hydrauliques voisines. La distribution des temps de transit dans le sous-sol s’est décalée vers des temps de transit significativement plus courts. Les cheminements d’écoulement souterrain présentant des temps de transit inférieurs à 200 jours ont augmenté de 10%, et ceux inférieurs à 300 jours de 15%. De manière générale, le signal périodique de température, et les extrêmes de température en été et en hiver, ont augmenté et pénétré plus profondément dans l’aquifère. Les effets cumulés de l’hydrologie et de la thermique dus à la reconstruction du canal seraient susceptibles d’accroître le potentiel de pénétration de composants biodégradables dans l’aquifère, en affectant également la zonation redox de l’aquifère.
Impacto de la restauración de un río en el flujo de agua subterránea durante la filtración en las márgenes, evaluado mediante la modelización tridimensional transitoria del flujo y el transporte de calor
La filtración en las márgenes de un río (BF) es una tecnología establecida para un tratamiento indirecto del agua. La calidad del agua obtenida a través del BF depende de la zona de captura en el subsuelo, la proporción de mezcla (agua de río versus agua subterránea), la distribución espacial y temporal de los tiempos de trayectorias y de los patrones de temperatura en el subsuelo. La infiltración de agua superficial en el acuífero adyacente está determinada por el gradiente hidráulico local y la permeabilidad del lecho del río, que podría verse alterada por los procesos naturales de obstrucción, lavado y descolmatación artificial. El comportamiento estacional de un sistema de BF en Alemania, y su desarrollo durante y aproximadamente seis meses después de la descolmatación (restauración del canal), se observó con un programa de seguimiento a largo plazo. Para cuantificar la variación espacial y temporal en el sistema BF, se implementó un modelo de transporte de flujo transitorio y calor y se generaron dos escenarios modelo, ‘con’ y ‘sin’ restauración de canales. En general, las cargas de agua simuladas y las temperaturas coincidieron con las observadas. El aumento de la conexión hidráulica entre el canal y el acuífero causado por la restauración del canal condujo a un aumento de ~23% en la ya alta proporción de agua BF extraída por las obras hidráulicas cercanas. La distribución del tiempo de la trayectoria subterránea cambió sustancialmente hacia tiempos de más cortos. Los recorridos de flujo con tiempos de trayectorias <200 días aumentaron en ~10% y los de <300 días en 15%. En general, la señal periódica de temperatura, y las temperaturas extremas de verano e invierno, aumentaron y penetraron más profundamente en el acuífero. Los efectos hidrológicos y térmicos conjuntos causados por la restauración del canal podrían aumentar el potencial de los compuestos biodegradables para penetrar aún más en el acuífero, también al afectar potencialmente la zonificación redox en el acuífero.
河岸渗滤作用(BF)是一项成熟的间接水处理技术。BF补给水的质量取决于地下截获区, 混合比(河水与周边的地下水), 地下运移时间的时空分布以及地下温度分布。进入相邻含水层的地表水入渗量取决于当地的水力梯度和河床渗透率, 可通过自然堵塞, 冲洗和人工清淤的方法来改变。通过长期监测项目, 观察到了德国BF系统的季节性行为, 以及在清淤(运河改造)期间和之后约六个月的情况。为了量化BF系统的空间和时间变化, 开发了瞬态水流和热传输模型, 并设置了两种模型场景:“有”和“无”运河改造。总体而言, 模拟的水头和温度与观察到的相符。运河改造增加了运河与蓄水层之间的水力联系, 导致附近自来水厂抽取的BF水比例已经提高了约23%。地下运移时间分布基本上转向了较短的运移时间。时间 < 200天的流路增加了〜10%, 而 < 300天的流路增加了15%。因此, 周期性的温度信号以及夏季和冬季的温度极值都增加了并影响到含水层。运河改造引起的水文和热的耦合效应会增加生物可降解化合物进一步进入含水层的可能性, 并潜在影响含水层的氧化还原带。
Impacto no fluxo de água subterrânea com a reconstrução do rio durante a filtração em margem avaliada pela modelagem transiente tridimensional do fluxo e transporte de calor
A filtragem em margem (FM) é uma reconhecida tecnologia indireta de tratamento de água. A qualidade da água obtida pela FM depende da zona de captura em subsuperfície, da taxa de mistura (água do rio versus água subterrânea local), distribuição espacial e temporal dos tempos de viagem em subsuperfície e padrões de temperatura em subsuperfície. A infiltração das águas superficiais no aquífero adjacente é determinada pelo gradiente hidráulico local e pela permeabilidade do leito do rio, que pode ser alterada pelos processos de colmatação natural, autolimpeza do leito do rio e descolmatação artificial. O comportamento sazonal de um sistema de FM na Alemanha, e seu desenvolvimento durante e cerca de seis meses após a descolmatação (reconstrução do canal), foram observados com um programa de monitoramento de longo prazo. Para quantificar a variação espacial e temporal no sistema FM, um modelo transiente de fluxo e transporte de calor foi implementado e foram gerados dois cenários, ‘com’ e ‘sem’ reconstrução do canal. No geral, as cargas hidráulicas e temperaturas da água simuladas corresponderam às observadas. O aumento da conexão hidráulica entre o canal e o aquífero, devido a reconstrução do canal, levou a um aumento de ~23% na já alta parcela de água da FM captada pelas estações próximas. A distribuição do tempo de viagem no subsolo mudou substancialmente para tempos de viagem mais curtos. Os caminhos de fluxo com tempos de viagem <200 dias aumentaram ~10% e aqueles com <300 dias em 15%. Geralmente, o sinal cíclico da temperatura, e as temperaturas extremas de verão e inverno, aumentavam e atingiram maiores profundidades no aquífero. Os efeitos hidrológicos e térmicos em conjunto, causados pela reconstrução do canal, podem aumentar o potencial de compostos biodegradáveis para penetrar ainda mais no aquífero, também afetando potencialmente a zona redox no aquífero.
Bank filtration refers to a process in which river water infiltrates at the riverbed, migrates through the aquifer and is finally extracted at pumping wells that are located adjacent to the river (Hiscock and Grischek 2002; Kuehn and Mueller 2000). It is a pretreatment option often used for conventional drinking water supply. In Europe, bank filtration is widely used, and contributes 16% of the potable water supply in the whole of Germany; it may even dominate as a treatment option locally (Hiscock and Grischek 2002; Massmann et al. 2008a). Along the subsurface flow paths from the river towards the pumping wells, raw water quality can be improved through several beneficial attenuation processes such as equilibration of the temperature variation, pathogen filtration, mixing, biodegradation and sorption; those processes act on suspended solids, biodegradable compounds, other dissolved organic carbon (DOC) or bacteria, viruses, parasites, etc. (Hiscock and Grischek 2002; Sharma et al. 2012). The quality of water gained via bank filtration strongly depends on the subsurface capture zone, and the mixing ratio between river water versus ambient groundwater, as well as the spatial and temporal distribution of subsurface travel times (Hiscock and Grischek 2002). For all properties mentioned in the preceding, the hydraulic conductivity of riverbed materials plays a decisive role (Hiscock 2005; Massmann et al. 2008a; Vogt et al. 2009). The formation of a clogging layer is almost inevitable during the long-lasting operation of bank filtration systems (Schubert 2006). It leads to a decrease of the riverbed hydraulic conductivity, limiting the infiltration rate of surface water and thus to a decreased recoverable ratio of surface water at the production wells (Goldschneider et al. 2007; Tufenkji et al. 2002; Ulrich et al. 2015); however, it would not decrease infinitely but rather reach a state of equilibrium (Gutiérrez et al. 2018). The formation of a clogging layer is commonly caused by physical (sedimentation of fine-grained particles) and biogeochemical (growth of microbes, alga and biofilms) factors, while the dynamic hydrological condition also plays an important role (Ghodeif et al. 2016; Gutiérrez et al. 2018; Rinck-Pfeiffer et al. 2000; Schubert 2002). Natural declogging processes (decolmation) involve an increase in flow velocity giving rise to removal and restorage of river sediments and, in part, by digging activities of the animal stock at the river bottom. In managed aquifer recharge schemes, the clogging layer is regularly abraded to re-establish a higher hydraulic conductivity (corresponding recharge rates may increase by the factor of 10; Greskowiak et al. 2005; Escalante et al. 2015), although artificial excavation in surface-water beds at bank filtration sites has rarely been studied.
Furthermore, the quality of water gained via bank filtration depends on subsurface temperature patterns (Munz et al. 2019). Temperature is a key factor controlling redox reactions, enzyme activity and biodegradation of organic substances (Maeng et al. 2010; Miettinen et al. 1996; Umar et al. 2017). Besides the short-term fluctuations of the daily temperature signal in the riverbed itself (e.g. Munz and Schmidt 2017), seasonal temperature changes penetrate to greater aquifer depth (Kalbus et al. 2006), strongly affecting the bioactivity of the microorganisms (Massmann et al. 2008b). If the subsurface temperature falls below 14 °C, biodegradation has been shown to be strongly constrained (Massmann et al. 2006; Prommer and Stuyfzand 2005). At an artificial recharge and bank filtration site at Lake Tegel in Berlin and at the site used for this study in Potsdam, Germany, an aerobic condition develops between the river and the pumping well in winter, while in summertime the oxygen is consumed within the first meter below the riverbed followed by the prevailing anaerobic condition (Munz et al. 2019; Gross-Wittke et al. 2010; Massmann et al. 2006).
Numerical models are widely adopted in water-resource studies (Umar et al. 2017) and are the most common tool for understanding the flow regime at a bank filtration site. By recalibrating a three-dimensional (3-D) groundwater flow model against pumping-test results, Schafer (2006) confirmed riverbed colmation at a bank filtration site at the Ohio River (USA) several months after commissioning. Shankar et al. (2009) quantified the contributing ratio between infiltrating water and natural groundwater to the production well, and also delineated the catchment area. May and Mazlan (2014) evaluated the change of the groundwater flow field under the influence of heavy abstraction along Langat River, Malaysia, by comparing the scenarios with and without pumping activities in a 3-D numerical model. By considering seasonal temperature variations in reactive transport studies, more accurate simulation results can be achieved for water quality changes (Prommer and Stuyfzand 2005) as well as for variation of O2 and NO3− during bank filtration (Henzler et al. 2016). Changes in the flow field (e.g. by colmation or decolmation) will influence the heat transport through the riverbed and further influence the temperature distribution along the subsurface flow path to pumping wells, but this issue has not been investigated quantitatively in the field so far.
The objective of this research was to determine the hydraulic connection between the river and groundwater at a bank filtration site with substantial geological heterogeneity and further investigate the temporal changes caused by a river reconstruction (artificial deepening by removal of till sediments and river bank renewal). A 3-D flow and heat transport model has been calibrated first against water heads and then against temperature data to (1) investigate the groundwater flow regime, (2) quantify the changes of infiltration rates, the mixing ratio between river water versus ambient groundwater, and the spatial and temporal distribution of subsurface travel times, and (3) quantify the changes in the subsurface temperature distribution brought on by the reconstruction. The coupled water flow and heat transport simulations have been performed as a transient 3-D model for a period of 4.5 years, delivering spatiotemporal distributions of the hydraulic and thermal properties.
The bank filtration takes place mainly through Quaternary aquifers, which are semi-confined aquifers stratified by glacial till layers from the later Pleistocene. Featured as rather impermeable material, the glacial tills, from the bottom to the top, were formed in the Saale I glacial period, Saale II glacial period and Weichsel I glacial period, but structure and extent are substantially different between them. The glacial tills of Saale-I (SI-GT) and Saale-II (SII-GT) cover most of the study area and join at some places forming thicker glacial blocks. The glacial till formed in the Weichsel I glaciation (WI-GT) is the youngest and uppermost glacial till in the study area, and it is not as continuous as the two former glacial deposits due to periglacial processes and erosion during the Holocene. Directly in the area below the ND, there is a lens of WI-GT, which covers half of the area of the canal bottom. Sand exists above the WI-GT, while mud and silt show up at the lake bottom. Stratified by semi-permeable glacial tills, the shallow aquifers could be divided into four sandy layers. At the bottom of the deepest of these Quaternary aquifers, a Miocene layer consisting of silt, fine sand, lignite, and clay exists, which separates these shallow aquifers from the deeper aquifers. Sapropel existed in the top few meters of Weisser See as well as Jungfernsee, which are located at the western and eastern end of the ND canal, respectively. Water exchange might occur at some locations due to the existence of hydrogeological windows possibly formed by erosion during the Pleistocene.
Observation wells and data collection
Group 1. Water heads were measured in 89 observation wells distributed within the study site—Fig. S2a of the electronic supplementary material (ESM)—in mid-April 2012. The filter screens are mostly 2 m long, while the bottom elevation varies from 19 to 34 m a.s.l. These water heads were used as the calibration targets for a numerical steady-state flow model.
Group 2. Water heads and temperatures were measured continuously at 15 observation wells using automatic data loggers with time intervals varying between 5 min and 6 h (Rugged TROLL 100, In-Situ, USA) with an accuracy of ±0.01 m for water head and ±0.3 °C for temperature. Measurements took place from June 2009 to December 2014. Most of these observation wells (except U100) are located between the canal and the pumping well group, and some of them are situated along the canal banks (Fig. 1b). Some of the observation wells (W03, W04, W07, and W09) comprise two adjacent wells with different filter screen elevations, which is denoted by HIGH (upper well screen) and LOW (lower well screen; Table 1). For the observation wells W03, W07 and W09, the data loggers in HIGH and LOW were set at the same elevation. These water heads were used as calibration targets for a transient flow model.
Group 3. Representative groundwater temperatures were measured in 12 observation wells weekly or biweekly, over a period of 18 months from June 2013 to December 2014. Except for two additional measurements in W05 and W06, the sampling points are identical with group 2. These temperatures were used as calibration targets for the heat transport model.
Configuration of the observation wells located between the canal and the pumping well group (except U100 which is located at the northern bank), sorted with increasing distance to the canal
Elevation (m a.s.l.)
Distance to the canal (m)
The river-water temperature for the simulation period was retrieved from two stations. A local data logger was put into operation in the canal from October 2012. Temperature data before that date were derived from a gauging station around 6 km upstream, in a parallel branch of the River Havel flowing through the city of Potsdam. The two temperature time series showed good data consistency and thus were used in combination to define the canal temperature for the entire simulation period. Temperature amplitudes were calculated as half of the difference between maximum and minimum temperature.
Daily precipitation and air temperature (2 m above ground) were measured at the nearby Leibniz Institute for Agricultural Engineering and Bioeconomy (about 2 km west of the study site). Average annual precipitation varied between a minimum value of about 382 mm in 2011 and a maximum value of 663 mm in 2010. For 2013 and 2014 the annual precipitation was about 468 and 436 mm, respectively. On average, 122 rainy days per year occurred between May 2009 and May 2015. Highest rainfall magnitudes and intensities were observed mainly in July/August each year. Maximum rainfall intensity was 42.8 mm/day. Also for high rainfall intensities, no surface runoff was noticed at the study site and thus also highly intense precipitation was assumed to recharge into the aquifer without substantial losses at the surface. Groundwater recharge was estimated based on a large-scale conceptual model (Landesamt für Umwelt Brandenburg 2019). In the study area, the evaluated ratio of (Precipitation minus Actual evapotranspiration)/Precipitation (1991–2010) varied from 18.7 to 19.7%, and a rounded-up value of 20% was taken as a fixed recharge ratio in the modelling work. Correspondingly, the estimated yearly groundwater recharge ranged between 76.1 mm/year in 2011, as driest year, and 138.2 mm/year in 2010, as wettest year. The pumping rates at the waterworks slightly varied between 6,000 and 7,000 m3/day. In July 2014 a remediation pumping (~2,090 m3/day) located east of the pumping wells of the waterworks (Fig. 1c) was started, but was not related to the reconstruction measures. Overall, groundwater recharge in the catchment area of the waterworks plays only a minor role due to the large contribution of river bank filtration to the water balance.
Groundwater equal potential lines were interpolated from synoptic head measurements of mid-April, 2011 (adapted from the environmental agency, MLUL Brandenburg) and used to define the model domain (Fig. 1c). About 1 km south of the pumping well group, a groundwater divide at a flat hilltop extends as a flowline at the hillslope towards the southern bank of Jungfernsee, forming a no-flow boundary at the south-eastern boundary. From the western end of the water divide, a flowline to the north, ending at the southern part of “Fahrländer See” [Lake Fahrländer], is taken as no-flow boundary of the model domain. Both are relatively far from the main area of interest between ND and the waterworks wells (Fig. 1b). The water levels of Jungfernsee and Fahrländer See were included in the model as fixed, but temporally variable water heads. The small northern area between these lakes is characterized by a hilly topography (Fig. 1c) formed by a glacial end-moraine. Here, local groundwater flow paths tend to directly end at the nearest surface water body and the groundwater divide approximately follows the surface-water divide delineating the northern groundwater subcatchment. Together, these boundaries define a model domain that includes more than the complete groundwater catchment area of the pumping wells and embeds the main area of interest between the pumping wells and the ND canal.
A total of 145 borehole profiles were available to construct the geological model (Fig. 1c). Most of the drilling locations are placed between the channel bank and pumping wells, while comparatively little information is available for the southern and western parts of the study site. Vertically, 67 of the drillings near the canal reached shallow depths only (mainly above 10 m a.s.l.), while 78 reached deeper layers, though only a few of them reached down to the Miocene layer (5.16 m to −12,8 m a.s.l).
Geological composition and layer attributes of the 3-D model
WI-GT lens, silt lens
Silt, sand, sapropel
Sapropel Weisser See/Jungfernsee
In respect to heat transport, in FEFLOW, the volumetric heat capacity (c) and heat conductivity (λ) are defined separately for solid and water phase, while the total porosity (ɛ) and longitudinal and transversal thermal dispersivities (λdisp) are set for the bulk porous medium. Total porosity mainly controls the volumetric ratio between the solid and liquid phases, as both phases contribute to heat transport, which is a difference to non-sorbing solute transport. The total porosity of the glacial tills and sapropel sludge were inferred from saturated gravimetric moisture content and grain density of sediment samples taken in the river section at the south end of the Fahrländer See. The average value of total porosity was 0.26. The thermal parameters for water are well known, while parameters of the solid phase could vary between sediments and have been subjected to calibration. The unsaturated zone on the top was set as ‘phreatic mode’ in FEFLOW, implying that its hydraulic conductivity adopted is taken as the saturated Kf-value times the ratio between the saturated thickness and full element height.
A steady-state model was run and calibrated with the aim of providing a consistent representation of the geological structure and their corresponding hydraulic parameters. The model calibration was based on the observed water heads at 104 locations distributed within the study site (data groups 1 and 2). A Dirichlet boundary of 29.5 m was chosen for the surface-water bodies. The pumping wells were set as multi-layer wells with a total pumping rate of 6,556 m3/day. The distribution of the total pumping rate to each layer in contact with the well screen was calculated based on the particular Kf-values and the hydraulic head distribution in this layer. In the steady-state model, water level and pumping rate represent the mean of values measured between mid-March and mid-April 2012. Groundwater recharge was set as 0.3 mm/day, which represents 20% of the mean precipitation of 2011 and 2012. Average daily groundwater recharge volume at the land surface of the model area was about 1,780 m3/day and thus only about 21% of the average volume pumped daily (sum of production wells of the waterworks and nearby groundwater remediation wells), reflecting a high fraction of bank filtration at this site. All outer boundaries not representing surface waters and the bottom of the model were set as no-flow boundaries according to the configuration of the model domain. Saturated hydraulic conductivities of the steady-state model were calibrated using FePEST, integrating the FEFLOW model into the PEST (Doherty 2015) framework. Parameter zones of homogeneous hydraulic conductivity were defined based on the existing geological units. The parameter ranges used in the calibration are given in Table S1 of the ESM). The objective function was based on the observed water heads and the anisotropy ratio of hydraulic conductivities, ensuring vertical conductivities (Kv) being linked to the horizontal ones (Kh, where Ky = Kx). A maximum limit was set so that Kh/Kv must remain larger than 2 in all geological units, according to Todd and Mays (2005). With respect to data quality and spatial relevance, different weighting factors were assigned to the hydraulic heads contributing to the objective function. The 12 observation points of group 2 that were directly located between the canal and the pumping well group were given a weighting factor of 10, while for the remaining observation points of group 1 it was set as 1.
Transient flow model
The transient flow model was set up to simulate the temporal variation in the groundwater flow system caused by the transient condition in surface water heads, pumping rates and groundwater recharge, and by changes in riverbed hydraulic conductivity induced by the river reconstruction (decolmation). The transient simulation started mid-June 2010 and covered a period of 4.5 years until the end of December 2014. The hydraulic heads of the steady-state model were used as initial condition for the transient simulation. The model boundaries were consistent in type with the steady-state model, but now applying daily average values obtained from the measured time series. The specific yield was set as 6% for glacial tills, 8% for silt and 23% for sand (Domenico and Schwartz 1998) and specific storage was taken as 1 × 10−4 m−1 (Freeze and Cherry 1979). Limited by the computation time, an automated calibration of the hydraulic conductivities was not feasible and therefore a manual calibration by trial and error was performed.
In the model, the river reconstruction was implemented in a way that all the grid elements located in the river area with a top elevation higher than 24.5 m were assumed to be influenced by the excavation. Since it was not possible to remove or replace inner grid cells during a transient simulation, the change in hydraulic resistance was instead implemented in an indirect way, i.e. all elements affected by the excavation kept their shape but obtained an increased hydraulic conductivity. Its value was fitted manually to reproduce the increasing water head after the start of the canal reconstruction and the finally adopted values were Kh = 1.16 × 10−4 and Kx,y = 2.31 × 10−5 m/s.
Transient heat transport model
The transient heat transport simulation was carried out from mid-June 2010 to the end of December 2014 corresponding to the transient flow model. Dirichlet boundary conditions were assigned for the surface water bodies (according to measured river temperatures) and land surfaces (according to measured air temperatures), while the remaining ones were defined as no-flux boundaries. All temperatures were applied as time series of daily-averaged values. The initial temperature in the model was set to 11.5 °C, which is the temporal average of all observed temperatures.
The thermal parameters, heat conductivity, volumetric heat capacity and thermal dispersivity (longitudinal and transversal) of the existing geological units were adapted during manual calibration. The parameter range used varied between 1.6 and 6.8 W m−1/K for thermal conductivity, between 1.8 and 3.2 MJ m−3/K for the volumetric heat capacity, and between 0.1 and 1 m for the longitudinal thermal dispersivity following established literature values (VDI 4640/1 2010; Abu-Hamdeh 2003; Stonestrom and Blasch 2003). Gelhar et al. (1992) summarized the studies existing at that time and have shown that longitudinal dispersivity is bigger than transversal dispersivity, in the horizontal as well as vertical direction. In this model, the transversal dispersivity was set to 10% of the longitudinal value, a common assumption and experimentally supported by the study of Huang et al. (2003). To illustrate the uncertainties in simulated temperatures resulting from the given range in thermal parameters, two extreme scenarios were adopted representing good ability (minimum c and maximum λ) and limited ability (minimum λ and maximum c) of heat transport through the saturated porous media.
The accuracy of temperature simulations was quantified by the deviation in timing of the maximum summer temperature or the minimum winter temperature (ΔΦ) by the availability and difference in temperature amplitude (ΔA) between simulated and observed temperatures. For these analyses, the measured groundwater temperatures of group 3 were used as they are assumed to be representative of the groundwater temperature and directly linked to the position of the filter screen of each observation well. Only for W05 and W06 no sampling data were available and instead logger temperatures were used for calibration.
Field observation of water heads and temperatures
As an annually periodic signal, the temperature time series take the form of a sine function. This original signal gets smoothed, damped and delayed on its way from the infiltration at the canal bottom and banks to the observation wells (Fig. 3b). The distance from wells W07LOW, W08, W09LOW to the canal is comparable, as well as the elevation of the data logger installed inside these wells (Table 1), but the temperature patterns in those wells feature significant differences in both amplitude and peak time. The average temperature amplitude in W07LOW was ~7.5 K, while the average amplitudes at W08 and W09LOW were much smaller (2.9 and 1.0 K, respectively). The peak time for W07LOW was postponed by about one month, but for W08 and W09LOW, the signals were shifted by about half a year.
Steady-state flow modelling
Calibrated hydraulic conductivities and anisotropy ratios (Kh/Kv) for the steady-state and transient models (WI refers to Weichselian, S to Saalian and GT to glacial till). Main changes are highlighted as italic
Sand layers on the top
8.0 × 10−5
8.0 × 10−5
Sand zone of WI-GT layers
1.0 × 10−4
1.0 × 10−4
Sand layers between WI-GT and SII-GT
5.6 × 10−5
8.1 × 10−5
Sand zone of SII-GT layers
8.4 × 10−5
8.4 × 10−5
Sand layers between SII-GT and SI-GT
2.9 × 10−5
1.9 × 10−4
Sand zone of SI-GT layers
3.5 × 10−4
1.2 × 10−4
Sapropel of Weisser See
8.9 × 10−6
8.9 × 10−6
Sapropel of Jungfernsee
1.3 × 10−5
1.3 × 10−5
4.4 × 10−7
2.0 × 10−6
1.1 × 10−6
1.1 × 10−6
Silty sand lens
1.1 × 10−6
1.1 × 10−6
2.0 × 10−6
2.0 × 10−6
2.0 × 10−7
1.2 × 10−6
Simulations of transient flow over 4 years
Pearson’s correlation coefficient and RMSE of the transient flow model. The observation wells with a correlation level smaller than 0.4 are highlighted as italic
Pearson’s correlation coefficient
The RMSE before reconstruction ranged between 0.09 and 0.33 m, with an average value of 0.24 m. Before the start of the reconstruction, three observation points (W03LOW, W07HIGH and W07LOW) were characterized by a moderate correlation level (<0.4). For W03LOW (Fig. S3f of the ESM), the mismatch is overemphasized by the uneven distribution of the sampling frequency. The observed dynamic changes before Oct 2012 were actually in good accordance with the simulation results; however, it takes much smaller weight on evaluating the simulating accuracy, due to the sparsity of the data points. For W07HIGH (Fig. 5b) as well as W07LOW (Fig. S3a of the ESM), the mismatch resulted from a period with an observed decreasing trend in October 2012 before an increasing trend started in April 2013. This deviation was also observed at W05, W06 (FS3c,d of the ESM), and, to a lesser extent, at W09HIGH (Fig. S3b of the ESM) and W09LOW (Fig. 5d); that is, at almost all observation points close to the canal except W08. This could indicate that part of the main driving force influencing the water heads in that winter period might be missing. The winter 2012/2013 was characterized by a long period in time with minimum daily temperatures below 0 °C, and average daily temperatures of only 2.4 °C. Potentially, freezing processes at the study site which were not represented in the numerical model might have caused this deviation between nominally observed and simulated water levels. Besides, for all simulations the hydraulic conductivity (also fluid viscosity and density) was constant in time and did not depend on temperature. This simplification may contribute to the general deviation of the simulation results compared to the observations. However, the temperature dependent variations in fluid viscosity and density for the given temperature range were rather small and thus the uncertainties caused by temperature dependent variations in hydraulic conductivity were much smaller than the general uncertainty in heterogeneous hydraulic conductivity itself.
Generally, the performance of the model including the reconstruction improved in respect to the scenario without reconstruction as expected. The increasing groundwater head after the start of reconstruction was captured at all observation points by the simulation including the hydraulic effect of reconstruction (Fig. 5 and Fig. S3 of the ESM). During the riverbed reconstruction, low-permeability riverbed sediments (including parts of the glacial tills) were dredged out, while the bank reconstruction led to the removal of lateral hydraulic resistances. This increased the hydraulic connection between the canal and aquifer. To maintain the consistent pumping rates after the reconstruction, a lower hydraulic gradient between the canal and pumping wells was sufficient due to the increased hydraulic connection. Therefore, the water heads in the pumping wells increased as well as in the area between the canal and the waterworks. Nevertheless, the RMSE on average increased for the period of reconstruction. This reflects that the system change is less well represented than the stable period previously, which could be attributed to fact that the real local changes in the riverbed are not known and that this period was a forward prediction and not subject to recalibration.
In the steady-state model, W03HIGH has shown the largest error among all the wells in group 2 and it continues this tendency in transient state (Fig. S3e of the ESM). After reconstruction, although the rising trend could be reflected in the scenario of including the reconstruction, the absolute error was smaller in the scenario without reconstruction. The deviation in W03HIGH could be caused by the existence of low-conductivity material along the groundwater flow path from the canal not being captured in the geological model. During and after reconstruction, W07HIGH and W07LOW were characterized by a moderate correlation level. For W07HIGH the short-term dynamics in the observed water level including a substantial drop of about 0.7 m on 12 January 2014, an increasing trend of around 0.2 m before a slow recovery of about 0.5 m until the end of February; after that, step changes were not captured by the model simulations (Fig. 5b). During this period the southern canal bank adjacent to W07 was renewed (see Fig. S1c of the ESM), i.e. a new geotextile was placed and covered with armour stones, potentially affecting the observed water levels at the nearby observation well W07 (distance between W07 and channel is only 13.2 m paired with a good hydraulic connection at this spot). From July 2014 onwards, the observed water level at W07HIGH continuously decreased until the end of the observation period (Fig. 5b). In December 2014 the water level at W07HIGH was close to the simulated water level of the model scenario without canal reconstruction, whereas the water level of the model scenario including canal reconstruction was about 0.6 m higher. This indicates that the better hydraulic connection between the canal and groundwater caused by the artificial decolmation of the canal bed at this canal section lasted for about 6 months only. The hydraulic connection there continuously decreased to its initial level before river reconstruction, probably by colmation. Comparable effects were observed for W07LOW (Fig. S3a of the ESM). These results agree with Schafer (2006) who confirmed that substantial riverbed colmation was observed at a bank filtration site of the Ohio River several months after commissioning. However, at this study site it was observed just for a local spot.
Influenced by the increasing infiltration rate, the travelling time distribution must also change. The results of forward particle tracking prove that the travel time distribution is shifted towards shorter travel times (shorter than 550 days); i.e. subsurface residence time of water in the aquifer is reduced (Fig. 6). Thus, flow paths with travel times between 100 and 550 days occurred more often in the model scenario including reconstruction compared to the scenario without reconstruction—for example, 55% of flow paths had travel times shorter than 300 days, whereas without reconstruction this would have been only 40% of the flow paths. Also an according decrease of long residence time (longer than 550 days) resulted, which has to be expected because pumping as the general driver was continuing on the same level as before.
The reconstruction work caused some major changes in subsurface flow regime. The northern capture zone shrinks towards the canal, i.e. no flow lines were traced back anymore to the northern lake “Krampnitzsee” [Lake Krampnitz] (Fig. 7b). The main reason for this substantial change could be attributed to removal of impermeable glacial tills below the canal bottom, which had so far impeded the vertical infiltration from the ND, and had made the nearby lake of same water level to deliver the water instead. After the reconstruction, part of the glacial tills has been removed or replaced by more conductive material and the direct infiltration from the canal became the origin of flow paths. In contrast, no main differences in subsurface capture zone and corresponding travel times were observed in the southern catchment. Overall, the increased hydraulic connection between canal and aquifer caused a reduction in subsurface capture zone in the area north of the canal and shortened subsurface travel times from canal to pumping wells.
Simulation of heat transport
Differences in time shift (ΔΦ) and temperature amplitude (ΔA) between simulated and observed temperatures for the time period after river reconstruction for all observation wells
Reference peaking time
The observation wells W04HIGH, W04LOW and U100 were characterized by almost constant temperatures throughout the year, which is well represented by both simulation scenarios (Fig. S4e–g of the ESM). The constant signal was not expected to be influenced by the canal reconstruction, as the average yearly temperature (equal to long-term average in air/river temperature) is independent of the groundwater flow velocity. Mean simulated temperatures were 1.17, 1.12 and 0.70 K lower than observed temperatures in W04HIGH, W04LOW and U100 respectively.
The greatest deviation between observed and simulated temperatures of the model scenario including canal reconstruction occurred in W09HIGH (Fig. S4b of the ESM). The overestimated amplitude and underestimated arrival time of the temperature signal were caused by overestimated groundwater flow rates (advective heat transport). The model is not reproducing the slow flow regime in W09HIGH, indicating that the hydraulic connection between the canal and the shallow aquifer is weaker than incorporated in the current geological model. Thus, the heat transport simulations here could indicate and rectify a local refinement of the geological model to be appropriate.
In comparison, the reconstruction scenario (Fig. 9b,d) shows a more continuous area of larger temperature variation range at the southern river bank induced by the annually cycling canal temperatures which represents a more direct hydraulic contact caused by the canal bed reconstruction. The snapshot of the vertical temperature distribution at 30th June 2014 (Fig. 9e,f) illustrates that in the reconstruction scenario, the warm water from the canal could penetrate much deeper into the aquifer, almost reaching the model bottom, with less damping during its propagation, and a similar situation would be observed in winter.
Generally, the periodic temperature signal travelled further before it became flat, both in the vertical and horizontal directions. It is the evident that the reconstruction has driven substantial changes to water flow and heat transport at the surface-water/groundwater interface.
Insights from heat transport simulation into the geological model
Featured as a naturally occurring periodic signal with easy and rapid measurement, the time series of temperature shows greater sensitivity to variations during transport, above potential environmental noise, compared to conventional or pulse tracers (Constantz 2008; Cox et al. 2007; Mahinthakumar et al. 2005). Research by Munz et al. (2017) illustrated that by taking heat transport into consideration, the model non-uniqueness could be better constrained and the accuracy of flux estimation across the interface between surface water and groundwater could be improved. Delsman et al. (2016) showed that conditioning the model to temperature measurements could worsen other model results such as electrical conductivity and salinity but also recognized that this could be mainly caused by errors in the model structure itself. However, few studies so far have used heat as a natural tracer for the numerical modelling of bank filtration. One of them is Henzler et al. (2016), who especially investigated its details with 3-D modelling of flow and heat transport in areas with geological heterogeneity.
By employing a 3-D numerical model for transient water flow and heat transport incorporating the heterogeneous geological structure, it is only partly possible to match both the seasonal temperature patterns and water heads at the same time. However, substantial deviations can be used to trace back shortcomings of the model structure, and heat transport can help as an additional test to evaluate model performance and realizing model limitations. The simulation results at W09 showed the lowest RMSE and the highest Pearson’s correlation coefficient (see section ‘Steady-state flow modelling’), but also substantial deviation between observed and simulated temperatures (see section ‘Simulations of transient flow over 4 years’). The poor match in the temperature pattern of W09 indicated an overestimation of water flux, which could not be compensated by reasonably adapting the thermal parameters (Fig. 9d); i.e. the hydraulic connections in the geological structural model, especially between the canal and the two aquifer layers leading to W09, are not representing the real situation well, and could be adapted based on this insight resulting from temperature observations and heat transport modelling.
Influence of river reconstruction on redox reaction
The quality of water gained via bank filtration strongly depends on the concentration of surface-water-born substances and pollutants, the subsurface capture zone, the mixing ratio between river water versus ambient groundwater, and the spatial and temporal distribution of subsurface travel times (Hiscock and Grischek 2002). Both the increased infiltration rate along the canal and shorter subsurface travel times may have implications for the occurrence, fate and distribution of biodegradable compounds, other DOC or even pathogens etc. in the adjacent aquifer. Generally, removal efficiency (purification capacity) of the underground passage will be diminished by shorter path lengths, higher groundwater flow velocities, and accompanying shorter travel times. Resulting concentrations of surface-water-borne substances and pollutants within the aquifer, as well as the mobility of these substances, might increase, especially for very quick flow paths. However, a sufficiently long setback distance could compensate this by supplying enough time for the adaption of biosystems and other slow processes (Schoenheinz and Grischek 2011). The results of this study indicate that a change in the heat transport by reconstruction can very well also be relevant for pollutant transport to the waterworks, especially for the canal bottom which acts as a sensitive interface between the groundwater and surface water. Generally, the periodic temperature signal, and finally the summer and winter temperature extrema in groundwater increased and penetrated deeper into the aquifer. The potential for attenuation and degradation (removal efficiency) of biodegradable compounds and other DOCs depends on the subsurface temperature, redox state, and the evolving microbial community; all of them are site-specific and strongly vary in time. Especially, temperature is recognized to control microbial processes in saturated sediments (Munz et al. 2019; Acuña et al. 2008; Thamdrup and Fleischer 1998). The more pronounced temperature variation within the aquifer may thus lead to changes in the redox zonation. The oxic zone may reach deeper into the aquifer by a reduced microbial activity in winter (colder temperature); vice versa, the penetration depth of the oxic zone may be reduced by the increased microbial activity in summer (higher temperatures). However, the aforementioned variation might be offset by the increasing flow velocity in summer, and intensified in winter. With short travelling distance (13~16 m), the change may be observed in the near-bank wells of W07, W08 and W09 but to different degrees, due to influence from geological heterogeneity.
With an intensive and long-term monitoring programme, it was possible to observe the seasonal behaviour of a bank filtration system, as well as its development, during and shortly after the reconstruction of the ND canal as the main supply of bank filtration water towards the pumping wells of the waterworks. Evidenced by the change of the groundwater heads in the aquifer close to the canal, the riverbed reconstruction by its artificial decolmation had substantially increased the hydraulic connection between the canal and the aquifer. The change started to take effect directly with the beginning of reconstruction, but intensified during the reconstruction period and had a clear effect after it was accomplished. Generally, this effect lasted for at least a year, until the end of this observation period, except for W07. Additionally, to these qualitative conclusions, a much more detailed and quantitative interpretation was enabled by employing a 3-D numerical model for transient flow. Initially, a steady-state model supplied a calibrated catchment-scale flow field and checked the feasibility and consistency of the geological model. Based on the set of calibrated parameters, a transient model was implemented that allowed for the development of the bank filtration groundwater system to be tracked over several years, including the reconstruction period. It showed to be a useful tool to quantify the effect of the reconstruction by directly comparing two model scenarios ‘with’ and ‘without’ reconstruction, which went far beyond just interpreting the measured time series. A key benefit is that this modelling approach eliminates the effects of changing hydrologic boundary conditions that otherwise would add a lot of uncertainty to the interpretation of the situation after reconstruction with inherently varying drivers hampering hard conclusions based merely on experimental data.
The increased hydraulic connection between the canal and the aquifer led to some spatial rearrangement of well catchments and an increase in the already high share of bank filtration water abstracted by the waterworks. The bank filtration rate along the canal has increased on average by 23%. Correspondingly, the increasing flow rate substantially changed the subsurface travel time distribution towards shorter travel times. Flow paths with travel times below 200 days increased by around 10% and those below 300 days by 15%. The potential of biodegradable pollutants and other DOCs to penetrate into the aquifer may thus be increased, especially in winter, due to the joint hydrological and thermal effects caused by the river reconstruction. Nevertheless, a significant deterioration of water quality due to the canal reconstruction was not observed so far in the waterworks but could well be the case at other locations.
Calibration performance limited the model accuracy on reflecting the flow field and thermal field. It could be attributed to both (1) the insufficient or incomplete interpretation of the geological conditions and (2) the calibration accuracy. Heterogeneity brought great difficulty and uncertainty in concluding on the geological condition, and further detailed investigation would be required, especially on the area where the modelling result deviated largely from the observation. Autocalibration may also reach a better parameter composition for the model, however, due to the large calculation load, it could be applied only for the steady-state model. Other simplifications and assumptions might also contribute to the inaccuracy in the model simulation, e.g. the fixed groundwater recharge, or not considering the influence from temperature on the viscosity, which would require another set of underlying equations and thus modelling environment. Pumping well settings in the model might also contribute to the possible error in calibration. As the current pumping rate distribution in a well is decided by the Kf-value of different layers, a clogging of the well screen may modify the real situation to show a different local flow field. Logger-recorded water head and biweekly sampling frequency for temperature was sufficient for describing the temporal dynamic. The number of wells and their distribution limited the spatial resolution of the study, but this will be the rule rather than the exception at other sites, too. Multi-level sampling could at least improve the vertical sampling resolution of both flow and thermal field in a given sparse set of groundwater observation wells.
The investigations of bank filtration and particularly the numerical modelling allowed for a quantitative assessment of the hydraulic impacts of river reconstruction within a dynamic, seasonally and annually changing hydrological system. Furthermore, the modelling was able to show how far the heat patterns penetrate into the aquifers during bank filtration and leave a seasonal imprint, even stronger after the canal reconstruction and a resulting higher hydraulic connection. Last but not least, such quantitative information is also required as the basis to interpret the fate of pollutants, e.g. micropollutants (Munz et al. 2019), during bank filtration, including their sensitivity to temperature and travel time distributions.
We gratefully acknowledge Dorit Tscherner and Karsten Zühlke from Stadtwerke Potsdam GmbH for supporting this study and providing geological information and pumping data. Meteorological data were provided by the nearby Leibniz Institute for Agricultural Engineering and Bioeconomy (ATB). Also, we appreciate the support from the DHI-WASY user service on technical issues of FEFLOW.
The first author acknowledges the Helmholtz graduate research school GeoSim for funding the PhD scholarship. This work was supported by the Federal Waterways Engineering and Research Institute (BAW) in Karlsruhe, Germany.
- Constantz J (2008) Heat as a tracer to determine streambed water exchanges. Water Resour Res 44 (4)Google Scholar
- Doherty J (2015) Calibration and uncertainty analysis for complex environmental models. Watermark, Brisbane, AustraliaGoogle Scholar
- Domenico PA, Schwartz FW (1998) Physical and chemical hydrogeology, 2nd edn. JWiley, New YorkGoogle Scholar
- Escalante EF, Calero R, de Borja González Herrarte F, Sauto JSS, del Pozo Campos E, Carvalho T (2015) MAR technical solutions: review and bata base. MARSOL demonstrating managed aquifer recharge as a solution to water scarcity and drought; MARSOL Deliverable 13-1; TRAGSA, Madrid, SpainGoogle Scholar
- Freeze RA, Cherry JA (1979) Groundwater. Prentice-Hall, Englewood Cliffs, NJGoogle Scholar
- Gross-Wittke A, Gunkel G, Hoffmann A (2010) Temperature effects on bank filtration: Redox conditions and physical-chemical parameters of pore water at Lake Tegel, Berlin, Germany. J Water Clim Change 1:55–66Google Scholar
- Hiscock KM (2005) Hydrogeology: principles and practice, 2nd edn. Blackwell, Oxford, 544 ppGoogle Scholar
- Landesamt für Umwelt Brandenburg (2019) https://maps.brandenburg.de/WebOffice/synserver?project=Hydrologie_www_CORE&client=core. Accessed April 2019
- Massmann G, Nogeitzig A, Taute T, Pekdeger A (2008b) Seasonal and spatial distribution of redox zones during lake bank filtration in Berlin, Germany. Environ Geol 54(1):53–65Google Scholar
- NASA’s Land Processes Distributed Active Archive Center (LP DAAC) (2019) The ASTER global digital elevation model (GDEM) version 3 (ASTGTM). https://doi.org/10.5067/ASTER/ASTGTM.003
- Schubert J (2006) Significance of hydrologic aspects on RFB performance. In: Hubbs SA (ed) Riverbank filtration hydrology. Springer, Dordrecht, The Netherlands, pp 1–20Google Scholar
- Stonestrom DA, Blasch KW (2003) Determining temperature and thermal properties for heat-based studies of surface-water ground-water interactions. In: Stonestrom DA, Constantz J (eds) Heat as a tool for studying the movement of ground water near streams. US Geol Surv Circ 1260, pp 73–80Google Scholar
- Todd DK, Mays LW (2005) Groundwater hydrology. Wiley, Hoboken, NJGoogle Scholar
- VDI 4640/1 (2010) VDI 4640-thermal use of underground, Blatt 1: fundamentals, approvals, environmental aspects. Verein Deutscher Ingenieure, Düsseldorf, GermanyGoogle Scholar
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.