Relative Efficiency Analysis of Biomass Agricultural Plants using Data Envelopment Analysis

Renewable energy has recently been a promising interest as a substitute for fossil fuels due to an increasing energy demand as well as a rising concern over the environmental impact of fossil fuel consumption around the globe. Biofuel, in particular, is a type of renewable energy, which can be derived from various biomass types. In this research, we analyze relative efficiencies using Data Envelopment Analysis (DEA) technique from three types of energy-related plants in the Northeastern region of Thailand, which are cassava, sugarcane, and palm. The relative efficiency of each province is further analyzed during 2017 to 2019 for a comparative study. Next, the input criteria are collected including allowable planting area, labor cost, and rainfall amount; whereas the included output criterion is the quantity of harvested product. Our initial analysis using CCR, BBC, and Scale Efficiency (SE) models of DEA provides the baseline of efficient provinces to be benchmarked and directions for improving inefficient provinces, given desired input and output criteria in this study. Keyword. Relative Efficiency, Data Envelopment Analysis, Biomass, Renewable Energy


Introduction and Motivation
Renewable energy, such as biomass, solar, and wind has recently been a promising interest as a substitute for fossil fuels, such as oil and coal, due to an increasing energy demand as well as a rising concern over the environmental impact of fossil fuel consumption around the globe. Many countries have taken a variety of actions through strategic policies aiming at meeting energy needs more securely and sustainably. For example, the United States mandates to have more than 20 billion gallons of biofuel under the Energy Security Act by 2022. The European Union (EU) also aims to achieve 20% of energy from renewable sources by 2020. Also, China issues a long-term development plan of renewable energy aiming to increase the capacity of biomass power generation for 30 million Kilowatt (kW) by 2020 [1][2]. Thailand has also promoted a new economic model towards Industry 4.0 development plan by focusing on 10 targeted, S-curve industries -three of them are agricultural, logistics, and biofuel sectors [3].
Biomass, in particular, can be obtained from several sources including edible crops, non-edible crops, crop residues, forests, and waste. In comparison to fossil fuels, biomass is easy to grow and replace quickly without depleting natural resources. The advantages of using biomass are noted for its ability to be stored and used on demand, clean energy, renewable, and no carbon dioxide side effect. In addition, biomass also has the potential to reduce the dependency on fossil fuels, which are the main source of carbon dioxide release in the atmosphere [4][5][6][7].
Biofuel supply chain, in particular, involves a number of stakeholders, including farms providing feedstocks from biomass, pre-processing facilities, transshipment depots, bio-refinery plants, fuel-blending facilities, and demanding points of gas stations. Thus, biomass can be viewed as the upstream of the biofuel supply chain, in which the efficiency evaluation needs to be properly addressed. Fig. 1 illustrates the differences and similarities between traditional industrial and bioenergy supply chain.
In this research, we collect and analyze biomass data of three major feedstock for biofuel in the Northeastern region of Thailand. In particular, energy plants are collected for cassava, sugarcane, and palm during 2017 to 2019. Then, the relative efficiency of each province is further analyzed using Data Envelopment Analysis (DEA) for a comparative study. The input criteria are collected including allowable planting area, labor cost, and rainfall amount; whereas the included output criterion is the quantity of harvested product. Our initial analysis using CCR, BBC, and Scale Efficiency (SE) models of DEA provides the baseline of efficient provinces to be benchmarked and directions for improving inefficient provinces, given desired input and output criteria in this study. Renewable energy has attracted the attention of researchers around the globe for ensuring future energy security and sustainability. Biofuel energy, in particular, is one of the renewable energy that has gained ground in this regard. According to REN21 [5], biofuels employment attracted around 2 million jobs in 2018, in which most of these jobs are in the agricultural supply chain in developing countries, especially in the case of Southeast Asia, including Thailand. REN21 (2019) also estimates annual capacity and production of ethanol production in 2018 and finds that the top five countries are United States, Brazil, China, Canada, and Thailand, respectively. Besides, the top five countries for biodiesel production are United States, Brazil, Indonesia, Germany, and Argentina, respectively. Thus, Thailand also has a high potential to enhance its economics through bioethanol process.
In Thailand, the Department of Alternative Energy Development and Efficiency (DEDE [8]) plays a key role, in which a mission is to promote and support sustainable and worthy energy consumption and production for exporting and domestic use and to build collaborative network for bringing the country into the knowledge based society with sustainable economic stability and social well beings. Two key performancerelated projects noted are 1) developing communitybased biomass power plants and 2) developing the biomass potential database in Thailand.
In addition, the report by DEDE [9] for Research and development (R&D) studies of renewable energy in Thailand suggests that there are four groups of research studies going on in Thailand -1) the research on the potential of materials focusing on assessing the overall potential of biomass as raw materials; 2) research on biomass preparation process focusing on finding a way to improve the quality of biomasses, such as chipping, grinding, pelletizing, and humidity reduction; 3) research on electricity and heat production technologies for improving production process and quality of technologies in producing power and heat from biomass; and 4) research on economics and environmental impacts of biomass. The authors also note that most of the research in Thailand falls under group 3 and there is a need to pursue studies in other research areas.
With regard to biomass and biofuel status in Thailand, according to DEDE [8], Thailand has the target for ethanol production in 2036 to be 11.3 million liters per day. The actual ethanol production, however, is below the target (i.e., 3.51 million liters in 2015, 3.67 in 2016, 3.94 in 2017, and 4.20 in 2018, respectively). Obviously, the trend of ethanol production in Thailand will continuously grow in the future and thus a proper evaluation of biomass efficiency for each agricultural regional area is required. Regarding biomass types in Thailand, studies from DEDE [8] also show the potential of the biomass for varied types of feedstocks with more or less capacity, in which cassava, sugarcane, and palm are among the top potential biomass types.

Efficiency Study with DEA applications
Multi-Criteria Decision Analysis (MCDM) is a subdiscipline of operations research and management science (OR/MS) that explicitly considers multiple criteria in a decision-making environment and has been used to support decision-makers facing decision and planning problems that a unique optimal solution does not exist and/or decision-maker's preferences are involved. Common methods, specifically, include various tools, such as Analytic Hierarchy Process (AHP), Data Envelopment Analysis (DEA), Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS), Multi-Attribute Utility Theory (MAUT), Multi-Objective Mathematical Programming (MOMP), and Goal Programming (GP). These tools have been applied and extended in a number of applications (e.g., [10][11][12][13][14][15][16][17][18][19]).
DEA, particularly, is a Linear Programming (LP) methodology to measure relative efficiency of multiple Decision-Making Units (DMUs) or so-called alternatives when the problem is presented with multiple input and output criteria. After the DEA linear programming model is solved, a particular DMU will be considered efficient if it obtains a score of one, whereas scores that are lesser than one imply relative inefficiency. It is also possible that more than one alternatives are found to be efficient. According to survey study from Liu et al. [20], the DEA literature's size is expected to continue to grow at least double the size of the existing literature. In addition, the DEA method has been applied in various applications [21][22]. We next discuss the three prevalent DEA models commonly used in the literature and the DEAP computer program.

CCR Model
The CCR model was early developed and named after the three researchers (Charnes, Cooper and Rhodes [23] to measure the overall technical efficiency (TEoverall), in which a Constant Return to Scale (CRS) assumption holds. That is, the CRS assumption holds true when the DMUs are operated under the condition of the optimal size and perfect competition. In particular, equations (1)-(5) present the CCR model of DEA in a linear programming form.

BCC Model
The BCC model was later developed by and named after Banker, Charnes, Cooper [24] to measure the pure technical efficiency (TEpure) of DMUs. The BCC model is formulated by extending from the dual model of the primal CCR model, which transforms the primal maximization to dual minimization problem. In contrast to CCR model, the BCC allows DMUs to be operated under imperfect condition and not necessarily at optimal size, which is more practical in real situations. That is, the Variable Return to Scale (VRS) assumption holds for the BCC model of DEA (Equations (6)-(10)), where  is the relative efficiency and k  is the dual decision variable for each DMU.

Mathematical model
Subject to:

SE Model
The SE can be computed to express whether a particular DMU is operating at optimal size (i.e., similar to the CRS assumption) or whether at imperfect condition (i.e., similar to the VRS assumption). That is, if the latter holds true, the value of SE can be used to indicate whether the DMU operates under Increasing Return to Scale (IRS) (i.e., the size is too large) or Decreasing Return to Scale (i.e., the size is too small). In particular, the SE can be computed as a ratio between the relative efficiency obtained from the CCR model and the BCC model as shown in Equation (11).

DEAP Computer Program
We next discuss the Data Envelopment Analysis Program (DEAP). The program consists of the instruction file, the data file, and the output file; in which the CCR model, the BCC model, and the SE model can be simultaneously computed to obtain relative efficiencies of DMUs of interest. The program is also capable of computing how much the input criteria should be decreased for inefficient DMUs to be efficient (i.e., input-oriented) and how much the output criteria should be increased for inefficient DMUs to be efficient (i.e., output-oriented) for benchmarking purpose. In this research, the computer program DEAP Version 2.1 is used for analyzing related efficiency data.

DEA Analysis and Results
We next analyzed results from the DEAP computer program as shown in Tables 4-6 for data from 2017-2019, respectively. The overall technical efficiency from the CCR model, the pure technical efficiency from the BCC model, and the analysis from the SE model are presented. Provincial DMUs with either IRS (too-large size) or DRS (too-small size) are also analyzed. Regardless, other techniques (i.e., heuristics, simulation) can also be used and integrated to solve the linear programming problem of DEA model as well [26][27][28][29][30][31][32][33][34].  Table 4, provincial DMUs that operate with an efficient condition (i.e., the score for relative efficiency of 1) and with the optimal size (i.e., the score for SE of 1) during 2017 and should be considered the benchmark units are A1, A2, A3, A5, A11, A13, A14, A18, A19, and A20. In addition, data analyzed for 2018 (Table 5) show that efficient DMUs with optimal sizes are A1, A3, A5, A11, A18, A19, and A20. Next, based on the 2019 data obtained in Table 6, efficient DMUs with optimal sizes are found to be A1, A3, A5, A18, A19, and A20, respectively.
Clearly, an operation for some provincial DMUs fluctuates during 2017-2019, whereas certain provincial DMUs can operate with all efficient conditions for three years. Additionally, the IRS condition for certain provincial DMUs suggest that scale inefficiency exists, in which the size is considered too large when comparing to other DMUs. These analyzed results are also categorized for CCR model, BCC model, and SE model across all progressive years to illustrate the trend with respect to time as shown in Figs 2-4, respectively.

Discussion
Analyzed results from the CCR model, the BCC model, and the SE model obtained earlier suggest that A1) Loei, A3) Udon Thani, A5) Bueng Kan, A18) Khon Kaen, A19) Chaiyaphum, and A20) Nakhon Ratchasima are efficient across three years from 2017 to 2019. This is due to that the analyzed relative efficiency scores are shown to be 1.00 for three consecutive years in  Overall, these efficient provinces are found to utilize lesser inputs (e.g., planting area, labor cost, rainfall) and/or obtain higher outputs (e.g., tons of products) when comparing to other peers. Thus, these provinces have been operated with efficient condition, in which they should be further used as a benchmark DMUs for other provinces.
In addition, other provinces operated at inefficient condition can consider whether a particular input criterion should be decreased with a fixed output requirement or a particular output criterion can be increased under a fixed input.

Conclusion and Future Research
Biomass represents a significant source of biofuel, which is a type of renewable energy getting attention from many countries nowadays. In this research, biomass data of three major feedstocks for biofuel in the Northeastern region of Thailand were collected and analyzed using DEA to analyze each provincial efficiency. The input criteria of allowable planting area, labor cost, and rainfall amount as well as the output criterion of the quantity of harvested product were, in particular, collected for the top energy crops of cassava, sugarcane, and palm during 2017 to 2019. Accordingly, the relative efficiency of each provincial alternative was analyzed using DEA analysis of CCR model, BCC model, and SE model, respectively.
Analyzed results showed that, among 20 provinces of the Northeastern region of Thailand, there were six provinces that operated efficiently under the selective criteria. These provinces were found to be Loei, Udon Thani, Bueng Kan, Khon Kaen, Chaiyaphum, and Nakhon Ratchasima, respectively. Thus, these efficient provinces could be further used as benchmark DMUs for other provinces. Regardless, it is important to note that the analyzed results are dependent on selected criteria for inputs and outputs, in which the caution should be noted.
Directions for future research of this study include 1) expanding the case study for other regional areas in Thailand for further comparative study, 2) exploring other types of crops related to energy feedstock, 3) investigating other time spans for different years or with other time units, such as monthly basis, and 4) assessing other criteria types inclusive of both inputs and outputs. That is, other economic aspects can be further included for the input criteria. In addition, outputs concerning the sustainability index can also be enhanced. Additionally, we note that this study is the first phase of our research framework to investigate the upstream process of the bioenergy supply chain. That is, the results obtained from this study will be used as input for further supply chain modelling study.