Dynamic head loss in pilot- and full-scale wastewater sand filtration with phosphorus removal, denitrification, and associated bumping

Sand filters were constructed at the Henriksdal wastewater treatment plant (WWTP) in Stockholm, Sweden, in anticipation of new effluent standards for nitrogen, phosphorus, and BOD₇ concentrations. The design of the full-scale sand filters was based on pilot studies with smaller sand filters operated at the Henriksdal WWTP before the extension of the plant. Furthermore, the concentration of mixed liquor suspended solids (MLSS) in the aeration tanks was increased from 1,250 to 2,500 mg MLSS/L and three additional aeration tanks were also constructed to increase the volume of the activated sludge step. These measures increased the nitrification and denitrification capacities of the plant. It also amplified the risk of sludge overflow from the secondary sedimentation tanks. Therefore, three additional sedimentation tanks were also built. The Henriksdal WWTP is the largest underground WWTP in Europe excavated into rock. It has 870,700 individuals connected to its network and manages an average hydraulic load of 273,000 m³/d. The effluent standards for the mixed effluent from the two central WWTPs in Stockholm (Henriksdal and Bromma) are ≤10 mg total N/L as the annual average value, and ≤0.3 mg total P/L and ≤8 mg BOD₇/L as the quarterly average values since the two WWTPs have the same receiving water. The expected demands were ≤6 mg total N/L and ≤0.2 mg total P/L in the effluent. The average concentrations in the effluent from the Henriksdal WWTP were 8.4 mg total N/L in the year 2022, 7.8 mg total N/L in 2021, 8.0 mg total N/L in 2020, and 8.4 mg total N/L in 2019. Further nitrogen reduction is needed if 6 mg of total N/L or lower would be reached in the future. Addition of a dosage of a carbon source to the influent of the filters to perform denitrification will accomplish that. Otherwise, high concentrations of nitrogen and phosphorus in the filtrate would have resulted in eutrophication and increased the primary production of the receiving water.

Deep-bed two-media down-flow sand filters for particle separation were built as a final treatment step in the plant. Applications of sand filters might create different operational problems mainly connected to head loss in the filter beds. After 5 years of operation, the initial head loss in the filters started to increase. The head loss in a filter consists of a dynamic part and an initial part. The dynamic part of the head loss was studied, and it consist of several parts. First, accumulated suspended solids (SS) that originated from the secondary sedimentation tanks were transported to the filter and, second, SS originated from primary settled wastewater (PW) that bypassed the biological step at high hydraulic loads, and, third, SS from nitrogen gas, carbon dioxide bubbles, and from cell growth produced in the filter during denitrification. Initial head loss was due to inorganic clogging (Jonsson & Björlenius 2022). For efficient functionality of the filters, it is most important that a filter cycle between backwashings lasts as long as possible, or else the filters would be taken out too often for backwashing. Several studies have investigated head loss in sand filters. For example, Al-Jadhai (2003) performed a filter study with sand of Ø 2.0–3.36 mm, which gave an operation time of 100 h at 8 m/h and 53 h at 18 m/h at a maximum available head loss of 2.25 mH₂O. Boller et al. (1997) reported an operation time of around 24 h at a hydraulic load of 9.2 m/h and accumulation of 2 kg SS/m³ in a deep-bed two-media down-flow filter without a dosage of carbon source. Altmann et al. (2016) found, after 22 h when the filter reached clogging, a head loss of 170 mbar corresponding to 1.8 mH₂O in a down-flow filter with a bed height of 2.0 m.

Our paper compares the pilot-scale and full-scale studies performed at the Henriksdal WWTP and describes the dynamic head loss with and without denitrification at different filter bed depths and after different operational times between backwashings of the filters. It is based on data from the full-scale study performed during 7 months after the initial 4 years of continuous operation and on data from the prior pilot plant study performed for little more than 2 years. To build pilot-scale filters and study them before the full-scale filters are constructed gives a unique possibility to both correctly design the full-scale filters and afterwards, by a full-scale study, compare the results from the two studies and find out if a pilot-scale study is able to predict the performance of the full-scale filters.

Overall, 66 composite samples were collected from the influent wastewater entering the filter (marked Inlet in Figure 1) and from the filtrate (marked Sampling in Figure 1). The concentration of phosphate and total phosphorus in the influent and the filtrate were analysed with a filter photometer LP 2W (Dr Lange, Germany) using cuvette tests LCK348 and LCK349. Organic material was analysed as chemical oxygen demand (COD) with cuvette test LCK114 and LCK614. Nitrate (NO₃-N) was analysed with LCK339 and nitrite (NO₂-N) with LCK341. Iron (Fe) was analysed with LCK321 and ammonium (NH₄-N) with LCK303 and LCK304. The precision for LCK348 was ±0.048 mg P/L, for LCK349 ± 0.013 mg P/L, for LCK114 ± 19.5 mg COD/L, for LCK614 ± 4.33 mg COD/L, for LCK339 ± 0.50 mg N/L, for LCK341 ± 0.040 mg N/L, for LCK321 ± 0.05 mg Fe/L, for LCK303 ± 0.90 mg N/L, and for LCK304 ± 0.0135 mg N/L. Munktell MGC filters (pore size 1.2 μm) were used to filter the sample before the analysis of phosphate phosphorus (PO₄-P). SS were analysed by a gravimetric method according to the Swedish standard methods SS028112-3 and SS EN 872 using Whatman GF/C filters (pore size 1.2 μm). All samples were analysed in duplicate. Although the analysis with the photometer is not accredited, it has a superior advantage of the analysed samples being fresh as they are directly obtained from onsite and not analysed after around 24 h or more when they are sent to a laboratory that is not situated at the WWTP.

Sodium acetate (NaAc) was dosed as an energy and carbon source to accomplish denitrification in the filter. The dosages used were mainly 33–34 mg COD/L (18 experiments), but also 35–38 mg COD/L (three experiments), 50–51 mg COD/L (two experiments), and 67 mg COD/L (two experiments) were applied.

The Henriksdal WWTP has a biological treatment step with a combination of conventional activated sludge (CAS) and pre-denitrification biological nitrogen removal (BNR). The last particle separation step in the plant constitutes of 60 parallel deep-bed, two-media, down-flow sand filters with a total surface area of 3,600 m² divided into four operational groups with 15 filters in each group. Each filter has a length of 10 m and a width of 6 m, corresponding to a horizontal surface area of 60 m² per sand filter, Figure 1. The upper layer in the filters consists of 1.0 m of crushed ceramic grains with a grain size of Ø 2.5–3.5 mm, d₁₀–d₉₀, and the lower layer consists of 0.5 m of sand with a grain size of Ø 1.2–1.8 mm, d₁₀–d₉₀, and a particle density ρ = 2,650 kg/m³. The particle density of the wet ceramic grains is around 1,013–1,200 kg/m³. The dry ceramic grains have a particle density close to the density of water. The porosities of the ceramic bed and the sand bed are 0.45 and 0.40, respectively. The 2,880 nozzles in the filter bottom of one filter have 24 slots of 1.0 mm each.

The cleaning of the filters is done by up-flow backwashing with a combination of 20 m/h filtrate from the filter outlet and 30 m/h air, repeated in two sequences, i.e. cycles, after each other. The used backwashing water is drained from the surface of the filter through a funnel in the centre of the filter after each of the two backwashing cycles. This is followed by sorting of the grains of the filter bed material after density and grain size by fluidisation, i.e. expansion of the filter bed, with 90 m/h filtrate. This water is drained through the filter and out through the nozzles. The design maximum hydraulic flow to the filters is 10 m³/s. During daytime and dry weather, the flow is around 3.2 m³/s in the WWTP. Approximately 2 g Fe/m³ of ferrous sulphate is dosed to the filters and approximately 13 g Fe/m³ is dosed in the pre-precipitation step.

The contribution to the dynamic head loss from denitrification in the biologically active sand filter was studied via the pressure meters in the full-scale filter No. 60, which was fed with a constant hydraulic flow of 200 m³/h corresponding to a filtration rate of 3.3 m/h to the 60-m² filter. The pressure meters were first mounted at 0.20, 0.40, 0.60, 0.80, and 1.50 m down in the filter bed. The pressure meter at 0.60 m was later moved to 1.20 m, but soon after that, the pressure meter at 1.20 m was moved to 1.40 m. This was done in order to find the depth in the filter bed where the most dominating clogging occurred contributing to an increased dynamic head loss. The depth in the filter bed is related to the surface of the filter bed that is set to 0 m, which is at the top of the ceramic layer. Negative values represent points in the wastewater above the filter bed top and positive values are points situated down in the filter bed. The time before clogging was determined to be the time from the start of each filter cycle to the time when the head loss in the filter bed had increased enough so that the filter went into the queue for backwashing.

In order to achieve denitrification in filter No. 60, ethanol or methanol was dosed as an energy and carbon source for the heterotrophic bacteria. No carbon source was dosed for the first period from March 16 to April 8. Ethanol was dosed for the second period of 1.5 months from April 9 to May 25. Methanol was dosed for the third period of 1.5 months from May 25 to July 13. The dosage was varied between 13 and 37 mg COD/L for ethanol, and 25 and 71 mg COD/L for methanol. Ethanol was first chosen as a carbon source, as it is not poisonous and thereby will be easier to handle. It was also sought as a carbon source to accomplish denitrification as fast as possible. When the microorganisms were adapted to the carbon source, methanol replaced ethanol. Methanol is cheaper and can be bought in larger quantities. No decrease in denitrification activity was seen after the change from ethanol to methanol as the microorganisms were adapted to a carbon source.

In total, 79 composite samples were collected during the study for analysis of influent and filtrate. Inlet samples were taken in the influent channel (Figure 1) before the inlet and filtrate was taken from a small valve on the tube near the control valve (Figure 1). All samples were analysed at the accredited laboratory of Stockholm Water Co. according to Swedish standard methods. The accuracy for the analysis of (NO₃ + NO₂)-N is 5%, for NO₃-N is 5%, for total P is 5%, for PO₄-P is 4%, for COD is 3%, for SS is 30%, for Fe is 4%, and for NH₄-N is 2%. The characteristics of sand filter No. 60 are presented in Table 1.

A newly installed filter has a head loss of 0 m, 0 mH₂O. After some time, inorganic clogging gave a small initial head loss noticed at the start of every filter cycle. The head loss that increases during the time of operation of one filter cycle is the dynamic head loss. When that head loss is equal to the head that existed at the beginning of the filter cycle, part of the wastewater flow bypasses the filter until the flow reaches zero and the filter is completely clogged. Then, the filter must be backwashed. In reality, the full-scale filters are backwashed before it starts to bypass parts of the flow, due to settings in the control system allowing 80% filter effluent valve opening before backwashing. The level of operation in the full-scale filter is +2.70 m as the level meter in Figure 1 was calibrated to zero at 0.60 m below the filter bottom.

In the pilot-scale filter, the level meter detected the end of the cycle when the water level reached around 2.75 m. At a water level of approximately 2.9 m, the filter is clogged and the filter starts to bypass parts of the influent water (marked Overflow in Figure 1).

In Figures 5 and 6, the scales are turned upside-down, as the values are taken from manual readings of water columns in the pilot plant filter and directly from the SCADA system in the full-scale filter of the Henriksdal WWTP. Here, a clean filter has a head of around 2.5 mH₂O and a clogged filter has a head reaching 0 mH₂O.

From a hydraulic load of 10 m/h to the Ø 0.4 m pilot plant filter tube and Equation (2), gas production of 8.5 L N₂/h and 24.0 L CO₂/h could be calculated when 0.3 parts of the carbon source produced sludge. With the value 0.6 from Equation (3), gas production of 5.9 L N₂/h and 19.5 L CO₂/h could be calculated. The gas production refers to a gas with a temperature of +15 °C and a pressure of 101,325 N/m². The filter bed had a volume of 188 L and a pore volume of 82 L. The evolved gas volume corresponds to 2.4 and 2.0% of the pore volume, respectively. The gas is successively accumulated in the filter and after some time there will be too much gas presumably accumulated around or below the nozzles and the filter will be totally clogged.

From a hydraulic load of 200 m³/h to the 60 m² full-scale filter and Equations (4)–(7), gas production of 1.9 m³ N₂/h and 3.1 m³ CO₂/h could be calculated for methanol if 0.3 parts of the carbon source produced sludge. The same values for ethanol became 1.7 m³ N₂/h and 2.8 m³ CO₂/h. If 0.6 parts of the carbon source produced sludge, the gas production with methanol as a carbon source could be calculated to 1.6 m³ N₂/h and 2.5 m³ CO₂/h and with ethanol as a carbon source to 1.4 m³ N₂/h and 2.3 m³ CO₂/h. The filter bed had a volume of 90 m³ and a pore volume of 39 m³. The evolved gas volume corresponds to 1.7, 1.5, 1.4, and 1.3%, respectively, but accumulation will occur with time in the lower parts of the filter.

In pilot plant filter settings 6, 7, and 8, different grain sizes and sand layer heights were compared, Table 1. The denitrification study was performed in filter setting 7, although, filter setting 6 had a grain size of the sand closer to the grain size in the full-scale filters. The smaller grain size, Ø 0.8–1.2 mm, from filter setting 7, was first chosen for the full-scale filters but this was changed to a larger size, Ø 1.2–1.8 mm, just before ordering the sand to the full-scale filters. Heavier sand grains were required to be separated from the light ceramic grains after fluidisation. The filtrates from the pilot-scale filters were analysed for little more than 2 years and a selection of the results is presented in Table 2.

When comparing filter setting 7 (Ø 0.8–1.2 mm) and filter setting 6 (Ø 1.2–1.5 mm) in period 1, it is obvious that there is not any significant difference between the concentrations of phosphorus in the filtrates, both having a concentration of 0.29 mg total P/L. In period 2, filter settings 7, 6, and 8 (Ø 0.6–0.8 mm) have a filtrate concentration of 0.32/0.33 mg total P/L. The lower SS values from filter setting 6, both in period 1 and period 2, may be a result of a deeper sand layer in that filter with a better removal and larger particle accumulation capacity in a filter with a thicker sand layer, 0.5 m, compared to 0.3 m in filter settings 7 and 8.

The major difference, however, is the time of operation. Both period 1 and the carbon source period in filter setting 7 have concentrations of 8.9 mg SS/L, but the time of operation was shortened from 41.6 h without a carbon source to 28.8 h with a dosage of NaAc. As assumed, denitrification decreased the time of operation considerably, hereby on average of 31% despite that the influent concentration of SS to the denitrifying filter was only around two-thirds of the SS to the same filter during non-nitrifying conditions. Comparing filter setting 6 and filter setting 7 in period 1, the time of operation was 43.5 h in filter setting 6 and 41.6 h in filter setting 7. In period 2, filter settings 6, 7, and 8 had operation times of 34.3, 30.6, and 22.0 h, respectively. This might point to a longer time of operation with larger grain sizes of the sand. Between filter settings 6 and 7, the difference was not too large but filter setting 6 has a sand layer of 0.5 m compared to 0.3 m in filter setting 7, which can compensate for the somewhat larger sand grains in filter setting 6. When filter settings 7 and 8 are compared, the difference in time of operation is large enough to be statistically significant, 30.6 and 22.0 h. The conclusion is that a smaller grain size gives the same concentrations of phosphorus and approximately the same concentration of SS in the filtrate but a large decrease in the time of operation. This also makes it possible to compare the concentrations in the filtrate from the full-scale filter No. 60 with the filtrate from filter setting 7 in the pilot plant study. A grain size of Ø 0.6–0.8 mm studied in filter setting 8 was considered too small to be chosen for the full-scale filters.

In the full-scale filters in the Henriksdal WWTP, simultaneous separation of SS, denitrification, and phosphorus precipitation take place. In this study, an evaluation of head loss measurements was done to identify the dynamic part of the total head loss, especially during denitrification when gas bubbles were clogging the filter.

The head at different depths in the sand filter bed during one filter cycle followed a similar pattern as in filter setting 7, Figure 5(a)–5(j). A level meter detected the end of the cycle when the water level reached around 2.75 m. At 2.9 m, the water level reached the overflow tube. This was detected by the meter placed at −0.15 m. The main head loss seemed to occur between −0.15 and 0.25 m i.e. on the surface of the filter bed. A smaller head loss was also detected between 0.25 and 0.70 m in the filter bed in some experiments according to Figure 5. In the full-scale filter, small head losses were also found deeper down in the filter bed, Figure 6. This might be due to a smaller area of the filter surface, which probably makes it easier to backwash the pilot filter.

The time of filter operation varied with hydraulic load, the concentration of SS in the influent, the dosage of NaAc and Fe, and addition of SS and PW for different experiments. This resulted in varying SS loads although the hydraulic load was constant. Comparing all the experiments, the time of operation for one cycle as average values with different concentrations of SS in the influent, varied between 4.3 and 175.2 h, Table 3. The addition of SS decreased the operational time by most. With a concentration of 60–77 mg SS/L in the influent, the operational time decreased to as low as 4.3 h. PW addition gave longer operational times in the size of 9–22 h, but here the SS concentration in the influent was around 29 mg SS/L. Both NaAc dosage and Fe dosage decreased the time of operation. With Fe dosage, a time of 42.6 h was found at 10 m/h and 49.3 h at 5 m/h with only a small difference between 5.8 mg SS_in/L compared to 5.2 mg SS_in/L, respectively. With no dosage, the time of operation was 80.3 h at 10 m/h and 4.8 mg SS_in/L compared to 175.2 h at 5 m/h and 8.1 mg SS_in/L. Despite a higher influent SS concentration at 5 m/h, the time of operation was more than twice as high. With both NaAc and Fe dosages, the time of operation was 29.3 h at 10 m/h and 9.7 mg SS_in/L compared to 30.8 h at 5 m/h and 3.2 mg SS_in/L. The values in Table 3 showed generally shorter operational times in most of the cases with 10 m/h compared to 5 m/h. With solely SS addition, the time of operation was 10.1 h at 10 m/h and 61.8 mg SS_in/L compared to 17.1 h at 5 m/h and 60 mg SS_in/L.

Without NaAc dosage, the times of operation were 218.7, 49.8, and 76.3 h at the hydraulic loads of 3.3, 5, and 10 m/h, respectively, Figure 5(a)–5(c). With NaAc dosage and at a hydraulic load of 5 m/h, the times of operation were 60.6 and 62.3 h, Figure 5(d)–5(e), and at 10 m/h, 31.9 h, 35.3 h, and 39.1 h (the last excluding the time after bumping), Figure 5(f)–5(h), all with different concentrations of SS in the influent. When SS and PW were dosed, the times of operation were 10.1 and 18.4 h, respectively, with a load of 10 m/h, Figure 5(i) and 5(j).

It was observed that the meter at 1.40 m down in the filter bed corresponding to 0.10 m above the filter bottom had lost the head from approximately 2.5 mH₂O to around 0.6–1.3 mH₂O at the time of clogging, Figure 5(a)–5(c), 5(i) and 5(j). This points to a head loss of 1.2–1.9 mH₂O in the filter bed and a remaining head loss of 0.5–1.2 mH₂O over the nozzles in experiments without carbon source dosage assuming a head of 0.1 mH₂O at clogging if comparing with the calculation of the head loss in the full-scale study also starting at approximately 2.5 mH₂O and reaching 0.1 mH₂O at clogging. In experiments with NaAc dosage and a hydraulic load of 10 m/h, partly different patterns regarding the time of operation were observed, Figure 5(f)–5(h). A head loss of 1.2–1.95 mH₂O in the filter bed and a head loss of 0.45–1.2 mH₂O over the nozzles was calculated from the values found in the experiments i.e. the same values as without NaAc dosage, but the time of operation became shorter, Table 7.

In the evaluation of the dynamic clogging, one of the full-scale filters (No. 60) at the Henriksdal WWTP was investigated. A head loss in the sand layer was first detected as a large difference in pressure between the probes at 0.8 and 1.5 m down in the filter bed, Figure 6(e)–6(h). The pressure meter at 0.6 m was first moved to 1.2 m to locate where the large head loss occurred in the filter bed. Figure 6(i) showed a large head loss between 1.2 and 1.5 m. Soon after that, the meter at 1.2 m was moved to 1.4 m to further locate the depth of the dominant head loss. The pressure meters still indicated a large head loss at the filter bottom, now between 1.40 and 1.50 m down in the filter bed, Figures 6(j) and 6(k).

The measured head loss could be interpreted in several ways. It could be a result of SS not being separated by the filter bed but transported by the water down in the filter. However, this was considered less probable. Previous pilot plant studies and a built-in observation glass window in one of the full-scale filters showed that almost all SS were accumulated on the surface of the uppermost part of the ceramic layer in the filter bed, which most probably also was the case for the pilot-scale filter. The two main factors for the increase of dynamic head loss are related to denitrification. Sludge, i.e. bacteria, was produced in the filter bed during denitrification and this sludge might have clogged the filter. Another possibility was that carbon dioxide and nitrogen gas produced in the filter bed during denitrification clogged the filter by trapping gas bubbles around the nozzles and evolving gas pockets in the bed material. After the termination of the study, parts of the bed in filter No. 60 were excavated. There was no higher concentration of SS found between 1.40 m and the bottom of the filter. The conclusion is that it probably was gas bubbles that mainly clogged the filter between 1.4 and 1.5 m. Boller et al. (1997) report that trapped nitrogen gas occupied 50% of the pore space at a depth of 0.80 m in the filter bed. According to their measurements, this was at the depth where the maximum concentration of nitrogen gas was found. Koch & Siegrist (1997) did similar observations, where the produced nitrogen gas during denitrification filled 50% of the pore volume in the filter bed when the filter was close to clogging. The head loss was increasing significantly when the gas bubble accumulation reached 30–40% of the pore volume. Jepsen & La Cour Jansen (1993) studied a down-flow sand filter with denitrification. They found that the head loss in the filter became larger than the available hydraulic capacity abruptly after approximately 15.3 h at a hydraulic load of 4.9 m/h and a nitrogen load of 1.3 kg NO₃-N/(m³·d). This was a result of the production of sludge, nitrogen gas, and carbon dioxide in the filter during denitrification. In their study, the time of operation before clogging reached 45 h for loads up to a maximum of 1.5 kg NO₃-N/(m³·d) and at 4.0 kg NO₃-N/(m³·d) the clogging took place after 12 h. At a yield of 0.5 g SS/g COD and a C/N ratio of 5 consumed, they found that 20–40% of accumulated SS in the filter came from the influent containing 65 mg SS/L and the rest arose from the SS production during denitrification. A ratio of 5 for C/N consumed is close to 4.4 g COD/g NO_x-N, 4.5 for ethanol and 4.3 for methanol, as average values in this study. Al-Saedi et al. (2019) studied clogging in an up-flow filter dosed with a carbon source. They found that the filter dosed with ethanol clogged first followed by the sucrose-dosed filter. The filter with only wastewater in the influent was the last to clog. It resulted in a decrease in the time of operation to 36% for ethanol and 66% for sucrose compared to the time of operation with only wastewater. In the study, the carbon source was first added to a C/N ratio of 2 and was later changed to 4, which was close to the carbon source dosed to a C/N ratio of 3.9 g COD/g NO_x-N, 3.9 for ethanol and 3.8 for methanol, as average values in this study. As average values in this study, the nitrogen load was 0.46 kg NO₃-N/(m³·d) at a yield of 0.11 g SS/g COD for an influent concentration of 16 mg SS/L. The abrupt clogging probably came from nitrogen gas and carbon dioxide produced during denitrification. At that point of filtration, the maximum available head was of low interest as the head loss due to accumulated gas bubbles was high enough to immediately and totally clog the filter. It was, instead, the total time of operation before clogging that determined the functionality of the filter. Chen et al. (2021a, 2021b) found that clogging was induced by biofilm growth in an intermittent sand filter. This clogging comprised dominantly of heterotrophic microorganisms and occurred mostly in the top layer of the filter. This is interesting as microorganisms involved in denitrification are heterotrophic bacteria.

In filter No. 60 in the Henriksdal WWTP, the dynamic head loss in the filter bed during a cycle of operation was evaluated. Both the carbon source dosage and the concentration of SS in the influent varied during the study, Figure 6. At the end of the operation cycle, the head had decreased from around 2.5 mH₂O in the filter to a value of around 0.1 mH₂O, i.e. around 2.4 mH₂O in head loss. The hydraulic load was kept at 3.3 m/h. In a filter with no carbon source dosage and thereby no denitrification, the head loss between 1.40 and 1.50 m was considerably smaller, Figure 6(a)–6(d). The time of operation before clogging was much shorter in a denitrifying filter, Figure 6(e)–6(f), compared to a filter without denitrification. With Fe and carbon source dosage, time of operation between 12 and 27 h were detected with ethanol or methanol dosed as an external carbon source to the filter, Figure 6(l).

In Figures 6(a)–6(d) without carbon source dosage, the time of operation was around 41–47 h. The head at clogging, measured from the 1.5 m pressure meter, reached 0.1 mH₂O, which gives a total head loss of 2.5–0.1 = 2.4 mH₂O. The head according to the pressure meter at 1.4 m reached 0.9–1.5 mH₂O giving a head loss of 1.0–1.6 mH₂O in the filter and a remaining head loss of 0.8–1.4 mH₂O over the nozzles and the bottom of the filter.

With ethanol as a carbon source, a head loss of 0.8–1.1 mH₂O was found in the filter bed not including the nozzles, Figure 6(e) and 6(f). The total head loss including the filter bottom with the nozzles was detected with the pressure meter situated below the filter bottom, 1.5 m in the figures, which reached 0.2 mH₂O at clogging. This gave a total head loss of 2.3 mH₂O and a time of operation of 24 h. The head loss over the filter bottom became 1.2–1.5 mH₂O. Here, nitrogen gas and carbon dioxide bubbles probably were mainly responsible for the head loss. The time of operation for all experiments with ethanol dosage was 12–27 h, Figure 6(l), excluding the first experiments when the microorganisms adapted to ethanol.

Head loss curves from the experiments with methanol as a carbon source showed similar patterns, Figure 6(g)–6(k). The total head in the filter reached 0.1–0.2 mH₂O at clogging giving a total head loss of 2.3–2.4 mH₂O. The lowest pressure meter in the filter bed was situated at 0.8 m showing a head loss of around 0.7–0.8 mH₂O in the filter bed, Figure 6(g) and 6(h), and a remaining head loss of 1.6 mH₂O over the nozzles. A head loss was calculated for experiments with pressure meters situated at 1.2 and 1.4 m to 1.0–1.5 mH₂O in the filter bed and 0.9–1.4 mH₂O over the nozzles, Figure 6(i)–6(k). With methanol as a carbon source, the time of operation became 16–27 h for all experiments with the dosage of methanol, Figure 6(l).

As can be seen from Figure 6, head loss in the experiments with carbon source dosages was smaller in the filter bed (the filter bottom with nozzles excluded) at clogging than in experiments without a carbon source dosage. All experiments clogged at 0.1–0.2 mH₂O giving a total head loss of 2.3–2.4 mH₂O. If the head loss in the total filter bed from the experiments without a carbon source is detected in Figure 6(a)–6(d) after 24 h of operation, values of around 0.9–1.3 mH₂O were found. This is close to the head loss in the filter bed (the filter bottom with nozzles excluded) at clogging (after 21–27 h) detected in Figure 6(e)–6(k) showing experiments with carbon source dosage. Here, a head loss of 0.7–1.5 mH₂O was found in the filter bed excluding the nozzles, Table 7. The conclusion is that the SS in the influent were not responsible for the remaining head loss found over the nozzles at clogging, Figure 6(e)–6(k). Instead, this head loss resulted from nitrogen gas and carbon dioxide bubbles trapped by the nozzles. This also explains why the clogging in a denitrifying filter seems so abrupt. The clogging in the upper part of the filter bed is still relatively small. From Figure 6(a)–6(k), an average decrease in time of operation of 46% was calculated when a carbon source was dosed. The influent SS concentration was higher, on average 20.5 mg SS_in/L, to the non-denitrifying filter compared to the denitrifying filter, on average of 16.3 mg SS_in/L. This suggested a corresponding decrease of 57% if the filters had been fed with the same SS concentration in the influent.

In Table 6, average values of different substances from the full-scale study with different dosages of COD are presented. The time of operation became around 22 h when a carbon source is dosed to the filter. Then the filter was abruptly clogged with gas. Without carbon source dosage, a time of 44 h was observed as an average value.

A high head loss over the nozzles was also found in experiments without denitrification, but the time of operation was much longer giving a higher total load of SS during these experiments, which might point to clogging by SS over the nozzles, i.e. approximately the same head loss after 24 h in the total filter without denitrification. This indicates that the clogging is caused by SS. The remaining head loss over the nozzles is probably caused by gas bubbles produced during denitrification. In Table 7, the head losses from Figures 5 and 6 are summarised. As can be seen, after 24 h in the full-scale filter without carbon source dosage a total head loss of 0.9–1.3 mH₂O was found. This can be compared with the experiments with ethanol/methanol dosage where a head loss of 0.7/0.8–1.1/1.5 mH₂O was found in the filter bed at clogging, not including the nozzles, Table 7.

A comparison with studies of one, two, and three media filters at other sites showed that the parameters in our study were contained in the previously reported broad range of process conditions. In comparison, the hydraulic load varied between 5 and 24.4 m/h, the concentration of SS in filter feed varied between 2.0 and 56 mg SS_in/L, and the time of operation varied between 7 and 106 h. The corresponding values from our studies were 3.3 m/h, 15–20 mg SS_in/L, and 25–72 h for the full-scale filter and 3.3–10 m/h, 4–63 mg SS_in/L, and 7.3–219 h for the pilot plant filter. Table S1 in supplementary material summarises data from our studies and data from Boller (1984), Al-Jadhai (2003), Angermüller et al. (1998), Brenner et al. (1994), Altmann et al. (2016), Williams et al. (2007), Tchobanoglous (1970), Zenz et al. (1973), and Oliva (1973).

One of our experiments, number 34 with bumping in the pilot plant filter setting 7, showed that 4.8 h of operational time was gained both after the first and the second bumping, which is in good agreement with the result of 4–5 h from the full-scale filter, Figure 8(c) and 8(d). The gain in operation time is also observed in Figure 5(h) as the time on the X-axis. The first bumping can be detected in Figure 5(h) from the point when the water level decreased from around 2.9 mH₂O to approximately 2.65 mH₂O. After that, a period of 4.8 h filter operation occurred when the water level increased from 2.65 to 2.9 mH₂O and the filter was clogged again. A second bumping was started and the water level decreased to 2.65 mH₂O. Finally, a new period of operation of 4.8 h took place. When the water level reached 2.9 mH₂O for the third time, the experiment ended. The time of operation increased from 39.1 to 48.7 h when two bumpings were included. A gain of 4.8 h at each bumping was, however, considered to be too short a time especially as the concentration of SS and corresponding total P in the filtrate increased directly after that the filter bed had been ‘swirled’ during the bumping, which was the case both for the full-scale and the pilot-scale filter.

Common operational times at loading of 3.3 m/h for the denitrifying full-scale filter with Fe dosage were approximately 12–27 h and most often 16–27 h. Without denitrification but with Fe dosage the time of operation could be as long as 72 h before clogging.

Operational time at 5 m/h for the denitrifying pilot plant filter with Fe dosage was 30.8 h and without Fe dosage 60.6–62.3 h. Without denitrification but with Fe dosage, the time of operation became 41.8–56.3 h and without Fe dosage 175.2 h. At 3.3 m/h without Fe dosage, the time of operation could last over 200 h. Operational times at 10 m/h for the denitrifying pilot plant filter with Fe dosage were 22.4–36.2 h and without Fe dosage 20.1–50.2 h. Without denitrification but with Fe dosage, the time of operation became 23.6–101.6 h and without Fe dosage 59.8–104.9 h.

Studies performed at the Henriksdal WWTP show that dynamic head loss in the sand filters consisted of several factors namely (1) from clogging with SS from the biological treatment, without or with sludge overflow, (2) from bypassed PW, (3) from accumulated nitrogen gas and carbon dioxide bubbles and possibly (4) from SS produced during denitrification in the biologically active filter. The measures taken to decrease the dynamic head loss involved interrupting the dosage of both carbon source and ferrous sulphate during bypassing of high flows of PW to the sand filter to prolong the time of filter operation but also perform bumping of the filter bed with water (filtrate) to remove trapped gas in the filter bed material.

The pilot studies gave relevant data and information for the design of the full-scale filters. Operational data were similar in pilot-scale and full-scale filters. The operational times decreased with higher hydraulic loadings, which was most notable in periods without Fe dosage.

Denitrification shortened the operational times significantly due to nitrogen gas and carbon dioxide bubbles that clogged the filter at the bottom of the filter bed during denitrification. The head decreased abruptly in this layer and this was not caused by particles, which was confirmed by excavation in the filter bed after the termination of the study.

In a denitrifying filter with abrupt clogging induced by gases, the total available head in the filter is of less interest. It is instead the time of operation that defines the functionality of the filter.

The operational time for a filter cycle was prolonged by 50% with one bumping and by 75% after two bumpings. After 6–7 bumpings, around 85% of the head remained giving a head loss of 15% in the full-scale filter. Bumping increased the time of operation between 4 and 5 h per bumping in both the full-scale and the pilot-scale filter.

Bumping was not introduced in the ordinary operation in the Henriksdal WWTP as the gain of operational time was considered too short as well as the increase of the SS and total P concentrations in the filtrate directly after a bumping was considered too high.

Valuable support and help in the preparation of the manuscript by Professor Elzbieta Plaza is greatly appreciated. This work was supported by Stockholm Water Co. and NUTEK (within the STAMP project).

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.