energies

Review
Energy Consumption on Dairy Farms: A Review of
Monitoring, Prediction Modelling, and Analyses
Philip Shine 1 , John Upton 2 , Paria Sefeedpari 3 and Michael D. Murphy 1, *
1 Department of Process, Energy and Transport Engineering, Cork Institute of Technology,
Cork T12 P928, Ireland; [email protected]
2 Animal and Grassland Research and Innovation Centre, Teagasc Moorepark Fermoy, Cork P61 C996, Ireland;
[email protected]
3 Wageningen Livestock Research, Wageningen University and Research, 6708 WD Wageningen,
The Netherlands; [email protected]
* Correspondence: [email protected]




Received: 19 February 2020; Accepted: 4 March 2020; Published: 10 March 2020


Abstract: The global consumption of dairy produce is forecasted to increase by 19% per person by
2050. However, milk production is an intense energy consuming process. Coupled with concerns
related to global greenhouse gas emissions from agriculture, increasing the production of milk must
be met with the sustainable use of energy resources, to ensure the future monetary and environmental
sustainability of the dairy industry. This body of work focused on summarizing and reviewing
dairy energy research from the monitoring, prediction modelling and analyses point of view. Total
primary energy consumption values in literature ranged from 2.7 MJ kg−1 Energy Corrected Milk on
organic dairy farming systems to 4.2 MJ kg−1 Energy Corrected Milk on conventional dairy farming
systems. Variances in total primary energy requirements were further assessed according to whether
confinement or pasture-based systems were employed. Overall, a 35% energy reduction was seen
across literature due to employing a pasture-based dairy system. Compared to standard regression
methods, increased prediction accuracy has been demonstrated in energy literature due to employing
various machine-learning algorithms. Dairy energy prediction models have been frequently utilized
throughout literature to conduct dairy energy analyses, for estimating the impact of changes to
infrastructural equipment and managerial practices.

Keywords: dairy; energy; review; modelling; efficiency; sustainable agriculture; machine-learning

1. Introduction
The Food and Agriculture Organization of the United Nations has forecasted the global
consumption of milk and dairy produce to increase by 19%, to 99 kg person−1 year−1 by 2050,
compared to 2005–2007 levels [1]. This increase is largely due to an increased forecasted demand
from developing countries (fueled by projected Gross Domestic Product (GDP) growth), where a
46% increase is projected across Latin America, Sub-Saharan Africa, South Asia, East Asia, and Near
East/North Africa [1]. Thus, a 22% growth in global milk production is forecasted in the ten year
period between 2018 and 2027 [2]. In 2018, the European Union (EU-28) was the largest producer
of milk worldwide producing 153 billion liters of milk, producing a 59% greater volume of milk
than the USA [3]. Since 2015, milk production volumes increased by 6% in the EU and USA to help
supply increased global demand [3]. From an environmental perspective, the increased dairy herd
and consequent increased milk production comes with its own significant challenges regarding the
consumption of on-farm energy resources and related greenhouse gases (GHG) [4,5].

Energies 2020, 13, 1288; doi:10.3390/en13051288 www.mdpi.com/journal/energies

Energies 2020, 13, 1288 2 of 25

Energy consumption on dairy farms is composed of direct (on-farm energy use) and indirect
energy use (energy required to produce farm inputs; e.g. concentrate feed) [4,6]. Thus, energy use can
be attributed to electricity consumption, liquid fuel use, fertilizer application, concentrate feed, and
other miscellaneous energy consumption. Naturally, an increase in milk production will result in an
increased energy consumption unless changes to management strategies and/or farm infrastructure are
carried out. The magnitude and efficiency (energy use per kg of milk produced or per head) of energy
consumption on dairy farms will vary according to a number of factors including (but not limited
to): type of production system (e.g. grazing, confined, conventional, organic, calving pattern, etc.),
type of milking system (e.g. conventional or automatic milking system (AMS)), milking schedules,
installed infrastructure, climate, etc. [4,5,7,8]. The importance of dairy energy consumption is twofold:
1) environmentally, the use of grid-sourced electricity, liquid fossil energy (e.g. kerosene, diesel),
and natural gas on dairy farms have a related GHG intensity with negative environmental impact.
While the production of milk, processing, and transport of dairy products contribute to 4% of global
anthropogenic emissions, the dairy industry must respond to recent increased scrutiny by minimizing
its environmental impact across all stages of the dairy supply chain [9]. 2) economically, grid-sourced
electricity may be exposed to rising energy costs, increasing financial concerns for dairy farmers, and
increasing interest in energy efficient and renewable energy technologies to help improve energy
independence and minimize energy usage [4,6].
The expansion of dairy farms globally must be met with considerations regarding minimizing
energy consumption to insure the future sustainability of dairy farms. Dairy energy research will
become increasingly important as researchers aim to identify new technologies and methods to improve
the energy efficiency of producing milk and dairy products. Recently, Baldini et al. [10] focused on
reviewing life cycle analyses strategies in the dairy industry. However, there exists a lack of secondary
research (literature review) which focuses specifically on energy use. This review focused on critically
assessing published literature related to the monitoring, prediction modelling and analyses of energy
consumption in dairy farming. In particular, this review placed specific emphasis on the portion
of energy consumption that can be controlled by the farmer, although research focusing on indirect
energy use is also incorporated. This review incorporated two primary components; 1) review of
research studies, which focusing on dairy energy assessment, prediction modelling and analyses; and
2) a discussion highlighting common trends throughout the dairy energy literature.

2. Dairy Energy
Research regarding energy on dairy farms has primarily focused on life cycle assessment (LCA),
developing prediction models, and analyzing various strategies to reduce energy consumption, related
costs, and GHG emissions. Interest regarding the energy intensity of dairy farming is due in part to
conflicting targets regarding increasing milk production and reducing GHG emissions (i.e. the EU
aims for a 30% reduction in GHG emissions by 2030, compared to 2005 levels) [11,12]. Concurrently,
due to the volatile nature of milk pricing, dairy farmers should adequately prepare for periods of
reduced revenue through minimizing the costs of all aspects of production. Monitoring and effectively
reducing the energy intensity of the milk production process may offer significant environmental
benefits, increase profits for dairy farmers and improve the long-term sustainability of the global
dairy industry. In this review, dairy energy related research studies will be differentiated according to
whether they report results from conventional, organic, confined or pasture-based systems [5]. This
literature review covers international studies related to dairy energy measurement and assessment,
prediction modelling, and the conservation/analysis of energy on dairy farms.

2.1. Dairy Energy Assessment

Total energy consumption on dairy farms can be categorized as direct and indirect energy
use [4,6,13]. Direct energy is composed of energy consumption carried out directly on the farm. For
example, electrical energy consumption and the consumption of liquid/gaseous fuels. Indirect energy is
Energies 2020, 13, 1288 3 of 25

composed of those energy uses whereby the direct energy use occurs outside the farm boundaries [13].
More specifically, indirect energy is the energy embodied in the products used on the farm. Examples
of indirect energy sources include the energy required to produce and transport feed, fertilizers, and
agricultural chemicals, which are produced outside the farm, but consumed on the farm.
In 2017, Baldini et al. [10] conducted a critical review of the recent evolution of LCA when
applied to milk production. They noted inconsistencies between functional units defined throughout
international literature and suggested the definition of a common functional unit. This would allow
for direct comparison between the results of international studies, with different assumptions. This
inconsistency was found throughout the dairy energy literature whereby functional units included
energy corrected milk (ECM), fat and protein corrected milk (FPCM), milk solids (MS), kg of milk,
fat-corrected milk (FCM), and liters of milk. The authors noted that the most common functional unit
throughout the literature was ECM, thus when possible, ECM was utilized for comparative purposes.
Key performance indicators, which utilized the remaining functional units were then converted to
their ECM equivalence when milk fat and protein percentage values were available. If milk fat and
protein values were unavailable in a research paper, the corresponding authors were contacted directly
via email.

2.1.1. Total Energy

A summary of 23 studies that were utilized for calculating average energy usage values (MJ kg−1
ECM), as displayed in Table 1. In total, 36 studies were gathered across 17 countries, however, 13
of these studies presented energy consumption values relative to either MJ kg−1 , MJ kg−1 FPCM,
or MJ kg–1 FCM, without reporting corresponding fat and protein percentage values to convert
to MJ kg–1 ECM using Equation (A1), Equation (A2), and Equation (A3). Further details related to the
36 research findings are available in Appendix B.
Total primary energy use was 54% greater on conventional farms compared to organic farms,
as shown in Table 1. This was primarily due to increased indirect energy use on conventional farms,
as conventional farms used 75% greater indirect energy, while organic farms used 13% less direct
energy, on average.

Table 1. Mean energy consumption and characteristics of studies found in literature.

Title Conventional Organic

Characteristic Conv-g * Conv-c Org-g Org-c
No. of studies 5 11 5 2
No. of countries 4 11 4 2
Mean no. farms per study 37 43 5 9
2.8 4.7 2.9 2.1
Total Energy (MJ kg−1 ECM)
4.1 2.7
* Conv-g = conventional grazing farm; Conv-c = conventional confinement farm; Org-g = organic grazing farm;
Org-c = organic confinement farm.

Across the 16 studies that reported energy use (MJ kg−1 ECM) on conventional dairy farms, an
average of 4.1 MJ kg–1 ECM was calculated. Conventional confinement farming systems (Conv-c)
had an average of 4.7 MJ kg−1 ECM across the literature, while conventional grazing farming systems
(Conv-g) had an average of 2.8 MJ kg−1 ECM. Thus, Conv-c dairy farms required 68% greater energy
resources compared to Conv-g dairy farms. Differences in energy requirements between Conv-c and
Conv-g dairy farms will be discussed further in Sections 2.1.2 and 2.1.3.
Across the seven studies that reported energy use on organic dairy farms, an average of 2.7 MJ kg−1
ECM was calculated. Organic confinement farming systems (Org-c) studies had an average of
2.1 MJ kg−1 ECM across the literature, while organic grazing or pasture-based systems (Org-g) studies
reported an average of 2.9 MJ kg−1 ECM. Interestingly, the mean energy consumption of Org-g farms
Energies 2020, 13, 1288 4 of 25

was only 4% greater than that of Conv-g farms. In contrary to conventional dairy farms, Org-c dairy
farms required 28% less energy resources compared to Org-g dairy farms. Differences in energy
requirements between Org-c and Org-g dairy farms will be discussed further in Sections 2.1.2 and 2.1.3.

2.1.2. Indirect Energy

Indirect energy consumption is sectioned into two categories: ancillary and embodied, as shown
in Tables 2 and 3. Ancillary energy encompasses energy utilized for the production of agricultural
products utilized on farm, such as fertilizer and feed. Concurrently, embodied energy refers to the
energy consumed in the mining of raw materials for, and production of capital dairy farm inputs
averaged across their expected working life, assuming straight-line depreciation [14]. Embodied
energy sources include buildings and facilities, and machinery and equipment. Due to the lack of
available data, embodied energy calculations are rarely included in LCA [10].

Ancillary Energy
On average, studies that reported on ancillary energy requirements on conventional dairy farms
(Conv-g or Conv-c) reported an average of 3.0 MJ kg−1 ECM, as may be calculated using data shown
in Table 2.

Table 2. Ancillary energy values found in literature.

Study System Country MJ kg−1 ECM

Fertilizer Meul et al. [15] Conv-c BEL 0.82
O0 Brien et al. [5] Conv-c IRL 0.49
Pagani et al. [16] Conv-c ITA/USA 0.21
Todde et al. [17] Conv-c ITA 0.84
O0 Brien et al. [5] Conv-g IRL 1.09
Pagani et al. [16] Conv-g ITA / USA 0.30
Upton et al. [4] Conv-g IRL 1.34
Wells [14] Conv-g NZ 0.66
Pagani et al. [16] Org-c ITA 0.00
Pagani et al. [16] Org-g USA 0.00
Fertilizer mean Conv-c n/a 0.59
Fertilizer mean Conv-g n/a 0.85
Feed Aguirre-Villegas et al. [18,19] Conv-c USA 1.54
Meul et al. [15] Conv-c BEL 0.24
Pagani et al. [16] Conv-c ITA / USA 2.28
Sefeedpari et al. [20,21] Conv-c IRN 6.30
Todde et al. [17] Conv-c ITA 3.90
Pagani et al. [16] Conv-g ITA / USA 1.44
Upton et al. [4] Conv-g IRL 0.49
Pagani et al. [16] Org-c ITA 0.85
Pagani et al. [16] Org-g USA 1.22
Feed mean Conv-c n/a 2.85
Feed mean Conv-g n/a 0.96
Key performance indicators were converted to ECM using Equation (A1), Equation (A2) and Equation (A3).

On these conventional farms, the use of fertilizer was responsible for 0.7 MJ kg−1 ECM, while the
production of feed required 2.3 MJ kg−1 ECM. On the only study involving organic dairy farms (Org-c
and Org-g) [16], fertilizer required 0.0 MJ kg−1 ECM. The production of feed utilized on conventional
farms required 131% greater energy to produce compared to the value of 1.0 MJ kg−1 ECM calculated
by Pagani et al. [16] on organic farms (Org-c and Org-g).
Comparing ancillary energy requirements of confinement and grazing system dairy farms,
fertilizer used on Conv-g farms required 44% greater energy to produce compared to Conv-c farms,
on average. Regarding the energy required for feed, Conv-c farms required 196% greater energy
Energies 2020, 13, 1288 5 of 25

compared to Conv-g farms. This difference in feed energy requirements is due to large portions of
cattle feed being met by pasture on Conv-g farms as opposed to concentrated feed. On organic farms,
the energy required for fertilizer production required 0.0 MJ kg−1 ECM for both the Org-c and Org-g
systems. Regarding feed, Pagani et al. [16] found that feed consumed by dairy cows on Org-g farms
required 43% greater energy for production compared to Org-c farms. Overall, Conv-c farms reported
90% greater ancillary energy compared to Conv-g farms.

Embodied Energy
Embodied energy requirements on conventional dairy farms (Conv-g or Conv-c) reported in
literature averaged 0.8 MJ kg–1 ECM, as may be calculated using data shown in Table 3.

Table 3. Embodied energy values found in literature.

Study System Country MJ kg−1 ECM

Buildings and Facilities Kraatz [22] Conv DEU 0.10
Todde et al. [17] Conv ITA 0.29
Wells [14] Conv NZ 0.25
Buildings and facilities mean Conv n/a 0.21
Machinery and Equipment Kraatz [22] Conv-c DEU 0.57
Meul et al. [15] Conv-c BEL 1.27
Pagani et al. [16] Conv-c ITA / USA 0.24
Sefeedpari et al. [20,21] Conv-c IRN 0.08
Todde et al. [17] Conv-c ITA 1.08
Pagani et al. [16] Conv-g ITA / USA 0.27
Pagani et al. [16] Org-c ITA 0.39
Pagani et al. [16] Org-g USA 0.62
Machinery and equipment mean Conv-c n/a 0.65
Key performance indicators were converted to ECM using Equation. (A1), Equation (A2) and Equation (A3).

On these conventional farms, embodied energy contained in the production of building and
facilities equaled 0.2 MJ kg−1 ECM, while no studies for organic farms were reported in literature.
Embodied energy contained within machinery and equipment equaled 0.6 MJ kg−1 ECM on
conventional farms, greater than the value of 0.50 MJ kg−1 ECM calculated by Pagani et al. [16]
on organic farms (Org-c and Org-g).
Regarding the energy embodied within dairy farm buildings and facilities, Conv-g farms contained
27% greater embodied energy compared to the Conv-c farm. Regarding the energy embodied within
machinery and equipment, the single Org-g farm had 60% greater energy embedded in machinery and
equipment compared to the single Org-c farm. On Conv-c farms, 0.6 MJ kg−1 ECM were embodied
within machinery and equipment, while 0.3 MJ kg−1 ECM were embodied on Conv-g farms. Overall,
Conv-c farms had 62% greater embodied energy compared to Conv-g farms.

2.1.3. Direct Energy

A summary of total on-farm electricity consumption usage metrics is displayed in Table 4. Electrical
energy is directly consumed on farm through milk cooling (refrigeration), milk harvesting (vacuum
pumps), water heating, water pumping, lighting, as well as other miscellaneous usage throughout the
farm. The other major source of direct energy use includes the consumption of liquid/gaseous fuels
such as diesel, kerosene, natural gas, liquefied petroleum gas (LPG), and lubricants. Liquid fuels may
be utilized on-farm for water heating or for powering mechanical machinery (e.g. tractors).

Electrical energy
On average, 48.9 watt-hours of electrical energy are consumed per kg of milk (Wh kg−1 ) (n = 4)
on conventional dairy farms, as shown in Table 4. On conventional farms (Conv-c and Conv-g), milk
Energies 2020, 13, 1288 6 of 25

cooling was the largest consumer of electrical energy on dairy farms, consuming 15.3 Wh kg−1 on
average. Milk harvesting was found to be the second largest consumer of electricity (14.0 Wh kg−1 )
followed by water heating (9.5 Wh kg−1 ), and water pumping (3.3 Wh kg−1 ). However, Edens et al. [24]
(n/a), Hörndahl [25] (n/a), and Todde et al. [29] (23%) found milk harvesting to be the largest consumer
of electricity, while additionally, although no specific electricity consumption values were reported,
Hartman and Sims [30] found water heating to be the largest consumer of electrical energy, consuming
31%, followed by milk cooling (21%), milk harvesting (18%), water pumping (18%), and miscellaneous
usage throughout the farm (12%).

Table 4. Electrical energy consumption breakdown statistics of studies found in literature (Wh kg−1 ).

Milk Milk Water Water

Study System Country Total
Cooling Harvesting Heating Pumping
Calcante et al. [23] AMS-c ITL n/a n/a 11.13 n/a n/a
Edens et al. [24] Conv-c USA n/a 21.17 22.82 13.50 n/a
Hörndahl [25] Conv-c SWE n/a 16.70 23.01 4.85 n/a
Hörndahl [25] AMS-c SWE n/a 13.20 20.49 5.05 n/a
Murgia et al. [26] Conv-c ITA 42.84 9.85 8.14 3.43 4.71
Rajaniemi et al. [27] Conv-c FIN n/a 21.70 12.00 16.30 1.51
Rajaniemi et al. [27] AMS-c FIN n/a 21.90 29.30 2.20 n/a
Shine et al. [8] Conv-g IRL 38.68 11.24 6.91 7.66 1.51
Shortall et al. [28] AMS-g IRL 60.78 10.97 20.10 4.27 2.62
Todde et al. [29] Conv-c ITA 73.00 13.87 16.79 10.95 6.57
Upton et al. [4] Conv-g IRL 41.11 12.64 8.19 9.54 2.07
Mean AMS-c n/a n/a 17.45 14.54 10.67 1.51
Mean Conv n/a 48.91 15.32 13.97 9.45 3.28
Mean Conv-c n/a 57.92 16.68 16.54 9.80 4.27
Mean Conv-g n/a 39.89 11.94 7.55 8.60 1.79

Concurrently, the one study that calculated total electricity consumption of AMS farms found
60.8 Wh kg−1 were consumed (24% greater than conventional farms) [28]. Four AMS farms reported on
sub-metered energy consumption values, whereby on these farms (AMS-c and AMS-g), milk harvesting
was the largest consumer of electrical energy, consuming 20.2 Wh kg−1 . Milk cooling was found to be
the second largest consumer of electricity (16.4 Wh kg−1 ) followed by water heating (3.2 Wh kg−1 ),
and water pumping (2.6 Wh kg−1 ). All four studies [23,25,27,28] found milk harvesting to be the
greatest consumer of electricity on AMS farms. Although none of the farms included in Table 4 are
organic farms, electrical energy on organic farms is of greater importance compared to conventional
farms because it represents 26% of total primary energy usage, compared to 17% on conventional
farms (calculated using energy values in Table A1). In Ireland, Upton et al. [4] found that electricity
consumption was the largest source of total direct energy use (60%), representing 12% of total energy
consumption on Irish dairy farms. They found that electricity use was the third largest contributor to
total energy use behind indirect energy related to fertilizer application (57%) and concentrates feed
(21%). On the 22 Irish study farms, Upton et al. [4] found 41.1 Wh kgm -1 were consumed on average
in 2011. Shine et al. [8] carried out a more focused assessment of dairy electricity consumption on
Irish dairy farms by utilizing 43 dairy farms monitored throughout the 2015 calendar year. Shine
et al. [8] reported an electricity usage value of 38.7 Wh kgm -1 , thus 6% less than Upton et al. [4]. Using
a day/night pricing tariff (day tariff of 0.18 € kWh−1 ; night tariff of 0.08 € kWh−1 from 00:00 to 09:00 h),
Upton et al. [4] calculated a mean cost of 0.51 € cent Lm −1 in 2011, while Shine et al. [8] calculated a
mean cost of 0.55 € cent Lm −1 in 2015. With production costs averaging 21.8 cent Lm −1 in 2016 [31], and
using values calculated by Shine et al. [8], electricity costs are responsible for 2.5% of overall production
costs, on average. Hartman and Sims [30] calculated electricity costs of between $0.06 per cow-1 day-1
and $0.12 per cow−1 day−1 , equaling between 1.5% and 3.0% of total costs. However, they employed
an assumed electricity tariff of $0.17 kWh-1 , as opposed to a day/night or day/night/peak pricing tariff.
Energies 2020, 13, 1288 7 of 25

To the authors’ knowledge, no other studies directly assessed with the monetary aspect of electrical
energy consumption on dairy farms, possibly due to varying electrical energy costs between countries.
However, Todde et al. [32] developed the dairy energy prediction (DEP) model, which allowed for the
estimation of diesel and electricity energy usage, CO2 emissions, and monetary costs (static electricity
price). This model is discussed further in Section 2.2.2.
In Italy, Murgia et al. [26] conducted a partial LCA to quantify the energy intensity associated
with electricity and diesel use on 20 dairy farms in 2011 (mean herd size = 320 cows; range = 158
− 500 cows). They found electricity use contributed 30% to direct energy consumption (401 kWh
per lactating cow, 42.84 Wh kgm −1 ), thus less than the average of 47% found between international
studies. Concurrently, Todde et al. [29] analyzed the direct energy use of 285 Italian dairy farms
through a detailed survey of electricity, diesel and liquefied petroleum gas consumption and related
on-farm activities and processes. They calculated an electricity consumption value of 73 Wh kgm −1 .
Similar to Murgia et al. [26] electricity accounted for 27% of direct energy use. Additionally, in Italy,
utilizing 60 dairy farms in the Emilia Romagna region in 2009, Rossi et al. [33] calculated values of
510 kWh cow−1 year−1 and 64 Wh kgm −1 , again, representing greater electricity consumption values
than Murgia et al. [26].
Meul et al. [15] utilized a representative set of Flemish farms to analyze the energy use efficiency
of dairy, arable and pig farms between the 1989–1990 and 2000–2001 periods. They found that energy
usage per hectare reduced significantly during the considered timeframe. They also showed greater
energy efficiency associated with more intense dairy farming practices, while electricity use contributed
to 9.5% of overall dairy farm energy use across the 2000 to 2001 period.
Similar to Upton et al. [4] and Shine et al. [8], Kraatz [22] found that milking equipment (vacuum
pumps, milk bulk tanks, and water heaters) were the major electricity consumers on German dairy
farms. In Finland, Rajaniemi et al. [27] calculated electricity consumption associated with milk
production (milk-cooling, milk harvesting and water-heating) to vary between 37 and 67 Wh kgm −1
when analyzing three dairy farms (mean herd size = 82 cows), one farm of which employed an AMS.
Rajaniemi et al. [27] reported milk harvesting and milk-cooling as the largest electricity consumers
on the AMS farm, (29% and 22%, respectively), while on the two Conv-c farms, milk-cooling and
water-heating were the two largest electricity consuming processes (22% and 16%, respectively).
In Denmark, Brøgger Rasmussen and Pedersen [34] analyzed the electricity consumption within the
milking parlor on 17 dairy farms, which each employed an AMS between June 2003 and February 2004,
making comparisons with a conventional herringbone parlor and 26-stall rotary milking parlor. Across
four AMS brands, Brøgger Rasmussen and Pedersen [34] found AMS to consume between 15.2 and
86.0 Wh kgm −1 (values converted from liters of milk using milk volumetric mass density value equal to
1.03 kg per liter), while the conventional milking parlors consumed between 19.0 and 21.5 Wh kgm −1 .
Concurrently, Brøgger Rasmussen and Pedersen [34] found (in diminishing order (no specific values
reported)) the vacuum pump, milk-cooling, electric water heater and automatic washing system to be
the greatest electricity consuming processes on AMS dairy farms.
Hörndahl [25] measured and analyzed the electricity consumption of two Swedish dairy farms
(herd sizes: 150 cows and 202 cows), finding that milk harvesting (24.4 Wh kgm −1 ), milk-cooling
(17.7 Wh kgm −1 ), and water-heating (5.2 Wh kgm −1 ) were the three largest electricity consuming
processes. Additionally, Edens et al. [24] conducted a thorough energy analysis of the four major
energy components on USA based dairy farms (vacuum pumps, refrigeration compressors, water
heaters (with water preheater), and an air compressor). Utilizing 14 years of data for a single farm
(herd size = 160 cows), they calculated the kWh requirement per 100 pounds of milk produced (values
were converted to Wh kgm −1 using milk volumetric mass density of 1.03 kg Lm −1 ). They found milk
harvesting (24.2 Wh kgm −1 ) to be the largest consumer of electrical energy, followed by milk cooling
(22.5 Wh kgm −1 ), and water heating (14.3 Wh kgm −1 ). Concurrently, Hartman and Sims [30] measured
the electrical energy consumption on three New Zealand dairy farms. They found that on average, total
energy use amounted to 47 MJ kg−1 MS, whereby 30% was associated to electricity (163 kWh cow−1
Energies 2020, 13, 1288 8 of 25

year−1 ). As they calculated and presented total energy use in MJ kg-1 MS and electricity efficiency
as kWh cow−1 year−1 , without reporting milk production, protein, and fat content values, it is difficult
to make comparisons with cognate studies. However, they did breakdown electricity consumption,
whereby water heating (31%) was the largest consuming process, followed by milk cooling (21%),
water pumping (18%), milk harvesting (18%), and miscellaneous usage (12%).

Liquid Fuel Energy

Liquid fuel energy usage values are summarized in Table 5. All studies presented in Table 5
focused on conventional dairy systems, except for one study that reported diesel use for farms that
had AMS installed. On average, 0.73 MJ kg−1 ECM of energy is required on conventional dairy farms
(Conv-c and Conv-g) in the form of diesel fuel. Conv-c farms consume 1.0 MJ kg−1 ECM, while
Conv-g farms consume 0.2 MJ kg−1 ECM. However, this comparison contains only a small number
of farms, thus making it difficult to make useful conclusions. This is particularly important as diesel
consumption would be expected to be greater on pasture-based farms due to the increased tractor
requirement for larger pasture production, as found by Cederberg and Mattson [35] when comparing
conventional and organic dairy farms. Cederberg and Mattson [35] also found that the use of crude
oil was greater on conventional dairy farms compared to organic farms. However, Cederberg and
Mattson [35] did not report on exact liquid fuel consumption values. For conventional dairy farms
(both Conv-c and Conv-g), diesel use equaled 0.7 MJ kg−1 ECM, lubricants equaled 0.03 MJ kg−1 ECM,
LPG equaled 0.02 MJ kg−1 ECM, while kerosene equaled 0.02 MJ kg−1 ECM. Thus, in total, liquid fuels
were responsible for 0.92 MJ kg−1 ECM.

Table 5. Liquid energy consumption breakdown statistics of studies found in literature.

Fuel Study System Country MJ kg−1 ECM

Diesel Hospido et al. [36] AMS-c ESP 0.16
Meul et al. [15] Conv-c BEL 0.79
Sefeedpari et al. [20,21] Conv-c IRN 1.09
Thomassen et al. [37] Conv-c NLD 0.29
Todde et al. [17] Conv-c ITA 1.89
Upton et al. [4] Conv-g IRL 0.19
Diesel mean Conv-c n/a 1.01
Kerosene Sefeedpari et al. [20,21] Conv-c IRN 0.04
Upton et al. [4] Conv-g IRL 2.28x10−3
LPG Todde et al. [17] Conv-c ITA 0.02
Lubricants Meul et al. [15] Conv-c BEL 0.05
Upton et al. [4] Conv-g IRL 2.47x10−3
LPG = liquefied petroleum gas. Key performance indicators were converted to ECM using Equation (A1), Equation
(A2), and Equation (A3).

In addition to Table 5, Murgia et al. [26] found diesel fuel contributed 70% to the total direct energy
consumption (92 kg per cow, 0.021 kg diesel kg−1 FPCM). Concurrently, Todde et al. [29] calculated a
liquid fuel consumption of 40 kg per tonne of milk produced. Similar to Murgia et al. [26], diesel fuel
accounted for 72% of total direct energy requirement. Wells [14] calculated liquid fuels to be responsible
for 20% of total primary energy consumption on the average New Zealand dairy farm, equating to
0.40 MJ kg−1 ECM. Concurrently, O’Brien et al. [5] calculated the liquid energy consumption of both
Conv-c and Conv-g dairy systems, finding that liquid fuels were responsible for 4.6% and 12.7% of total
primary energy, respectively. This equated to 0.18 MJ kg−1 ECM and 0.30 MJ kg−1 ECM for Conv-c
and Conv-g systems, respectively.
Energies 2020, 13, 1288 9 of 25

2.1.4. Dairy Energy Assessment Summary

A breakdown of total primary energy for conventional and organic farming systems is displayed
in Figure 1.

Figure 1. Pie charts showing the breakdown of total primary energy consumption on: a) conventional
dairy farm systems and b) organic systems (b).

Balance equations were used for the development of pie chart in Figure 1, while liquid fuel
usage values, energy embodied in buildings, and facilities were assumed to be equal on conventional
and organic farms, due to unavailability of data for organic farms. These pie charts contain energy
consumption associated with electricity, diesel, kerosene, LPG, lubricants, fertilizer, feed, buildings and
facilities, and machinery and equipment. In summary, across the studies considered in this review, the
production of feed requires the greatest energy consumption on both conventional and organic dairy
systems, requiring 43% and 34%, respectively. On conventional dairy farms, electricity and diesel use
were both the second greatest consumers of energy, each responsible for 14% of total energy, followed
by the energy required for the production of fertilizer (13%), and energy embodied in machinery and
equipment (11%), and buildings and facilities (4%). On organic farms, electricity and diesel use were
each responsible for 24% of total primary energy, followed by energy embodied in machinery (12%)
and energy embodied in buildings and facilities (5%). On conventional farms, direct energy use was
responsible for 28% of total primary energy whereby, electrical energy use was responsible for 48% of
direct energy use. On organic farms, direct energy use was responsible for 50% of total primary energy,
whereby electrical energy use was found responsible for 49% of direct energy use.
It is clear from the literature that numerous international research studies have focused on
quantifying energy consumption both directly and indirectly related to the production of milk.
However, albeit useful for comparing cognate studies, averaged electricity consumption per liter of
milk produced (or FPCM, kgm , etc.) is a generalized consumption metric. This may be problematic
when single farm consumption predictions may be required, whereby consumption per liter of milk
can vary considerably depending upon country, milk-cooling equipment, hot water-heating strategies,
vacuum pump types, managerial strategies, etc. Concurrently, LCAs often require the physical
metering of dairy farm inputs such as liquid fuel usage and electricity consumption. Often, the
installation of metering equipment may require high capital and maintenance costs and can be quite
time consuming. Thus, literature has also covered the development of prediction models for dairy
farm-related energy consumption.

2.2. Dairy Energy Prediction Modelling

Many methods have been utilized for the prediction of energy consumption on dairy farms.
Methods that are covered in this review include: mechanistic modelling, energy balance modelling,
multiple linear regression (MLR), polynomial regression, and machine-learning. Similar to the varying
functional units utilized for assessing energy consumption (as detailed in Section 2.1), there also exist
Energies 2020, 13, 1288 10 of 25

a number of different model assessment criteria and methods used in dairy energy literature. The
most common model assessment criteria were the coefficient of determination (R2 ), root mean square
error (RMSE), and relative prediction error (RPE). Other common accuracy metrics included (but not
presented) are mean absolute percentage error (MAPE) and mean percentage error (MPE).

2.2.1. Mechanistic Modelling

A mechanistic model for electricity consumption on dairy farms (MECD) was developed in 2014,
capable of predicting electricity consumption, cost and related GHG emissions [38]. The MECD was
developed through Microsoft Excel as a mathematical representation of the seven major infrastructural
systems on a dairy farm (milk-cooling, water-heating, milk harvesting, lighting, water pumping, wash
pumping, and winter housing). Due to the MECD calculating electricity consumption from first
principals, a large number of input variables are required, including but not limited to: milk to water
ratio, water temperatures, milking times, milk collection interval, hot water temperature set point, and
water pump motor sizes. Mathematical equations representing the seven infrastructural systems were
constructed using a 12-month × 24-hour matrix structure and the accuracy of the MECD was validated
on three farms sizes (small (45 cows)–medium (88 cows)–large (195 cows)). The MECD was found
to predict annual electricity consumption to less than 10% (RPE). This level of prediction accuracy
represented excellent prediction capability according to Fuentes-Pila et al. [39]. The model accuracy
scale proposed by Fuentes-Pila et al. is commonly utilized throughout the dairy domain [40–45],
whereby additionally, RPE values between 10% and 20% represent acceptable levels of prediction error,
while values greater than 20% suggest poor prediction capability.
Breen et al. [46] adapted the MECD model [38] to include wind turbine and solar photovoltaic
models. They assessed the applicability of installing solar photovoltaic modules and wind turbines on
a dairy farm across four technology scenarios (i.e. four different dairy farm infrastructural situations)
using three electricity tariffs, and three feed-in-tariffs. They found that for wind turbine and photovoltaic
systems, a farm with an ice bank (IB) milk-cooling system, nighttime electric water-heating schedule
using a day/night electricity tariff was the optimum scenario for all three feed-in-tariffs considered.
Additionally, monetary savings attributed to wind turbines were highly sensitive to feed-in-tariffs,
while savings associated with photovoltaic cells varied depending upon the technology scenario (i.e.
electrically heated water start time). It was found that the electricity output of the photovoltaic system
matched the load profile of the direct expansion (DX) scenario well, which resulted in a grid connected
photovoltaic system having similar monetary savings to a standalone system.
Murphy et al. [47] developed a decision support system for energy use on dairy farms (DSSED) to
open-source the work carried out by Upton et al. [38] and Breen et al. [46] in Ireland. DSSED allows dairy
farmers, policymakers, academics, and other stakeholders to assess the potential monetary payback
(ROI), and environmental impact (CO2 emissions) of various energy efficient and renewable energy
technologies. Users may assess the financial or environmental applicability of a plate cooler, variable
speed drive (VSD), heat recovery system, solar thermal water heating system, solar photovoltaic system,
and/or a wind turbine. DSSED allows for various user inputs to be adjusted, so that calculations are
carried out unique to each farm scenario. User inputs related to dairy farm infrastructure, managerial
procedures, energy technology parameters (e.g. solar PV size (kWp)), investment cost, and the level
of available grant aid, inflation rate, feed-in-tariff, and electricity tariff rates may be adjusted. Shine
et al. [48] carried out a hypothetical case study utilizing DSSED, assessing the applicability of a solar PV
system on a hypothetical farm with 210 dairy cows. They found that under the hypothetical conditions,
the solar PV system would save over 96 tons of CO2 over the course of 20 years, with a calculated
payback period of 6.1 years.

2.2.2. Regression Modelling

Regression models has been widely applied in literature in order to find a pattern for energy use
in dairy farming systems. Sefeedpari et al. [21] used linear regression to predict output energy (energy
Energies 2020, 13, 1288 11 of 25

of milk produced and outputted cow manure) via inputs related to the fossil fuel and electrical energy
consumption of 50 Iranian dairy farms between 2011 and 2012. However, linear regression was found
to be an inadequate method of predicting dairy farm output energy, with a resulting R2 value equaling
0.11.
Edens et al. [24] utilized empirical modelling, developing four multiple linear regression (MLR)
models to predict the annual electricity consumption of each major electricity component in the USA.
These included vacuum pumps, water heaters, refrigeration compressors, and air compressors, as well
as a fifth MLR model to predict their combined consumption. Using data from a single farm (herd
size = 160 cows) throughout a 14 year period, Edens et al. [24] developed the MLR models using data
related to the volume of milk produced, the number of lactating cows, the percentage of butterfat in
milk, pounds of fat-corrected milk, monthly low temperature, monthly high temperature, average
monthly high temperature, and average monthly low temperature. They utilized five variable selection
methods to help maximize the prediction capability of each MLR model to predict monthly unseen
electricity consumption for each individual component. The five variable selection methods were: 1)
forward variable selection, 2) backward variable selection, 3) stepwise regression, 4) R-squared, and
5) Mallows Cp. The final subset of variables selected for each component was then utilized for MLR
model development through SAS software [49]. For the vacuum pumps, milk production alone was
capable of explaining 44% of the variation (R2 = 0.44), as shown in Table 6.

Table 6. Summary of models developed for predicting energy consumption on dairy farms.

Model Prediction Validation

Study Country Res R2 RMSE RPE (%)
Type Variable Method
Edens et al. Milk harvesting
USA MLR M n/a 0.44 n/a n/a
[24] (kWh)
Edens et al.
USA MLR M Milk cooling (kWh) n/a 0.74 n/a n/a
[24]
Edens et al. Water heating
USA MLR M n/a 0.34 n/a n/a
[24] (kWh)
Edens et al. Air compressors
USA MLR M n/a 0.18 n/a n/a
[24] (kWh)
Edens et al.
USA MLR M Combined (kWh) n/a 0.62 n/a n/a
[24]
Sefeedpari Output energy of Test set
IRN ANN A 0.88 0.015 n/a
et al. [50] milk (MJ Cow-1 ) (20%)
Upton et al. Total electricity use 3 typical
IRE Mech M n/a 125.0 7.5
[38] (kWh) farms
Sefeedpari Output energy of Test set
IRN Linear A 0.11 0.2 n/a
et al. [20] milk (MJ Cow-1 ) (20%)
Sefeedpari Output energy of Test set
IRN ANFIS A 0.79 0.1 n/a
et al. [20] milk (MJ Cow-1 ) (20%)
Mhundwa Morning milk Test set
SA MLR D 0.92 n/a n/a
et al. [51] cooling (kWh) (30%)
Mhundwa Evening milk Test set
SA MLR D 0.90 n/a n/a
et al. [51] cooling (kWh) (30%)
Todde et al. Total electricity use
ITL PR A LOOCV n/a n/a 11.4
[32] (kWh)
Todde et al.
ITL PR A Total diesel use (kg) LOOCV n/a n/a 15.0
[32]
Shine et al. Total electricity use
IRE MLR M 10-fold CV 0.72 543.0 16.1
[41] (kWh)
Shine et al. Total electricity use
IRE SVM M 10-fold CV 0.94 241.0 12.0
[40] (kWh)
USA = United States of America; IRN = Iran; IRE = Ireland; Sa = South Africa; ITL = Italy. MLR = multiple
linear regression; ANN = artificial neural network; Mech = mechanistic; Linear = linear regression model; ANFIS
= adaptive neuro-fuzzy inference system; PR = polynomial regression; SVM = support vector machine. Res =
prediction resolution; M = monthly; A = annual; D = daily. LOOCV; leave-one-out cross-validation; 10-fold CV =
10-fold cross-validation. n/a = information not available
Energies 2020, 13, 1288 12 of 25

Regarding milk-cooling, milk production and monthly high temperature together were capable of
explaining 74% of the variation in electricity consumption of the refrigeration compressors. Regarding
the electricity consumption of the water heater, a large proportion of the variability could not be
explained, with all variables only capable of explaining 34% of the total variability. Concurrently,
milk production and the number of lactating cows explained only 18% of the variability of energy
consumed by the air compressors. They noted leaks in the air system could have been the primary
cause of the poor predictive capability of the data. Finally, regarding the total energy use of the major
components, the number of cows milked, the pounds of milk produced, the monthly average high
temperature the monthly high temperature, and together explained 62% of the total variability of
electricity consumption. They concluded that the quantity of milk had a greater impact on dairy farm
electricity consumption over quality of milk produced with the number of cows milked and ambient
temperature having lesser statistical impacts. The specific model coefficients for each major electrical
energy consuming process were not presented by Edens et al. [24].
Shine et al. [41] also employed MLR modelling to predict dairy electricity consumption. As opposed
to predicting annual electricity consumption, they predicted monthly electricity consumption data
remotely monitored on 56 dairy farms throughout the January 2014–May 2016 period in conjunction
with farm details related to milk production, cow numbers (herd size and number of lactating cows),
farm infrastructural equipment, managerial processes, and environmental conditions. This resulted in
12 individual regression model equations being developed for each month, allowing for consumption
trends throughout the year to be modelled. In total, 15 farm variables were assessed for their ability to
predict dairy farm electricity consumption. The subset of farm variables that maximized the prediction
accuracy of unseen electricity consumption was selected through applying a univariate variable
selection technique and Variance Inflation Factor (VIF) in conjunction with all subsets’ regression and
10-fold cross validation. The final subset of variables was found to be herd size, the volume of milk
produced, whether an IB or DX milk bulk tank was used, whether ground water was utilized for
pre-cooling milk, the number or air compressors, the frequency of hot washing (HzHW), and the total
water heater volume. This MLR model was found to predict monthly dairy electricity consumption to
within 26% (RPE). Through a standardized regression analysis, milk production and herd size had the
largest impact on electricity consumption. Mhundwa et al. [51] developed MLR models to predict the
milk-cooling related electricity consumption of a DX bulk tank without pre-cooling, utilizing a single
South African dairy farm (mean of 500 lactating cows) monitored over three months. Two MLR models
were developed; the first predicted the electricity requirements of the morning milking, while the
second predicted the electricity requirements of the evening milking. The MLR models were developed
using input data related to the volume of milk produced (liters), the milk temperature (◦ C), bulk tank
room temperature (◦ C), ambient temperature (◦ C), and relative humidity. Model development was
carried out using a 70% of the overall dataset, while the remaining 30% was utilized as a test set to
calculated model accuracy. They report R2 values of 0.92 and 0.90 for the MLR models to predict the
milk-cooling related electricity consumption of the morning and evening milking, respectively. The
developed MLR models to predict the milk-cooling related electricity consumption of the morning and
evening milking are shown in Equations (1) and (2), respectively.

EAM = 23.7 + 0.33 × Tamb − 6.4 × 10−2 × RH + 9.0 × 10−3 × Vm + 0.22 × Tm − 0.64 × Tr (1)

EPM = 12.0 + 1.53 × Tamb − 0.30 × RH + 7.0 × 10−3 × Vm + 2.2 × Tm − 0.62 × Tr (2)

where EAM and EPM are the milk-cooling electricity consumption of the morning and evening milking,
respectively, Tamb is the ambient temperature, RH is the relative humidity, Vm is the volume of milk
harvested, Tm is the milk temperature, and Tr is the temperature of the bulk tank room.
Todde et al. [32] developed the dairy energy prediction (DEP) model (polynomial regression
models) for predicting annual electricity and diesel fuel consumption on Italian dairy farms using
empirical data from 285 farms. For polynomial regression model development, they considered herd
Energies 2020, 13, 1288 13 of 25

size, the number of lactating cows, milk production (kg FPCM) and land area (hectares) as input
variables for both electricity and diesel models. Unlike Edens et al. [24], their methodology employed
a univariate statistical method (correlation matrix) for variable selection and analyzed second and
third order polynomial terms for herd size and number of lactating cows. The inclusion of the second
and third order polynomial terms for herd size and number of lactating cows were assessed through
Akaike Information Criterion (AIC) and VIF. Todde et al. [32] prescribed a maximum VIF value of ten.
As a result, the final electricity polynomial regression model only included the number of lactating
cows as an input variable, as shown in Equation (3).

TEi T = 14.3 + 0.19 × LCi − 1.9× 10−6 × LCi 3 (3)

where TEi T represents the annual electricity consumption (kWh) of the ith farm, LCi is the number of
lactating cows on the ith farm, and LCi 3 is number of lactating cows cubed on the ith farm.
Leave-one-out cross-validation (LOOCV) was employed to measure the capability of the developed
models to predict unseen electricity consumption (data not used for model development). Through
this LOOCV method, the electricity model was found to predict annual electricity consumption of the
Italian farms to within 11.4% (RPE).
Using an identical methodology, Todde et al. [32] also developed a polynomial regression model
to predict diesel energy consumption (kg) on Italian dairy farms, as shown in Equation (4). Using
LOOCV, their diesel model was found to predict annual diesel consumption of the Italian farms to
within 15.0% (RPE).

TFi T = 23.5 + 0.18 × TCi − 1.3× 10−7 × TCi 3 + 0.34 × Landi + 15.4 × MF (4)

where TFi T represents the annual diesel usage (kg) of the ith farm, TCi is the total number of cows on
the ith farm, and TCi 3 is the total number of cows cubed on the ith farm, Landi is the total hectares (ha)
of land used, and MF is the presence (1) or absence (0) of feed mechanization.
Concurrently, Todde et al. [17] developed a polynomial regression model to predict the embodied
energy per year per farm, as shown in Equation (5). Using LOOCV, their model was found to predict
annual diesel consumption of the Italian farms to within 13.3% (RPE).

EEi T = 90.4 + 0.52 × TCi − 8.5 × 10−7 × TCi 3 (5)

where EEi T represents the annual embodied energy (MJ) of the ith farm, TCi is the total number of cows
on the ith farm, and TCi 3 is the total number of cows cubed on the ith farm.

2.2.3. Machine-Learning
Studies by Sefeedpari et al. [50] and Sefeedpari et al. [20] also applied machine-learning to
predict energy output on Iranian dairy farms. These studies involved the development of an artificial
neural network, and an adaptive neural-fuzzy inference system (ANFIS) model. The data for model
development were attained through detailed questionnaires including: the amount of fossil fuel (diesel,
gasoline, kerosene, natural gas) and electricity consumed, milk produced, amount of cow manure,
farm area, and cow numbers. The total energy consumption values were then calculated using energy
intensity values found in literature. Each model was developed and tested utilizing a milk production
year equal to the lactation period of 305 days to improve the prediction capabilities of each model. For
model training, Sefeedpari et al. [50] refers to 60% of the original dataset used for model training, 20%
used for validation and 20% for testing. Sefeedpari et al. [38] found that the artificial neural network
model with 16 hidden neurons in the hidden layer with the Levenberg–Marquardt training algorithm
maximized the prediction accuracy of energy consumption with a R2 value of 0.88. When considering
the R2 values of the developed models, values of 0.79 and 0.11 were calculated for the ANFIS and MLR
models, respectively [20].
Energies 2020, 13, 1288 14 of 25

Shine et al. [40] also looked at the capability of improving the prediction accuracy of dairy farm
electricity consumption achieved by MLR modelling through employing various machine-learning
algorithms. Using the same dataset to Shine et al. [8], Shine et al. [40] assessed the applicability of
predicting monthly dairy farm electricity consumption using a support vector machine algorithm,
a CART decision tree algorithm, a random forest ensemble algorithm, and an artificial neural
network. Their methodology excluded variables that added little predictive power through employing
backward sequential variable selection. Additionally, hyper-parameter tuning was carried out in
conjunction with nested cross-validation to calculate the capability of each model to predict unseen
electricity consumption. Backward sequential variable selection allowed for those variables with
little predictive power to be removed, while hyper-parameter tuning allowed for a machine-learning
model that generalized well on unseen data. They found the support vector machine algorithm
maximized prediction accuracy of dairy farm electricity consumption, resulting in an RPE value of
12% (an improvement of 54% compared to MLR modelling). Shine et al. [40] commented on the
improved capabilities of machine-learning algorithms to improve prediction accuracy stems from their
ability to quantify non-linearities and interactions between input variables while offering an increased
flexibility regarding data multicollinearity, input data distributions, missing data points and pattern
recognition compared to standard statistical methods (such as MLR). Shine et al. [52] further assessed
the ability of the support vector machine model to predict dairy farm electricity consumption at an
annual resolution, both at the farm-level and catchment-level (combined consumption of multiple
farms). They found a negative correlation between prediction resolution and RPE. More specifically,
prediction error (RPE) reduced from 12%, to 10%, and to 5%, as the prediction resolution increased from
predicting monthly consumption, to annual farm-level consumption, and to annual catchment-level
consumption, respectively. This result demonstrated the models’ potential effectiveness as a simulation
tool for macro-level analyses.

2.2.4. Prediction Modelling Summary

It is clear from the literature that numerous international research studies have focused
on developing prediction models for dairy farm energy and/or electricity consumption. These
prediction models vary from mechanistic, MLR, polynomial regression, balance equation-based
models, to machine-learning models. In conjunction with different data from different countries being
employed for model training, the methods utilized for the development of these energy models vary.
These methods vary from the response variable considered (i.e. dairy farm electricity consumption or
diesel consumption), to the number of farms utilized for model training, to variable selection and model
validation methods. Thus, reported accuracies throughout the dairy energy prediction modelling
literature also vary. Therefore, it is difficult to determine which methodology provides the greatest
prediction capability for a particular objective. There is a need for a common methodology for model
development and assessment to be developed. However, the development of a prediction model for
a particular purpose is highly problem specific. Thus, defining a common methodology would be
difficult. However, a minimum methodology standard may be defined to ensure, at the very least that
the accuracy of a prediction model is not over-estimated. Although model prediction accuracy may not
be maximized in all cases, these prediction models may still offer adequate capability when forecasting
the potential impact of various strategies to reduce dairy energy consumption in different scenarios.

2.3. Dairy Energy Analysis

With increasing electricity demand, there exists the potential for improvements in efficiency
(usage per liter of milk) as dairy farm facilities may not be optimally configured for milk production
increases [4,27]. For example, utilizing the MECD [38], Upton et al. [53] found a farm with a herd
size of 88 dairy cows had a 24% greater electricity consumption per liter compared to a farm with
a herd size of 45 cows. Additionally, in Ireland, through carrying out a detailed statistical analysis,
Shine et al. [8] calculated that farms that employed ice chiller or IB milk cooling systems consumed
Energies 2020, 13, 1288 15 of 25

32% greater electricity compared to farms that employed DX systems. Thus, consideration for
electricity consumption per liter of milk produced must be considered, especially during periods of
milk production expansion.
Hartman and Sims [30] conducted an analysis of potential energy reduction strategies on three
New Zealand dairy farms. They highlighted that the majority of farms could reduce their electrical
energy usage and costs by between 10% and 20% through implementing various electrical energy
reduction and load shifting strategies. They highlighted that as electricity is a minor component
of overall production cost, simply reducing electrical energy usage may have a marginal impact on
profit. However, relatively inexpensive methods of load shifting or levelling may result in a greater
return on investment, especially as electricity pricing tariffs that charge different prices depending
on the time of day become more commonplace. To improve energy efficiency, they suggested that
a 70% reduction in milk harvesting energy use (20% of total electricity use) could be achieved by
employing a VSD. Concurrently, they highlighted the significant impact of pre-cooling milk prior
to entry into the milk bulk tank, as 90% of the energy required for milk chilling is used for initial
milk cooling, while the remaining 10% is required to maintain the milk temperature below 4◦ C. More
specifically, each 1 ◦ C drop in milk temperature set point prior to entry to the milk bulk tank would
result in a 5–7% reduction in chiller load [30]. Additionally, Hartman and Sims mention potential
water heating energy savings due to installing a heat recovery system to pre-heat the cold inlet water
by: 1) extracting heat from the milk flow, 2) using a small heat exchanger on the chiller condenser
coils, or 3) recycling residual heat from the used washing water. Regarding diesel fuel savings, they
identify the importance of dairy farmers learning the correct use of the hydraulic systems, keeping the
machinery well maintained, matching tractor size and ballast to the task in hand, and checking tyre
pressures. Although Hartman and Sims mentioned the greater potential for energy cost savings due
to load shifting or load levelling, they only identify the use of heat recovery and storage systems in
conjunction with IBs for the production and storage of heated and chilled water.
Barnett and Russell [13] reviewed strategies to reduce energy use on dairy farms. They highlighted
an increase in the proportion of total energy use associated with direct sources from 41% to 53% due
to irrigation practices on New Zealand farms [14]. They listed potential strategies to reduce water
heating, water cooling, milk harvesting, water pumping, and lighting energy usage and costs on dairy
farms. They report that the implementation of these strategies may reduce dairy energy usage by
43.6%. Barnett and Russell [13] also assessed the potential for alternative energy supplies including:
methane from manure (anaerobic digestion), wind power, hydro-electric power (small-scale), solar
water heating, and bio-diesel. They highlight that anaerobic digestion is more economically suited
to confinement dairy systems, whereby nearly 100% of the manure can be digested, compared to
between 10% and 20% in outdoor grazing systems. Thus, confinement systems require a minimum
herd size of between 250 and 300 cows to be economically viable, whereby an outdoor grazing system
requires a minimum herd size of 1,000 cows. Regarding wind power, they report that small-scale wind
energy (example used: 10kW) is not economical in the short to medium term. Barnett and Russell
also reported that small-scale hydroelectric power generation would only be suitable for very few
dairy farms that have the required flow capable of generating the magnitude of electricity required to
be economically viable. Regarding solar water heating, the potential for a 50% reduction in energy
required for water heating on a typical dairy farm is reported, however, the economic viability is not
discussed. Although they mention the potential for biodiesel to reduce fossil fuel consumption and
emission of GHGs, they fail to mention any studies that directly focus on this aspect of dairy energy
reduction. They conclude their study by mentioning that although energy savings are oftentimes given
low priority, this is likely to change in the future as the cost of energy rises and greater incentive is
given to reduce GHG emissions.
Rajaniemi et al. [27] highlighted that milk-cooling related electricity savings of approximately
30% may be achieved through modest improvements to plant infrastructure such as: improving the
ventilation of the milk tank room, placing the refrigeration condenser outside the building to improve
Energies 2020, 13, 1288 16 of 25

the operational COP, and installing milk pre-cooling and heat recovery systems. However, it should be
noted that these calculations were carried out through utilizing only three dairy farms, while electricity
saving potential was calculated using an un-validated model. Improvements in animal welfare, growth,
and milk production were reported by Crill et al. [54], whereby increases in cow milk production of
between 5% and 16% were calculated when exposed to between 16 and 18 hours of adequate light
over cows exposed to less than 13.5 hours of adequate lighting in a confinement system. Harner
and Smith [55] also recommend subjecting lactating cows to between 16 and 18 hours of continuous
light. In addition, Rajaniemi et al. [56] recommended illumination intensity of the milking parlor of
200–250 lux, in line with lighting level recommendations by Harner and Smith [55], who suggested
illumination levels of 215 lux (20 foot-candles). However, the lighting level of the milking pit must also
be considered, whereby ASABE [57] recommend an illumination level of 500 lux to ensure adequate
lighting for controlling hygiene and the attachment of milking clusters. In either case, the energy
intensity of the lighting equipment must be considered, whereby the installation of energy efficient
lighting may reduce overall electricity consumption by 9% [58].
Regarding milk harvesting, Dunn and Butler [59] reported milk harvesting savings of between 40%
and 50% due to the installation of VSD for milk harvesting in the United Kingdom, while Ludington
et al. [58] calculated milk harvesting savings of 33% across 32 dairy farms in the United States. Similarly,
on Danish AMS farms, Brøgger Rasmussen and Pedersen [34] highlighted farms that employed a
VSD for milk harvesting had 20 kWh lower electricity consumption per 24 hours compared to regular
vacuum pumping systems. In Ireland, Upton et al. [60] reported that the installation of a VSD for milk
harvesting reduced milking costs by 60%, on average. Conventional vacuum pumps are primarily
designed to operate at higher air flows than is required for milking due to higher air flows being
required for the adequate washing of milking equipment. This may result in air bleeding within the
system and energy wastage [61]. As highlighted by Rajaniemi et al. [56], Upton et al. [62], and Morison
et al. [61], VSDs reduce electricity consumption over conventional milking machine vacuum pumps
due to their ability to vary the speed of the motor to balance the rate of air flow removed from the
system and admitted to the system, while still producing adequate vacuum stability. In addition, the
installation of a VSD allows for milk pre-cooling to be carried out with a greater efficiency, whereby
Morison et al. [61] reported an improvement of 5 ◦ C.
With regards to milk-cooling, Ludington et al. [58] calculated milk-cooling energy reduction of
15% due to the installation of a pre-cooling system, whereby milk was pre-cooled prior to entry to
the milk bulk tank, reducing the electrical energy required for cooling, i.e. reducing the temperature
differential. Similarly, Karlsson and Nordman [63] measured a reduction in milk temperature from
37 ◦ C to 17 ◦ C, resulting in about a 50% reduction in milk-cooling electricity consumption when
pre-cooling milk through a PHE using a 1:1 milk to water flow rate [56]. This is comparable to similar
results in Ireland, whereby Upton et al. [62] recommended milk to water ratios of between 1:1 and 1:3
with the optimum ratio depending upon the size of the PHE and power of the milk pump.
Upton et al. [53] employed the MECD to investigate the impact of dynamic pricing environments
on dairy farm electricity costs. Electricity usage on Irish dairy farms is primarily bi-modal during the
milking season, due to the common 2-day milking schedule of dairy cows (once early in the morning
and once late in the afternoon) [4]. With 62% of daily electricity consumption occurring during the peak
electricity demand period (17:00–19:00), the impact of dynamic pricing tariffs may result in increased
electricity costs whereby higher tariffs are imposed on times of high consumption on the electricity grid
which may coincide with evening milking [4,53]. In response, Upton et al. (2015) found that for a twice
a day milking strategy, adjusting milking times to milk earlier in the morning and later at night offered
the greatest scope for energy and cost savings (flat, day/night, two time of use tariffs, and a real time
pricing tariff) due to variations in milk-cooling system COP. More specifically, adjusting milk starting
times (from default milking times) on a time-of-use tariff (€0.13 kWh−1 (range 0.08–0.23 € kWh−1 ))
resulted in electricity cost reduction of 39%, 34%, and 33% for small (herd size = 45 cows), medium
(herd size = 88 cows), and large (herd size = 195 cows) sized representative farms, respectively [53].
Energies 2020, 13, 1288 17 of 25

However, a real-time pricing environment may also offer opportunities for dairy farmers that have IB
milk bulk tanks installed, whereby pre-designed ice charging strategies may be employed to minimize
milk-cooling costs [64]. IB storage units may offer the electricity grid operators a variable load for
demand side management purposes during periods of high or low grid frequency or system voltage,
in particular with an increased penetration of intermittent power sources (e.g. wind and solar energy
systems) to the electricity grid [64].
In conjunction with defining electricity usage metrics, Shine et al. [8] also conducted a detailed
statistical analysis to determine key relationships between dairy farm characteristics as well as potential
differences between 45 dairy farms. A correlation analysis found that electricity consumption was
largely associated with milk production, herd size, and the number of lactating cows. They found
that across IB and DX systems, employing ground water for milk pre-cooling reduced electricity
consumption by 25% on average. They also found interesting results associated with employing IB
milk bulk tanks for milk-cooling. More specifically, they found that systems that utilized ice chiller
units or IB milk tanks consumed 21% less electrical energy during day-time hour’s due to their load
shifting capabilities (thus taking advantage of lower electricity rates). However, this decrease was met
with a 32% increased energy consumption per liter resulting in no difference in milk cooling cost per
liter of milk when under a day and night electricity tariff (day tariff of 0.18 € kWh−1 ; night tariff of
0.08 € kWh−1 from 00:00 to 09:00 h).
Shine et al. [52] utilized the support vector machine model [40] to conduct a macro-level analysis,
whereby the impact of increased milk production on electricity use was assessed on 16 Irish dairy farms.
Relative to a base scenario (no change in infrastructural equipment as milk production increased),
they found the greatest reduction in electrical energy requirement per liter of milk occurred when
all farms pre-cooled milk in conjunction with the installation of two additional parlor milking units.
The addition of two parlor units resulted in electricity savings due to reduced milking times, which
outweighed the increased electricity usage of the additional vacuum units.
Breen et al. [65] developed an optimization strategy to maximize return on investment in dairy
farm infrastructural equipment over a specific time horizon. This was referred to as the discrete
infrastructure optimization model for economic assessment on dairy farms (DIOMOND) and assessed
the performance of five optimization algorithms. These included: Dynamic Programming the Genetic
Algorithm, Particle Swarm Optimization, Simulated Annealing, and Tabu Search. The Genetic
Algorithm was found to offer greater efficiency and performance compared to the other optimization
algorithms for a test scenario involving a 195-cow farm. The optimal combination of farm technology,
management practices, and electricity tariff was identified, resulting in a 26.3% improved return on
investment over a ten year time horizon compared to a base investment scenario. Concurrently, Breen
et al. [44] developed a multi-objective optimization (DAIRYMOO) method (employed similar methods
to those used for DIOMOND) to identify farm infrastructure and managerial practices to 1) maximize
farm net profit and 2) minimize farm electricity related CO2 emissions, over a ten year time horizon.
This study incorporated solar thermal heating and heat recovery models developed and validated
using empirical data. However, heat recovery systems were only selected in the optimal scenario when
the combined objective function was weighted heavily towards minimizing CO2 emissions, while solar
thermal heating was never selected in the optimal farm configuration. These results suggested poor
financial performance for both heat recovery and solar thermal heating technologies.
Previous studies have utilized prediction models to quantify the impact of numerous methods to
reduce energy/electricity consumption and related costs on dairy farms such as: altering the placement
of the refrigeration system, heat recovery, installation of VSDs, milk pre-cooling, and the installation
of an IB milk-cooling system to allow for load shifting of electricity consumption to cheaper cost
rates. Pre-cooling milk with water through a PHE can reduce milk temperature from 37 ◦ C to 17 ◦ C,
substantially reducing energy demand and reducing the time taken to reduce milk temperature to
below 4 ◦ C. However, consideration must be given to the impact on the related water consumption,
Energies 2020, 13, 1288 18 of 25

as increases in milk production can result in unsustainable levels of water consumption during periods
of high stress.

3. Discussion and Perspective

The expansion of the global dairy industry poses challenges regarding minimizing environmental
impacts while ensuring dairy farmers can adjust their farming strategies to minimize production costs
in the volatile milk price environment [66]. Thus, literature related to the consumption of energy on
dairy farms was reviewed under the headings of monitoring, prediction modelling and analyses.
The monitoring of dairy farm energy consumption is well documented in numerous LCA research
articles. Although consumption metrics vary between studies (e.g. Lm , FPCM, ECM, kgm , etc.),
the reporting of fat and protein percentage values of the milk production used in the respective
studies allowed for the effective comparison between international studies. Concurrently, although
reporting energy usage metrics are useful for international comparisons, monitoring consumption can
be quite time-consuming, financially expensive and not offer much in terms of prediction accuracy
on a particular farm. Thus, literature has also focused on developing effective prediction models to
predict dairy farm electricity and water consumption.
Numerous prediction and analyses methodologies have been employed to predict dairy farm
energy consumption. Prediction modelling has involved mechanistic modelling, MLR modelling,
polynomial regression modelling and machine-learning. Thus, prediction modelling has largely
focused on empirical prediction modelling over mechanistic modelling. Although highly suitable to
simulating single dairy farms, a mechanistic model such as the MECD [38] may not be suitable to
predict large-scale dairy energy consumption (which may be required for environmental reporting)
due to requiring a large number of input variables that dairy farmers may not know without the use of
specialized equipment (i.e. milk to water ratio, water temperatures, water pump motor sizes, etc.).
However, empirical modelling can replace metering equipment and/or mathematical models (that
require a large number of input variables) with a small number of empirically derived coefficients, or
black box machine-learning models. It should be noted that the final developed models are country
specific due to the unique climatic conditions and farming practices
Balancing coarse input variables with acceptable prediction accuracy is difficult through standard
regression methods such as MLR, polynomial regression etc. Thus, recent research carried out
by Shine et al. [40] looked at the applicability of machine-learning algorithms to provide the dual
benefit of requiring a reduced number of input variables compared to mechanistic modelling without
compromising on prediction accuracy. Further work may look towards applying a larger range
of machine-learning algorithms to dairy farm energy consumption data collected internationally,
potentially stemming to deep-learning algorithms if/when applicable (large number of data points
required). Concurrently, the development of a global database containing international dairy energy
consumption values and descriptive variables could be developed, and models built to generate a
global dairy energy model. If such a model were to be developed, each country would require a
cohort of dairy farms to be selected that are representative of the country’s dairy farm demographic.
These farm’s energy consumption could then be monitored in conjunction with farm characteristics,
managerial strategies, and environmental data, and all collected data shared to a central database for
model development. This would greatly reduce any financial cost associated with calculating the
energy consumption on dairy farms globally as well as offer a means for international comparison.
In conjunction with national surveys carried out throughout each country (to collect data required
for energy and water models), such a model would offer countries the opportunity to continually
monitor dairy energy consumption per liter of milk while also assessing the continual impact of various
strategies aimed to reduce energy use on dairy farms.
Some studies have focused on identifying the most significant areas for reducing overall energy
consumption associated with the production of milk. For example, Wells [14] identified the following
areas as most important to consider for improving overall energy efficiency on New Zealand dairy
Energies 2020, 13, 1288 19 of 25

farms: fertilizer management, water management (irrigation), farm vehicle selection and operation,
insulation of hot water cylinders, and pre-cooling milk to reduce electrical energy use and related
costs. It is difficult to quantify the potential energy savings due to the implementation of best practice
methods on dairy farms as all farms are unique (i.e. the potential energy savings will vary from farm
to farm). However, from analyzing dairy energy literature, some conclusions can be made. More
specifically, the literature shows that on average, switching from conventional farming system to
organic farming system, will result in a 31% reduction in energy required per kg ECM. However, further
analyses may be required on a country-wide basis to calculate the economic viability of switching to
an organic farming system. If switching to an organic farming system is unfeasible, another option for
some conventional confinement dairy farms could be to switch to a pasture-based farming system (if
possible). On average, literature shows that energy consumption may be reduced by 37% by switching
from a conventional confinement system to a pasture-based system. This switch may also result in
financial savings, whereby Finneran et al. [67] identified a cost saving of 61% in switching a cows diet
from concentrate to grazed grass.

4. Conclusions
Literature related to the consumption of energy on dairy farms was reviewed with respect to
monitoring, prediction modelling, and data analyses. The future of the global milk production must be
sustained through the minimal impact to energy resources, while ensuring dairy farmers adjust their
farming strategies minimize production costs. Total primary energy consumption values ranged from
2.7 MJ kg−1 ECM on organic dairy farming systems to 4.1 MJ kg−1 ECM on conventional dairy farming
systems, whereby variances in energy intensity between farms that employed either confinement or
pasture-based farming systems were also identified. Across all studies (conventional and organic
farms), direct energy consumption was responsible for 32% of total primary energy. On average,
electrical energy made up 48% of direct usage, while other liquid fuels were responsible for the
remaining 52%. Concurrently, indirect energy sources were responsible for the remaining 68% of
total energy consumption. On conventional dairy farms, ancillary energy is responsible for 56% of
total energy needs on average across the literature, whereby feed is responsible for 76% of ancillary
energy representing the largest energy requirement on dairy farms. As the largest consumer of energy,
altering dairy cow feed practices may offer the greatest reduction to dairy energy requirements. For
example, a conventional confinement dairy farm may reduce feed energy requirements by 66% by
switching to a pasture-based system (if possible). Concurrently, considerable energy savings may be
achieved by switching from conventional farming to an organic farming system, whereby studies have
calculated minimal energy requirements associated with organic fertilizer production. Other strategies
to reduce dairy farm energy consumption include technologies which aim to reduce on-farm electricity
consumption, including pre-cooling milk through a plate cooler, variable speed drives, improving
hot water tank insulation, switching to energy efficient lighting, etc. Developed prediction models in
literature have largely focused on predicting dairy farm electrical energy consumption, however, some
studies have focused on modelling output energy, as well as diesel use and embodied energy on dairy
farms. In addition, varying modelling techniques have also been employed including mechanistic,
regression, and machine-learning models. It is difficult to determine which method is most applicable
to simulate dairy energy consumption on dairy farms, instead, the most applicable modelling technique
is largely dependent on its intended application. For example, to simulate farm-level electrical energy
consumption, whereby a large number of input data are available, mechanistic modelling may be
desirable. Concurrently, to simulate dairy farm energy consumption on a large scale, whereby a small
number of easily attainable input data are desirable, then regression or machine-learning methods
would be best suited. From a research perspective, increased prediction capabilities of dairy energy
prediction models associated with the utilization of machine-learning algorithm, will result in an
increased confidence in predictions, while also potentially allowing for a greater number of dairy farms
to be included in future lifecycle analyses as the need for electricity monitoring equipment is reduced.
Energies 2020, 13, 1288 20 of 25

Author Contributions: Conceptualization, P.S. (Philip Shine) and M.D.M.; methodology, P.S. (Philip Shine) and
J.U.; data curation, P.S. (Philip Shine), J.U., and P.S. (Paria Sefeedpari); writing—original draft preparation, P.S.
(Philip Shine); writing—review and editing, P.S. (Philip Shine), J.U., P.S. (Paria Sefeedpari), M.D.M.; visualization,
P.S. (Philip Shine); project administration, M.D.M.; funding acquisition, P.S. (Philip Shine) and M.D.M. All authors
have read and agreed to the published version of the manuscript.
Funding: This work has been funded by the Sustainable Energy Authority of Ireland under the SEAI Research,
Development and Demonstration Funding Program 2018, Grant number 18/RDD/317.
Conflicts of Interest: The authors declare no conflict of interest.

Appendix A
A range of usage metrics relating to milk analyses are utilized throughout dairy energy consumption
literature. These include: liters of milk (Lm ), kilograms of milk (kgm ) (Equation (A1)), fat and protein
corrected milk (FPCM) (Equation (A2)) [68], and energy corrected milk (ECM) (Equation (A3)) [69].
The units of FPCM and ECM are commonly utilized for international comparisons as it assures a
fair evaluation between farms with different breeds or feed regimes [68]. However, studies utilizing
different usage metrics may be compared through reporting of milk data variables such as the mean
percentage of fat (% or g/kg), the percentage of protein (% or g/kg), and/or the amount of lactose (g/kg),
depending on the key performance indicator required.

kgm = Lm × (milk volumetric mass density) (A1)

FPCM = kgm × ((0.1226 × % Fat) + (0.0776 × % Protein) + 0.2534) kg (A2)

ECM = Lm × ((0.383 × % Fat) + (0.242 × % Protein) + 0.7832)/3.1138 kg (A3)

where milk volumetric mass density equals 1.03 kg per liter of milk, Lm is the volume of milk in liters,
% Fat is the percentage of fat in milk, % Protein is the percentage of protein in milk.

Appendix B
In total, 36 research findings related to total, direct and electrical energy usage on dairy farms is
presented in Table A1. These studies were carried out across 17 countries including: Belgium (BEL),
Canada (CAN), France (FRA), Germany (DEU), Denmark (DNK), Spain (ESP), Estonia (EST), Finland
(FIN), Great Britain (GBR), Iran (IRN), Ireland (IRL), Italy (ITA), Japan (JPN), the Netherlands (NTL),
New Zealand (NZL), Sweden (SWE), and the United States of America (USA). The mean number of
farms per study was 24, while some studies (notated by “n/a” in Table A1) used national or regional
level parameters with no on-site analysis. The exact energy sources considered when calculating total
energy consumption in each study may be identified within each article, as some energy sources (such
as machinery and facilities) may not be considered due to unavailability of data. Energy values from
11 studies are presented in MJ kg−1 , one finding is presented in MJ kg−1 FPCM, while one finding is
presented in fat-corrected milk (FCM) due to unavailability of fat and protein percentage values for
conversion purposes. When possible (i.e. when fat and protein percentage values were available),
energy values per liter of milk, kg of milk, kg of MS, or kg of FPCM were converted to MJ kg−1 ECM
using Equations (A1), (A2), and (A3). Where direct or electrical energy consumption was given as
a percentage of the overall energy value, these values were calculated accordingly. Where electrical
energy values were presented in terms of electrical energy (i.e. kWh, Wh, etc.), these values were
converted to MJ using 3.6 MJ kWh−1 [22], in conjunction with either country specific or average EU-28
primary energy factor of electricity values [70,71]. Therefore, reporting of fat and protein percentage
values when describing the energy intensity of dairy farming is critical to ensure future provision of
comparisons between international studies.
Energies 2020, 13, 1288 21 of 25

Table A1. Energy consumption (MJ kg−1 ECM unless stated otherwise) values found in literature.

Study System5 Country n Total Energy Direct Electrical

Arsenault et al. [72] Conv-c CAN 1 4.871 n/a n/a
Arsenault et al. [72] Conv-g CAN 1 4.991 n/a n/a
Basset-Mens et al. [73] Conv-g NZL 1 1.511,4 n/a n/a
Cederberg and Flysjö [74] Org-g SWE 6 2.10 n/a 0.74
Cederberg and Flysjö [74] Conv-c SWE 17 2.66 0.93 0.59
Cederberg and Mattson [35] Org-g SWE 1 2.51 n/a n/a
Cederberg and Mattson [35] Conv-c SWE 1 3.55 n/a n/a
Frorip et al. [75] Conv-c EST 1 5.36 n/a n/a
Haas et al. [76] Org-g DEU 6 1.201 n/a n/a
Haas et al. [76] Conv-g DEU 6 1.301 n/a n/a
Haas et al. [76] Conv-g DEU 6 2.701 n/a n/a
Hartman and Sims [30] Conv-g NZL 62 3.903 2.03 1.17
Hospido et al. [36] AMS-c ESP 2 6.031 0.73 0.58
Kraatz [22] Conv-c DEU n/a 3.54 1.24 0.39
Meul et al. [15] Conv-c BEL 74 3.581 1.20 0.34
Mikkola and Ahokas [77] Conv-c FIN n/a 3.201 1.60 0.70
Nguyen et al. [78] Conv-g FRA 1 3.977 n/a n/a
O0 Brien et al. [5] Conv-g IRL 1 2.37 0.30 n/a
O0 Brien et al. [5] Conv-c IRL 1 4.02 0.21 n/a
Ogino et al. [79] Conv-c JPN 1 5.538 n/a n/a
Pagani et al. [16] Org-c ITA 3 1.97 0.80 n/a
Pagani et al. [16] Org-g USA 3 4.07 2.23 n/a
Pagani et al. [16] Conv-c ITA/USA 5 4.32 1.62 n/a
Pagani et al. [16] Conv-c ITA/USA 4 3.35 1.38 n/a
Refsgaard et al. [80] Org-c DNK 14 2.16 n/a 0.66
Refsgaard et al. [80] Conv-c DNK 17 3.34 n/a 0.66
Sefeedpari et al. [20,21] Conv-c IRN 50 8.05 1.57 0.26
Thomassen et al. [37] Conv-c NLD 10 5.15 0.62 0.356
Thomassen et al. [37] Org-g NLD 11 3.19 0.99 0.556
Todde et al. [17,29] Conv-c ITA 285 8.91 2.60 0.27
Upton et al. [4] Conv-g IRL 22 2.37 0.48 0.29
Van der Werf et al. [81] Org-g FRA 6 2.68 n/a n/a
Van der Werf et al. [81] Conv-c FRA 41 2.88 n/a n/a
Wells [14] Conv-g NZL 96 1.98 0.87 0.47
Williams et al. [82] Conv-g GBR n/a 2.441,2,4 n/a n/a
Williams et al. [82] Conv-g GBR n/a 1.551,2,4 n/a n/a
1 Values from study are in MJ kg−1 milk. 2 Value converted from liters to kg using milk volumetric mass density of

1.03 kg Lm −1 . 3 Value provided by Pagani et al. [16]. 4 Value provided by Upton et al. [4]. 5 Conv-c = conventional
confinement farm; Conv-g = conventional grazing farm; Org-c = organic confinement farm; Org-g = organic grazing
farm; AMS-c = conventional automatic milking system. 6 Converted from Wh kg−1 ECM using 3.6 MJ kWh−1 [22]
and mean EU-28 primary energy efficiency factor between 2010 and 2013 [71] 9 and using standard equations. 7
Values from study are in MJ kg−1 FPCM. 8 Values from study are in MJ kg−1 FCM (Fat-corrected milk)

References
1. Bruinsma, J.; Alexandratos, N. World Agriculture towards 2030/2050: The 2012 Revision. Available online:
http://www.fao.org/docrep/016/ap106e/ap106e.pdf (accessed on 9 March 2020).
2. FAO. Agricultural Outlook 2018-2027 Dairy and Dairy Products 2018. Available online: http://www.agri-
outlook.org/commodities/Agricultural-Outlook-2018-Dairy.pdf (accessed on 8 March 2020).
3. Agriculture and Horticulture Development Board-World Milk Deliveries 2019. Available online: https:
//dairy.ahdb.org.uk/non_umbraco/download.aspx?media=26233 (accessed on 1 August 2019).
4. Upton, J.; Humphreys, J.; Groot Koerkamp, P.W.G.; French, P.; Dillon, P.; De Boer, I.J.M. Energy demand on
dairy farms in Ireland. J. Dairy Sci. 2013, 96, 6489–6498. [CrossRef] [PubMed]
5. O’Brien, D.; Shalloo, L.; Patton, J.; Buckley, F.; Grainger, C.; Wallace, M. A life cycle assessment of seasonal
grass-based and confinement dairy farms. Agric. Syst. 2012, 107, 33–46. [CrossRef]
Energies 2020, 13, 1288 22 of 25

6. von Keyserlingk, M.A.G.; Martin, N.P.; Kebreab, E.; Knowlton, K.F.; Grant, R.J.; Stephenson, M.; Sniffen, C.J.;
Harner, J.P., III; Wright, A.D.; Smith, S.I. Invited review: Sustainability of the US dairy industry. J. Dairy Sci.
2013, 96, 5405–5425. [CrossRef] [PubMed]
7. Shortall, J.; Shalloo, L.; Foley, C.; Sleator, R.D.; O’Brien, B. Investment appraisal of automatic milking
and conventional milking technologies in a pasture-based dairy system. J. Dairy Sci. 2016, 99, 7700–7713.
[CrossRef] [PubMed]
8. Shine, P.; Scully, T.; Upton, J.; Shalloo, L.; Murphy, M.D. Electricity & direct water consumption on Irish
pasture based dairy farms: A statistical analysis. Appl. Energy 2018, 210, 529–537. [CrossRef]
9. Food and Apicultural Organization. Food and Agricultural Organization. Greenhouse Gas Emissions from the
Dairy Sector A Life Cycle Assessment; Food and Apicultural Organization: Rome, Italy, 2010. [CrossRef]
10. Baldini, C.; Gardoni, D.; Guarino, M. A critical review of the recent evolution of Life Cycle Assessment
applied to milk production. J. Clean. Prod. 2017, 140, 421–435. [CrossRef]
11. European Commission. Energy Union and Climate Action: Driving Europe’s Transition to A Low-Carbon Economy;
European Commission: Brussels, Belgium, 2016.
12. Lanigan, G.; Donnellan, T.; Hanrahan, K.; Paul, C.; Shalloo, L.; Krol, D.; Forrestal, P.; Farrelly, N.; O’Brien, D.;
Ryan, M.; et al. An Analysis of Abatement Potential of Greenhouse Gas Emissions in Irish Agriculture
2021–2030. Available online: https://www.teagasc.ie/media/website/publications/2018/An-Analysis-
of-Abatement-Potential-of-Greenhouse-Gas-Emissions-in-Irish-Agriculture-2021-2030.pdf (accessed on
8 March 2020).
13. Barnett, J.; Russell, J. Energy Use on Dairy Farms. Bull. Int. Dairy Fed. 2010, 443, 23–32.
14. Wells, C. Total Energy Indicators of Agricultural Sustainability: Dairy Farming Case Study; Technical paper;
Ministry of Agriculture and Forestry: Wellington, New Zealand, 2001.
15. Meul, M.; Nevens, F.; Reheul, D.; Hofman, G. Energy use efficiency of specialised dairy, arable and pig farms
in Flanders. Agric. Ecosyst. Environ. 2006, 119, 135–144. [CrossRef]
16. Pagani, M.; Vittuari, M.; Johnson, T.G.; De Menna, F. An assessment of the energy footprint of dairy farms in
Missouri and Emilia-Romagna. Agric. Syst. 2016, 145, 116–126. [CrossRef]
17. Todde, G.; Murgia, L.; Caria, M.; Pazzona, A. A Comprehensive energy analysis and related carbon footprint
of dairy farms, Part 2: Investigation and modeling of indirect energy requirements. Energies 2018, 11, 1–14.
[CrossRef]
18. Aguirre-Villegas, H.A.; Passos-Fonseca, T.H.; Reinemann, D.J.; Larson, R. Corrigendum to “Grazing intensity
affects the environmental impact of dairy systems”. J. Dairy Sci. 2019, 102, 923–925. [CrossRef] [PubMed]
19. Aguirre-Villegas, H.A.; Passos-Fonseca, T.H.; Reinemann, D.J.; Larson, R. Grazing intensity affects the
environmental impact of dairy systems. J. Dairy Sci. 2017, 100, 6804–6821. [CrossRef] [PubMed]
20. Sefeedpari, P.; Rafiee, S.; Akram, A.; Komleh, S.H.P. Modeling output energy based on fossil fuels and
electricity energy consumption on dairy farms of Iran: Application of adaptive neural-fuzzy inference system
technique. Comput. Electron. Agric. 2014, 109, 80–85. [CrossRef]
21. Sefeedpari, P.; Rafiee, S.; Akram, A.; Chau, K.-W.; Komleh, S.H.P. Modeling Energy Use in Dairy Cattle Farms
by Applying Multi-Layered Adaptive Neuro-Fuzzy Inference System (MLANFIS). Int. J. Dairy Sci. 2015, 10,
173–185. [CrossRef]
22. Kraatz, S. Energy intensity in livestock operations-Modeling of dairy farming systems in Germany. Agric.
Syst. 2012, 110, 90–106. [CrossRef]
23. Calcante, A.; Tangorra, F.M.; Oberti, R. Analysis of electric energy consumption of automatic milking systems
in different configurations and operative conditions. J. Dairy Sci. 2016, 99, 1–5. [CrossRef]
24. Edens, W.C.; Pordesimo, L.O.; Wilhelm, L.R.; Burns, R.T. Energy use analysis of major milking components
at a dairy experiment station. Appl. Eng. Agric. 2003, 19, 711–716. [CrossRef]
25. Hörndahl, T. Energy Use in Farm Buildings—A Study of 16 Farms with Different Enterprises 2008. Available
online: https://pub.epsilon.slu.se/3396/1/Eng-rapport145-v1.pdf (accessed on 8 March 2020).
26. Murgia, L.; Todde, G.; Caria, M.; Pazzona, A. A partial life cycle assessment approach to evaluate the energy
intensity and related greenhouse gas emission in dairy farms. J. Agric. Eng. 2013, 44, 186–190. [CrossRef]
27. Rajaniemi, M.; Jokiniemi, T.; Alakukku, L.; Ahokas, J. Electric energy consumption of milking process on
some finish dairy farms. Agric. Food Sci. 2017, 26, 160–172. [CrossRef]
28. Shortall, J.; O’Brien, B.; Sleator, R.D.; Upton, J. Daily and seasonal trends of electricity and water use on
pasture-based automatic milking dairy farms. J. Dairy Sci. 2018, 101, 1565–1578. [CrossRef]
Energies 2020, 13, 1288 23 of 25

29. Todde, G.; Murgia, L.; Caria, M.; Pazzona, A. A comprehensive energy analysis and related carbon footprint
of dairy farms, part 1: Direct energy requirements. Energies 2018, 11, 1–14. [CrossRef]
30. Hartman, K.; Sims, R. Saving Energy on the dairy farm makes good sense. In Proceedings of the 4th Dairy3
Conference, Palmerston North, New Zealand; 2006; pp. 11–21.
31. Teagasc. Teagasc National Farm Survey 2016 Dairy Enterprise 2016. Available online: https://www.teagasc.
ie/media/website/publications/2017/NFS-2016-Dairy-Enterprise-Factsheet.pdf (accessed on 17 May 2018).
32. Todde, G.; Murgia, L.; Caria, M.; Pazzona, A. Dairy Energy Prediction (DEP) model: A tool for predicting energy
use and related emissions and costs in dairy farms. Comput. Electron. Agric. 2017, 135, 216–221. [CrossRef]
33. Rossi, P.; Gastaldo, A. Consumi energetici in allevamenti bovini da latte. Inf. Agrar 2012, 3, 45–47.
34. Brøgger Rasmussen, J.; Pedersen, J. Electricity and Water Consumption at Milking; Danish Agricultural Advisory
Service: Aarhus, Denmark, 2004.
35. Cederberg, C.; Mattsson, B. Life Cycle assessment of Swedish milk production—A comparison of conventional
farming. J. Clean. Prod. 2000, 8, 49–60. [CrossRef]
36. Hospido, A.; Moreira, M.T.; Feijoo, G. Simplified life cycle assessment of galician milk production. Int. Dairy
J. 2003, 13, 783–796. [CrossRef]
37. Thomassen, M.A.; van Calker, K.J.; Smits, M.C.J.; Iepema, G.L.; de Boer, I.J.M. Life cycle assessment of
conventional and organic milk production in the Netherlands. Agric. Syst. 2008, 96, 95–107. [CrossRef]
38. Upton, J.; Murphy, M.; Shalloo, L.; Groot Koerkamp, P.W.G.; De Boer, I.J.M. A mechanistic model for
electricity consumption on dairy farms: Definition, validation, and demonstration. J. Dairy Sci. 2014, 97,
4973–4984. [CrossRef]
39. Fuentes-Pila, J.; DeLorenzo, M.A.; Beede, D.K.; Staples, C.R.; Holter, J.B. Evaluation of equations based on
animal factors to predict intake of lactating Holstein cows. J. Dairy Sci. 1996, 79, 1562–1571. [CrossRef]
40. Shine, P.; Murphy, M.D.; Upton, J.; Scully, T. Machine-learning algorithms for predicting on-farm direct
water and electricity consumption on pasture based dairy farms. Comput. Electron. Agric. 2018, 150, 74–87.
[CrossRef]
41. Shine, P.; Scully, T.; Upton, J.; Murphy, M.D. Multiple linear regression modelling of on-farm direct water and
electricity consumption on pasture based dairy farms. Comput. Electron. Agric. 2018, 148, 337–346. [CrossRef]
42. Murphy, E.; De Boer, I.J.M.; van Middelaar, C.; Holden, N.; Curran, P.; Upton, J. Predicting fresh water
demand on Irish dairy farms using farm data. Clean. Prod. 2017, 166, 58–65. [CrossRef]
43. Hanrahan, L.; Geoghegan, A.; O’Donovan, M.; Griffith, V.; Ruelle, E.; Wallace, M.; Shalloo, L. PastureBase
Ireland: A grassland decision support system and national database. Comput. Electron. Agric. 2017, 136,
193–201. [CrossRef]
44. Breen, M.; Murphy, M.D.; Upton, J. Development of a Dairy Multi-Objective Optimization (DAIRYMOO)
method for economic and environmental optimization of dairy farms. Appl. Energy 2019, 242, 1697–1711.
[CrossRef]
45. Ruelle, E.; Shalloo, L.; Wallace, M.; Delaby, L. Development and evaluation of the pasture-based herd
dynamic milk (PBHDM) model for dairy systems. Eur. J. Agron. 2015, 71, 106–114. [CrossRef]
46. Breen, M.; Murphy, M.; Upton, J. Development and validation of photovoltaic and wind turbine models to
assess the impacts of renewable generation on dairy farm electricity consumption. In 2015 ASABE Annual
International Meeting; American Society of Agricultural and Biological Engineers: New Orleans, LA, USA,
2015; pp. 1–11. [CrossRef]
47. Murphy, M.D.; Shine, P.; Breen, M.; Upton, J. DSSED: Decision Support System for Energy Use in Dairy
Production. Available online: https://www.seai.ie/resources/publications/SEAI-DSSED-Final-Report.pdf
(accessed on 9 March 2020).
48. Shine, P.; Breen, M.; Upton, J.; O’Donovan, A.; Murphy, M.D. A decision support system for energy use on
dairy farms. In Proceedings of the 9th European Conference on Precision Livestock Farming, Cork, Ireland,
26–29 August 2019; pp. 45–52.
49. SAS User’s Guide 2015. Available online: https://www.sas.com/en_ie/home.html (accessed on 16 May 2018).
50. Sefeedpari, P.; Rafiee, S.; Akram, A. Application of artificial neural network to model the energy output of
dairy farms in Iran. Int. J. Energy Technol. Policy 2013, 9, 82. [CrossRef]
51. Mhundwa, R.; Simon, M.; Tangwe, S.L. Modelling of an on-farm direct expansion bulk milk cooler to
establish baseline energy consumption without milk pre-cooling: A case of Fort Hare Dairy Trust, South
Africa. African J. Sci. Technol. Innov. Dev. 2017, 1338, 62–68. [CrossRef]
Energies 2020, 13, 1288 24 of 25

52. Shine, P.; Scully, T.; Upton, J.; Murphy, M.D. Annual electricity consumption prediction and future expansion
analysis on dairy farms using a support vector machine. Appl. Energy 2019, 250, 1110–1119. [CrossRef]
53. Upton, J.; Murphy, M.; Shalloo, L.; Groot Koerkamp, P.W.G.; De Boer, I.J.M. Assessing the impact of changes
in the electricity price structure on dairy farm energy costs. Appl. Energy 2014, 137, 1–8. [CrossRef]
54. Crill, R.L.; Hanchar, J.J.; Gooch, C.A.; Richards, S.T. Net Present Value Economic Analysis Model for
Adoption of Photoperiod Manipulation in Lactating Cow Barns. Pro-Dairy 2000, 1–6. Available online: https:
//ecommons.cornell.edu/bitstream/handle/1813/36961/photoperiod.pdf;sequence=1 (accessed on 8 March
2020).
55. Harner, J.P.; Smith, J.F. Lighting Low Profile Cross Ventilated Dairy Houses 2008. Available online:
https://www.asi.k-state.edu/doc/dairy/lighting-low-profile-cross-ventilated-dairy-houses.pdf (accessed on
8 March 2020).
56. Rajaniemi, M.; Turunen, M.; Ahokas, J. Direct energy consumption and saving possibilities in milk production.
Agron. Res. 2015, 13, 261–268.
57. Lighting Systems for Agricultural Facilities. Available online: https://www.spar.msstate.edu/class/EPP-2008/
Chapter%201/Reading%20material/Solar%20Radiation/Lighting%20Systems%20for%20Agricultural%
20Facilities.pdf (accessed on 8 March 2020).
58. Ludington, D.; Johnson, E. Dairy Farm Energy Audit Summary. Available online: https://www.nyserda.ny.
gov/-/media/Files/Publications/Research/Energy-Audit-Reports/dairy-farm-energy.pdf (accessed on 8 March
2020).
59. Dunn, P.; Butler, G.; Bilsborrow, P.; Brough, D. Energy + Efficiency—Renewable Energy and Energy Efficiency
Options for UK Dairy Farms 2010. Available online: https://docplayer.net/6454694-Energy-efficiency-
renewable-energy-and-energy-efficiency-options-for-uk-dairy-farms.html (accessed on 8 March 2020).
60. Upton, J.; Murphy, M.; De Boer, I.J.M.; Groot Koerkamp, P.W.G.; Berentsen, P.B.M.; Shalloo, L. Investment
appraisal of technology innovations on dairy farm electricity consumption. J. Dairy Sci. 2015, 98, 898–909.
[CrossRef] [PubMed]
61. Morison, K.; Gregory, W.; Hooper, R. Improving Dairy Shed Energy Efficiency. Available online: https:
//ir.canterbury.ac.nz/bitstream/handle/10092/11588/Dairy_Technical_Report.pdf;sequence=1 (accessed on
8 November 2019).
62. Upton, J.; Murphy, M.; French, P.; Dillon, P.; Systems, L. Dairy Farm Energy Consumption. In Proceedings of
the Teagasc National Dairy Conference, Cork, Ireland, 17–18 November 2010; pp. 87–97.
63. Karlsson, A.E.; Hörndahl, T.; Nordman, R. Energy recover from milk cooling. In Report 401; Agriculture &
Industry; JTI-Swedish Institute of Agricultural and Environmental Engineering: Uppsala, Sweden, 2012.
64. Murphy, M.D.; O’Mahony, M.J.; Upton, J. Comparison of control systems for the optimisation of ice storage
in a dynamic real time electricity pricing environment. Appl. Energy 2015, 149, 392–403. [CrossRef]
65. Breen, M.; Upton, J.; Murphy, M.D. Development of a discrete infrastructure optimization model for economic
assessment on dairy farms (DIOMOND). Comput. Electron. Agric. 2019, 156, 508–522. [CrossRef]
66. Donnellan, T.; Hennessy, T.; Thorne, F. The End of the Quota Era: A History of the Irish Dairy Sector and Its Future
Prospects; Teagasc: Galway, Ireland, 2015.
67. Finneran, E.; Crosson, P.; O’Kiely, P.; Shalloo, L.; Forristal, D.; Wallace, M. Simulation modelling of the cost of
producing and utilising feeds for ruminants on Irish farms. J. Farm Manag. 2010, 14, 95–116.
68. International Dairy Federation. Bulletin of the International Dairy Federation—A common carbon footprint
approach for dairy. Bull. Int. Dairy Fed. 2010.
69. Sjaunja, L.O.; Baevre, L.; Junkkarinen, L.; Pedersen, J.J.; Setälä, A. Nordic proposal for an energy corrected
milk (ECM) formula. In Proceedings of the 27th Session International Committee for Recording and
Productivity of Milk Animals, Paris, France, 2–6 July 1990.
70. Energy in Ireland 2018. Available online: https://www.seai.ie/resources/publications/Energy-in-Ireland-2018.
pdf (accessed on 8 March 2020).
71. Fritsche, U.R.; Greß, H.-W. Development of the Primary Energy Factor of Electricity Generation in the EU-28
from 2010–2013. Available online: http://www.iinas.org/tl_files/iinas/downloads/GEMIS/2015_PEF_EU-28_
Electricity_2010-2013.pdf (accessed on 8 March 2020).
72. Arsenault, N.; Tyedmers, P.; Fredeen, A. Comparing the environmental impacts of pasture-based and
confinement-based dairy systems in Nova Scotia (Canada) using life cycle assessment. Int. J. Agric. Sustain.
2009, 7, 19–41. [CrossRef]
Energies 2020, 13, 1288 25 of 25

73. Basset-Mens, C.; Ledgard, S.; Boyes, M. Eco-efficiency of intensification scenarios for milk production in
New Zealand. Ecol. Econ. 2009, 68, 1615–1625. [CrossRef]
74. Cederberg, C.; Flysjo, A. Life cycle inventory of 23 dairy farms in South-Western Sweden.
Available online: https://pdfs.semanticscholar.org/b930/2a75bd2a17e4398ca3c880927b7924d66911.pdf?_ga=2.
226958749.1090255613.1575905885-2041938671.1575905885 (accessed on 8 March 2020).
75. Frorip, J.; Kokin, E.; Praks, J.; Poikalainen, V.; Ruus, A.; Veermäe, I.; Lepasalu, L.; Schäfer, W.; Mikkola, H.;
Ahokas, J. Energy consumption in animal production—Case farm study. Agron. Res. 2012, 10, 39–48.
76. Haas, G.; Wetterich, F.; Köpke, U. Comparing intensive, extensified and organic grassland farming in
southern Germany by process life cycle assessment. Agric. Ecosyst. Environ. 2000, 83, 43–53. [CrossRef]
77. Mikkola, H.J.; Ahokas, J. Energy ratios in Finnish agricultural production. Agric. Food Sci. 2009, 18, 332–346.
[CrossRef]
78. Nguyen, T.T.H.; Doreau, M.; Corson, M.S.; Eugène, M.; Delaby, L.; Chesneau, G.; Gallard, Y.; van der
Werf, H.M.G. Effect of dairy production system, breed and co-product handling methods on environmental
impacts at farm level. J. Environ. Manag. 2013, 120, 127–137. [CrossRef] [PubMed]
79. Ogino, A.; Ishida, M.; Ishikawa, T.; Ikeguchi, A.; Waki, M.; Yokoyama, H.; Tanaka, Y.; Hirooka, H.
Environmental impacts of a Japanese dairy farming system using whole-crop rice silage as evaluated by life
cycle assessment. Anim. Sci. J. 2008, 79, 727–736. [CrossRef]
80. Refsgaard, K.; Halberg, N.; Kristensen, E.S. Energy utilization in crop and dairy production in organic and
conventional livestock production systems. Agric. Syst. 1998, 57, 599–630. [CrossRef]
81. van der Werf, H.M.G.; Kanyarushoki, C.; Corson, M.S. An operational method for the evaluation of resource
use and environmental impacts of dairy farms by life cycle assessment. J. Environ. Manag. 2009, 90, 3643–3652.
[CrossRef] [PubMed]
82. Williams, A.G.; Audsley, E.; Sandars, D.L. Determining the Environmental Burdens and Resource Use in the
Production of Agricultural and Horticultural Commodities; Cranfield University: Bedford, UK; Department for
Environment, Food and Rural Affairs (Defra): London, UK, 2006.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).