CO2 production, dissolution and pressure dynamics during silage production: multi-sensor-based insight into parameter interactions

Silage is a critical global feedstock, but is prone to aerobic deterioration. The dominant mechanism of O2 transport into silage remains unresolved. Here, multiple sensors tracked O2 and CO2, gas pressure (ΔP) between internal silage and ambient air, pH and silage temperature (Tsi) during the ensilage of maize and ryegrass. We report the first observation that CO2 produced from microbial respiration was partially dissolved in silage water, with evidence of negative or positive ΔP depending on the changing balance between CO2 production and dissolution. The ΔP < 0 reflected an apparent respiratory quotient (RQ) < 1. Net CO2 production was much greater in anaerobic fermentation stage than in initial aerobic phase or later aerobic feed-out phase. O2 transport into silage is intimately linked to the dynamics of net CO2, ΔP, microbial activity, pH and Tsi. These results suggested that both gas diffusion (based on Fick’s law) and advective transfer (Darcy’s law) play equally important roles in governing the complex temporal progression of inward and outward gas fluxes to and from the silage interior. Even though low pH suppressed microbial activity and supported aerobic stability, the negative ΔP increased the risk of O2 entry and aerobic deterioration during feed-out phase.

While the biochemical process of silage-making has been mostly characterized, determination of conditions within the developing silage has been challenging. For gas composition and concentration inside the silage, a common method has been to extract gas samples from the silage for later analysis using a laboratory gas chromatograph 8,18,19 . Despite sufficient accuracy of the gas chromatography method, two technical issues are encountered. First, sample extraction may result in additional O 2 entering the silage. Second, the method is time consuming and thus unable to track the dynamics of silage gasses. Measurement of pH within the silage during its solid-state fermentation has also been difficult to achieve 20 .
Despite these challenges, obtaining and characterizing these process level data will contribute to understanding and improving silage yield and quality. Here we present the first multi-sensor-based, simultaneous measurements of internal O 2 , CO 2 , ΔP, pH, and internal silage temperature (T si ) over the whole silage-making process. The major objective of this study was to reveal the previously unknown dynamic interactions of the key parameters governing silage development and quality.

Fresh plant materials (Test-A).
A unique suite of sensors was placed in a test silo (Fig. 1) such that naturally occurring changes could be monitored under realistic gas-tight conditions. Details are presented in Methods (below). The testing period consisted of the initial aerobic phase (about 3 h) and the anaerobic fermentation period (about 100 h), repeated three times with each of two fresh materials (maize and ryegrass).
The time courses of O 2 , CO 2 , pH, ΔP and T si (three independent replicates; Test-A) and relationships between ΔP and CO 2 in Test-A were obtained from fresh-chopped maize (Fig. 2a,c) and fresh ryegrass (Fig. 2b,d). During the aerobic period, the initial pH was 5.8 for maize and 6.1 for ryegrass (Table 1; Fig. 2a,b), which was conducive to high activity of microorganisms and plant proteolytic enzymes 5,8,18,21 . While O 2 was rapidly depleted to near zero in each test silo, the increase in CO 2 was minimal.
Most of the CO 2 was produced during the anaerobic period, rather than the aerobic period. During the anaerobic phase, O 2 remained depleted, but CO 2 increased 3-to 5-fold, accompanied by a pronounced decline of pH (maize to pH 4.6 and ryegrass to pH 5.2). The decline in pH lagged the depletion of O 2 in both cases (Fig. 2a,b), reflecting increasing anaerobic lactic acid production by lactic acid bacteria (LAB) 22 .
The rate of CO 2 production was linearly related to pH for both fresh maize (Fig. 3a, R 2 = 0.8902, p= 6.04 × 10 −30 ), and fresh ryegrass (Fig. 3b, R 2 = 0.7638, p= 6.04 × 10 −40 ). When pH declined to 4.6 in maize and 5.2 in ryegrass, CO 2 production and further changes in pH and ΔP were greatly reduced (Figs 2 and 3). This reflects suppression of microbial activity and further acid production by the reduced pH, despite the modest increase in counts of LAB observed in maize (Table 1), and the substantial increase observed in ryegrass.  ΔP exhibited a modest initial decline to a negative peak of −2.6 kPa in maize and −1.8 kPa in ryegrass, followed by a large increase to +45.4 kPa in maize and +19.1 kPa in ryegrass (Fig. 2a,b). Reasoning from the ideal gas law, observations of ΔP < 0 could result either from a loss of mass in the gas phase or from a decline in T si , or a combination of both. Similarly, ΔP > 0 could result from an increase of gas phase mass and/or elevation of T si . The increase in ΔP beginning as O 2 declined to near zero (Fig. 2a,b) closely tracked the time course of rising   CO 2 in each test silo, demonstrating the essential role of gas phase CO 2 production exceeding O 2 consumption ( Fig. 2a,b; upper panels). The CO 2 produced in the fresh maize system was considerably greater than that in the fresh ryegrass, reflected in both CO 2 and ΔP ( Fig. 2a,b). The relationship between CO 2 and ΔP was biphasic in both systems (Fig. 2c,d). Piecewise regression analysis revealed strong linear relationships in both phases for maize ( Fig. 2c; R 2 = 0.9220, p = 5.94 × 10 −40 for y 1 and R 2 = 0.9952, p = 7.12 × 10 −80 for y 2 ) and for ryegrass ( Fig. 2d; R 2 = 0.8856, p = 3.57 × 10 −61 for y 1 and R 2 = 0.9952, p = 3.70 × 10 −90 for y 2 ).
An increase in T si was observed in both systems during the aerobic phase ( Fig. 2a,b). This stabilized at an elevated level in maize, but declined to original values in ryegrass during the anaerobic phase. More heat was generated during aerobic than during anaerobic respiration 8,23 . At the end of the ensiling process (about 100 h), pH was low, ΔP was elevated, and all parameters were stable. The experimental system remained anoxic.

Ensiled plant materials (Test-B).
The time courses of O 2 , CO 2 , pH, ΔP and T si and relationships between ΔP and CO 2 in Test-B were obtained with three repetitions (Fig. 4). The aerobic period for ryegrass was longer than for maize. This may be partially attributed to the lower packing density and resulting greater porosity in the ensiled ryegrass, leading to more O 2 molecules entrapped but fewer microorganisms within the test silo 9 . The initial populations of LAB in ensiled ryegrass (10 6 Cfu g −1 ) and total-bacteria (10 6 Cfu g −1 ) were significantly lower than in ensiled maize (10 9 Cfu g −1 and 10 9 Cfu g −1 , respectively; Table 1). Molds were initially below detection threshold (10 2 Cfu g −1 ) in maize, while both yeasts and molds were below detection threshold (10 2 Cfu g −1 ) in ryegrass throughout the experiment ( Table 1). The fresh maize and the ensiled maize had similar (10 8 Cfu g −1 ) microbial populations, whereas the fresh ryegrass had higher (10 8 Cfu g −1 ) populations than the ensiled ryegrass (10 6 Cfu g −1 ; Table 1).
In contrast to the biphasic time courses observed in Test-A (cf. Fig. 2), the CO 2 concentration in both ensiled materials was lower than in the same material in the fresh state (cf. Figs 2 and 4) and ΔP only exhibited negative values throughout Test-B. The CO 2 concentration in the ensiled ryegrass was greater than in the ensiled maize (Fig. 4), even though the microbial counts in the ensiled ryegrass were lower ( Table 1). As a result, the decline to negative ΔP in the ensiled maize was greater than the ensiled ryegrass.
As with the fresh material, observations of ΔP < 0 reflected a decline in gas phase mass inside the test silo. The simultaneous measurements of O 2 and CO 2 over time indicate that the respiratory quotient (RQ, ratio of mole CO 2 evolution per mole O 2 uptake 24 ), was significantly below 1.0 for both ensiled materials (Fig. 5a,b; blue line). While aerobic carbohydrate metabolism exhibits a theoretical RQ = 1.0, i.e. 1 mole of CO 2 evolved per 1 mole of O 2 consumed 5,25 , in Test-B the observed RQ < 1.0 suggested the concurrence of CO 2 production by aerobic respiration and simultaneous partial CO 2 dissolution in interstitial silage water. This process was dominated by CO 2 dynamics since the solubility of CO 2 in H 2 O (39.2 mmol L −1 at 20 °C) is much greater than that of O 2 (1.38 mmol L −1 ) 26 .
The relatively low CO 2 concentrations in both ensiled substrates are likely to reflect several interacting factors. First, the produced CO 2 was partially dissolved in the interstitial silage water. Second, after the fermentation (anaerobic phase), the residual plant sugar in the ensiled materials was sufficiently low that the CO 2 produced from microbial respiration was minimal. In addition, microbial metabolic activity was further depressed by low pH 5,8,18,21 . The greater CO 2 in ensiled ryegrass than ensiled maize reflects the lower density and mass of silage, greater porosity and volume of O 2 entrapped, despite the similar O 2 concentrations, and less available silage water for gas dissolution. Therefore, more CO 2 was produced through aerobic respiration and less CO 2 was dissolved in the interstitial silage water.    (Fig. 5a) and for the re-ensiled ryegrass (Fig. 5b) during Test-B (aerobic phase). The reference line of RQ = 1 is the expected value from carbohydrate metabolism. The O 2 -CO 2 course below RQ = 1 (i.e. actual RQ < 1) in Fig. 5a or Fig. 5b reflects the CO 2 not detected by the sensor and apparently dissolved in the aqueous phase. Note the declining axes for O 2 .

Discussion
In Test-A, ΔP < 0 was observed over a quite short period (around 5 h) following introduction of fresh plant material. The transition of ΔP from negative to positive indicated that the CO 2 produced by microbial respiration rapidly exceeded the capacity for CO 2 dissolution, while high pH allowed sustained metabolic activity during the early stages of ensilage. Therefore, ΔP is a potentially useful indicator not only of CO 2 dynamics, but also of microbial activity. On the other hand, the solubility of CO 2 in H 2 O is a function of numerous interacting factors, including pH, partial pressure of CO 2 (Henry's law), ambient temperature (van't Hoff equation), interaction between internal gas pressure and temperature (the ideal gas law), sugar content in water and water content in the substrate 26-30, . Adding further complexity, dissolution of CO 2 in water is exothermic 27 , contributing to the observed increases in both T si and ΔP. Since dissolution of CO 2 in water is the outcome of the chemical reaction, , the decline of silage pH over the silage-making process resulted not only from the anaerobic fermentation of plant sugars such as glucose and fructose to lactic acid by LAB, but also from the dissolution of CO 2 in water. As a result, the time courses of net CO 2 production and ΔP are not readily simulated using current methods. Here we characterize the trajectory of variation in ΔP (ΔP = 0) → (ΔP < 0) → (ΔP > 0) → (ΔP < 0) throughout the silage production process. Observations of ΔP < 0 indicate apparent RQ < 1, reflecting the sequestration of CO 2 , and ΔP > 0 reflects the greater amount of CO 2 produced during anaerobic respiration. While other gases may contribute to these effects, CO 2 dominates because concentrations of H 2 , CH 4 and N 2 O in silage are very low compared to N 2 , O 2 , and CO 2 5,18,19,31-33 , and H 2 and N 2 exhibit low solubility in silage water. For a field-scale bunker silo during feed-out phase, the working face is exposed to air as the silage is unloaded for feeding to livestock. This results in re-ensiling of the remaining material and unavoidable aerobic deterioration in the surface zone as O 2 penetrates into the silage. With the two ensiled plant materials, Test-B provided an analog of these biochemical and physical processes of aerobic respiration and partial dissolution of CO 2 in the interstitial silage water. Here we infer changes in the dissolved CO 2 in the silage water, but carefully evaluate the role of gas phase advective transfer associated with CO 2 dissolution. Our unique sensor combination allowed us to follow O 2 and CO 2 , in situ, while tracking the variations of ΔP and T si simultaneously, with ΔP employed as an indicator of the net CO 2 dynamics based on the ideal gas law. For this reason, Test-B was conducted under gas-tight and constant volume conditions. Most real-world silage-making processes, including the maize bunker silo and ryegrass bale system used here, rarely exhibit this level of seal 5,17,34 . Our experimental design revealed the substantial and contrasting levels of ΔP generated by two common plant materials at the early ensiling stage and in the ensiled state. The rigorous vacuum sealing of our test silos allowed ΔP to vary over a range of −7.5 to + 45.4 kPa for maize and of −3.1 to 19.1 kPa for ryegrass.
To our knowledge, we provide the first evidence of the physical and chemical interactions involved in the production of CO 2 , its partial dissolution in silage water, and resulting negative and positive pressures, depending on stage of ensilage. The ΔP > 0 ('balloon') phenomenon has been reported previously [17][18][19]35 . One study 19 observed an increase in O 2 and N 2 but a decrease in CO 2 , measured at the end of the eight-day fermentation phase, and inferred a reversal of gas movement from efflux to influx, as a result of the reversal of ΔP from positive to negative. However, fundamental processes and causative factors were not explored. Our multi-sensor measurements provide insight into these parameters with essential interactions and dynamics, and the resulting relationships between net CO 2 production, O 2 consumption, and variations in ΔP, pH and T si .

Conclusions
In situ tracking of O 2 depletion, CO 2 production and dissolution, pH lowering-processes and ΔP dynamics has been realized during early and late phases of silage production, using multi-sensor techniques. CO 2 production was much greater from anaerobic than aerobic processes. The ΔP exhibited both positive and negative values, indicative of net values of CO 2 production and dissolution. For the two common silage materials tested here, the resulting ΔP is of the order of tens of kPa. Since the negative ΔP allowed more O 2 entry through the silo face, risk of aerobic deterioration in the feed-out phase could be potentially higher than anticipated. These interactions are diagnostic of the previously unknown behavior during ensilage and re-ensiling processes. These results suggest that both gas diffusive and advective transfer play equally important roles in governing the complex temporal progression of inward and outward gas fluxes to and from the silage interior. Future efforts to develop simulation of the ensilage process should incorporate both Fick's law and Darcy's law. The fusion of our data with improved process models will support development of improved silage management.

Methods
General information on the sensors used. Table 2 lists general information of all sensors used for this study. These sensors employ different principles, each providing sufficient temporal response (<1 min.) to meet the data sampling requirement. This level of performance is critical for tracking O 2 as it is converted to CO 2 , in order to interpret the resulting ΔP and T si dynamics. Cross calibration of the O 2 and the CO 2 sensors were carried out in advance of the experiment 36 .
During each test period the pH sensor tip (diameter 2 mm) was enclosed in a small piece of well-wetted porous plastic to improve hygroscopic contact with the test material. The porous plastic was rinsed with deionized water and then dried naturally. Initial pH was determined using a standard protocol for silage 37 , by extracting 25 g of maize or ryegrass material with 0.225 L deionized water in a glass vessel for 30 min. This provided an initial reference value for comparison with the initial value of the inserted pH sensor when encased in the porous plastic wetted with the same solution. In addition, the pH sensor was repeatedly calibrated against three buffer pH-solutions (ZB98PHPL4, pH = 4.0, ZB98PHPL7, pH = 7.0, and ZB98PHPL10, pH = 10.0; AHLBORN GmbH, Germany) at 20 °C prior to and after the experiment.
Experimental test silo. Two PVC cylinders (inner diameter: 370 mm, height: 600 mm, wall thickness: 15 mm) together with a lid (diameter 400 mm, thickness 20 mm) were used as the experimental silos (Fig. 1). To make the cylinder gas-tight, a rubber membrane (diameter: 400 mm, thickness: 10 mm) was affixed by a layer of vacuum grease to the inside of the lid, which was secured using six equidistantly-spaced fasteners. Five access ports through the lid and membrane allowed connection of the sensors inside the cylinder to a data logger (ALMEMO-2890-9, 9-channel, AHLBORN GmbH, Germany) placed outside. After each cable was placed through a hole, the space around the cable and all cable terminals were sealed with vacuum grease. The data logger operated at a sampling interval of 3 min throughout the experiment.

Test-A.
This was an analogue to the ensiling phase using fresh chopped whole shoots of maize (Zea mays) including the cobs (dry matter content, DMC of 27%) and fresh-cut ryegrass (Lolium perenne; DMC of 45%). The resulting biomass was finely chopped and packed to high bulk density (BD) in separate cylinders (maize: 625 kg m −3 , ryegrass, 425 kg m −3 ) using a hydraulic ram (360 mm diameter circular foot), in approximately six layer increments to ensure uniform density. Five holes (diameter 10 mm, depth 100 mm) through the top layer of silage, aligned with the holes in the lid, were drilled using a hole-saw for the installation of O 2 , CO 2 , ΔP, pH and T si sensors.

Test-B.
Test-B used previously ensiled maize with DMC of 35% and ryegrass with DMC of 43%. Ensiled maize (harvested and ensiled autumn 2015) was obtained from 20 cm behind the freshly cut silage face in a concrete bunker silo (40 × 6 × 3.5 m), located at the Frankenforst research farm (University of Bonn, Germany). Ensiled ryegrass was obtained from a silage bale (harvested and ensiled autumn 2015), which was stored outside, wrapped in four layers of plastic film (thickness: 0.025 mm). The maize and ryegrass silages were re-ensiled in the test silos, and packed with the hydraulic ram to a high BD of 625 kg m −3 and 425 kg m −3 , respectively. The sensors were installed as for the fresh materials. To achieve similar temperatures for the two experiments and to minimize ambient temperature fluctuations, both tests (each with three repetitions) were conducted in a large insulated workshop in the summer of 2016.  Table 2. Sensor specifications.