Systems for farm animal husbandry are subject to critical

Transcription

1 SPECIAL SUBMISSIONS Development of a Dynamic Model to Predict Emissions from Swine Houses Angelika Haeussermann Christian-Albrechts University Kiel Annamaria Costa University of Milan Jean-Marie Aerts Catholic University Leuven Eberhard Hartung Christian-Albrechts University Thomas Jungbluth University of Hohenheim Marcella Guarino University of Milan Daniel Berckmans* Catholic University Leuven Influences on dust emissions from livestock operations are number, weight, and kind of animals and characteristics of the housing system. Differences between facilities cannot be explained solely by mechanistic input variables. The objective of this study was to characterize the main input variables for modeling emissions of particulate matter with a mass median diameter 10 μm ( ) from swine facilities using a databased model. Investigations were performed in mechanically ventilated facilities for weaning, growing-finishing, and sows in Italy and Germany. The measurements included inside and outside concentration of airborne particles (scatter light photometry), ventilation rate (calibrated measuring fans), indoor air climate at a measuring frequency of 60 s, feeding times, and animal-related data such as weight and animal activity. Dust concentration and emission were simulated using a dynamic transfer function. The results indicated that the average emission rate was influenced considerably by housing system. The simulation of the emission rate resulted in a mean percentage error per data set of 21 to 39%, whereas the average simulated and measured emission rate per data set differed by about 4 to 19%. High prediction errors occurred especially during situations in which the absolute level and spatial location of the measured activity peaks did not correspond with the measured dust peaks. Further recommendations of the study were to improve continuous and accurate measurements of input variables, such as the activity level in animal houses, and to optimize the amount of measuring days in relation to the model accuracy. Copyright 2008 by the American Society of Agronomy, Crop Science Society of America, and Soil Science Society of America. All rights reserved. No part of this periodical may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher. Published in J. Environ. Qual. 37: (2008). doi: /jeq Received 29 Sept *Corresponding author (Daniel.Berckmans@biw.kuleuven.be). ASA, CSSA, SSSA 677 S. Segoe Rd., Madison, WI USA Systems for farm animal husbandry are subject to critical and conflicting requirements of the society and agricultural legislation for their health impact and for their environmental and greenhouse effects. Management and ventilation requirements in animal facilities are to ensure that the quality of indoor and outgoing air is of an acceptable risk for animal and human health. In addition, the mitigation of environmental impacts of gaseous and particulate emissions from farm systems is required. Data-based modeling approaches can be used to quantify the impact of management, ventilation, seasonal, and diurnal variations on atmospheric emissions (Vranken et al., 2004). The parameters of the emission model are determined based on frequent measurements of the emission rate and the main influencing variables. The emission model is then used to estimate emission rates due to variations of the latter during days with no emission measurements. The release and composition of particulate substances from livestock operations are influenced by the farm management and characteristics of the respective housing system, specified, for example, by feeding practices, bedding materials, fouled surfaces, number of animals, animal species, ventilation, and inside and outside climatic conditions (Takai et al., 1998; Guarino et al., 1999; Aarnink et al., 2004, Hartung et al., 2004). The impacts of aerosols depend mainly on their physical size and on their biological and chemical composition. Due to biologically active components, aerosols in animal husbandry are labeled in general as bioaerosols, which may lead to infections, allergic, toxic, or pharmacological reactions (Cox and Wathes, 1995). Origins of aerosols in swine facilities are mainly feed, bedding materials, skin, and excrement (Aarnink et al., 1999; Schneider et al., 2001). They consist of a complex mixture made of organic dust and relatively high concentrations of NH 3, microorganisms, and endotox- A. Haeussermann, J.-M. Aerts, and D. Berckmans, M3-BIORES, Catholic Univ. Leuven, Kasteelpark Arenberg 30, 3001 Heverlee, Belgium. A. Costa and M. Guarino, Dep. of Veterinary and Technological Sciences for Food Safety, Faculty of Veterinary Medicine, Univ. of Milan, via Celoria 10, Milan, Italy. A. Haeussermann and T. Jungbluth, Livestock Systems Engineering, Inst. of Agricultural Engineering, Univ. of Hohenheim, Garbenstrasse 9, Stuttgart, Germany. A. Haeussermann and E. Hartung, Inst. of Agricultural Engineering, Christian-Albrechts Univ., Max-Eyth-Strasse 6, Kiel, Germany. 50 Abbreviations: AU, animal unit (1 AU = 500 kg); d ae, aerodynamic diameter; d ae, 50% cut-off at aerodynamic diameter;, particulate matter with a mass median diameter 10 μm; R 2, coefficient of determination adjusted for number of adj parameters and data. 557

2 Table 1. Investigated time periods and research facilities in Italy. No. of animals Avg. animal weight Food supply and feeding times Floor material kg Sows Farm 1, 18 Feb. 25 Mar to 43 sows 180 dry (ad libitum) slatted concrete floor Weaning Farm 1, 25 Sept. 9 Nov piglets 7 36 dry (ad libitum) grooved plastic slats Weaning Farm 2, 5 Sept. 10 Oct piglets 7 30 liquid + dry (manually) at 0740, 1340, and 1700 grooved plastic slats Growing-finishing Farm 1, 5 23 Nov pigs liquid (automatically) at 0930, 1230, 1630, and 1930 slatted concrete floor ins (Seedorf et al., 1998; Takai et al., 2002; Aarnink et al., 2004; Rieger, 2004). Donham et al. (1995) recommended clear exposure limits for aerosols, endotoxins, and NH 3 in swine facilities, which were based on dose response relationships between work exposure and respiratory symptoms. According to a cross-sectional study, undertaken by Radon et al. (2002), pig farmers were at high risk for the development of work-related respiratory symptoms related to asthma-like syndrome and chronic bronchitis. Aerosols emitted from livestock farming can be a nuisance to the neighborhoods in the vicinity of the farm (Seedorf, 2004; 2007). Beside odorous compounds and supposed health effects, they add significantly to the atmospheric content of particulate ammonium (gas-to-particle conversion) and organic carbon (Lammel et al., 2004). Equally important, they are relevant to climate change issues, such as cloud formations, radiative forcing, and coupling with photochemistry of greenhouse gases (IPCC, 2005). The emission rate of particles (i.e., aerosols with a mass median diameter of 10 μm) (USEPA, 1987) differs seasonally and diurnally and is related to management aspects such as ventilation and working activity. In mechanically ventilated facilities, a high ventilation rate during summer results in a low indoor dust concentration but may increase dust emission rate (Takai et al., 1998; Jacobson et al., 2004). The influence of ventilation rate on dust emissions can be explained by the transport, settlement, and resuspension of airborne particles in relation to airflow patterns in the building. Gustafsson (1999) reported that, depending on air velocity, less than 50% of the total amount of settled and airborne dust is ventilated to the outside. According to Haeussermann et al. (2006), 64 and 96% of airborne particles measured indoors were transported to the outside in winter and summer, respectively (i.e., during measuring days with low and high ventilation rate). Keck et al. (2004) found an increased impact of open yard exercises on the level of the absolute emission rate from pig husbandry during summer and a significant influence on emissions with increasing growth stage of growing-finishing pigs. Control activities of farmers, feeding operations, and animal activity can affect diurnal variations in particle concentration and emission. High peak-to-mean ratios of airborne particle concentrations were found especially in husbandry systems that use straw as bedding material (Hartung et al., 2004). The effect of activity on the dust concentration and emission is evident when evaluating characteristic daily patterns of the measured dust emission rate (e.g., differences between day and night) (Zeitler-Feicht et al., 1991; Takai et al., 1998; Hinz and Linke, 1998; Gustafsson, 1999) or by correlating animal activity and concentration in swine facilities, both recorded at highly frequent intervals (Pedersen, 1993; Pedersen and Pedersen, 1995; Schneider et al., 2001). In general, it is assumed that the resuspension of settled dust does not only lead to a higher dust indoor concentration but also to an increase in the dust emission rate. Because the effective dust emission rate is influenced by a number of dust sources and by varying transport characteristics of dust particles and because input variables and dust emissions vary due to management, place, and time, a dynamic databased model was developed in this study to estimate dust emissions from individual swine facilities. In general, data-based or site-specific models represent conditions similar to those used for the model parameter estimation. They are a useful tool to simulate and predict dust emissions over a time period that extends beyond the measuring period. Data-based models have to be adjusted regularly and cannot be transferred easily to another facility. The primary use is to improve estimated emission factors while reducing the required number of measuring days per farm. In addition, a better understanding of the relationship between input variables and dust emissions enables improved control measures for reducing environmental impacts of farms. The study focuses on emission rates from mechanically ventilated swine facilities. The objective of this study was to improve the ability for estimating or predicting average concentrations and emissions that occur in mechanically ventilated swine facilities and their variation due to diurnal and seasonal impacts. Main input variables (i.e., generally available or easily measurable variables, which have a major influence on the emission rate of ), were characterized. Possible applications are improved emission factors and better control measures. Materials and Methods Dust indoor concentrations and emissions were investigated in facilities for weaning, growing-finishing, and sows in Italy (Pianura Padana, Milan, Farm 1: N, E; Farm 2: N, E) (Table 1) and Southern Germany (Hohenheim, N, E) (Table 2). The swine facilities were equipped with mechanical ventilation systems, separately controllable ventilation fans, and calibrated measuring fans. Temperature and humidity were measured and recorded via sensors of the ventilation control system and additional sensors. Further recordings included the group animal activity in one swine facility and the animal weight and feeding times in all investigated houses. Measurements in Research Facilities for Weaning, Growing-Finishing, and Sows in Italy The measurements in Farm 1 in Italy (weaning and sows) included inside and outside concentrations of airborne particles, ventilation rate, and indoor air climate with a measuring frequency of 1 min. Dust particles were sampled continuously once per minute in the incoming air stream (weaning) and 558 Journal of Environmental Quality Volume 37 March April 2008

3 Table 2. Investigated time periods in the research pig facilities in Germany. No. of animals Avg. animal weight Food supply and feeding times measurements kg Dec pigs 50 liquid (sensor) automatically; 20 feeding times: 0600 until 2200 indoors and exhaust shaft Feb pigs 101 liquid (sensor) automatically; 20 feeding times: 0600 until 2200 indoors and exhaust shaft May pigs 61 liquid (sensor) automatically; 20 feeding times: 0600 until 2200 indoors and exhaust shaft June pigs 90 liquid (sensor) automatically; 20 feeding times: 0600 until 2200 indoors and exhaust shaft Dec pigs 58 mash (ad libitum) automatic food supply at six feeding times: 0610 until 2010 indoors Jan pigs 85 mash (ad libitum) automatic food supply at six feeding times: 0610 until 2010 indoors Apr pigs 35 mash (ad libitum) automatic food supply at six feeding times: 0610 until 2010 indoors and exhaust shaft June pigs 73 mash (ad libitum) automatic food supply at six feeding times: 0610 until 2010 indoors and exhaust shaft Particulate matter with a mass median diameter 10 μm. Control corridor (height: 1.70 m). No measurements of temperature, humidity, ventilation rate, and animal activity (25 26 May 2004). indoors at a distance of 60 cm from one of the exhaust chimneys (all facilities). To ensure isokinetic sampling conditions, the dustmeasuring instrument for the latter was positioned in such a way that the airflow rate, checked with a hot wire anemometer, was in general less than 0.5 m s 1. Sampling was performed in isoaxial conditions to avoid particle size distribution distortion of the sampled aerosol. Samples were taken by the instrument each second and averaged by the sample interval time. Farm 2 (weaning) and Farm 1 (growing-finishing) included measurements of concentrations at a frequency of 1 min. Ventilation rate, outside and inside temperature, and humidity were recorded every 15 min. The number of valid measuring days within the investigation periods, listed in Table 1, were 21, 15, and 9 for sows, weaning and growingfinishing, Farm 1, respectively, and 36 for weaning, Farm 2. In each of these facilities, the concentration was monitored using calibrated scatter light photometers (accuracy: ±3 μg m 3 ) (EPAM 5000, HAZ-Dust; Environmental Devices Corporation, Plaistow, NH). Temperature and humidity were recorded via the ventilation control system (Fancom B.V., Panningen, The Netherlands). Sensors were placed in the middle of the rooms, far from inlets, at a height of approximately 1.5 m. The measurements of the ventilation rate were performed separately for each exhaust chimney using calibrated measuring fans with high accuracy (±45 m 3 h 1 ) (Fancom B.V., Panningen, The Netherlands) (Berckmans et al., 1991). The ventilation principle for the research facilities in Italy was generally based on a high air inlet, high air outlet. Incoming air was supplied through a perforated ceiling at the weaning facility, Farm 1, and through wall openings at the other stables. The floor of the compartments was covered with grooved plastic slats (weaning) or featured slatted concrete floor elements (growing-finishing, sows). Feeding was supplied dry or liquid ad libitum or at three to four fixed feeding times per day (Table 1). A detailed description of the facilities is provided in Costa et al. (2004; 2005). Measurements in the Research Facility for Growing- Finishing Pigs in Germany Investigations in the research facility for growing-finishing pigs in Hohenheim were performed randomly in two compartments and throughout two growing-finishing periods (18 Nov until 24 Feb and 13 Apr until 20 July 2004). Dust concentrations and emissions were measured at four measuring periods per growing-finishing period, in compartment 1 or in compartment 2, each resulting in a data set of 18 to 42.5 h (Table 2). The compartments consisted of two pens with a concrete slatted floor and a slurry pit underneath and housed 54 pigs each (0.9 m 2 per pig). Fresh air entered via two air inlet pore channels per compartment. The outgoing air was extracted under-floor, according to a high-air-inlet, low-air-outlet ventilation principle. The compartments were equipped alternatively with two different feeding systems, whereby the feeding type influenced the consistency of the feed and the activity pattern of the animals. Feed was provided as liquid or as mash at 20 fixed feeding times per day or ad libitum (Table 2). Straw was supplied weekly via an occupation equipment, type Porky Play (straw capacity: approx. 3 4 kg wk 1 per pen). Airborne particle concentrations were monitored simultaneously at one representative measuring point indoors (all measuring periods) and in the exhaust shaft (six out of eight measuring periods). At both positions, the dust particles were monitored using calibrated scatter light photometers (accuracy: ±1 μg m 3 ) (DustTrak 8520 Aerosol-Monitor; TSI Inc., Shoreview, MN). Further measurements included air temperature, relative humidity, ventilation rate, and animal activity, which were recorded continuously with a frequency of one average value per minute. Temperature and relative humidity were measured with a combined sensor (accuracy ±1 C and ±1%, respectively) (PT 100 and capacitive, HygroClip; Rotronic Messgeräte GmbH, Ettlingen, Germany). Measuring locations were indoors 1 m above the pen area and in the incoming and exhaust air stream. The group animal activity per pen (27 pigs) was recorded with infrared sensors (Pedersen and Pedersen, 1995), and the ventilation rate was measured using calibrated measuring fans (accuracy: ±20 m 3 h 1 ) (Multifan; Vostermans Ventilation B.V., Venlo, The Netherlands) (Hartung, 2001). The experimental lay-out, including results on respirable and inhalable particle size fractions, sampling of endotoxins and bacteria at the indoor air, a vertical projection of the research facility, and detailed tables about the accuracies of the measuring instruments, is provided in Haeussermann (2006). Modeling Approach and Model Validation The aim of modeling in this study, using a data-based modeling approach, was to cope with high variations between Haeussermann et al.: Model to Predict Emissions from Swine Houses 559

4 individual housing systems when estimating yearly emission rates. The first step involved evaluating the linear relationship between the numerous input variables and dust concentration in the two facilities for growing-finishing pigs using a multiple linear regression model: y l = β 0 + β 1 x l1 + β 2 x l2 + + β p x lp + ε l [1] where y l is the model output of the linear regression model; x lj and β j are the input variable and regression coefficient, respectively; p is the number of input variables in the model; ε l is a serially uncorrelated random variable with a zero mean; and l is the sampling size. Input variables were evaluated by significance of coefficients and partial correlation coefficients. A dynamic data-based modeling approach was used for modeling and validating dust concentration and emission in one facility. The randomly recorded data in this facility were subdivided into separate data sets (Table 2). Assuming that the system is discrete-time, linear, and time-invariant, the multipleinput, single-output transfer function model can be described as follows (Young, 1984; Aerts and Berckmans, 2004): m 1 Bi ( z ) yk ( ) = u( ) ( ) 1 i k nki + ξ k i= 1 Az ( ) [2] where y(k) is the output of the dust concentration (mg m 3 ) or the dust emission rate (g h 1 ) at time k; u i (k) is the ith input of the measured variable at time k; nk i is the time delay between the inputs i and their first effects on the output; ξ(k) is additive noise (assumed to be a zero mean, serially uncorrelated sequence of random variables with variance σ 2 ) accounting for measurement noise, modeling errors, and effects of unmeasured inputs to the process; and A(z 1 ) and B(z 1 ) are as shown in Eq. [3] and Eq. [4], respectively: A(z 1 ) = 1 + a 1 z 1 + a 2 z a na z na [3] B(z 1 ) = b 0i + b 1i z 1 + b 2i z b nbi z nbi [4] where a j and b j are the model parameters to be estimated; z 1 is the backward shift operator, with z 1 [y(k)] = y(k 1); y and k are defined as in Eq. [2]; and na and nb are the orders of the respective polynomials. The model parameters were estimated using a simplified refined instrumental variable approach (Young, 1984), whereby the resulting models were selected by means of the Young Identification Criterion, the standard error of the model parameters, and the coefficient of determination. Time step of the model was one value per 60 s. The modeling approach was tested on four data sets of the research facility for growing-finishing pigs, Germany (Table 2). Selection criteria for the data sets were completeness of data, including indoor and exhaust concentration; ventilation rate; temperature; humidity; and animal activity. The data set of 26 to 27 Apr was not selected for model evaluation. This data set differed considerably compared with the rest of the data sets. Although the this difference was explained mainly by the early measuring period (growing-finishing Day 13 14) and related disturbances. As a result, the model had to be limited to the period between growing-finishing Day 24 (50 kg average weight; Dec. 2003) and growing-finishing Day 85 (101 kg average weight; 9 11 Feb. 2004). The model parameters were estimated at three out of four data sets each time. The validation of each model took place at the respective data set, which was not taken to build the model. Validation criteria were (i) the ability to simulate the average measured dust concentration and emission per validation set; (ii) R 2 value to evaluate the linearity of measured and simulated values per minute per validation set (i.e., how far the input variables account for the variability of the output); and (iii) the agreement of the measured and simulated values per minute per validation set, expressed by RMSE, with: 1 N RMSE = y() k t () k 2 N = k 1 where y(k) is the simulated output at time k, t(k) is the target or measured value at time k, and N is the number of values in the dataset according to the duration of the measured time slot. Results and Discussion Variation of the Dust Concentration and Emission The average indoor concentration of the different swine facilities varied between 0.11 mg m 3 (median: 0.09 mg m 3 ) and 0.73 mg m 3 (median: 0.57 mg m 3 ) (Table 3). The average and median particle concentrations were lowest in the research facility for weaning (Farm 1) in Italy and highest in the research facility for growing-finishing in Germany. In comparison, the average concentration at the three remaining facilities (weaning Farm 2, growing-finishing and sows, Farm 1, Italy) featured a narrower range of 0.31 to 0.47 mg m 3 (median, mg m 3 ) (Table 3). Peak concentrations were noticeably high, compared with the general level of the dust concentration, but relatively infrequent. Thus, the median particle concentration was in general lower than the average concentration (Table 3). Average indoor concentrations found in the literature in swine facilities for growing-finishing without straw bedding range from 0.46 to 0.74 mg m 3 (Zeitler-Feicht et al., 1991; Gallmann et al., 2002; Koziel et al., 2004). Dust indoor concentrations reported by Takai et al. (1998) from intensive measurements in Northern Europe in swine facilities for weaning, growing-finishing, and sows averaged 0.23 mg m 3 for respirable dust (50% cut-off at aerodynamic diameter [d ae 50 ] = 5 μm) and 2.19 mg m 3 for inhalable dust (d ae 50 = 100 μm). Compared with values in the Northern European study, the concentration (aerodynamic diameter [d ae ] <10 μm) measured in the weaning facility (Farm 1) was rather low, whereas concentrations measured in the other swine facilities fit well, considering the different particle size fractions. According to the dust indoor concentration, the average emission rate per building showed a high variation throughout the facilities for sows, weaning, and growing-finishing, ranging from 0.7 to 6.0 g d 1 AU 1 (median, g d 1 AU 1 ) (Table 3; Animal Unit: 1 AU = 500 kg). Because the main studies on dust emission were performed for respirable and inhalable par- [5] 560 Journal of Environmental Quality Volume 37 March April 2008

5 Table 3. Mean, median, and total range of model input variables and concentration and emission of particulate matter with a mass median diameter 10 μm ( ) measured at different swine facilities (Animal Unit: 1 AU = 500 kg animal weight). Input variables mean/median (total range) Sows Farm 1, Milan Weaning Farm 1, Milan Weaning Farm 2, Milan Grow-Fin Farm 1, Milan Grow-Fin Hohenheim Date 18 Feb Mar Sept Nov Sept Oct Nov Nov Dec June 2004 Air temperature indoors ( C) 16.8/17.0 ( ) 24.7/24.4 ( ) 28.3/28.2 ( ) 19.0/19.0 ( ) 20.2/20.8 ( ) Air relative humidity indoors (%) 70/72 (28 92) 52/53 (31 82) 53/52 (37 78) 62/62 (45 74) 47/46 (27 64) Animal activity (V ) 0.83/0.82 ( ) Ventilation rate (m 3 h 1 AU 1 ) 96/73 (20 577) 273/269 ( ) 245/236 (81 827) 174/167 ( ) 481/372 ( ) concentration indoors (mg m 3 ) 0.31/0.24 ( ) 0.11/0.09 ( ) 0.40/0.37 ( ) 0.47/0.40 ( ) 0.73/0.57 ( ) emission rate (g d 1 AU 1 ) 0.70/0.48 ( ) 0.71/0.59 ( ) 2.38/2.14 ( ) 2.04/1.58 ( ) 5.99/5.52 ( ) Grow-Fin, growing-finishing (T) Eight random measuring days (see Table 2). Volts. Table 4. Multiple linear regression model, facility for growingfinishing pigs, Germany. Nonstandardized coefficients Correlation coefficients Input variables β SE (β) Pearson Partial p Value Constant Indoor humidity (%) Ventilation rate (m 3 s 1 ) Animal activity (V ) Animal weight (AU ) Output variable: particulate matter with a mass median diameter 10 μm concentration. Mean (measured): 0.49 mg m 3 ± Model evaluation: R 2 adj: 0.603; SE: 0.234; F value = 2746 (p = 0.000); mean: 0.49 mg m 3 ± Volts. Animal unit (1 AU = 500 kg). ticle size fraction, numbers on emission rate from swine facilities are still very sparse. Preliminary results for average emission rates reported by Jacobson et al. (2004), Hoff et al. (2005), and Koziel et al. (2005) varied between 0.7 and 4.3 g d 1 AU 1 in facilities for growing-finishing pigs without straw bedding. Differences occurred with regard to diet, facilities, and season. Jacobson et al. (2005) and Jerez et al. (2005) found average emission rates of 1.2 g d 1 AU 1 in gestation and dry sow barns and in a farrowing facility, with maximum emission rates up to 8.6 g d 1 AU 1. Average emission rates of 1.4 to 2.7 g d 1 AU 1 were reported in Haeussermann et al. (2007a) from deep-bedded growing-finishing pigs. Differences were due to growing-finishing stage and ventilation rate per AU. Feeding operations, level of animal activity, ventilation rate, indoor temperature, animal weight, growing-finishing day, housing system, and management of the facility exerted an influence on the level of dust indoor concentration and dust emission rate in this investigation (Tables 4 and 5). The difference in the average emission rate between the research facilities for weaning and growing-finishing in Italy and Germany, respectively (Table 3), can be associated with the higher ventilation rate per AU at the latter and hence is partly due to seasonal differences. On the other hand, the lower average weight per animal in the weaning stable, in combination with a different farm management, led to a very low dust production, which was mirrored by a relatively low indoor concentration at this farm (Table 3). Similarly, Gallmann et al. (2002) reported a mean indoor concentration of 0.17 mg m 3 in a growingfinishing kennel housing with natural ventilation during mea- Table 5. Multiple linear regression model, facility for growingfinishing pigs, Italy. Nonstandardized coefficients Correlation coefficients Input variables β SE (β) Pearson Partial p Value Constant Indoor humidity (%) Ventilation rate (m 3 s 1 ) Feeding times (1/0) Animal weight (AU ) Output variable: particulate matter with a mass median diameter 10 μm concentration. Mean (measured): 0.47 mg m 3 ± Model evaluation: R 2 adj: 0.225; SE: 0.224; F value = 43.4 (p = 0.000); mean: 0.47 mg m 3 ± Animal unit (1 AU = 500 kg). Haeussermann et al.: Model to Predict Emissions from Swine Houses 561

6 Table 6. Validation (leave-one-out ) of the transfer function model on concentration and emission of particulate matter with a mass median diameter 10 μm ( ), research facility for growing-finishing pigs, Hohenheim (input variables: ventilation rate, animal activity, and animal weight) Dec (liquid feeding) surements outside the kennels, where the air quality is influenced to a minor extent by the animals. The ambient dust concentration outside the weaning facility in Italy in this investigation averaged 0.02 mg m 3 with a maximum value of 0.17 mg m 3. Multiple Linear Regression Model on Dust Concentration The linear relation between the measured input variables and the dust concentration was estimated with a multiple linear regression model. The proportion of variability in the data that were accounted for in the regression model, evaluated by the adjusted coefficient of determination, was R 2 = 0.60 in the facility adj for growing-finishing pigs in Germany (Table 4) and R 2 = 0.22 adj in Italy (Table 5). Mean dust concentration and SE of the estimated values were 0.49 mg m 3, 0.23 (Table 4) and 0.47 mg m 3, and 0.22 mg m 3 (Table 5). Significant coefficients were found for indoor humidity, ventilation rate, animal activity, and animal weight. The partial correlation between indoor temperature and dust concentration was lower than 0.1 and resulted in no additional improvement of the model accuracy. Measurements of animal activity were not available in the facility for growing-finishing pigs in Italy and hence were replaced by the information on feeding times. Thus, R 2 was clearly lowered. Similarly, adj R2 adj and SE were lowered to 0.36 and 0.30, respectively, if animal activity was replaced by feeding times in the facility in Germany. According to the partial correlations of the input variables (Tables 4 and 5), the variability in the dust concentration was mainly explained by animal activity or feeding times, animal weight, and ventilation rate, and the limitation of the model on these three input variables had only a minor influence on the evaluation criteria (R 2 : 0.59; SE: 0.24 and adj R2 : 0.20; adj SE: 0.23). No changes with regard to the model accuracy were registered when animal weight was replaced by fattening day, which is in accordance with a r value of and hence a nearly linear relationship between these two variables. One remarkable aspect when comparing the two linear regression models is the difference in the mathematical sign for indoor humidity and ventilation rate (Tables 4 and 5). In general, a reduced concentration of airborne dust would be expected in combination with an increased relative humidity (bounding forces, increasing diameter of particles; Seedorf and Hartung, 2002) or due to a dilution effect when ventilation rate is increased (Heber et al., 1988). However, in both cases, their influence on the concentration of airborne dust has to be differentiated between the short-term effects occurring within a few days and the long-term 9 11 Feb (liquid feeding) 8 9 June 2004 (mash feeding) June 2004 (liquid feeding) Animal weight (AU) Ventilation rate (m 3 h 1 AU 1 ) mean: 371; range: mean: 243; range: mean: 854; range: mean: 752; range: indoor concentration (mg m 3 ) 0.57/0.72 ± 0.29 ; R 2 : /1.01 ± 0.65; R 2 : /0.39 ± 0.29; R 2 : /0.39 ± 0.20; R 2 : exhaust concentration (mg m 3 ) 0.37/0.52 ± 0.19; R 2 : /0.58 ± 0.41; R 2 : /0.30 ± 0.16; R 2 : /0.33 ± 0.15; R 2 : emission rate (g h 1 ) 0.75/0.89 ± 0.34; R 2 : /2.11 ± 1.07; R 2 : /2.05 ± 0.60; R 2 : /2.34 ± 0.84; R 2 : Model is built on three data sets. Validation is performed at the respective fourth data set. AU: Animal unit (1 AU = 500 kg). Measured/simulated ± RMSE. effects occurring over several months. With regard to indoor relative humidity, the effect on the concentration of airborne dust particles was minor, and at the same time a co-linearity with indoor temperature, ventilation rate, animal weight, and animal activity resulted in an apparently positive correlation with dust indoor concentration. An increased ventilation rate increases air movement in the room and thereby resuspends settled dust, which results in a higher concentration of airborne particles. According to Gustafsson (1999), ventilation has only a limited effect on the concentration of airborne dust compared with settlement. In addition, the effect of the ventilation rate on the spatial dust distribution depends largely on the respective airflow pattern in the building (Wang et al., 2002), which explains why the effect of ventilation rate can differ between individual facilities. A negative correlation between the ventilation rate and the concentration of airborne dust can be found if ventilation rate is considerably increased for a longer period. In this case, a higher proportion of originally settled dust is ventilated to the outside, and the amount of settled and airborne dust in the building is reduced. Particulate emissions are increased in both cases. Dynamic Data-Based Model on Dust Concentration and Emission The concentrations and emissions in the facility for growing-finishing pigs in Germany were estimated for each data set with a data-based dynamic model. Input variables were ventilation rate, animal activity, and animal weight. The modeling approach is advantageous for describing individual dynamic biological systems (e.g., the dynamic heat loss of chicken) (Aerts and Berckmans, 2004) or the evaporation of water droplets before steady state is reached (Haeussermann et al., 2007b). Model parameters for each input variable were calculated based on three data sets each time. The validation of the model in terms of mean value, R 2 value, and RMSE between measured and simulated dust concentration and emission are listed in Table 6. Considering the emission rate, the agreement of the simulated and measured values per minute is visualized for the two sampling periods with the lowest percentage error (Fig. 1) or the highest R 2 value (Fig. 2). In general, the dynamic modeling approach simulated the mean indoor and exhaust dust concentration fairly well (Table 6). The average emission rate per data set was simulated with an error of 4 to 19%. The mean percentage error of the simulated dust emission rate per data set per minute ranged from 21% (Fig. 1) to 39% (Fig. 2). Although the diurnal 562 Journal of Environmental Quality Volume 37 March April 2008

7 Fig. 1. Measured and simulated emission rate of particulate matter with a mass median diameter 10 μm ( ), 8 9 June 2004 (54 pigs, average weight: 73 kg; mash feeding, ad libitum). Input variables: ventilation rate, animal activity, and animal weight (mean measured/simulated: 2.13/2.05; RMSE: ± 0.60; R 2 : 0.529) variability in the dust concentration and emission was estimated reasonably well for the data set of 12 to 13 Dec (Fig. 2), resulting in an R 2 of 0.70 to 0.77, the bias between measured and simulated values was highest for this data set. This bias can be explained by the missing information regarding the influence of the low animal weight when estimating the model parameters from the three other data sets (Table 6). The simulation of the mean dust concentration and emission was clearly improved if the transfer function, built on the respective three other data sets, included the information of the total variation of the ventilation rate and the animal weight (Table 6; Fig. 1: 8 9 June 2004). To improve the simulation of the concentration and emission rate, main future goals are to optimize measurements in relation to the model accuracy (i.e., to optimize the database that is taken to build the model, including the total variation of the respective input variables) and to perform measurements on input variables, such as animal activity at a high temporal frequency and at a high spatial accuracy. Summary The average emission rate, measured in swine facilities for sows, weaning, and growing-finishing in Italy and Germany, ranged from 0.7 to 6.0 g d 1 AU 1, which was generally in accordance with particle emission rates reported in literature. The main influencing variables on the concentration and emission within individual housings were ventilation rate, (group) animal activity, feeding operations, indoor humidity, weight of the animals, and growing-finishing day. Taking these variables into account, the variability of the exhaust concentration was accounted for with a R 2 value of Housing characteristics and farm management exerted a considerable influence on the absolute value of concentrations and emissions and on the relations between input and output variables. The ventilation rate had a negative correlation with the dust concentration in one building, whereas in another building a positive correlation occurred. Different airflow patterns in the buildings and differences between short-term (diurnal) and long-term (seasonal) dilution effects of high ventilation rates influence the direction of these correlations. The concentration and emission rate in one of the facilities was simulated using a dynamic data-based model. Fig. 2. Measured and simulated emission rate of particulate matter with a mass median diameter 10 μm ( ), Dec (53 pigs, average weight: 50 kg; sensor-liquid feeding). Input variables: ventilation rate, animal activity, and animal weight (mean measured/simulated: 0.75/0.89; RMSE: ± 0.34; R 2 : 0.772) Three input variables (ventilation rate, animal weight, and animal activity) were used to simulate the average emission rate per validation set with an error of 4 to 19%. The mean percentage error of the measured and simulated dust emission rate per data set per minute was 21 to 39%. Further improvements on the modeling approach are necessary to realize an accurate simulation of the particle emission rate from pig facilities at a high time resolution. This includes an optimization of the data sampling period and an improved spatial accuracy for measurements on input variables such as the activity level. Acknowledgments The project was partly funded by the EU in the framework of a Marie Curie training site (reference number: HPMT- GH ). For the study conducted in Italy, funding was provided by the Italian Agency for Environmental Protection and Technical Services for a research initiative into agricultural air quality and emission into atmosphere. Data sampling in Germany was funded by Deutsche Forschungsgemeinschaft in the framework of the research training group Mitigation strategies for the emission of greenhouse gases and environmentally toxic agents from agriculture and land use. References Aarnink, A.J.A., P.F.M.M. Roelofs, H. Ellen, and H. Gunnink Dust sources in animal houses. p In Dust control in animal production facilities. Proc Int. Symposium, Aarhus, Denmark. 30 May 2 June Danish Inst. of Agricultural Sciences (DIAS), Research Centre Bygholm, Horsens, Denmark. Aarnink, A.J.A., N. Stockhofe-Zurwieden, and M.J.M. Wagemans Dust in different housing systems for growing-finishing pigs. In Engineering the Future. Proc. AgEng2004 [CD-ROM], Leuven, Belgium Sept Technological Inst., K.U., Leuven, Belgium. Aerts, J.-M., and D. Berckmans A virtual chicken for climatic control design: Static and dynamic heat loss. Trans. ASAE 47: Berckmans, D., P. Vandenbroeck, and V. Goedseels Sensor for continuous ventilation rate measurement in Livestock Buildings. J. Indoor Air 3: Costa, A., J. Dekock, M. Guarino, E. Vranken, and D. Berckmans On-line monitoring of emissions from a weaning room. 98th Annual Conf. & Exhibition, Paper-No.1189, Minneapolis, MN June AWMA, Pittsburgh, PA. Costa, A., M. Guarino, J. Dekock, M.A. Meda, P. Navarotto, and D. Berckmans On-line monitoring of PM10 emission in a swine intensive breeding. p In Emission inventories and projections. Proc. 5th JOINT Haeussermann et al.: Model to Predict Emissions from Swine Houses 563

8 UNECE Task Force & EIONET, Pallanza, Italy Oct JRC, Ispra, Italy. Cox, C.S., and C.M. Wathes Bioaerosol handbook. Lewis Publ., CRC Press, Boca Raton, FL. Donham, K.J., S.J. Reynolds, P. Whitten, J.A. Merchant, L. Burgmeister, and W.J. Popendorf Respiratory dysfunction in swine production facility workers: Dose response relationship of environmental exposures and pulmonary-function. Am. J. Ind. Med. 27: Gallmann, E., E. Hartung, and T. Jungbluth Comparison of the indoor dust concentration between two different housing and ventilation systems for fattening pigs. Landbauforsch. Völkenrode 235: Guarino, M., A. Caroli, and P. Navarotto Dust concentration and mortality distribution in an enclosed laying house. Trans. ASAE 42: Gustafsson, G Factors affecting the release and concentration of dust in pig houses. J. Agric. Eng. Res. 74: Haeussermann, A Stallklimaregelung und Emissionen Entwicklung und Evaluierung Sensorgestützter Komplexer Regelstrategien für die Mastschweinehaltung. (In German with English abstract.) Forschungsbericht Agrartechnik VDI-MEG-Schrift 443, Ph.D. diss. Univ. of Hohenheim, Hohenheim, Germany. Haeussermann, A., M. Götz, and E. Hartung. 2007a. Particulate emissions from deep-bedded growing-finishing pigs. In How to improve air quality. Proc. Dust Conf [CD-ROM; Maastricht, The Netherlands, Apr. 2007, Animal Science Group, Wageningen UR, Wageningen, The Netherlands. Haeussermann, A., E. Hartung, T. Jungbluth, E. Vranken, J.-M. Aerts, and D. Berckmans. 2007b. Cooling effects and evaporation characteristics of fogging systems in an experimental piggery. Biosyst. Eng. 97: Haeussermann, A., T. Jungbluth, and E. Hartung Emissions of particulate matter (PM10) from housing systems for fattening pigs. p In Workshop on agricultural air quality: State of the science. Proc., Potomac, MA. 5 8 June Dep. of Communication Services, North Carolina State Univ., Raleigh, NC. Hartung, E Konzeption, realisierung und evaluierung einer versuchseinrichtung zur entwicklung und differenzierten beurteilung von haltungssystemen für mastschweine. (In German with English abstract.) Forschungsbericht Agrartechnik VDI-MEG Schrift 392, Prof. diss. Univ. of Hohenheim, Hohenheim, Germany. Hartung, E., K. Ollesch, A. Haeussermann, M.A. Rieger, H. Diefenbach, A. Sundrum, M. Ebke, and M. Lohmeyer Stallluftqualität und arbeitsplatzbelastung in unterschiedlichen mastschweinehaltungsverfahren (In German). Agrartech. Forsch. 10: Heber, A.J., M. Stroik, J.L. Nelssen, and D.A. Nichols Influence of environmental factors on concentration and inorganic content of aerial dust in swine finishing buildings. Trans. ASAE 31: Hinz, T., and S. Linke A comprehensive experimental study of aerial pollutants in and emissions from livestock buildings: Part 2. Results. J. Agric. Eng. Res. 70: Hoff, S., B.C. Zelle, M.A. Huebner, and A.K. Gralapp NH 3, H 2 S, CO 2, PM, and odor animal emission data from the Six-State (APECAB) Project: Swine deep-pit finishing building. 98th Annual Conf. and Exhibition, Paper No. 05-A-648. Minneapolis, MN June AWMA, Pittsburgh, PA. IPCC Report of the IPCC expert meeting on emission estimation of aerosols relevant to climate change. Geneva, Switzerland, 2 4 May Intergovernmental Panel on Climate Change, Geneva, Switzerland. Jacobson, L.D., B.P. Hetchler, and V.J. Johnson Particulate emissions from pig, poultry, and dairy facilities located in Minnesota. In Engineering the future. Proc. AgEng2004 [CD-ROM], Leuven, Belgium Sept Technological Inst., K.U., Leuven, Belgium. Jacobson, L.D., B.P. Hetchler, V.J. Johnson, D.R. Schmidt, R.E. Nicolai, A.J. Heber, J.-Q. Ni, J.A. Koziel, J.M. Sweeten, S.J. Hoff, Y. Zhang, and D.B. Beasley Aerial pollutants emissions from confined animal buildings: Dry sow buildings in Minnesota. 98th Annual Conf. and Exhibition, Paper no. 53. Minneapolis, MN June AWMA, Pittsburgh, PA. Jerez, S.B., Y. Zhang, J.W. McClure, A.J. Heber, J.A. Koziel, S.J. Hoff, L.D. Jacobson, and D. Beasley Aerial pollutant concentration and emission rate measurements from a swine farrowing building in Illinois. 98th Annual Conf. and Exhibition, Paper no. 1026, Minneapolis, MN June AWMA, Pittsburgh, PA. Keck, M., K. Zeyer, and L. Emmenegger Emissions of NH 3 and from closed housing and housing combined with open exercise yards for growing-finishing pigs. In Engineering the future. Proc. AgEng2004 [CD- ROM], Leuven, Belgium Sept Technological Inst., K.U., Leuven, Belgium. Koziel, J.A., B.H. Baek, and K.J. Bush, A. Balota C.L. Bayley, and J.M. Sweeten Emissions of particulate matter from swine finish barns in Texas. In Engineering the future. Proc. AgEng2004 [CD-ROM], Leuven, Belgium Sept Technological Inst., K.U., Leuven, Belgium. Koziel, J.A., B.-H. Baek, C.L. Bayley, K.J. Bush, J.P. Spinhirne, A. Balota, D.B. Parker, J.M. Sweeten, L.D. Jacobson, A.J. Heber, J.-Q. Ni, S.J. Hoff, Y. Zhang, and D.B. Beasley Aerial pollutant emissions from confined animal buildings: Swine finisher buildings in Texas. 98th Annual Conf. and Exhibition, paper no. 1043, Minneapolis, MN June AWMA, Pittsburgh, PA. Lammel, G., F. Schneider, E. Brüggemann, T. Gnauk, A. Röhrl, and P. Wiesner Aerosols emitted from a livestock farm in Southern Germany. Water Air Soil Pollut. 154: Pedersen, S Time-based variation in airborne dust in respect to animal activity. p In Livestock Environment IV, Proc. ASAE, St. Joseph, MI. Pedersen, S., and C.B. Pedersen Animal activity measured by infrared detectors. J. Agric. Eng. Res. 61: Radon, K., E. Monso, C. Weber, B. Danuser, M. Iversen, U. Opravil, K. Donham, J. Hartung, S. Pedersen, S. Garz, D. Blainey, U. Rabe, and D. Nowak Prevalence and risk factors for airway diseases in farmers: Summary of results of the European farmers project. Ann. Agric. Environ. Med. 9: Rieger, M.A Biological hazards at work: specific conditions in livestock farming. In Engineering the future. Proc. AgEng2004 [CD-ROM], Leuven, Belgium Sept Technological Inst., K.U., Leuven, Belgium. Schneider, F., T. Engelhardt, and P.H. Wieser Characterisation of aerosol particles from animal husbandry with single particle analytic techniques. ASAE Annual International Meeting, paper no , Sacramento, CA. 29 July 1 Aug. ASAE, St. Joseph, MI. Seedorf, J An emission inventory of livestock-related bioaerosols for Lower Saxony, Germany. Atmos. Environ. 38: Seedorf, J Emissions and dispersion of livestock-related biological aerosols: An overview. In How to improve air quality. Proc. Dust Conf [CD-ROM; Maastricht, The Netherlands, Apr Animal Science Group, Wageningen UR, Wageningen, The Netherlands. Seedorf, J., and J. Hartung Stäube und mikroorganismen in der tierhaltung. KTBL-Schrift 393, Darmstadt, Germany. Seedorf, J., J. Hartung, M. Schröder, K.-H. Linkert, V.R. Phillips, M.R. Holden, R.W. Sneath, J.L. Short, R.P. White, S. Pedersen, H. Takai, J.O. Johnsen, J.H.M. Metz, P.W.G. Groot Koerkamp, G.H. Uenk, and C.M. Wathes Concentrations and emissions of airborne endotoxins and microorganisms in livestock buildings in Northern Europe. J. Agric. Eng. Res. 70: Takai, H., S. Pedersen, J.O. Johnsen, J.H.M. Metz, P.W.G. Groot Koerkamp, G.H. Uenk, V.R. Phillips, M.R. Holden, R.W. Sneath, J.L. Short, R.P. White, J. Hartung, J. Seedorf, M. Schröder, K.H. Linkert, and C.M. Wathes Concentrations and emissions of airborne dust in livestock buildings in northern Europe. J. Agric. Eng. Res. 70: Takai, H., K. Nekomoto, P.J. Dahl, E. Okamoto, S. Morita, and S. Hoshiba Ammonia contents and desorption from dust collected in livestock buildings. CIGR Ejournal IV:BC USEPA CFR50. Revision to the National Ambient Air Quality Standards for Particulate Matter and Appendix J: Reference method for the determination of particulate matter as in the atmosphere. Fed. Regist. 52:24634, Vranken, E., J. Dekock, M. Guarino, E. Gallmann, E. Hartung, T.G.M. Demmers, P.L. Navarotto, and D. Berckmans Validation of the intermittent measurements method to determine ammonia emissions from livestock buildings. In Engineering the future. Proc. AgEng2004 [CD-ROM], Leuven, Belgium Sept Technological Inst., K.U., Leuven, Belgium. Wang, X., Y. Zhang, and G.L. Riskowski Measurements and analysis of dust spatial distribution in a mechanically ventilated pig building. Biosyst. Eng. 81: Young, P.C Recursive estimation and time-series analysis. Springer-Verlag, Berlin, Germany. Zeitler-Feicht, M.H., G.J. Praml, and H. Riedel Die zeitliche Variation der feinstaubbelastung der luft im mastschweinestall: Tierhygienische und arbeitsmedizinische Aspekte. Berl. Münch. Tierärztl. Wschr. 104: Journal of Environmental Quality Volume 37 March April 2008