Characterizing and modeling hydrogen sulfide production in anaerobic digestion of livestock manure, agro‐industrial wastes, and wastewater sludge

Hydrogen sulfide (H2S) is the most undesirable inorganic gas in biogas from anaerobic digestion (AD). However, H2S production in AD is complex and understanding of its processes is still limited. This study performed six controlled batch anaerobic co‐digestion experiments to investigate H2S production. Materials were obtained from four field anaerobic digester systems and co‐digestion feedstocks from agroindustry. An additional precipitation experiment was conducted to further examine H2S production dynamics. Digesters containing highly soluble, carbohydrate‐based wastes had a high H2S final specific production (FSP) value. Additionally, the FSP values were negatively correlated with the initial Fe(II):S ratios in the digester liquid of the batch tests. The precipitation experiment indicated that iron sulfide precipitation was preferred in the presence of an anaerobic community. The H2S production as a time series was successfully modeled using a generalized additive model (R2 > 0.82). This study revealed that sulfate, phosphorus, and iron concentrations are important predictors and potential inhibitors of H2S production in AD. Further examination of real‐time H2S modeling in AD is warranted.


| INTRODUCTION
There is an increasing global interest in renewable energy production to reduce dependency on fossil fuels and to mitigate emissions of greenhouse gases (GHG) to combat climate change. Manure management and landfills accounted for 17.4% and 9.7% of total United States methane (CH 4 ) emissions in 2018, respectively . Fossil fuel combustion alone accounted for 92.7% of carbon dioxide (CO 2 ) emissions in the United States .
Anaerobic digestion (AD) is a solution for these issues. It is a mature and cost-effective technology that uses a mixed microbial community to convert pre-existing wasted biomass to biogas. Biogas is composed mainly of CH 4 (50%-70%) and CO 2 (30%-50%). This biogas can then be used as a renewable energy source to provide heat or generate electricity, thereby replacing conventional fossil fuels (NREL, 2013). Capturing biogas for energy use also reduces GHG emissions from the wasted biomass. In 2019, AD systems on livestock farms were able to reduce GHG emissions by 4.63 million metric tons of CO 2 equivalent (U.S. EPA, 2020).
Hydrogen sulfide (H 2 S) is the most undesirable inorganic gas generated during the AD process (Gerardi, 2003) with typical concentrations of less than 10,000 ppm in the biogas. It is extremely toxic and can cause nearly instant death to humans and animals at concentrations from 1000 to 2000 ppm (OSHA, n.d.). Moreover, H 2 S mixtures in air between 4.3% and 46% by volume are explosive (Smith et al., 1979). To protect the equipment, H 2 S concentrations must be kept below 1000 ppm for combined heat and power generation and boiler applications, and removed almost entirely for vehicle fuel or natural gas use (Rasi et al., 2011).
During AD, H 2 S is formed through sulfate reduction and the degradation of sulfur-containing organic compounds. Modeling H 2 S production in AD systems via sulfate reduction is of interest as a means to predict digester failure (Barrera et al., 2015). During sulfate reduction, sulfate-reducing bacteria (SRB) can convert the acetate generated during the hydrolysis process to H 2 S (Vavilin et al., 1994) as shown in Equation (1).
If H 2 S concentrations become high enough (>200 mg L −1 HS − ), methanogenesis may be inhibited (Gerardi, 2003). One cause of this inhibition is when SRB outcompete methanogens for acetate and hydrogen to produce H 2 S. Sulfide can also permeate the cell membranes of anaerobic bacteria and disrupt metabolic activity.
Inhibition can also occur when sulfide precipitates metals and thus inhibits methanogenesis by depriving microorganisms of essential nutrients (Gupta et al., 1994).
There are several proposed methods for preventing H 2 S production in AD. These include the addition of iron salts, sulfur scavenging microorganisms, adsorbents, or oxygen into the digester (Appels et al., 2008;Cirne et al., 2008;Song et al., 2001). However, these methods can be expensive and inefficient or cause other side effects (Peu et al., 2012). Hence, H 2 S is typically scrubbed from the biogas rather than preventing its formation in the AD process (Choudhury et al., 2019;Peu et al., 2012). Hydrogen sulfide production in AD is usually studied in controlled batch lab-scale digestion, as opposed to full-scale systems.
Connecting H 2 S production only to the sulfate concentration in the AD system can oversimplify the other influences on H 2 S production. For example, the presence of metals can influence the conversion of sulfur. Batch digester tests of dairy manure (DM) demonstrate that iron and copper play a role in reducing H 2 S concentrations in the biogas through sulfide-metal precipitation (Lin et al., 2017). Additionally, the pH, chemical oxygen demand (COD):sulfate ratio, and sulfate concentrations of the digester liquid can influence the sulfate reduction process and subsequent inhibition thresholds (Guerrero et al., 2013;Vavilin et al., 1994). For example, sulfide toxicity intensifies at acidic pH values (Koster et al., 1986). Moreover, H 2 S production in AD could be complicated by anaerobic co-digestion. A model to predict H 2 S production in batch tests was developed from the C:S ratio in anaerobic mono-digestion on 37 different feedstocks (Peu et al., 2012). However, the model did not account for sulfur form or feedstock mixtures.
The goal of this study is to gain a new understanding of H 2 S production. The specific objectives of the study are to (1) identify influential factors on H 2 S production in co-digested AD systems that can serve as indicators for potential digester inhibition; (2) reveal dynamic behaviors of H 2 S production and digester liquid characteristics; and (3) determine influential factors for controlling H 2 S production.

| Overview of four field digesters
The substrate and inoculum used in the experiments were collected from four different industrial scale field digesters (Digesters F, B, L, and W) located in the Midwestern United States (Table 1). Collections occurred on seven dates between January 2018 and November 2019 for seven controlled lab experiments (Section 2.2). The substrate and inoculum were taken at different locations of the field digesters. After collection, they were transported to Purdue University and immediately used in laboratory studies.

| Digester F
Digester F is a mixed plug-flow digester treating DM and dairy farm wastewater, including milk processing wastewater. It operated at mesophilic conditions (~42°C) and a hydraulic retention time (HRT) of 15 days (Table 1). Digester F produced about 34,000 m 3 of biogas daily. Three types of materials were collected in January 2018. These included DM, digester effluent (EF), and effluent after solid separation and phosphorus recovery (EF-R) from the EF.

| Digester B
Digester B is also a mixed plug-flow design, consisting of parallel and isolated digester bodies. It operated at ~38°C and an HRT of 28-32 days to produce about 61,000 m 3 of biogas daily (Table 1). Digester B received beef cattle manure (~50%) with other co-digestion materials including food waste, glycerin, and biodiesel waste.
Samples were collected from Digester B on four dates, that is, June 5, 2018;August 9, 2018;October 24, 2018;and February 18, 2019, and used in Tests 2, 3, 4, and 5, respectively. Digester B was reported as experiencing foaming at the first three collection times. Additionally, glycerin loads were reported as being reduced during the last collection.
Six types of materials were taken at different locations of the digester system. These materials were (1) digester influent (INF) in the equalization pit after feedstock mixing, (2) and (3) digester liquid in the middle of the east (DL-E) and west (DL-W) digesters, (4) and (5) digester effluent in the effluent pits of the east (EF-E) and west (EF-W) digesters, and (6) effluent with solids removal (EF-L).

| Digester L
Digester L was at a wastewater treatment plant (WWTP), operating at 36°C and an HRT of 15-30 days. It treated waste primary sludge (PS), which is the product after initial solids removal (Tchobanoglous & Schroeder, 1987), and waste activated sludge (WAS). Digester L produced about 1900 m 3 of biogas daily. Influent and effluent at Digester L were collected in May 2019 and used in lab Test 6.
Additionally, three types of dry and powdered cornstarchbased co-digestion wastes (identified as S1, S2, and S3) from the facility were used as the substrate for Test 6. Substrate SM was an equal mix by mass of S1, S2, and S3. T A B L E 1 Characteristics of the four industrial-scale field anaerobic digesters, and the materials collected from each digester. The laboratory studies were conducted in customized digesters. For experiments involving materials from Digesters F and B (Tests 1-5), the digesters were made of 500 mL (with 6.35 mm OD barb side outlet port) sealed with #8 rubber stoppers or 1000 mL (with 6.75 mm OD barb side outlet port) flasks (Bomex borosilicate glass filtering flasks) sealed with #10 rubber stoppers. A piece of pipe, cut from a 10-mL Falcon Pipet (#357551), was inserted through each rubber stopper and used as an operation port for sampling and measurement. When the digesters were sealed with the stoppers, the lower end of the pipe extended to below the digester liquid so the biogas produced in the digesters could not escape from the operation port. The side outlet port in the digester body was used as a biogas collection port. Two 500-mL Tedlar gas bags or re-enforced gas bags in parallel were attached to the side outlet port using a Tee to collect biogas (Table 2). For experiments involving materials from field Digester L (Test 6), the lab digesters were made of 1000-mL Corning polycarbonate square bottles. The original 45mm screw caps of the bottles were replaced with #8 rubber stoppers. In addition to the operation port, a biogas port was added by inserting a short piece of pipe, cut from a 2-mL polystyrene Falcon Pipet (#357507), through the stopper. The lower end of the pipe opened at the headspace of the digester to release the biogas. The top end of the pipe was connected to a 3-L or a 1-L Tedlar bag, depending on the biogas production rates during different stages of AD, to collect the biogas. To ensure air tightness, the pipes were sealed to the stopper using a silicone sealant.

| Lab-digester operation
Immediately after loading the digesters with influent, that is, the mixture of substrate, inoculum, and reverse osmosis (RO) water when necessary, the digesters were randomly placed in several water baths (Model YCW-010; Gemmy Industrial Co.; and Model WB28A11B; PolyScience). The temperatures of the water baths were controlled at 38.3°C ± 0.1 for Tests 1-5 and at 36.0°C ± 0.1 for Test 6 ( Table 2). The tests ended when the CH 4 daily production was less than 1% of its cumulative production.
Tests 1-6 were conducted with different combinations of substrate and inoculum. The substrate included DM, EF-R, INF, and mixture of PS and WAS. The inoculum was either EF, EF-R, or DL to ensure the necessary microbial consortia for CH 4 . Additionally, digesters with 100% inoculum to determine the activity of the different stages of the AD system were used. The DL from field Digester B experienced foaming issues in Test 2 so, in subsequent tests, the DL was diluted with RO water to 50% in the digesters.

Test 1 Test 2 Test 3 Test 4 Test 5 Test 6
Digester volume (mL) 500 1000 1000 1000 1000 1000 Total group (n) 6 6 6 5 5 17 Blank group (n) Tests 1-5 were all performed with digester groups in triplicate, except for blank groups and Test 6, which were in duplicate. "Blank" control groups consisting of inoculum and RO water were used to determine the production of CH 4 from the inoculum (Table 2).

| Precipitation experiment
Based on the results of Tests 1-6, an additional experiment (Test 7) was designed to gain more insight into the role of iron, phosphorus, and sulfide in AD (Table 3). The 1000-mL Corning square bottle digesters were used in this test, and the digestion temperature was controlled at 38.3°C ± 0.1. Treatments were designed to determine whether changes in iron, phosphorus, and sulfide in these digesters were biologically mediated or solely dependent on physical-chemical reactions. Therefore, half of the treatments, except for the control, contained 10% of their volume as inoculum. The remaining volume in the digesters was filled with RO water. The inoculum was obtained from field Digester W as described in Section 2.1.5. To test FeS precipitation, sulfide was the limiting reagent. To test Fe-PO 4 precipitation, Fe was the limiting reagent. Ferric chloride hexahydrate (VWR Analytical) was the iron source, phosphoric acid 85% (Mallinckrodt Chemicals) was the phosphorus source, and sodium sulfide nonahydrate (ACS, 98.0% min, Crystalline, Na 2 S · 9H 2 O; Alfa Aesar) was the sulfide source. On Day 0, each digester received 3 g of sodium bicarbonate (Mallinckrodt Chemicals) and 0.001 g of resazurin sodium salt (Thermo Fisher Scientific) as a buffer and oxygen indicator, respectively. Each treatment was performed in duplicate.
On Day 0 and 26, the concentrations of Fe(II), TP, and sulfate in each digester were measured as described in Section 2.3. The Fe(II), TP, and sulfate concentration data were first tested for normality using the Shapiro-Wilk test (Ghasemi & Zahediasl, 2012). If the data were not normally distributed (p < 0.05), then a Wilcoxon signed-rank test was applied. If the data were normally distributed, then a paired t-test was applied (McDonald, 2014). The p values were corrected with a Bonferroni correction method (Jafari & Ansari-Pour, 2019).

| Biogas measurement
The biogas volume and composition were measured daily for the first 3 days, and then at intervals of no more than 3 days until the end of the tests. Before detaching from the digesters, the gas bags were sealed with metal screw clamps (16 mm × 19 mm). The bags were immediately replaced with new bags.
The volumes of biogas in the bags were measured using a custom-made device and a 200 mL syringe. All biogas volumes were converted to standard temperature and pressure (1 atm, 0°C) before data processing. Biogas compositions were measured with a 5000 gas analyzer (LAND-TEC North America, Inc.), which has detection ranges of 0%-100% CH 4 ; 0%-100% CO 2 ; 0%-25% oxygen (O 2 ), and 0-10,000 ppm H 2 S. The specific methane yield was calculated by dividing the final cumulative CH 4 volume by the grams of the initial volatile solids (VS).

| Substrate, inoculum, and digester effluent characterization
The substrate and inoculum from the field digesters (Day 0 of each test) and the final liquid (effluent) from the lab-digesters at the end of each lab experiment were thoroughly characterized to determine several chemical and physical parameters. The total chemical oxygen demand (TCOD), soluble chemical oxygen demand (SCOD), total volatile fatty acids (TVFAs), total alkalinity (TALK), total sulfate, total phosphorus (TP), total Kjeldahl nitrogen (TKN), total nitrogen (TN), inorganic N, iron (Fe(II)), copper, and nickel were measured using Measurements necessitating particulate samples (pH, solids, conductivity, TCOD, sulfate, tannic acid, TAN, and TP) were thoroughly mixed with a magnetic stir bar to ensure consistency. Measurements requiring soluble samples were filtered with a cellulose acetate filter of either 0.2 μm (SCOD) or 0.45 μm (TKN, TVFA, TALK, OP, Fe, Cu, Ni). Dilutions were performed on a g g −1 basis.
The pH of the digester liquid was measured through the operation port on the same days when the biogas volume and concentration were measured. The liquid in the digesters was thoroughly mixed with a magnetic stir bar before pH measurement.

| H 2 S production calculation
To have a uniform comparison between the different digesters, the FSP value was calculated using Equation (2): where FSP is the H 2 S final specific production, mL g VS −1 ; C H 2 S is the H 2 S concentration in the biogas, ppm; V Biogas is the volume of the biogas, L; W VS is the initial weight of VS fed into the digester, g; n is the number of biogas collections during the test; and i is ith biogas collection.
Pearson and Spearman correlation tests between the FSP values and initial characterizations were performed using the cor() function in Rstudio (R Studio Team, 2022).
The total dissolved sulfide (TDS) concentration was calculated from the H 2 S concentration measured in the biogas and the digester liquid pH using Equation (3): where [H 2 S] g is the free H 2 S gas (mols L −1 ), TDS is H 2 S(l) + HS − + S 2− (mols L −1 ), and K d is the equilibrium dissociation constant for H 2 S which was 1.639 × 10 −7 at 38.3°C and 1.535 × 10 −7 at 36.0°C (Isa et al., 1986;Lawrence et al., 1964).
The total soluble sulfide (TSS) concentration (HS − + S 2− ) was then calculated using Henry's Law as shown in Equation (4): where α is the absorption coefficient, which is 1.73 at 38.3°C and 1.80 at 36.0°C, and H 2 S concentrations are expressed in moles per liter of liquid and moles per liter of gas, respectively.
The ratio of [H 2 S] g to TSS was plotted against pH to determine the effect of pH on TSS concentration and the gaseous form of H 2 S (Lawrence et al., 1964).

| COD removal
The COD removal by SRB was calculated as shown in Equation (5) (Guerrero et al., 2013).
where %COD removed is the percent change in the initial and final TCOD concentrations.

| Model development
Local regression (loess) and generalized additive model (gam) were used for modeling the specific H 2 S production due to their ability to fit smooth curves to complex data (Thakur et al., 2018;Wood & Augustin, 2002). Additionally, they were selected for their ability to capture the complexity of environmental modeling (Wood, 2001). Loess is a nonparametric method that uses data points of less than 1000 to predict the local y value through fitting multiple regressions in a local neighborhood of numerical data. Gam uses nonlinear regression to fit local y values in a local neighborhood of numeric data and is a more generalized version of loess. The gam model is shown in Equation (7): where β 0 is the intercept; f m is the smoothing function; and s is the link function (identity). The loess and gam curves were generated using the mcgv package (Wood, 2011) in RStudio. The generated model parameters and coefficients form this study can be found in Daly (2023 The FSP values from 45 digester groups in the six tests showed considerable variability. The lowest FSP ranged from nondetectable (F-1-6) to 1.29 mL g VS −1 (L-6-41) ( Table 4). The lab-digesters that produced ≥0.2 mL H 2 S g VS −1 were all from Test 6 and one from Test 2 (B-2-10). The lab-digesters from Test 6 received co-digestion substrates S1, S2, S3, or SM, which were cornstarch-based wastes. Lab-digester groups L-6-41 and L-6-43 had the highest FSP values with 1.29 and 0.75 mL g VS −1 , respectively. The material in these digesters underwent extensive solubilization as evidenced by increases in the average SCOD:TCOD ratios (0.02-0.44; 0.03-0.51) and VS:TS ratios (0.57-0.63; 0.59-0.67) over the digestion period. This solubilization most likely increased the availability of acetate for sulfate reduction, thus increasing H 2 S production (Equation 1).
Correlation calculation results showed that the most important influences on FSP values, disregarding those characteristics containing solids measurements, were the Day 0 Fe(II):S and OP concentrations (Table 5). Day 0 sulfate concentrations were not highly correlated with FSP values, but Fe(II) was. Additionally, Day 0 Cu concentrations were negatively correlated with H 2 S production (Table 5). Because the majority of H 2 S was produced relatively early in the experiment (Figure 1), the initial concentrations of these metals and nutrients most likely had a complex effect on H 2 S production. The FSP values and TCOD:sulfate ratios were not significantly correlated ( Table 5). The TCOD:sulfate ratio has been commonly used as an indicator for the likelihood of SRB or methanogen predominance. It has been suggested that values from 1.7 to 2.7 indicate competition, with SRB predominance at lower values (Choi & Rim, 1991). In this study, none of the digesters had initial or final average TCOD:sulfate ratios below 4, suggesting that H 2 S production could not be attributed to the low TCOD:sulfate ratio.
The H 2 S gas production was pH dependent. For most tests, pH varied between 6 and 8 during the test period.
When the pH was below 7, the fraction of [H 2 S] g TSS −1 approached 0.50 (Figure 2). Higher pH values, particularly in Test 4, corresponded with lower [H 2 S] g TSS −1 fractions. In test 4, pH was largely above 7 for most of the test, which explained the lower FSP values (Table 4). 3.1.2 | TCOD:sulfate ratio predicted COD removal by SRB The amount and type of COD removal varied considerably. In most tests, COD was removed by methanogenesis (Figure 3). The digesters in Test 5 had some of the lowest COD removal rates, with rates of less than 22%, except for lab-digester group number B-5-28 (Figure 3). Lab-digester group number B-5-28 was the blank and subsequently had a low initial TS concentration (12 g L −1 ) compared with the other digesters in the test (34-138 g L −1 ). Overall, COD removal was not significantly correlated to the FSP values (Table 5). Test 5 had some of the highest percentages of COD removal by SRB (6.22% and 7.59% in digesters B-5-27 and B-5-28, respectively), most likely due to the low initial TCOD:sulfate ratio (average 5.67 and 7.62 in digesters B-5-27 and B-5-28, respectively) (Tables S4 and S7). The average FSP values were 0.0026 and 0.0064 for digesters B-5-27 and B-5-28 (Table 6).
Several important characteristics of sulfate behavior in the lab-digesters were shown. Overall, the starting sulfate concentrations varied from 0.57 to 10.48 g L −1 for different lab-digester groups (Table S3). However, most starting concentrations exceeded 2 g L −1 (Table S3). Notably, Test 5 had sulfate concentrations exceeding 4 g L −1 for lab-digester groups B-5-24 through B-5-27 (Table S3). The H 2 S concentration in the biogas exceeded 1000 ppm for lab-digester groups B-5-24, B-5-26, B-5-28 ( Figure 1). However, this did not necessarily result in high FSP values (Table 4). Influent sulfate concentrations in AD above 2-5 g SO 4 2 L −1 can result in high concentrations of undissociated H 2 S and subsequent methanogenesis inhibition T A B L E 4 Final specific productions (FSP) of hydrogen sulfide from 45 digester groups in the six experimental studies. Note: Each digester group contains three individual digesters, except for blanks (marked with "*") that contains two digesters.

| Evidence of FeS precipitation in the presence of an anaerobic community
The t-tests revealed significant changes in Fe(II), TP, and sulfate concentrations over the digestion period in the precipitation experiment (Table 6). When iron reduction was tested, the Fe(II) concentrations only significantly increased in the treatments without inoculum (1B, p = 0.026) ( Table 6). Increases in Fe(II) concentrations suggested that iron reduction (Fe 3+ → Fe 2+ ) was occurring. Phosphorus precipitation occurred in treatments with and without inoculum. Decreases in PO 3− 4 concentrations indicated that FePO 4 precipitation was occurring (Fe 3+ + PO 3− 4 → FePO 4 ). For example, TP concentrations significantly decreased (3A and 3B, p < 0.001) in the phosphorus treatment with and without inoculum (Table 6). Sulfate significantly decreased in the FeS treatment with inoculum (2A, p = 0.003) without a significant decrease in Fe(II) (2A, p = 0.18) ( Table 6). In the control (Treatment 4, Table 6), which only contained inoculum, there were significant decreases in both Fe(II) and sulfate. Therefore, observed decreases in Fe(II) concentrations with decreases in sulfate concentrations suggested that an FeS precipitation was occurring (Fe 2+ + H 2 S → FeS + 2H + ). This was the evidence that Fe-S was mediated by the anaerobic community while FePO 4 precipitation was not.

| Modeling H 2 S production in batch digestion
In most digesters, the H 2 S production increased rapidly during the first 10 days and then leveled off. The gam models were the best fit for the H 2 S production data (Table S8). The gam model had the ability to fit the H 2 S production data even when the production curves varied considerably between tests ( Figures S1-S7). Labdigester group number B-3-15 experienced a drop due to one of the replicates being removed during the test and had lower goodness-of-fit values (Table S8). The coefficients from the gam model can be obtained to predict H 2 S production in future studies. The results benefitted from these modeling processes, which do not rely on comparative selection processes for defined models but, rather on smoothing processes with penalties (Wood, 2001).

| Processes that inhibit H 2 S production
The results in this study provided some insight into the role that Fe(II) had in curtailing H 2 S production. The FeS precipitation was an explanation for the negative correlation with the Fe(II) S −1 ratio and the final specific H 2 S productions ( (0.043-1.29 mL g VS −1 for digesters L-6-29 through L-6-45) (Table 4) had Fe(II):S ratios that were 0 (Table S5). This suggested that increased Fe(II):S ratios could have inhibited H 2 S production. The primary way would have been through an FeS precipitation process (Liu et al., 2017). If an iron precipitation process is favored in an anaerobic community, then ensuring there is an adequate concentration of soluble iron (0.20-16.0 mg L −1 Fe) (Parkin et al., 1990) can prevent H 2 S production while avoiding microbial inhibition. Furthermore, the results of the precipitation tests, namely the differences in Fe(II) concentrations between inoculum treatments (Table 6), revealed that the inoculum-only control (sample 4) did not undergo iron reduction (Fe 3+ → Fe 2+ ). However, the anaerobic community did not seem to prevent Fe-P precipitation because phosphorus concentrations were observed to decrease in both the absence and presence of an anaerobic community ( Table 6). The anaerobic community appeared to be necessary in the case of Fe-S precipitation as a decrease in sulfate concentration was not observed in the non-inoculum treatment (Table 6). In other words, iron reduction and iron-phosphorus precipitation seemed to occur without the presence of an anaerobic community, whereas FeS precipitation seemed to require an anaerobic community. Further investigations of FeS and Fe-P precipitation as an H 2 S inhibitor in AD systems is warranted.

| H 2 S production reveals potential digester failure
There have been few studies modeling H 2 S production kinetics from batch digester systems. The H 2 S production in one mesophilic batch study was 0.76 mL g VS −1 for DM and 2.23 mL g VS −1 for radish (Belle et al., 2015).
In that study, the H 2 S concentration initially increased and then rapidly decreased. This increase then decline was attributed to the depletion of sulfate concentration in the digester. In this study, there were additional "spikes" of H 2 S concentrations past day 10 (Tests 1, 5, and 6) ( Figure 1). Moreover, these spikes in H 2 S may indicate less favorable conditions for methanogenesis in the digesters. Increases in H 2 S concentrations have been shown to trigger a positive feedback loop with sulfate and acetate resulting in a subsequent decline in pH in chemostats (Vavilin et al., 1994). Furthermore, methanogens and SRB could exhibit oscillating behavior near system failure with even small shifts in pH (Vavilin et al., 1994). However, the oscillations stopped when pH was kept constant (Vavilin et al., 1994). Given the complexity of H 2 S production, typical Biochemical methane potential (BMP) models (Strömberg et al., 2015) are unsuitable for modeling H 2 S production. The environmental models used in this study can provide more flexibility for modeling and predicting H 2 S production. Further examination and validation of models for H 2 S production in AD is also warranted.

F I G U R E 2
The average fraction of free hydrogen sulfide (H 2 S) gas to TSS versus pH for the six lab-digester groups. TSS, total soluble sulfides.

| CONCLUSIONS
The following conclusions were drawn from this study: 1. Hydrogen sulfide production demonstrated a considerable variability, ranging from nondetectable to 1.29 mL g VS −1 . Higher H 2 S concentrations in the produced biogas occurred within the first 10 days of AD.
2. There were no significant correlations between the initial sulfate concentrations and the H 2 S FSP values. The commonly cited indicator of final specific H 2 S production, that is, TCOD:sulfate ratio, was not shown as a reliable predictor. This ratio was generally a better indicator of TCOD removal by SRB. 3. The most important influences on the final specific H 2 S productions were the initial Fe(II):S ratio and OP concentrations. Sulfate, phosphorus, and iron in the F I G U R E 3 The average final percent TCOD removal by methanogens and sulfate-reducing bacteria (SRB) for each lab-digester group. FD-T-LDG, field digester-test number-lab digester group number; TCOD, total chemical oxygen demand. anaerobic microbial community were important for the understanding of H 2 S production. 4. Iron reduction and iron-phosphorus precipitation occurred without the presence of an anaerobic community, whereas FeS precipitation required an anaerobic community. 5. The gam model could be used to model the H 2 S production for a variety of different substrate types with high R 2 values (>0.82). sulfate AUTHOR CONTRIBUTIONS Sarah E. Daly was involved in conceptualization; methodology; investigation; data curation; formal analysis; visualization; writing-original draft. Ji-Qin Ni was involved in project administration; resources; supervision; methodology; writing-review and editing. Both authors approved the final manuscript.