Subnational energy planning for Turkey’s net-zero transition

Gulaydin, Oguzhan ORCID: https://orcid.org/0000-0002-1795-7939 2025. Subnational energy planning for Turkey’s net-zero transition. PhD Thesis, Cardiff University. Item availability restricted.

	PDF - Accepted Post-Print Version Restricted to Repository staff only until 6 May 2027 due to copyright restrictions. Download (7MB)
	PDF (Cardiff University Electronic Publication Form) - Supplemental Material Restricted to Repository staff only Download (315kB)

Abstract

As awareness of climate change grows, net-zero pathways have become a priority for researchers and policymakers. Achieving these targets requires aligning national climate commitments with spatially varied energy planning, yet most countries set overall capacity targets without systematically considering regional differences in de mand or renewable resource availability. The mismatch between nationally aggregated targets and regionally diverse energy conditions is especially pronounced in emerging economies, where data limitations restrict traditional forecasting methods and infrastruc ture development varies across provinces. Effective decarbonisation requires analytical frameworks that operate at subnational levels where infrastructure decisions are actually made and that remain underdeveloped in data-scarce contexts. Turkey illustrates these challenges: a large, fast-growing economy with diverse climate zones, significant renew able resources located away from demand centres, and ambitious decarbonisation goals that must be balanced against heavy import dependence and ongoing demand growth. This dissertation addresses the mismatch between aggregate national planning and provincial realities by analysing both supply-side renewable potential and demand-side consumption patterns across Turkey’s 81 provinces. The dissertation follows a paper-based format, comprising three self-contained studies. The first offers a meta-analysis of renewable energy potential across solar, wind, hydropower, geothermal, and biomass, quantifying implementation gaps and assessing deployment barriers. The second creates a machine learning framework for predicting provincial residential electricity demand to 2050 under Shared Socioeconomic Pathway scenarios, comparing tree-based, artificial neural network, and kernel-based methods. The third expands this framework to residential natural gas, testing its transferability under severe data constraints, where ten years of observations support projections extending 2.5 times beyond the training period. The renewable assessment shows significant underutilisation: solar generation is only 6.6% of the estimated potential, wind capacity is at 26%, and climate variability increasingly threatens hydropower reliability, with reservoir levels in south-western basins declining by over 40% since 2010. The geographic mismatch between ideal generation sites and demand centres worsens these issues. Technical, economic, and regulatory barriers together explain why plentiful resources remain underused. For demand forecasting, Random Forest achieves the highest accuracy (R2 = 0.936 for electricity), outperforming artificial neural networks and gradient boosting methods that exhibited overfitting under data constraints. Feature importance analysis shows that demographic variables (population, household composition, and GDP) exert a greater influence on residential demand than climatic indicators. The same pattern emerges for both electricity and natural gas, despite the latter being heating-dominated. Residential electricity demand is projected to increase by 78% to approximately 117 TWh by 2050, while natural gas demand nearly doubles to 358–463 TWh, with associated emissions of 74–89 Mt CO2-equivalent annually. Provincial trajectories vary widely: some regions show growth exceeding 270%, while others are below 30%. For gas forecasting, algorithmic uncertainty exceeds scenario uncertainty, indicating that methodological choices matter more than socioeconomic assumptions when data is limited. The dissertation offers a transferable framework for subnational forecasting in data-limited contexts, showing that simpler ensemble methods outperform complex ar chitectures under extreme extrapolation. The discovery that demographic factors surpass climatic ones challenges existing assumptions in energy demand modelling. Provincial projections highlight spatial heterogeneity that national averages conceal, emphasising that subnational detail is crucial rather than optional for effective infrastructure planning and decarbonisation strategies.

Item Type:	Thesis (PhD)
Date Type:	Completion
Status:	Unpublished
Schools:	Schools > Engineering
Uncontrolled Keywords:	1. Subnational energy planning 2. Net-zero transition 3. Machine learning 4. Residential energy demand forecasting 5. Renewable energy potential 6. Turkey
Date of First Compliant Deposit:	6 May 2026
Last Modified:	06 May 2026 13:12
URI:	https://orca.cardiff.ac.uk/id/eprint/186671

Actions (repository staff only)

Edit Item