ARTICLE IN PRESS. Sensitivity Analysis and Application of a Dynamic Simulation Model of Nitrogen Fluxes in Pig Housing and Outdoor Storage Facilities

Transcription

1 Biosystems Engineering (2007) 96 (4), doi: /j.biosystemseng IT Information Technology and the Human Interface ARTICLE IN PRESS Sensitivity Analysis and Application of a Dynamic Simulation Model of Nitrogen Fluxes in Pig Housing and Outdoor Storage Facilities P. Berthiaume 1 ; M. Bigras-Poulin 1 ; A.N. Rousseau 2 1 Faculté de médecine vétérinaire, Université de Montréal, CP 5000, St-Hyacinthe, Québec, Canada, J2S 7C6; of corresponding author: philippe_berthiaume@phac-aspc.gc.ca 2 Centre Eau, Terre et Environnement. Institut national de la recherche scientifique Université du Que bec, INRS-ETE, 490 de la couronne, Québec, Canada, G1 K 9A9; alain_rousseau@ete.inrs.ca (Received 20 January 2006; accepted in revised form 12 December 2006; published online 23 February 2007) This article presents the sensitivity analysis of a deterministic model proposed by Berthiaume et al. (2005) for the prediction of daily nitrogen concentrations N conc in kg [N] t 1 [slurry] and loads N load in kg [N] inside buildings and storage facilities at the production site scale. This model makes use of many parameters and therefore, it is important to evaluate the impact of each of these. Identification of those parameters which most affect the output values allows for the rationalisation of resources when establishing a sampling protocol for determining more precise parameter values. The most important parameters identified were the proportion of proteins in feed P, the temperature of the slurry T, the ph of the slurry h, and, the air speed over slurry v. It therefore confirmed the already acknowledged high importance of feed content and methods of distribution information that can be easily obtained from producers and, thus, can be used in the determination of regional amounts of nitrogen produced by swine production systems (e.g, municipality, county or watershed level). In addition, this sensitivity analysis confirmed that some characteristics that are seldom known to producers slurry ph and air speed over slurry are also of great importance. Finally, two sets of simulation scenarios were used to illustrate potential applications of this model as a management tool and to further demonstrate its coherent behaviour over different sets of parameter values. r 2007 IAgrE. All rights reserved Published by Elsevier Ltd 1. Introduction In recent years, intensive pig production has been associated with nitrogen non-point sources of water pollution resulting from spreading of slurry in excess of crop requirements. This situation has prompted experimental studies aiming at a better understanding of the role of feeding, genetic, slurry management and building characteristics on nitrogen loads (Canh, 1998; Canh et al., 1997, 1998a, 1998b; Dourmad and Henry, 1994; Dourmad and van Milgen, 1998; Portejoie et al., 2004; Portejoie et al., 2003; Quiniou et al., 1994). It has also inspired the development of mathematical equations for these factors but these have not been integrated simultaneously at the production site scale (Aarnink and Elzing, 1998; Dourmad et al., 1992). A deterministic mathematical model was recently developed by Berthiaume et al. (2005) to predict the effect of these factors at the production site level, and therefore facilitate management. This model allows for the prediction of daily nitrogen concentrations N conc in kg [N] t 1 [slurry] and loads N load in kg [N] inside buildings and in storage facilities. Although the model represents a particularly well-adapted tool intended to take into account the impact of major pig farming characteristics at the production site level, it necessitates many parameter values; hence the need to evaluate theimpactofthelackof precision associated to each of these. A sensitivity /$ r 2007 IAgrE. All rights reserved Published by Elsevier Ltd

2 456 P. BERTHIAUME ET AL. Notation A area of emitting ammonia solution, m 2 B output value of a simulation B output value when simulated using reference values for every parameters b 1 b 12 coefficients C dej daily dejection volume correction coefficient C tan total ammonia nitrogen concentration in slurry or urine, mol m 3 E daily ingested digestible energy, MJ day 1 sow 1 F wasted proportion of feed wasted, % f un-ionised ammonia-nitrogen fraction in solution H Henry s constant h ph of slurry h e effective corrected ph of slurry h u ph of urine k mass transfer coefficient, m s 1 N conc total nitrogen concentration, kg [N] t 1 [slurry] N I ingested nitrogen, g pig 1 day 1 N load total nitrogen load, kg [N] N R nitrogen retained, kg animal 1 day 1 N U proportion of total nitrogen excreted in urine, decimal N V ammonia nitrogen volatilisation, mol [N- NH 3 ]s 1 n lit litter size, number of piglets P proportion of proteins in feed, % p parity number correction (+05, 06, 17, 25. respectively, for parity 1, 2, 3 and 4 or more), g day 1 sow 1 Q daily volume of slurry produced, m 3 R corr daily precipitation correction coefficient R evap daily evaporation correction coefficient relative sensitivity, dimensionless S R S S scenario sensitivity, % T temperature, K v air velocity, m s 1 W ADG average daily weight gain, kg day 1 W evapo daily water evaporation correction coefficient, dimensionless X p the reference value for the i th parameter x p the value of the i th parameter Z cf 088, correction factor, dimensionless a ratio of the value of the parameter used for the sensitivity assessment to the standard value of the same parameter DB difference between B and result with the studied parameter modified Dx p corresponds to a positive increase (5%) in the value of the targeted parameter Subscript D dry sows DGL the combination of dry sows, gestating sows and lactating sows dej faeces and urine falling on the floor F floor G gestating sows L lactating sows lagoon slurry lagoon p parameter pig growing-finishing pig pen pen tank concrete slurry tank TS temporary storage structure UF area receiving slurry under the slatted floor analysis, that is an analysis focussing on how variations in the output of a model can be apportioned to the different sources of variation and how the model depends on the information fed into it (Saltelli et al., 2000), was performed. The identification of those parameters exerting the most significant effect should allow for the rationalisation of resources when deciding on a sampling protocol for getting the parameter values. Many important characteristics in swine production such as genetics, feeding efficiency, feed content, feed distribution methods, slurry management systems and building characteristics are included in the aforementioned mathematical model. The main objectives of the sensitivity analysis presented in this paper were: (1) to ascertain the relative impact of each parameter on model outputs and (2) to study the behaviour of the model in relation to the modification of the input parameter values. The presentation of the sensitivity analysis constitutes the first section of this paper while the second section aims at illustrating the usefulness of the model as a decision support tool through simulation of two potential applications scenarios.

3 DYNAMIC SIMULATION MODEL OF NITROGEN FLUXES IN PIG HOUSING Methods 2.1. Sensitivity analysis Two different methods of sensitivity analysis were used. The first method, the scenario sensitivity S S was used in order to identify the most significant parameters. The second method, relative sensitivity S R, was used to test whether the sensitivity remained constant or not (i.e., varied) over the domain of values for the most important parameters. In both cases, sensitivity analyses were performed using a one at-a-time approach that is the impact of changing the value of each parameter was evaluated sequentially while the values for all other parameters remained constant and set to standard or reference values. The standard values were determined using scientific literature and the Quebec pig farm census (Gilbert et al., 1998) or obtained from the opinions of experts in pig production when no values were available Scenario sensitivity analysis A simulation run which uses the standard values for all parameters, was defined as the control experiment (Saltelli et al., 2000). For each parameter, results of the simulations using extreme values (high and low) were compared with results from the control experiment and the magnitude of residuals was assessed. Results from the comparisons are presented in percentage points for all the parameters where the absolute value of the sensitivity was found to be greater than 1% (rounded results). The scenario analysis was performed for a reproduction site and a growing-finishing site. The formula used to obtain the results in the tornado graphs is given by Eqn (1), where: S S,p represents the scenario sensitivity in % for the p th parameter; DB p represents the difference between output value obtained when simulating using reference values for every parameter and the result using the modified value of the studied parameter p; B is the reference output value obtained when simulating using reference values for every parameter. S S;p ¼ 100 DB p (1) B Relative sensitivity Relative sensitivity S R was performed to verify whether the model was equally sensitive to any variation of parameter value over the plausible domain of values that can be assumed by the targeted parameters. This sensitivity is defined by Eqn (2), (Robert et al., 1992), where: S R,p expresses the degree of variation in the output resulting from 1% unit change at a specific level a, the ratio between the value of the parameter x p and X p ; where x p is the value of the p th parameter; X p is the reference value for the p th parameter; Dx p corresponds to a positive increase of 5% in the value of the targeted parameter; DB p corresponds to the difference in the simulation output resulting from using x p plus 5% and x p ; a corresponds to the ratio between the value of the parameter p and the reference value for the same parameter p; Da correspond to the difference between the value obtained for a using the parameter p and the value of a obtained using the parameter p plus 5%; and B is the output result using reference values for all parameters S R;p ¼ DB p B ¼ DB p (2) Dx p X p BDa where the specific level a is a ¼ x p X p (3) Unlike scenario sensitivity, where simulation results were compared with a reference simulation, the relative sensitivity compares the result of each simulation with that of another simulation using only a slightly different value of the studied parameter. Therefore, two distinct simulation runs were necessary to obtain each of the S R estimates. It should be noted that the resulting S R coefficient is dimensionless and independent of the order of magnitude of DB and Dx p. This form of relative sensitivity is considered a local analysis, since it provides information on the effect of variation around a specific value of the parameter but also allows exploration of this effect in as many locations (or points) as desired. The minimum, central and maximum values used for the different parameters were the same as for the scenario sensitivity analysis Simulation characteristics For both types of sensitivity analyses, the general characteristics of two existing production systems were simulated in order to take into account the structural differences in design between reproduction sites and growing-finishing sites. Two output variables were considered: (1) N load in the exterior storage tank or lagoon; and (2) N conc in the exterior storage tank or lagoon. On the reproduction site, the reproduction building was the only animal production system of the site. It consisted of a single building with a capacity of at least 800 sows. The growing-finishing system simulated consisted of three joint buildings localised on one unique production site. The three buildings of this production system totalised 2250 pigs and used a specific lagoon to store slurry. There were also two other buildings of at the site but these were not included in the simulation as they were using a different lagoon for

4 458 P. BERTHIAUME ET AL. Table 1 Parameter values used for the theoretical simulation of the presence or absence of a roof and presence or absence of an acidifying diet Scenarios Air velocity over stored slurry (v tank ), m s 1 Evaporation coefficient, (R evap ) Precipitation coefficient, (R corr ) ph of slurry in tank (h tank ) and under floors (h UF ) +Roof +Acidifying diet Roof +Acidifying diet +Roof Acidifying diet Roof Acidifying diet Table 2 Scenario sensitivity for the maternity site for parameters having an impact of more than 1% Parameter Impact on N concentration, % Impact on N load, % Use of minimum value Use of maximum value Use of minimum value Use of maximum value Feed wasted, F wasted,g ph of slurry in tank, h tank ph of urine, h u,g ph of slurry under floor (h UF ) h UF,D h UF,G h UF,L Protein in feed (P) P D P G P L Daily slurry production (Q) Q G Q L Correction coefficients Rainfall, R corr Evaporation, R evap Temperature (T) T F,G T UF,D T UF,G Air velocity (v) v F,G v UF,D v UF,G D, G, L, common subscripts for dry sow, gestating sow and lactating sow. slurry storage. The duration of simulation was 365 days beginning May 1 st for both sites. A daily time step was used. Parameter values are given in Appendices A and B for the reproduction site and the growing-finishing site, respectively. the Centre de Référence en Agriculture et Agroalimentaire du Québec (CRAAQ, 2006) and values from samples taken in the slurry contained in the tank for the reproduction site and in the lagoon for growingfinishing system Comparison of simulated concentrations with literature values A comparison was made between N conc estimated using the mathematical model, the values proposed by 2.3. Illustration of potential applications Two potential applications are presented to exemplify the ability of the model to generate useful information for producers at the production site scale. In both of

5 DYNAMIC SIMULATION MODEL OF NITROGEN FLUXES IN PIG HOUSING 459 these theoretical experiments, the production site characteristics were based on existing sites that were visited. The number of animals and the feed content in protein, for each types of animals (lactating, dry or gestating sows, growing pigs) were obtained directly from the producers. Visits to the production sites were made to obtain values for the following factors: floor dimensions of the different rooms, area of slurry under floors, dimensions of temporary storage structure and slurry tank, type of floor in the different rooms, and type of distribution system for feed and water. Values for the other necessary parameters were estimated from the literature. Key parameter values are presented in Appendix C. Impact of acidifying diet and roof over slurry tank in reproduction site The first application consisted in a theoretical experiment for which four simulations were performed. Each simulation was made up of one realisation of any combination of the use or not of an acidifying diet and the presence or not of a roof over the slurry tank for an existing reproduction site of 830 sows with main characteristics corresponding to the system studied in the sensitivity analyses. The use of a roof cover over the slurry tank was taken into account only through precipitation, air velocity immediately over stored slurry and evaporation, although it is clear that it constitutes a simplification. In this experiment, when simulating the slurry tank without a roof, rainfall and evaporation values obtained from the nearest meteorological station were used. To simulate the effect of the presence of a roof over the slurry tank, daily evaporation values were multiplied by a correction coefficient R evap of an arbitrary value of 05 and rainfall accumulation in the slurry was considered to be null (values were multiplied by a correction coefficient R rain of 0). Both R evap and R rain were kept constant during the entire simulation. Two different values for air velocity over the slurry tank v tank were also used to account for the presence or absence of a roof cover. The v tank parameter was given a constant value of 003 m s 1 in presence of a roof cover and a constant value of 277 m s 1 (approximately 10 km h 1 ) in the absence of one. To simulate the use of an acidifying diet the parameters h UF and h slurry both were given a value of 6 when simulating the use of the acidifying diet and a value of 75 for regular diet. These ph values were considered to represent a realistic impact of the use of an acidifying diet. Both h UF and h slurry were kept constant during the entire simulation. The values for these scenario parameters are presented in Table 1. The key parameter values common to all four scenarios are given in Appendix C. Implementation of a multi-phase feeding system in a growing-finishing site For the second application, the model was used to study the impact of switching from a two-phase feeding system (two diets) to a multi-phase feeding system (10 different diets). In this experiment, a growing-finishing site of approximately 2250 pigs, with characteristics similar to those used for the sensitivity analyses was simulated. In the two-phase feeding system, the proportion of protein P took values of 17% for pigs with a weight W B ranging from 25 kg to less than 60 kg and 15% for pigs of kg. For the multiple phases, the value of the parameter P was modified from 17% to 15% in a total of 10 consecutive steps during the growing period. Other key parameters are given in Appendix C. 3. Results 3.1. Scenario sensitivity As mentioned earlier, the sensitivity of the model was assessed both on the basis of the variables N load in slurry and N conc in slurry. Sensitivity results are presented in Table 2 for the reproduction site and in Table 3 for the growing-finishing site. Only the parameters for which the sensitivity was superior to 1% are presented. It can clearly be seen in Tables 2 and 3 that for some parameters, the model shows high sensitivity in both reproduction and growing-finishing system. The most significant of these parameters are: the proportion of proteins in feed P, the ph of slurry h and, the temperature of slurry T. In the reproduction system, the number of gestating sows was substantially larger than that of dry or lactating sows. For this reason, the impact of modifying characteristics related to gestating sows was generally higher than that for the other types of sows. For gestating sows, the use of high (P G ¼ 18%) and low (P G ¼ 12%) proportions of proteins caused a 15% increase and a 17% decrease in the value of N load in the tank, respectively, when compared with the results obtained using the standard value (P G ¼ 147%). In the simulated growing-finishing system, the high (P pig ¼ 18%) and low (P pig ¼ 16%) values used for the parameter representing the amount of proteins induced an increase of 22% and a decrease of 22%, respectively, when compared with the prediction using the standard

6 460 P. BERTHIAUME ET AL. Table 3 Scenario sensitivity for the growing-finishing site for parameters having an impact of more than 1% Parameter Impact on N concentration, % Impact on N load, % Use of minimum value Use of maximum value Use of minimum value Use of maximum value Area of dejections, A dej Proportion of protein in feed, P Daily slurry production, Q Evaporation correction factor, R evap Rainfall correction factor, R corr Temperature of lagoon, T lagoon Temperature under floor, T UF ph of slurry under floor, h UF ph of slurry in lagoon, h lagoon Air velocity on floor, v F Air velocity under floor, v UF value (P pig ¼ 17%). The effect on the value of N conc concentration was similar. It was commented by Lenis (1989) that a reduction of 2 percentage units in crude protein content in starter diet, grower diet and finishing diet could result in about 25% reduction in nitrogen excretion per pig. To make a comparison with the value of Lenis, we can use the result obtained with the maximum value (P pig ¼ 18%) and find a reduction of 36%. But in the case presented here, this reduction corresponded to the impact in the slurry, for the site defined in the sensitivity analysis and not only in the nitrogen excreted by pig. Inside buildings, the ph value of slurry temporarily stored under slatted floors h UF, had an important effect on the values of N load and N conc. For the reproduction, the use of the maximum value of ph for the slurry in an under floor pit (h UF ¼ 8) of the gestating room resulted in a 25% decrease in the value of N load compared to the standard value while the use of the minimum value of ph (h UF ¼ 5) resulted in a 24% increase. For the growing-finishing site, the upper value for the ph parameter (h UF ¼ 8) caused a decrease of 21% in the N load value while the use of the lower value (h UF ¼ 5) caused an increase of 13% when compared with the standard value (h UF ¼ 74). Outside the buildings, the impact of the ph in the slurry tank h tank and in the slurry lagoon h lagoon were very asymmetrical and of a lower magnitude than the impact of under floor slurry ph for both the reproduction system and the growingfinishing system. For the reproduction site, the value of N conc in the tank was highly affected by the amount of precipitation but the value of N load was not. The sensitivity of the model to the variation in precipitation, obtained through the use of the parameter R corr, ranged from ¼ +62% to 55% when using the minimum (R corr ¼ 05) and the maximum value of R corr, respectively (R corr ¼ 15). The impact of R corr was much more important in the growing-finishing pig system than for the reproduction system, with an impact on the nitrogen concentration ranging from +184% to 134%. This higher impact reflects the much higher area of the slurry lagoon when compared to concrete tank used on the reproduction site. The low impact on nitrogen load reflects the fact that direct dilution is the main causing factor. In the reproduction building, a maximum temperature value (25 1C) and a minimum (15 1C) value for the slurry under the floors of gestating sows resulted in a decrease of 110% and an increase of 88% in the value of N load in the slurry tank, respectively. In the growingfinishing buildings, the high temperature for the slurry stored under the floor caused a decrease of 78% in the value of N load while the lowest value caused an increase of 53% when compared with the standard value. The higher impact of temperature in the reproduction building follows from the fact that the area of emitting surface under the floor was more significant since all the space under the gestating room was used for temporarily stored slurry while in the growing-finishing building only the slatted area (one-third of the total area) was used. For the gestating sows, the use of maximum and minimum values for air velocity under floors parameter v UF resulted in a decrease of 59% and an increase of 82% in the value of N load, respectively. For the growing-finishing pigs it was found that the use of maximum and minimum values of the v UF parameter resulted in a decrease of 16% and an increase of 19% of N load and N conc in the lagoon, respectively. The air velocity above slurry under floor is the parameter for which the highest disparity

7 DYNAMIC SIMULATION MODEL OF NITROGEN FLUXES IN PIG HOUSING 461 was found between the growing-finishing site and the reproduction site. For the reproduction site, the use of the lowest value for daily volume of dejections from the gestating sows Q G caused a 59% augmentation in the value of N conc but a reduction N load by 24%. On the opposite, the use of the largest value caused a reduction of 24% of N conc and a 14% augmentation of N load Relative sensitivity Relative sensitivity was evaluated for all parameters but only those for which an important effect was observed are presented and discussed here. Results of the sensitivity analysis for N laod and N conc are presented in Figs 1 and 2 respectively. Unless otherwise noted, results were similar between the reproduction site and the growing-finishing site. It should also be noted that the more a value of S R deviates from zero, either negatively or positively, the greater the sensitivity. Relative sensitivity S R Ratio α of the value tested to the standard value of the same parameter Fig. 1. Relative sensitivity S r of the model when predicting total nitrogen load N load for slurry in concrete tank on the reproduction site:, ph of slurry in the tank h tank ;, ph of slurry under floors h UF ;, ph of urine h U ;, percent of wasted feed F wasted ;, amount of proteins in feed P ;, temperature of slurry under floors T UF ; slurry under floor v UF ;, air velocity over daily volume of slurry produced per pig Q Relative sensitivity S R Ratio α of the value tested to the standard value of the same parameter Fig. 2. Relative sensitivity S r of the model when predicting nitrogen concentration N conc for slurry in concrete tank on the reproduction site:, ph of slurry in the tank h tank ;,ph of slurry under floors h UF ;, ph of urine h U ;, percent of wasted feed F wasted ;, amount of proteins in feed P;, temperature of slurry under floors T UF ;, air velocity over slurry under floor v UF ; daily volume of slurry produced per pig Q;, rainfall coefficient R corr (i) Protein content in feed It can be seen in both Figs 1 and 2 that S R values for the parameter P are positive for all three values of a are distant from zero and show a slight increase in sensitivity when going from the minimum to the maximum value of a. This result indicates that the impact of a slight increase in the protein content of the diet produces an important positive effect at any of the three tested values of a. This result could be explained by the fact that nitrogen excretion is estimated from a subtraction of the calculated nitrogen retained by the animals. Therefore, once the ingested nitrogen exceeds the retained quantity calculated, any further nitrogen goes into excreted nitrogen. The slope does not equal zero since nitrogen retention is not calculated entirely independently of nitrogen ingestion as can be seen in Eqn (D1) in the Appendix D. Model behaviour also seems to be coherent with animal biology since, when the needs of the animal for amino acids and energy are met, excess nitrogen should go into excrements.

8 462 P. BERTHIAUME ET AL. (ii) ph value for slurry stored under the floor The value of S R for the parameter h UF is always negative. At the minimum value of a, S R is close to zero, which means that the positive increments of this parameter (Dx ¼ 5%) near the minimum value of h UF have very little impact on both the value of N load (Fig. 1) and the value of N conc (Fig. 2) but has a strong impact near the reference and the maximum value. (iii) ph value for slurry stored in the outdoor slurry tank For the ph of slurry in the tank h tank, S R is negative at each of the parameter values tested, with a smaller departure from zero for the minimum value, a maximum departure value around the median value (a ¼ 1) and an almost null departure from zero at the maximum value of the parameter. (iv) ph of urine puddles on the floor The value for the ph of urine h u is negative at all three values of a. Therefore, an increase in ph value causes a decrease in the value of N load and N conc at all ph values. The sensitivity for this parameter increases with higher values of ph. It should be noted that the range of a is very limited and corresponds to the very limited range of urine puddle ph values. This parameter represents the ph of urine on the floor which has already chemically reacted with enzyme urease as opposed to that of the normal animal s urine ph, which would be closer to a value of seven. The behaviour of this parameter is governed by Eqn (D4) predicting the equilibrium between non-ionised ammonia and ammonium. Both extreme values produced a values near unity because of the relative lack of variation of this parameter. (v) Temperature of slurry stored in under floor slurry pit The values of S R for temperature of slurry stored under floor T UF is negative for all three points. The slope is negative, indicating that near the maximum value of a the impact is greater than near the standard value or the minimum value. (vi) Air velocity above slurry stored under floor The value of S R for the parameter v UF is negative at all three points, meaning that a positive Dx always reduces the value of N load and N conc. Furthermore, the significance of this effect is greater for the lower value of a than for the maximum value. (vii) Precipitation A significant sensitivity to changes in the amount of precipitation, in mm day 1, tested through the use of a dimensionless precipitation correction coefficient R corr was observed for N conc but not for N load indicating that it is almost solely an effect of dilution. The S R value is almost a constant negative value. (viii) Volume of excretions A parameter for the average volume of excretions Q in m 3 was used to simulate the impact of water reduction systems (e.g., wet feeding). For this parameter, S R is different when considering either N load or N conc as the decision variable. For N conc, S R is negative at all three values and otherwise positive for N load Comparison of simulated values with mean values taken from literature Table 4 shows that for the specific cases of the three systems simulated, the model produced a better estimation of the total nitrogen concentration than the mean values from literature. In all cases, values reported in the literature were substantially larger than measured concentrations Potential applications scenarios For the first application scenario, it can be observed from Fig. 3 that the impact of using an acidifying diet is very important. In the case of the roofed tank, the use of an acidifying diet caused a 39% increase of the final value of N conc when compared to no use of an acidifying diet. Without roofing, the impact is even more significant with an increase of 49% in the value of N conc. Comparatively, the impact of roofing appears much less important. In the presence of an acidifying diet the inclusion of the roof caused an increase of 12% while in absence of an acidifying diet, the inclusion of roofing caused an increase in the value of N conc of 21%. For the second application scenario, it can be seen from Fig. 4 that going from a two-phase diet to a multiphase diet has a significant impact on the decrease in the value of N conc for the slurry stored in a lagoon (decrease of 10%).

9 DYNAMIC SIMULATION MODEL OF NITROGEN FLUXES IN PIG HOUSING 463 Table 4 Comparison of simulated values with mean values form literature for a 365 days simulation Site Concentrations kg [N] t 1 [slurry] Simulated (departure from measured data) CRAAQ (departure from measured data) Measured data (s.d.) y Reproduction (+14%) (+42% to+63%) (012) Growing-finishing, system (+8%) (+35% to+62%) (033) Growing-finishing, system ( 2%) (+133% to+35%) (042) Centre de Re férence en Agriculture et Agroalimentaire du Québec (CRAAQ, 2006). y s.d., standard deviation. Total nitrogen concentration, kg [N] t 1 [slurry] Days from 1 May 2002 Fig. 3. Impact of the use of an acidifying diet and roof covering of slurry tank on nitrogen concentration N conc :, with cover but without acidifying diet; with cover and with acidifying diet;, without cover and without acidifying diet; without cover but with acidifying diet Total nitrogen concentration, kg [N] t 1 [slurry] Days from 1 May 2002 Fig. 4. Impact of two different diets on nitrogen concentration N conc in stored slurry:, two-phase feeding system (two diets); 10-phase feeding system (10 diets) 4. Discussion 4.1. Sensitivity analyses Scenario sensitivity analysis allowed us to identify parameters with a large effect the value of the output variables N load and N conc. Proteins in diet and ph of slurry are two of the most important parameters in that regard. High sensitivity of the model to both parameters is coherent with results already published (Lenis, 1989). To reduce the amount of protein ingested in the diet, different approaches have been proposed in the literature. The use of phase feeding could allow for a better adjustment of the content of proteins given to pigs by taking into account the specific needs corresponding to their mass and physiological status (Everts, 1994; Guillou et al., 1993; Henry & Dourmad, 1993; Henry et al., 1979; Jean dit Bailleul et al., 1997). The use of synthetic amino acids favors equilibrium and reduces excess of nitrogen in protein (Henry, 1993). Using a formulation algorithm including an environmental parameter rather than solely optimising on the lowest price is also a possible way to reduce excess amount of excreted nitrogen (Jean dit Bailleul et al., 1997). The observed sensitivity to the ph value is coherent with literature reporting the important effect of lowering ph on ammonia volatilisation in animal production buildings (Aarnink & Elzing, 1998; Canh et al., 1998a). The usual ph of pig slurry is slightly larger than seven (7) and according to Aarnink and Elzing (1998) the surface of the slurry is approximately 11 unit of ph greater. The impact of change in the ph value comes from the fact that it affects the equilibrium between ammonium and ammonia. For example, at a temperature of 20 1C, and a ph value of 65 for the surface layer of the slurry of 65, the unionised fraction is of while at a ph value of 85 this fraction is 002. Many

10 464 P. BERTHIAUME ET AL. experimental trials have shown the high potential of diet in reducing ph value of slurry without negative effect on animal growth. This has been done for example with acidifying salt in diet (Canh et al., 1998c). The important asymmetric effect of the ph of slurry stored in the outdoor storage structure that was observed for both sites is the result of a constraint included in the model to consider the limiting aspect of diffusion when surface volatilisation is large due to significant air velocity caused by wind over an open storage structure. The fact that the model is so sensitive to the ph value of slurry, and, that to date the prediction of ph value of slurry is not included in the model indicates a clear need for more research on this topic. The increase of volatilisation with temperature has already been reported by many authors. In an experiment with weaned piglets, Aarnink et al. (1993) reported an increase of 12% in volatilisation for 1 1C of slurry temperature augmentation. In the case of field applied manure, Sommer and Hutchings (2001) depict that as much as 50% of the total amount of nitrogen is volatilised as ammonia at a temperature of 30 1C compared with 35% at a temperature of 25 1C. Muck and Steenhuis (1982) presented graphically a relation showing a very important positive impact of temperature on nitrogen volatilisation from urine when temperature was over 10 1C. In addition to the impact on the rate of volatilisation, the results obtained in the sensitivity analysis illustrate as well the impact of changing the temperature value on the nitrogen load in the slurry at the end of the slurry accumulation period. Hence, the results can not directly be compared quantitatively with the articles cited but altogether they all confirm a significant positive impact of temperature. Furthermore, our modelling approach is similar to the approach reviewed by Ni (1999) to simulate ammonia volatilisation. Indeed his review summarised that in 30 publications of mechanistic models, temperature affects both the convection mass transfer coefficient and the dissociation constant of ammonia into ammonium. The observed impact of a change in air velocity over slurry was much less important in growing-finishing because of the much smaller area of slurry under the floors; stored slurry occupied almost the entire area under the gestating sows room, while only approximately one-third of the space in the growing-finishing pens. The daily volume of dejection produced Q had a different impact on the value of N conc than on that of N load. This can be explained by the fact that when the value of Q augments, the dilution becomes more important, hence a lower concentration, but at the same time it also reduces the volatilisation since this process is partly concentration driven. The reduced importance of the ph of urine comes from the fact that all simulated systems where using slatted floor and that only a small part of the urine remained on the slates. In a different type of production system, this should be reevaluated. Relative sensitivity analysis allowed us to identify those parameters for which the sensitivity is highly dependent on value of the parameter within in the possible domain of values. The parameter for which this effect is by far the most important is the ph of slurry. It seems that in a context where acidification of slurry ph would occur, the exact value would not be critical (Figs 1 and 2). Nonetheless, in actual context where slurry ph is more neutral or alkaline, there should be an important impact of the uncertainty on the value of this parameter. The modification of the ph value of slurry has a strong effect on the value of the equilibrium between unionised ammonia nitrogen and ammonium ions. In the vicinity of the minimum value of this parameter, however, almost all of total ammonia nitrogen exists in the form of ammonium ions (Eqn (D4)), and a small increment in the ph value does not change this equilibrium, which strongly favours the ammonium ion. Therefore, not much free ammonia is available for volatilisation. At the value of a ¼ 1, sensitivity is strongly negative (S R o 3) which means that any incremental change in the value of the parameter around the standard value (a ¼ 1 at ph ¼ 74) will have a strong negative effect on nitrogen load and nitrogen concentration. Near this value of ph, Eqn (D4) is very sensitive to an increase of the parameter value. As an example, raising the value of ph from 74 to 77 [equivalent to effective ph values h e of 85 and 88, respectively, in Eqn (D4)] produces an increase in the non-ionised fraction by a factor of two. This, combined with the fact that the amount of volatilisation near this ph value is sufficiently important to have a major impact on the total nitrogen load explains the high negative value of S R. For the higher value of a represented by the third point on the graph, the diminution of the significance seems to come from the fact that volatilisation is important enough to act fast enough on total nitrogen concentration and partly counterbalance the effect of the equilibrium now going further in the direction of free ammonia (although still in favour of the ammonium). The behaviour of the model is coherent with observations from other authors (Aarnink & Elzing, 1998) and actual chemical knowledge. In the case of ph of slurry stored in an outdoor tank, mechanisms explaining the first and second points are the same as for the ph of the slurry stored under floors, but their importance is of smaller magnitude because of the lower volatilisation rate occurring in

11 DYNAMIC SIMULATION MODEL OF NITROGEN FLUXES IN PIG HOUSING 465 slurry tank rather than from under floor inside the buildings. The very small value observed at the highest value of a results from the largest daily volatilisation constraint (9 g m 2 day 1 ) that was mentioned earlier. Temperature is included in the calculation of the mass transfer coefficient k [Eqn (D3)], the calculation of the equilibrium between un-ionised ammonia and ammonium f [Eqn (D4)] and in the calculation of the Henri s constant H [Eqn (D5)]. An increase in the temperature value produces increase in the value of f and a decrease in the values of k and H. Both the augmentation in the value of f and the reduction in the value of H are increasing volatilisation. While the reduction in k tends to decrease volatilisation. It is therefore the fact that the combined effect on f and H is more important than the effect on k that causes the increase in volatilisation and consequently, the decrease of both N load and N conc. The behaviour of the parameter for air velocity v UF above slurry could be explained by the fact that an increase in the value of this parameter also causes an increase of the mass transfer coefficient k. The decrease of the impact of this parameter at the larger value of a might result from the fact that at larger velocity, the ammonia concentration in slurry is reduced by the enhanced volatilisation. Since it is the concentration gradient that is one of the driving forces for further volatilisation, a reduced value might counterbalance part of the action of air velocity. Globally, an increase in volatilisation following an increase of air velocity above emitting surfaces makes sense since it causes increased turbulence and mixing with ammonia-poor air of the higher air stratum. This, in turn dilutes ammonia in the air directly above the surface, and results in an increase of the ammonia concentration gradient forcing ammonia in the slurry solution to move into free air. The fact that for the parameter of the volume of excretions (Q), S R is negative at all three values of a is nothing but the result of a direct dilution effect. In the case of the nitrogen load, it is also the dilution effect that is implicated but in a non-direct way. When dilution occurs, total ammonia nitrogen concentration is reduced and therefore volatilisation is reduced, causing an increase in the remaining nitrogen load. Finally, both scenario sensitivity and relative sensitivity analyses further demonstrate that the model is mainly sensitive to those factors that have been identified in the literature has having an important impact on either volatilisation or nitrogen excretion. This correspondence seems to indicate coherence of the model with current knowledge of the effect of several characteristics of pig production Assessment of potential usefulness Comparison with mean values from literature Simulation results for both the reproduction site and the growing-finishing site corroborate with measured concentrations better than the mean values from literature (CRAAQ, 2006) and it is most likely because both studied sites differ from the average farm. Indeed, in the case of the reproduction site, distributed feeding quantities were adjusted on a daily basis and specific formulations were used for lactating and gestating sows, respectively both procedures are meant to reduce excess nitrogen. Since many farms do not use these measures yet, it was to be expected that concentration would be smaller than the average value. It was possible to include these characteristics in the model. In the case of the growing-finishing farms, the literature value was again larger than the measured value. In these two growing-finishing systems, a threephase feeding system was used, reducing nitrogen excretion when compared to single-phase. Furthermore, the permanent slurry storage structures consisted of lagoons, which tend to accumulate more precipitation water than concrete tanks. So again, it was to be expected that nitrogen concentration might be lower than average. It seems clear that the mathematical model offers the possibility to include farm characteristics that will affect nitrogen content of slurry. The model can provide important advantages when in need of an estimate of slurry concentration for regional load assessment compared to literature values, especially when there is a tendency towards reducing nitrogen loads by the farming community of a region Potential application scenarios For the first example, the simulated impact on nitrogen concentration using an acidifying diet is coherent with current knowledge since a lower slurry ph modifies the equilibrium between ammonia and ammonium ions in favour of the latter. By reducing the ammonia concentration, the volatilisation is also reduced, hence, the increase in total nitrogen concentration. In this example, the reduction on the amount of nitrogen volatilised from the slurry tank caused by the acidification of the slurry from a ph of 75 to a ph of 6 was of 86% with the roof and 85% without the roof. In comparison, Stevens et al. (1989) observed that a decrease from a ph value of 7 to a ph value of 6 could reduce volatilisation by approximately 80%. Based on the results of an in vitro experiment, Canh et al. (1998c) suggested that supplying Ca-benzoate instead of CaCO 3 could reduce ammonia emission from the slurry by about 48%, a value lower than the results presented here. However, Canh et al. (1998c) based their

12 466 P. BERTHIAUME ET AL. prediction on results measured over a short period (7 days), in an in vitro essay that did not take into account the dynamic situations present in commercial pig houses, in which fresh urine and faeces are continuously added to the slurry pit. Our results also show that, although the use of a roof had an important impact the nitrogen concentration, it had little impact on the simulated, volatised nitrogen load. Results of the second example, on the effect of the number of diets in phasefeeding systems, also corroborate with current knowledge, although some studies have shown an even more important impact, which in some cases could go up to 25 35% (Latimier & Dourmad 1993). It appeared from the sensitivity analyses that the model behaviour is coherent with the chemical knowledge and results from other investigations reported in the literature and therefore in itself, this does not provide major knowledge gain and should be considered a prerequisite to the use of the model. The real interest of modelling over the use of general values is that it becomes possible to refine the prediction of the nitrogen concentration and load in the slurry of a production site by using representative, that is, site-specific parameter values. In this sense, the two examples illustrated that the model shows great flexibility and could be used both as a scientific research tool and as a decision support tool in a management perspective at the farm level. 5. Conclusions The object of this work was to study the behaviour of the model of nitrogen fluxes in pig production sites proposed by Berthiaume et al. (2005). Two different methods of sensitivity analysis were used to: (i) identify the most significant parameters and (ii) verify whether the model was equally sensitive to those parameters over the range of plausible values that these could take when simulating either a reproduction system or a growingfinishing system. In the reproduction system, lowering the proportion of protein in feed P from 147% to 12% caused a 17% reduction in the value of the output variable nitrogen load N load while raising the proportion caused an augmentation of 15%. The impact of the parameter was even more important in the growingfinishing system where proportions of proteins in feed of 18% and 16% raised and lowered the nitrogen load in the slurry stored in the lagoon by 22%, respectively. The ph value of the slurry h slurry also caused major changes with a maximum impact of +24% for a value of 5 when compared to 74. The use of such a low ph value could represent the use of an acidifying diet. On the opposite, a high ph value (h slurry ¼ 8) caused a reduction of 26% of the total predicted nitrogen load in the stored slurry. The model also showed an important sensitivity to other parameters such as temperature of the slurry, air velocity over slurry or the use of water reducing system. These results are consistent with the already acknowledged importance of parameters representing feed content, slurry ph, temperature, and air velocity over slurry and constitute a confirmation of the coherence of the model. The two potential application scenarios illustrated well model flexibility and potential as a tool for management purposes. Nevertheless, further empirical validation of the model is needed specifically for farms characterised by extreme values for those significant parameters identified in sensitivity analyses. Acknowledgements This research was supported in part by the Fond québécois de la recherche sur la nature et les technologies (FCAR) and by a grant from the Faculté de médecine vétérinaire de l Université de Montréal. We are grateful to Dr. Jean-Pierre Villeneuve (INRS-ETE) for academic support. References Aarnink A J A; Elzing A (1998). Dynamic model for ammonia volatilization in housing with partially slatted floors, for fattening pigs. Livestock Production Science, 53(2), Aarnink A J A; Wagemans M J M; Keen, A (1993). Factors affecting ammonia emission from housing for weaned piglets. Proceedings of the Congress on Nitrogen Flow in Pig production and Environmental Consequences (Verstegen M W A; Den Hartog L A; van Kempen G J M; Metz J H M, eds), pp EAAP Publishers, Wageningen, The Netherlands. Anonymous (1992). Structures d entreposage des fumiers, lisiers et purins [Manure and slurry storage structures]: AGDEX 710. Inc. (Conseil des productions ve ge tales du Que bec, eds), Quebec, Canada Berthiaume P; Bigras-Poulin M; Rousseau A N (2005). Dynamic simulation model of nitrogen fluxes in pig housing and outdoor storage facilities. Biosystems Engineering, 92(4), , doi: /j.biosystemseng Canh T T (1998). Ammonia emission from excreta of growingfinishig pigs as affected by dietary composition PhD Thesis. Wageningen Institute of Animal Sciences and Agricultural university, Wageningen, The Netherlands Canh T T; Aarnink A J A; Mroz Z; Jongbloed A W; Schrama J W; Verstegen M W A (1998a). Influence of electrolyte balance and acidifying calcium salts in the diet of growingfinishing pigs on urinary ph, slurry ph and ammonia volatilisation from slurry. Livestock Production Science, 56(1), 1 13 Canh T T; Aarnink A J A; Schrama J W; Haaksma J (1997). Ammonia emission from pig houses affected by pressed sugar beet pulp silage in the diet of growing-finishing pigs. Proceedings of the International Symposium on ammonia and Odour control from animal production facilities