Emissions of greenhouse gases from energy use in agriculture, 1 forestry and fisheries: 1970-2019 2

. Fossil-fuel based energy use in agriculture leads to CO 2 and non-CO 2 emissions. We focus on emissions 14 generated within the farm gate and from fisheries, providing information relative to the period 1970-2019, for both 15 energy use, as input activity data and the associated greenhouse gas (GHG) emissions. Country-level information 16 is generated from UNSD and IEA data on energy in agriculture (including forestry and fisheries), relative to use 17 of: gas/diesel oil, motor gasoline, liquefied petroleum gas (LPG), natural gas, fuel oil and coal. Electricity used 18 within the farm gate is also quantified, while recognizing that the associated emissions are generated elsewhere. 19 We find that in 2019, annual emissions from energy use in agriculture were about 523 million tonnes 20 (Mt CO 2eq yr - 1 ), while including electricity they were 1,029 Mt CO 2eq yr -1 , having increased 7% from 1990. The 21 largest emission increasesincrease from on-farm fuel combustion werewas from LPG (32%), whereas significant 22 decreases were observed for coal (-55%), natural gas (-50%), motor gasoline (-42%) and fuel oil (-37%). 23 Conversely, use of electricity and the associated indirect emissions increased three-fold over the 1990-2019 period, 24 thus becoming the largest emission source from energy use in agriculture since 2005. Overall, the global trends were a result of counterbalancing effects: marked decreases in developed countries in 2019 compared to 1990 (- 26 273 CO 2 eq yr - 1 ) were masked by slightly larger increases in developing and emerging economies (+ 339 Mt 27 CO 2 eq yr -1 ). The information used in this work is available as open data at: https://zenodo.org/record/5153241 28 (Tubiello and Pan, 2021). The relevant FAOSTAT (FAO, 2021b) emissions database is maintained and updated 29 annually by FAO. for aquaculture and for powering fishing vessels. We include additional estimates of the emissions associated to the off-site generation of electricity used on the farm, tracking results both separately for electricity and on-site fossil fuel use, as well as in the aggregate. The analysis does not include all other indirect energy uses that are typically addressed in life-cycle analyses, such as embedded energy for manufacturing of agriculture machinery (FAO, 2011; Sims et al. 2015; FAO, 2018).


31
Meanwhile, Aagricultural production more than doubled over the period 1990-2019, with additional increases of 32 more than 50% expected to 2050, to meet projected increases in food demand (

23
Energy use in agriculture, forestry and fisheries nonetheless deserves more attention than paid in current reporting 24 and associated studies, because it is an important food production component deserving analysis in its own right 25 alongside the biophysical crop and livestock processes mentioned above. Additionally, it offers significant 26 opportunities for on-farm mitigation actions directly focussed on CO2 (Dyer et al., 2014). This paper therefore 27 focuses on quantifying the GHG emissions that arise from the combustion of fossil fuels for energy use in 28 agriculture, forestry and fisheries (capture fishing and aquaculture). As detailed in the methods section, our 29 quantification will focus mostly on the farm and on fishing activities, assuming that emissions associated to energy 30 used in forestry is negligible-i.e., it will focus on energy use for farm operations, for aquaculture and for powering 31 fishing vessels. We include additional estimates of the emissions associated to the off-site generation of electricity 32 used on the farm, tracking results both separately for electricity and on-site fossil fuel use, as well as in the 33 aggregate. Agricultural production more than doubled over the period 1990-2019, with additional increases of 34 more than 50% expected to 2050, to meet projected increases in food demand (FAO, 2018; Calicioglu et al., 2019).

35
Historically, productivity increases were achieved through transitions from traditional, extensive agri-food systems 36 to modern, intensive production systems, characterized by greater energy use within the farm (Smil, 2008). Direct 37 on-farm energy inputs include fuel to power tractors and other agricultural field machinery, irrigation pumps, heat 38 to warm greenhouses and animal shelters. Other uses beyond the farm may include power for forestry machinery 39 and fishing vessels. We consider herein additionally the energy used to generate electricity that may be used on   available literature and global estimates at 17 EJ, of which 5 EJ to power machinery; 4 EJ for animal husbandry, 26 aquaculture, and fisheries; 2 EJ to manufacture and maintain agricultural machinery; 5 EJ to extract, synthesize 27 and distribute fertilizers; 0.5 EJ to manufacture pesticides and herbicides; and 0.3 to manufacture irrigation 28 systems. Direct energy use in agriculture was a bit more than half this total, about 9 EJ. In addition to these 29 amounts, energy use in agriculture includes electricity from the grid, decentralized renewable sources including 30 bioenergy, conventional technologies, mechanical and thermal energy and biodiesel/biofuels. In many traditional 31 systems, human labour and draught animal power add significant energy inputs.

32
As opposed to GHG emission estimates from global analysis (top-down analysis), tThis paper therefore focuses 33 on quantifying the GHG emissions that arise from the combustion of fossil fuels for energy use in agriculture, 34 forestry and fisheries (i.e. capture fishing and aquaculture) with a "bottom-up" approach, i.e. using official 35 statistical data reported by countries to the UN Statistics Division. It also provides an overview of total emissions 36 and key trends at the global, regional and country level.

37
The dataset and the related analysis refers to one single 'agriculture' sector, which covers the three agricultural 38 sub-sectors: agriculture, forestry and fisheries. Some additional disaggregated information is provided for fishing 39 alone.  Historically, productivity increases were achieved through transitions from traditional, extensive agri-food systems 9 to modern, intensive production systems, characterized by greater energy use within the farm (Smil, 2008). Direct 10 on-farm energy inputs include fuel to power tractors and other agricultural field machinery, irrigation pumps, heat 11 to warm greenhouses and animal shelters. Other uses beyond the farm may include power for forestry machinery 12 and fishing vessels. We consider herein additionally the energy used to generate electricity that may be used on   25 EJ, of which 5 EJ to power machinery; 4 EJ for animal husbandry, aquaculture, and fisheries; 2 EJ to produce and 26 maintain agricultural machinery; 5 EJ to extract, synthesize and distribute fertilizers; 0.5 EJ to manufacture 27 pesticides and herbicides; and 0.3 to manufacture irrigation systems. Hence direct energy use in agriculture was a 28 bit more than half this total, about 9 EJ. In addition to these amounts, energy use in agriculture includes electricity 29 from the grid, decentralized renewable sources including bioenergy, conventional technologies, mechanical and 30 thermal energy and biodiesel/biofuels. In many traditional systems, human labour and draught animal power add 31 significant energy inputs. 32 33 34

35
Data on energy use in agriculture forestry and fisheries, by fuel type, over the annual time series 19790-2019, were 36 available from UNSD and IEA. These Agencies regularly collect energy data from member countries, including 37 for use in agriculture, forestry and fishing. Biofuels, renewables, and other energy carriers derived from biomass, 38 were analyzed but not considered for calculating GHG emissions, since they were assumed to be carbon neutral 39 (IPCC, 2006). In particular, UNSD energy consumption data were used to estimate GHG emission from agriculture as a whole, while IEA data were used to provide a breakdown for GHG from fisheries for information purposes. 1 UNSD data are publicly available through the UNDATA portal, while access to IEA data is restricted, and the 2 latter was kindly made available by IEA for this analysis. The dataset has used IEA energy data as a reference 3 since the UNSD database covers IEA data and expand its data from OECD countries to all the countries. Energy 4 use data from the UNSD Energy Statistics Database (UNSD, 2020) included the following fuels, over the period  time series for countries with no data were generated with a multivariate approach, i.e., by computing the sub-23 regional energy use in agriculture divided by the sub-regional total energy use, and applying the coefficient to the 24 time series of national total energy use, which was available in the UNSD database without major gaps. We 25 validated our gap-filling method by performing random substitutions of existing values and computing the 26 associated error, which was on average below 5%. 27 28

29
The activity data on energy use described in previous sections served as input for estimates of GHG emissions,

22
There are limitations and uncertainties associated with the estimates presented herein. First, we note that the input 23 data on energy refers to use in agriculture, forestry and fisheries, without further breakdown. While we refer often 24 to the associated emissions as generated within the farm gate, they include components of unknown relative 25 magnitude that are in fact generated through forestry and fisheries activities. For the latter, we have provided a 26 partial and incomplete breakdown in the database, using IEA fisheries data. Second, the underlying data on energy 27 use have significant geographical gaps, especially in Africa, as well as temporal gaps, particularly before 1990.

28
Out of 233 countries and territories, 51 were imputed in the energy emissions FAOSTAT database. However, these 29 are all small countries and their total share of global GHG emissions from energy use in agriculture is less than 30 1%. As mentioned above, the error associated with activity data gap-filling was on average below 5%. For The GHG emission data presented herein cover the period 19790-2019, at the country level, with regional and 10 global aggregates. Significant gaps in some countries and regions, especially Africa, imply that specific regional     country agricultural size, both in terms of area and economy. We defined GHG emission intensity per unit cropland 2 as the total GHG emissions from energy use in agriculture divided by total cropland area of a country.. Likewise, 3 energy GHG intensity per production value was computed by dividing total GHG from national energy use in 4 agriculture by total agricultural value added. . (Fig. 68). Data for denominators of both indicators were taken from 5 FAOSTAT (FAO, 2021a,bb, c).

6
Our results indicate that energy GHG emissions per unit cropland have been fluctuating but have been substantially 7 stable over the last two decades. Nonetheless, significant differences can be noted among regions (Fig. 10791).
8 While Europe has decreased significantly its energy-related GHG emission intensity in agriculture (-57%) in the 9 period 1990-2018, Africa, Central America and Asia have increased it substantially (+88%, +51% and +44% 10 respectively). This means that more GHG emissions are associated with the cultivation of one unit of cropland in 11 these regions. In absolute terms, the lowest energy intensity per unit of cropland in 2018 was achieved in Africa 12 (0.16 t CO2eq ha -1 ), followed by Oceania (0.38 t CO2eq ha -1 ), South America (0.42 t CO2eq ha -1 ) and Europe (0.48 13 t CO2eq ha -1 ). A clear diverging trend can be noticed between Annex I and non-Annex I countries, with the former 14 significantly decreasing the energy-related agricultural emissions intensity, and the latter significantly increasing 15 them (Fig. 8).

16
In terms of energy-related GHG emissions to agricultural value added, the picture is substantially different, with 17 Europe having significantly improved its energy intensity since 1990 (-68%), followed by Asia (-61%), Latin 18 America and the Caribbean (-54%), Northern America (-53%) and Oceania (-45%), while Africa's intensity 19 remained substantially stable over the last two decades.

20
This picture is significantly different when analyzing energy-related emission per capita (Fig. 9). Per capita, the 21 emission intensity is lowest in most African countries and India, while it is high in Canada, Australia and 22 Argentina, among others.

23
In 2019, high levels of GHG emissions per capita (from energy used in agriculture) were estimated for Faro Islands, 24 Greenland and Iceland (Fig. 1012). (Fig. 123). In those territories, emissions from gas/diesel oil take more than

20
The reason is that we focused only on electricity and on the most relevant fuels consumed in agriculture, but not 21 all. Specifically for fisheries, the relatively low coverage is also due to the fact that still few countries report 22 disaggregated energy consumption statistics for fisheries alone.

23
Electricity generation and gas/diesel oil used in agriculture were the two most important emissions sources, 24 responsible for roughly 40% of the total on average during the period 1990 -2019. Electricity is used for different 25 agriculture purposes: irrigation, processes that require heat or mechanical power, such as drying or milling. LPG, 26 natural gas, and heavy fuel oil are typically used for heat generation and, in some rare cases, for motive power.

27
Apart from some sharp variation of their total consumption in agriculture between consecutive years, mainly at the 28 beginning of the '90s, probably due to reporting issues of important consumer countries such as India and the 29 dissolution of the USSR, their emissions remained relatively stable. Compared to other emissions, coal and fuel 30 oil emissions decreased over the last few years, while agricultural production still increased. This can be explained 31 by updated energy use structure -the increased uptake of cleaner energy carriers such as electricity and LPG over 32 fuel oil and coal for heating. China, for example, one of the major emitting countries, decreased emissions from 33 fuel oil use by 48%, while increased emissions due to diesel use by around 59 % and emissions due to electricity 34 use by over 170% over the same period 1990-2019. There is anyway still a long way to go to decrease emissions 35 in the agricultural sector in China, due to its still very high reliance on coal as a heat source.