2.1 Case Study

The focus of the case study is a real Uruguayan dairy farm, where the farmer (producer) keeps records of both management and productivity, what enabled to make the N and P budgets and to develop a reliable calibrated ABM for this case. The database analysis was complemented with interviews and farm’s visits to gather further, required information. As is represented in Figure 1, the farm is located at the southern limit of the department of Soriano, Uruguay, in a historic dairy farm region -this means that the focus farm is situated in a Uruguayan representative dairy region³⁸.

This dairy farm has produced artisan cheese since 1973, spanning several generations of the same family. Since its beginning, they have expanded the area, increased the number of cows, and boosted production volume. The family currently consists of five members, all of whom are involved in both farming and cheese production. In addition to the family owners, thirteen workers are employed in the dairy farm and cheese production.

Figure 1: Farm location³⁹  

The 2020 case study indicators summary is represented in Table 1, compared with INALE typology to establish a benchmark. It is situated in the Uruguayan typology groups with higher milk productivity, represented as the H1 and H2 groups. The difference between these groups is that H1 farms have a pasture intake minor to 3000 kg DM/ha affected by milking and dry cows, while H2 farms have a pasture intake of more than 3000 kg DM/ha⁴⁰.

The farm under study can be considered a typical Uruguayan dairy farm because it is characterized by a pastoral-based diet, with grass and legume pastures as the main diet’s components, and it is on a medium scale referring to the farm size in ha³.

The farm has an extension of 260.5 ha divided into two distinct areas that are in turn subdivided into multiple paddocks. The area used by the dairy cows is named tambo (milking yard) and has 126 ha. The other area of the farm is composed of four sectors used by the rest of the animals and has 134.5 ha (called “support area”). It is worth remembering that parts of the feed reserves (meadow bales) and concentrate (wet corn grain silage) are produced in the support area. Because of this, those feed reserves are considered inputs to the studied and modeled system. The tambo area and its paddocks are represented in Figure 2.

Figure 2: Tambo area and its paddocks³⁹  

The farm under study is part of a local group of dairy farms called “Los 30,” located in the departments of Colonia, San José, and Soriano in a dairy region of Uruguay⁴¹. These dairy farms primarily produce artisan cheese, and the group is coordinated by an agronomist, a veterinarian, and a sociologist. This is an important point to highlight because the research involved communication with the farmers’ group to support peer learning.

2.2 Nutrients Budgets

To begin a nutrient assessment of the farm system, it’s important to note that, unlike greenhouse gas emissions regulated by the IPCC, there is no international legal framework for N and P budgets¹⁶. Consequently, there is no standardized international terminology or methodology for developing nutrient budgets, leading to varying terms used by different authors. However, according to Gourley and others³¹, a nutrient budget is defined as “an accounting approach for nutrient inputs, stores, and outputs.” Similarly, Cherry and others¹⁴ describe nutrient budgeting as a valuable and straightforward tool for generating, analyzing, and evaluating farm inputs and outputs. Farm-gate budgets are the most common, as they are easier to calculate using available farmer data and reliable sources¹⁴.

In the case of N, it is generally accepted that the main entrance of this nutrient to the agroecosystem is N mineral fertilizer, fixation of atmospheric N by legumes and soil bacteria, deposition from the atmosphere, and livestock feed. There are also minor inputs such as animal bedding and seeds¹⁶. N is retained in animal products, milk and meat, crop products, and exported manure¹⁶. Furthermore, N outputs are the losses to the environment by volatilization, denitrification, and lixiviation. According to Eurostat¹⁶, gaseous losses occur from effluent and manure management systems (EMS) and paddock management.

In reference to P, the inputs are mineral fertilizer and livestock feed. Like N, the P outputs can be divided into animal products, crop products, and environmental losses¹⁶. Nutrient losses to the environment can be separated into punctual and diffuse losses. While punctual refers to the loss from the EMS, diffuse refers to the paddock losses²⁶.

Based on the literature reviewed, the main inputs, outputs, and terms used for the nutrient budget developed in the present research are, as P and N input: mineral fertilizer, imported livestock feed, and BNF; and atmospheric deposition for N only. As N and P output we considered milk and meat production. For environmental N losses (at paddocks and EMS) we took into account the N₂O direct losses, volatilization, and leaching. For P environmental losses we considered leaching for EMS, and particulate and soluble runoff at paddocks.

In our work, the nutrients budgets for 2020 were generated using the farm’s records and bibliographic review. When using other research, assumptions were made to estimate the data needed to calculate the budget variables. The data, assumptions, and terminology used in the study are presented in Table 2.

To evaluate productive alternatives and their impact on nutrient budgets, some management scenarios were discussed and validated with the farmers’ group in two online meetings held in 2021. The research formulated three farm management scenarios to evaluate possible alternatives and solutions to the problem situation with the nutrient budget method. These scenarios intend to answer the questions: which is the impact of the different dairy farm management practices on the nutrient dynamics, and how those practices can reduce nutrient losses to the environment?

The management scenarios evaluated in the research are:

It is important to note that the annual nutrient budget for current situation and scenarios are a static approach. Therefore, ABM provides a complementary method to recreate dynamically both the spatial and temporal distribution of N and P on the farm, linked with farmers grazing and feeding management practices.

2.3 NPM Agent-based Model

The ABM simulation was developed on the NetLogo platform version 6.2.⁵³, a multi-agent programmable modeling environment suitable for modeling complex systems.

The development of the model in the case study was also based on the register’s records provided by the dairy farmer and from literature review. It was adjusted using the nutrient budget previously described (see Section 2.2). The objective of building the ABM is to dynamically recreate as well as to explicitly represent the spatial distribution of N and P across farm paddocks, as well as the EMS. This model helps to visualize annual dynamics that cannot be captured by the nutrient budget methodology previously described.

The farmer’s registers and monitoring corresponded to year 2020, which included (on a monthly scale) the number of animals, milk composition and production, diet profile (proportion of pastures, reserves, and concentrates), and land use (paddock land use records). Only the pasture area of the case study was modeled (see 2.1 section). The central dynamics of feeding, grass management, and milking in the ABM carried out a daily step (see Table 2).

The allotment of animals to paddocks was discussed and validated with the farmer as the main grazing decision rule. The ABM was implemented assigning the animals to the pasture paddock (set of patches at the ABM) with the highest average biomass, keeping the cows in that paddock up to a remnant of 1500 kg DM/ha. The animals consume the DM of each diet component according to the monthly diet profile records provided by the farmer. N and P excretes at each step were proportionally assigned to time spent at pasture paddocks and milking parlor (20 h/24 h, and 4 h/24 h, respectively). The ABM model calculates daily the N and P budget for each patch. Also, the different pastures have an individual daily growth rate, and have a protein (N) and P content. The farmer provided the soil P content by soil analysis information that was used at initialization, as described below.

The ABM is presented in the next paragraphs, as suggested by Grimm and others⁵⁴, using a summarized version of ODD (Overview, Design concept, and Details) protocol:

Purpose - The overall purpose of the NPM model is to evaluate dynamically dairy farm management practices that influence N and P budget and its spatial distribution. Specifically, the model addresses the following questions:

• According to farming practices in the case study, which is the spatial and temporal pattern of nutrients at the modeled system?

• Which is the contribution of each nutrient budget component (excreta, paddock environmental losses, effluent environmental losses, and biological N fixation) to whole system results?

Entities and states variables - The model includes the following entities: Patch, Cow, EMS. The Entities and States Variables of NPM are described in Table 2. The spatial and temporal resolution and extent: a time step in the model represents one day, and the simulation can be for the year of evaluation, emulating the 2020-year farm performance. NPM model is spatially explicit and represents the area of the dairy farm case study. Simulations are based on 13 paddocks with different extents each, representing a total area of 126 hectares. The landscape or “world” is a grid of 50×20 patches with 2600 m² each, but only 846 patches represent the tambo area, to fit to GIS information provided by the farmer.

Process overview and scheduling - The core process of the NPM model is the nutrient budget in each patch. This process is repeated every time step and is based on these other processes: pastures growth according to sown species, nutrients dynamics and presence, and permanence of cow in the patch. It is important to highlight that the NPM model represents the scale of colors according to the variable value of each patch, which helps state visualization. Also, the interface represents the patches average variable value of the same paddock, as presented in the Results section.

Scheduling is very significant in the NPM model because it represents the farmer management. The scheduling refers to number of dairy cows, milk production and composition, dairy cows’ diet, fertilizer application, and pasture growth rate, according to farmer registry information.

Design concepts - The most important design concepts of the model are:

• Objective: The design model objective is to represent the farm case study, using GIS and land use information.

• Observation: The number of cows, the animal diets levels, the cow’s production, and the total amount of biomass in each paddock are recorded daily.

• Communication: The main communication variables are represented in the model interface. Its interface includes the use of “on/off switch” from each nutrient budget component, the N and P use efficiency (NUE and PUE, respectively) graphs, and the dynamics of the N and P separately in paddock and EMS. The interface represents in a color scale the values of the main variables of N and P dynamics. These specific design aspects are to facilitate collective learning and communication between farmers, researchers, and other actors.

• Interaction: There exist two kinds of interactions: cow-patch and cow-EMS.

• Stochasticity: By default, there is no stochasticity. When initializing, there are variables generated with random values like initial DM availability of paddocks.

• Adaptation: Cows react to decreasing pasture biomass levels in paddocks by being less attracted to them, and after a limit level, they move to another paddock with more biomass.

Initialization - The NPM model is initialized with the information provided by the farm, like the land use and the number of dairy cows (see Table 2). Also, it is essential to highlight that the GIS is included in the initialization. Furthermore, the simulations were set to start on 1^st March.

Input data - Model dynamics are driven by input data representing data from different sources, including extensive literature review and data from the farm case study. The data from the model development is described in Table 2.

3.1 Nutrient Budgets

Starting with N annual budget, Table 3 summarizes the N annual budget for tambo paddocks considering the year 2020 information.

Table 3: Tambo annual N budget 

As shown in Table 3, the N inputs are divided into mineral fertilizer, imported livestock feed, BNF, and atmospheric deposition. Considering the proportion of these N inputs, where the main N inputs are the imported livestock feed (48%), the second most important is the BNF (29%), thirdly the mineral fertilizer (22%), and the contribution of atmospheric deposition is minor (1%).

The analysis of N outputs in the dairy farm nutrient budget indicates that 66% of the N leaving the system is retained in milk, while the remaining 34% corresponds to environmental losses. Furthermore, the tambo paddocks' environmental N outputs can be considered losses divided into paddock and EMS losses.

Regarding the P budget from the tambo paddocks, Table 4 summarizes the annual results.

Table 4: Tambo annual P budget 

Since the farmer did not apply P fertilizers to the paddocks in 2020, the only P inputs corresponded to imported livestock feed; unlike N, no other forms of inputs were identified. This critical point is explained in the Discussion section. Approximately two-thirds of P outputs are in milk, and the balance one-third consists of environmental outputs.

Considering the system N outputs proportion, classified as volatilization, lixiviation, and direct N₂O nutrient losses, it represents 81.9%, 5.3%, and 2.7%, respectively, for Paddocks (being 89% of total N system outputs); and 4.5%, 6.4%, and 0.1% for EMS, respectively, (being 11% of total N outputs). For P environmental losses, paddocks have 53.9% of soluble and 33.1% of particulate P runoff, totalizing 87% of environmental losses. The remaining 13% corresponds to EMS direct losses to the environment.

The next section shows the result of the management scenarios in the nutrient budget methods. As was explained before, the management alternatives were discussed and validated with the farmers’ group.

3.2 Nutrient Budgets with Different Management Scenarios

Table 5 summarizes how the farm management scenarios assessed impact nutrient inputs, outputs, and cow’s nutrient efficiency.

Table 5: Farm management practices scenarios 

Scenario 1 - Land use: Represents the change in land use (see Table 6). This modification of land use does not change the NUE but changes the dairy system's N inputs and outputs. The proportion of N inputs varied compared with current situation, being 50% for imported livestock feed, 25% for mineral fertilizers, 24% for BNF, and 1% for atmospheric depositions. The proportion of mineral fertilizer increased and the N fixed by the Leguminosae decreased. In addition, the environmental N outputs increment 560 kg N/year (see Table 6).

Table 6: Annual N and P budgets at Scenario 1 - Land use 

Scenario 2 - Type of diet: Refers to the cow's diet proportion of pasture, supplement, and concentrate. As shown in Table 7, while the NUE maintains the current situation value, the PUE increases 3% compared to the current PUE. Furthermore, in this management scenario, the N input drops by 50 kg N/ha and the N output increases by 2 kg N/ha. Also, the proportion of N inputs changes because the imported livestock feed decreases compared with the current farm situation. At this scenario, N inputs proportion is 42% for imported livestock feed, 32% for mineral fertilizer, 25% for BNF, and 1% for atmospheric depositions.

The P input drops by 9 kg P/ha, and the P outputs maintain the same value per hectare. However, the farm P environmental outputs (kg/year) change in the value of the punctual losses of the EMS (see Table 7).

Table 7: Annual N and P budgets at Scenario 2 - Diet 

Scenario 3 - Stocking rate: Refers to a decrease of stocking rate, showing that the total N output is lower because not only the N on milk decreased, but also the environmental N losses did (see Table 8). Still, the P outputs decreased due to the lower milk production. Moreover, this scenario covers a lower effluent generation due to the stocking rate reduction (see Table 8).

Table 8: Annual N and P budgets at Scenario 3 - Stocking rate 

Since the nutrient budget method has an annual time scale and does not represent dynamical interactions between system components, the following section analyzes the nutrient flows in a dynamic way only for the current farm situation by agent-based approach.

3.3 Agent-Based Model NPM

Before examining the different model nutrient dynamic results, it is fundamental to understand how the NPM model represents the dairy farm case study. Figure 3 (screen capture at model initialization stage) shows that the NetLogo interface displays the dairy farm paddocks with their name using the GIS information provided by 2021 conversation with Case Study Farmer. The milking parlor (red house) is situated in its actual place and the cows are in the paddock at initialization. In addition, Figure 3 includes the interface generated graphs for monthly cow number, milk production and protein, and cow intake across the year simulated.

Figure 3: NPM model at initialization stage. At worlds view the paddock identification is labeled at the centroid of each one⁵³  

Considering the cows’ nutrient use efficiency calculated by the model simulation, Figure 4 shows that there has been an intra-annual variation in the NUE and PUE across the year 2020 simulation. Both curves had a gradual increment in the first trimester and then a diminution of the efficiency in the last months of the simulated year. The month variation is associated with the different monthly diets and the milk production in volume and in protein concentration (see graphs at Figure 3; milk protein and intake).

Figure 4: Cows’ nutrient use efficiency  

The cows’ nutrient use efficiency (Figure 4) presents monthly variations along the year, according to the animals’ diet profile and milk production and composition. The annual average and standard deviation were 24.6% ± 1.7% and 32.9% ± 1.7% for N and P, respectively, reaching a maximal value of 27.6% for NUE in October and 34.5% for PUE in June.

Figure 5 shows N budget evolution at the paddocks at different land uses across the year simulation.

Figure 5: Paddocks N budget (kg/ha) 

As shown in Figure 5, N accumulates in paddocks throughout the year, particularly in prairies older than one year, which were grazed more frequently than annual forage crops and prairies established in 2020. The average paddock accumulated 98.7 kg N/ha, predominantly attributed to the immobile fraction. Extreme values ranged from a minimum accumulation of 33.7 kg N/ha in annual forage crops to a maximum of 100.9 kg N/ha in second-year prairies.

The environmental losses fractions are presented in Figure 6.

Figure 6: Paddock environmental N loses (kg/ha) 

As shown in Figure 6 , volatilization and leaching are the main environmental outputs. Annual dynamics of environmental losses represent mainly fertilization events according to farmer’s scheduling linked to land use and animals’ diet profile. For example, in June, the highest pick of volatilization and leaching occurs when farmers fertilize 56% of the area (on annual grass crops and 2-, 3-, and 4-year old prairie paddocks) with an average of 41.2 kg N/ha.

On the other hand, the accumulation of N at the EMS presented a barely linear increment over the 2020-year simulation (EMS N kg = 20.499 * day -214.83; R² = 0.9966), reaching 6975 kg N/year with a proportion of 70-30% for mobile and immobile fractions, respectively.

Related to P dynamics at paddocks, Figure 7 shows soil P content and particulate and soluble environmental losses evolution across the yearly simulation.

Figure 7: Soil P content (ppm) indicating particulate and soluble P soil levels 

Figure 7 shows a relative constant quantity of that nutrient across the 2020-year simulation, averaging 40.3 ppm in soil. Even if the farmer decides not to fertilize with P, cows’ deposition compensates pasture extraction for growing. The 60-day simulation presents a fall in P soil content, reaching 39.4 ppm, but it represents less than 1.5% of difference with initial soil P concentration.

Regarding P accumulation at the EMS, it presented a linear increment (EMS P kg = 1.968 * day - 35.18; R² = 0.9925), reaching 664.4 kg P/year, and its loss to the environment by leaching totalized 166.1 kg P/year.

As was pointed out, ABM allows the spatialization of complex system dynamics. Figure 8 illustrates the soil P dynamic, including the initial (upper) and final (lower) soil P situation after one year of simulation. The model interface represents the P soil change by color, and the graphs allow the understanding of the spatial distribution in the ground.

As shown in Figure 8, spatial distribution of P is influenced by paddocks’ soil use and grazing strategy. Since farmer did not apply P fertilizer in 2020, and there is no biological fixation for P either, only forage P absorption and environmental losses by lixiviation are considered as outputs at patch budget assessment. The initial average P soil analysis of 41 ppm with spatial heterogeneity was recorded by farm soil analysis and the variation across the simulated year. While only considering cows’ excretion, the soil P average value increases across the year adding 7 ppm to initial soil content. Nevertheless, at the full paddock budget, the P soil concentration is relatively stable across the year simulation (see Figure 7).

Spatial heterogeneity of P soil in the system is explained by cows grazing sequence, leaded by the entering-leaving paddock criteria defined at 2.3 section. Prairies of 2- and 3-year sown were more frequently grazed due to its DM availability. The soil P in paddock with those frequently used land use reached 58 ppm at simulation considering only cows’ excretion.

Figure 8: Initial (upper) and final (lower) Paddock P content (ppm) across yearly simulation 

Regarding N simulated dynamics, Figure 9 illustrates paddock N budget.

Figure 9: Paddock N accumulation (kg/ha) across yearly simulations 

The paddock colors (Figure 9) show the value of N on soil -the darker magenta, the higher level of N-. As for P nutrient, the spatial heterogeneity of N in the system is explained by grazing circuit in both partial and complete budgets. The more frequently grazed paddocks (2 and 3-years old prairies, see Figure 5) presented the highest values of N budgets. For those more frequently grazed paddocks, the accumulation was 245 kg N/ha only by cows’ depositions, and 98 kg N/ha in the complete budget simulation. In average terms, considering only cows’ excretion at paddock budget, the N content reached 152.2 kg N/ha, whereas in the complete budget it reached 60.7 kg N/ha. It is worth mentioning that, different from the P analysis, the initial N on soil assumed in this simulation is 0. This means that values at Figure 9 represent cumulated N at paddocks, since all of them finalize simulations with large and positive values.