The Impact of Electricity Generation on CO₂ Emissions in Türkiye: An Agent-Based Simulation Approach

Abstract

This study investigates the impact of energy, economic, and environmental policies on Türkiye’s energy dynamics and CO₂ emissions using climate models and an agent-based simulation (ABM) framework. By integrating climate projections with policy scenarios, it assesses how energy transitions and climate change affect renewable energy sources (RES), cooling demands, and CO₂ emissions from electricity generation. Methods include selecting suitable Global Climate Models (GCMs), evaluating climate change impacts on RES performance, and simulating policy effects through ABM across one base and nine policy scenarios from 2023 to 2040. The study highlights the critical role of policy interventions in influencing emissions trends and energy prices. It identifies renewable energy subsidies and low-carbon strategies, such as nuclear power support, as effective tools for reducing emissions and stabilizing energy costs. The methodologies and findings provide actionable insights for policymakers globally, emphasizing the importance of integrating climate data with policy planning.

1. Introduction

The escalating concerns over climate change and global warming have compelled governments worldwide to introduce stringent regulations and policies aimed at mitigating environmental impacts. In 2022, electricity generation contributed to 39.3% (14.6 GtCO₂) of global CO₂ emissions, underscoring its significance in emission reduction strategies [1]. Various approaches have been adopted, including carbon taxes and renewable energy subsidies, to incentivize cleaner energy production. Carbon taxes impose financial disincentives on high-emission activities, while renewable energy incentives, such as feed-in tariffs and tax credits, encourage investments in sustainable technologies [2]. Success stories from countries like Sweden, which achieved notable emission reductions through a carbon tax introduced in the 1990s, and Germany, where renewable energy incentives fostered a thriving clean energy sector, highlight the effectiveness of these measures.

Crafting impactful climate policies, however, requires a profound understanding of human behaviour and social interactions. Traditional models often simplify these complexities, assuming rational, self-interested agents and overlooking social dynamics [3]. Agent-Based Modelling (ABM) has emerged as an effective tool to overcome these limitations, offering simulations of complex systems by depicting interactions among heterogeneous agents [4]. This approach captures both market and non-market relationships within a unified framework. By enabling the analysis of how market mechanisms, such as taxes, interact with non-market mechanisms like information campaigns, ABM facilitates a nuanced exploration of policy synergies and trade-offs [5,6]. Moreover, its ability to simulate emergent behaviours and decentralized interactions provides deeper insights into electricity market dynamics compared to traditional equilibrium, optimization, econometric, or system dynamics models [7,8,9].

In Türkiye, a developing country responsible for 1.2% of global CO₂ emissions in 2022, these issues take on critical importance. Türkiye’s contribution to emissions has almost tripled over the past 50 years [10]. Despite a 15% decline in the carbon intensity of electricity generation since 2000, the country emitted 148.57 MtCO₂ from electricity production in 2023 [11]. Population growth, increasing energy demand, and Türkiye’s vulnerability to climate impacts in the Mediterranean Basin emphasize the urgency of addressing its energy policies [12].

Türkiye has made significant strides in renewable energy adoption, increasing non-hydro renewable capacity from 3.5% to 22.96% of total installed capacity within a decade [13]. However, renewable energy production remains heavily influenced by variability in wind speed, temperature, and solar irradiance. This study establishes a climate–energy–economy nexus using an agent-based simulation model, integrating future projections from Global Climate Models (GCMs) as inputs. By incorporating these projections, the model evaluates the long-term effects of diverse policy scenarios on energy supply, demand, and CO₂ emissions, considering dynamic and complex interactions.

2. Literature Review

In the literature, the effects of government policies on energy demand and production have been examined by many different methods. ABM stands out as one of the best methods to investigate the effects of single or multiple combinations of various policies on the energy sector. Since this paper investigates the nexus of climate and energy policies through ABM, this literature review section only focuses on the specific studies which utilized ABM.

Initially, studies like Cong and Wei [14] and Chen et al. [15] employed ABM to assess the implications of carbon emissions trading systems and investment preferences on China’s power sector. These models demonstrate the complex interplay between policy interventions and market dynamics, showing how risk attitudes and technological inclinations influence the transition toward low-carbon strategies.

Gerst et al. [16], Rengs et al. [17], and Chappin et al. [18] extended this analysis by introducing multi-level modelling frameworks that capture the evolving interactions between international agreements and local policy outcomes. By incorporating behavioural assumptions and diverse climate policies, these models provide insights into the combined impacts on climate and economic factors, revealing nuances not captured by traditional rational agent models.

Further advancements include the development of integrated assessment models such as the Dystopian Schumpeter Keynes (DSK) model by Lamperti et al. [19] and agent-based methodologies presented by Czupryna et al. [20]. These models incorporate a broader spectrum of empirical patterns, accounting for micro- and macro-level economic and climate dynamics. They highlight the significance of individual agent choices, technological advancements, and the complexity of damage functions in shaping emergent economic behaviour and climate outcomes.

Studies by Karimi and Vaez-Zadeh [21], Li et al. [22], Zhou et al. [23], and Babatunde et al. [24] focused on assessing sustainable energy policies and renewable energy initiatives using ABM. These analyses revealed the potential consequences of inadequate policies, the strategic decision-making of market participants, and the challenges in achieving ambitious emission reduction goals. More recently, Foramatti et al. [25] used an ABM to show that combining carbon pricing with low-carbon infrastructure and progressive revenue recycling reduces emissions and improves well-being. Similarly, Amendola et al. [26] found that direct technological investments, evaluated through an ABM, are the most effective energy efficiency policies, requiring long-term commitment and immediate complementary measures. These studies emphasize the importance of diverse, integrated policy approaches that consider economic and social impacts to address climate and energy challenges effectively.

The integration of ABM into energy system studies has greatly enhanced the understanding and management of modern energy challenges. For instance, Yang et al. [27] used a multi-agent game-based approach for planning electricity–gas systems, addressing wind power uncertainties and ensuring fair profit and security. Similarly, Zhou and Lund [28] emphasized the role of blockchain in secure and transparent P2P energy sharing. Additionally, Stennikov et al. [29] highlighted the need for balanced centralized and distributed energy generation for efficiency and reliability. Furthermore, Ahmadi et al. [30], Yu et al. [31] and Mussawar et al. [32] showcased an ABM to model of urban energy demand and supply, improving profit and reliability with advanced storage. Meanwhile, Howell et al. [33] proposed the EOS agent-based model to analyze neighbourhood energy consumption and assess peak-hour energy shaving across varying EV adoption rates and energy demands. Moreover, Zhang et al. [34] revealed that carbon peak and neutrality goals, strategic bidding, and carbon trading promote cleaner electricity markets in China. Finally, Yao et al. [35] reviewed ABM in community energy systems, calling for interdisciplinary approaches. Collectively, these studies demonstrate ABM’s effectiveness in modelling complex interactions, addressing uncertainties, and advancing sustainable energy futures.

Although the global literature on ABM applications is vast, there is a clear gap in studies from Türkiye and other developing countries. Notable works addressing this gap include Calikoglu and Koksal [36], who modelled Türkiye’s electricity and heat production sector to explore pathways for achieving net-zero emissions by 2053. Their results highlight the significant increase in capacity, generation, and investment required. Similarly, Telli et al. [37] compared the energy transitions of Turkey and Germany, analyzing the effectiveness of Germany’s renewable energy policies for Turkey’s context. Shahbaz et al. [38] investigated the impact of financial development on renewable energy consumption in 34 upper-middle-income countries, showing that financial growth drives demand for cleaner energy sources. More recently, Bi and Khan [39] examined the role of environmental, social, and governance investing, eco-innovation, and renewable energy in the BRICS nations, concluding that these countries can achieve carbon neutrality by aligning economic strategies with renewable energy investments and policies.

While ABM has been widely applied in energy studies, existing research rarely incorporates climate data to comprehensively link climate and energy systems. This study bridges this gap by integrating multi-model ensemble climate projections into ABM to analyze the effects of long-term climate change on energy demand and supply. It pioneers the modelling of interconnected dynamics between climate, economy, and energy generation, considering both economic and environmental impacts. Furthermore, it introduces a novel approach by dynamically estimating renewable energy capacity factors under changing climate conditions, moving beyond traditional static coefficients.

By employing this advanced framework, the study addresses several critical questions:

• Which Global Climate Models (GCMs) are the best for mimicking the climate conditions of Türkiye?
• How will electricity production be affected by climate change in the future? How will factors such as changing wind speeds, solar radiation, and temperature affect the amount of electricity to be obtained from renewable energy sources?
• How will the space cooling need to change (cooling-degree-days) depending on climate change?
• How will the energy policies affect CO₂ emissions from electricity generation when climate change is also taken into account?
• How will the electricity generation capacity change in different time horizons considering the climate, government regulations, and policies?

3. Materials and Methods

To investigate the impacts of energy, economic, and environmental policies on energy demand, production, and CO₂ emissions in Türkiye, this study follows the outlined steps:

• Collecting and regridding GCM-based climate data from the CMIP6 (Coupled Model Intercomparison Project Phase 6) experiment, and comparing them with observation-based data, specifically ERA 5 (ECMWF Reanalysis v5), which are bias-corrected with CRU (Climatic Research Unit) (hereinafter ERA5), using three performance criteria, namely Kling–Gupta efficiency, normalized root mean squared error, and modified index of agreement
• Evaluating the performance of GCMs and selecting the top four models.
• Training the selected top four GCMs with ERA5 data using the Extreme Gradient Boosting Tree (XGBoost) method and forecast the future values of climate variables under SSP5-8.5 climate scenario.
• Estimating electricity demand, cooling-degree-days (CDD), and electricity generation from wind and solar power systems.
• Establishing utility function weights for technology investment decisions through Analytical Hierarchical Process (AHP) and Multi-Attribute Utility Technique (MAUT).
• Developing an ABM and conducting various simulation scenarios from 2023 to 2040.

The structure of proposed GCM-ABM framework is given in Figure 1.

3.1. Climate Model

Climate models serve as crucial tools for understanding past and future climate changes. These models simulate the physics, chemistry, and biology of the atmosphere, land, and oceans with high detail, requiring powerful supercomputers. As modelling groups worldwide refine these tools by improving spatial resolution and incorporating new processes, their results are coordinated under the Coupled Model Intercomparison Projects (CMIP), timed with the Intergovernmental Panel on Climate Change (IPCC) reports. The 2013 IPCC fifth assessment report (AR5) utilized CMIP5 models, while the 2021 sixth report (AR6) featured the advanced CMIP6 models [40]. CMIP6 includes projections from 49 modelling groups across about 100 runs.

In this study, historical data from 13 CMIP6 GCMs (

Table A1. Information about GCMs and their horizontal resolutions in CMIP6 experiment.

No	Model	Institute, Country	Horizontal Resolution
M1	ACCESS-CM2	Commonwealth Scientific and Industrial Research Organisation, Australia	1.9° × 1.3°
M2	ACCESS-ESM-1-5		1.9° × 1.2°
M3	BCC-CSM2-MR	Beijing Climate Center (BCC) and China Meteorological Administration (CMA), China	1.1° × 1.1°
M4	CMCC-ESM2	Euro-Mediterranean Centre on Climate Change coupled climate model, Italy	1.25° × 0.938°
M5	CMCC-CM2-SR5
M6	GFDL-ESM4	Geophysical Fluid Dynamics Laboratory, US	1.3° × 1°
M7	HadGEM3-GC31-LL	Met Office Hadley Centre, UK	1.25° × 1.875°
M8	IITM-ESM	Centre for Climate Change Research, Indian Institute of Tropical Meteorology, India	1.875° × 1.9°
M9	INM-CM4–8	Institute of Numerical Mathematics, Russia	2° × 1.5°
M10	INM-CM5–0
M11	MIROC6	Japanese Modelling Community, Japan	1.4° × 1.4°
M12	MPI-ESM1–2-HR	Max Planck Institute for Meteorology, Germany	0.9° × 0.9°
M13	UKESM1–0-LL	Met Office Hadley Centre, UK	1.9° × 1.3°

in the Appendix A) were analyzed to identify those best representing Türkiye’s climate. The selected GCMs offer diverse and robust projections, considering temperature (2 m), wind speed (m/s), and surface downwelling shortwave solar radiation (W/m²), which are key variables for agent-based modelling. By ensuring a balance of computational efficiency and spatial detail, the chosen models provide a comprehensive basis for evaluating future climate scenarios.

To standardize model resolutions, all data were regressed to a 1° × 1° grid using the conservative remapping method via the Linux-based cdo programme. Daily data for 120 grid points across Türkiye during 2010–2014 were extracted from NetCDF files using R software (version 4.4.1). Performance evaluations of these GCMs were conducted by comparing their data with the ERA5 dataset, applying metrics such as Kling–Gupta efficiency (KGE), the modified index of agreement (md), and normalized root mean square error (nRMSE), as detailed in [41].

For each grid and variable, GCMs were ranked from 1 (best) to 13 (worst), based on their performance across the metrics. The final rankings were processed using the Multiple-criteria Decision Analysis (MCDA) method, where a GCM’s score reflects its performance across all grid points. According to this method, if a GCM achieves rankings of 1, 2, 3, …, n at grid points X₁, X₂, X₃, X_n, respectively, the MCDA score for the GCM is calculated as follows: X₁ + X₂ (1/2) + X₃ (1/3) + … + X_n (1/n). To illustrate, if a GCM secures the top rank in all n grid points (indicating the highest performance), it receives (n/1) points; if it attains the second place in m grid points, it receives (m/2) points, and so on. The overall score is determined by summing the values assigned for all the rankings that a GCM has received. These rankings were further consolidated into a single measure using the Comprehensive Rating Metrics (MR) method, integrating all performance criteria and variables. The mathematical expression of MR is given as

$$MR=1-{\displaystyle \frac{1}{nm}}\sum _{i=1}^n{rank}_i,$$

(1)

where m and n represent the number of evaluation criteria (in this case m is 9:3 KGE, 3 md, and 3 nRMSE) and the number of GCMs (in this case n is 13), respectively. An MR value closer to 1 means that the GCM has a greater ability to mimic observed data.

3.2. Extreme Gradient Boosting

Extreme Gradient Boosting (XGBoost) is a sophisticated tree-based artificial intelligence method that stands out by harnessing the combined power of boosting and gradient boosting to generate remarkably precise and efficient predictions. While techniques like Random Forest (RF) and Gradient Boosting Regression Trees (GBRT) can be demanding in terms of computational resources during the learning process, XGBoost distinguishes itself as one of the most computationally efficient approaches in the field [42]. Unlike the GBRT methodology, XGBoost utilizes a distinctive algorithm for constructing trees, leveraging the Similarity Score (Equation (2)) and Gain (Equation (3)) to identify the optimal node splits.

$$Similarity Score={\displaystyle \frac{{\left(\sum _{i=1}^n{Residual}_i\right)}^2}{\sum _{i=1}^n\left[{Previous Probability}_i\times \left(1-{Previous Probability}_i\right)\right]-\lambda }}$$

(2)

The term ‘previous probability’ refers to the probability of an outcome that was estimated in the prior step. Initially, a probability of 0.5 is assigned to all observations to construct the initial tree. As additional trees are built, the previous probability is re-evaluated, taking into account the initial prediction and predictions gathered from all prior trees. The parameter Lambda (λ) serves as a regularization factor. Once the Similarity Score has been calculated for each leaf, it is possible to determine the Gain through the following procedure:

$$Gain={Left leaf}_{similarity}+{Right leaf}_{similarity}-{Root}_{similarity}.$$

(3)

The highest Gain indicates the best point for splitting the tree.

The xgboost package in R was utilized to assemble the top four selected GCMs, assembling them to forecast future climate variables. This approach ensured robust predictions by integrating the strengths of individual models in a data-driven framework.

3.3. Analytical Hierarchical Process

The Analytic Hierarchical Process (AHP) is a well-established and structured approach for decision-making, widely utilized in the realms of operations research and multi-criteria decision analysis. AHP was conceptualized by Saaty [43] to create a systematic framework that effectively addresses intricate decision dilemmas by organizing them into a hierarchical structure encompassing criteria and alternatives.

AHP utilizes pairwise comparisons and mathematical algorithms to determine priority weights for criteria and alternatives, thereby assisting in the identification of the most suitable choice. This method has found practical use in a diverse array of fields, including project selection, resource allocation, supplier evaluation, and environmental impact assessment, owing to its capability to capture subjective preferences and facilitate structured, well-informed decision-making. AHP’s stringent and transparent nature renders it an invaluable instrument for confronting intricate decision challenges spanning a broad spectrum of disciplines. For more information, the study of Saaty [43] can be examined.

3.4. Multi-Attribute Utility Technique

Multi-Attribute Utility Technique (MAUT) is an analytical method and decision-making approach employed in diverse domains, including economics, engineering, and management. Its purpose is to evaluate and compare multiple choices or alternatives by considering a predefined set of attributes or criteria (See Figure 2). MAUT is strategically crafted to assist decision-makers in making well-informed decisions, particularly in situations involving intricate, multi-criteria scenarios.

A utility score of 1.0 is attributed to the most favoured option, while 0.0 is ascribed to the least favoured choice. Experts then pinpoint a midpoint value with a utility score of 0.5, situated precisely between the most and least favoured options. Once the midpoint value is established, experts proceed to identify a “quarter point” value with a utility score of 0.25, positioned halfway between the midpoint and the least preferred alternatives. Ultimately, experts determine a value function with a utility score of 0.75, which is positioned midway between the most favoured and the midpoint values [44].

3.5. Agent-Based Model

Given the capabilities and advantages inherent in this approach, ABM emerges as one of the most suitable methods for precisely simulating the proposed model. Consequently, the ensuing study employs the ABM structure depicted in Figure 3. Furthermore, the detailed flowchart (Figure S1) and pseudocode of this structure are provided in the Supplementary Materials.

The proposed model comprises three distinct agent categories: government, independent power producers (IPP), and market maker. The government agent is tasked with establishing carbon taxes, carbon allowances, and subsidies, while the market maker agent oversees bid collection and determines electricity prices. IPP agents represent electricity producers with diverse portfolios and profit margin expectations, submitting bids for each electricity generation technology within their portfolios. To enact the proposed ABM, this study utilizes four modules:

• Electricity demand module
• Electricity generation module
• Capacity addition/shutdown module
• Carbon module

The agent-based model was developed and executed using the AnyLogic software package (version 8.9.1), a versatile platform for simulation modelling. AnyLogic enables the integration of multiple modelling approaches, including agent-based, discrete event, and system dynamics methods, providing a comprehensive framework for analyzing complex systems. Its ability to incorporate spatial and temporal dynamics, along with robust visualization tools, allows for detailed scenario testing and real-time feedback. These capabilities make AnyLogic particularly well-suited for modelling interactions among agents and assessing the emergent behaviours within dynamic systems.

3.5.1. Electricity Demand Module

The analysis of Türkiye’s electricity consumption reveals that three primary sectors account for over 94 percent of the total electricity demand [45]. These sectors include (i) residential buildings, (ii) commercial buildings, and (iii) industry. Each sector’s electricity demand exhibits varying sensitivities to changes in electricity prices, income levels, population, and temperature. Given these sensitivities, Equations (4)–(6) depict the electricity demand for the residential sector, commercial sector, and industry, respectively. These equations represent an enhanced and more advanced formulation compared to those presented by Guven et al. [46], incorporating additional considerations to improve their applicability and accuracy.

$$\begin{gathered} {\mathrm{E}\mathrm{C}\mathrm{A}}_t^{res}={\mathrm{E}\mathrm{C}\mathrm{A}}_{t-1}^{res}·\left[1+\left({\displaystyle \frac{Y_t}{Y_{t-1}}}-1\right)·{\rho }_{gdp}^{res}\right]·\left[1+\left({\displaystyle \frac{P_{t-1}^{elec.cons.res}}{P_{t-2}^{elec.cons.res}}}-1\right)·{\rho }_p^{res}\right]· \\ \left[1+\left({\displaystyle \frac{{CDD}_t}{{CDD}_{t-1}}}-1\right)·{\rho }_{cdd}^{res}\right]·\left[1+\left({\displaystyle \frac{{Pop}_t}{{Pop}_{t-1}}}-1\right)·{\rho }_{pop}\right] \end{gathered}$$

(4)

$$\begin{gathered} {\mathrm{E}\mathrm{C}\mathrm{A}}_t^{com}={\mathrm{E}\mathrm{C}\mathrm{A}}_{t-1}^{com}·\left[1+\left({\displaystyle \frac{Y_t}{Y_{t-1}}}-1\right)·{\rho }_{gdp}^{com}\right]·\left[1+\left({\displaystyle \frac{P_{t-1}^{elec.cons.com}}{P_{t-2}^{elec.cons.com}}}-1\right)·{\rho }_p^{com}\right]· \\ \left[1+\left({\displaystyle \frac{{CDD}_t}{{CDD}_{t-1}}}-1\right)·{\rho }_{cdd}^{com}\right]·\left[1+\left({\displaystyle \frac{{Pop}_t}{{Pop}_{t-1}}}-1\right)·{\rho }_{pop}\right] \end{gathered}$$

(5)

$$\begin{gathered} {\mathrm{E}\mathrm{C}\mathrm{A}}_t^{ind}={\mathrm{E}\mathrm{C}\mathrm{A}}_{t-1}^{ind}·\left[1+\left({\displaystyle \frac{Y_t}{Y_{t-1}}}-1\right)·{\rho }_{gdp}^{ind}\right]·\left[1+\left({\displaystyle \frac{P_{t-1}^{elec.cons.ind}}{P_{t-2}^{elec.cons.ind}}}-1\right)·{\rho }_p^{ind}\right]· \\ \left[1+\left({\displaystyle \frac{{CDD}_t}{{CDD}_{t-1}}}-1\right)·{\rho }_{cdd}^{ind}\right]·\left[1+\left({\displaystyle \frac{{Pop}_t}{{Pop}_{t-1}}}-1\right)·{\rho }_{pop}\right] \end{gathered}$$

(6)

$${\displaystyle \frac{Y_t}{Y_{t-1}}}=\left(1+{\eta }_t^Y\right)·\left(1-{\eta }^{CDF}·\left(T_t^{POP}-T_0^{POP}\right)\right).$$

(7)

$$T_t^{POP}={\displaystyle \frac{\sum _{i=1}^{81}{Pop}_{i,t}·T_{i,t}}{\sum _{i=1}^{81}{Pop}_{i,t}}},$$

(8)

where $T_t^{POP}$ is the population-weighted mean temperature of Türkiye. ${Pop}_{i,t}$ and $T_{i,t}$ stand for the population and annual mean temperature of i^th city in year t. $T_0^{POP}$, which is the population-weighted mean temperature of Türkiye for the base year 2023, is found to be 285.43 K as an output of the established climate models.

$${Pop}_{i,t}={\gamma }_i·{Pop}_{i,t-1}·{\beta }_t,$$

(9)

$${\beta }_t={\displaystyle \frac{\sum _{i=1}^{81}{Pop}_{i,t}}{{Pop}_t^{TS}}},$$

(10)

where ${\beta }_t$ is the population correction coefficient in year t, and ${Pop}_t^{TS}$ is the total population of Türkiye projected by TurkStat [47]. The descriptions, values, and references of variables are provided in

Table 1. Variables of electricity demand calculations.

Variable	Description	Value	Reference
${\mathit{\rho }}_{\mathit{g}\mathit{d}\mathit{p}}^{\mathit{r}\mathit{e}\mathit{s}}$	Income elasticity of residential electricity demand	0.227	[46]
${\mathit{\rho }}_{\mathit{p}}^{\mathit{r}\mathit{e}\mathit{s}}$	Price elasticity of residential electricity demand	−0.126
${\mathit{\rho }}_{\mathit{c}\mathit{d}\mathit{d}}^{\mathit{r}\mathit{e}\mathit{s}}$	CDD elasticity of residential electricity demand	5.397
${\mathit{\rho }}_{\mathit{g}\mathit{d}\mathit{p}}^{\mathit{c}\mathit{o}\mathit{m}}$	Income elasticity of commercial electricity demand	0.219
${\mathit{\rho }}_{\mathit{p}}^{\mathit{c}\mathit{o}\mathit{m}}$	Price elasticity of commercial electricity demand	−0.147
${\mathit{\rho }}_{\mathit{c}\mathit{d}\mathit{d}}^{\mathit{c}\mathit{o}\mathit{m}}$	CDD elasticity of r commercial electricity demand	4.55
${\mathit{\rho }}_{\mathit{g}\mathit{d}\mathit{p}}^{\mathit{i}\mathit{n}\mathit{d}}$	Income elasticity of industrial electricity demand	0.548
${\mathit{\rho }}_{\mathit{p}}^{\mathit{i}\mathit{n}\mathit{d}}$	Price elasticity of industrial electricity demand	−0.145
${\mathit{\rho }}_{\mathit{c}\mathit{d}\mathit{d}}^{\mathit{i}\mathit{n}\mathit{d}}$	CDD elasticity of industrial electricity demand	3.25
${\mathit{\rho }}_{\mathit{p}\mathit{o}\mathit{p}}$	Population elasticity of electricity demand	5.198	[48]
${\mathit{\eta }}_{\mathit{t}}^{\mathit{Y}}$	Real GDP per capita potential growth rate		[49]
${\mathit{\eta }}^{\mathit{C}\mathit{D}\mathit{F}}$	Capital damage factor for natural disasters	0.061%	[20]
${\mathit{P}\mathit{o}\mathit{p}}_{\mathit{i},\mathit{t}}$	Projected population of cities		Author’s calculation based on TurkStat [47,50] data

.

3.5.2. Electricity Generation Module

In this study, real portfolios of Independent Power Producers (IPP) in Türkiye will be used. However, company names are not used, and each company has a code. The biggest 50 power generation enterprises and the cumulative installed capacity of the rest of the facilities as the 51st power generation enterprise is simulated. These remaining capacities are treated as a single entity since they are relatively small-scale, constituting 21.1% of the total capacity. Compared to the larger enterprises, these smaller enterprises are more likely to adapt to market conditions and base their actions on the average behaviour patterns of the entire sector, particularly concerning their preferences for technology, risk, and investment thresholds [51].

The GCM outputs are only climate variables, and for this reason, these data must be used as input to various mathematical equations in order to be associated with energy production in the agent-based simulation model.

Wind speeds are model outputs showing speeds at 10 m above the surface. However, today the towers of wind turbines could exceed 150 m. Therefore, wind speeds need to be revised according to height. In this study, the average wind turbine height was taken as 100 m and the wind speed at this height was estimated using the equation below.

$$wspd\left(z\right)=wspd\left(z_{ref}\right){\left({\displaystyle \frac{z}{z_{ref}}}\right)}^{\alpha },$$

(11)

where z and z_ref is the wind turbine hub height (100 m) and GCMs output height (10 m), respectively. wspd(z_ref) represents the wind speed at z_ref, while α is the coefficient of the power law exponent. α was considered as 1/7 because it is suitable for open areas and was taken as 1/7 in numerous studies [41,52,53].

Wind Power Density (WPD) is an extensively used metric to evaluate the wind power generation potential for a specific location and is calculated as:

$$WPD={\displaystyle \frac{1}{2}}·\rho ·{wspd\left(z\right)}^3,$$

(12)

where ρ stands for the air density (1.225 kg/m³). Since the WPD varies proportionally with the cube of the wind speed, even a slight alteration in wind speed can lead to a significant rise or fall in wind energy output.

The performance of a photovoltaic (PV) solar system is contingent upon the surface downwelling shortwave solar irradiation it receives and the efficacy of the PV system. Notably, the effectiveness of a PV system fluctuates in response to the surrounding temperature [54]. To gauge PV efficiency in relation to ambient temperature, the Evans–Florschuetz PV efficiency correlation coefficients are employed in this study [55].

$${\eta }_c={\eta }_{ref}\left[1-{\beta }_{ref}\left(T_c-T_{ref}\right)\right],$$

(13)

where η_ref and β_ref represent the standard PV efficiency (0.20) at the reference temperature (T_ref = 25 °C) and temperature coefficient (0.0045), respectively.

$$T_c=c_1+c_{2}tas+c_{3}rsds+c_{4}wspd,$$

(14)

c₁ = 4.3 °C, c₂ = 0.943, c₃ = 0.028 °C m²W⁻¹, and c₄ = −1.528 °C m⁻¹s [56]. The energy that can be obtained from the PV panel depends on solar radiation and panel efficiency (Equation (15)).

$$E_{PV}=Radiation \left({\displaystyle \frac{W}{m^2}}\right)\times {\eta }_c.$$

(15)

Each company’s production facility portfolio and capacities differ. The electricity production from solar PV depends on the amount of shortwave radiation coming to the surface and the panel efficiency, which is revised according to the temperature and wind speed, as mentioned before. In this context, the amount of electricity obtained from 1 m² PV panel in year t is given in Equation (16).

$$Q_{i,PV,j,t}={\sum }_{d=1}^{365}{\eta }_{c_d}\times {rsds}_d\times 24 [\mathrm{kWh}/{\mathrm{m}}^2].$$

(16)

Here, i is the IPP number, j is which plant in the selected IPP portfolio, and rsds_d is the average surface downwelling shortwave solar radiation on day d. Similarly, electricity generation from the wind turbine is calculated according to the equation below, depending on the wind power density.

$$Q_{i,WT,j,t}={\sum }_{d=1}^{365}\pi \times R^2\times {\beta }_{WT}\times {WPD}_d\times 24 \left[\mathrm{kWh}\right],$$

(17)

where R is the radius of the rotor of the wind turbine and β_WT is the efficiency of the wind turbine. In this study, turbine efficiency was taken as 0.4 considering the literature [57].

The first step in the process following the calculation of energy production from renewable resources, which varies depending on the climate, is the calculation of “residual generation”. Here, “residual generation” is the difference between the total electricity demand and the energy produced from climate dependent renewable sources, and this difference will be met by fossil fuel-based (coal, natural gas) or high-availability (hydroelectric, biogas, biomass, nuclear, geothermal) power plants.

$$Residual Generation=Electricity Demand-Generation from PV and WT.$$

(18)

IPPs make annual production planning for each power plant in their portfolio. While planning for power generation (G^planned), the capacity of the power plant (Cap_i,tech,j) and the capacity factor (CF_tech) are taken into consideration (Equation (19)).

$$G_{i,tech,j,t}^{planned}={8760\times CF}_{tech}\times {Cap}_{i,tech,j}.$$

(19)

IPPs prepare an annual electricity sale price offer (bid) for each plant in their portfolio. They give this electricity sales price offer according to the equation below.

$$\begin{gathered} {bid}_{i,tech,j,t}=\left\{\left[P_{tech,t}^{fuel}·G_{i,tech,j,t-1}^{actual}·f_{tech}^{consumption}+{(G}_{i,tech,j,t-1}^{actual}·f_{tech} \\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}-C_{i,techj,t}^{allowance})·{tax}_t^{carbon}+{OPEX}_{i,techj,t}+{Depr}_{i,techj,t}\right] \\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\times \left(1+{\displaystyle \frac{{\varnothing }_{t-1}}{1-{tax}_{tech,t}}}\right)/G_{i,techj,t-1}^{actual} \right\}-{subs}_{i,tech,j,t}. \end{gathered}$$

(20)

Table 2. Parameters of bidding calculation.

Parameter	Description	Parameter	Description
${\mathit{P}}_{\mathit{t}\mathit{e}\mathit{c}\mathit{h},\mathit{t}}^{\mathit{f}\mathit{u}\mathit{e}\mathit{l}}$	Fuel price	${OPEX}_{i,techj,t}$	OPEX of power plant
${\mathit{G}}_{\mathit{i},\mathit{t}\mathit{e}\mathit{c}\mathit{h},\mathit{j},\mathit{t}-1}^{\mathit{a}\mathit{c}\mathit{t}\mathit{u}\mathit{a}\mathit{l}}$	Actual generation of previous year	${Depr}_{i,techj,t}$	Depreciation of power plant
${\mathit{f}}_{\mathit{t}\mathit{e}\mathit{c}\mathit{h}}^{\mathit{c}\mathit{o}\mathit{n}\mathit{s}\mathit{u}\mathit{m}\mathit{p}\mathit{t}\mathit{i}\mathit{o}\mathit{n}}$	Unit fuel consumption of power plant	${\varnothing }_{t-1}$	Expected profit margin of IPP
${\mathit{f}}_{\mathit{t}\mathit{e}\mathit{c}\mathit{h}}$	Unit carbon emission of power plant	${tax}_{tech,t}$	Corporation Tax
${\mathit{C}}_{\mathit{i},\mathit{t}\mathit{e}\mathit{c}\mathit{h}\mathit{j},\mathit{t}}^{\mathit{a}\mathit{l}\mathit{l}\mathit{o}\mathit{w}\mathit{a}\mathit{n}\mathit{c}\mathit{e}}$	Carbon allowance of power plant	${subs}_{i,tech,j,t}$	Government subsidy
${\mathit{t}\mathit{a}\mathit{x}}_{\mathit{t}}^{\mathit{c}\mathit{a}\mathit{r}\mathit{b}\mathit{o}\mathit{n}}$	Carbon tax ($/ton CO2)

presents the descriptions of parameters used in electricity sales price offer calculation.

After Equations (19) and (20) are obtained for each power plant, the average electricity sales price offer is calculated according to Equation (21).

$${bid}_t^{avg}={\displaystyle \frac{\sum _i\sum _{tech}\sum _j{G}_{i,tech,j,t}^{planned}\times {bid}_{i,tech,j,t}}{\sum _i\sum _{tech}\sum _j{G}_{i,tech,j,t}^{planned}}}.$$

(21)

While the actual generation is equal to the planned generation for renewable energy power plants, the actual production of the power plants using fossil resources is shown in Equation (22).

$$G_{i,tech,j,t}^{actual}=\left\{\begin{array}{c}{G}_{i,tech,j,t}^{planned}, tech:Renewables\\ G_{i,tech,j,t}^{planned}\times {\displaystyle \frac{D_t^{total}-\sum G_{i,RES,j,t}}{\sum _i\sum _{tech}\sum _j{G}_{i,tech,j,t}^{planned}}}, tech:Fossil based\end{array}\right..$$

(22)

Finally, the electricity sales price determined by the market maker agent is given in Equation (23).

$$P_t^{elec}={bid}_t^{avg}{e}^{\left[\tau \left({\displaystyle \frac{D_t^{total}-\sum _i\sum _{tech}\sum _j{G}_{i,tech,j,t}^{planned}}{\sum _i\sum _{tech}\sum _j{G}_{i,tech,j,t}^{planned}}}\right)\right]},$$

(23)

where τ is the proportional coefficient reflecting price fluctuations when there is an imbalance between supply and demand. It controls the sensitivity of electricity prices to supply–demand imbalances, amplifying price changes based on the severity of the mismatch. It is used to capture the non-linear price behaviour typical in electricity markets, where even small imbalances can lead to significant price fluctuations. In this study, it is taken as 0.001 based on the literature [14,15].

The agent-based model in this study operates on a yearly time step, chosen to capture long-term economic trends and policy impacts on Türkiye’s energy system. This approach aligns with similar studies, including [14,15,17,21,24], which also used yearly resolutions for macroeconomic evaluations. While finer resolutions (e.g., hourly or daily) are essential for analyzing intra-day electricity price variations and renewable intermittency, this study emphasizes macroeconomic shifts and structural changes under policy scenarios.

The yearly resolution allows the integration of key parameters like GDP and annual emissions, balancing computational efficiency with the availability of Turkish data. Though this approach does not explicitly model short-term renewable variability or supply–demand mismatches, these dynamics are indirectly represented via annualized capacity factors and aggregated policy impacts. Consequently, the model’s outcomes are best interpreted as long-term trends, not detailed operational insights.

3.5.3. Capacity Addition/Shut-Down Module

Considering the future electricity demand, electricity producers may decide to invest in new power plants. As a first step, each electricity producer predicts the future electricity demand and supply. It is assumed that each electricity producer applies an ARIMA model to forecast electricity demand in year t + 3.

$${\displaystyle \frac{D_{i,t+3}^{pred.total}}{Q_t^{serv.}+Q_t^{constr.}-Q_t^{ret.}}}>{\epsilon }_i,$$

(24)

where $D_{i,t+3}^{pred.total}$ is the predicted future electricity demand, $Q_t^{serv.}$ is the generation capacity in the service, $Q_t^{constr.}$ is the power plants under construction, and $Q_t^{ret.}$ is the power plant capacity expected to complete their lifespans before year t + 3. Considering the security of electricity supply, ε_i is taken as 0.9 to constitute a reserve capacity.

Once the investment decision is made, an electricity producer establishes the capacity of newly allocated power plants by considering projected electricity shortages and its portion within the electricity market based on Equation (25).

$$Q_{i,t}^{inv.}=\left({\displaystyle \frac{D_{i,t+3}^{pre.total}}{{\epsilon }_i}}+Q_t^{ret.}-Q_t^{serv.}-Q_t^{constr.}\right)·{\displaystyle \frac{\sum _{tech}\sum _j{G}_{i,tech,j,t}^{actual}}{\sum _i\sum _{tech}\sum _j{G}_{i,tech,j,t}^{actual}}}.$$

(25)

After determining the capacity of newly allocated power plant, an electricity producer evaluates the utility of various power plant technologies by taking into account factors including return on investment (ROI), associated investment risks, as well as its preferences for risk, public acceptance, environmental impact (Global Warming Potential trough the life cycle) and technology. The utility function applied in this study is presented in Equation (26).

$$U_{i,tech,j,t}^{inv.}={(\omega }_{tech}^{eco}{\theta }_{i,tech,j,t}^{eco}+{\omega }_{tech}^{soc}{\theta }_{tech}^{soc}{+{\omega }_{tech}^{env}{\theta }_{tech}^{env})·\phi }_{i,tech,t,}$$

(26)

where $\omega$ is the weight of the criteria calculated based on the AHP analysis, while $\theta$ stands for the environmental, economic, and social utility score of the technology. Social acceptance percentages and environmental impacts of technologies are given in

Table 3. Social acceptance percentages and environmental impacts of technologies.

Technology	Social Acceptance (%)	Environmental Impact (g CO2-eq/kWh)
Wind	23.1	13.45
Geothermal	19.89	37.4
PV	18.44	38.8
Hydro	10.73	22.7
Biomass	8.47	62.4
CCGT	7.02	502
Coal	3.01	936
Nuclear	1.76	26.9

[58,59,60].

Economic utility function for energy investments is given as [15]:

$${\theta }_{i,tech,j,t}^{eco}=\left(1-e^{-{\gamma }_i·W_{i,tech,j,t}^{inv.}}\right),$$

(27)

where $w_{i,tech,j,t}^{inv.}$ is the discounted investment return rate (See Equation (28)), and γ_i, the Arrow–Pratt risk aversion coefficient, quantifies the risk inclination of a power generation company. When γ_i is positive, it signifies that the company is risk-averse, and a higher γ_i indicates a stronger aversion to risk. In this study, it is taken as 0.08 which is the typical rates of return on investment for technologies used to generate power [61]. φ_i,tech,t represents the proportion of this technology within the enterprise’s portfolio and its contribution to the overall national share of the same technology.

Furthermore, if a power plant incurs ongoing financial losses over a five-year period or reaches the conclusion of its operational lifespan, the power generation company will proceed to decommission it.

$$w_{i,tech,j,t}^{inv.}={\displaystyle \frac{\sum _{\stackrel{\prime }{t}=t}^{T_{tech+t}}\left\{\frac{\begin{array}{c}{(Q}_{i,t}^{inv.}·8760·{CF}_{tech}·P_t^{elec.avg})-{(Q}_{i,t}^{inv.}8760·{CF}_{tech}·f_{tech}-C_{new.i,techj,t}^{allowance})·{tax}_t^{carbon}\\ +{subs}_{tech,t}·Q_{i,t}^{inv.}-P_{tech,t}^{fuel}·Q_{i,t}^{inv.}·f_{i,tech,j,t}^{consumption}-{OPEX}_{i,techj,t}\end{array}}{{(1+r)}^{\stackrel{\prime }{t-t}}}\right\}}{{Inv}_{new.i,tech,j,t}}}+{Decom}_{tech}.$$

(28)

In light of the finite resources available for low-carbon energy sources, environmental constraints prevent their excessive utilization, as illustrated in Equation (29). Once a particular technology’s total installed capacity reaches its resource threshold, further investments in that technology are prohibited unless some power plants of the same technology are decommissioned.

$$\sum _i\sum _j{Cap}_{i,tech,j}\le {Cap}_{tech}^{limit}$$

(29)

Considering the continuous technological improvement and economic of scale, the cost of energy generation technologies is decreasing year by year. Equation (30) shows the average investment cost of a certain technology in the future through a learning-by-doing model.

$$\mathrm{l}\mathrm{n}({Inv}_{tech,t}^{avg})={\mathrm{l}\mathrm{n}(Inv}_{tech,t_0}^{avg})-{\sigma }_{tech}·\mathrm{l}\mathrm{n}\left({\displaystyle \frac{{\left(\sum _i\sum _j{Q}_{i,tech,j}\right)}_t}{{\left(\sum _i\sum _j{Q}_{i,tech,j}\right)}_{t_0}}}\right),$$

(30)

where σ_tech represent the experience index for the technology tech.

3.5.4. Carbon Module

Annually, the government assigns carbon quotas for the upcoming year to all power plants and establishes the subsidy policy for that year. Initially, the government calculates each power plant’s carbon emission allowances for the following year using the grandfathering allocation mechanism. The equation of the grandfathering allocation mechanism is given as [15]:

$$C_{i,tech,j,t+1}^{allowance}=\left(1-{\eta }_{rr}\right)·G_{i,tech,j,t}^{actual}·f_{i,tech,j,t}·\left(1-reservedrate\right).$$

(31)

The parameter η_rr refers to the percentage by which the total number of carbon permits issued is reduced each year, aiming to limit overall emissions. This parameter is critical for modelling the impact of emissions reduction policies, as it influences the availability of permits and drives changes in industry behaviour over time. It is set at 5 percent based on the literature [14,15].

In this study, it is assumed that the existing carbon tax rate is influenced by the cumulative quotas and emissions from the previous period. Given the absence of a carbon taxation system in Türkiye, this module explores a hypothetical scenario where such a mechanism is implemented exclusively for electricity generation, a major contributor to the nation’s greenhouse gas emissions. By focusing the carbon tax on the electricity sector, the study aims to assess its potential impact on emissions reduction and its implications for energy system dynamics within the Turkish context [62]. The existing carbon tax rate can be estimated as:

$${tax}_t^{carbon}={tax}_{t-1}^{carbon}·e^{\left({\lambda }^c·\left({Total emissions}_{t-1}-{Total allowance}_t\right)\right)}·\left(1+{ϵ}_t\right)~N\left(0,{0.01}^2\right),$$

(32)

where λ^c represents a proportional coefficient which is 0.005 [14]. It is the sensitivity coefficient that adjusts the carbon tax rate based on the difference between past emissions and current allowances. It determines how responsive the tax rate is to deviations in emissions from the allocated limits. Furthermore, a random disturbance denoted as ${ϵ}_t$ is included to account for various unpredictable factors (e.g., economic shocks, technological innovations, international dynamics, etc.). The initial value for the carbon tax rate, ${tax}_{t=0}^{carbon}$, is fixed at 75 $/ton CO_2-eq, in accordance with the average carbon tax rate in Europe in 2023 [63].

3.6. Data

In order to execute the proposed model effectively, a comprehensive array of data are compiled from diverse sources. The process involves the aggregation of information from multiple channels to ensure accuracy and reliability.

Table A2. Technical parameters of technologies.

	PV	Wind	Hydro	Geothermal	Biomass	Coal	Natural Gas	Nuclear	Reference
Fuel Consumption (ton/MWh)	-	-	-	-	-	0.4	0.26	0.08	[64,65]
CAPEX (M$/MW)	0.92	1.1	2.574	5.3875	4.332	4.9	0.975	11.2
Variable OPEX ($/MWh)	-	-	-	16.375	5.8	4.25	3.75	4.5
Fixed OPEX (k$/MW-year)	10.5	27.5	64	14.5	150.85	65.375	13.5	142
Carbon emission (gCO2/kWh)	-	-	-	-	-	900	460	-	[66]
Experience index	0.15	0.15	0.03	0.1	0.1	0.07	0.03	0.3	[15]
Construction time (year)	1	1	4	3	2	4	2	6
Life span (year)	30	30	100	30	45	30	30	60	[65]

(in Appendix A) [15,64,65,66] serves as a concise repository, delineating the collected data alongside their respective sources.

Future uranium prices and calorific values of fuels are sourced from [65], while future coal and natural gas prices are obtained from [67].

3.7. Assumptions and Scenarios

This study assumes that Türkiye’s CO₂ emissions, constituting approximately 1.2% of the global total, have a negligible impact on the global climate [10]. IPPs are presumed to strategically allocate investments in PV and wind projects in regions with the highest efficiency, using capacity factors derived from GCM projections to estimate power generation. Hydropower capacity is assumed to be nearly maximized at 38 GW, limiting new development opportunities [68]. Similarly, offshore wind projects are presumed to face significant barriers, further challenging the renewable energy transition. Nevertheless, IPPs are expected to continue advancing renewable energy solutions, with anticipated profit margins modelled as a normal distribution (mean: 10%, SD: 1) [69].

Capacity constraints also play a critical role. Türkiye’s PV systems are capped at 387 GW [70], wind energy at 83 GW [71], geothermal at 5 GW [72], and biomass energy at 9.5 GW [73].

In addition to these assumptions, one base scenario and nine policy scenarios will be studied to analyze the impact of energy policies on capacity additions, electricity prices, and CO₂ emissions based on electricity generation. In this regard, the features of these scenarios are presented in

Table 4. Features of policy scenarios.

Scenario	Carbon Tax	Renewable Subsidy	Corporation Tax Reduction for RES	Nuclear Power Plant
Base	-	-	-	-
1	-	5 $/MWh (escalated with inflation rate) at t = 0	-	-
2	75 $/ton CO2-eq at t = 0	-	-	-
3	75 $/ton CO2-eq at t = 0	5 $/MWh (escalated with inflation rate) at t = 0	-	-
4	-	-	10% reduction	-
5	-	-	40% reduction	-
6	-	-	-	4800 MW t = 3
7	-	5 $/MWh (escalated with inflation rate) at t = 0	-	4800 MW t = 3
8	75 $/ton CO2-eq at t = 0	5 $/MWh (escalated with inflation rate) at t = 0	-	4800 MW t = 3
9	75 $/ton CO2-eq at t = 0	Transfer of half of the carbon tax revenue as a subsidy for RES	-	4800 MW t = 3

.

In the model, subsidies are allocated solely to renewable energy sources, with a fixed amount of $5 per MWh for electricity generated, in accordance with Presidential Decree No. 7189, issued on 1 May 2023. This subsidy is applied uniformly across renewable technologies, with no differentiation based on capacity or efficiency. This approach ensures consistency and transparency in the economic mechanisms while incentivizing renewable energy generation.

In Scenarios 6–9, the nuclear power plant in Akkuyu, Mersin, is modelled with a capacity of 4800 MW, reflecting the ongoing construction project. To account for potential uncertainties, such as construction delays, public opposition, and geopolitical factors, we have assumed a delayed operational start, with the plant beginning operation in three years instead of in the originally planned two years. This conservative assumption ensures the robustness of the model under varying circumstances and aligns with the uncertainties commonly associated with large infrastructure projects.

Although Türkiye is not an EU member, applying a carbon tax of $75/ton CO₂ can be justified by referencing the average carbon price in the EU, which reflects current market conditions and the environmental cost of carbon emissions. This approach provides a benchmark for evaluating potential impacts on electricity prices and emissions reductions, offering a comparative analysis that helps understand how a similar policy could influence Türkiye’s energy sector. Adopting this rate allows for a consistent assessment of potential economic and environmental outcomes, despite Türkiye’s non-membership in the EU.

4. Results and Discussion

4.1. Climate Projection

After obtaining the results of KGE, md, and nRMSE for each grid, all GCMs are ranked for each climate variables using the MCDA method (See

Table A3. MCDA results for each climate variable and performance criterion of the models.

		M1	M2	M3	M4	M5	M6	M7	M8	M9	M10	M11	M12	M13
RSDS	KGE	37.854	38.616	9.809	18.417	14.825	11.361	16.631	42.045	67.704	52.396	14.690	33.367	17.539
	md	40.265	33.807	14.755	14.861	12.308	12.971	13.067	45.093	72.987	47.002	14.803	42.161	11.176
	nRMSE	44.890	41.898	9.995	14.820	12.146	13.883	15.669	40.693	76.595	44.477	13.918	31.474	14.798
TAS	KGE	25.792	34.153	34.857	31.659	29.364	41.905	32.337	9.686	28.443	41.922	25.616	20.408	19.112
	md	36.151	54.162	30.420	29.462	36.214	27.557	16.853	28.065	29.314	35.401	12.656	26.033	12.968
	nRMSE	32.724	42.571	45.371	27.796	34.250	29.017	24.815	9.563	29.849	39.559	12.045	33.203	14.493
SFCWIND	KGE	61.167	29.524	9.139	27.731	29.663	13.932	56.596	15.241	30.607	26.669	25.766	14.502	34.718
	md	75.553	24.913	9.182	24.533	25.700	13.341	60.061	13.711	25.672	23.314	25.283	12.334	41.660
	nRMSE	52.423	21.354	9.083	30.802	27.594	14.102	59.612	14.007	36.534	34.289	19.907	11.398	44.149

in Appendix A). The rankings of GCMs for each climate variable and performance criterion are presented in

Table 5. MCDA results rankings of models for each climate variable and performance criterion.

		M1	M2	M3	M4	M5	M6	M7	M8	M9	M10	M11	M12	M13
RSDS	KGE	5	4	13	7	10	12	9	3	1	2	11	6	8
	md	5	6	9	7	12	11	10	3	1	2	8	4	13
	nRMSE	2	4	13	8	12	11	7	5	1	3	10	6	9
TAS	KGE	9	4	3	6	7	2	5	13	8	1	10	11	12
	md	3	1	5	6	2	9	11	8	7	4	13	10	12
	nRMSE	6	2	1	9	4	8	10	13	7	3	12	5	11
SFCWIND	KGE	1	6	13	7	5	12	2	10	4	8	9	11	3
	md	1	7	13	8	4	11	2	10	5	9	6	12	3
	nRMSE	2	8	13	6	7	10	1	11	4	5	9	12	3

.

The selection of GCMs was performed using three performance criteria—KGE, nRMSE, and md—all weighted equally based on their significance in assessing model accuracy and bias. This equal weighting approach, consistent with findings in the literature, ensures that no single metric disproportionately influences the final rankings [41,74]. The MCDA method was employed to aggregate the rankings, resulting in a balanced and comprehensive evaluation across all grid points. This methodology guarantees that the selected GCMs represent the most reliable models for simulating Türkiye’s unique climate conditions, supporting robust projections for policy analysis.

As shown in Figure 1, the comprehensive ranking method was applied to identify the most suitable GCMs for simulating Türkiye’s climate data (See Equation (1)). The top-performing climate models for Türkiye, as shown in

Table 6. The most successful climate models for modelling Türkiye.

	M1	M2	M3	M4	M5	M6	M7	M8	M9	M10	M11	M12	M13
MR	0.709	0.641	0.291	0.453	0.462	0.265	0.513	0.350	0.675	0.684	0.248	0.342	0.368
Rank	1	4	11	7	6	12	5	9	3	2	13	10	8

, include ACCESS-CM2, INM-CM5-0, INM-CM4-8, and ACCESS-ESM-1-5. These models were selected based on their consistently superior performance across key metrics (KGE, nRMSE, md), ensuring robust and reliable climate projections. Their ability to accurately represent Türkiye’s diverse climatic conditions makes them the most appropriate choices for this study.

After detecting the top four GCMs, the Extreme Gradient Boosting Tree algorithm is used to assemble the outputs of these GCMs. The future projection of climate variable for each grid is obtained by assembling the SSP5-85 scenario data of the selected climate models. Projections are made for the years between 2023 and 2040.

The projections show how the future of a variable changes compared to its current state. Figure 4 shows the changes in the averages of the energy potential for the 2025–2030, 2031–2035, and 2036–2040 periods compared to the average of that during the years 1985–2014, as a percentage for each grid.

It is predicted that there will be a decrease in the amount of electricity produced from solar power plants throughout Türkiye between 2031 and 2035 due to loss of efficiency with increasing temperatures. The biggest decrease is expected in the Mediterranean Region and Eastern Black Sea Region. However, since the solar potential of the Eastern Black Sea Region is still very low today, it is not an economically suitable region for the installation of photovoltaic solar power plants. The results of this study suggest that this region is not suitable for photovoltaic solar power plants in the future. The regions where electricity production from photovoltaic solar power plants increased the most are the Marmara Region (especially Thrace) and the Western Black Sea Region. The results obtained from this study are parallel to other studies in the literature [57,75].

When the electricity produced from wind turbines is examined, an increase in wind production is expected, especially in Thrace, the northern parts of Central Anatolia (around Çorum, Tokat cities), and around the cities of Ağrı and Van in the Eastern Anatolia region. However, there appears to be a decrease in wind production potential in the Eastern Black Sea, the cities of Uşak, Kütahya, Eskişehir, and Bolu in western Anatolia and the cities of Mardin, Batman, and Şırnak in Southeast Anatolia. These results are parallel to those of the previous studies in the literature [76].

The average CDDs for each city in Türkiye are computed across three distinct timeframes utilizing temperature projections of the GCM (See Figure 5). It becomes glaringly evident that the average CDDs are poised to surge significantly across nearly all cities, particularly in the Mediterranean region and the south-eastern portion of Türkiye, owing to the impacts of global warming. It can be revealed that these findings are in agreement with those in the literature (i.e., [77,78]).

4.2. AHP and Utility Function

As previously discussed in Section 3.5.3, the weights for the utility functions are established through AHP analysis, utilizing the insights of 11 experts from academia (professors and researchers) and private sector (energy and environment) to determine the importance of each environmental criterion. While there is no strict minimum for the number of experts needed, increasing the group size can provide a broader range of perspectives, enhance the reliability of results, and reduce the impact of individual biases [79]. The experts are chosen based on their academic backgrounds (primarily in environmental engineering, energy science, and environmental science), research experience, quality and quantity of publications, experience in the sector, field reputation, and the relevance of their expertise to the focus of the AHP analysis.

Since the Consistency Ratio (CR) with the value of 0.088 is lower than the threshold value (0.1) given by Saaty [43], it can be concluded that the result of the AHP analysis is reliable.

Depending on the AHP analysis,

Table 7. Weights of utility functions.

Utility	Weight
Return of Investment	0.755
Social acceptance	0.150
Environmental impact	0.095

shows the weights of each utility functions. As expected, the economic criteria are detected as the most important factor with a weight of 0.755.

Considering the MAUT, after the ranges and units of utility functions of environmental impact and social acceptance, the utility scores are determined by these experts.

Table 8. Utility scores for each quartile.

Criteria/Score	Range	0	0.25	0.5	0.75	1.0
Environmental	13.45–936	13.45	290.12	420.21	520.34	936
Social acceptance	1.76–23.21	1.76	7.02	11.42	17.42	23.21

presents the utility scores, and Figure A1 (in the appendices) shows the utility curves.

As a result of utility curves established based on the MAUT and the weights resulted from AHP analysis, the combined utility scores of environmental impact and social acceptance of each electricity generation technology are given in

Table 9. Combined utility scores of electricity generation technologies.

Technology	Combined Utility Scores
Wind	0.245
Geothermal	0.220
PV	0.210
Hydro	0.157
Biomass	0.137
CCGT	0.081
Coal	0.008
Nuclear	0.094

considering Equation (33) which is a part of Equation (31).

$$0.150\left(0.0468x_{soc}-0.0686\right)+0.095\left(-0.0011x_{env}+0.9946\right),$$

(33)

where x_soc and x_env are scores of environmental impact and social acceptance given in Table 9, respectively.

4.3. Agent-Based Simulation

The results of 10 different energy-climate policy scenarios, implemented using a mathematically described agent-based simulation model outlined in Section 3.5, are presented in this section.

In this context, Figure 6 displays the future electricity demand output of the ABM for the base scenario. Considering variations in CDDs (a proxy for temperature change), electricity prices, income, and population, the future electricity demand is projected to rise to 456.2, 521.4, and 571 TWh in 2030, 2035, and 2040, respectively. It can be said that electricity demand almost has a linear trend.

Figure 7 illustrates the installed capacities required to meet Türkiye’s electricity demand across various policy conditions. Under the base scenario, it is projected that the installed PV capacity will reach 28.7, 50.7, and 79.5 GW by 2030, 2035, and 2040, respectively. This suggests an almost tenfold increase in the current installed PV capacity by 2040. Moreover, across all policy scenarios, PV technology emerges as the most favoured choice for IPPs.

The anticipated growth in wind power capacity is also noteworthy across all scenarios. In each scenario, the installed wind power capacity is expected to increase by 1.5-fold every five years. These figures indicate that solar and wind power systems will form the cornerstone of future electricity generation investments. Conversely, biomass, hydroelectric, and geothermal power plants exhibit limited expansion potential compared to PV and wind power systems due to capacity constraints.

Regarding fossil fuel-based power plants, the capacity expansion in natural gas power plants surpasses that of coal power plants. Under the base scenario, the installed capacity of natural gas power plants is anticipated to rise to 27.7, 36.1, and 47.2 GW by 2030, 2035, and 2040, respectively, whereas the installed capacity of coal power plants remains largely unchanged after 2030. This indicates that although coal power plants persist as the primary base-load power sources, the significance of natural gas power plants for load-following and peak demand will significantly rise owing to the intermittent electricity generation from wind and solar power plants.

Figure A2 (in the Appendix A) illustrates the capacity distribution among electricity generation technologies for each scenario. Without governmental policy influence on the electricity market, it is projected that wind and solar power plants’ combined installed capacity will represent half of Türkiye’s total installed capacity by 2040.

However, each policy exerts a distinct influence on capacity development across various timeframes. It is observed that renewable energy subsidies boost capacity additions in wind and solar power plants in the short and medium terms, but their impact diminishes in the long term. Additionally, carbon-tax systems foster a more consistent growth trajectory for wind and solar power plants. Despite these policy variances, the share of RES capacity is forecasted to not surpass 71 percent across all scenarios.

Carbon taxing within the energy sector operates as a market-driven approach designed to encourage carbon emission reduction by levying taxes on the carbon content of fossil fuels utilized in energy generation. Figure 8 depicts the progression of carbon taxes across Scenarios 3, 4, and 9. There is an anticipation that by 2040, the carbon tax could potentially exceed $271.1 per ton of CO₂ if implemented independently, without integration with other policy measures. If carbon taxing is coupled with other policy instruments, there is a possibility for the carbon tax to reach $257.3 per ton of CO₂.

Electricity prices are projected to undergo a significant decrease until 2029, driven by capacity expansions (See Figure 9). Subsequently, they are anticipated to stabilize across all scenarios until the end of the forecast period. The most favourable electricity rates are forecasted when both carbon tax and RES subsidy policies are implemented concurrently. Conversely, in Scenario 6, where there are no policy interventions except for the integration of a nuclear power plant into the grid, the highest prices are expected.

Taking into account capacity installations and energy–climate policies, Figure 10 illustrates the annual and cumulative electricity generation-based CO₂ emissions. In the baseline scenario, it is anticipated that annual CO₂ emissions will peak at 174.6 million tons in 2032, followed by a decline as the renewable energy capacity increases. By 2040, annual CO₂ emissions are projected to decrease to 118 million tons without any policy interventions. However, the influence of energy–climate policies on CO₂ emissions is unmistakable. Through the implementation of appropriate policies, there is a clear potential to reduce cumulative CO₂ emissions during the period of 2022–2040 by more than 11% compared to the baseline scenario.

While this study primarily focuses on assessing renewable energy potential and simulating policy implementation, it is important to acknowledge the variability of wind and solar resources and their implications for grid stability. Established studies emphasize that hybrid systems, which combine wind, solar, and energy storage technologies, are effective in mitigating intermittency and ensuring a stable energy supply. Additionally, grid modernization and smart grid technologies play a critical role in managing the integration of variable renewable energy sources. Future studies can build on the findings of this research by exploring these integration challenges in greater depth and evaluating their applicability to Türkiye’s unique energy landscape. Such efforts would provide a more holistic approach to renewable energy development, addressing both resource potential and system reliability.

4.4. Discussion

This study examines the potential impact of climate-energy policies on electricity demand, renewable electricity generation, capacity expansions, electricity prices, and CO₂ emissions from electricity generation in Türkiye, taking into account the influence of future climate change. As the initial step, the top four GCMs capable of accurately simulating Türkiye’s unique climate conditions are detected applying three methods, namely nRMSE, KGE, and md. Out of all GCMs, ACCESS-CM2, INM-CM5-0, INM-CM4-8, and ACCESS-ESM-1-5 have been identified as the most promising models. Consequently, they are employed to predict the future climate of Türkiye under SSP5-8.5 climate scenario, which is the most pessimistic scenario (business-as-usual scenario).

Considering the future climate condition of Türkiye, it can be revealed that increasing temperature due to the climate change will impact the electricity demand for space cooling especially in Mediterranean and south-east regions of Türkiye. To decrease the impact of space cooling needs on the electricity demand, the government should take some steps including promoting energy-efficient building practises, offering incentives for sustainable cooling technologies, and raising public awareness about efficient cooling habits to reduce electricity demand for cooling purposes. By incorporating passive thermal management systems like thermochromic smart windows (TSW) and daytime passive radiative coolers (DPRC), both significant technologies in this domain, it becomes feasible to conserve up to 17% of electricity typically consumed for air conditioning [80].

This study examines one foundational scenario alongside nine policy scenarios to assess how energy policies influence capacity expansions, electricity pricing, and CO₂ emissions related to electricity generation. In all policy scenarios, there is notable growth in solar and wind power plant capacities. As renewable energy generation grows, particularly from wind and solar, advanced grid management systems are essential. Policies that encourage the adoption of smart grid technologies and demand response (DR) can improve supply–demand efficiency, reduce fossil fuel dependence during peak periods, and enhance grid stability. Offering incentives for utilities and consumers to invest in smart metres, real-time monitoring, and flexible pricing can boost energy efficiency, lower costs, and cut CO₂ emissions.

Following the capacity expansions in RES, 2032 marks a significant milestone where electricity generation-based CO₂ emissions peak and then begin to decline across all scenarios. However, the extent of the impact of each energy-climate policy on CO₂ emissions varies. To minimize cumulative CO₂ emissions, the deployment of nuclear power plants alongside the implementation of both carbon taxing and RES subsidies by the government is recommended. In this scenario, the government could offer a subsidy adjusted for inflation in addition to carbon taxing.

While carbon taxing has the potential to reduce cumulative CO₂ emissions by 1.52% compared to the base scenario, RES subsidies may decrease CO₂ emissions by 4.14%. If both policies are implemented simultaneously, CO₂ emissions could decrease by over 6%, surpassing the total reduction potential of carbon taxing and RES subsidies individually. Combining these policies with the deployment of nuclear power plants could result in a reduction in cumulative CO₂ emissions by over 11% throughout the projection period. On the other hand, the impact of corporation tax reductions for RES on CO₂ emissions is negligible. Therefore, the study suggests effect of carbon taxing could be enhanced if complemented by additional policy mechanisms such as carbon credits or a cap-and-trade system. Implementing a more comprehensive carbon pricing framework, combining both carbon taxes and emissions trading schemes, would create a stronger financial incentive for industries to reduce emissions. This can stimulate the adoption of cleaner technologies, including renewable energy and energy-efficient systems, and further accelerate the transition to a low-carbon economy.

Deploying nuclear power plants without any climate–energy policy implications could reduce cumulative CO₂ emissions by 5.3%, underscoring the importance of nuclear power in emissions reduction efforts. However, due to the significant initial investment costs and low social acceptance rates, IPPs are hesitant to invest in nuclear power plants. To address this, the government could take an active role in investing in nuclear power plants, either through direct state involvement or by creating a public–private partnership model to reduce the financial burden on IPPs. Additionally, the government could incentivize research and development in advanced nuclear technologies to enhance safety and public acceptance, ensuring nuclear energy plays a role in Türkiye’s low-carbon future.

Another outcome of this study involves forecasting electricity prices while considering the influence of energy-climate policies. With the ongoing increase in capacity additions, particularly in RES, electricity prices are projected to undergo a significant decline until 2029, followed by a period of stability, fluctuating between $25–31/MWh across all scenarios. While reductions in corporation taxes for RES have a minor effect on lowering prices, RES subsidies hold substantial potential to reduce electricity prices, as RES power plants can offer lower bids under such circumstances. Additionally, the deployment of nuclear power plants and the implementation of RES subsidies adjusted for inflation emerge as the most effective combination for simultaneously minimizing both CO₂ emissions and electricity prices.

Validating the results through comparison with other studies or governmental projections is of utmost importance. In this context, it is feasible to juxtapose the findings of this study with the National Energy Plans issued in 2022 by the Ministry of Energy and Natural Resources [81]. While this study predicts total installed capacities ranging from 147.6 to 162.8 GW for 2030, and 190.2 to 204.3 GW for 2035, MENR anticipates total capacities of 149.1 GW and 189.7 GW for the same time periods, respectively. Furthermore,

Table 10. Comparison of capacity projections for 2035 (GW).

Scenario	Solar	Wind	Hydro	Other	Natural Gas	Coal	Nuclear	Total
Base	50.67	31.04	37.35	10.34	36.14	27.41	0	192.96
1	52.85	31.85	31.64	10.84	37.63	25.43	0	190.24
2	53.98	31.03	37.35	10.34	36.09	26.46	0	195.24
3	56.38	32.12	31.64	10.85	36.93	25.13	0	193.06
4	62.07	33.45	31.72	10.5	40.21	26.34	0	204.28
5	62.17	33.45	31.72	10.5	41.87	26.34	0	204.28
6	50.11	31.89	31.72	11.09	41.87	26.16	4.8	197.63
7	64.01	30.78	31.64	10.47	33.59	25.56	4.8	200.84
8	64.11	30.75	31.64	10.5	33.47	25.42	4.8	200.71
9	56.21	33.4	31.72	10.93	37.44	25.29	0	194.96
MENR	52.9	29.6	35.1	7.5	35.5	24.3	4.8	189.7

illustrates a detailed comparison of capacities between the results of this study and the National Energy Plans for 2035. It can be observed that the outputs of the ABM and the MENR projections exhibit close alignment. This underscores the robustness of the established GCM-ABM framework in estimating capacity additions, considering future electricity demand and projections from wind and solar power plants. Consequently, this model holds promise for estimating the impact of various energy-climate policies, beyond those explored in this study, on electricity prices, capacity additions, and electricity generation-based CO₂ emissions.

This study’s methodologies and findings hold significant potential for global application, transcending the context of Türkiye. By integrating climate projections and simulating diverse policy scenarios through an ABM, the research provides a flexible framework adaptable to various regions with different climatic, economic, and regulatory conditions. This adaptability ensures the transferability of insights, allowing countries to analyze how specific policies interact with their unique energy systems under evolving climate conditions. The ability of the model to simulate policy interventions, such as carbon taxes or renewable energy incentives, enables stakeholders to evaluate potential synergies and trade-offs.

The findings of this study align with Türkiye’s National Energy Plans, particularly the goal to achieve 75% renewable energy capacity by 2035. Additionally, the implementation of carbon taxation and renewable subsidies, as simulated in this study, supports the international commitment to reduce CO₂ emissions under the Paris Agreement. By integrating the technical results with practical energy policy objectives, this study bridges the gap between modelling outcomes and actionable policy recommendations.

To illustrate, the adoption of a carbon tax at $75/ton CO₂ aligns with the average EU carbon pricing level, positioning Türkiye’s policies on a competitive global stage. Similarly, renewable subsidies modelled in this study reflect the successful implementation strategies seen in nations like Germany, where such incentives have led to substantial renewable capacity growth. These parallels emphasize the feasibility and scalability of the proposed policies.

Furthermore, the projected growth of solar and wind capacities directly addresses the variability challenges in Türkiye’s renewable energy grid. By highlighting the role of hybrid systems, grid modernization, and policy-driven investment in energy storage, the study provides a roadmap for ensuring grid stability while maximizing renewable integration. This practical approach ensures that the recommendations are not only theoretically sound but also implementable within the existing policy and infrastructure framework.

Türkiye’s energy policy evolution offers crucial insights for shaping future strategies. The implementation of feed-in tariffs through Law No. 5346 in 2005, along with its 2010 amendments, significantly boosted the adoption of renewable energy, especially in solar and wind sectors. Despite these achievements, gaps remain, such as the lack of a national carbon tax and insufficient incentives for hybrid systems. Drawing from these lessons, this study proposes expanding subsidies to include energy storage and hybrid technologies, alongside the gradual introduction of carbon taxation. These recommendations mirror effective global practises, such as Germany’s Energiewende, and outline a pathway for Türkiye to meet its ambitious renewable energy goals.

By providing a clear connection to international agreements such as the Paris Agreement and Türkiye’s Nationally Determined Contributions (NDCs), the study underscores its global relevance. It also highlights Türkiye’s potential to act as a regional leader in renewable energy adoption, serving as a case study for other developing countries in the Mediterranean and Middle Eastern regions. These insights enhance the study’s applicability and provide policymakers with evidence-based strategies to achieve long-term sustainability goals.

Moreover, this study makes valuable contributions to global climate strategies by showcasing how localized analyses can inform broader international efforts to combat climate change. The insights gained into the impact of climate change on energy demand and supply, combined with the evaluation of policy effectiveness, offer lessons that extend beyond regional boundaries.

5. Conclusions

This study evaluates the potential impacts of climate–energy policies on electricity demand, renewable electricity generation, capacity expansions, electricity prices, and CO₂ emissions in Türkiye, accounting for future climate change influences. By integrating multi-model ensemble climate projections with an ABM, this research provides a comprehensive assessment of long-term climate–energy interactions, including the economic effects of climate change-induced natural hazards on income.

Key findings include:

• Climate Model Selection: ACCESS-CM2, INM-CM5-0, INM-CM4-8, and ACCESS-ESM-1-5 are identified as the most suitable GCMs for simulating Türkiye’s unique climate conditions.
• Solar Efficiency and Cooling Needs: Rising temperatures reduce solar power plant efficiency, while CDDs increase across most cities, particularly in the Mediterranean and south-eastern regions.
• Electricity Demand: Demand is projected to rise to 456.2 TWh in 2030, 521.4 TWh in 2035, and 571 TWh in 2040, driven by CDDs, electricity prices, income, and population growth.
• Renewable Energy Constraints: RES shares are unlikely to exceed 71%, with fossil fuel plants remaining essential as baseload and load-following sources. Coal is projected to remain the primary baseload source, while natural gas will play a crucial role in addressing intermittent renewable generation.
• CO₂ Emissions Trends: RES capacity expansion will lead to a peak in electricity generation-based CO₂ emissions by 2032, followed by a decline. Proper policy implementation could reduce cumulative CO₂ emissions by over 11% from 2022 to 2040 compared to the baseline.
• Electricity Prices: Prices are expected to decline significantly until 2029, stabilizing afterward. The most effective strategies involve combining carbon tax and RES subsidy policies, with optimal results achieved through nuclear power deployment and inflation-adjusted RES subsidies.

The proposed GCM-ABM framework effectively evaluates the implications of diverse policy scenarios on electricity prices, capacity additions, and CO₂ emissions. However, certain limitations remain. Expanding the number of GCMs used could improve the robustness of results. Additionally, the study excludes the impact of electric vehicles (EVs) on electricity demand due to their negligible share in Türkiye. Future research could integrate EV-related demand projections as their adoption increases and data reliability improves.