Pilot-scale field study for vanadium removal from mining-influenced waters using an iron-based sorbent

This study investigated the removal of vanadium from mining waters at a closed mine site (Mustavaara, Finland) using granular ferric oxyhydroxide (CFH-12) on pilot scale. Two filter systems, pilot A and pilot B, were placed in different streams, where the influent in pilot A contained a higher and very variable vanadium concentration (6.46 – 99.1 mg/L), while the pilot B treated influent had lower vanadium concentrations (0.443 – 2.33 mg/L). The operation periods were 51 days for pilot A and 127 days for pilot B. Water quality analyses revealed that va- nadium was efficiently captured in the filter system in both pilots. X-ray fluorescence analysis revealed that the filter beds were not fully saturated with vanadium. X-ray photoelectron spectroscopy confirmed that oxidised vanadium (5 + ) existed in the used CFH-12 and the carbon content in the used material had increased due to the adsorbed organic compounds. For comparison, lab-scale coagulation experiments were conducted using ferric sulphate for the influent of pilot A (the sampled batch contained 15.9 mg/L V). The optimum coagulant dosage was 350 mg/L ( > 93% vanadium removal) at the original


Introduction
Vanadium has been extensively used as an additive in the steel industry thanks to its particular physical properties such as high tensile strength, hardness and fatigue resistance (Moskalyk and Alfantazi, 2003). The worldwide production of vanadium has approximately doubled in the past 20 years to 76,600 tons/year in 2017 (U.S. Geological Survey, 2018). However, vanadium is a well-known toxic metal and can exist in different oxidation states, with the pentavalent (V 5+ ) form the most common and stable in aqueous solution. Vanadium has been included in the contaminant candidate list (CCL4) by the United States Environmental Protection Agency (USEPA, 2016). The California Office of Environmental Health Hazard Assessment proposed a notification level of 50 µg/L for vanadium in drinking water (Fakhreddine et al., 2019). It has been confirmed that a higher concentration of vanadium can impair plants through chlorosis and stunted growth (Kabata and Pendias, 1984). Short-term and long-term exposure to vanadium has caused various toxic effects on animals, such as haematological and biochemical changes, abnormalities in reproduction or hepatotoxicity (Castellini et al., 2009). In addition, occupational exposure to vanadium can alter neurobehavioural outcomes for humans (Li et al., 2013). In recent decades, vanadium consumption has dramatically increased. As a consequence, the over-limit discharge of vanadium into water bodies has raised growing concern. Elevated vanadium concentration in groundwaters has been reported in many places (Arena et al., 2015;Meng et al., 2018;Zhang et al., 2019aZhang et al., , 2019b. Therefore, the need for the removal of vanadium from different waste streams has become an essential issue. Many methods have been investigated for treatment of vanadium-contaminated waters, such as ion exchange (Chen et al., 2020;Dabrowski et al., 2004;Keränen et al., 2015), adsorption (Leiviskä, 2021), membrane filtration (Chen et al., 2020;Lazaridis et al., 2003;Melita and Gumrah, 2010) and bioreduction Shi et al., 2020;Wang et al., 2021;Zhang et al., 2021).
Among the different treatment methods, adsorption technology has been regarded as a cost-efficient method for many types of wastewater. Various materials have been reported to effectively adsorb vanadium from aqueous solutions, such as activated carbon (Doǧan and Aydin, 2014), clay materials (Zhu et al., 2018), biomass waste (Kaczala et al., 2009) and different types of functionalized low-cost materials (Bello et al., 2019;Bhatnagar et al., 2008;Liao et al., 2008;Zhang et al., 2019b;Zhang and Leiviskä, 2020). In addition, iron materials have been widely used for water purification due to their large surface area and abundant surface hydroxyl groups. Commercial iron products such as GFH/GEH 101 (Fe(OH) 3 and β-FeOOH), E-33 (α-FeOOH) and  have been successfully used for vanadium removal from aqueous solutions (Lazaridis et al., 2003;Leiviskä et al., 2017aLeiviskä et al., , 2017bNaeem et al., 2007). Lazaridis et al. (2003) reported that vanadium sorption capacity onto GFH was 100 mg/g. Naeem et al. (2007) observed a similar capacity (107.80 mg/g) onto GFH, but a lower capacity was achieved onto E-33 (25.20 mg/g). Leiviskä et al. (2019) found that the maximum sorption capacity was 34 mg/g onto CFH-12 and 22 mg/g onto GEH 101. Besides synthetic solutions, removal of vanadium from real waters by CFH-12 was also tested by Leiviskä et al. (2017aLeiviskä et al. ( , 2017b where the vanadium removal was over 95% from vanadium-polluted natural water (dosage 2 g/L; initial V 1.2-1.9 mg/L) and close to 100% from industrial wastewater (dosage 10 g/L; initial V 52.2 mg/L), respectively.
Although vanadium removal by adsorption has been widely investigated and also a few studies have investigated the removal of vanadium from real industrial waters, most of these studies are restricted to laboratory scale. In general, lab-scale studies provide the opportunity to optimise adsorption parameters and investigate uptake mechanisms in more detail. However, field studies are essential for evaluating the performance of the system in realistic environmental conditions, for example when there are fluctuations in water quality and particularly in target species concentration. Long-term continuous-mode field studies have rarely been reported. This study provides new information about the suitability of granulated iron sorbent to remove vanadium from mining-influenced water streams in the field. Pilot filter systems packed with iron sorbent were placed to at a closed vanadium mine site to treat two different water streams (referred to as A and B). CFH-12 was selected for this study because of its good vanadium sorption in a wide pH range (3− 9) and its regeneration possibility (Leiviskä et al., , 2017a. Water quality, in terms of vanadium concentration, pH, conductivity and turbidity, was monitored during the operation and the used sorbents were characterised for elemental composition, surface area, pore size, mineralogy, particle size and chemical bond information. For comparison, lab-scale coagulation experiments were conducted on the mining water of pilot A using ferric sulphate, providing valuable information for evaluating material efficiency.

Pilot experiments
The pilot studies were conducted at the closed Mustavaara mining site. The Mustavaara V-Fe-Ti oxide deposit is located in northern Finland and vanadium occurs in magnetite mineral with a grade of 0.91 wt% (Karinen and Törmänen, 2016). Reopening of the Mustavaara mine is possible in the future. The physical and chemical properties of CFH-12 are presented in Table 1. Two filter boxes (pilot A and pilot B) were packed with 25 kg of CFH-12 and rinsed with tap water for a few hours (referred to hereinafter as fresh CFH-12). Pilot A was installed next to a ditch carrying the water away from the mining area. Pilot B was installed next to a small natural pond into which the water flowed from the mining area (including the ditch water from pilot A). The operation period for pilot A was from 10th June to 30th July 2019 (51 days), while pilot B was operated for a longer period, 10th June-14th October 2019 (127 days).

Design and operation
The pilot systems (pilot A and pilot B) used in this study were the same as those used in the previous work by Postila et al. (2019). The experimental setup of pilot A is shown in Fig. 1 (Pilot B had the same setup but was installed in a different place). The system included a pump, water container (90 L) and filter box. The container was placed above a plastic box to ensure a height difference between the filter box so that the water flowed into the filter box via gravity. The water container and the filter box were connected to each other through an inlet tube (inner diameter 3 mm). The plastic filter box (55 × 35 × 30 cm) was divided into three sections: the influent, filter and effluent sections. Each section was isolated by a perforated plastic wall. The filter section was filled with CFH-12 in a layer measuring 35 × 35 × 15 cm. The pump was connected to a timer, which was timed to pump water every five hours from upstream of the ditch to fill the water container. The influent flow rate was controlled by the water level height in the container. The flow rate decreased with decreasing water level height and reached maximum value again when the container was filled up during the pumping period. The pump was protected by a net with a fine mesh (diameter ~1 mm) to prevent larger particles from entering the pump. The filter box was placed on a slight longitudinal slope to ensure that water flowed through the filter material via gravity. The treated water was then routed to the effluent section of the filter box and finally discharged downstream of the ditch via the outlet pipe.

Water sampling and analyses
The redox and temperature of the water streams were measured using a handheld pH/mV metre (IQ Scientific ISFET) and thermometer on each sampling date. The redox potential varied between 273 and 350 mV in the pilot A stream and 275-530 mV in the pilot B stream, which reflects the systems' contact with the atmosphere (readings were converted to the standard hydrogen electrode). The temperature in the pilot A stream was between 9 and 17 • C (June-July) and 0-17 • C in the pilot B stream (June-October). Water samples were taken approximately once per week at the beginning and once every two weeks during the later stages of the studies. The influent sample was collected from the water container and the effluent water was collected from the outlet water (without taking the residence time into account).
The influent flow rate in the filter box was measured using a volumetric flask (average of three repeats). The average influent flow rate in pilot A was 83 mL/min and 229 mL/min before and after pumping water in the water container, respectively. The flow rates in pilot B were 86 mL/min and 189 mL/min before and after pumping, respectively. The flow rate variations were stable during the operation periods.
The influent and effluent pH were measured with a pHenomenal®  pH 1000 L pH metre (VWR). A Mettler Toledo conductivity metre and a Hach 2100Q Portable turbidity metre were used to measure the conductivity and turbidity of the samples. Vanadium concentration was analysed by inductively coupled plasma optical emission spectrometry. Detailed water quality analyses were performed on the selected dates (day 1 and day 33 in pilot A; day 1, day 33 and day 93 in pilot B). Elements were analysed using inductively coupled plasma mass spectrometry. Ion chromatography was used to measure sulphate, chloride and fluoride. Ammonium, nitrate, nitrite and phosphate were measured with a continuous flow analyser. The chemical oxygen demand (COD) was analysed using a Hach Lange cuvette (LCK 314 15-150 mg/L).

Characterisation of the fresh and used sorbent
The used CFH-12 after the pilot study was sampled at different points in the filter beds (A and B). Ten samples were taken from both pilots, which included five samples from the top layer (a depth of ~1 cm) and five samples from the bottom layer (a depth of ~8 cm). In each layer, the used sorbents were sampled from close to the corners (~5 cm from the corners) and one sample was taken from the middle. Samples taken from the top-left corner were numbered 1 (closer to the influent section) and the other corners were marked 2-4 in anti-clockwise order. The middle sample was referred to as 5. The collected samples were rinsed with Milli-Q ultrapure water and dried at 60 • C for 24 h. The rest of the used CFH-12 sorbent was mixed and dried at 60 • C (referred to hereinafter as mixed sample).
The chemical composition of the used CFH-12 samples (collected from different points in the filter beds) was determined using an X-ray fluorescence (XRF) spectrometer (Bruker AXS S4 Pioneer). For the XRF analysis, the samples were ground into a fine power and a pressed pellet was prepared using boric acid as binder under a hydraulic pressure of 10 metric tons before analysis. The X-ray diffraction (XRD) analyses were performed with a Rigaku Smartlab rotating anode diffractometer using Co Kα radiation for fresh CFH-12 and used CFH-12 (collected from different points in the filter beds). The ground samples were measured at room temperature in the 2-theta range from 5 • to 120 • with a step size of 0.02. The diffraction patterns were analysed using PDXL2 (Rigaku) analysis software with an ICDD PDF-4 database. X-ray photoelectron spectroscopy (XPS) was also conducted on the ground samples (fresh and used CFH-12; used CFH-12 collected from different points in the filter beds) using a Thermo Fischer Scientific ESCALAB 250xi with a monochromatic Al Kα source (1486.6 eV). The charge correction was performed by setting the binding energy of adventitious carbon to 284.8 eV.
The surface area and pore size distribution of the used CFH-12 (mixed sample) from both pilots were measured with an ASAP 2020 surface area and porosity analyser (Micromeritics). The surface area was measured by the Brunauer-Emmett-Teller (BET) nitrogen adsorption technique. Before analysis, the samples were outgassed in a vacuum for 12 h at a temperature of 100 • C. Two repeats were done for each pilot. The pore size distribution was measured by analysing the desorption branches of the isotherm using the Barrett-Joyner-Halenda (BJH) method. To evaluate the influence of the treatment process on the CFH-12 particle size, the fresh CFH-12 and used CFH-12 from both pilots (mixed sample) were fractionated into particle size intervals of 1-2 mm, 0.5-1 mm and < 0.5 mm by sieving tests using Retac 30 (Germany). The sieving tests were performed in duplicate.

Geochemical modelling
The PHREEQC geochemical modelling programme (Parkhurst and Appelo, 1999) was used to evaluate the vanadium speciation, mineral saturations, and saturation index (SI) for the ten water analyses of the influent and effluent solutions. The calculated Eh from NO 2 -/NO 3 was used as the pe in the PHREEQC modelling of speciation and mineral saturation. Additions of Na and/or Cl were used to charge balance the solution. The nitrate and nitrite were below the detection limit, so the calculated pe value for the pilot B sample at 93 days was used. The Minteq.V4 database was used for the calculations.

Coagulation experiments
Coagulation experiments were performed by using a Kemira flocculator 2000 jar test apparatus with a 1 L glass beaker. Each beaker was filled with 800 mL mining ditch water (the same ditch as where pilot A was installed). Ferric sulphate (PIX-322, 12.5 wt% Fe 3+ , Kemira Oyj) was used as a coagulant. The effect of dosage (50-600 mg/L) was studied first without pH adjustment (the initial pH of the water was 7.8-7.9). Then the effect of dosage (50-300 mg/L) was studied after adjusting the pH with HCl (Sigma-Aldrich), so that the pH was 4.6-4.8 during the coagulation stage. After adding the coagulant, the water was rapidly mixed (300 rpm) for 30 s and then slowly mixed (50 rpm) for 25 min. The coagulation pH was measured after five min of slow mixing using a Metrohm 744 pH metre. After the slow mixing stage, the sample was left to settle for 30 min before the supernatant was sampled at a depth of four cm below the water surface for analysis. The residual vanadium concentration was measured by the phosphorus-tungstenvanadium acid spectrophotometry method (Li, 2011) with a Shimadzu spectrophotometer (UV-1800). Turbidity was measured with a HACH 2100Q portable turbidimeter. The total surface charge was analysed with a Mütek particle charger detector by titrating 10 mL of the sampled supernatants with cationic Poly-Dadmac or anionic PesNa titrants. The ultraviolet (UV) absorbance at 254 nm (UV 254 ) was determined using a Shimadzu spectrophotometer (UV-1800). Prior to the UV 254 analyses, the supernatants were filtered with a 0.45 µm filter membrane (VWR, polyethersulphone membrane).

Vanadium removal
The vanadium concentrations in the influent and effluent samples are shown in Fig. 2. The influent concentration in pilot A fluctuated strongly during the operation period, between 6.46 and 99.1 mg/L (Fig. 2a). This can be attributed to the variation in rainfall precipitation, which might have affected the vanadium concentration in the ditch water. Moreover, the ditch was closely connected to the mining area and thus showed higher concentrations of vanadium and other elements, whereas the natural pond (pilot B) received many streams from the area and the concentrations of elements were at a lower level. In pilot B, the vanadium concentration in the influent varied between 0.443 and 2.33 mg/L (Fig. 2b), which was more stable compared to pilot A.
The operation time of pilot A was 51 days and the effluent vanadium concentration was relatively low (0.62-1.83 mg/L) from day one until day 19. After that a higher vanadium concentration was observed in the effluent on day 26 and day 33. This was probably due to high influent concentrations and therefore the impaired ability of the filter bed to adsorb all vanadium species. At the end of the operation period, vanadium concentration remained at a lower level in the effluent. Pilot A was stopped on day 51, because the pumping no longer worked reliably due to the increased solids concentration in the ditch water. This might have caused the partial clogging of the pump.
Pilot B had a longer operation time (127 days) and the pump worked smoothly during the whole operation period. In pilot B, vanadium concentration in the effluent samples was mostly below 0.5 mg/L, indicating high removal rates. However, a higher vanadium concentration was detected on day 59 (1.04 mg/L), and then some higher concentrations were observed at the end of the operation period.
The variation in effluent vanadium concentration in both pilots can be attributed to the fluctuating influent vanadium concentration and flow rate variation. Just after pumping when the water container was full, the water flow rate was at its maximum and thus the contact time in the sorbent bed was at its lowest. A shorter contact time resulted in somewhat lower vanadium removal efficiency. Nevertheless, the water flow variation in the pilots mimicked a real situation where rainfall variation may cause a variation in water flow if the filter systems were installed in a ditch. The low temperature in the pilot B at the end of the operation period might also have affected the kinetics (e.g. 0 • C on day 127 in pilot B) and thus vanadium removal efficiency.
The kinetics of vanadium sorption from synthetic vanadium solution onto CFH-12 was reported in a previous study . The sorption process was fast in the early stage, when the reaction was mostly based on the interaction of vanadate ions with surface hydroxyl groups. Then the sorption process became slow and governed by pore diffusion . Several researchers have reported the sorption of different anionic pollutants onto iron sorbents, and similar results were observed, i.e. that the sorption mechanism included fast sorption followed by slow diffusion into the pores (Banerjee et al., 2008;Luengo et al., 2006;Zhang et al., 2019b). In addition, according to Leiviskä et al. (2019), the sorption kinetics of vanadium onto CFH-12 was best described by the Elovich model, which is known to describe chemisorption (Tran et al., 2017).

pH, conductivity and turbidity
The influent and effluent pH values are shown in Fig. 2c (pilot A) and Fig. 2d (pilot B). The influent pH in pilot A remained quite stable, at between 7.3 and 7.7, and varied slightly within the range of 6.4-7.8 in pilot B during the operation period. In both pilots, the effluent pH was at a lower level at the beginning of the operation period, and then gradually reached the influent pH level. Leaching of impurities from the sorbent (discussed below) was the probable cause of the decrease in the pH of the treated water, and it occurred despite the fact that the sorbent was rinsed for a few hours. A similar result was also reported by Leiviskä et al. (2017a), in which a drop in pH was observed in both batch and pilot experiments using the same CFH-12 sorbent.
The influent conductivity varied in the range of 379-1086 µS/cm in pilot A (Fig. 2e), whereas a lower influent conductivity (41.1-79.7 µS/ cm) was observed in pilot B (Fig. 2f). However, the effluent conductivity increased in both pilots, with the highest values observed on the first day. In pilot B, the effluent conductivity reached a low level quickly, whereas the effluent conductivity in pilot A remained at a higher level and varied between 690 and 1181 µS/cm after the first sampling day. This increase in effluent conductivity can be attributed to the leaching of impurities.
Influent turbidity values in pilot A (15.9-54.6 NTU) were at a higher level compared to pilot B (3.5-32.5 NTU) ( Fig. 2g and h). The ditch where pilot A was installed contained more suspended solids, which resulted in turbid water when mixing occurred, for example when reindeer crossed the ditch or the wind blew more strongly. In pilot A, a decrease in effluent turbidity was generally observed with few exceptions. In pilot B, effluent turbidity values were always lower or at a similar level to the influent turbidity values. It is also noteworthy that influent and effluent turbidity values followed each other closely in pilot B.

Detailed water quality
The influent and effluent water quality was characterised in detail on the selected dates in both pilots; the results of selected parameters are shown in Tables 2 and 3 (complete data in Tables S1 and S2). Graphs of the elements with mg/L level concentration are shown in Fig. S1. Besides vanadium, both influents contained a considerable amount of aluminium and iron, a small amount of manganese and zinc, and trace amounts of other elements. Anions including sulphate, chloride, nitrite, nitrate, fluoride and a small amount of phosphate were also found in both influents, among which the concentrations of sulphate and chloride in the influents were at a relatively higher level. In general, the influent in pilot B had lower ion concentrations than the influent of pilot A. COD was measured only for the first influent samples. Pilot A had a higher COD (71 mg/L) than pilot B (33 mg/L).
In pilot A, the Al and Fe concentrations in the effluent were at a significantly lower level (> 90%) than in the influent on the first operation day, while on day 33 the concentration of Al was only slightly lower and the concentration of Fe was clearly higher in the effluent than in the influent (Table 2). In the effluent samples, Mn, Zn, NO 2 and other elements were generally present in lower concentrations. Elements such as Ba, Cu, Mo and U were constantly at lower levels in the effluent samples. However, a significantly higher concentration of SO 4 2was observed in both effluent samples. In addition, the concentrations of NO 3 -, Ni and B were also at a higher level in both the analysed effluent samples, although on day 33 the concentrations were not significantly higher. Higher Cl -, NH 4 + and Sr concentrations in the effluent were observed on the first operation day but this was no longer the case on day 33. In pilot B, the effluent samples showed lower concentrations of Fe, Al, Cu and Ba, as presented in Table 3, although their concentrations in the influent and effluent were at quite the same level in the later stage. The concentration of SO 4 2and Clsignificantly increased in all of the analysed effluent samples, whereas the effluent concentrations of NO 3 -, NH 4 + , Sr, B, Ni, Mn and Zn were at a clearly higher level only on the first operation day. Other elements either existed at a very low level or remained quite stable.
Although a good performance of CFH-12 has been reported in previous studies (Biela et al., 2017;Funes et al., 2018), CFH-12 leaching issues have been rarely studied. Backman et al. (2007) observed that the mean concentrations of sulphate, cobalt, nickel, calcium and manganese clearly increased in the treated groundwater when using CFH-12. Also, leaching of impurities from CFH-12 was reported by Leiviskä et al. (2017a) in the treatment of real industrial wastewater formed in the oil gasification process. Sulphate leaching was confirmed to be mainly due to the gypsum impurity contained in CFH-12. The same batch of CFH-12 was also used in this study, and therefore a significant leaching of sulphate was observed. Although this study showed leaching of Cl -, NO 3 -, Ni, B and Sr in both pilots, Leiviskä et al. (2017a) observed Ni sorption and no significant change in Clconcentrations in the pilot study with CFH-12. The difference can be attributed to the different water quality (pH, speciation and concentrations). Also, the temperature was much higher (80 • C) in their study (Leiviskä et al., 2017a) than in the present study. Nevertheless, leaching issues can be prevented by rinsing the fresh sorbent thoroughly. In addition, if the sorbent is regenerated and used many times, leaching is no longer a problem. The PHREEQC geochemical modelling programme (Parkhurst and Ferrihydrite and amorphous aluminium hydroxide were supersaturated (log SI = 4.6 and 0.9, respectively) in the influent and slightly supersaturated (log SI = 0.6 and 0.1, respectively) in the effluent for day one in pilot A. Assuming that the reduced concentration from influent to effluent is due to mineral precipitation, this would release 3 * 10 − 4 mole of H + following these mineral reactions: This would result in a pH of 3.0 if there were no other sinks for the hydrogen ions in pilot A on day one. The buffering capacity of the solution together with the potential adsorption of Fe 3+ and Al 3+ prevented the pH from going more acidic than the measured 5.8 in the effluent on day one. All of the samples were within the stability field of ferrihydrite (Fig. 3).
On day 33 (in pilot A), the influent concentration of aluminium and iron was much lower than on day one and the concentration of Al and Fe was approximately the same as in the influent solution. There was no reduction in pH from the influent to the effluent solutions. The concentration of Al and Fe, however, was three to four times higher in the effluent sample from day 33 compared with day one. Amorphous aluminium hydroxide was at saturation, while the ferrihydrite was supersaturated (log SI = 4.5). This lack of reduction could indicate that the sorption capacity had reached its limit.
Pilot B showed a similar loss of Al and Fe as in pilot A on day one. If Al and Fe precipitated as ferrihydrite and aluminium hydroxide, this would result in pH values of approximately 3.8, 5.0 and 5.5 for pilot B on days 1, 33 and 93 days, respectively. The effluent pH was 5.3, 6.6 and 6.7, i.e. lower than the influent solution, but higher than without a neutraliser and/or sorption of these constituents.
Vanadium occurred primarily as vanadate (V) anions: H 2 VO 4 -, 3-, HVO 4 2and constituted > 95% of the vanadium ions for all 10 analysed water samples. There were, however, some major constituents that were not analysed in these water samples, e.g. carbonate, calcium and magnesium.

Characterisation of the used sorbent
In both pilots, according to the XRF results (Tables S3 and S4), vanadium was detected in all of the sampled materials taken from the top and bottom layers of the CFH-12 beds. The vanadium content varied between 0.22% and 4.713% in pilot A and 0.017% and 0.396% in pilot B. In pilot A, it was observed that the top layer contained a higher amount of vanadium than the bottom layer. A similar result was observed in pilot B with one exception. The sample in position B-4 had a slightly higher vanadium amount in the bottom layer (0.038%) than in the top layer (0.034%). The highest vanadium content observed in pilot A sample (4.7%) was at a similar level to that observed in the pilot study made with industrial wastewater using CFH-12 (Leiviskä et al., 2017a), as the highest amount observed in the latter study was 4.2% V. The maximum vanadium sorption capacity was reported to be 34 mg/g onto CFH-12 with synthetic vanadium solution (vanadium concentration 10-300 mg/L; initial pH 6; contact time 72 h) , which corresponds to 3.4% vanadium content in the saturated sorbent. Three samples in pilot A contained a higher vanadium amount than the saturated sorbent. This might be due to the long operation period of the pilot tests or the inaccuracy of the XRF method. Apart from these three samples, the other analysed samples in pilot A and all the samples in pilot B contained less than 3.4% vanadium. The results indicated that the filter beds were not totally saturated by vanadium, especially in pilot B.
The amounts of magnesium and sulphur in the used CFH-12 were significantly lower than in the fresh sorbent, which is in agreement with Table 3 Influent and effluent quality for pilot B (selected) (complete table is presented in Table S2).  Fig. 3. Eh-pH phase diagram for the iron-sulphur system with the ten water analyses for sorption testing. Each mark type has the influent and effluent together. The plot is based on -Fe tot = 0.1 mmol; S tot = 1 mmol; temperature = 25 • C; and excludes magnetite, haematite and goethite from the calculation (plot produced with Medusa: 2016 from KTH, Royal Institute of Technology, Sweden). the water quality data, i.e. the sulphate leaching and high effluent conductivity values observed during the first day of operation (Mg was not analysed from water samples). The X-ray diffraction results confirmed that no new phases were formed during the pilot study and that goethite was the only phase observed in the used CFH-12 samples of both pilots (Fig. 4). The impurity gypsum had effectively leached from the filter beds. A slight increase in goethite crystallinity was also observed in the used CFH-12 samples of both pilots.
The elements detected in the ground fresh and used CFH-12 samples by XPS are presented in Table 4 (only two samples were analysed after the pilot study). The fresh CFH-12 contained Fe and O as major components, while small amounts of C, Si, S and F were also found. After treatment with mining waters, the atomic % of C 1s increased; the increase was more pronounced in the samples taken from the top. The proportion of C 1s increased more in pilot B than in pilot A, which could be attributed to the higher concentrations of humic substances in the natural pond. Sulphur was not detected in the used CFH-12, which is in agreement with the water quality data and is related to the fact that sulphate was efficiently leached during the operation periods. Besides, the content of O 1s slightly decreased during the treatment, which relates to the higher leachability of oxygen-containing compounds (e.g. SO 4 2-) than compounds that were adsorbed (e.g. vanadates and organics). After treatment, the V 2p peak was detected in the survey spectra, and the vanadium content in pilot A samples was about ten times higher than in pilot B samples. This result agrees with the XRF results.
Fresh CFH-12 presented three peaks in the C 1s spectra, related to C-C/C-H (284.8 eV), C-O (286.1 eV) and inorganic carbonate/bicarbonate (289.3 eV) (Heuer and Stubbins, 1999;Sun et al., 2006). C-C/C-H and C-O may have derived from adventitious carbon, whereas the inorganic carbonate/bicarbonate may have originated from the washing procedure. After treatment, three peaks were observed at 284.8 eV, 286.4-286.5 eV and 288.6-288.9 eV in the used CFH-12 (Fig. 5), which can be assigned to C-C/C-H, C-O/C-N and C-O-C/C--O, respectively. Moreover, the carbonate/bicarbonate probably presented a small peak at ~290 eV but it overlapped with the C-O-C/C--O peak. The increase of C 1 s (atomic %) mostly enriched the C-O/C-N and C-O-C/C--O peaks, originating from the adsorbed organic compounds from mining streams. The result was in good agreement with previous studies in that the organic compound may form strong multiple bonds with the functional groups on iron sorbents (Kaiser and Guggenberger, 2007;Leiviskä et al., 2017b).
Two peaks were observed in the O 1s spectra of fresh CFH-12 at 530.3 eV and 531.7 eV, corresponding mostly to the Fe-O bond and Fe-OH bonds. The used CFH-12 showed three peaks related to the Fe-O (529.9 ± 0.1 eV), Fe-OH/C--O (531.3 ± 0.1 eV) and C-O (533.0 ± 0.2 eV) bonds after treatment (Fig. 6). The new C-O peak originated from the adsorbed organic compounds. In addition, a V-O bond should exist at ~530 eV, having a similar BE value to the Fe-O bond (Ureña-Begara et al., 2017). The proportion of the -OH/C--O bond decreased in the used CFH-12, which may have been due to the reaction between the hydroxyl group and other impurities.
The used CFH-12 in pilot A presented V 2p 3/2 and V 2p 1/2 peaks at 516.9 eV and 524.6-524.8 eV (Fig. S2), suggesting oxidised vanadium (V 5+ ) was bound to the sorbent (Demeter et al., 2000). This is in agreement with previous studies . However, these peaks were very weak in the V 2p spectra of the used material in pilot B. This confirmed that a smaller amount of vanadium was captured in pilot B due to the lower vanadium concentration in the stream. In the Fe 2p high-resolution XPS spectrum of fresh CFH-12, Fe 2p 3/2 and Fe 2p 1/2 peaks were located at binding energies of 711.7 eV and 725.2 eV (Fig. S3). After the treatment process, the Fe 2p 3/2 and Fe 2p 1/2 peaks shifted 0.5-0.6 eV towards lower binding energies (711.2 ± 0.2 eV and 724.6 ± 0.2 eV), which might have originated from increased goethite crystallinity (Khalid et al., 2017).
The surface area, pore diameter and pore volume of the fresh CFH-12 and used CFH-12 are presented in Table 5. After the treatment, the BET surface area and pore volume were enlarged and a slight increase in average pore diameter was also observed. These changes may be attributed to the leaching of the impurities from the sorbent during the operation periods . Additionally, in pilot B, the BET surface area and pore volume of the used sorbent increased less than in pilot A. This was probably due to the higher concentrations of organic compounds in the pilot B stream and more organic compounds being adsorbed onto the sorbent. This was in agreement with the XPS result as the proportion of C 1s increased more in pilot B than in pilot A. Moreover, the particle size distribution of fresh CFH-12 was as follows: 83-86% particles 1-2 mm in size, 13-16% particles 0.5-1 mm in size and 1% particles < 0.5 mm in size. It was observed that the proportion of granules having a particle size of 1-2 mm increased after the field tests (Table S5).

Coagulation experiments
As significantly higher vanadium concentrations were observed in the pilot A stream than those observed during the preliminary sampling, it was decided to carry out coagulation experiments using an iron coagulant for comparison. In the sampled batch from the pilot A stream the vanadium concentration was 15.9 mg/L and the turbidity was 16.3 NTU. The effect of the coagulant dosage on vanadium removal is shown in Fig. 7a. Without pH adjustment (pH 7.8-7.9), a dosage of 350 mg/L was needed to remove 92% of the vanadium, the residual vanadium concentration being 1.2 mg/L. Some higher coagulant dosages resulted in even lower residual vanadium concentrations (~ 0.7 mg/L). Turbidity first increased with the increased coagulant dosage at the  lower dosage range (50-260 mg/L), as shown in Fig. 7b, but then decreased to a low level. A slight increase in turbidity was observed when the dosage was over 500 mg/L. Underdosing of coagulant may have caused the turbidity to increase due to the formation of unsettleable microflocs, formed in the initial phase of the coagulation. Turbidity can also increase when dosing is too high and restabilization occurs. The total surface charge (TSC) of the pilot A ditch water was − 126 µeq/L and thus had a moderate cationic demand. The TSC decreased as the coagulant dosage increased and was close to zero when the optimal dosage range (375-500 mg/L) was reached (Fig. 7c). The UV 254 value decreased to a very low value and remained stable in the optimal dosage range (Fig. 7d). After exceeding a dosage of 500 mg/L, the value of UV 254 started to increase, which follows the trend observed for turbidity. The TSC and UV data confirmed that organic compounds were removed in the coagulation and that the charge neutralisation mechanism played an important role in organics removal. Duan and Gregory (2003) proposed that charge neutralisation and adsorption on the amorphous hydroxide precipitate are the two main mechanisms involved in the removal of humic substances with metal salts. Vanadium is presumably adsorbed on the formed iron flocs while complexation with organics or direct reaction with dissolved cationic iron species might also occur (Leiviskä et al., 2017b;Roccaro and Vagliasindi, 2015). These results indicate that the optimal dosage range (375-500 mg/L) is a combination of vanadium removal efficiency, turbidity and UV 254 . A dosage of 375 mg/L was considered to be the optimal coagulant dosage in the pH range of 7.8-7.9. Besides coagulant dosage, coagulation pH is also one of the most important factors governing the coagulation process. The optimal pH range for iron-based coagulant was reported to be between 4.5 and 7 for the treatment of river water and municipal wastewater (Umar et al., 2016;Zhao et al., 2014). When the pH of the ditch water was adjusted to a lower level (4.6-4.8), a higher vanadium removal rate was achieved at a lower dosage (Fig. 7a). For example, vanadium removal efficiency was significantly enhanced to 82% at a lower pH (4.8) compared to only 2.6% at a higher pH (7.8) at the same coagulant dosage of 200 mg/L, which might be related to the higher affinity of vanadium to iron flocs at lower pH. The speciation of vanadium is strongly affected by the pH, and several species can coexist in the pH range of 4.5-7.0 (Guzmán et al., 2002;Liao et al., 2008). The higher affinity of vanadium species to iron sorbents at slight acidic pH values has been reported earlier (Zhang et al., 2019b). In addition to vanadium concentration, turbidity and UV 254 quickly reached lower values with the lower coagulant dosage, and the TSC was close to zero already at a dosage of 150 mg/L. This result suggests that the coagulation pH strongly affected vanadium removal alongside the turbidity, TSC and UV 254 .
This study confirmed that both the coagulation and pilot filter systems can decrease vanadium concentration to a low level. It is noteworthy, however, that higher material efficiency can be obtained in Fig. 6. O 1s spectra of the fresh and used CFH-12.

Table 5
Surface area, pore diameter and pore volume of fresh and used CFH-12 (mixed sample). coagulation. A dosage of 375 mg/L PIX-322 coagulant was required to remove 14.8 mg of vanadium (93% vanadium removal) in the coagulation test at pH 7.8-7.9. The capacity obtained by the coagulant (PIX-322) was 39.5 mg/g (dosage 375 mg/L, pH 7.9). A dosage of 6 g/L CFH-12 sorbent was required to remove 4.6 mg V (99% vanadium removal) from ditch water sampled from the same area (data not shown), which resulted in a capacity of 0.77 mg/g. Nevertheless, the filter system has other advantages such as easy scalability, low maintenance and more reliable operation even if the vanadium concentration varies widely. In addition, CFH-12 can be regenerated and reused , which promotes sustainability.

Conclusion
The pilot filter systems studied here efficiently removed vanadium from real mining streams although the high variation in vanadium concentration in pilot A resulted in a shorter operation period compared to pilot B. In both pilots, leaching of non-toxic impurities was evident on the first sampling day but then the leaching trend slowed down. The sorbent was stable during the operation period, although some crystallisation into goethite could be observed. XRF analyses revealed that the filter beds were not fully saturated with vanadium, and pilot B could have been operated for even a longer period if the weather had allowed it. The results demonstrate the potential of an easily operable filter system to treat a vanadium-containing mining stream, especially when the concentration of the target component fluctuates moderately. The coagulation method is recommended for vanadium removal from water containing a higher concentration of vanadium and natural organic matter.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. acknowledge the support from Maa-ja vesitekniikan tuki ry (Finland). A special thanks goes to Heini Postila at the Water, Energy and Environmental Engineering research units of the University of Oulu for providing the filter equipment.