Odour Emissions and Dispersion from Digestate Spreading

Abstract

Odour emissions from digestate applied on 21 October 2020 in a 2.4 ha field in the Po Valley (Casalino, 28060, Novara, Italy) were measured using dynamic olfactometry and a six-specialist odour panel, and two application techniques were compared. The measured odour emissions were 3024 and 1286 ou m⁻² h⁻¹, corresponding to the digestate application with surface spreading and direct injection, respectively. The odour dispersion for the different emission values was modeled to a distance of approx. 500 m from the center of the field and 15 m from the ground using a Lagrangian puff model (SCICHEM) in different meteorological conditions. The meteorological variables were measured at the closest station during the whole month in which the digestate application took place, mimicking a “worst-case scenario” characterized by the frequent applications along the considered period. The maximum odour concentrations within one square km area from the center of the field occurred in calm wind and stable atmospheric conditions. This study also evaluated the effect of a barrier downwind from the source. In the worst-case scenario (spreading technique with maximum emissions, no barriers), the average and maximum estimated odour concentrations were 3.2 and 18.9 ou m⁻³, respectively. The calculated probabilities of exceeding the threshold value of 1 ou m⁻³ were 36% and 47% for the whole period and the episodes of calm winds, respectively, and 14% on average for the episode of maximum wind gust. In the best emission scenario (direct injection), the average and maximum odour concentrations were 1.5 and 8.6 ou m⁻³, respectively, while the probabilities of exceeding 1 ou m⁻³ were 26% and 36% for the whole period and the episodes of calm winds, respectively, and 0.016% for the maximum wind gust episode. In the presence of a solid barrier downwind from the source and for the wind gust episode, the peak values of the concentrations and exceedance probabilities at the sampling height were found to be reduced by a factor close to 2.5 and 5 × 10⁵, respectively. The study also evaluated the concentration field’s vertical distribution, showing that the odour plume’s vertical and horizontal dispersion slightly increased with the barrier. This is not a cause of concern unless the emitted substances causing odour nuisance are also atmospheric pollutants with potential harm to far-field ecosystems and human settlements at low concentration levels.

1. Introduction

A sensation from inhaling volatile chemicals (e.g., sulfur, nitrogen, and volatile organic compounds) is called an odour [1]. Anthropogenic activities are mainly responsible for odour issues in the environment, including agricultural and industrial activities, wastewater treatment plants, landfills, livestock buildings, petrochemical parks, foundries, slaughterhouses, composting activities, and paper and pulp facilities [2,3,4,5]. odours adversely affect health, depreciate property prices, and increase annoyance [6,7,8]. Recent studies considered odour as an atmospheric contaminant [9,10]. Odour nuisances are often reported to public authorities in agricultural areas, especially with intense livestock activities.

Odour modeling is a challenging problem from both a scientific and regulatory perspective [11]. Nowadays, odour dispersion is treated with tools similar to those with better characterized atmospheric pollutants. However, some widely accepted assumptions used in air pollution modeling do not hold for odour dispersion [12]. For example, mass conservation is not strictly applicable when assimilating an odour to a transported scalar in a model, and the chemical nature of odourous compounds is seldom known. Additionally, most common atmospheric dispersion models provide mean concentrations at the integration time step, generally 15 min to 1 h, which are much larger than the relevant time scale for odours perception (a few seconds). This problem is mainly related to the intrinsic limitations in the models’ abilities to represent the instantaneous features of turbulent atmospheric flows. It is, therefore, not a straightforward task to address. Several approaches rely on empirical estimations of the output concentration probability distribution and assume a particular value for the “peak-to-mean ratio”, which can possibly vary in time and space [13,14]. Higher-order closure methods are considered more reliable as they derive the concentration variance by solving equations that describe the physical properties of atmospheric turbulence [15].

In addition, the measurements needed to feed odour dispersion models are also affected by several challenges and uncertainties, thus adding complexity to the problem. From the regulatory point of view, institutions and authorities face the challenge of establishing objective criteria and limitations despite the subjective nature of odour nuisances that are mediated by human perception. Despite the difficulties, several approaches and tools have been developed and are currently used to measure, model, and regulate atmospheric odour dispersion [7,16]. Atmospheric dispersion models are often used as a support for authorization decisions and regulation procedures and in case of litigation due to reported odour nuisances. Their use in agricultural areas is nevertheless mostly limited to the livestock sector. At the same time, odour emissions from manure spreading are seldom measured and modeled without any published dataset on the issue.

The present work focuses on odour dispersion modeling, with input data derived from field measurements over the Po Valley, Italy. Northern Italy, and in particular the Po Valley, is one of the areas with the highest levels of intensive farming and concentrations of biogas plants worldwide [17]. The sole Lombardy region in this area accounts for a large part of the total Italian livestock, in particular more than 27% of cattle and 51% of pigs [18]. We used a Lagrangian puff model for the odour dispersion estimation. The simulations were performed for varied wind intensities and considered the effect of a solid barrier downwind from the source. The paper is structured as follows. Section 2 describes the experimental site and the measurements of both the odours and meteorological variables, Section 3 describes the selected model and the simulation scenarios, and Section 4 presents the results of the simulations.

2. Materials and Methods

2.1. The Experimental Site

The experimental site was located in Casalino (45.34598 N, 8.497738 E), Italy, and the surface of the fertilized field was 2.4 ha. This area was characterized by its intensive agriculture, with significant production and use of organic fertilizers. The orography was smooth and the agricultural fields were often contiguous or at a small distance from human settlements. Figure 1 shows the topographic map of the area, with the measurement site highlighted in blue and red (the different colors correspond to the two tested spreading techniques explained in the Section 2.2). The soil was bare during the fertilization event.

2.2. Odour Measurements

The odour emission values used for the simulations were part of a dataset belonging to an experiment whose purpose was to compare the odourous impacts of the different methods of distribution of livestock manure, funded by the EU LIFE PrepAir Project. In this context, three field experiments were carried out in three different locations in northern Italy and in three different periods, where a “local” solution (typically surface distribution followed by incorporation with a harrow) and a “commercial” solution were compared (direct slurry injection via an all-in-one combined spreader (Hydro Tryke by Vervaet, Biervliet, The Netherlands)). For each experimental thesis, five odour emission measurement replicates were carried out for 5–10 min and 4 h after the slurry distribution. The sampling points were chosen using a non-systematic X-shaped sampling (see Figure 2).

The measurements were carried out in accordance with the UNI EN 13725: 2004 standard: “Air quality—determination of odour concentration by dynamic olfactometry”, official Italian version of the European Standard EN 13725 (Edition April 2003) [19]. The surveys were carried out using a stainless steel low speed wind tunnel (LSWT) (0.125 m² surface covered) through which technical air (grade 5.0) was flown at a flow rate of 30 L/min for a period of 5 min. At the end of the flushing period, a nalophan bag for the air sampling was connected to the LSWT via Teflon tubing. The sampling was carried out using a vacuum sampler. The samples were subsequently transferred to the analysis laboratory (Osmotech, Pavia, Italy) where the olfactometric analysis was performed within 12 h using the dynamic olfactometer Scentroid SS600 (Scentroid, Stouffville, ON, Canada) and six-specialist odour panel in a binary forced choice mode to determine the odour concentration (ou_e/m³). Starting with the odour concentration data, the specific odour emission rate (SOER, Ou_e m² h) was calculated as a function of the flow intensity and the surface covered by the LSWT. The emission flux measurements took place on 21 October 2020.

Two spreading techniques were tested: direct injection (blue area in Figure 1 and Figure 2) and surface spreading, followed by incorporation (combined disc harrowing of a 10 cm depth) within 12 h (red area in Figure 1 and Figure 2). In the case of the odour emissions measurements at 0 h and 4 h, the harrowing was not relevant since it was performed after the sampling. For the sake of our study, the second technique should, therefore, be intended as “surface spreading”. Direct injection is commonly indicated as a technique for mitigating odour and other pollutant emissions, and our measurements confirmed this behavior. The measured emissions in ou (odour units) m⁻² h⁻¹ used in the following as an input for the model are shown in

Table 1. Odour emission measurements at the moment of digestate application (t1) and after 4 h (t2) for the two studied application techniques.

Treatment	SOER (ou m−2 h−1)
	t1 = 0 h	t2 = 4 h
Surface spreading	3024.0	2611.2
Direct injection	1373.0	1262.0

. The digestate came from the anaerobic digestion of cattle slurry, poultry manure, and energy crops, and details about its chemical composition and application can be found in Appendix A.

The values shown in Table 1 were in line with previously measured emissions in the Po Valley [5,17]. Considering that a reduction in the emissions was observed after 4 h from the application, we kept the first measurement for the simulations and used it as a constant value throughout the month during which the application took place. This model configuration represented a “worst-case” but still realistic scenario (corresponding to frequent applications on different fields of the same region, during the whole month of October 2020). Therefore, only the data from the column t1 = 0 h of Table 1 were used in our simulations.

2.3. Meteorological Measurements

To model the typical odour emission–dispersion scenarios in the studied region, the simulations were carried out using hourly meteorological data (temperature, wind, humidity, and precipitation) for the whole month of October 2020, during which the slurry application took place. These data were measured at the station “Casello Ruggerina” (Vercelli), 8 km from the experimental site. The meteorological measurements were routinely carried out by Arpa Piemonte—Dipartimento Sistemi Previsionali on a network of stations in the Piemonte region, and they are made available upon request.

Figure 3 shows the time series of the most relevant measured variables for the dispersion model (air temperature, wind speed, and direction), together with the turbulence parameters u* and L derived from the measurements by using the meteorological routines of SCICHEM. Figure 4 shows the average diurnal cycles of the abovementioned variables throughout the month of October 2020. The wind rose in Figure 5a shows that wind most frequently blows from North/Northwest, but the strongest winds most frequently blow from Southeast/East. The wind rose diagram in Figure 5b was drawn using the dataset of the gust events (maximum value recorded by the anemometer in a one-hour interval) and shows a remarkable prevalence of strong wind events from the East/Southeast. In particular, the maximum wind gust event occurred on 3 October 2020, at 2 a.m., with a wind speed value of 13.6 m s⁻¹ and a wind direction of 101 deg from the North. Considering that the odour plume is susceptible to reaching longer distances in stable conditions with moderate to intense wind speeds, we analyzed a nocturnal, statically stable wind gust event and evaluated, in this case, the mitigation potential of a barrier, employing the building downwash algorithm implemented in SCICHEM.

On the other hand, “calm winds” conditions are also relevant for pollutant and odour dispersion. Indeed, in these conditions, the dilution effects of the winds and turbulence are reduced, and the pollutants can accumulate near the source regions. The upper threshold for the “calm winds” detection was chosen according to the requirements of the Italian regulation [20], as the value of the wind for which the model applies a dedicated algorithm for dispersion calculations. The rationale for this choice was that these algorithms are generally less accurate than those used in “normal wind” conditions and could, therefore, affect the final estimation of the odour concentrations and exceedance probabilities. The threshold was chosen to be 0.6 m s⁻¹ as this corresponded to the “calm winds” flag in SCICHEM. According to this definition, 19% of the records during the month of the simulation fell in the category of “calm winds”.

Table 2. Main statistics of the meteorological variables used in the present simulations: minimum, 1st and 3d quartiles, median, mean, and maximum.

Statistics	Temperature (°C)	Wind Direction (° from N)	Wind Speed (m s−1)	Wind Gust (m s−1)
Min.	2.5	0	0	0.8
1st Qu.	9.2	81.5	0.7	1.8
Median	11.9	205	1	2.4
Mean	11.91	186.7	1.238	2.921
3rd Qu.	14.5	288	1.4	3.1
Max.	22.7	357	6.9	13.6

summarizes the main statistics of the principal variables used by the model in our simulations. The values were in line with the typical meteorological conditions of the region and the considered season.

2.4. Odour Dispersion Modeling

2.4.1. The Selected SCICHEM Model

The atmospheric dispersion model suitable for odour modeling was selected based on the requirements and preference criteria listed below.

• In line with the Italian regulations, particularly that of the Lombardy Region that commissioned the study [20].
• In line with the standard UNI 10796:2000 that specified the classes of the models that are recommended as applicable to the problem of odour nuisances (i.e., non-stationary puff models/segment models, 3D Lagrangian puff or particle models, 3D Eulerian models).
• Spatial and temporal scales compatible with the problem under analysis.
• Transport and turbulent schemes adequately representing the peculiarities of odour dispersion.

The preference criteria for selecting SCICHEM were as follows.

• Able to represent the aerodynamic effect and mitigation potential of the barriers.
• Able to represent the inhomogeneous surfaces and the effects of canopies.
• Actively used and maintained by the scientific community, multi-platform and open source.
• User friendly and with low–moderate computational resources requirements for easy use in non-academic contexts (e.g., administrative/regulatory).

SCICHEM, which stands for SCIPUFF (Second-order Closure Integrated Puff) with CHEMistry [21,22,23,24] was selected after reviewing more than 50 models with similar purposes [7,25,26] and several methodological guidelines that often contained the models’ categorization and reviews [27,28,29]. More details about the selection criteria and the analyzed models are provided in Appendix B.

SCICHEM, SCIPUFF, and other Lagrangian puff dispersion models calculate the pollutant concentration from the superposition of three-dimensional puffs, which can be viewed as packages of material released from the source, advected by the mean wind, and dispersed according to the turbulence. The representation of the dispersion can be analogous to that of classic Gaussian plume models, with more or less empirical parametrizations of the dispersion coefficients. At the same time, advection (advance in the puff centroid) is coupled with a puff splitting/merging scheme that ensures adequate resolution and mass conservation. The governing equations of the SCICHEM model were widely documented [22,30,31]. Therefore, the present work only highlights the main features of the SCICHEM model that are particularly suitable for odour modeling (i.e., flexibility in meteorological input requirements and second-order closure scheme for turbulence).

Concerning the meteorology, SCICHEM is very flexible in the required input, since it embeds a series of meteorological routines that estimate/calculate all the necessary boundary layer variables when they are not provided in input. The minimum requirement is a single wind vector, but when available, the model can consider observations from a network of sensors measuring the turbulence characteristics, boundary layer height, and other variables in the form of vertical profiles or gridded data typically provided by meteorological models or pre-processors. Additionally, SCICHEM can use the specified topography to interpolate the measured fields and ensure mass conservation. Our case is somehow intermediate, and the meteorological variables at our disposal consisted of the following.

• Mean wind (intensity and direction)
• Wind gust (intensity and direction)
• Air temperature

All the variables above were provided as hourly means from a measurement height of 10 m for the wind variables and 2 m for the temperature. The air temperature used in the boundary layer calculations of the model should be representative of the surface layer mean temperature. Without additional measurements, we considered that the measurement at 2 m was adequate for the purpose of this study. The stability parameters, such as the friction velocity u* and the Monin–Obukhov length L, are estimated from the input variables (considering also the geographical location, date, and time of the day for the estimation of radiation), with an iterative technique that started from a “first guess” value for L. An iterative scheme is needed since L depends on u* and u* on L [32]. Details on this methodology are explained in the technical documentation [33], in particular the section 10.1.1 “Monin–Obukhov Length, L” and 10.1.2 “Friction Velocity, u*”. The parameters L and u* that were derived in this way for our simulations are reported together with the measured meteorological variables in Figure 3 and Figure 4.

The atmospheric turbulent dispersion in SCICHEM is modeled using a second-order closure scheme, which allows for simulating the variance of the concentration field. This renders SCICHEM particularly suitable for modeling odours and other substances where the short time fluctuations are equally or more relevant than the average values of highly toxic, flammable, and explosive materials [34].

The main equations governing the puff dynamics can be derived from the conservation principles and can be expressed in terms of the first and second spatial moments of the concentration.

$$\frac{d{\overline{x}}_i}{dt}={\overline{u}}_i\left(\overline{x}\right)+\frac{\langle \overline{u_i^{\prime }{c}^{\prime }}\rangle }{Q}$$

(1)

where $x_i$ are the coordinates of the puff centroid, ui represents the local atmospheric velocity field, and Q is the total mass (integrated concentration of the puff).

The second-moment equation can be obtained by multiplying Equation (1) by x’i x’j, with x’i = xi − <x> i, and integrating it in parts

$$\frac{d{\sigma }_{ij}}{dt}={\sigma }_{ik}\frac{\partial {\overline{u}}_j}{\partial x_k}+{\sigma }_{jk}\frac{\partial {\overline{u}}_i}{\partial x_k}+\frac{\langle x_i^{\prime }\overline{u_j^{\prime }{c}^{\prime }}\rangle }{Q}+\frac{\langle x_j^{\prime }\overline{u_i^{\prime }{c}^{\prime }}\rangle }{Q}$$

(2)

where $x_i$ are the coordinates of the puff centroid and ${\sigma }_{ij}$ is the matrix of the second moments that describe the spatial dispersion and distortion of the puff. As can be seen from Equations (1) and (2), the third-order moments appear in the equations for the first- and second-order moments. This is the typical “turbulence closure problem” introduced by Reynolds that averages the conservation equations and is addressed in SCIPUFF/SCICHEM by parametrizing the third-order moments as in [35]. This allows the system of equations to be closed.

The final concentration at each point is calculated as a superposition of the contributions from all the puffs in that point, eventually considering the following contributions: the reflections at the top and bottom boundaries, chemical reactions, deposition, and background species concentrations. The concentration variance is also provided in the output using the second-moment equation. The model then obtains the probability distribution by assuming a clipped or truncated Gaussian shape that is obtained from the Gaussian normal distribution by replacing the non-physical negative values with zero values. The atmospheric observations show that this distribution is the most suitable to describe the short-term atmospheric fluctuations [36,37].

Another aspect that made SCICHEM particularly suitable for our purposes was the great flexibility in the input data requirements. The model can be run either with a minimal set of simple meteorological data, most likely for routine monitoring/regulatory purposes, but also with complex measurements derived from state-of-the-art measurement and modeling techniques (mostly for research purposes).

In addition, SCICHEM can be used from the local (<1 km) to the continental scale and it allows for the simultaneous activation of multiple sources, chemistry simulations, dry and wet deposition, and can account for the effect of the barriers using the PRIME (plume rise and building downwash) scheme [38].

The PRIME algorithm allows for the evaluation of the potential impact of solid obstacles on the pollutant concentrations, simulating the aerodynamic perturbations of the wind and turbulence fields. The barrier induces multiple disturbances on the wind and turbulence fields. First, it decreases the horizontal wind speed at low heights, both upwind and downwind, while it deflects the wind streamlines by adding a vertical component to the wind on the windward side and a negative component on the leeward side (plume rise and downwash effects). An increase in the turbulent kinetic energy comes along with the reduction in the mean wind as a result of large atmospheric eddies being broken into smaller eddies by the obstacle (see [39] and references therein).

The development of the tool was initially motivated by the search for criteria to establish minimum stack heights in urban and industrial contexts. Therefore, it was designed for point sources. At present, the models implementing PRIME (AERMOD, ISC, SCIPUFF/SCICHEM) do not allow for simulations with barriers/buildings in the presence of the area or volume sources. For this reason, when evaluating the effects of the barrier on odour dispersion, the field source was replaced by a point source of a large diameter (40 m). SCICHEM can indeed attribute a finite diameter to point/stack sources and specify their height (0 in our case). We used the same total emission as the one measured from the field in the “best-case scenario” to mimic the combined effect of the injection and the presence of the barrier to mitigate the dispersion.

To our knowledge, this is the first application of the SCICHEM model in the framework of odour dispersion.

2.4.2. Selected Simulation Scenarios

Four different scenarios were simulated, and for each of these, different wind and stability conditions were analyzed. The simulation scenarios referred to the different fertilizer application techniques (thus different emission values) and the eventual presence of a barrier.

Table 3. Description of the simulation scenarios.

Scenario Name	Fertilizer Application Technique	Emissions	Notes
Scenario 1	Surface spreading	Per unit surface: 3024 ou m−2 h−1	Area source
		Total: 2 × 104 ou s−1
Scenario 2	Direct injection	Per unit surface: 1286 ou m−2 h−1	Area source
		Total: 8.5 × 103 ou s−1
Scenario 2bis0	As Scenario 2	As Scenario 2	Area source replaced by a “point” (stack type) source of 40 m diameter and zero height
Scenario 2bis	As Scenario 2	As Scenario 2	As Scenario 2bis0 + barrier

summarizes the characteristics of each scenario: the application technique, emission intensity, and eventual presence of a barrier. As already mentioned, the presence of a barrier could only be simulated in SCICHEM using point sources. Therefore, in addition to the “real scenarios” (1 and 2), two hypothetical scenarios (2bis and 2bis0) were simulated, where the source was replaced by one type “point” with a relatively large diameter (40 m) and the same total emission as scenario 2. In this way, we tested the mitigation potential of the combined effect of direct injection (reduced emissions with respect to scenario 1) and the barrier. Using a very large diameter with a point source posed numerical problems. Therefore, the total area of the source was smaller than the real field.

SCICHEM is also very flexible in the output format, which can be provided at a maximum of 500 receptors and whose location can be specified by the user. In our simulations, the odour concentrations were simulated for a grid of 484 receptors uniformly distributed on a 1 km × 1 km region centered on the fertilized field for scenarios 1 and 2, and a 200 m × 200 m region for scenarios 2bis0 and 2bis. Additionally, vertical slices of the concentrations were simulated by placing the 484 model receptors along a transect crossing the fertilized field in the East–West direction, passing by the field’s center and reaching a 15 m height. In the horizontal field simulations, the receptors were placed at a 1.7 m height, representative of human perception. The exceedance probabilities of the threshold value of 1 ou m⁻³ were also calculated for the same horizontal and vertical networks of the model receptors. The value of 1 ou m⁻³ is the lowest among the three thresholds identified by the regional regulation on odour monitoring: 1, 3, and 5 ou m⁻³, corresponding to the concentrations at which 50%, 85%, and 90–95% of the population perceive the odour sensation [20]. The choice of the lowest threshold represented a “worst case scenario” in terms of the exceedance probabilities.

Acknowledging the limitations of representing a field with a point source in the model, we adopted the following measures to mimic the real scenario as closely as possible.

• The diameter of the point source was set as large as possible (40 m diameter, about 0.1 ha), compatibly with numerical stability.
• The height of the point source was set to zero.
• The ejection velocity and buoyancy effects, typical of a point/stack source, were set to zero as the field was a passive source.
• The total emission value was set equal to that of scenario 2.
• The meteorological data were those of the wind gust event occurring in scenario 2.

The choice of the nocturnal wind gust event was motivated by the fact that, in these conditions, the pollutant/odour plume is susceptible to reach larger distances and the barrier to affect the plume transport and dispersion to a greater extent.

The other relevant configuration parameters (common across all the scenarios) were as follows.

• The fertilized field shape (Figure 1 and Figure 2) was approximated by a square of the same area as that of the experimental field.
• The mass effects (inertia, gravity) of the transported quantity were not considered.
• The chemical reactions were not considered.
• The deposition was switched off.
• The surface roughness length was set to 0.01 m (rough, bare soil).
• The maximum number of receptors allowed by the model is 500, therefore, we set a grid of 22 × 22 equally spaced receptors across the domain.
• The output time step was set to 15 min.

For scenarios 2bis0 and 2bis, where the field was replaced by a circle source of 40 m diameter, the total domain was reduced to 200 × 200 m, zoomed around the source. For these scenarios, vertical profiles were also simulated. Therefore, in this case, a vertical grid was defined from the ground up to a 15 m height at a constant latitude corresponding to the center of the circular source. For the vertical slices, 22 points were used on the x-axis and 22 vertically to get a good resolution of the zoomed region.

3. Results

3.1. Modeled Odour Concentration Plume

To analyze an event of potential odour nuisance reaching significant distances from the source, we chose the date and time (3 October 2020 at 2 a.m.) when the maximum wind gust speed was registered (see Section 2.3. Meteorological Measurements).

The simulation domain extent was chosen considering that concentrations larger than 1 ou m⁻³ were observed up to approx. 500 m from the center of the source field (Figure 6). On average over the whole period, the choice of this domain size led to a boundary/peak concentration ratio of 5%. For the 2bis0 and 2bis scenarios, the domain was reduced accordingly, considering that the source extent was smaller in this case.

Figure 6 shows that the maximum concentrations along the whole period (a), for the case study of 3 October 2020, at 2 a.m. (b), and during the episodes of calm winds (c) were observed in correspondence with the source and over a roughly elliptical area oriented towards the direction of the prevailing winds (see Figure 5). Averaging only the calm winds episodes, the maximum concentrations increased by a factor of 1.5 due to accumulation near the source. The whole period concentrations and the concentrations during the calm winds decreased by a factor of 0.2 at approx. 500 m distance from the source. The above-threshold values (>1 ou m⁻³) were also observed at these distances due to the large emission area and the high value of emissions in this scenario. As for the concentrations observed during the episode of a maximum wind gust, these were generally lower (see Figure 6a,c), likely due to the increased dilution by the transport and turbulence. However, the reduction in the concentration values with the distance was much less than for the calm or average winds. The downwind concentrations at the domain border were still above 1 ou m⁻³, a value that was only reduced from the peak value at the center of the field by a factor of 0.7. These differences were reflected in the probabilities of exceeding the chosen threshold (1 ou m⁻³). The mean decreased to approx. 0/0.4 close to the borders of the simulation domain. The gust episode remained close to one and 500 m downwind from the source (Figure 6d–f).

The vertical slices of the concentrations and exceedance probabilities are shown in Figure 7 for the same meteorological cases shown in the horizontal maps. The maximum concentrations (up to 25 ou m⁻³) were observed for larger heights (up to 2.5 m) in the case of the calm winds compared to both the average period and the wind gust event. The effects of accumulation near the source is visible in Figure 7 and is related to the reduced advection in these meteorological conditions. The shape of the plume was strongly modified by the wind gust, as already observed in the horizontal section, with more pronounced vertical gradients and decreased horizontal gradients in the direction of advection.

3.2. Effect of Stability

Figure 8 and Figure 9 show the SCICHEM results for scenario 1, disaggregated by the stability conditions. The disaggregation was achieved by filtering the output fields on the basis of the model-calculated Monin–Obukhov length L, shown in Figure 3 and Figure 4. Considering the limited amount of input data and the approximations made in the model in order to calculate L, we only differentiated between the stable (L > 0) and unstable (L < 0) cases without going into finer disaggregation (e.g., according to Pasquill stability classes). Both figures show the highly relevant impact of the stability on the concentration fields, with strong dilution effects induced by the daytime convective turbulence. The horizontal maps (Figure 8) show that, at the selected receptors’ height (1.7 m), the maximum concentration was reduced by approx. one order of magnitude in the unstable conditions. Figure 9 shows that the plume was more diluted, with lower vertical gradients in the unstable cases, indicating the effect of the increased vertical dispersion.

3.3. Effect of Reduced Emissions

Figure 10 corresponds to scenario 2 with lower emissions and shows the same patterns as Figure 6, with decreased (roughly halved) concentrations in all the meteorological conditions. The exceedance probabilities were also reduced on average by factors of 1.4 and 1.3 for calm winds and 2 and 3 for the wind gust episodes. Notably, the relationship between the exceedance probability and the concentration isnon-linear and strongly depended on the selected threshold.

Table 4. Odour concentration and exceedance probability statistics for scenarios 1 and 2: minimum, median, mean, 1st and 3d quartiles, maximum. The statistics are provided for the whole period, for the wind calm conditions, and for the wind gust episode.

	Concentration [ou m−3]			Exceedance Probability [-]
	Whole Period	Calm Winds	Wind Gust	Whole Period	Calm Winds	Wind Gust
Scenario 1
Min.	0.12	0.14	0	0.04	0.07	0
1st Qu.	0.79	1.08	4 × 10–5	0.19	0.31	0
Median	1.90	2.50	0.01	0.33	0.46	0
Mean	3.23	4.48	0.37	0.36	0.47	0.14
3rd Qu.	3.85	5.33	0.27	0.46	0.61	0.03
Max.	18.86	26.80	3.90	0.96	0.98	0.98
Scenario 2
Min.	0.05	0.06	0	0.02	0.02	0
1st Qu.	0.33	0.44	1.93 × 10–5	0.11	0.17	0
Median	0.83	1.13	2.5 × 10–3	0.24	0.34	0
Mean	1.46	2.01	0.17	0.26	0.36	0.06
3rd Qu.	1.76	2.38	0.12	0.37	0.5	3 × 10–7
Max.	8.55	12.03	1.66	0.81	0.88	0.84

summarizes the main statistics of both the concentrations and exceedance probabilities over the whole domain for both scenarios and under the considered meteorological conditions (average, calm, wind gust).

3.4. Effect of Barriers

As explained above, the PRIME model embedded in SCICHEM and in other dispersion models does not simulate barriers for area or volume sources. For this reason, the emitting area of the fertilized field was replaced by a source of type “point” but with a finite diameter. The use of a very large diameter that would allow for the representation of the same area as the real field nevertheless posed some numerical problems. Therefore, the area of the source was reduced to a diameter of 40 m but maintained the same total emission value of scenario 2.

This replacement allowed for the activation of the PRIME module in SCICHEM and, therefore, the quantification of the effect of a barrier by comparison of concentration fields with and without it, for the hypothetical circular source (scenarios 2bis0 and 2bis as defined in Table 3). The effect of the barrier was studied only in the event of the wind gust and placed 1 m downwind from the western edge of the field, which was normal to the incident wind direction. The barrier width (along-wind), length (across-wind), and heights were 1, 40, and 10 m, respectively. The emissions were those corresponding to the application technique that already significantly reduced the emissions and, consequently, the dispersion of odours. In this way, the combined abatement potential of the application technique and the presence of a barrier was evaluated.

The choice of the nocturnal wind gust event was motivated by the fact that during stable and high wind speed conditions the pollutants/odour plume is susceptible to reach the greatest distances, as demonstrated in Figure 6 and Figure 10 (compare panel d with a,b). Additionally, in this condition, and for a barrier placed perpendicularly to the wind direction, its effect was expected to be maximum. As shown in Figure 11 and Figure 12, placing a barrier downwind of the source and perpendicular to the wind direction reduced the maximum odour concentrations approximately by a factor of 2.5. The exceedance probability close to the source was more drastically reduced by five orders of magnitude in its maximum value and became essentially zero on average (Figure 11 and Figure 12, and

Table 5. Odour concentration and exceedance probability statistics for scenarios 2bis0 and 2bis: minimum, median, mean, 1st and 3d quartiles, maximum. Scenarios 2bis0 and 2bis are simulated for the wind gust episode only.

	Scenario 2bis0		Scenario 2bis
	Concentration [ou m−3]	Exceedance Probability [-]	Concentration [ou m−3]	Exceedance Probability [-]
Min.	0	0	2.3 × 10–4	0
1st Qu.	2.99 × 10–5	0	0.01	0
Median	7.1 × 10–3	0	0.09	0
Mean	0.13	0.03	0.15	6.79 × 10–9
3rd Qu.	0.18	4.9 × 10–5	0.28	0
Max.	1.20	0.67	0.50	1.26 × 10–6

). The very large difference between the exceedance probabilities with and without a barrier rendered the exceedance probabilities indistinguishable from 0 in Figure 12 (a common color scale was used between Figure 11 and Figure 12 for both the concentrations and exceedance probabilities).

These results highlighted the high potential of using barriers to reduce the odour nuisance at the considered height (1.7 m, representative of the human sampling height) in the vicinity of the source, especially if combining them with techniques that favor the capture or deposition of the pollutants upwind or within the barrier itself. Several factors influence the effect of the barrier, including its physical features, location, orientation with respect to the wind, and the physicochemical properties of its material.

The plume appeared slightly more dispersed horizontally in the scenario 2bis (with a barrier), and this was reflected by the statistics of the concentration all over the domain (Table 5). Despite the maximum concentration values being drastically reduced with the barrier, the mean values were slightly higher in this scenario.

Our results show a significant decrease in both the concentration and exceedance probability at the sampling height with the barrier. Nevertheless, considering that the deposition was switched off in our simulations, these results were likely related to a different spreading of the odour in the atmosphere, as also suggested by the statistics across the simulation domain (Table 5). To investigate the fate of the simulated odour, we performed additional simulations by placing the receptors on a vertical slice parallel to the x-axis (along-wind direction for the wind gust event). The slice crossed the simulation domain of scenarios 2bis0 and 2bis at the location of the source center and reached 15 m in height. The results are shown in Figure 13 and Figure 14. These figures confirmed the overall reduction in the concentrations in the vicinity of the source but also highlighted the higher level of the horizontal and especially vertical dispersion induced by the barrier. Similar to the horizontal graphs, the very large difference between the exceedance probabilities with and without a barrier rendered the exceedance probabilities indistinguishable from zero in Figure 14 (a common color scale was used between Figure 13 and Figure 14 for both the concentrations and exceedance probabilities). The barrier indeed induces multiple perturbations on the wind and turbulence fields, decreasing the horizontal component of the mean wind, introducing a vertical velocity component by deflecting the wind streamlines, and increasing the turbulence and turbulent mixing. The algorithm PRIME within SCICHEM considers all these effects and allowed us to appreciate the increased dispersion that occurred due to the reduction in the odour concentration at the 1.7 m height. If the substance responsible for the odour presents no danger to human health or ecosystems, the increased dispersion and concentration reduction would not represent a cause of concern and the barrier could be considered effective for the reduction in the odour nuisance.

Nevertheless, caution should be used when implementing barriers as a mitigation tool, especially if the emitted gases or particles could harm the environment and human health at very low concentrations in the far field and/or if a significant increase in the pollutant concentration in the immediate vicinity of the barrier occurs close to the ground. A similar concern arises in the industrial and urban context for areas close to stacks and buildings (building downwash). In these cases, optimal solutions could be vegetated barriers or specific materials that favor the deposition/capture of the pollutant, resulting in a net removal of the pollutant from the atmosphere. It should be noticed, however, that vegetation and porous structures in general do not act on the flow as non-permeable barriers, and this should be considered when selecting the modeling approach.

The PRIME algorithm was initially designed in the industrial context, but it contains all the physical mechanisms (plume rise, increase in turbulent mixing, etc.) induced by a barrier. Therefore, it is also applicable in the agricultural context with ground and passive sources. SCICHEM and PRIME could, therefore, be powerful tools in designing opportune barriers to reduce the odour nuisance, especially when the prevailing wind direction of the region/season and fertilizer application is known. It would be desirable to further develop the PRIME algorithm within SCICHEM in order to identify when the plumes are generated by area (and not only point) sources.

4. Conclusions

This work showed how dispersion modeling with a state-of-the-art research tool, such as the SCICHEM model, can support institutions in predicting and regulating odour nuisances while contributing to scientific research purposes and developing promising synergies between academic and regulatory works. The framework of the present project consisted of the Italian and international regulations on odours, and SCICHEM demonstrated the ability to inform and support decision-makers in estimating and regulating the odour concentrations and exceedance probabilities of specific thresholds. On the other hand, the use of the model in this context can potentially trigger new model developments for regulatory applications.

The results of our simulations, combined with the measurements of odour emissions using dynamic olfactometry, showed that a typical application of organic manure to an agricultural field of 2.4 ha can produce odour concentrations that exceed the minimal threshold identified by the national reference law up to a 500 m distance from the source. The most critical situation occurred in the vicinity of the source, under calm winds, stable conditions, and following the surface spreading method. The direct injection technique was highly performant in reducing the overall emissions and atmospheric concentrations. For intense winds, the plume elongated more than under light winds but with lower values of concentrations, as expected from the enhanced advection and mixing in these conditions.

The probability of exceeding the defined threshold of 1 ou m⁻³ was close to 0.7 on average at a distance of 500 m from the center for the field using the surface spreading technique, potentially raising a cause of concern from a regulatory point of view. On the other hand, we considered a very conservative scenario that included the following:

• The chosen threshold was the lowest considered by the Italian regulation.
• The emissions was considered constant during the whole month, thus constituting a “worst-case scenario” where fertilizer applications are either replicated very frequently or performed in neighbors’ fields almost at the same time and for an extended period.

Considering the total emission rate (Table 3), we estimated a concentration/source ratio C/Q(s m⁻³) in the direction of the prevailing wind. At approx. 400 m from the source, this ratio ranged between 10⁻⁶ s m⁻³ (unstable cases, see Figure 8) and 10⁻³ s m⁻³ (calm winds and stable conditions, see Figure 6 and Figure 8). Our results were in line with previous findings ([40], Table 3) where the concentrations/source ratios were measured and simulated using different models. Even if the classification of the stable/unstable cases was finer in [40], the results were consistent with our C/Q ratios, being of the same order of magnitude, smaller for unstable cases, and larger for stable cases and calm winds.

The direct injection technique showed great mitigation potential, with a reduction in both the emissions and concentrations by a factor of two on average.

The concentrations observed during the calm wind episodes increased by a factor of 1.5 in their maximum values, while during the maximum wind gust event, the peak was reduced approximately by a factor of 4 as the odour plume reached greater distances. Additionally, the atmospheric stability had a great impact on the odour dispersion, with concentrations in the stable conditions being on average one order of magnitude larger than in the unstable conditions in the considered domain and at the sampling height of 1.7 m.

The PRIME algorithm in the model also allowed for the estimation of the effect of a barrier downwind from the source during the maximum wind gust event, showing a further reduction of approx. a factor 2.5 for the peak concentrations (close to the source), which became essentially zero for the exceeding probabilities averaged across the simulation domain. Even though this scheme is available only for point sources at present, we designed the source in the simulations to mimic the real field and the presence of the barrier as close as possible. The generalization of the algorithm for the area sources was identified as a promising perspective, particularly for applications in the agricultural context.

The results for the circular source with and without a barrier showed the strong potential of reducing odour concentrations in the vicinity of the source at the sampling height of 1.7 m. A vast literature exists on the potential of barriers in reducing the atmospheric dispersion of pollutants, both in urban/industrial and agricultural contexts [41]. Our results confirmed the promising mitigation potential of barriers against odour nuisances, especially if combining them with techniques that favor the capture or deposition of the pollutants upwind or within the barrier itself.

At the same time, our results showed how this reduction occurred along with the increased vertical mixing and dispersion. Optimal barriers for mitigation purposes should, therefore, be designed by considering several factors, including all the pollutant’s characteristics and the region’s prevailing meteorological conditions. If the odour compound emitted by the source is not harmful to human or ecosystem health, a slight increase in the vertical dispersion might be desirable as it reduces the level of human perception of the nuisance without causing major concern in the far-field region. On the contrary, if the substances causing the odour nuisance are atmospheric pollutants, the increase in horizontal and vertical dispersion might be a cause of concern. In these cases, better solutions are most likely vegetated barriers that can capture the particles/gases responsible for the odour nuisance. The combined effect on the wind, turbulence fields, and on the deposition of the pollutant would maximize the mitigation potential in these cases. Nevertheless, as long as only the odour nuisance is considered, the mitigation potential of the simulated barrier shown by our results was highly satisfactory.

Our results globally show the potential of using SCICHEM in the agricultural and regulatory context, particularly regarding odours, for its capability of simulating the exceedance probabilities and its implementation of the PRIME algorithm, since the use of barriers (either artificial or natural) is becoming a common practice to mitigate the dispersion of pollutants from various agricultural activities.