The modeling and optimization of energy inputs and greenhouse gas emissions in watermelon production using artificial neural network and multi objective genetic algorithm

Yelmen, B., Çakır, M. T., and Çakır, M. F. (2026). "The modeling and optimization of energy inputs and greenhouse gas emissions in watermelon production using artificial neural network and multi objective genetic algorithm," BioResources 21(3), 6498–6517.

Abstract

This study modeled and optimized energy consumption and greenhouse gas emissions (GHGE) for watermelon (Citrullus lanatus L.) production in Adana, Turkey. Artificial Neural Networks (ANN) and Multi-Objective Genetic Algorithms (MOGA) were employed for the analysis. The findings revealed that chemical fertilizers accounted for the largest share of energy use (77.0%), followed by diesel fuel (8.4%), with a total energy consumption of 50,100 MJ ha⁻¹. The ANN 10-8-2 architecture provided the most accurate performance (R²). Using the MOGA method, optimum values were determined for minimum total GHGE and maximum watermelon production. The highest amount of production with minimum energy usage was approximately 10,900 MJ ha^-1. The GHGE of the best production were calculated as approximately 282 kg CO₂_eq ha^-1. The GHGE reduction potential using MOGA was calculated as 903 kg CO₂_eq ha^-1. Furthermore, the highest reduction in GHGE occurred in nitrogen fertilizer by 52.0%. The results also indicated that the highest amount of production with minimum energy usage is approximately 10,900 MJ ha^-1. The GHGE of the best production were calculated as approximately 282 kg CO₂_eq ha^-1. The GHGE reduction potential using MOGA was calculated as 903 kg CO₂_eq ha^-1. Furthermore, the highest reduction in GHGE occurred in nitrogen fertilizer by 52.0%.

Full Article

The Modeling and Optimization of Energy Inputs and Greenhouse Gas Emissions in Watermelon Production Using Artificial Neural Network and Multi Objective Genetic Algorithm

Bekir Yelmen ,^a,* Mutlu Tarık Çakır ,^b and Musa Faruk Çakır ^c

DOI: 10.15376/biores.21.3.6498-6517

Keywords: Artificial neural network; Genetic algorithm; Energy consumption; Greenhouse gas emissions

Contact information: a: Department of Environmental Protection Control, Adana Metropolitan Municipality, Adana 01355, Turkey; b: Department of Mechanical Engineering, Sivas University of Science and Technology, Sivas 58000, Turkey; c: Çankırı Karatekin University, Vocational School, Electronics and Automation, Çankırı 18200, Turkey; Corresponding author: bekiryelmen@gmail.com

INTRODUCTION

Watermelon is an annual plant that thrives in warm and temperate climates. It can be utilized entirely from its seeds to its rind and is used in various industries such as food, pharmaceuticals, and cosmetics. Approximately 3 million hectares of land worldwide are devoted to watermelon cultivation, producing about 100 million tons annually. Watermelon is the second most cultivated fruit in the world after bananas. China produces 60% of the world’s watermelon production, with 60.68 million tons. Turkey follows China with a production of 3.8 million tons. This is followed by India, Brazil, Algeria, Iran, Russia, and the United States follow Turkey, respectively. Turkey’s share in global watermelon production is about 3.9%. The average global watermelon yield is 33.6 tons ha^-1, while in Turkey this value is higher at 44.5 tons ha^-1 (FAOSAT 2022). Approximately 22.6% of Turkey’s total watermelon production occurs in Adana, where yield values reach up to 70 tons per hectare (TUIK 2021).

A sustainable approach to agriculture is achieved through efficient fuel use. This is also important for reducing greenhouse gas emissions (GHGE). Efficient energy (EE) use in agriculture can save fossil fuels and reduce air pollution. Therefore, both production costs and GHGE are reduced while also protecting the surroundings. GHGE occur due to agricultural activities, such as biocides, chemical fertilizers, machinery, electricity, and fuel use (Klikocka et al. 2019). Artificial Neural Networks (ANN) deal with event examples, generalize from them, gather information, and use the acquired knowledge to make decisions when encountering unseen examples. ANN has been employed to model energy balance and indices in cultivation (Niedbała 2019). There has been some analysis of energy usage in agricultural producing processes. This has included citrus fruits (Yılmaz and Aydın 2020), mandarin (Dagtekin et al. 2019; Bilgili 2021), maize (Farjam et al. 2014), winter rapeseed (Niedbala 2019), almond (Yılmaz and Bayav 2022), vetch (Seydosoglu et al. 2023), watermelon (Demir 2023), garlic (Baran et al. 2023), peach (Demir and Gokdogan 2023), and cherry (Gökdogan et al. 2024), orange ( Mohammadshirazi et al. 2015; Yelmen 2025).

One of the methods used to solve the optimization process is the Multi-Objective Genetic Algorithm (MOGA) method. The Genetic Algorithm (GA) is a searching method that uses random selection for a function optimization with parameter-space coding. Genetic algorithms have been developed by Holland (1975), with the primary references given by Bäck (1998) and investigations carried out by Goldberg (1997). Few studies have considered optimizing agricultural energy with genetic algorithms. An investigation on the optimization of energy usage in the cultivation of sugar beet had been carried out by Hemmati et al. (2013). Cultivation results showed an optimized total energy used for sugar beet production of 32,700 MJ ha^-1. Other investigations focused on modeling and the optimization of energy usage and GHGE with ANN and MOGA in eggplant crop production (Pelesaraei et al. 2013) and peanut crop production (Nabavi-Pelesaraei et al. 2013).

Based on the literature review, there appears to have been insufficient research on ANN modeling and MOGA-based improvement of EE and reduction of GHGE in watermelon farming in Adana county. Therefore, this study aimed to identify energy input (EI) indices for watermelon cultivation, model the energy consumption (EC), and GHGE using ANN, optimize them with MOGA, and identify the most topology for utmost prediction correctness.

MATERIALS AND METHODS

The data used in this study were obtained from interviews conducted with 68 farmers engaged in watermelon (Citrullus lanatus L.) cultivation during the March-May 2024 season in Adana. Hundreds of parameters for agricultural activities were compared in 68 farms. Adana is located (36°32′ – 38°25′ N; 34°39′-36°24′ E) in the Mediterranean Region of Turkey (GDM 2024).

Calculation of Energy Indices

To determine energy equivalence of agricultural sector, amount of performance per hectare for watermelon production involving of seeds, pesticides, machinery, fuel, farm yard manure, human work force, chemical fertilizers, and electricity as shown in Table 1 (Demir and Gokdogan 2023).

Table 1. Energy Equivalent of Inputs and Output in Agricultural Production

Equations 1 to 5 were used to determine the energy Indices including those for energy use efficiency (EUE), energy productivity (EP), specific energy (SE), energy density (ED), and net energy (NE). These equations were calculated using Table 1 (Nabavi-Pelesaraei et al. 2016; Macedo et al. 2021).

In these equations, EO is energy output (MJ ha^-1), EI is energy input (MJ ha^-1), crop yield has units of (kg ha^-1), and production cost has units of ($ ha^-1). Samples were classified into 3 categories (<1 ha, 1-3 ha, >3 ha). In agricultural activities, EUE and SE are complementary values. They are mentioned as direct energy (DE), indirect energy (IDE), renewable energy (RE), and non-renewable energy (NRE) (Demir 2023). The GHGE value for watermelon farming was calculated using Table 2 (Ozbek et al. 2024).

Table 2. Greenhouse Gas Emission Parameters of Agricultural Input

The GHGE for watermelon culture was determined by Eq. 6 (Karaağaç et al. 2019):

where GHG_ha is Greenhouse gas emission (kgCO_2eq.ha^-1), R(i) is the amount of i input (unit_input ha^-1), and EF(i) is the GHG emission equivalent of i input (kgCO_2eq.unit^-1_input). The value of I_GHG, based on GHG ratio, and Y crop in kg ha^-1, was computed utilizing Eq. 7 (Houshyar et al. 2015).

Modeling of Artificial Neural Networks

Artificial neural networks (ANN), possessing non-algorithmic and highly parallel processing capabilities, facilitate complex and nonlinear computations easily and rapidly through its learning ability and parallel distributed memory (Fig. 1) (Benti et al. 2023). Normalization in artificial neural networks is performed to enable the model to learn faster, more stably, and more efficiently.

Fig. 1. Artificial neural network model

ANN is a data input similar to a biological neural system. To solve specific problems, consists of many interconnected neurons. In this research, ANN model was used to train and test 52 and 16 units of watermelon crops, respectively. The selected units were obtained incidental from all examples. Several network structures were selected by evaluating empirical data to identify best predictive model. Numeral of neurons was defined based on input and output layers for watermelon cultivation and GHGE. Various neural network models with multiple layers and structures for the hidden layers have been proposed. Among them is the Levenberg-Marquardt (LM) learning model. All connections between input layers and hidden layers form the weight input matrix. The weight (W) at each node controlling the output, the input value (X), and evaluated output (O) are adjusted using Eq. 8 (Nabavi-Pelesaraei et al. 2016):

Here, T is a certain threshold value for each node and f is a nonlinear sigmoid function. In artificial neural networks, the sigmoid function is an activation function that compresses a neuron’s output between 0 and 1. In this study, Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and coefficient of determination (R²) functions were used to evaluate the model errors (Bhatti and Do 2019). The models were trained using training output and tested using the output and the RMSE criterion was calculated using Eq. 9:

In Eq. 10, m represents the number of tested outputs, Oᵢ is the predicted output from the ANNs, and yᵢ is the calculated (observed) output value. The symbol y denotes the mean of the calculated output values (yᵢ), while O represents the mean of the predicted output values (Oᵢ) from the ANN. MAPE values were computed using Eq. 11.

N represents number of training vectors, while the output of training vector “y” and “y′ “ denote the values that are tracked and simulated.

Multi-Objective Genetic Algorithm (MOGA)

When selecting MOGA parameters, a population size of 50-500, a crossover rate of 0.60-1.0, and a mutation rate of 0.001 to 0.1 were chosen. The main objective of this study was to investigate the integration of ANN and MOGA for modeling and optimization. While comparative modeling could provide additional insights, it is outside the scope of the current study. Genetic algorithms, classified as population-based methods, can efficiently be applied to deal with multiobjective optimization tasks. A typical single-objective genetic algorithm can easily be employed to derive multiple non-dominated solution sets in one execution. The characteristic of genetic algorithms to concurrently explore other parts of the solution space makes them suitable for finding diverse solutions to complex tasks with non-convex, discontinuous, or multimodal solution spaces (Konak et al. 2006). The first step for the optimization process using the MOGA method included calculating the production functions. The production functions were established through the energy inputs-output (watermelon production and total greenhouses gas emissions) described in Eqs. 12 and 13:

Here, Xᵢ represents the corresponding energy inputs: X₁ for human labor, X₂ for machinery, X₃ for diesel fuel, X₄ for nitrogen, X₅ for phosphate, X₆ for potassium, X₇ for biocides, X₈ for farmyard manure, X₉ for electricity, and X₁₀ for seeds. Yᵢ represents watermelon yield, and Gᵢ represents total greenhouse gas emissions.

Subsequently, the boundaries of the functions were calculated based on min and max EC values for each input. Fundamental information regarding the EI and GHGE in watermelon farming was coded into Excel 2019, SPSS 20, and MATLAB (R2024a).

RESULTS AND DISCUSSION

Analysis of Energy Use in Watermelon Culture

The energy demand for watermelon farming was roughly 50,100 MJ ha^-1. The results indicated that large-scale farm models exhibited best efficiency and productivity in watermelon farming. Furthermore, the findings showed that there was no major distinctness in EI among the three farm size groups (ANOVA). However, when the three farm models were compared in terms of watermelon yield, a significant difference was observed. In other words, while there was a notable difference in EI for large-scale watermelon production, there was no significant difference between small and medium scale farms on account of total EI and output. Table 3 demonstrates that large-scale farms achieve greater productivity than other scales of farms with the same total EI.

Table 3. Energy Input – Energy Output Relationship in the Watermelon Cultivation

In watermelon production, nitrogen fertilizer had the highest EC value, accounting for approximately 70.0%. This was followed by fuel 8.42% and electricity consumption 7.75%. Additionally, the shares of other inputs in watermelon farming were lower than the total average EI. Therefore, by managing nitrogen consumption and using farmyard manure instead of chemical fertilizers, EUE in watermelon production can be improved.

In some related studies, the total EI in watermelon farming was reported to be 46,300 MJ ha^-1, with the highest EC values in chemical fertilizers (28,600 MJ ha^-1) and fuel (3,030 MJ ha^-1) (Banaeian and Namdari 2011). For comparison, the total EI was reported as 45.5 MJ ha^-1 for tomatoes (Cetin and Vardar 2008), 30.3 GJ ha^-1 for kiwifruit (Mohammadi et al. 2010), and 45.2 GJ ha^-1 for grapes (Hamedani et al. 2011). Other examples include 81.4 GJ ha^⁻¹ for greenhouse tomatoes (Ozkan et al. 2011), 173 GJ ha^⁻¹ for pears (Tabatabaie et al. 2013), and 80.2 GJ ha^⁻¹for wheat farming (Safa and Samarasinghe 2011). On the basis of the levels of farm size given in the current study, EUE, EE, SE, NE, and ED results have been seen in Table 4. EUE for growing watermelon is determined to be 1.26. Other similar studies have given EUE of 0.8 for tomatoes (Ozkan et al. 2011), 0.96 for cherries (Kizilaslan 2009), 0.15 for strawberries (Banaeian et al. 2010), 1.16 for apples (Rafiee et al. 2010), 0.665 for garlic (Samavatean et al. 2011), 0.51 for pear (Tabatabaie et al. 2013), and 0.45 for wheat (Safa and Samarasinghe 2011).

Table 4. Energy Input-Energy Output Ratio in Watermelon Cultivation

As shown in Table 4, small-size farms in watermelon production have the highest EUE value of 1.29, due to the fact that the ratio of EO to EI is higher. Increase in EO for small farms is significantly greater than the increase in EI compared to medium and large farms. In contrast, the lowest EUE value of 1.24 was calculated for large-scale farms. Therefore, small-scale farms in Adana province, Turkey, are considered more suitable in terms of EUE for watermelon production.

Additionally, EE, SE, NE, and ED were computed to be 0.79 kg MJ^⁻¹, 1.59 MJ kg^⁻¹, 13,220.61 MJ ha^⁻¹, and 10.34 MJ $^⁻¹. The distribution of inputs used in watermelon cultivation for three farm sizes across the DE, IDE, and RE, NRE groups is presented in Table 4. The results indicate that IDE accounts for 80.2% of the total EI, while DE represents 19.8%. Furthermore, only 3.83% of sum EI utilized in watermelon cultivation was derived off renewable energy origined. The main renewable energy sources in watermelon cultivation are chemical fertilizers, fuel, and electricity. Therefore, using renewable inputs such as farmyard manure can increase the share of RE, while productive utilize of diesel fuel and irrigation water can help diminish NRE dependency.

Greenhouse Gas Emissions from Watermelon Cultivation

The total GHGE were found to be 1,250 kg CO₂_eq ha^⁻¹. In a similar study, the CO₂ emission from organic fig production was reported to be approximately 1,109.02 kg CO₂_eq ha⁻¹ (Oguz et al. 2022). Another study calculated total CO₂ emission from okra production as approximately 875 kg CO₂_eq ha^-1 (Sarkar et al. 2023). Greenhouse gas emissions from tomato cultivation were found to be 323 kg CO₂_eq ha^-1 (Sarkar et al. 2023). Greenhouse gas emissions are shown in Table 5.

Table 5. Greenhouse Gas Emission of inputs used in Watermelon Cultivation

Note: Different letters show significant difference of means at 5% level.

In watermelon cultivation, the GHGE share was highest in large-scale farms, although the lowest share was observed in medium-scale farms. However, there was not any statistically (ANOVA) considerable segregation in GHGE amongst three farm groups. With respect to the table, nitrogen fertilizer had the highest share within total GHGE at 54.3%, which was followed by fuel (16.5%) and electricity consumption (15.8%). The contribution of watermelon cultivation inputs to greenhouse gas emissions (GHGE) is illustrated in Fig. 2.

Fig. 2. Contribution of watermelon production inputs to the use of greenhouse gas emission

Through better agricultural management of nitrogen application, lower greenhouse gas emissions (GHGE) can be achieved in the studied area. Considering that Adana province receives sufficient rainfall, reducing irrigation water use electricity consumption in watermelon production could lead to significant improvements in lowering greenhouse gas emissions.

ANN Model Structure and Evaluation

For the purpose of modeling the production efficiency of watermelon and the GHGE, neural networks were developed. These networks were developed using the MATLAB software (R2024a). This modeling was done based on several parameters such as the range of the layers and neurons in the networks. Additionally, there was involvement of the learning parameters. For this modeling to be done effectively, there is a Levenberg-Marquardt learning algorithm that is used. This training is done with 75% of the data for the training datasets. Further testing is done with 52 samples. As shown in Table 6, statistical scales were used to evaluate the ANN models and predict watermelon production efficiency and GHGE.

Table 6. Results of Different Models

Through test and fault, top capacity was obtained with use of an ANN model of 10-8-2 structure. The texture of the selected ANN is schematic illustrated in Fig. 3. The MATLAB interface and training data are shown in Fig. 3. The practice and testing consequences are presented in Table 6. The architectural choice was made through a comparative analysis of multiple artificial neural network structures (10-9-2, 10-8-2, and 10-7-2), as presented in the Table. The 10-8-2 model was chosen based on its superior performance metrics (highest R² and lowest RMSE and MAPE). For watermelon performance and GHGE, the best topology with the highest R² and the lowest RMSE and MAPE values demonstrated that both practice and testing stages of the ANN models closely followed the predicted watermelon yield and GHGE values. The R² values ranged between 0.912 and 0.982 for training data and 0.901 and 0.991 for testing data.

Fig. 3. Artificial neural networks model with 10-8-2 topology

Fig. 4. The MATLAB interface and training data

The R² value for training data was between 0.978 and 0.982, while the R² value for test data was between 0.964 and 0.991. This indicates that the training and test results were very close, suggesting there was no overfitting (if there were, the training data would be larger than the test data). Therefore, the similarity between test and training performances shows that the model did not exhibit overfitting. In artificial neural networks, an epoch refers to the time it takes for the entire training dataset to pass through the network exactly once. The artificial neural network model created uses a training stopping criterion of 1000 iterations, an error value of “0”, and a gradient of 1e-7. Training reached 1000 iterations in 3 seconds. The MATLAB interface and training data are shown in Fig. 4.

The ANN model 6-9-1 capable of predicting jute cultivation in Bangladesh was developed by Rahman and Bala (2010). For basil production, the best topology was reported as 7-20-20-1 (Pahlavan et al. 2012). Another study advanced an ANN to estimate watermelon production efficiency and GHGE (Nabavi-Pelesaraei et al. 2016). The outcomes showed that 12-9-9-2 ANN was the best topology for estimating performance and GHGE in watermelon production. The training and testing datasets are illustrated in Figs. 5 and 6. As shown the estimated and actual values are consistent.

Fig. 5. Correlation between predicted and real outputs energies based on the best topology

Fig. 6. Correlation between predicted and real outputs GHG emission based on the best topology

The coefficient of determination showed that the advanced network for watermelon cultivation was quite suitable in terms of output energy and GHG emission. The R² parameters for watermelon performance and GHGE were found to be 0.988 and 0.997 for the training data, respectively. Furthermore, for the test data, the factor of determination R² for watermelon performance and GHGE was 0.980 and 0.998, respectively.

Fig. 7. Predicted and real energy output values

In Fig. 7, the y-axis is used to display normalized outputs. The x-axis is used to display the no of estimated data group. The blue and green curves demonstrate actual outputs. On the other hand, the red and magenta curves represent how well the model tracked the predicted outputs.

In ANNs, sensitivity analysis measures how much small changes in input variables affect the model’s output. Sensitivity analysis is performed to test accuracy of a model’s results. In sensitivity analysis conducted using ANN, the input variants were listed using a fractional differential study (Table 7).

Table 7. Sensitivity Analysis Results for Input Energies

Optimization of Energy Inputs and Greenhouse Gas Emissions

The models for yield and GHGE were optimized with MOGA-based energy inputs. The limitations of the energy inputs are shown in Table 8. The minimum amount of energy consumed is taken as the lower limit, and the largest amount of energy that can be consumed is taken as the upper limit.

Table 8. Limits of Functions for Multi-Objective Genetic Algorithm (MJ ha^-1)

In addition, the production function was calculated according to Eqs. 12 and 13:

In this study, with MOGA calculated 52 optimal production cases based on maximum watermelon yield and minimum total greenhouse gas emissions (Table 9). Moreover, the best production scenario with minimum energy usage was identified. Thus, production No. 26 was identified as the best production scenario with minimum energy usage. This means that this production scenario had the highest production amount with low production costs and low production-related GHG emissions. The total energy usage and GHG emissions of the best cultivation scenario were identified to be 10,900 MJ ha^-1 and 282 kg CO₂_eq ha^-1. In the final part of this investigation, the potential for the reduction of greenhouses gases was computed utilizing the MOGA method. Figure 8 above denotes the percentage contribution to the overall greenhouse gases reduction.

Fig. 8. Distribution of GHGE reduction for each input in watermelon production

According to the results, nitrogen 52.0% had the highest contribution to greenhouse gas reduction, followed by diesel fuel (17.1%) and machinery 16.4%. Therefore, as a first step, reducing diesel fuel consumption and optimizing machinery use through timely maintenance and proper machine selection is recommended. Adopting minimum tillage, no-tillage systems, and the use of biofertilizers in place of chemical fertilizers can significantly reduce both energy consumption and GHGE in the research field.

Table 9. Multi-Objective Genetic Algorithm Results for Optimization of Energy Inputs and GHGE in Watermelon Production

CONCLUSIONS

This study aimed to evaluate artificial neural net (ANN) models to make predictions based on energy input (EI) consumption by optimizing the EI and greenhouse gas emissions (GHGE) in Adana province, where watermelon is grown in the Mediterranean region of Türkiye.

Evaluations were made in terms of EI and energy output (EO), yield, Renewable energy (RE), energy efficiency (EE), specific energy (SE), and net energy (NE). For watermelon cultivation, the total EI was computed as 50,100 MJ ha^-1, with 77.0% attributed to chemical fertilizers. The EO was computed at 63,300 MJ ha^-1. Large-scale farms showed high EC and EO at 53,100 and 66,100 MJ ha^-1, respectively. Differences between EI and EO were not significant among the three farm size levels. The mean measures of EUE, EE, SE, NE, and ED were determined to be 1.26, 0.79 kg MJ^-1, 1.59 MJ kg^-1, 13,200 MJ ha^-1, and 10.30 MJ$^-1.
Results revealed that medium sized farms had significantly higher ER, EE, and NE values in comparison to other farms, indicating that energy was used efficiently in watermelon cultivation. It was also concluded that ER can be further improved by enhancing vegetative performance or reducing energy consumption.
The EUE index indicates that watermelon production is an energy-intensive process. In terms of energy generation, farm productivity can be increased by using less EI through reduced diesel fuel and natural gas consumption, electricity savings, and various energy-saving techniques while increasing output energy.
In watermelon production, the share of NRE sources 96.2% was much higher than that of RE sources (3.83%). The total EI composed of DE, IDE, RE and NRE was 9,900, 40,200, 1,920, and 48,200 MJ ha^-1, respectively.
The GHG analysis showed that the total GHG emissions were 1,240 kg CO₂_eq ha⁻¹. Chemical fertilizers had the greatest part of GHGE (58.9%), followed by fuel (16.5%) and electricity (15.8%). Moreover, differences in GHGE among the three farm sizes were not statistically significant.
In watermelon growing, the proportion of NRE (96.2%) was substantially larger than that of RE (3.83%). The total EI including DE, IDE, RE, and NRE was 9,900, 40,200, 1,920, and 48,200 MJ ha^-1, respectively. Results of the GHGE indicated that total GHG emissions amounted to 1,240 kg CO₂_eq ha^-1. Among those components, chemical fertilizers (GHGE) had the largest percentage (58.9%) contribution to overall GHGE followed by fuel (16.5%) and electricity (15.8%). Furthermore, differences in GHGE were not significant across the three farm sizes.
For this purpose, multiple ANN models were developed to estimate energy consumption. The results showed that the 10-8-2 Levenberg-Marquardt ANN model was the most accurate model to estimate watermelon production efficiency and GHGE. R² for training datasets for both watermelon performance and GHGE were 0.978 and 0.982. RMSE was 0.149 and 0.067 for watermelon performance and GHGE, respectively. MAPE was 0.006 and 0.005 for watermelon performance and GHGE, respectively. A well-trained ANN model can be applied to estimate energy consumption of other crops. In accordance with EUE examine results, the ANN model was deemed to be advantageous in modeling EI in watermelon production with high accuracy.
Multi-objective optimization results showed that MOGA evaluated 52 production units, including maximum yield and minimum total greenhouse gas emissions, as optimal units; however, the best production was selected based on minimum energy consumption. The total EC and GHGE of the best output were estimated at approximately 10,900 MJ ha^-1 and 282 kg CO_2eq ha^-1. Furthermore, nitrogen fertilizer accounts for the largest portion of greenhouse gas reductions, at 51.96%. This saves on fertilizer consumption and significantly reduces greenhouse gas emissions.

Conflict of Interest

The authors declare no conflict of interest.

REFERENCES CITED

Bäck, T. (1998). Evolutionary Algorithms in Theory and Practice: Evolution Strategies, Evolutionary Programming, Genetic Algorithms, Oxford Academic. https://doi.org/10.1108/k.1998.27.8.979.4

Banaeian, N., Omid, M., and Ahmadi, H. (2010). “Energy and economic analysis of greenhouse strawberry production in Tehran province of Iran,” Energy Conversion Management 52(2), 1020-1025. https://doi.org/10.1016/j.enconman.2010.08.030

Banaeian, N., and Namdari, M. (2011). “Effect of ownership on energy use efficiency in watermelon farms: A data envelopment analysis approach,” International Journal of Renewable Energy Research 1(3), 75-82. https://dergipark.org.tr/en/pub/ijrer/issue/16086

Baran, M. F., Demir, C., Elicin, A. K., and Gokdogan, O. (2023). “Energy use efficiency and greenhouse gas emissions (GHG) analysis of garlic cultivation in Turkey,” Int. J. Agric. Biol. Eng. 16(4), 63-67. https://doi.org/10.25165/j.ijabe.20231604.7599

Benti, N. E., Chaka, M. D., and Semie, A. G. (2023). “Forecasting renewable energy generation with machine learning and deep learning: Current advances and future prospects,” Sustainability 15(9), article 7087. https://doi.org/10.3390/su15097087

Bhatti, M. I., and Do, H. Q. (2019). “Recent development in copula and its applications to the energy, forestry and environmental sciences,” Int. J. Hydrogen Energy 44(36), 19453-19473. https://doi.org/10.1016/j.ijhydene.2019.06.015

Bilgili, M. E. (2021). “Energy use efficiency of mandarin production: A case study from Adana province,” Erwerbs-Obstbau 63(2), 61-64. https://doi.org/10.1007/s10341-021-00548-8

Cetin, B., and Vardar, A. (2008). “An economic analysis of energy requirements and input costs for tomato production in Turkey,” Renewable Energy 3(3), 428-433. https://doi.org/10.1016/j.renene.2007.03.008

Dagtekin, M., Bilgili, M. E., and Beyaz, A. (2019). “The effects of good agricultural practises on energy use effectiveness on mandarin production and environmental,” Applied Fruit Science, Erwerbs-Obstbau 61(1), 55-60. http://doi.org/10.1007/s10341-018-0395-5

Demir, C. (2023). “Analyses of energy use and greenhouse gas emissions (GHG) in watermelon production,” International Journal of Agricultural and Biological Engineering 16(5), 221-225. https://doi.org/10.25165/j.ijabe.20231605.7918

Demir. C., and Gokdogan, O. (2023). “Determination of energy use efficiency and greenhouse gas emissions in peach production,” International Journal of Agricultural and Biological Engineering 16(2), 165-170. https://doi.org/10.25165/j.ijabe.20231602.7917

FAOSAT (2022). “Food and agriculture organization,” Statistical Database. https://www.fao.org/faostat/en/#data/QI

Farjam, A., Omid, M., Akram, A., and Fazel-Niari, Z. (2014). “A neural network based modeling and sensitivity analysis of energy inputs for predicting seed and grain corn yields,” Journal of Agricultural Science and Technology 16, 767–778. https://jast.modares.ac.ir/article_15285.html

GDM (2025). “Republic of Türkiye ministry of agriculture, forestry general directorate of meteorology,” https://mgm.gov.tr/?il=Adana.

Gokdogan, O., Demir, C., and Baran, M. F. (2024). “Energy balance and greenhouse gas emissions of cherry production in Turkey,” Applied Fruit Science, Erwerbs-Obstbau 66, 1269-1274. https://doi.org/10.1007/s10341-024-01113-9

Goldberg, D. E. (1997). Genetic Algorithms, in Search, Optimization and Machine Learning, Addison Wesley. Publishing Co Reading Mass. 97-125. https://api.semanticscholar.org/CorpusID:38613589

Hamedani, S. R., Keyhani, A., and Alimardani, R. (2011). “Energy use patterns and econometric models of grape production in Hamadan province of Iran,” Energy 36, 6345-6351. https://doi.org/10.1016/j.energy.2011.09.041

Hemmati, A., Tabatabaeefar, A., and Rajabipour, A. (2013). “Comparison of energy flow and economic performance between flat land and sloping land olive orchards,” Energy 61, 472-478. https://doi.org/10.1016/j.energy.2013.09.006

Holland, J. H. (1975). Adaptive in Natural and Artificial Systems, Ann Arbor, MI: University of Michigan Press. https://link.springer.com/chapter/10.1007/0-387-28356-0_4#ref-CR70

Houshyar, E., Dalgaard, T., Tarazgar, M. H., and Jorgensen, U. (2015). “Energy input for tomato production what economy says, and what is good for the environment,” Journal of Cleaner Production 89, 99-109. https://doi.org/10.1016/j.jclepro.2014.11.022

Karaagac, H. A., Baran, M. F., Mart, D., Bolat, A., and Eren, O. (2019). “Determination of energy usage efficiency and greenhouse gas (GHG) emissions in chickpea production (the case of Adana province),” European Journal of Science and Technology 16,41-50. https://doi.org/10.31590/ejosat.523080

Klikocka, H., Kasztelan, A., Zakrzewska, A., Wyłupek, T., Szostak, B., and Skwaryło-Bednarz, B. (2019). “The energy efficiency of the production and conversion of spring triticale grain into bioethanol,” Agronomy 9(8), article 423. https://doi.org/10.3390/agronomy9080423

Kizilaslan, H. (2009). “Input-output energy analysis of cherries production in Tokat Province of Turkey,” Applied Energy 86(7), 1354-1358. https://doi.org/10.1016/j.apenergy.2008.07.009

Konak, A., Coit, D. W., and Smith, A. E. (2006). “Multi objective optimization using genetic algorithms: A tutorial,” Reliability Engineering and System Safety 91(9), 992-1007. https://doi.org/10.1016/j.ress.2005.11.018

Macedo, I., Terra, J.A., Siri-Prieto, G., Velazco, J. I., and Carrasco-Letelier, L. (2021). “Rice pasture agroecosystem intensify cation affects energy use efficiency,” Journal of Cleaner Production 278, article 123771. https://doi.org/10.1016/j.jclepro.2020.123771

Mohammadi, A., Rafiee, S., Mohtasebi, S., and Rafiee, H. (2010). “Energy inputs-yield relationship and cost analysis of kiwifruit production in Iran,” Renewable Energy 35(5), 1071-1075. https://doi.org/10.1016/j.renene.2009.09.004

Mohammadshirazi, A., Akram, A., Rafiee, S., and Kalhor, E. B. (2015). “On the study of energy and cost analyses of orange production in Mazandaran province,” Sustainable Energy Technologies and Assessments 10, 22-28. https://doi.org/10.1016/j.seta.2015.01.007

Nabavi-Pelesaraei, A., Abdi, R., and Rafiee, S. (2013). “Energy use pattern and sensitivity analysis of energy inputs and economical models for peanut production in Iran,” International Journal of Agriculture and Crop Sciences (IJACS) 5(19), 2193-2202. https://ijagcs.com/

Nabavi-Pelesaraei, A., Abdi, R., and Rafiee, S. (2016). “Neural network modeling of energy use and greenhouse gas emissions of watermelon production systems,” Journal of the Saudi Society of Agricultural Sciences 15(1), 38-47. https://doi.org/10.1016/j.jssas.2014.05.001

Niedbala, G. (2019). “Simple model based on artificial neural network for early prediction and simulation winter rapeseed yield,” Journal of İntegrative Agriculture 18(1), 54-61. https://doi.org/10.1016/S2095-3119(18)62110-0

Oguz, H. I., Baran, M. F., Gokdogan, O., Eren, O., and Solak, M. (2022). “Energy balance and greenhouse gas (GHG) emissions of organic fig (Ficus carica L.) production in Turkey,” Erwerbs-Obstbau 64, 61-67. https://doi.org/10.1007/s10341-021-00617-y

Ozbek, O., Celik, S. A., Colak, I. A., and Gokdogan, O. (2024). “Energy balance and greenhouse gas (GHG) emissions of sesame (Sesamum indicum L.) production in Turkey,” Turkish Journal of Agriculture Food Science and Technology 12(6), 911-917. https://doi.org/10.24925/turjaf.v12i6.911-917.6756

Ozkan, B., Ceylan, R. F., and Kizilay, H. (2011). “Comparison of energy inputs in greenhouse double crop (fall and summer crops) tomato production,” Renewable Energy 36(5), 1639-1644. https://doi.org/10.1016/j.renene.2010.11.022

Pahlavan, R., Omid, M., and Akram, A. (2012). “Energy input–output analysis and application of artificial neural networks for predicting greenhouse basil production,” Energy 37(1), 171-176. https://doi.org/10.1016/j.energy.2011.11.055

Pelesaraei, A. N., Koohi, S. S., and Dehpour, M. B. (2013). “Modeling and optimization of energy ınputs and greenhouse gas emissions for eggplant production using artificial neural network and multi-objective genetic algorithm,” International Journal of Advanced Biological and Biomedical Research 1(11), 1478-1489. https://www.ijabbr.com/article_7973.html

Rafiee, S., Mousavi-Avval, S. H., and Mohammadi, A. (2010). “Modeling and sensitivity analysis of energy inputs for apple production in Iran,” Energy 35(8), 3301-3306. https://doi.org/10.1016/j.energy.2010.04.015

Rahman, M. M., and Bala, B. K. (2010). “Modelling of jute production using artificial neural networks,” Biosystems Engineering 105(3), 350-356. https://doi.org/10.1016/j.biosystemseng.2009.12.005

Safa, M., and Samarasinghe, S. (2011). “Determination and modelling of energy consumption in wheat production using neural networks: A case study in Canterbury province, New Zealand,” Energy 36(8), 5140-5147. https://doi.org/10.1016/j.energy.2011.06.016

Samavatean, N., Rafiee, S., Mobli, H., and Mohammadi, A. (2011). “An analysis of energy use and relation between energy inputs and yield, costs and income of garlic production in Iran,” Renewable Energy 36(6), 1808-1813. https://doi.org/10.1016/j.renene.2010.11.020

Sarkar, B., Das, B., Sundaram, P., Mali, S., Anurag, A., Upadhyaya, A., Chandra, N., Bhatt, B., and Kumar, A. (2023). “Energy input output analysis and greenhouse gas emission in okra and tomato production in Chotanagpur plateau region of India,” Environment Development and Sustainability 25(1), 12945-12964. https://doi.org/10.1007/s10668-022-02596-w

Seydosoglu, S., Baran, M.F., Turan, N., Alfarraj, S., and Albasher, G. (2023). “Greenhouse gas emission and energy analysis of vetch (Vicia sativa L.) cultivation,” Jour. King. Saud. Univ. Science 35(3), article 102541. https://doi.org/10.1016/j.jksus.2023.102541

Tabatabaie, S. M. H., Rafiee, S., Keyhani, A., and Heidari, M. D. (2013). “Energy use pattern and sensitivity analysis of energy inputs and input costs for pear production in Iran,” Renewable Energy 51, 7-12. https://doi.org/10.1016/j.renene.2012.08.077

TUIK (2021). “Turkish Statistical Institute,” Plant Production Statistics. https://data.tuik.gov.tr/Bulten/Index?p=Bitkisel-Uretim-Istatistikleri-2021-37249

Yelmen, B. (2025). “Modeling of energy use and greenhouse gas emissions in orange production with artificial neural networks: Case study of Turkey,” Int. Agrophys 39(2), 165-173. https://doi.org/10.31545/intagr/200128

Yilmaz, H., and Aydın, B. (2020). “Comparative input-output energy analysis of citrus production in Turkey: Case of Adana province,” Applied Fruit Science Erwerbs-Obstbau 62, 29-36. https://doi.org/10.1007/s10341-019-00447-z

Yılmaz, A., and Bayav, A. (2022). “Determination of energy efficiency in almond production according to variety: A case study in Turkey,” Erwerbs-Obstbau 65(4),971-979. https://doi.org/10.1007/s10341-022-00728-0

Article submitted: January 1, 2026; Peer review completed: April 17, 2026; Revised version received: April 23, 2026; Accepted: May 6, 2026; Published: May 28, 2026.

DOI: 10.15376/biores.21.3.6498-6517