CC BY-NC-ND
0Information Processing in Agriculture
— Se, * ^
Combined application of Artificial Neural Networks and life cycle assessment in lentil farming in Iran
Behzad Elhami, Majid Khanali *, Asadollah Akram
Department of Agricultural Machinery Engineering, Faculty of Agricultural Engineering and Technology, College of Agriculture and Natural Resources, University of Tehran, Karaj, Iran
ARTICLE INFO
ABSTRACT
Article history: Received 9 September 2016 Accepted 31 October 2016 Available online xxxx
Keywords:
Artificial Neural Network Energy balance Environmental impacts Lentil production Sensitivity analysis
In this study, an Artificial Neural Network (ANN) was applied to model yield and environmental emissions from lentil cultivation in Esfahan province of Iran. Data was gathered from lentil farmers using face to face questionnaire method during 2014-2015 cropping season. Life cycle assessment (LCA) was applied to investigate the environmental impact categories associated with lentil production. Based on the results, total energy input, energy output to input ratio and energy productivity were determined to be 32,970.10 MJ ha1, 0.902 and 0.06 kg MJ1, respectively. The greatest amount of energy consumption was attributed to chemical fertilizer (42.76%). Environmental analysis indicated that the acidification potential was higher than other environmental impact categories in lentil production system. Also results showed that the production of agricultural machinery was the main hotspot in abiotic depletion, eutrophication, global warming, human toxicity, fresh water aquatic ecotoxicity, marine aquatic ecotoxicity and terrestrial ecotoxicity impact categories, while direct emissions associated with lentil cultivation was the main hotspot in acidification potential and photochemical oxidation potential. In addition, diesel fuel was the main hotspot only in ozone layer depletion. The ANN model with 9-10-6-11 structure was identified as the most appropriate network for predicting yield and related environmental impact categories of lentil cultivation. Overall, the results of sensitivity analysis revealed that farmyard manure had the greatest effect on the most of the environmental impacts, while machinery was the most affecting parameter on the yield of the crop.
© 2016 China Agricultural University. Publishing services by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-
nd/4.0/).
Introduction
Lentil (Lens culinaris) as a cool-season annual bushy plant or pulse crop is a member of the legume family. Lentils are grown for their high protein content (about 25%) and supply specially the essential amino acids lysine and leucine for human diet. The crop can play a major role in sustaining soil
fertility because of its symbiotic nitrogen fixing ability, especially in cereal-based cropping systems. In addition, its straw can be used as animal feed [1,2].
Lentil as a human diet is one of the most common legumes in the regions of Middle East and Asia. Iran ranks tenth in the world in production of lentil with total production of 78,500 tons and a world share of 1.6%. The cultivation area of lentil in Iran is 140,000 ha which ranked the sixth most cultivated area in the world. Esfahan province is one of the most important areas of lentil production in the country.
* Corresponding author. Fax: +98 2632808138. E-mail address: khanali@ut.ac.ir (M. Khanali). Peer review under responsibility of China Agricultural University. http://dx.doi.org/10.1016/jinpa.2016.10.004
2214-3173 © 2016 China Agricultural University. Publishing services by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Nomenclature
ANN Artificial Neural Network LCA life cycle assessment
ADP abiotic depletion potential LCI life cycle inventory
ACP acidification potential LM Levenberg-Marquardt
DCB dichlorobenzene ME machinery energy
DE direct energy MAPE mean absolute percentage error
ER energy ratio MAEP marine aquatic ecotoxicity potential
EP energy productivity NMVOC non-methane volatile organic compounds
EUP eutrophication potential NRE nonrenewable energy
FAEP freshwater aquatic ecotoxicity potential NEG net energy gain
FYM farmyard manure OLDP ozone layer depletion potential
FU functional unit PHOP photochemical oxidation potential
GWP global warming potential RMSE root mean square error
GHG greenhouse gas RE renewable energy
HTP human toxicity potential SE specific energy
INE indirect energy TEP terrestrial ecotoxicity potential
IE irrigation energy
About 12% of the total lentil production in Iran is supplied in this province [3].
In Iran, despite the high cultivation area of lentil, its average yield is about 700 kg ha-1 (Anonymous, 2014) which is relatively low in comparison to the average yield of 1800 kg ha-1 in Canada [4]. This comparison shows that increasing the average yield can increase the profitability of lentil cultivation and subsequently increase production [4]. Iranian lentil is produced in a semi-mechanized agricultural system. Land preparation, seeding and fertilization operations are generally conducted in a mechanized manner while weeding, spraying and harvesting operations are performed manually. The manual harvesting operations of Iranian lentil is due to the short height of the plant and lack of suitable harvesting machinery that could cut the plant near the soil surface in the generally stony land of farms in Iran. Also the right amount of chemical fertilizers is not consumed in lentil cultivation and their consumption is determined based on the farmer's experience that resulted in relatively high consumption of fertilizers. Therefore the cost, energy use and environmental damages in lentil production are unreasonably high. Regarding these issues, more efficient use of energy and better environmental management in lentil cultivation are important to provide a sustainable production, therefore modeling the environmental impacts associated with lentil production was recognized as important tool for both farmers and decision makers in agriculture.
Life cycle assessment (LCA) methodology is applied for environmental analysis of a product by establishing the inventory of the energy and material inputs vs environmental emissions brought about from each stage of the life cycle of the product, from resource extraction until processing, application, disposal and expressing the results in terms of impact categories [5]. Nowadays, global warming is considered as one of the most serious environmental impact categories man is confronted with. Greenhouse gas (GHG) emissions from agricultural production systems account for 11% of all manmade GHG emissions [4,6]. Thus, LCA is becoming more and more
important in the agro-food sector. A review of the literature demonstrated that several researchers have assessed the environmental impacts related to different agricultural products throughout their life cycle using LCA based on cradle to grave approach [7-14].
Despite the importance of lentil production in Iran and considerable amount of research work that has been assessed and predicted the energy use and environmental impacts of agricultural products, the number of publications accessed these topics in the cultivation of lentil or even other legumes is rather small. Abeliotis et al. [15] conducted an LCA study to compare the production of three different varieties of bean in Greece based on three different cultivation methods, i.e., conventional, integrated and organic. Overall results showed that integrated agricultural method could preferably be used to establish the most environmentally friendly production system among the three. Romero-Gamez et al. [8] evaluated the environmental impacts attributed to green beans production in three different cropping systems in Spain including screen house, screen house equipped with misting system and cropping in the open field by applying LCA. Koocheki et al. [16] performed an energy input-output analysis of pulses (lentil, bean and chickpea) production in Khorasan Razavi province of Iran. It must be mentioned that they considered the embodied energy in straw as the output energy, while in the present study only the energy of lentil was considered as the output energy.
Artificial Neural Network (ANN) technique has proved to be of several applications for modeling, simulation and forecasting in the complex nonlinear systems in which there is not any linear or simple relationship between inputs and out-put(s). Capturing the underlying relationship is known as the process of learning the network [17,18]. ANN models were used to predict energy usage, yield and environmental emissions related to agricultural products in various studies. Khoshnevisan et al. [19] developed an ANN model for estimating output energy and GHG emissions in terms of global warming potential (GWP) of potato production in Esfahan
province of Iran as a function of input energies, i.e., human labor, diesel fuel, electricity, seed, machinery, farmyard manure, chemical fertilizers, biocides and irrigation water. They deduced that modeling of both output energy and GWP was performed with a high accuracy. Nabavi-Pelesaraei et al. [20] applied ANN in modeling energy use and GWP of kiwifruit production in Iran based on all inputs within the area. Pahla-van et al. [21] developed an ANN model to estimate the production yield of greenhouse basil in Iran based on energy inputs. Safa and Samarasinghe [18] developed an ANN model for predicting energy consumption in wheat production based on farm conditions, farmers' social considerations and energy inputs in New Zealand. They mentioned that ANN model can predict energy consumption relatively better than the applied multiple linear regression model. In another study, ANN model with Levenberg-Marquardt (LM) training algorithm was applied to predict yield and GWP of watermelon production in Guilan province of Iran [22]. Taghavifar and Mardani [23] developed an ANN model to predict the yield and GWP of apple production in West Azarbayjan of Iran on the basis of input energies. They highlighted that ANN is a powerful and robust tool for studding energy and environmental emissions in agricultural systems.
To the best of authors' knowledge, there is no study up to date on the prediction of yield (output energy) and environmental impact categories of lentil production using ANN models in Iran and even all over the world. Although in all the conducted studies in this field, only one environmental impact category, i.e., GWP was considered. Therefore it is essential to develop an ANN model that can predict simultaneously a number of environmental impact categories and yield based on input energies. Therefore, the main objective of the present study was to estimate the ten environmental impact categories presented by CML2 baseline method and yield of lentil production in Iran using ANN modeling technique. Accordingly, several ANN models were structured and their performance for prediction of output parameters evaluated using the statistical quality parameters. Finally, the sensitivity analysis of the energy inputs on lentil yield and the environmental impact categories were investigated.
Materials and methods
2.1. Case study region and data collection
This research was conducted in Esfahan province, located between 30°42' and 34°30' N latitudes and 49°36' and 55°32' E longitudes, in the center of Iran [24]. The study covered the rural areas in the five regions of the province including Chadegan, Fereydonshahr, Fereydan, Tiran and Semirom. The data was collected from 140 lentil farmers using a face to face questionnaire method in 2014-2015 cropping season. The total area of the investigated lentil farms in the studied area was 163.5 ha. The average size of the lentil farms in Chadegan, Fereydonshahr, Fereydan, Tiran and Semirom regions were 1.12, 1.22, 1.38, 0.91 and 0.85 ha, which were not statistically significant. It must be noted that in these regions, apart from lentil, other important crops such as wheat, sugar beet and chickpea were cultivated. Before the
data being collected, a pre-test survey was done; thus, a group of farmers randomly selected and interviews conducted. For sampling, simple random sampling method was used. The sample size was determined using Cochran method as follows [25,26]:
N x S2 x r2
(N - 1)e2 + (S2 x t2)
where 'n' denotes the calculated sample size, 'N' stands for the number of lentil farmers in target population, 'S' presents the standard deviation for the pre-tested data, 'r' denotes the reliability coefficient (1.96 which represents 95% confidence) and 'e' stands for the acceptable error, which was defined to be 5% for a confidence level of 95%.
Energy balance in lentil cultivation
The input energy sources for lentil production in the region included human labor, machinery, diesel fuel, farmyard manure (FYM), chemical fertilizer, electricity, chemicals (pesticides) and seeds while the produced lentil accounted as the output energy.
In order to convert inputs and output materials into energy forms, the energy equivalent coefficients was used as detailed in Table 1. In this study, the corresponding energy coefficients were extracted from the literature. These coefficients are constant values that do not depend on the product type. For example, diesel fuel and human labor in the production of different products are of the same nature and have constant coefficients for the conversion to their energy forms. Thus, the energy consumption in various agricultural products differs in input values. Expressing the energy consumption in lentil production using standard coefficients resulted in the unique pattern of energy consumption of the crop. Therefore, it will be possible to compare the energy consumption in different products or production systems.
To assess the energy consumption by agricultural machinery in different farm operations, it was assumed that energy use for the manufacturing of agricultural implements and tractors be depreciated during their economic life time [27]. Therefore, the following formula was used to estimate machine energy (ME) per hectare [27,28]:
G x M„ x t
ME = -
where 'ME' is the machine energy (MJ ha-1), 'T' is the economic life of the machine (h), 'G' stands for the mass of the machine (kg) and 't' denotes the operation time of the machine per unit area (h ha-1).
Irrigation energy (IE) was expressed as below [29]:
IE = d X 3 X H X Q (3)
g1 X g2
where 'IE' is irrigation energy (J ha-1), 'g' is gravitational acceleration (9.81 m s-2), 'd' stands for the density of water (1000 kg m-3), 'Q' presents the overall quantity of water (m3 ha-1) including losses by evaporation, drainage run-off, etc., 'H' denotes the total dynamic head (m), 'ga' is the pump efficiency and 'g2' is representing the efficiency of the powering system, either electric motor or diesel engine.
Table 1 - Energy equivalent of inputs and output in lentil production.
Input-output (Unit) Energy equivalent (MJ per unit) References
1. Inputs
Labor (h) 1.96 [21]
Machinery (kg)
Tractor 138 [29]
Plow 180 [29]
Disk 149 [29]
Boundaries 160 [29]
Leveler 149 [29]
Planter 133 [29]
Sprayer 129 [29]
Rotary Hoes 148 [29]
Thrashing (h) 62.7 [29]
Seed (kg) 14.7 [16]
Chemicals (kg)
Herbicide 238 [11]
Insecticide 101.2 [11]
Diesel (L) 47.8 [29]
Electricity (kWh) 11.93 [26]
Chemical fertilizer (kg)
Nitrogen (N) 78.1 [30]
Phosphate (P2O5) 17.4 [30]
Potassium (K2O) 13.7 [30]
Farmyard manure (kg) 0.3 [54]
2. Output (kg)
Lentil 14.7 [16]
Other inputs including diesel fuel, human labor, electricity, seed, chemicals, FYM and chemical fertilizers used throughout lentil production were multiplied by their corresponding energy equivalents (Table 1) to calculate their relevant energy consumptions in unit of MJ ha-1.
The energy balance in agricultural crop production is expressed in terms of some energy indices including the energy ratio (ER), energy productivity (EP), specific energy (SE) and net energy gain (NEG). Implementing energy balance of agricultural products can be a very useful tool for decision makers to compare and analyze various alternative products with lentil in the study area. Based upon the energy taken from the inputs vs that derived from output, ER (which is indicative of the energy use efficiency defined as the ratio of output energy to input energies), EP, SE and NEG were calculated as follows [29]:
output Energy (MJ ha 1) Input Energy(MJ ha-1)
lentil out put(Kg ha-1) Energy input(MJ ha-1)
Se Energy input(MJ ha-1) lentil output(Kg ha-1)
NEG = Output Energy (MJ ha-1) - Input Energy (MJ ha-1)
Based on the type of energy sources, energy demand in agriculture can be classified into direct (DE) and indirect (IDE), renewable (RE), and non-renewable (NRE) energies. DE
is used directly in agriculture comes from a fossil origin such as diesel fuel, gasoline, liquid petroleum gas, coal and from electricity. IDE refers to the energy used to produce equipment and other materials that are used on the farm. The major IDE is contributed to chemical fertilizers, machinery and water used in irrigation [29]. In this study, DE includes energy derived from human labor, diesel fuel, water used in irrigation and the electricity to power irrigation pumps while INE covers energy that is embodied in seeds, FYM, machinery, chemical fertilizers and chemicals.
RE and NRE are other forms of energy. RE is used to describe energy sources that are replenished by natural processes on a sufficiently rapid time-scale. Thus RE can be used by humans more or less indefinitely, provided the quantity taken per unit of time is not too great. On the other hand, NRE term is used to describe energy sources that exist in a limited amount on earth [30]. In this study, RE sources consist of human labor, seeds, water used up in irrigation and FYM, while NRE in the production of the crop is resulted from the use of diesel fuel, chemicals, electricity, chemical fertilizers and machinery.
2.3. Life cycle assessment methodology
LCA of any product is performed based on the cradle to grave approach, i.e., from production of input materials using raw materials to the produced lentil in the farms. This means that the whole process of production is analyzed by considering all inputs (raw materials and energy consumption) and their interactions [31]. LCA specifies the environmental impacts considering all materials emitted into air, soil and water cause environmental burdens [32]. Based on ISO 14040, every LCA methodology consists of four stages i.e. goal and scope definition, inventory analysis of materials or processes, environmental impact assessment and interpretation of the results [33].
2.3.1. Goal and scope definition
Goal and scope definition is the first stage in an LCA study. It defines the purpose of the study, describes the functional unit and expected product of the study, the product system and its boundaries, the approach of data collection and its processing and finally the considered environmental impact categories. To achieve a sharper understanding of the goal in LCA studies, the boundaries of the system must be clearly defined. Therefore all operations which contribute to the life cycle of the product, process, or activity of interest are considered within the system boundaries [34]. In this study, the total inputs from the cradle (i.e., production of machinery, fertilizer and pesticide from raw materials) to the farm gate (harvested lentil) was considered as system boundary (Fig. 1). Determining the functional unit (FU) in LCA is a key concept that makes it possible to compare different products in a unique scale [33]. In agricultural systems, generally two functional units are considered, namely the mass-based and land-based. The mass-based FU deals with the unit of mass of a product, e.g. ton or kg of dry material, and land based FU is concerned with the unit of cultivated area, i.e. one cultivated hectare per year [35]. Based on the relatively equal farm size about one hectare and lentil yield in the studied area, considering FU as one cul-
Fig. 1 - The farm gate as system boundary of lentil production.
tivated hectare is useful in analysis of the farms. Therefore, these two FUs were considered simultaneously in this study in order to be able to effectively clarify the environmental performance of LCA of lentil production [19].
2.3.2. Life cycle inventory (LCI)
The inventory analysis corresponds to all resources required for lentil production and all the emissions generated from the production process considering the specific FU. In this study, environmental emissions of lentil production were divided into two parts. The first part called indirect emissions refers to environmental impacts of inputs during their production phase. The second part encompassed the direct emissions associated with consumption of inputs in lentil production as presented in Table 2. Direct emissions were due to the diesel fuel consumption, application of fertilizers (chemical fertilizers and FYM) and use of chemicals adopted from literature, environmental reports and EcoInvent database center [36,37], which is explained in the following.
The application of chemical fertilizers resulted in direct emissions including emissions of ammonia, nitrogen monoxide and nitrogen oxides into the air and nitrate leaching to groundwater. Several methodologies have been introduced to estimate direct emissions of chemical fertilizers but
EMEP/EAA guidelines from the European Environmental Agency [38] and IPCC guidelines [39] are the most relevant ones.
Crop production in the study region is extensively related to the application of nitrogen fertilizers. Based on IPCC guidelines [39], by application of 100 kg of nitrogen fertilizers, 1.25 kg of N2O is emitted into the air. Also, Galloway et al. [40] reported that 2% of the total nitrogen fertilizer is emitted in the form of NOx and likewise, 8% of the total nitrogen applied is emitted in the form of NH3. In addition, it was assumed that 30% of nitrogen fertilizers in the form of nitrate (NO-) are leaching from soil into the groundwater [41]. The use of phosphorus (P) fertilizers resulted in emissions to soil and water. Phosphate (P2O5) emissions in the form of phosphorus is calculated through an equilibrium, in which seed and fertilizers are inputs and lentil and accumulated phosphorus in the soil are considered as outputs. About 2.9% of the total phosphorus fertilizers in the soil leaches from soil profile in the form of phosphate. The average amount of phosphorus leached to groundwater was considered as 0.22 kg P-based fertilizers per ha [42]. Pesticides may contain either a single or a combination of two or more active ingredients. Throughout the present study, herbicides and insecticides were considered as a single input referred to as
Table 2 - Life cycle inventory data for lentil production.
Inputs Units Average Max Min SD
Seed kg 74.78 90 60 6.37
Chemical fertilizers
Nitrogen (N) kg 134.92 200 100 27.41
Phosphate (P2O5) kg 131.35 180 100 23.81
Potassium (K2O) kg 75.28 150 0 42.35
FYM kg 892.75 5000 0 1921.84
Herbicide kg 2.07 3.26 1.39 0.47
insecticide kg 3.05 4.66 1.99 0.77
Machinery kg 5752.96 7320 4430 895.27
Diesel fuel L 108.39 140 80 611.7
Labor h 201.10 245 170 29.69
Water for irrigation m3 366.82 350 260 132.55
Electricity kWh 565 800 450 107.07
"pesticides". Van den Berg and Ashmore [43] have estimated that 30-50% of applied pesticides in agricultural crop spraying are emitted into the air due to spray drift and volatilization.
Within the lentil production, diesel fuel was used up by tractors in different farm operations. Direct emissions from combustion of diesel fuel into air in farm operations was calculated by multiplying the amount of consumed energy of diesel fuel per hectare by the emission factors based on Ecoinvent database due to its completeness. The values of various emission factors applied in this study derived from data given by Nemecek and Kagi [44] are presented in Table 3. Accordingly, all of the emissions from diesel fuel combustion can be released into air which is obtained by multiplying the emission factors by the amount of consumed energy from diesel fuel per hectare.
2.3.3. Life cycle impact assessment
Life cycle impact assessment as the third LCA step investigates the environmental impacts associated with emissions and consumption of resources in a production system. This
Table 3 - Emission factors for 1 MJ energy production from diesel fuel based on EcoInvent.
Emission Amount (g/MJ diesel)
Carbon dioxide (CO2) 74.5
Sulfur dioxide (SO2) 2.41E- 02
Methane (CH4) 3.08E- 03
Benzene 1.74E- 04
Cadmium (Cd) 2.39E- 07
Chromium (Cr) 1.19E- 06
Copper (Cu) 4.06E 05
Dinitrogen monoxide (N2O) 2.86E 03
Nickel (Ni) 1.67E 06
Zinc (Zn) 2.39E 05
Benzo(a)pyrene 7.16E 07
Ammonia (NH3) 4.77E 04
Selenium (Se) 2.39E 07
PAH (poly cyclic hydrocarbons) 7.85E 05
Hydro carbons (HC, as NMVOC) 6.80E 02
Nitrogen oxides (NOx) 1.06
Carbon monoxide (CO) 1.50E 01
Particulates (<2.5 im) 1.07E 01
step consists of a number of compulsory vs voluntary steps. The compulsory steps involve translating the inventory data of input materials and production processes into their contributions to a number of specified environmental impact categories (impact characterization). The voluntary steps are traditionally directed at evaluating the results of impact categories while considering each other (normalization) [45]. Literature review indicated that, CML 2 baseline 2000 V2.05/world 1997/characterization method developed by Leiden University is commonly used in LCA studies of agricultural products. Additionally, application of this method had been the most frequent approach to analysis of the life cycle in the production systems [46]. To perform impact assessment, CML 2 baseline 2000 V2/world method and its ten environmental impact categories were applied in this study. The selected impact categories were eutrophication potential (EUP), abiotic depletion potential (ADP), acidification potential (AP), human toxicity potential (HTP), global warming potential (GWP), freshwater aquatic ecotoxicity potential (FAEP), marine aquatic ecotoxicity potential (MAEP), terrestrial eco-toxicity potential (TEP), photochemical oxidation potential (PHOP) and ozone layer depletion potential (OLDP). The measurement units for these impact categories can be found in Table 4. The prevalence of the selected impact categories was observed in most of the studies [8,10,11].
The index for each impact category is calculated using Eq. (8) [47] as follows:
ICIi = ^[(Ej or Rj) x CFj (8)
where ICIi is indicator value per functional unit for impact category i; Ej or Rj is the emission of j mixture or the consumption of j resource on each functional unit; CFj is the characterization factor for j mixture in impact category i. The characterization factor in each impact category shows the mixture potential for creating the impact.
The LCA analysis was conducted using SimaPro V8.03 software as one of the most common LCA software for analysis of the environmental burdens of a product through its life cycle.
2.3.4. Interpretation of the LCA results
In the fourth stage of the LCA, all the results will be analyzed
in order to investigate the environmental conditions resulted
Table 4 - Environmental indices categories and measurement units for each category.
Impact categories Nomenclature Measurement units
Abiotic depletion potential ADP kg Sb eq.
Acidification potential ACP kg SO2 eq.
Eutrophication potential EUP kg PO|- eq.
Global warming potential3 GWP kg CO2 eq.
Ozone layer depletion potential OLDP kg CFC-11 eq.
Human toxicity potential® HTP kg 1,4-DCB eqb
Freshwater aquatic ecotoxicity potential FAEP kg 1,4-DCB eqb
Marine aquatic ecotoxicity potential MAEP kg 1,4-DCB eqb
Terrestrial ecotoxicity potential8 TEP kg 1,4-DCB eqb
Photochemical oxidation PHOP kg C2H4 eq.
a Considering 100 years. b DCB = dichlorobenzene.
from production system and provide solutions. LCA results obtained in this study will be discussed in the results section.
2.4. Development of ANN models
Selection of the appropriate inputs parameters of the ANN model is the key step of model development. Nine input energies including human labor, diesel fuel, machinery, chemical fertilizers, chemicals, FYM, electricity, water for irrigation and seeds were considered as inputs to the ANN model, while eleven output parameters, i.e. lentil yield and ten environmental impact categories were considered as model outputs. To ensure the suitability of this selection, the relationship between dependent variables (input energies) and independent variables (outputs of the model) was analyzed statistically by SPSS software. Based on the evaluation results, there were significant correlations between inputs and outputs while the correlations between inputs were not statistically significant.
ANN models are excellent nonlinear modeling tools which can efficiently find the existing deterministic relation between input and output variables by composition of activation functions and weights. In this study, several feed-forward back-propagation neural networks with an input layer, one or more hidden layers and a single layer of output neurons were evaluated and trained using the collected data. In this study, the sigmoid and linear transfer functions were respectively applied for the hidden layers and the output layer. LM training algorithm as one of the most common learning rules in ANN was used for network training. In feed-forward back-propagation neural network, the information flows only in the forward direction, from inputs to outputs. The input vector is directly passed to the node activation output of input layer without any computation. The hidden layer with sig-moid activation function performs intermediate computations. Then, the linear output layer generates the network output. Neurons of the hidden layer with suitable nonlinear transfer functions are applied to process the information by the input nodes received [48].
In this study, the output of the network is given by following equation [49]:
where 'n' is the number of hidden nodes, 'm' is the number of input nodes and 'f stands for a transfer function, i.e., sig-moid function in this study which is defined as
f (x) - 1+exp(-x). Also, {bj, i = 1,2,..., m; j = 0,1,.., n} are weights from the input to the hidden nodes, while, the vectors of weights from the hidden to the output nodes are represented as {aj,j = 0,1,...,n}, moreover 'a0' and 'bj denote the weights of arcs leading from the bias terms, which are of values always equal to 1.
Basic information on inputs and outputs in lentil farms in the form of input and output matrixes was entered into Mat-lab V7.14 (R2012a) software package to perform ANN analysis. MATLAB software was used to train and test the developed ANNs on a personal computer. The input and output data sets are matrixes composed of vectors specific to each farm. The input vector includes nine input energies while output vector covers eleven output parameters, i.e., yield and ten impact categories in any farm. Moreover, inputs and outputs data were normalized in the range of 0-1 and then returned to original values after the simulation. In this study, data collected from 80, 20 and 40 lentil farmers were respectively used for training, cross validation and testing of the developed ANN models.
For the development of ANN models, several networks were built up and tested using the experimental data to determine the most appropriate ANN arrangement for predicting the output parameters. In this research, 80, 20 and 40 units were respectively used for training, cross validation and testing of ANN models. Accordingly, the most acceptable topology was identified by the highest R2 value vs the lowest RMSE as well as MAPE values.
To assess the performance of the developed ANN models for estimating the desired output in lentil production, some statistical quality parameters including mean absolute percentage error (MAPE), root mean square error (RMSE) and coefficient of determination (R2) were employed as follows [18,21]:
MAPE(%)^£|(ti - z0'
yt = ^0+J2af b-yt-i+boj + et
RMSE Jti - zi)2
3. Results and discussion
3.1. Analysis of input-output energy use in lentil production
Various energy inputs used in the production of lentil and their percentage share from the total energy inputs are given in Table 5. The average of total input energy for cultivating one hectare of lentil and the energy output calculated were 32,970.10 MJ ha-1 and 29,746.50, respectively. Also, a detailed description of the share of each input energy to the total input energy is shown in Fig. 2. Based on the results, the energy related to chemical fertilizers amounting to 13,855 MJ ha-1 contributed the highest share (42.76%) from the total energy input in lentil production within the region. Energy contribution related to chemical fertilizers N, K2O and P2O5 amounted to 76.05%, 16.47% and 7.48% of total energy of chemical fertilizers, respectively. Following chemical fertilizers, the parameters of electricity, diesel fuel and irrigation water were the main energy consuming inputs with values of 20.92%, 15.99% and 12.21%, respectively in lentil production. The total share of all five input energies related to seed, chemicals, machinery, human labor and FYM was 8.12% of total input energy.
Koocheki et al. [16] reported that diesel fuel energy made up 24.36% of total energy, followed by irrigation water (18.79%), chemical fertilizers (18.52%) and electricity (13.27%) for lentil production in Khorasan Razavi province of Iran. As mentioned before, energy and environmental analyses in pro-
Table 5 - Amounts of energy inputs and output in lentil production.
Inputs/output Min (MJ/ha) Max (MJ/ha) SD Average (MJ/ha) Percentage (%)
A. Inputs
1. Human labor 333.2 480.2 29.69 394.17 1.21
2. Machinery 246.33 1144.53 199.45 631.7 1.94
(a) Tractor 66.59 517.03 102.38 236.37 0.72
(b) Plow 50 105 9.80 66.85 0.20
(c) Disk 20 85 14.19 47.46 0.14
(d) Boundaries 0 14.4 3.62 7.38 0.02
(e) Leveler 0 6.5 2.38 2.72 0.00
(f) Planter 70 150 18.52 106.41 0.32
(g) Sprayer 0 135 31.26 74.53 0.23
(h) Rotary Hoes 0 75 19.26 45.91 0.14
(i) Thrashing
10 32.8 4.68 18.17 0.05
3. Diesel fuel 3824 6692 611.70 5181.17 15.99
4. Chemical Fertilizers 9550 20,807 2983.84 13854.87 42.76
(a) Nitrogen 7810 15,620 2141.41 10537.92 32.52
(b) Phosphorus (P2O5) 1740 3132 414.40 2285.61 7.05
(c) Potassium (K2O) 0 2055 580.22 1031.34 3.18
5. Farmyard manure (FYM) 0 1500 576.55 267.85 0.82
6.Chemical 535 1250 180.51 803.75 2.48
(a) Herbicide 333 778 112.87 494.37 1.52
(b) (b) Insecticide 202 472 78.19 309.38 0.96
7. Water for irrigation 2849.73 4964.47 484.52 3957.21 12.21
8. Seed 882 1323 93.66 1099.35 3.39
9.Electricity 5400 9600 1284.84 6780 20.92
Total energy input 23713.13 45901.91 5901.19 32970.10 100
B. Output
Total energy output 26,460 33,075 1751.00 29746.50 100
sr=1(ti - Zi)2
where 'ti' and 'zi' are respectively the actual and predicted values by the ANN model and 'n' denotes the total number of data. Relative percentage deviation between the predicted and measured values was evaluated by MAPE. The MAPE value smaller than 10% was considered to be the acceptable value. The smaller the values of MAPE and RMSE are, the better performance of the ANN model is achieved. The coefficient of determination was used to determine that how well the model approximates the real data points; That is, a model acts more efficiently and accurately when R2 values are closer to unity.
Sensitivity analysis investigates the influence of input parameters of the model on the model outputs. It can rank and specify the influential input parameters on yield and environmental emissions. To analyze the sensitivity of energy inputs on yield and ten environmental impact categories in lentil cultivation, sensitivity analysis via ANN was conducted using the NeuroSolutions V5.07 software package [50]. In this study, the sensitivity analysis reveals clearly the contribution of input parameters of the best ANN model on the desired outputs, i.e., lentil yield and assessed environmental impact categories. By considering the analysis, it becomes evident that the analysis is of a great assistance in making it feasible to judge what parameters should be considered as the most significant vs the least significant ones during the generation a satisfactory model [21].
Fig. 2 - The share of total mean energy inputs in lentil production.
duction of crops from legume family are very sparse, therefore the energy use profile for production of some agricultural products were assessed. Many studies presented similar results and revealed that chemical fertilizers and diesel fuel are the most energy consuming inputs in production of agricultural products [19,21,26,51,52]. Excessive use of chemical fertilizers in agricultural systems generates such environmental burdens as nitrogen loading and carbon emissions in the environment causing degradation of water quality [53]. In Iran, the primary fuel source for electricity generation is fossil fuels, and since the electric power transmission system is outdated; thus, the efficiencies in electricity production and transmission are low. Also, use of old and inefficient agricultural tractors and implements in field operations increases the diesel fuel consumption. For time management and economy in fuel consumption, it is essential that the machinery and equipment work at their highest field capacities.
Table 6 presents the values of lentil yield (kg ha-1) and values of energy indices of ER, EP, SE and NEG for lentil production. Also, total input energy consumed in different forms as DE, IDE, RE and NRE (MJ ha-1) are given in this table. The average value of ER index in lentil cultivation was calculated on 0.902 indicating that energy use in lentil production is virtually inefficient in the study region. Other researchers have reported similar results i.e. 0.72 for lentil by considering only lentil in calculation of output energy [16], 0.96 for cherry [25], 1.16 for apple [54] and finally 1.1 for potato [55]. The average yield in the study region was calculated as 2,023.57 kg ha-1
while the average yield of 696.6 kg ha-1 in Khorasan Razavi province of Iran was reported by Koocheki et al. [16]. This comparison shows that the lentil yield in Esfahan province is relatively high with respect to other regions of the country. The average values of energy indices i.e. SE, NEG and EP in lentil production were calculated as 16.82 MJ kg-1, -3,223.61 MJ ha-1 and 0.06 kgMJ-1, respectively. NEG is negative, therefore it could be concluded that, in lentil production, energy is being lost.
The total energy use in the form of DE and IDE were calculated as 16,312.55 (49.47%) and 16,657.54 (50.53%), respectively. It is clear that DE and IDE have same contribution in input energy of lentil cultivation. The share of RE was 17.34% (5718.58 MJ ha-1) while that of NRE form was 82.66% (27,251.52 MJ ha-1), respectively. It is clear from Table 6 that in comparison with RE, the contribution of NRE is higher, thus lentil production is most dependent on NRE sources (such as chemical fertilizers and fossil fuels). Several researchers presented similar results that the contribution of NRE was higher than that of RE for different agricultural products [16,23,53,56].
3.2. Interpretation ofLCA results in lentil production
On the basis of the models presented by SimaPro software, more than 1600 emissions from raw materials were generated including emissions emitted into air, soil and water. Accordingly, a part of inventory emissions to air, soil and water asso-
Table 6 - Lentil yield, energy indices and different form of energy in lentil production.
Items Unit Min Max Average SD
Yield kg ha-1 1800.00 2250.00 2023.57 119.11
Energy use efficiency - 0.70 1.11 0.902 0.10
Specific energy MJ kg-1 13.18 20.86 16.29 2.00
Energy productivity kgMJ-1 0.04 0.07 0.06 0.007
Net energy MJ ha-1 -13561.90 2722.86 -3223.60 4296.49
Direct energy MJ ha-1 12426.54 21511.02 16312.55(49.47%) 2264.72
Indirect energy MJ ha-1 11221.01 25747.1 16657.54(50.53%) 3727.22
Renewable energy MJ ha-1 4084.53 7929.31 5718.58(17.34%) 981.19
Non-renewable energy MJ ha-1 19563.01 37972.70 27251.52(82.66%) 5043.70
Table 7 - Some environmental emissions of lentil production per hectare.
Type of emissions Emission source Unit Amount (Unit ha 1
A. To air
1. Carbon dioxide (CO2) Diesel fuel g 385997.1650
2. Sulfur dioxide (SO2) Diesel fuel g 124.8661
3. Methane (CH4) Diesel fuel g 15.9580
4. Benzene (C6H6) Diesel fuel g 0.9015
5. Di nitrogen monoxide (N2O) Diesel fuel g 14.8182
6. Ammonia (NH3) Diesel fuel g 2.4714
7. Hydrocarbons (HC, as NMVOC) Diesel fuel g 352.3195
8. Nitrogen oxides (NOx) Diesel fuel g 5492.0402
9. Carbon monoxide (CO) Diesel fuel g 777.1755
10. Particulates (<2.5 mm) Diesel fuel g 544.3851
11. Di nitrogen monoxide (N2O) Fertilizer kg 1.686
12. Nitrogen oxides (NOx) Fertilizer kg 2.698
13. Ammonia (NH3) Fertilizer kg 10.793
B. To soli
1. Pesticide pesticide kg 2.048
2. Nitrate (NO3) Fertilizer lg 81.1108
3. Cadmium (Cd) Fertilizer mg 62.6834
4. Cobalt (Co) Fertilizer mg 3.3794
5. Zinc (Zn) Fertilizer mg 187.1091
6. Lead(Pb) Fertilizer mg 35.8289
C. To water
1. Nitrate(NO3) Fertilizer kg 40.48
2. Phosphorus Fertilizer kg 29.68
Table 8 - Life cycle impact impacts per two distinctive FUs.
Impact category Mass based FU: 1 ton Land based FU: 1 ha
Abiotic depletion 17.98 36.38
Acidification 81.97 165.57
Eutrophication 2.12 4.28
Global warming (GWP100) 3593.72 7259.31
Ozone layer depletion (ODP) 0.00037 0.00074
Human toxicity 2289.75 4625.29
Fresh water aquatic ecotoxicity 62.53 126.31
Marine aquatic ecotoxicity 230051.2 464703.42
Terrestrial ecotoxicity 12.61 25.47
Photochemical oxidation 7.21 14.56
ciated with inputs used in lentil production are tabulated in Table 7. As shown, the emission values related to CO2, SO2, CH4, N2O and CO were determined as 385997, 124, 15, 14 and 777 g ha-1, respectively. The type of fertilizer is the main determinant of emissions at all the farm levels. N2O, NOx and NH3 emitted by the fertilizers at 1.68, 2.69 and 10.79 kg ha-1 significantly and negatively affect the air in the studied region (Table 7). Nemecek et al. [57] demonstrated that, N2O and CO2 emissions from chemical fertilizers made high contributions to GWP. Emissions from pesticide were assumed to end up in the agricultural soils, thus, pesticide emission to soil was tested and found out to be 2.048 kg ha-1. Elements, such as NO3, Cd and Pb that are released from fertilizers, affect both water and soil (Table 7).
The values of environmental impact categories on the basis of the mass based and land based FUs in lentil cultivation are presented in Table 8. The values of environmental
impact categories related to one ha of lentil cultivation were approximately two times the relevant impact categories for one ton of produced lentil. This is due to the fact that the yield of lentil is approximately 2 tons per ha. Based upon the obtained results, GWP was estimated at 4284.87 kg CO2 eq. t-1. Considering the lack of availability of similar research on lentil production in the literature, the results are compared with those of other agricultural crops produced. In a study in Chile, GWP for sunflower and rapeseed productions were estimated about 890 and 820 kg CO2 eq. t-1, respectively [58]. Bartzas et al. [59] determined that production of barely in Spain and open field production of fresh lettuce in Italy created the total GWP impacts of 171 and 243 kg CO2 eq. t-1, respectively. Abeliotis et al. [15] reported that the calculated GHG emissions related to the production of three bean varieties in different cultivation methods varied in the range of 86-438 kg CO2 eq. per ton of product. Romero-Gamez et al. [8] demonstrated that GHG emission varied from 101 to 2890 kg CO2 eq per ton of bean. The highlighted that the use of both screen house and screen houses equipped with misting systems produced the high air emissions due to the manufacture of steel structures, the processing of concrete, and the manufacture of plastics that constituted these systems. It must be noted that in a similar cropping system to lentil, total GHG emissions of green bean cropping in the open field is 136 kg CO2 eq t-1 which is substantially lower than that of lentil production in the present study. This difference to some extent is due to the different moisture content of green bean and lentil. Overall comparisons shows that the impact categories in this study are different from other studies. This high difference can be interpreted by large application of such agri-
Fig. 3 - Percentage contribution of inputs and processes per environmental impact categories in lentil cultivation.
cultural inputs as fertilizers and direct emissions in lentil cultivation.
A percentage contribution of production processes and inputs involved in lentil farming to the selected impact categories is presented in Fig. 3. The production of agricultural machineries was the one that mostly contributed in the six impact categories, contributing for 67.78%, 62.63%, 58.62%, 42.14%, 36.93% and 27.27% to HTP, GWP, ADP, TEP, MAEP and FAEP respectively. In a similar study on evaluation of the environmental impacts as regards chickpea production in Iran, application of LCA revealed that GWP, ADP, HTP, MAEP and TEP were dominated by agriculture machinery [11]. In order to reduce the environmental burdens related to agricultural machineries, it will be necessary to increase the sizes of the farms by integration, to prevent farms shrinking when a farm is transferred to the next generation and to perform different agricultural operations with combined machineries such as combined equipment for plowing and seed bed preparation. Also in the impact categories of ACP and PHOP, the direct emissions from diesel fuel and chemical fertilizers associated with lentil cultivation were important among all input categories with the shares of 78.62% and 62.49%, respectively. Diesel fuel with the share of 62.21% had the highest environmental impact on OLDP followed by agricultural machinery with 17.16% contribution.
Marucci et al. [60] concluded that the environmental impact from the use of agrochemicals was greater in greenhouse crop production as compared with open-field farming; also, they showed that, on opposite trend existed in terms of herbicide use, with greater quantities applied in the open-field. The MAEP impact category was dominated by machinery and Nbased fertilizer while in FAEP, the use of machinery and FYM was important. In LCA of rose cultivation in Ethiopia by Sahle and Potting [45], the production of fertilizers was the main contributor to MAEP, HTP, ADP and TEP. The use of right amount of chemical fertilizers and FYM at different growth stages of lentil cultivation based on the soil testing results and expert's opinions will have a significant impact in reducing direct emissions associated with these inputs. Also, the use of compost produced from agricultural wastes for the fertilization of crops was investigated as a promising alternative waste management option [61]. Regarding the consumption of diesel fuel, the use of clean fuels such as biodiesel and bio-ethanol instead of fossil fuels not only will reduce the negative impacts to environment, but also will provide the higher energy use efficiency [29,62]. In terms of environmental burdens, irrigation water and K2O chemical fertilizer seemed the least impacting inputs approximately in all of the impact categories.
To better determine the relative magnitude of each impact category within the production of lentil, normalized values of
Fig. 4 - Normalized impact categories of lentil production.
impact categories were utilized. The normalized values of selected impact categories are presented in Fig. 4. Normalization is the calculation of the magnitude of the impact category results with respect to reference values where the different impact potentials and consumption of resources are expressed on a common scale through relating them to a common reference, in order to facilitate comparisons between impact categories. The normalized values of all impact categories are dimensionless, thus their comparison is more readily applicable [34,63]. The magnitude of ACP was significantly higher than that of other impact categories followed by MAEP and PHOP. Since ACP was dominated by direct emission resulted from the application of chemical fertilizers, FYM and diesel fuel, any savings made in consumption of diesel fuel and fertilizers would cause a reduction in the ACP impact category. The normalization in this study moves the attention toward reduction of ACP impact category and reduction of other impact categories at the same time.
ANN model development
Investigation of different ANN models revealed that the best fitted ANN model consisted of an input layer with nine input variables, two hidden layers of each ten and six neurons, respectively, and one output layer with eleven output variables, i.e., 9-10-6-11 structure. The statistical criteria of the best ANN model for predicting yield and environmental impact categories of lentil is tabulated in Table 9. According
to the statistical criteria of the developed ANN model, namely R2 values in the range of 0.8993-0.9956, MAPE in the range of 0.0003-0.2085%, RMSE related to impact categories in the range of 0.0574-0.1292 and RMSE related to yield about 0.1493 kg, it can be concluded that all the considered ANN model provide a very satisfactory prediction results. On the other hand, lentil yield and environmental impact categories predicated by the best ANN model tended to quite closely follow the corresponding actual ones. Accordingly, this model was identified as the most appropriate solution for estimating the lentil production yield and related environmental impact categories.
Table 9 - Network performance of lentil yield and environ-
mental prediction for the best topology.
Item R2 MAPE (%) RMSE
Yield 0.9039 0.0382 0.1493
ADP 0.9726 0.1209 0.1276
ACP 0.9823 0.0041 0.1145
EUP 0.9850 0.0197 0.1109
GWP 0.9834 0.0003 0.1024
OLDP 0.9956 0.0111 0.0574
HTP 0.8993 0.2085 0.1944
FAEP 0.9641 0.1781 0.1244
MAEP 0.9793 0.1042 0.1053
TEP 0.9844 0.0901 0.0948
PHO 0.9744 0.0659 0.1292
Fig. 5 - Comparison between measured and estimated values of yield and environmental impact categories of lentil production using best developed ANN model.
Table 10 - Sensitivity analysis results for input energies.
Sensitivity Yield ADP ACP EUP GWP OLDP HTP FAEP MAEP TEP PHO
Seed 5.150 0.139 0.218 0.051 17.003 0.00030 14.752 0.176 2323.010 0.122 0.021
Fertilizers 10.098 0.509 0.614 0.198 79.230 0.00070 71.333 5.755 9195.634 0.956 0.048
FYM 11.0800 2.452 4.454 1.012 446.166 0.00370 79.410 23.890 21660.367 3.464 0.316
Chemical 1.3035 0.588 0.975 0.221 86.709 0.00020 28.066 5.264 6259.28 0.851 0.075
Machinery 17.8013 1.528 1.692 0.407 119.936 0.00015 155.743 12.057 25316.397 2.243 0.181
Diesel fuel 11.2343 0.295 0.436 0.137 49.490 0.00060 54.276 0.439 6596.823 0.321 0.040
Labor 6.4066 0.266 0.292 0.084 46.796 0.00220 46.197 0.273 6146.356 0.269 0.036
Irrigation 4.4657 0.461 0.685 0.170 17.494 0.00250 44.454 2.967 7078.624 0.616 0.060
Electricity 0.9541 0.387 0.510 0.108 31.656 0.00001 26.755 2.812 5295.476 0.517 0.502
Pahlavan et al. [21] reported that an ANN model with 7-2020-1 structure was the best network for predicting basil production yield. Khoshnevisan et al. [10] demonstrated that an ANN model including an input layer with 11 neurons, two hidden layers with six neurons in the first hidden layer and ten neurons in the second hidden layer and an output layer with two neurons was the best network for estimating the total yield and GWP in the strawberry production system. Their results revealed that the obtained structure can predict the desired outputs with high accuracy. Khoshnevisan et al. [19] for predicting the output energy and GWP of potato production, applied the best fitted ANN model consisted of an input layer with twelve inputs, one hidden layer with eight neurons and an output layer with two output variables, i.e., 12-8-2 ANN structure. This network had the least MAPE for output energy and GWP and the highest R2 and the least RMSE for GWP. In an ANN model developed by Taghavifar and Mar-dani [23] the best network was the 8-16-2 structure. The R2 values of 0.9879 and 0.9827 were obtained for yield and GWP prediction of apple production in Iran, respectively. Nabavi-Pelesaraei et al. [20] predicted energy use and GWP of kiwi-
fruit production using an ANN model with 12-9-9-2 structure. The R2 values of the best network were calculated as 0.987 and 0.992 for yield and GHG emissions, respectively, demonstrating the high accuracy of the model. Nabavi-Pelesaraei et al. [22] predicted yield and GWP of watermelon production using ANNs. They reported that selected ANN model was of the potential of predicting yield and GWP by respective coefficients of determination of 0.96 and 0.99.
Fig. 5 demonstrates the scatter plots of the predicted yield and environmental impact categories versus actual values for the training and testing data sets. The predicted and actual values were found out in good agreement with each other. Coefficients of determination for these indices demonstrated the potential capability of the developed ANN model for prediction of yield and environmental impacts in lentil production in the studied area.
3.4. Sensitivity analysis
Considering the best selected ANN model, a sensitivity analysis was performed to assess the prediction validity and capa-
bility of the developed models. The results of the sensitivity analysis are given in Table 10. The sensitivity values of the most effective input parameter on each output parameter are shown in bold type. As clearly shown, machinery related energy had the highest effect on lentil yield with sensitivity value equal to 17.80, followed by diesel fuel and FYM energies. It was also found that the sensitivity concerning electricity on lentil yield was the lowest among all inputs. In the case of environmental impact categories, all the indices except HTP and MAEP were discerned as sensitive to the FYM energy. Furthermore, the highest sensitivity was determined for agricultural machinery for both impact categories of HTP and MAEP.
4. Conclusions
The total input energy and output energy in lentil production were calculated as 32,970.10 and 29,476.50 MJ ha-1, respectively. On the average, the share of DE was 49.47% of total energy input expended in lentil production, while the contribution of IDE being 50.53%. The share of input as RE and NRE energies were recorded as 17.34% and 82.66%, respectively. Chemical fertilizers (42.76%), electricity (20.92%) and diesel fuel (15.99%) demonstrated their pivotal roles in total energy consumption. The high contribution of chemical fertilizers energy in total energy consumption (42.76%) revealed the high potential for reducing fertilizer application. The most significant impact categories are related to agricultural machinery employed in seedbed preparation and in sowing operations. Therefore, an application of either no-tillage or reduced tillage systems could reduce the use of machinery, thus diminishing some of these impacts. The direct emissions in lentil cultivation resulted from high application of chemical fertilizers and diesel fuel contribute considerably to some environmental impacts, notably ACP and PHOP. Also, diesel fuel would considerably dominate in OLDP. Therefore, it suggested establishing a sustainable and environmental friendly lentil production system in the region with application of alternatives such as no-till and reduced tillage systems, use of clean fuels instead of fossil fuels and more efficient fertilizers application by integrated nutrient management. The ANN model with 9-10-6-11 structure was determined as the most appropriate model for predicting the lentil yield and its related environmental impacts.
Acknowledgment
The authors would like to acknowledge the University of Tehran for providing financial support for this research.
REFERENCES
[1] Roy F, Boye JI, Simpson BK. Bioactive proteins and peptides in pulse crops: pea, chickpea and lentil. Food Res Int 2010;43 (2):432-42.
[2] Sravanthi B, Jayas DS, Alagusundaram K, Chelladurai V, White ND. Effect of storage conditions on red lentils. J Stored Prod Res 2013;53:48-53.
[3] MAJ. Annual agricultural statistics. Ministry of Jihad-e-Agriculture of Iran, <www.maj.ir> 2013.
[4] FAO F. Agriculture organization. Poultry Sector Ghana; 2014.
[5] Guinee J. Handbook on life cycle assessment—operational guide to the ISO standards. Int J LCA 2001;6(5):255.
[6] WRI (World Resources Institute). 2014. Climate Analysis Indicators Tool (CAIT) 2.0: WRI's climate data explorer. <http://cait.wri.org>.
[7] Mourad AL, Coltro L, Oliveira PA, Kletecke RM, Baddini JP. A simple methodology for elaborating the life cycle inventory of agricultural products. Int J LCA 2007;12(6):408-13.
[8] Romero-Gamez M, Suarez-Rey EM, Anton A, Castilla N, Soriano T. Environmental impact of screenhouse and open-field cultivation using a life cycle analysis: the case study of green bean production. J Clean Prod 2012;28:63-9.
[9] Martínez-Blanco J, Muñoz P, Anton A, Rieradevall J. Assessment of tomato Mediterranean production in open-field and standard multi-tunnel greenhouse, with compost or mineral fertilizers, from an agricultural and environmental standpoint. J Clean Prod 2001;19(9):985-97.
[10] Khoshnevisan B, Rafiee S, Mousazadeh H. Environmental impact assessment of open field and greenhouse strawberry production. Eur J Agron 2013;50:29-37.
[11] Elhami B, Akram A, Khanali M. Optimization of energy consumption and environmental impacts of chickpea production using data envelopment analysis (DEA) and multi objective genetic algorithm (MOGA) approaches. IPA 2016;3 (3):190-205.
[12] Nikkhah A, Emadi B, Soltanali H, Firouzi S, Rosentrater KA, Allahyari MS. Integration of Life Cycle Assessment and Cobb-Douglas modeling for the environmental assessment of kiwifruit in Iran. J Clean Prod 2016;137:843-9.
[13] Nikkhah A, Emadi B, Taheri-Rad A, Khorramdel A. Environmental impacts of peanut production system using life cycle assessment methodology. J Clean Prod 2015;92:84-90.
[14] Soltanali H, Emadi B, Rohani A, Khojastehpour M, Nikkhah A. Life cycle assessment modeling of milk production in Iran. IPA 2015;2:101-8.
[15] Abeliotis K, Detsis V, Pappia C. Life cycle assessment of bean production in the Prespa National Park. Greece. J Clean Prod 2013;41:89-96.
[16] Koocheki A, Ghorbani R, Mondani F, Alizade Y, Moradi R. Pulses production systems in term of energy use efficiency and economical analysis in Iran. IJEEP 2011;1(4):95-106.
[17] Ermis K, Midilli A, Dincer I, Rosen MA. Artificial neural network analysis of world green energy use. Energy Policy 2007;35(3):1731-43.
[18] Safa M, Samarasinghe S. Determination and modeling of energy consumption in wheat production using neural networks: a case study in Canterbury province, New Zealand. Energy 2011;36(8):5140-7.
[19] Khoshnevisan B, Rafiee S, Omid M, Mousazadeh H, Rajaeifar MA. Application of artificial neural networks for prediction of output energy and GHG emissions in potato production in Iran. Agric Syst 2014;123:120-7.
[20] Nabavi-Pelesaraei A, Rafiee S, Hosseinzadeh-Bandbafha H, Shamshirband S. Modeling energy consumption and greenhouse gas emissions for kiwifruit production using artificial neural networks. J Clean Prod 2016;133:924-31.
[21] Pahlavan R, Omid M, Akram A. Energy input-output analysis and application of artificial neural networks for predicting greenhouse basil production. Energy 2012;37(1):171-6.
[22] Nabavi-Pelesaraei A, Abdi R, Rafiee S. Neural network modeling of energy use and greenhouse gas emissions of watermelon production systems. J Saudi Soc Agric Sci 2016;15:38-47.
[23] Taghavifar H, Mardani A. Prognostication of energy consumption and greenhouse gas (GHG) emissions analysis of apple production in West Azarbayjan of Iran using Artificial Neural Network. J Clean Prod 2015;87:159-67.
[24] Statistical Yearbook of Esfahan province in Iran, amar.org.ir/ English/Iran-Statistical-Yearbook, 2013.
[25] Kizilaslan H. Input-output energy analysis of cherries production in Tokat Province of Turkey. Appl Energy 2009;86 (7):1354-8.
[26] Ghasemi-Mobtaker H, Keyhani A, Mohammadi A, Rafiee S, Akram A. Sensitivity analysis of energy inputs for barley production in Hamedan Province of Iran. Agric Ecosyst Environ 2010;137(3):367-72.
[27] Beheshti-Tabar I, Keyhani A, Rafiee S. Energy balance in Iran's agronomy (1990-2006). Renew Sustain Energy Rev 2010;14 (2):849-55.
[28] Gezer I, Acaroglu M, Haciseferogullari H. Use of energy and labour in apricot agriculture in Turkey. Biomass Bioenergy 2003 Mar 31;24(3):215-9.
[29] Kitani, O. Energy and biomass engineering. In: CIGR handbook of agricultural engineering. st. joseph michigan: ASAE, V5. 1999; p. 330 (<www.cigr.org/documents>).
[30] Pishgar-Komleh SH, Keyhani A, Mostofi-Sarkari MR, Jafari A. Energy and economic analysis of different seed corn harvesting systems in Iran. Energy 2012;43(1):469-76.
[31] Lazzerini G, Lucchetti S, Nicese FP. Green House Gases (GHG) emissions from the ornamental plant nursery industry: a Life Cycle Assessment (LCA) approach in a nursery district in central Italy. J Clean Prod 2016;112:4022-30.
[32] McDougall F, White P, Franke M, Hindle P. Life cycle assessment in integrated solid waste management: a life cycle inventory. Malden of USA: Blackwell Science Publishing; 2001. p. 103-28.
[33] ISO 14040. Environmental management: life cycle assessment; principles and framework. 2nd ed. Geneva, Switzerland: International Organization for Standardization; 2006.
[34] Khoshnevisan B, Rafiee S, Omid M, Mousazadeh H, Clark S. Environmental impact assessment of tomato and cucumber cultivation in greenhouses using life cycle assessment and adaptive neuro-fuzzy inference system. J Clean Prod 2014;73:183-92.
[35] Gabel VM, Meier MS, Kopke U, Stolze M. The challenges of including impacts on biodiversity in agricultural life cycle assessments. J Environ Manage 2016;181:249-60.
[36] ISO 14041. Environmental management: life cycle assessment; principles and framework. 2nd ed. Geneva, Switzerland: International Organization for Standardization; 2006.
[37] Ecoinvent Center. Ecoinvent 2.0 database. Swiss centre for life cycle inventories, Du'bendorf. <http://www.ecoinvent. ch>; 2007.
[38] Umweltagentur E. EMEP/EEA air pollutant emission inventory guidebook. Technical guidance to prepare national emission inventories. InEEA Technical Report; 2013.
[39] Eggleston HS, Buendia L, Miwa K, Ngara T, Tanabe K. IPCC guidelines for national greenhouse gas inventories V2; 2006.
[40] Galloway JN, Schlesinger WH, Levy H, Michaels A, Schnoor JL. Nitrogen fixation: anthropogenic enhancement-environmental response. Global Biogeochem Cycles 1995;9 (2):235-52.
[41] EricksonJE, CisarJL, VolinJC, Snyder GH. Comparing nitrogen runoff and leaching between newly established St. Augustinegrass turf and an alternative residential landscape. Crop Sci 2001;41(6):1889-95.
[42] Dalgaard R, Schmidt J, Halberg N, Christensen P, Thrane M, Pengue WA. LCA of soybean meal. Int J LCA 2008;13(3): 240-54.
[43] VandenBerg LJL, Ashmore MR. Ecotoxicology: nitrogen. In: Jorgensen SE, Fath B, editors. Encyclopedia of
ecology. Netherlands: Elsevier; 2008. p. 267-75.
[44] Nemecek T, Kagi T. Life cycle inventories of agricultural production systems. Ecoinvent report No. 15 DUbendorf, CH: Swiss Centre for Life Cycle Inventories, 2007. (<www. EcoInvent.org/> documentation/reports).
[45] Sahle A, Potting J. Environmental life cycle assessment of Ethiopian rose cultivation. Sci Total Environ 2013;443:163-72.
[46] Anonymous. PRe Consultants. SimaPro 5 Database Manual 2003.
[47] Brentrup F, Küsters J, Kuhlmann H, Lammel J. Environmental impact assessment of agricultural production systems using the life cycle assessment methodology: I. Theoretical concept of a LCA method tailored to crop production. Eur J Agron 2004;20(3):247-64.
[48] Dreyfus G. Neural networks: methodology and applications. Springer Science and Business Media; 2005.
[49] Azadeh AG, Ghaderi SF, Sohrabkhani S. Annual electricity consumption forecasting by neural network in high energy consuming industrial sectors. Energy Convers Manage 2008;49(8):2272-8.
[50] Anonymous.NeurosolutionsV5.07 for excel, neurodimension. 2011. (<www.Neurosolutions.com>).
[51] Erdal G, Esengün K, Erdal H, Gündüz O. Energy use and economical analysis of sugar beet production in Tokat province of Turkey. Energy 2007;32(1):35-41.
[52] Khanali M, Movahedi M, Yousefi M, Jahangiri S, Khoshnevisan B. Investigating energy balance and carbon footprint in saffron cultivation-a case study in Iran. J Clean Prod 2016;115:162-71.
[53] Khan S, Khan MA, Hanjra MA, Mu J. Pathways to reduce the environmental footprints of water and energy inputs in food production. Food Policy 2009;34(2):141-9.
[54] Rafiee S, Avval SH, Mohammadi A. Modeling and sensitivity analysis of energy inputs for apple production in Iran. Energy 2010;35(8):3301-6.
[55] Hamedani SR, Shabani Z, Rafiee S. Energy inputs and crop yield relationship in potato production in Hamadan province of Iran. Energy 2011;36(5):2367-71.
[56] Unakitan G, Hurma H, Yilmaz F. An analysis of energy use efficiency of canola production in Turkey. Energy 2010;35 (9):3623-7.
[57] Nemecek T, Dubois D, Huguenin-Elie O, Gaillard G. Life cycle assessment of Swiss farming systems: I. Integrated and organic farming. Agric Syst 2011;104(3):217-32.
[58] Iriarte A, Rieradevall J, Gabarrell X. Life cycle assessment of sunflower and rapeseed as energy crops under Chilean conditions. J Clean Prod 2010;18(4):336-45.
[59] Bartzas G, Zaharaki D, Komnitsas K. Life cycle assessment of open field and greenhouse cultivation of lettuce and barley. IPA 2015;2(3-4):191-207.
[60] Marucci A, Pagniello B, Campiglia E, Rouphael Y, Colla G. Environmental impact of pesticides in vegetable crop production under the Mediterranean climate of central Italy. Greensys: In International Symposium on High Technology for Greenhouse System Management; 2007. p. 1583-90.
[61] Komnitsas K, Zaharaki D. Assessment of human and ecosystem risk due to agricultural waste compost application on soils: a review. Environ Foren 2014;15:312-28.
[62] Zangeneh M, Omid M, Akram A. A comparative study on energy use and cost analysis of potato production under different farming technologies in Hamadan province of Iran. Energy 2010;35:2927-33.
[63] Goedkoop M, Oele M, Schryver AD, Vieira M. SimaPro Database Manual Methods Library. Report Version 2.2; 2008 (<www.Pre-sustainability.com>).
