Detailed analysis of animal manure-based CO2 emissions, coal, electricity, thermal energy, and CH4 emissions and using machine learning as a forecasting method: a study from Turkey

doi:10.21203/rs.3.rs-1648218/v1

Download PDF

Research Article

Detailed analysis of animal manure-based CO₂ emissions, coal, electricity, thermal energy, and CH₄ emissions and using machine learning as a forecasting method: a study from Turkey

https://doi.org/10.21203/rs.3.rs-1648218/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

In this study, the animal manure-based biogas potentials of Antalya, Isparta, and Burdur provinces in the Western Mediterranean region of Turkey were calculated. CO₂ emission, coal, electricity, and thermal energy, CH₄ emission values were calculated and machine learning modeling was performed. A total of 550305 tons/year of manure can be obtained from this animal's existence. As a result of the calculation, 130929541 m³/year of biogas can be produced from manure. Depending on biogas production; 93484 tons/year of coal, 235673174 kWh/year of electrical energy, and 2828078093 MJ/year of thermal energy can be provided and 153653 tons/year of CO₂ emissions will be prevented. In this study, special information on cattle, sheep, and poultry, and animal age, number, and manure amount information were used in detail. To determine the model with the best results, machine learning algorithms Support Vector Machine (SVM), Multi-Layer Perceptron (MLP), and Linear Regression (LR) were used and hyper-parameter optimization was performed. According to the results of biogas potential, CO₂ emission, electricity production, and thermal energy estimations SVM models are seen as the best models with R² = 0.999. When the coal amount estimation is examined, the LR models produce better results than SVM and MLP with R² = 0.997. In the estimation of CH₄ using the Tier 1 approach, the MLP model can perform the best estimation with R² = 0.977. In the CH₄ modeling obtained using the Tier 2 approach, the LR models were superior to the other models with the performance value of R² = 0.962.

Manure

Biogas

Energy Potential

Methane

Carbon dioxide

Machine Learning

In recent years, events such as population growth and technological developments have led to an increase in global energy demand. Fossil fuels have the largest share in the current energy supply in the world and constitute approximately 80% (Safieddin Ardebili 2020; Khoshgoftar Manesh et al. 2020). It is widely stated that the visible increase in greenhouse gases resulting from the widespread use of fossil fuels in the last decade has caused significant changes in the global climate. In addition, there is concern that these non-renewable fuels will run out in the future (Alatzas et al. 2019; Le et al. 2020). Renewable energy is the fourth largest energy source after oil, coal, and natural gas. The use of renewable energy sources is increasing to meet the world's energy demand by reducing the damage to the ecosystem, improving the conversion methods, and increasing their efficiency (Razmjoo et al. 2021; Aravani et al. 2022). Many developed countries are actively becoming less dependent on fossil fuels by replacing natural gas with biofuels such as biogas, and biochar instead of coal (Chowdhury et al. 2020; Zamri et al. 2021; Siddiki et al. 2021). The conversion of waste materials into biogas is a biological process. As a sustainable carrier, biogas usually consists of methane (CH₄) (35–40%) and carbon dioxide (CO₂) (60%). It also consists of various gases such as ammonia, hydrogen sulfide, hydrogen, oxygen, nitrogen, and carbon monoxide (Khalil et al. 2019; Pramanik et al. 2019; Zabed et al. 2020; Ramírez-Islas et al. 2020).

1.1. Renewable energy and biomass from animal manure in Turkey

Turkey is very suitable for energy crops due to both climate and terrain conditions. In addition, agricultural and animal wastes are also quite high. However, the role of biomass in installed power is less than hydroelectric, geothermal, and wind energies. According to the Electricity Transmission Company, the ranking of Turkey's gross electricity production by primary energy sources as of 2020 is as follows: 78094 GWh hydroelectric, 70931 GWh natural gas, 67873 GWh hard coal and imported coal, 45806 GWh geothermal/wind/solar, 37938 GWh lignite and 5736 GWh renewable wastes and waste heat. Accurately determining the amount and distribution of biomass resources is very important in making strategic decisions such as energy management policies (Erdin and Ozkaya 2019). Different incentives are given to businesses that produce from different renewable energy sources to reduce carbon emissions and energy costs in Turkey. Considered an important mechanism in the increase of renewable energy installed capacity in Turkey, within the scope of the Renewable Energy Resources Mechanism for 2022, 7.3 dollars/cent for wind and hydroelectricity, 10.5 dollars/cent for geothermal, 13.3 dollars/cent for biomass and solar energy incentive is provided (TETC 2021; EPDK 2022; MI&T 2022; RTME&NR 2022).

Turkey's electricity consumption is increasing by 9% every year. Renewable energy sources that contribute to the total electricity need in Turkey have a relatively low share. It is thought that Turkey's electricity demand will exceed 424 TWh in 2023, and for this reason, it is aimed that the electricity generation rate from renewable energy sources will be at least 30% by 2023. It is estimated that Turkey's biomass waste potential is approximately 8.6 Mt of oil equivalent (MTEP) and the amount of biogas that can be produced is 1.5-2 MTEP (Melikoglu 2017; Rincon et al. 2019; Ocak and Acar 2021). Petroleum and natural gas are the primary sources on which Turkey is dependent on energy. In this context, the use of biomass as an alternative source to meet the increasing energy demand and reduce the dependence on foreign conventional sources deserves detailed analysis (Bakay and Ağbulut 2021; Yurtkuran 2021; Şenol et al. 2021). It is noteworthy that Turkey has significant biomass potential as it is the world's 7th largest agricultural biomass producer. The International Energy Agency estimates that the total installed power of biomass power plants in Turkey will increase by 630 MW from 2020 to 2025 (Erdin and Ozkaya 2019; TETC 2021; Erat et al. 2021; IEA 2022; Gündoğan and Koçar 2022).

As of 2022, Turkey's population is over 84.5 million. In 2021, there are 18240000 cattle, 57519000 small ruminants, and 98115000 poultry in Turkey (TUIK 2022).

Depending on the diversity of climate, vegetation, and landforms, many animal husbandry species have developed in Turkey. Cattle, small ruminants, poultry, beekeeping, silkworm breeding, and fishing are the main types of livestock (IOPRT 2021; World Bank 2021).

Cattle-raising is in the form of pasture and livestock breeding. Stock farming is done with modern methods. This type of animal husbandry has developed mostly in big cities that are consumption centers or close to such cities. The small ruminant in Turkey are sheep, hair goats, and angora goats. The number of sheep is constantly increasing and they are generally raised in the interior of Turkey. Hair goat breeding has been raised in every region of Turkey, Angora goat breeding has been raised in the Mediterranean region. Agricultural production and animal husbandry are widespread in Turkey due to both climate and land conditions (Font-Palma 2019; Şenol et al. 2021; Ilbas et al. 2022). The number of cattle breeding enterprises registered in the Turkvet Animal Registration System in Turkey is 1295632, and the number of sheep and goat breeding enterprises is 444446 (World Bank 2021; RTMAF 2022). Accordingly, agricultural and animal wastes are quite high.

Livestock has a significant share in Turkey's greenhouse gas emissions. Therefore, it is aimed to find innovative approaches and solutions to reduce greenhouse gas emissions from manures. Sustainable manure management systems should be implemented on farms that reduce the risk to the environment and allow the storage, transportation, and use of manure. Animal manure is a type of biomass and biogas can be produced. When using the obtained biogas, digested substrate or decay product residues can also be used as a valuable fertilizer (Cheng et al. 2021; Karaaslan and Gezen 2022).

Turkey aims to reduce natural gas imports and increase biogas production to meet its energy needs and reduce foreign dependency. If Turkey's 2023 energy targets are realized, it is estimated that Turkey's natural gas need will reach 73400 Mm³ in 2026. In the case of effective use of agricultural and animal waste potential, it is thought that a significant amount of natural gas will be saved, and thus natural gas imports can be significantly reduced (Melikoglu and Menekse 2020; Gündoğan and Koçar 2022). The number of power plants producing electricity from biomass in Turkey is 199. As of 2022, the capacity of the 106 biogas power plants operating in Turkey is approximately 588 MW, contributing 1.21% of the installed power capacity. Data on electricity generation based on renewable resources obtained from the Electricity Transmission Company are given in Fig. 1 (TETC 2021; Ocak and Acar 2021; IEA 2022).

1.2. Livestock production in the Western Mediterranean Region, Turkey

In Isparta, Burdur, and Antalya, which are located in the Mediterranean region of the country, cattle-raising (culture-hybrid -native-buffalo) is carried out. Sheep (native-merino), and goats (hair-angora) are raised as small ruminants. As poultry, laying hen, goose, turkey, and the duck is preferred (TUIK 2022).

The Western Mediterranean Region includes the provinces of Antalya, Burdur, and Isparta, designated TR61 in IBBS Level 2. With an area of 36797 km², the region constitutes approximately 4.7% of Turkey's area and 3.8% of its population. The total population of Antalya, Isparta, and Burdur is 3134694. While the annual population growth in Turkey was 14.7 per thousand in 2018, this value was 25.9 per thousand in Antalya, the 5th city with the highest population, 17.3 per thousand in Isparta, and 19.3 per thousand in Burdur. According to 2018 data from Turkstat, 1.20% of the total animal products value was realized by Antalya, 1.57% by Burdur, and 0.95% by Isparta province (RTMAF 2018; TUIK 2022). Animal husbandry in the region is generally in the form of small-scale family businesses and is carried out together with plant production activities. The combination of animal husbandry and plant production limits professionalization in animal production. Most animal shelters have closed and fixed attachment features. This situation adversely affects animal welfare and animal health. In the country, the Mediterranean region constitutes 4% of the sheep and goat population. In addition, the region ranks ninth among Level 2 regions in terms of sheep and goat population (WMDA 2022).

In 2018 for Antalya, it was reported that the number of cattle was 85833 (55.99% pure culture, 32.22% cultural hybrids, 6.57% native, 5.06% hybrid, 0.05% buffalo), the number of small ruminants was 1245651 (39.65% sheep, 60.35% goat) and the number of poultry was 530582 (92.25% of laying hen, 0.87% duck and guinea fowl, 1.8% goose, 5.15% turkey). The livestock support for Antalya in 2018 is 69491039 in Turkish Lira. In addition, 445 cattle and 1190 sheep and goats were donated to Antalya within the scope of the young farmer project (TUIK 2022).

Approximately 70% of the population of Burdur is engaged in agriculture and animal husbandry. Culture breed cattle constitute 98.4% of the cattle. There are 23000 registered farmers and 16000 active livestock enterprises in the province. According to Turkstat data, there are 14810 cattle and 5902 small ruminant farms in Burdur. The livestock support for 2018 in Burdur is 67019938.29 in Turkish Lira. For 2018 in Burdur, the number of cattle is 222843 (89.69% pure culture, 8.84% culture cross, 0.46% native, and 0.85% crossbreed), the number of small ruminants is 410449 (45.07% goat, 54.92% sheep) and the number of poultry is 205813 (4.46% turkey, 0.54% goose, 0.2% duck and guinea fowl, 96.71% laying hen). In addition, 515 cattle and 1020 small ruminants were donated to Burdur within the scope of the young farmer project in 2018 (TUIK 2022).

In Isparta for 2018, the number of cattle is 145012 (65.80% pure culture, 15.79% culture cross, 15.17% native, 3% hybrid, and 0.21% buffalo), the number of small ruminants is 491550 (56.9% sheep, 43.10% goat), the number of poultry is 445574 (3.42% turkey, 0.42% goose, 0.80% duck and guinea fowl, 95.34% laying hen). According to the data of the Turkish veterinary information system, there were 13223 cattle and 4577 small ruminant enterprises in Isparta. 490 cattle and 1020 small ruminants were donated to Isparta within the scope of the young farmer project in 2018 (TUIK 2022).

1.3. Related studies in the literature

Studies on theoretical and experimental biogas production, conversion, and calculation for various countries and regions of the world are available in the literature. There are also studies on artificial intelligence in this field. Some of these studies from recent years are given here.

Najafi et al. (Najafi and Faizollahzadeh Ardabili 2018) used ANFIS, ANN models to predict small-scale biogas production. When the results of the models are compared, it is stated that the ANFIS predicts the output values better. Avcıoğlu et al. (Avcıoğlu et al. 2019) calculated the energy potential of agricultural biomass residues in Turkey. It has been stated that the total agricultural residue in Turkey is 75084 kt and the theoretical energy potential is 998473 TJ. De Clercq et al. (De Clercq et al. 2019) created a machine learning model that can predict biogas output based on waste input to improve production in the industry. The machine learning model used consists of logistic regression, SVM, random forest, extreme gradient boosting, and k-near neighbor regression. Beltramo et al. (Beltramo et al. 2019) used an optimized ANN model to predict the biogas production rate of an agricultural biogas plant. Bao et al. (Bao et al. 2019) have calculated the biogas potential that can be obtained from China's animal manure. According to the livestock statistical data of China, it is stated that the biogas potential was 61000 Mm³ in 2015, it was estimated that this amount could change between 86000 and 110000 Mm³ by 2030. Melikoglu and Menekse (Melikoglu and Menekse 2020) estimated the biomethane potential of Turkey from cattle and sheep manure until 2026. It has been stated that approximately 2140 Mm³ of biomethane can be obtained in 2026. It has been reported that 6600 GWh of electricity can be obtained from biomethane production obtained, and it can meet approximately 2.9% of the natural gas need in 2026. Stolarski et al. (Stolarski et al. 2020) examined the development of bioenergy technologies in some European countries. It has been determined that the greatest potential for agricultural biomass is in Germany and Poland. It is stated that 92% of all biogas plants are in Germany. Oliveira et al. (Oliveira et al. 2020) examined the energy potential of manure and municipal waste in Brazil. In the study, a mathematical model based on multiple LR was created to predict the energy-related electricity generation potential. Elmaz et al. (Elmaz et al. 2020) used the machine learning method to predict the results of biomass gasification. In the study, estimation methods of polynomial regression, SVM, decision tree regression (DTR), and multilayer heuristics were used. MLP and DTR showed the best performance compared to other methods. Ulusoy et al. (Ulusoy et al. 2021) analyzed the biogas and energy production potential from chicken manure in Balıkesir, Turkey. The pilot plant processes 110 kt of waste and produces 8.58 Mm³/year of biogas. With this biogas production, 17 GWh/year of electricity and 16 GWh/year of thermal energy can be obtained. Seo et al. (Seo et al. 2021) applied ANN to predict the biogas output of dry anaerobic digestion of food waste. The model presented a derived R² of 0.82 for validation data and 0.85 for test data, indicating a high linear correlation between datasets. Siddiki et al. (Siddiki et al. 2021) investigated the manure, biogas, and energy production of Bangladesh in 2018–2019. As a result, 27923.72 Mm³/year of biogas will be produced from 486.77 Mt of manure. The thermal energy equivalent of the biogas potential is 512 PJ/year, and the electrical energy equivalent is 50 TWh/year. Ceylan et al. (Ceylan et al. 2021) developed a hybrid optimization model for determining the optimum installation location of a biogas power plant for Manisa, Turkey. The mathematical model of the process was determined by the neuro-regression approach. The traditional and hybrid models were compared and it was concluded that the values of the hybrid model were more acceptable. Ocak and Acar (Ocak and Acar 2021) evaluated the energy production potential of the Marmara region, Turkey. It was concluded that the conversion of agricultural and animal wastes into biogas and then into electricity is an economical option. Jeong et al. (Jeong et al. 2021) estimated biogas production from a municipal wastewater treatment plant in South Korea. The anaerobic digestion process was tried to be modeled using deep learning-based models. The highest success in terms of R² value was found to be 0.76. Oliveira (Oliveira et al. 2021) created a mathematical model for the estimation and optimization of the energy potential from animal manure and sewage in the Brazilian state of Minas Gerais. In the study, a nonlinear deterministic constructive algorithm was used. Şenocak et al. (Senocak and Guner Goren 2022) estimated the agricultural and animal amount, yield, and energy potential expected to emerge in the coming years of various agricultural and animal-based biomass resources of Acıpayam - Denizli, Turkey, using SVM and performed spatial analysis.

It is predicted that biogas production based on manure may play an important role in the future of renewable energy in Turkey. Turkey has forward-looking goals to increase renewable energy production and reduce greenhouse gas emissions. In addition, investments in this area in Turkey are mostly in the establishment phase and are expected to continue increasing. With the growth in the software system in Turkey, it is foreseen that there will be one hundred thousand additional employment and an export potential of ten billion dollars by 2025. The software industry, with its unique global dynamics, creates a qualified and high-income employment opportunity in the country. Turkey's studies on the use of artificial intelligence methods continue (TÜSİAD 2022). For the use of livestock operators, researchers, and policy planners, accurate estimation of energy conversions according to Turkey's manure-based production potential are of great importance. With such estimation methods used in the study, the feasibility of animal manure-oriented biogas production in Turkey can be analyzed in detail, and its impact on energy and the economy can give an idea to researchers and policy planners. Energy conversion calculations based on fertilizer potential for the coming years can contribute to the planning of Turkey's future energy needs.

In this study, the provinces of Antalya, Isparta, and Burdur, which are located in the TR61 region of Turkey, were selected. This region is also named the Western Mediterranean and Lakes Region and is one of Turkey's important livestock bases. For example, 40% of Burdur's economy is based on milk production (RTMAF 2018). Biogas amount, CO₂ emission, coal, electricity, thermal energy, and CH₄ values were modeled by using general and special information about cattle, small ruminants, and poultry, and animal age, number, and waste amount information. No detailed analysis has been found in such studies in which artificial intelligence techniques are used in the literature. In the theoretical biogas calculations for Turkey, artificial intelligence applications were not included, although detailed. Popular machine learning algorithms in the literature were used for modeling in the study.

The study is given in four parts. In the first part, statistical information about the biogas and the study area is given and a literature review is included. In the second part, the parameters used in the calculation of biogas and energy potentials and the proposed methods for estimation are explained in detail. In the third part, the findings of the experimental results of machine learning algorithms are compared under various scenarios. In the last part, the results are given in detail and suggestions have been made.

2.1. Theoretical determination of animal manure, biogas, and energy potential

In the study, 2018 data from the Turkstat of Antalya, Isparta, and Burdur were used. As it is known, the Covid-19 Pandemic started in 2019. Therefore, data from 2018 were used in this study. The post-pandemic situation and its effect on livestock are the subjects of future studies. Information on the number of animals is given in Fig. 2. Each animal species was evaluated separately in its category. In the calculation of the manure amount, the data obtained from the farms of the relevant provinces were used for the live mass values according to the animal species and breed for each age group. In determining the daily amount of fresh manure, the percentage of live weight values was used, since a value that can represent Turkey, in general, is not available. These values were taken as 6% for cattle, 5% for small ruminants, and 4% for poultry. Using these values, the daily fresh manure values were calculated separately for each province according to the age and type of cattle and small ruminants, and poultry separately, and the total amount of manure was determined. The amount of animal manure varies according to the feeding, climatic conditions, and reproduction type. Availability coefficient (AC) according to animal species was taken as 50% for cattle, 13% for small ruminants, and 99% for poultry, respectively (Avcioǧlu and Türker 2012; Afazeli et al. 2014; Scarlat et al. 2015).

Details of the animal species are given in Tables 2, 3, and 4 (Dong et al. 2006). In these tables, VS, B₀, MCF, MS (defined in Table 5) data used as parameters in CH₄ calculation are also given. These parameters were used in the Tier 2 approach. Table 1 is a summary of Tables 2, 3, and 4 (Avcioǧlu and Türker 2012).

Table 1. Manure characteristics and biogas yields by animal breeds (Avcioǧlu and Türker 2012).

Animal	Age Range Month (Categorical)	Live Mass (kg)	Fresh Manure Amount		Solid Manure (SM) (%)	Availability (AC)	Biogas yield l/kg
Animal	Age Range Month (Categorical)	Live Mass (kg)	% Mass	kg/day	Solid Manure (SM) (%)	Staying time in the barn (%)	Biogas yield l/kg
Cattle	x<12 12<x<24 X<24	200-900	5-6	10-20	5-25	Dairy 65 Beef 25	200-350
Small Ruminant	x<6 6<x<12 12<x<24 X<24	20-100	4-5	2	30	13	100-310
Poultry		2-10	3-5	0.08-0.1	10-35 50-90	99	310-620 550-650

Table 2

Parameters and values used for cattle (Avcioǧlu and Türker 2012).
Animal	Age Range Month (Categorical)	Live Mass (kg)	VS	B₀	MCF(%)	MS(%)
Dairy Cattle (Pure culture, female)	x < 12	300	2.8	0.13	0.2	0.38
Dairy Cattle (Pure culture, male)	x < 12	350
Dairy Cattle (Pure culture, heifer)	12 < x < 24	400
Dairy Cattle (Pure culture, cow)	24 < x	600
Dairy Cattle (Culture hybrid, female )	x < 12	350
Dairy Cattle (Culture hybrid, male)	x < 12	350
Dairy Cattle (Culture hybrid, heifer)	12 < x < 24	500
Dairy Cattle (Culture hybrid, cow)	24 < x	600
Dairy Cattle (Native, male)	x < 12	200
Dairy Cattle (Native, female)	x < 12	200
Dairy Cattle (Native, male)	12 < x < 24	250
Dairy Cattle (Native, cow)	24 < x	275
Dairy Cattle (Buffalo, heifer)	12 < x < 24	400
Dairy Cattle (Buffalo, cow)	24 < x	450
Pure Culture Cattle (Female)	x < 12	300	2.3	0.1	0.35	1
Pure Culture Cattle (Calf)	x < 12	350
Pure Culture Cattle (Bullock)	12 < x < 24	500
Pure Culture Cattle (Ox)	24 < x	850
Pure Culture Cattle (Bull)	24 < x	900
Hybrid Cattle (Bullock )	12 < x < 24	600
Hybrid Cattle (Bull)	24 < x	800
Hybrid Cattle (Ox)	24 < x	900
Native Cattle ( Bullock )	12 < x < 24	475
Native Cattle (Ox)	24 < x	475
Native Cattle (Bull)	24 < x	600
Buffalo (Male)	x < 12	250
Buffalo (Female)	x < 12	250
Buffalo (Bulllock)	12 < x < 24	400
Buffalo (Ox)	24 < x	500

Table 3

Parameters and values used for small ruminants (Avcioǧlu and Türker 2012).
Animal	Age Range Month (Categorical)	Live Mass (kg)	VS	B₀	MCF(%)	MS(%)
Sheep (Merino, female-male, lamb)	x < 6	25	0.32	0.13	0.015	1
Sheep (Merino, female-male, yearling)	06 < x < 12	45
Sheep (Merino, female-male, yearling)	12 < x < 24	65
Sheep (Merino, female sheep)	24 < x	80
Sheep (Merino, ram)	24 < x	100
Sheep (Native, female-male, lamb)	x < 6	20
Sheep (Native, female-male, yearling )	6 < x < 12	35
Sheep (Native, female-male, yearling )	12 < x < 24	55
Sheep (Native, female)	24 < x	70
Sheep (Native, ram )	24 < x	90
Goat (Hair goat, female-male, yearling)	x < 6	20	0.35
Goat (Native, female-male,yearling )	6 < x < 12	35
Goat (Native, female-male, yearling)	12 < x < 24	55
Goat (Native, female)	24 < x	60
Goat (Native, male)	24 < x	80

Table 4

Parameters and values used for poultry (Avcioǧlu and Türker 2012).
Animal	Live Mass (kg)	VS	B₀	MCF(%)	MS(%)
Turkey	10	0.02	0.24	0.015	1
Goose	4
Duck and Guinea fowl	2
Laying hen	2

Figure 3. was used in the calculation of biogas production from manure (Scarlat et al. 2015; Abdeshahian et al. 2016; Khan et al. 2021; Şenol et al. 2021). It also gives the standard coal, CO₂ emission (Gao et al. 2019; Khalil et al. 2019) and the estimated electricity conversion (Scarlat et al. 2015; Benito et al. 2015; Khalil et al. 2019).

If animal manure is not collected and processed in a biogas production system, CH₄ gas is naturally produced and released into the atmosphere. Agriculture and livestock production has a significant impact on the formation of greenhouse gas emissions, especially CH₄, into the atmosphere (Riaño and García-González 2015).

Different methods are used to calculate CH₄ emissions. Tier 1 is the simplest approach in which just the number of each animal type and the emissions per animal are multiplied. The more advanced approach is Tier 2, which is used in most developed countries. It is the product of several parameters per animal species. The assumed emission factors based on average annual temperature are given by the IPCC for each of the proposed livestock categories. Emission factors represent the range in manure volatile solids content and manure management application of each region and were evaluated based on the annual temperature for each climatic region. The formula in Table 5 and the emission factors of the relevant regions in table 10.11 of IPCC-2006 were used to calculate the CH₄ emission with the Tier 1 approach. The formulas in Table 5 and the parameter values in Tables 2–4 were used in the calculation with the Tier 2 approach (Dong et al. 2006; Vanderzaag et al. 2013; Baek et al. 2014; Noorollahi et al. 2015; Shin et al. 2016; Ngwabie et al. 2018; Chen et al. 2020; Herrera et al. 2021; Zubir et al. 2022; Basak et al. 2022).

Table 5. Formulas for CH₄ emission (Tier 1, Tier 2 approximations) (Dong et al. 2006; Vanderzaag et al. 2013; Baek et al. 2014; Noorollahi et al. 2015; Shin et al. 2016; Ngwabie et al. 2018; Chen et al. 2020; Herrera et al. 2021; Zubir et al. 2022; Basak et al. 2022).

2.2. Machine learning algorithms and performance evaluations

In this study, the characteristics and theoretical calculations of cattle, small ruminants, and poultry belonging to the provinces of Antalya, Isparta, and Burdur in 2018 were used. Modeling of biogas amount, CO₂ emission, coal, electricity-thermal energy, and CH₄ values was carried out by using general and specific information about animals, age, number, and manure of animals. To determine the model with the best results, machine learning algorithms SVM, MLP, and LR were used and hyper-parameter optimization was performed. The proposed algorithms have been preferred because of their popularity in the literature, their successful results in problem-solving in many fields, and their easy-to-understand structure. Matlab R2019a and PyCharm 2021.1 programs and scikit-learn 0.24 library were used for hyper-parameter optimizations and training of algorithms. Parameters of machine learning algorithms can greatly affect model success. For this reason, while determining the most suitable model for the related problem, training should be carried out with algorithms with appropriate optimum parameters. In this study, grid search was used to perform hyper-parameter optimization of machine learning algorithms. The grid search method creates a model for each determined combination of hyper-parameters and evaluates the performance so that the most optimal parameters of the relevant algorithm are determined (Pillai et al. 2019).

2.2.1. Support vector machine (SVM)

SVM is one of the most common machine learning algorithms and is used for both classification and regression problems (Smola and Schölkopf 2004). The SVM algorithm is a method based on pre-training, and it tries to create a linear or nonlinear kernel called a hyperplane to separate the classes of the data or to make a regression-based value estimation. In the literature on the SVM algorithm, kernels such as linear, polynomial, and radial basis functions (RBF) are used, and support vectors that can express the data most optimally are tried to be determined (Pisner and Schnyer 2020). v-SVM which is a different variant of SVM uses the nu parameter for controlling the number of support vectors for regression tasks.

2.2.2. Multi-Layer perceptron (MLP)

MLP is a supervised learning algorithm that learns target values by training on the data it receives as input. MLP uses the input layer, hidden layer, and output layer for classification or regression. In the feedforward neural network structure, which is preferred in this study, the cells are arranged in layers and the outputs of the cells in the layer can only be given as inputs to the next layer's overweights (Goodfellow et al. 2016). The output value of the cells is calculated with activation functions such as sigmoid, hyperbolic tangent, and rectifier linear unit (ReLU) (Cui et al. 2017). For the training of the MLP model, algorithms such as scaled conjugate gradient, Levenberg-Marquardt, and Bayesian editing methods are available from backpropagation methods. In addition to these, L-BFGS-B, SGD (Stochastic Gradient Descent) and adaptive moment estimation (Adam) methods have also achieved very good results in recent years. The scaled matched gradient method is an algorithm developed to reduce the direct search time and is based on merging reliable regions (Møller 1993). The Levenberg-Marquardt algorithm is a method based on the maximum neighborhood structure. The bayesian method updates the weight and bias values using the Levenberg Marquardt method (MacKay 1992; Dan Foresee and Hagan 1997). The SGD method provides gradient estimates using a specified number of samples from the data distribution. The Adam algorithm was created by increasing the momentum of the Rmsprop method (Goodfellow et al. 2016). L-BFGS-B is a gradient-based approach that has limited memory and is based on the trust region technique for solving large-scale optimization problems (Byrd et al. 1995).

2. 2. 3. Linear regression (LR)

LR allows the modeling of the output values by fitting a linear function to the input data. Regulation (regularization) is used in linear regression to solve the problem of multicollinearity and increase efficiency. Lasso regression (Least absolute shrinkage and selection operator), which is one of the frequently used regulation types, adds a penalty to the least-squares loss function by using the L1-norm penalty; Ridge regression, on the other hand, uses the L2-norm penalty to reduce the multicollinearity problem in linear regression that arises in models with many parameters (Tibshirani 1996). Ridge regression is seen in Eq. (1).

$$c={\left({\text{X}}^{\text{T}}\text{X}+\lambda \text{I}\right)}^{-1}{\text{X}}^{T}y$$

In Eq. (1), y is the output, X is the Vandermonde matrix, I is the identity matrix, and the ridge parameter λ ≥ 0 serves as the constant shifting of the diagonals of the moment matrix (Khalaf and Shukur 2005). One of the differences between Lasso and Ridge regression is that Lasso can discard other attributes while selecting important attributes, while ridge regression does not discard attributes completely. Elastic net, on the other hand, is another regulation technique and overcomes the limitations of the Lasso method by using a penalty function (Zou and Hastie 2005).

2. 2. 4. Grid search parameters

In this study, for SVM, one of the hyper-parameters determined for Grid search; {SVM, v-SVM} as algorithm type, {Linear, 2-3-4 degrees Polynomial, RBF} as kernel, {0.25, 0.45, 0.65, 0.85, 1.0} as ε/ν parameters and {100, 500, 1000} values were used as the number of iterations. For MLP, {L-BFGS-B, Sgd, Adam, Scg, Br, Lm} were used as the learning algorithm, {4, 8, 12, 16} as the number of hidden layer neurons, and {100, 500, 1000} as the number of iterations. For LR; {Ridge Regression, Lasso Regression, Elastic net, No regularization} methods including regularization or its variant were used. Other parameters that are not included in the grid search and are considered fixed are; Tolerance or learning rate of 0.005 was used in all algorithms with a Cost (C) value of 1 for SVM and an activation function of ReLU in MLP. The hyper-parameter values according to the models and algorithms are shown in Table 6.

Table 6

Hyper-parameter values for models and algorithms.
Model	Algorithm	Kernel Type/Neuron size	ε/ν parameter	Iteration
SVM	SVM, v-SVM	Linear, 2nd-3rd-4th degree polynomial, RBF	0.25, 0.45, 0.65, 0.85, 1.0	100, 500, 1000
MLP	L-BFGS-B, Sgd, Adam, Scg, Br, Lm	4, 8, 12, 16	-	100, 500, 1000
LR	Ridge Regression, Lasso Regression, Elastic net, No regularization	-	-	-

2 .2. 5. Performance metrics

Various metrics are used in the literature to measure the performance values of models trained with machine learning. Mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²) were used in this study to determine the best model. MSE, RMSE, MAE, and R² formulas are given in (Eqs. (2)-(5)), respectively.

$$MSE=\frac{1}{n}\sum _{i=1}^{n}{\left({Y}_{i}-{\widehat{Y}}_{i}\right)}^{2}$$

$$RMSE=\sqrt{\frac{1}{n}\sum _{i=1}^{n}{\left({Y}_{i}-{\widehat{Y}}_{i}\right)}^{2}}$$

$$MAE=\frac{1}{n}\sum _{i=1}^{n}\left|{Y}_{i}-{\widehat{Y}}_{i}\right|$$

In Eq. (2–4), Y is the target value, $\widehat{Y}$ is the predicted value, and n is the number of samples.

$${R}^{2}=1-\frac{\sum _{i=1}^{n}{\left({Y}_{i}-{\widehat{Y}}_{i}\right)}^{2}}{\sum _{i=1}^{n}{\left({Y}_{i}-{\stackrel{-}{Y}}_{i}\right)}^{2}}$$

In Eq. (5), Y represents the target value, $\widehat{Y}$ the predicted value, $\stackrel{-}{Y}$ the mean of the target value, and n the number of samples.

In this study, Leave One Out Cross Validation (LOOCV), which is used to separate training and test data, was used to obtain more statistically accurate values while measuring the performance of machine learning algorithms. Cross-validation is an indication of how well a model can predict when it is trained to make new predictions for data it has not seen before. To overcome the problem known as overfitting in model training, when training a model, the entire dataset should not be used for training, but some of it should be used as test data, which creates data that the learner has never seen. Especially in small data sets containing less than 100 samples, the LOOCV method is used (Wong 2015). In the LOOCV method, only one sample is taken as the test set, and all the remaining samples in the data set are used as the training set. This process is repeated for each of the samples in the current data set, and the error of the model is found by averaging the results (Pasini 2015).

In this study, SVM, MLP, and LR algorithms were used for the modeling of biogas, CO₂ emission, coal, electric thermal energy, and CH₄ amount. To determine the model with the best results and their parameters, hyper-parameter optimization was carried out with the grid search method. While MSE, RMSE, MAE, and R² metrics were used to compare the success of the models, the Wilcoxon Rank Sum Test was also used to compare the estimates and actual values. Table 7 shows the scores for the parameters that give the best performance as a result of the biogas amount modeling according to the grid search parameters with SVM.

Table 7

Best results with hyper-parameter optimization for SVM and MLP in biogas amount modeling.
Model	Alg.1	Alg.2 / Neuron size	ε/ν	iter	MSE	RMSE	MAE	R²
SVM	v-svm	2nd degree polynomial	0.25	1000	2.16e + 11	4.59e + 05	1.61e + 05	0.997
				500	4.33e + 11	6.63e + 05	2.12e + 05	0.994
				100	8.66e + 11	9.43e + 05	3.48e + 05	0.987
			0.45	1000	2.16e + 11	4.76e + 05	1.70e + 05	0.997
				500	4.33e + 11	6.37e + 05	1.95e + 05	0.994
				100	1.59e + 12	1.27e + 06	3.99e + 05	0.977
			0.65	1000	1.44e + 11	3.99e + 05	1.53e + 05	0.998
				500	2.89e + 11	5.44e + 05	1.95e + 05	0.996
				100	2.74e + 12	1.65e + 06	4.84e + 05	0.962
			0.85	1000	2.16e + 11	4.25e + 05	1.70e + 05	0.997
				500	3.61e + 11	6.20e + 05	2.04e + 05	0.995
				100	2.60e + 12	1.61e + 06	5.27e + 05	0.964
			1.0	1000	7.21e + 10	3.06e + 05	1.19e + 05	0.999
				500	2.89e + 11	5.35e + 05	1.78e + 05	0.996
				100	2.89e + 12	1.70e + 06	5.18e + 05	0.959
		4th degree polynomial	0.25	1000	4.33e + 11	6.63e + 05	2.12e + 05	0.994
				500	1.08e + 12	1.03e + 06	3.14e + 05	0.985
				100	4.91e + 12	2.22e + 06	5.95e + 05	0.930
			0.45	1000	7.21e + 10	3.06e + 05	1.44e + 05	0.999
				500	5.77e + 11	7.73e + 05	2.80e + 05	0.992
				100	5.77e + 11	7.64e + 05	3.23e + 05	0.992
			0.65	1000	1.44e + 11	4.16e + 05	1.87e + 05	0.998
				500	2.89e + 11	5.61e + 05	2.38e + 05	0.996
				100	7.21e + 11	8.41e + 05	3.06e + 05	0.990
			0.85	1000	7.21e + 10	3.23e + 05	1.70e + 05	0.999
				500	7.21e + 10	2.72e + 05	1.53e + 05	0.999
				100	2.31e + 12	1.53e + 06	4.76e + 05	0.967
			1.0	1000	7.21e + 10	2.72e + 05	1.61e + 05	0.999
				500	1.44e + 11	3.40e + 05	1.87e + 05	0.998
				100	5.77e + 11	7.56e + 05	3.23e + 05	0.992
MLP	L-BFGS-B	4	-	1000	2.16e + 11	4.25e + 05	1.70e + 05	0.997
				500	2.45e + 12	1.56e + 06	3.06e + 05	0.965
				100	3.64e + 13	6.03e + 06	1.11e + 06	0.484
		8		1000	4.33e + 11	6.29e + 05	1.61e + 05	0.994
				500	2.89e + 11	5.52e + 05	1.70e + 05	0.996
				100	5.77e + 11	7.81e + 05	2.63e + 05	0.991
		12		1000	2.89e + 11	5.35e + 05	1.70e + 05	0.996
				500	2.89e + 12	1.70e + 06	3.57e + 05	0.959
				100	1.44e + 11	3.65e + 05	2.04e + 05	0.998
		16		1000	2.89e + 11	5.61e + 05	1.87e + 05	0.995
				500	3.32e + 12	1.83e + 06	3.48e + 05	0.953
				100	4.33e + 11	6.37e + 05	2.38e + 05	0.994

When Table 7 is examined, it is seen that the SVM method gives the best performance values with the v-svm algorithm, and the R² value was obtained as 0.999 in 4th and 2nd degrees polynomial. In the MLP model, the L-BFGS-B method as a learning algorithm was able to reach higher R² values and lower MSE, RMSE, and MAE values compared to other learning methods. The best scores in the MLP model were determined with the parameters that the R² value was obtained as 0.997 after 1000 iterations of training using 4 hidden neurons with the L-BFGS-B method, and the R² value was obtained as 0.998 after 100 iterations of training using 12 hidden neurons. Table 8 shows the performance scores of the modeling of the biogas amount according to the grid search parameters with LR.

Table 8

Hyper-parameter optimization for LR in biogas amount.
Model	Alg.1	Alg.2	ε/ν	iter	MSE	RMSE	MAE	R²
LR	Ridge Regression	-	-	-	3.61e + 11	5.78e + 05	2.72e + 05	0.995
	Lasso Regression				4.33e + 11	6.71e + 05	2.55e + 05	0.994
	Elastic net				5.05e + 11	6.97e + 05	2.80e + 05	0.993
	No regularization				3.61e + 11	5.78e + 05	2.72e + 05	0.995

It is seen that in Table 8, the LR model generally gives good results for different parameter combinations. The highest R² values of 0.995 were obtained for Ridge Regression and No regularization.

When Table 7 and Table 8 are examined, considering the best values of the algorithms, two-parameter combinations from each algorithm were determined and used in other models. While determining the parameter combinations for MLP, two parameters with the same R² values, one with the lowest RMSE value and the other with the lowest MAE value, were determined. The algorithms giving the best results and their parameters are shown in Table 9.

Table 9

Algorithms with optimal parameters.
Algorithm	Method	Alg.1	Alg.2/Neuron Size	ε/ν	Iter
SVM-v1	SVM	v-svm	2^nd -degree polynomial	1.0	1000
SVM-v2	SVM	v-svm	4^th -degree polynomial	0.85	500
MLP-v1	MLP	L-BFGS-B	4 hidden neuron	-	1000
MLP-v2	MLP	L-BFGS-B	12 hidden neuron	-	100
LR-v1	LR	Ridge Regression	-	-	-
LR-v2	LR	No regularization	-	-	-

Algorithms with the most suitable parameters for modeling biogas amount and statistical test results are shown in Table 10. Wilcoxon Rank Sum Test results have been indicated with h and its value has been expressed with p. Acceptance of the null test in the 5% confidence interval has been indicated with a “+” sign and rejection with a “-” sign.

Table 10

Modeling of biogas amount with algorithms with optimal parameters.
Algorithm	MSE	RMSE	MAE	R²	Wilcoxon Rank
Algorithm	MSE	RMSE	MAE	R²	h	p
SVM-v1	7.21e + 10	3.06e + 05	1.19e + 05	0.999	+	0.850
SVM-v2	7.21e + 10	2.72e + 05	1.53e + 05	0.999	+	0.892
MLP-v1	2.16e + 11	4.25e + 05	1.70e + 05	0.997	+	0.910
MLP-v2	1.44e + 11	3.65e + 05	2.04e + 05	0.998	+	0.791
LR-v1	3.61e + 11	5.78e + 05	2.72e + 05	0.995	+	0.910
LR-v2	3.61e + 11	5.78e + 05	2.72e + 05	0.995	+	0.910

The graphic of biogas modeling is seen in Fig. 4. The thick line in Fig. 4 shows the theoretical values obtained from the formulation in Fig. 3. Thin lines show the modeling results. The biogas unit on the vertical axis is m³/year.

Algorithms with the most suitable parameters determined after hyper-parameter optimization were used in modeling CO₂ emissions, coal, electricity, thermal energy, and CH₄ emissions. Algorithms with the most suitable parameters for CO₂ emission modeling and statistical test results are shown in Table 11.

Table 11

Modeling of CO₂ emissions with algorithms with the most suitable parameters.
Algorithm	MSE	RMSE	MAE	R²	Wilcoxon Rank
Algorithm	MSE	RMSE	MAE	R²	h	p
SVM-v1	9.94e + 04	3.59e + 02	1.40e + 02	0.999	+	0.850
SVM-v2	9.94e + 04	3.19e + 02	1.79e + 02	0.999	+	0.892
MLP-v1	3.28e + 06	1.79e + 03	4.19e + 02	0.967	+	0.904
MLP-v2	2.98e + 05	5.68e + 02	2.99e + 02	0.997	+	0.844
LR-v1	4.97e + 05	6.78e + 02	3.19e + 02	0.995	+	0.910
LR-v2	4.97e + 05	6.78e + 02	3.19e + 02	0.995	+	0.910

As seen in Table 11, it is seen that the model and parameter values that give good results for biogas modeling can also successfully perform CO₂ emission modeling. This shows that the hyper-parameter optimization has been carried out successfully and the parameters for the related problem can be optimized. The graph of the CO₂ emission modeling is seen in Fig. 5. The thick line shows the theoretical values obtained from the formulation in Fig. 3. Thin lines show the modeling results. The CO₂ emission on the vertical axis is ton /year.

Algorithms with the most suitable parameters for coal amount modeling and statistical test results are shown in Table 12.

Table 12

Coal amount modeling with algorithms with optimal parameters.
Algorithm	MSE	RMSE	MAE	R²	Wilcoxon Rank
Algorithm	MSE	RMSE	MAE	R²	h	p
SVM-v1	1.47e + 05	3.93e + 02	1.94e + 02	0.996	+	0.774
SVM-v2	2.20e + 05	4.90e + 02	2.24e + 02	0.993	+	0.768
MLP-v1	4.40e + 05	6.72e + 02	1.94e + 02	0.987	+	0.994
MLP-v2	2.57e + 05	5.08e + 02	2.36e + 02	0.993	+	0.995
LR-v1	1.10e + 05	3.45e + 02	2.18e + 02	0.997	+	1.0
LR-v2	1.10e + 05	3.45e + 02	2.18e + 02	0.997	+	1.0

In the modeling of coal amount, it is seen that the LR method gives the highest R² values with 0.997. In addition, as a result of the Wilcoxon Rank Sum Test, it is seen that the LR method obtained the highest statistical value with p = 1.0. While the SVM method came to the fore in biogas amount and CO₂ emission modeling, the LR method gave better results in coal amount modeling. The graph of the coal amount modeling is seen in Fig. 6. The thick line shows the theoretical values obtained from the formulation in Fig. 3. Thin lines show the modeling results. The coal amount on the vertical axis is ton /year.

Algorithms with the most suitable parameters for electricity generation modeling and statistical test results are shown in Table 13.

Table 13

Electricity generation modeling with algorithms with optimal parameters.
Algorithm	MSE	RMSE	MAE	R²	Wilcoxon Rank
Algorithm	MSE	RMSE	MAE	R²	h	p
SVM-v1	2.34e + 11	5.50e + 05	2.29e + 05	0.999	+	0.856
SVM-v2	2.34e + 11	4.89e + 05	2.75e + 05	0.999	+	0.892
MLP-v1	3.27e + 12	1.83e + 06	5.50e + 05	0.985	+	0.982
MLP-v2	1.87e + 12	1.35e + 06	4.59e + 05	0.992	+	0.797
LR-v1	1.17e + 12	1.04e + 06	4.89e + 05	0.995	+	0.910
LR-v2	1.17e + 12	1.04e + 06	4.89e + 05	0.995	+	0.910

It is seen in Table 13 that SVM-v1 and SVM-v2 models give the highest R² value of 0.999 for electricity generation modeling. It can be seen from the h values in Table 8 that the statistical test results of the models in which the SVM method is used are also accepted. The graph of the electricity generation modeling is seen in Fig. 7. The thick line shows the theoretical values obtained from the formulation in Fig. 3. Thin lines show the modeling results. The electricity generation on the vertical axis is kWh /year.

Algorithms with the most suitable parameters for thermal energy modeling and statistical test results are shown in Table 14.

Table 14

Thermal energy modeling with algorithms with optimal parameters.
Algorithm	MSE	RMSE	MAE	R²	Wilcoxon Rank
Algorithm	MSE	RMSE	MAE	R²	h	p
SVM-v1	3.37e + 13	6.60e + 06	2.75e + 06	0.999	+	0.856
SVM-v2	3.37e + 13	5.87e + 06	3.30e + 06	0.999	+	0.892
MLP-v1	1.01e + 14	1.01e + 07	3.67e + 06	0.997	+	0.856
MLP-v2	1.68e + 14	1.27e + 07	5.69e + 06	0.995	+	0.751
LR-v1	1.68e + 14	1.25e + 07	5.87e + 06	0.995	+	0.910
LR-v2	1.68e + 14	1.25e + 07	5.87e + 06	0.995	+	0.910

It is seen in Table 14 that SVM-v1 and SVM-v2 models give the highest R² value of 0.999 for thermal energy modeling. The graphic of the thermal energy modeling is seen in Fig. 8. The thick line shows the theoretical values obtained from the formulation in Fig. 3. Thin lines show the modeling results. The thermal energy on the vertical axis is MJ/year.

Algorithms with the most suitable parameters for methane emission (Tier 1 approximation) modeling and statistical test results are shown in Table 15.

Table 15

CH₄ emission (Tier 1) modeling with algorithms with optimal parameters.
Algorithm	MSE	RMSE	MAE	R²	Wilcoxon Rank
Algorithm	MSE	RMSE	MAE	R²	h	p
SVM-v1	1.10e + 07	3.32e + 03	7.46e + 02	-1.855	+	0.660
SVM-v2	5.52e + 07	7.43e + 03	1.33e + 03	-13.254	+	0.540
MLP-v1	9.11e + 04	3.01e + 02	1.57e + 02	0.977	+	0.789
MLP-v2	2.65e + 05	5.16e + 02	2.51e + 02	0.932	+	0.743
LR-v1	1.90e + 05	4.38e + 02	2.77e + 02	0.950	+	0.647
LR-v2	1.94e + 05	4.40e + 02	2.79e + 02	0.950	+	0.647

It is seen that the SVM method obtains high MSE, RMSE, and MAE values and is not suitable for methane (Tier 1 approximation) modeling. The MLP-v1 algorithm gave the best modeling result with an R² value of 0.977. The graph of the methane emission modeling is shown in Fig. 9. The thick line shows the theoretical values obtained from the formulation in Table 5. Thin lines show the modeling results. The CH₄ emission on the vertical axis is ton CH₄/year.

Algorithms with the most suitable parameters for CH₄ emission (Tier 2 approximation) modeling and statistical test results are shown in Table 16.

Table 16

CH₄ emission (Tier 2) modeling with algorithms with optimal parameters.
Algorithm	MSE	RMSE	MAE	R²	Wilcoxon Rank
Algorithm	MSE	RMSE	MAE	R²	h	p
SVM-v1	2.25e + 04	1.50e + 02	6.46e + 01	0.888	+	0.352
SVM-v2	1.23e + 07	3.51e + 03	6.08e + 02	-59.760	-	0.022
MLP-v1	3.82e + 04	1.95e + 02	6.77e + 01	0.811	+	0.300
MLP-v2	1.14e + 04	1.07e + 02	5.23e + 01	0.944	+	0.473
LR-v1	7.65e + 03	8.73e + 01	5.59e + 01	0.962	+	0.530
LR-v2	7.65e + 03	8.73e + 01	5.59e + 01	0.962	+	0.530

It is seen that the SVM-v2 model achieved high MSE, RMSE, and MAE values and low R² values. In addition, the statistical test result was rejected and it is seen that it is not suitable for the CH₄ emission modeling obtained using the Tier 2 approach. The best modeling result was obtained by LR-v1 and LR-v2 algorithms with R² values of 0.962. The graph of the CH₄ emission modeling is seen in Fig. 10. The thick line shows the theoretical values obtained from the formulation in Table 5. Thin lines show the modeling results. CH₄ emission on the vertical axis is ton CH₄/year.

Turkey is a developing country whose energy needs are based on fossil fuels, which are the primary source, and imports most of its energy needs from abroad. In Turkey, facilities for the use of biomass resources in energy production are increasing and new conversion facilities are commissioned every year to provide environmentally friendly and clean energy production. Therefore, reliable energy potential estimates are needed. Agriculture and animal husbandry are carried out in every region of Turkey. The geographical location of the country is effective in the abundance of agriculture and animal husbandry, and animal husbandry is among the important sources of income, especially in rural areas. Livestock support packages consisting of many parameters are provided for animal enterprises. Some of these packages can be specified as feed support, milk support, and animal shelter support. Some of the most important problems of animal husbandry in Turkey are known animal breeding, animal health, and epidemic diseases. Burdur Mehmet Akif Ersoy University has been selected by the Turkish Higher Education Institution as a research university focused on regional development in the field of animal husbandry. The University provides technical support and training to the farmers engaged in animal husbandry activities in the region on many issues such as animal breeding, care, and feeding, to increase domestic animal production, revitalize the economy, and make modern animal husbandry widespread.

This study focuses on biogas production from animal manure and its conversion to other energy sources to provide an efficient and reliable estimation of resource availability. The biogas potential and CO₂ emission, coal, electrical energy, thermal energy, and CH₄ emission that may occur according to cattle, small ruminant, and poultry data for 2018 in different categories and age groups of the provinces of Antalya, Isparta, and Burdur, which are called the Western Mediterranean and Lakes Region, predicted by machine learning algorithms. According to the data of Turkstat (2018), there are a total of 3883307 animals belonging to different animal species (cattle, small ruminants, poultry) in the provinces of Antalya, Isparta, and Burdur. A total of 550305 tons/year of animal manure can be obtained from this animal's existence. As a result of the theoretical calculation, 130929541 m³/year of biogas can be produced from animal manure. Depending on biogas production; 93484 tons/year of coal, 235673174 kWh/year of electrical energy, 2828078093 MJ/year of thermal energy can be provided. In addition, 153653 tons/year of CO₂ emissions will be prevented.

SVM, MLP, and LR methods from machine learning algorithms were used for estimation. The reasons for the preference of these algorithms are that they are popular in the literature, can be applied to problem solutions belonging to different fields, and are easy to understand.

As a result of hyper-parameter optimization with grid search, the six most successful models were identified and named as SVM-v1, SVM-v2, MLP-v1, MLP-v2, and LR-v1, LR-v2. Although all models generally produce good results in the estimation of biogas potential, SVM-v1 and SVM-v2 are seen as the best model with R² = 0.999.

For the estimation of CO₂ emission, it can be seen that SVM models produce the best results with R² = 0.999 values. When the coal amount estimation is examined, it is seen that the LR-v1 and LR-v2 models produce slightly better results than SVM and MLP with R² = 0.997 values. In terms of electricity production and thermal energy estimations, it is seen that SVM models produce the best results with R² = 0.999 values. In the estimation of methane using the Tier 1 approach, SVM models achieve the worst result compared to other models, while the MLP-v1 model can perform the best estimation with R² = 0.977. In the methane modeling obtained using the Tier 2 approach, the LR-v1 and LR-v2 models were superior to the other models, and these models reached the performance value of R² = 0.962.

It is seen that machine learning algorithms are successful in modeling biogas, energy, and emission transformations. It is seen that the SVM-v1 model is more successful than other models in energy and emission conversions, with performances in the range of R² = 0.996–0.999. In the estimation of methane emissions, MLP-v1 and LR models were the models with the highest performance values.

The study carried out covers only some provinces in the lakes region and is seen as a limitation of the study. Future studies, it is aimed to create more general machine learning-based models that will cover all provinces of Turkey.

It is seen that there is no study in the literature on the biogas potential calculated according to different animal categories and age-related weight values in Turkey and the estimation of the energy and emission conversions of this potential. In this study, coal, electrical energy, CO₂ emissions, thermal energy, and CH₄ conversions related to biogas were examined for the first time with machine learning algorithms and 6 different prediction models were created and compared. Due to these features, the study carried out in a wide scope is very important for Turkey. Considering the use of biogas potential in the coming years, the share of renewable energy in total energy consumption in Turkey can be increased. This, in turn, may enable the country to reduce its energy imports. The models used in the study can be applied to different regions of Turkey for different or the same biomass sources.

Conflict of interest The authors declare no confict of interest.

No funds, grants, or other support was received.

Ethical Approval

Not applicable

Consent to Participate

Not applicable

Consent to Publish

Not applicable

Availability of data and materials

Not applicable

Author Contributions Statement

Ihsan Pence: Machine Learning Methodology, Software, Investigation.

Kazım Kumaş: Invesitagation, Data curation, Conceptualization.

Melike Sisesi Cesmeli: Machine Learning Methodology, Visualization.

Ali Akyüz: Conceptualization, Writing- Reviewing and Editing.

Abdeshahian P, Lim JS, Ho WS, et al (2016) Potential of biogas production from farm animal waste in Malaysia. Renew Sustain Energy Rev 60:714–723. https://doi.org/10.1016/j.rser.2016.01.117
Afazeli H, Jafari A, Rafiee S, Nosrati M (2014) An investigation of biogas production potential from livestock and slaughterhouse wastes. Renew Sustain Energy Rev 34:380–386. https://doi.org/10.1016/J.RSER.2014.03.016
Alatzas S, Moustakas K, Malamis D, Vakalis S (2019) Biomass potential from agricultural waste for energetic utilization in Greece. Energies 12:1095. https://doi.org/10.3390/en12061095
Aravani VP, Sun H, Yang Z, et al (2022) Agricultural and livestock sector’s residues in Greece & China: Comparative qualitative and quantitative characterization for assessing their potential for biogas production. Renew Sustain Energy Rev 154:111821. https://doi.org/10.1016/J.RSER.2021.111821
Avcioǧlu AO, Türker U (2012) Status and potential of biogas energy from animal wastes in Turkey. Renew Sustain Energy Rev 16:1557–1561. https://doi.org/10.1016/J.RSER.2011.11.006
Avcıoğlu AO, Dayıoğlu MA, Türker U (2019) Assessment of the energy potential of agricultural biomass residues in Turkey. Renew Energy 138:610–619. https://doi.org/10.1016/j.renene.2019.01.053
Baek CY, Lee KM, Park KH (2014) Quantification and control of the greenhouse gas emissions from a dairy cow system. J Clean Prod 70:50–60. https://doi.org/10.1016/j.jclepro.2014.02.010
Bakay MS, Ağbulut Ü (2021) Electricity production based forecasting of greenhouse gas emissions in Turkey with deep learning, support vector machine and artificial neural network algorithms. J Clean Prod 285:125324. https://doi.org/10.1016/j.jclepro.2020.125324
Bao W, Yang Y, Fu T, Xie GH (2019) Estimation of livestock excrement and its biogas production potential in China. J Clean Prod 229:1158–1166. https://doi.org/10.1016/j.jclepro.2019.05.059
Basak JK, Arulmozhi E, Moon BE, et al (2022) Modelling methane emissions from pig manure using statistical and machine learning methods. Air Qual Atmos Heal 1–15. https://doi.org/10.1007/s11869-022-01169-0
Beltramo T, Klocke M, Hitzmann B (2019) Prediction of the biogas production using GA and ACO input features selection method for ANN model. Inf Process Agric 6:349–356. https://doi.org/10.1016/j.inpa.2019.01.002
Benito M, Ortiz I, Rodriguez L, Munoz G (2015) NieCo bimetallic catalyst for hydrogen production in sewage treatment plants: Biogas reforming and tars removal. Int J Hydrogen Energy 40:14456–14468. https://doi.org/10.1016/j.ijhydene.2015.06.163
Byrd RH, Lu P, Nocedal J, Zhu C (1995) A Limited Memory Algorithm for Bound Constrained Optimization. SIAM J Sci Comput 16:1190–1208. https://doi.org/10.1137/0916069
Ceylan AB, Aydın L, Nil M, et al (2021) A new hybrid approach in selection of optimum establishment location of the biogas energy production plant. Biomass Convers Biorefinery 1–16. https://doi.org/10.1007/s13399-021-01532-8
Chen Z, An C, Fang H, et al (2020) Assessment of regional greenhouse gas emission from beef cattle production: A case study of Saskatchewan in Canada. J Environ Manage 264:110443. https://doi.org/10.1016/j.jenvman.2020.110443
Cheng D, Liu Y, Shehata E, et al (2021) In-feed antibiotic use changed the behaviors of oxytetracycline, sulfamerazine, and ciprofloxacin and related antibiotic resistance genes during swine manure composting. J Hazard Mater 402:123710. https://doi.org/10.1016/j.jhazmat.2020.123710
Chowdhury T, Chowdhury H, Hossain N, et al (2020) Latest advancements on livestock waste management and biogas production: Bangladesh’s perspective. J Clean Prod 272:122818. https://doi.org/10.1016/J.JCLEPRO.2020.122818
Cui J Le, Qiu S, Jiang MY, et al (2017) Text classification based on ReLU activation function of SAE algorithm. In: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Springer, Cham, pp 44–50
Dan Foresee F, Hagan MT (1997) Gauss-Newton approximation to bayesian learning. In: IEEE International Conference on Neural Networks - Conference Proceedings. pp 1930–1935
De Clercq D, Jalota D, Shang R, et al (2019) Machine learning powered software for accurate prediction of biogas production: A case study on industrial-scale Chinese production data. J Clean Prod 218:390–399. https://doi.org/10.1016/j.jclepro.2019.01.031
Dong H, Mangino J, McAllister TA, et al (2006) Emissions from livestock and manure management. In: Guidelines for national greenhouse gas inventories. Intergovernmental Panel on Climate Change (IPCC 2006)
Elmaz F, Yücel Ö, Mutlu AY (2020) Predictive modeling of biomass gasification with machine learning-based regression methods. Energy 191:116541. https://doi.org/10.1016/j.energy.2019.116541
EPDK (2022) Republic of Turkey Energy Market Regulatory Authority. https://www.epdk.gov.tr/Detay/Icerik/3-0-0-122/yenilenebilir-enerji-kaynaklari-destekleme-mekanizmasi-yekdem. Accessed 18 Feb 2022
Erat S, Telli A, Ozkendir OM, Demir B (2021) Turkey’s energy transition from fossil-based to renewable up to 2030: milestones, challenges and opportunities. Clean Technol Environ Policy 23:401–412. https://doi.org/10.1007/s10098-020-01949-1
Erdin C, Ozkaya G (2019) Turkey’s 2023 energy strategies and investment opportunities for renewable energy sources: Site selection based on ELECTRE. Sustain 11:2136. https://doi.org/10.3390/su11072136
Font-Palma C (2019) Methods for the Treatment of Cattle Manure—A Review. C 5:27. https://doi.org/10.3390/c5020027
Gao M, Wang D, Wang H, et al (2019) Biogas potential, utilization and countermeasures in agricultural provinces: A case study of biogas development in Henan Province, China. Renew Sustain Energy Rev 99:191–200. https://doi.org/10.1016/J.RSER.2018.10.005
Goodfellow I, Bengio Y, Courville A (2016) Deep Learning. MIT Press
Gündoğan B, Koçar G (2022) Potential Usability of Cynara cardunculus L. Residues in Biogas Production in Various Regions of Turkey. Bioenergy Res 1:. https://doi.org/10.1007/s12155-021-10375-0
Herrera AMN, Esteves EMM, Morgado CRV, Esteves VPP (2021) Carbon footprint analysis of bioenergy production from cattle manure in the Brazilian Central-West. Bioenergy Res 14:1265–1276. https://doi.org/10.1007/s12155-020-10216-6
IEA (2022) International Energy Agency. https://www.iea.org/reports/turkey-2021. Accessed 25 Jan 2022
Ilbas M, Antari LOA, Sahin M (2022) Biogas production from domestic resources as an alternative energy source: a comprehensive feasibility study. Int J Energy a Clean Environ 23:97–107. https://doi.org/10.1615/interjenercleanenv.2021037381
IOPRT (2021) Investment Office of the Presidency of the Republic of Turkey. https://www.invest.gov.tr/en/pages/home-page.aspx. Accessed 20 Sep 2021
Jeong K, Abbas A, Shin J, et al (2021) Prediction of biogas production in anaerobic co-digestion of organic wastes using deep learning models. Water Res 205:117697. https://doi.org/10.1016/j.watres.2021.117697
Karaaslan A, Gezen M (2022) The evaluation of renewable energy resources in Turkey by integer multi-objective selection problem with interval coefficient. Renew Energy 182:842–854. https://doi.org/10.1016/j.renene.2021.10.053
Khalaf G, Shukur G (2005) Choosing ridge parameter for regression problems. Commun Stat - Theory Methods 34:1177–1182. https://doi.org/10.1081/STA-200056836
Khalil M, Berawi MA, Heryanto R, Rizalie A (2019) Waste to energy technology: The potential of sustainable biogas production from animal waste in Indonesia. Renew Sustain Energy Rev 105:323–331. https://doi.org/10.1016/j.rser.2019.02.011
Khan MU, Ahmad M, Sultan M, et al (2021) Biogas production potential from livestock manure in Pakistan. Sustain 13:6751. https://doi.org/10.3390/su13126751
Khoshgoftar Manesh MH, Rezazadeh A, Kabiri S (2020) A feasibility study on the potential, economic, and environmental advantages of biogas production from poultry manure in Iran. Renew Energy 159:87–106. https://doi.org/10.1016/j.renene.2020.05.173
Le T-H, Chang Y, Park D (2020) Renewable and nonrenewable energy consumption, economic growth, and emissions: International evidence. Energy J 41:73–92. https://doi.org/10.5547/01956574.41.2.THLE
MacKay DJC (1992) Bayesian Interpolation. Neural Comput 4:415–447. https://doi.org/10.1162/neco.1992.4.3.415
Melikoglu M (2017) Vision 2023: Status quo and future of biomass and coal for sustainable energy generation in Turkey. Renew Sustain Energy Rev 74:800–808. https://doi.org/10.1016/j.rser.2017.03.005
Melikoglu M, Menekse ZK (2020) Forecasting Turkey’s cattle and sheep manure based biomethane potentials till 2026. Biomass and Bioenergy 132:105440. https://doi.org/10.1016/j.biombioe.2019.105440
MI&T (2022) Ministry of Industry and Technology. https://www.yatirimadestek.gov.tr/haber/enerji-ve-tabii-kaynaklar-bakanligi. Accessed 15 Feb 2022
Møller MF (1993) A scaled conjugate gradient algorithm for fast supervised learning. Neural Networks 6:525–533. https://doi.org/10.1016/S0893-6080(05)80056-5
Najafi B, Faizollahzadeh Ardabili S (2018) Application of ANFIS, ANN, and logistic methods in estimating biogas production from spent mushroom compost (SMC). Resour Conserv Recycl 133:169–178. https://doi.org/10.1016/j.resconrec.2018.02.025
Ngwabie NM, Chungong BN, Yengong FL (2018) Characterisation of pig manure for methane emission modelling in Sub-Saharan Africa. Biosyst Eng 170:31–38. https://doi.org/10.1016/j.biosystemseng.2018.03.009
Noorollahi Y, Kheirrouz M, Farabi-Asl H, et al (2015) Biogas production potential from livestock manure in Iran. Renew Sustain Energy Rev 50:748–754. https://doi.org/10.1016/j.rser.2015.04.190
Ocak S, Acar S (2021) Biofuels from wastes in Marmara Region, Turkey: potentials and constraints. Environ Sci Pollut Res 28:66026–66042. https://doi.org/10.1007/s11356-021-15464-3
Oliveira ACL de, Milagres RS, Orlando Junior W de A, Renato N dos S (2020) Evaluation of Brazilian potential for generating electricity through animal manure and sewage. Biomass and Bioenergy 139:105654. https://doi.org/10.1016/j.biombioe.2020.105654
Oliveira ACL de, Renato N dos S, Martins MA, et al (2021) Modeling for estimating and optimizing the energy potential of animal manure and sewage in small and medium-sized farms. J Clean Prod 319:128562. https://doi.org/10.1016/j.jclepro.2021.128562
Pasini A (2015) Artificial neural networks for small dataset analysis. J Thorac Dis 7:953–960. https://doi.org/10.3978/j.issn.2072-1439.2015.04.61
Pillai N, Schwartz SL, Ho T, et al (2019) Estimating parameters of nonlinear dynamic systems in pharmacology using chaos synchronization and grid search. J Pharmacokinet Pharmacodyn 46:193–210. https://doi.org/10.1007/s10928-019-09629-4
Pisner DA, Schnyer DM (2020) Support vector machine. In: Mechelli A, Vieira S (eds) Machine Learning: Methods and Applications to Brain Disorders. Academic Press, Elsevier, pp 101–121
Pramanik SK, Suja FB, Zain SM, Pramanik BK (2019) The anaerobic digestion process of biogas production from food waste: Prospects and constraints. Bioresour Technol Reports 8:100310. https://doi.org/10.1016/j.biteb.2019.100310
Ramírez-Islas ME, Güereca LP, Sosa-Rodriguez FS, Cobos-Peralta MA (2020) Environmental assessment of energy production from anaerobic digestion of pig manure at medium-scale using life cycle assessment. Waste Manag 102:85–96. https://doi.org/10.1016/j.wasman.2019.10.012
Razmjoo A, Gakenia Kaigutha L, Vaziri Rad MA, et al (2021) A Technical analysis investigating energy sustainability utilizing reliable renewable energy sources to reduce CO2 emissions in a high potential area. Renew Energy 164:46–57. https://doi.org/10.1016/J.RENENE.2020.09.042
Riaño B, García-González MC (2015) Greenhouse gas emissions of an on-farm swine manure treatment plant - Comparison with conventional storage in anaerobic tanks. J Clean Prod 103:542–548. https://doi.org/10.1016/j.jclepro.2014.07.007
Rincon L, Puri M, Kojakovic A, Maltsoglou I (2019) The contribution of sustainable bioenergy to renewable electricity generation in Turkey: Evidence based policy from an integrated energy and agriculture approach. Energy Policy 130:69–88. https://doi.org/10.1016/j.enpol.2019.03.024
RTMAF (2022) Republic of Turkey Ministry of Agriculture And Forestry. https://hbs.tarbil.gov.tr/. Accessed 6 Feb 2022
RTMAF (2018) Burdur Directorate of Provincial Agriculture and Forestry. https://burdur.tarimorman.gov.tr/. Accessed 20 Nov 2018
RTME&NR (2022) Republic of Turkey Ministry of Energy and Natural Resources. https://www.enerji.gov.tr/. Accessed 20 Jan 2022
Safieddin Ardebili SM (2020) Green electricity generation potential from biogas produced by anaerobic digestion of farm animal waste and agriculture residues in Iran. Renew Energy 154:29–37. https://doi.org/10.1016/j.renene.2020.02.102
Scarlat N, Dallemand JF, Monforti-Ferrario F, et al (2015) Renewable energy policy framework and bioenergy contribution in the European Union - An overview from National Renewable Energy Action Plans and Progress Reports. Renew Sustain Energy Rev 51:969–985. https://doi.org/10.1016/j.rser.2015.06.062
Senocak AA, Guner Goren H (2022) Forecasting the biomass-based energy potential using artificial intelligence and geographic information systems: A case study. Eng Sci Technol an Int J 26:100992. https://doi.org/10.1016/J.JESTCH.2021.04.011
Şenol H, Ali Dereli̇ M, Özbilgin F (2021) Investigation of the distribution of bovine manure-based biomethane potential using an artificial neural network in Turkey to 2030. Renew Sustain Energy Rev 149:111338. https://doi.org/10.1016/J.RSER.2021.111338
Seo KW, Seo J, Kim K, et al (2021) Prediction of biogas production rate from dry anaerobic digestion of food waste: Process-based approach vs. recurrent neural network black-box model. Bioresour Technol 341:125829. https://doi.org/10.1016/j.biortech.2021.125829
Shin J, Hong SG, Kim SC, et al (2016) Estimation of potential methane production through the mass balance equations from agricultural biomass in Korea. Appl Biol Chem 59:765–773. https://doi.org/10.1007/s13765-016-0224-1
Siddiki SYA, Uddin MN, Mofijur M, et al (2021) Theoretical calculation of biogas production and greenhouse gas emission reduction potential of livestock, poultry and slaughterhouse waste in Bangladesh. J Environ Chem Eng 9:105204. https://doi.org/10.1016/j.jece.2021.105204
Smola AJ, Schölkopf B (2004) A tutorial on support vector regression. Stat Comput 14:199–222. https://doi.org/10.1023/B:STCO.0000035301.49549.88
Stolarski MJ, Warmiński K, Krzyżaniak M, et al (2020) Bioenergy technologies and biomass potential vary in Northern European countries. Renew Sustain Energy Rev 133:110238. https://doi.org/10.1016/j.rser.2020.110238
TETC (2021) Turkish Electricity Transmission Company, Electricity production-consumption statistics for Turkey. https://www.teias.gov.tr/tr-TR/turkiye-elektrik-uretim-iletim-istatistikleri. Accessed 28 Feb 2021
Tibshirani R (1996) Regression Shrinkage and Selection Via the Lasso. J R Stat Soc Ser B 58:267–288. https://doi.org/10.1111/j.2517-6161.1996.tb02080.x
TUIK (2022) Turkish Statistical Institute. https://data.tuik.gov.tr/Kategori/GetKategori?p=Nufus-ve-Demografi-109. Accessed 5 Jan 2022
TÜSİAD (2022) Turkish Industry and Business Association. https://tusiad.org/tr/yayinlar/raporlar/item/10709-turkiye-de-yazilim-ekosisteminin-gelecegi. Accessed 15 Feb 2022
Ulusoy Y, Ulukardesler AH, Arslan R, Tekin Y (2021) Energy and emission benefits of chicken manure biogas production: a case study. Environ Sci Pollut Res 28:12351–12356. https://doi.org/10.1007/s11356-018-3466-0
Vanderzaag AC, MacDonald JD, Evans L, et al (2013) Towards an inventory of methane emissions from manure management that is responsive to changes on Canadian farms. Environ Res Lett 8:035008. https://doi.org/10.1088/1748-9326/8/3/035008
WMDA (2022) West Mediterranean Development Agency. https://baka.ka.gov.tr/dokuman-merkezi/dokumanlar/bolge-plani/tr61-duzey-2-bolgesi-2014-2023-bolge-plani. Accessed 10 Jan 2022
Wong TT (2015) Performance evaluation of classification algorithms by k-fold and leave-one-out cross validation. Pattern Recognit 48:2839–2846. https://doi.org/10.1016/j.patcog.2015.03.009
World Bank (2021) The World Bank in Turkey. https://www.worldbank.org/en/country/turkey/overview#1. Accessed 15 Nov 2021
Yurtkuran S (2021) The effect of agriculture, renewable energy production, and globalization on CO2 emissions in Turkey: A bootstrap ARDL approach. Renew Energy 171:1236–1245. https://doi.org/10.1016/j.renene.2021.03.009
Zabed HM, Akter S, Yun J, et al (2020) Biogas from microalgae: Technologies, challenges and opportunities. Renew Sustain Energy Rev 117:109503. https://doi.org/10.1016/j.rser.2019.109503
Zamri MFMA, Hasmady S, Akhiar A, et al (2021) A comprehensive review on anaerobic digestion of organic fraction of municipal solid waste. Renew Sustain Energy Rev 137:110637. https://doi.org/10.1016/J.RSER.2020.110637
Zou H, Hastie T (2005) Regularization and variable selection via the elastic net. J R Stat Soc Ser B Stat Methodol 67:301–320. https://doi.org/10.1111/j.1467-9868.2005.00503.x
Zubir MA, Bong CPC, Ishak SA, et al (2022) The trends and projections of greenhouse gas emission by the livestock sector in Malaysia. Clean Technol Environ Policy 24:363–377. https://doi.org/10.1007/s10098-021-02156-2

Download PDF

Editorial decision: Major Revision
17 Aug, 2022
Reviewers agreed at journal
25 Jun, 2022
Reviewers invited by journal
17 Jun, 2022
Editor assigned by journal
23 May, 2022
First submitted to journal
11 May, 2022

You are reading this latest preprint version

Detailed analysis of animal manure-based CO₂ emissions, coal, electricity, thermal energy, and CH₄ emissions and using machine learning as a forecasting method: a study from Turkey

Status:

Version 1

Abstract

Figures

1. Introduction

1.1. Renewable energy and biomass from animal manure in Turkey

1.2. Livestock production in the Western Mediterranean Region, Turkey

1.3. Related studies in the literature

2. Material And Method

2.1. Theoretical determination of animal manure, biogas, and energy potential

2.2. Machine learning algorithms and performance evaluations

2.2.1. Support vector machine (SVM)

2.2.2. Multi-Layer perceptron (MLP)

2. 2. 3. Linear regression (LR)

2. 2. 4. Grid search parameters

2 .2. 5. Performance metrics

3. Results And Discussion

4. Conclusion

Declaration

References

Status:

Version 1