UNIVERSIDAD DE COSTA RICA SISTEMA DE ESTUDIOS DE POSGRADO ROBUST ENERGY SYSTEM PLANNING FOR DECARBONIZATION UNDER TECHNOLOGICAL UNCERTAINTY: FROM TRANSPORT ELECTRIFICATION TO POWER SYSTEM INVESTMENTS. Tesis sometida a la consideración de la Comisión del Programa de Posgrado en Ingeniería Eléctrica para optar al grado y título de Maestría Académica en Ingeniería Eléctrica LUIS FERNANDO VICTOR GALLARDO Ciudad Universitaria Rodrigo Facio, Costa Rica 2022 Dedication For my Family, Thank you ii Acknowledgements I express my gratitude to my supervisor Prof. Jairo Quirós-Tortós. He has provided me with nu- merous learning opportunities at the Electric and Power Energy Research Laboratory at the University of Costa Rica. I thank all the professors from the School of Electrical Engineering and the Postgraduate Program for their dedication to us, the students. In particular, I want to thank my advisors, Dr. Gustavo Valverde and Mag. Tony Méndez, for their help in completing and assessing this research. I also thank Adriana Carvajal and Dr. rer. Nat. Francisco Siles for their patience and kindness when I asked for advice. I also thank the professionals who have been part of the research projects that fed this thesis. Finally, I thank my mother Marta, my father Luis, my sister Natalia, and my girlfriend Daniela. Their love and support sustain me through difficulties and make me try to give my best always. I also thank my grandmothers Isabel and Odilia, who passed away while I attended this program. They shaped the present and future of my family. iii Esta tesis fue aceptada por la Comisión del Programa de Estudios de Posgrado en Ingeniería Eléctrica de la Universidad de Costa Rica, como requisito parcial para optar al grado y título de Maestría Académica en Ingeniería Eléctrica. Dr. Oscar Núñez Mata Representante de la Decana Sistema de Estudios de Posgrado Dr. Jairo Quirós Tortós Director de Tesis Dr. Gustavo Valverde Mora Asesor M.Sc. Tony Alonso Méndez Parrales Asesor Dr. Mauricio Espinoza Bolaños Representante del Director del Programa de Posgrado Luis Fernando Victor Gallardo Sustentante iv Table of contents Cover page i Dedication ii Acknowledgements iii Hoja de Aprobación iv Table of contents v Abstract viii Resumen ix List of tables x List of figures xi 1 Introduction 1 1.1. State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 1.1.1. Strategies for the Long-term Energy Transition . . . . . . . . . . . . . . . . . . . 4 1.1.2. Applications, Capabilities, and Limitations of ESOMs . . . . . . . . . . . . . . . 6 1.1.3. Robustness for the Energy Transition . . . . . . . . . . . . . . . . . . . . . . . . 7 1.2. Justification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 1.3. Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 1.4. Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 1.5. Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 1.5.1. Main Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 1.5.2. Specific Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 2 Energy System Modeling and Robust Decision Making 12 2.1. Energy System Optimization Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 v 2.1.1. Modeling Demand in ESOMs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 2.2. Robust Decision Making . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 2.3. Transfers in Energy Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 3 Methodology 25 3.1. Scenarios and Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 3.1.1. Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 3.1.2. Versions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 3.1.3. Relationships . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 3.1.4. Ranking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 3.2. Transfer Estimation Module . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 3.2.1. Electricity Prices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 3.2.2. Bus Prices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 3.2.3. Taxes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 3.3. Model Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 3.3.1. Wide Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 3.3.2. Narrow Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 3.3.3. Tax Adjustment Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 3.4. Robustness Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 3.4.1. Hierarchical Patient Rule Induction Method . . . . . . . . . . . . . . . . . . . . . 47 4 Results and analysis 50 4.1. National Costs, Benefits, and Emissions . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 4.2. Ranking Policy Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 4.3. Uncertainty Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 4.3.1. Power Sector Impacts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 4.4. Actor Impacts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 4.4.1. Fiscal Impacts and Tax Reform . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 4.5. Decision Insights for Robust Energy Planning . . . . . . . . . . . . . . . . . . . . . . . . 75 4.5.1. Robust Pathways for Nationwide Benefits . . . . . . . . . . . . . . . . . . . . . . 75 4.5.2. Robust Pathways for Nationwide Prices . . . . . . . . . . . . . . . . . . . . . . . 77 4.5.3. Robust Pathways for Nationwide Emissions . . . . . . . . . . . . . . . . . . . . . 78 vi 4.5.4. Robust Pathways for Nationwide CAPEX . . . . . . . . . . . . . . . . . . . . . . 79 4.5.5. Robust Pathways Nationwide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 4.5.6. Robust Pathways per Actor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 5 Conclusions, recommendations, and future work 85 5.1. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 5.1.1. National Costs, Benefits, and Emissions . . . . . . . . . . . . . . . . . . . . . . . 86 5.1.2. Ranking Policy Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 5.1.3. Power Sector Impacts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 5.1.4. Fiscal Impacts and Tax Reform . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 5.1.5. Actor Impacts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90 5.1.6. Robust Drivers Nationwide and per Actor . . . . . . . . . . . . . . . . . . . . . . 91 5.2. Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94 5.2.1. Invest in this Decade . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94 5.2.2. Make Transport Electrification a Priority . . . . . . . . . . . . . . . . . . . . . . 94 5.2.3. Decouple Transport from Economic Growth Soon . . . . . . . . . . . . . . . . . . 94 5.2.4. Develop and International Strategy for Freight Transport . . . . . . . . . . . . . 94 5.2.5. Design a Reform with Progressive Taxes and Externality Pricing . . . . . . . . . 95 5.2.6. Take Advantage of Existing Power Assets . . . . . . . . . . . . . . . . . . . . . . 95 5.2.7. Finance Assets at Low Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 5.2.8. Price Services with Lifetime Perspective . . . . . . . . . . . . . . . . . . . . . . . 95 5.2.9. Search for (or Develop) Assets with Low Unit Costs . . . . . . . . . . . . . . . . 96 5.3. Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 6 Bibliography 98 Appendix A Cost Trajectories of Energy Infrastructure 114 Appendix B Patient Rule Induction Method per Actor 116 Appendix C Transport Energy Consumption and Fleet 119 Appendix D Validation of the Hierarchical PRIM 124 Appendix E Wide Experiment Values 126 vii Abstract This work develops energy system modeling tools that identify features of a robust energy policy: a policy that performs well relative to alternatives. The tools are based on the Open Souce Modeling System (OSeMOSYS), are named the Multipurpose OSeMOSYS-based Framework (MOMF), and are applied to Costa Rica´s energy transition through the lens of its National Decarbonization Plan (NDP). The MOMF can support energy decarbonization planning exercises, and it is suitable to address the uncertainty involved in a decades- long process. It compares possible NDP futures -quantitative combinations of uncertainties and sectoral policy objectives- to a business-as-usual (BAU) scenario without decarbonization. The MOMF also evaluates actors within a country, including the fiscal impacts of decarbonization, following the best practices of applied energy modeling for policy support. This work finds that the NDP has high economic benefits (avoided costs relative to the BAU) in the long term, equivalent to 5.5% of GDP yearly in the 2041-2050 decade. In 2031-2040, the benefits are 0.8% of GDP yearly; in 2022-30, the NDP faces net costs (more costs than the BAU) of 0.9% of GDP yearly. These results are averages across futures and can be higher or lower. The government will have lower direct tax revenue of about 0.87% of GDP yearly in 2041-2050 and will need to redistribute benefits to compensate for this. It can use vehicle-kilometer taxes (VKT), property taxes, or energy taxes for the redistribution, mainly taxing private transport owners -who have the highest benefits-. However, to facilitate the decarbonization of freight firms in 2022-2030 and 2031-2040, the government could subsidize their zero-emission vehicles (ZEV) adoption. High benefits, low emissions complying with net-zero targets, and low electricity and public transport prices are desirable policy outcomes. Low costs for ZEVs and energy infrastructure -including renewables and storage- are crucial uncertain conditions for desirable outcomes. The robust levers the government can adopt to achieve desirable outcomes must decouple economic growth from transport activity. The specific levers include public transport investments, digitalization, non-motorized transport, ride-sharing, logistics hubs, and city densification. Moreover, low electricity prices need a low cost of capital to finance investments in the power sector. viii Resumen Este trabajo desarrolla herramientas de modelado de sistemas energéticos que identifican las características de una política energética robusta: una política con buenas métricas de desempeño en relación con alternativas. Las herramientas se basan en el Sistema de Modelado de Código Abierto (OSeMOSYS, por sus siglas en inglés), se denominan Multipurpose OSeMOSYS-based Framework (MOMF) y se aplican a la transición energética de Costa Rica plasmada en su Plan Nacional de Descarbonización (PND). El MOMF puede respaldar ejercicios de planificación de descarbonización energética y es adecuado para abordar la incertidumbre que implica un proceso de décadas. También compara posibles futuros del PND -combinaciones cuantitativas de incertidumbres y objetivos de políticas sectoriales- con un escenario de negocio habitual (BAU, del inglés business-as-usual) equivalente sin descarbonización. El MOMF también evalúa a los actores dentro de un país, incluidos los impactos fiscales de la descarbonización, siguiendo las mejores prácticas de modelado energético aplicado para el apoyo de políticas. Este trabajo encuentra que el NDP tiene altos beneficios económicos (costos evitados en relación con el BAU) en el largo plazo, equivalentes al 5,5 % del PIB anual en la década 2041-50. En 2031-2040, los beneficios son solo el 0,8 % del PIB anual; en 2022-2030, el NDP enfrenta costos netos del 0,9 % del PIB anual. Estos resultados son promedios de futuros y pueden ser mayores o menores. El gobierno tendrá ingresos fiscales directos más bajos de alrededor del 0,87 % del PIB anual en 2041-2050 y deberá redistribuir los beneficios para compensar esto. Puede utilizar impuestos por vehículo-kilómetro (VKT, por sus siglas en inglés), impuestos a la propiedad o impuestos a la energía para la redistribución, gravando principalmente a los propietarios de transporte privado -que tienen los mayores beneficios-. Sin embargo, para facilitar la descarbonización de las empresas de carga en 2022-2030 y 2031-2040, el gobierno podría subsidiar la adopción de vehículos de cero emisiones (ZEV, por sus siglas en inglés). Altos beneficios, bajas emisiones que cumplen con los objetivos de cero emisiones netas y bajos precios de la electricidad y el transporte público son resultados de política deseables. Los bajos costos de los ZEV y la infraestructura energética, incluidas las energías renovables y el almacenamiento, son condiciones inciertas cruciales para obtener resultados deseables. Los objetivos de política robustos que el gobierno puede adoptar para lograr resultados deseables deben desvincular el crecimiento económico de la actividad de transporte. Los objetivos de política específicos incluyen inversiones que logren aumentar el transporte público, la digitalización, el transporte no motorizado, los viajes compartidos, los centros logísticos y la densificación de las ciudades. Además, alcanzar bajos precios de la electricidad requiere un bajo costo del capital para financiar las inversiones del sistema de eléctrico. ix List of tables 3.1. Measures and interventions per parameter of the NDP scenario. . . . . . . . . . . . . . . . . 29 3.2. Description of energy sector mitigation measures for ranking. . . . . . . . . . . . . . . . . . 30 3.3. Actor classification and transactions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 3.4. TEM inputs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 3.5. XLRM matrix for the wide experiment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 3.6. Ranges of input values to produce the experiment from random value matrix. . . . . . . . . 43 3.7. XLRM matrix for the narrow experiment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 x List of figures 2.1. Connection of topics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 2.2. Reference Energy System (RES) of OSEMOSYS-CR. . . . . . . . . . . . . . . . . . . . . . . 14 2.3. Outputs of RES elements. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 2.4. Inputs of RES elements. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 2.5. Transport modeling in OSEMOSYS-CR. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 2.6. Iterative steps of a robust decision making analysis. . . . . . . . . . . . . . . . . . . . . . . 20 3.1. Overview of best practices and modeling framework for robust planning analysis. . . . . . . 26 3.2. Reference energy system for the energy and transport sectors. . . . . . . . . . . . . . . . . . 31 3.3. Reference energy system for the energy, transport, and industry sectors. . . . . . . . . . . . 31 3.4. Relationships among actors in the transaction estimation module. . . . . . . . . . . . . . . 35 3.5. Hierarchical PRIM application for national financial impacts. . . . . . . . . . . . . . . . . . 48 3.6. Hierarchical PRIM application for emissions, bus, and electricity prices. . . . . . . . . . . . 48 3.7. Hierarchical PRIM application for gross capital expenses. . . . . . . . . . . . . . . . . . . . 49 4.1. Financial expenses per sector of the BAU and NDP scenarios. . . . . . . . . . . . . . . . . . 50 4.2. Economic benefits for the NDP scenario. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 4.3. Emissions of carbon dioxide equivalent (CO2e). . . . . . . . . . . . . . . . . . . . . . . . . . 52 4.4. Economic benefits by mitigation measures. . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 4.5. Cumulative emissions reduction by mitigation measures. . . . . . . . . . . . . . . . . . . . . 54 4.6. Excess CAPEX and fixed costs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 4.7. Ranking of mitigation measures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 4.8. Overview of national metrics under uncertainty. . . . . . . . . . . . . . . . . . . . . . . . . . 56 4.9. Context of electricity prices. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 4.10. Context of bus prices. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 4.11. Financial impacts for the NDP in the 2022-50 period. . . . . . . . . . . . . . . . . . . . . . 59 4.12. Relationship between the transport financial benefits, electricity prices, discount rates, and profit margin. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 4.13. Yearly capacity and generation of BAU and NDP the power sectors. . . . . . . . . . . . . . 61 xi 4.14. The capacity and generation of the power sector of the NDP in 2050 across futures. . . . . 62 4.15. Net revenue of transport actors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 4.16. Financial impact for transport actors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 4.17. Net revenue of public transport operators. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 4.18. Net revenue of public transport operators. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 4.19. National public transport expenses. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 4.20. Net revenue of energy firms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 4.21. Net revenue and financial impact of energy firms under uncertainty. . . . . . . . . . . . . . 67 4.22. Government net revenue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 4.23. Government metrics under uncertainty. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69 4.24. Tax expenses and fiscal costs per period and scenario. . . . . . . . . . . . . . . . . . . . . . 70 4.25. Comparison of tax expenses across scenarios. . . . . . . . . . . . . . . . . . . . . . . . . . . 71 4.26. Fiscal costs per transport sector actor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 4.27. Tax rates versus financial impacts per actor. . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 4.28. Drivers for desirable and risk outcomes for national financial impacts. . . . . . . . . . . . . 76 4.29. Drivers for desirable and risk outcomes for electricity and bus prices. . . . . . . . . . . . . . 78 4.30. Drivers for desirable and risk outcomes for national emissions. . . . . . . . . . . . . . . . . . 79 4.31. Drivers for desirable and risk outcomes for national CAPEX. . . . . . . . . . . . . . . . . . 79 4.32. Drivers for desirable and risk outcomes across national metrics. . . . . . . . . . . . . . . . . 80 4.33. Drivers for desirable and risky financial impacts for private transport owners. . . . . . . . . 81 4.34. Drivers for desirable and risky financial impacts for freight firms. . . . . . . . . . . . . . . . 81 4.35. Drivers for desirable and risky financial impacts for public transport operators. . . . . . . . 82 4.36. Drivers for desirable and risky financial impacts for electricity firms. . . . . . . . . . . . . . 83 4.37. Drivers for desirable and risky financial impacts for hydrocarbon firms. . . . . . . . . . . . 84 4.38. Drivers for desirable and risky financial impacts for the government. . . . . . . . . . . . . . 84 A.1. Cost trajectories of renewable power generation. . . . . . . . . . . . . . . . . . . . . . . . . 114 A.2. Cost trajectories of electricity distribution technologies. . . . . . . . . . . . . . . . . . . . . 115 B.1. Hierarchical PRIM application for private transport owners. . . . . . . . . . . . . . . . . . . 116 B.2. Hierarchical PRIM application for public transport operators. . . . . . . . . . . . . . . . . . 116 B.3. Hierarchical PRIM application for freight firms. . . . . . . . . . . . . . . . . . . . . . . . . . 117 xii B.4. Hierarchical PRIM application for energy firms. . . . . . . . . . . . . . . . . . . . . . . . . . 117 B.5. Hierarchical PRIM application for the government. . . . . . . . . . . . . . . . . . . . . . . . 118 C.1. Transport energy consumption for the NDP scenario. . . . . . . . . . . . . . . . . . . . . . . 119 C.2. Distribution of energy consumption in the NDP scenario. . . . . . . . . . . . . . . . . . . . 120 C.3. Distance driven by transport mode. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120 C.4. Light duty transport metrics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 C.5. Light freight transport metrics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 C.6. Heavy freight transport metrics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 C.7. Bus and minibus metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 D.1. Validation of the hierarchical PRIM analysis for national metrics. . . . . . . . . . . . . . . . 125 E.1. Wide experiment values for the 2022-2030 period. . . . . . . . . . . . . . . . . . . . . . . . . 126 E.2. Wide experiment values for the 2031-2050 period. . . . . . . . . . . . . . . . . . . . . . . . . 127 xiii Autorización para digitalización y comunicación pública de Trabajos Finales de Graduación del Sistema de Estudios de Posgrado en el Repositorio Institucional de la Universidad de Costa Rica. Yo, _______________________________________, con cédula de identidad _____________________, en mi condición de autor del TFG titulado ___________________________________________________ _____________________________________________________________________________________________ _____________________________________________________________________________________________ Autorizo a la Universidad de Costa Rica para digitalizar y hacer divulgación pública de forma gratuita de dicho TFG a través del Repositorio Institucional u otro medio electrónico, para ser puesto a disposición del público según lo que establezca el Sistema de Estudios de Posgrado. SI NO * *En caso de la negativa favor indicar el tiempo de restricción: ________________ año (s). Este Trabajo Final de Graduación será publicado en formato PDF, o en el formato que en el momento se establezca, de tal forma que el acceso al mismo sea libre, con el fin de permitir la consulta e impresión, pero no su modificación. Manifiesto que mi Trabajo Final de Graduación fue debidamente subido al sistema digital Kerwá y su contenido corresponde al documento original que sirvió para la obtención de mi título, y que su información no infringe ni violenta ningún derecho a terceros. El TFG además cuenta con el visto bueno de mi Director (a) de Tesis o Tutor (a) y cumplió con lo establecido en la revisión del Formato por parte del Sistema de Estudios de Posgrado. FIRMA ESTUDIANTE Nota: El presente documento constituye una declaración jurada, cuyos alcances aseguran a la Universidad, que su contenido sea tomado como cierto. Su importancia radica en que permite abreviar procedimientos administrativos, y al mismo tiempo genera una responsabilidad legal para que quien declare contrario a la verdad de lo que manifiesta, puede como consecuencia, enfrentar un proceso penal por delito de perjurio, tipificado en el artículo 318 de nuestro Código Penal. Lo anterior implica que el estudiante se vea forzado a realizar su mayor esfuerzo para que no sólo incluya información veraz en la Licencia de Publicación, sino que también realice diligentemente la gestión de subir el documento correcto en la plataforma digital Kerwá. Luis Fernando Victor Gallardo 1-1588-0266 DECARBONIZATION UNDER TECHNOLOGICAL UNCERTAINTY: FROM TRANSPORT ELECTRIFICATION TO POWER SYSTEM INVESTMENTS ROBUST ENERGY SYSTEM PLANNING FOR X 1 Chapter 1 Introduction The Paris Agreement, a legally binding international treaty on climate change, sets a global temperature increase target of 1.5°C above pre-industrial levels. The Intergovernmental Panel on Climate Change (IPCC), a United Nations body responsible for advancing knowledge on anthropogenic climate change (see IPCC, n.d.), recommends in IPCC, 2018a and IPCC, 2021 that world emissions reach net zero by 2050 to meet the goal. Net-zero emissions are equivalent to carbon neutrality: carbon dioxide equivalent (CO2e) emissions1 put into the atmosphere equal the CO2 removed, e.g., by forests. Many countries have already pledged to reach important emission reduction or containment milestones by 2030, including Costa Rica in its Nationally Determined Contribution (NDC) Government of Costa Rica 2018-2022, 2020. To achieve the final state of net-zero emissions by 2050, the current policy instrument in Costa Rica is the National Decarbonization Plan (NDP), laying out a pathway of sectoral instruments and objectives that transform the economy. While the adverse effects of climate change are already palpable IPCC, 2021, particularly in vulnerable communities exposed to extreme weather events like flooding and drought, only ten countries were responsible for 68.7% of global greenhouse gas (GHG) emissions in 2018, according to World Resources Institute, 2020. World Resources Institute, 2020 shows estimations for global emissions at 47,515.3 MTon in 2018, while Costa Rica’s net emissions were 11.5 MTon in 2017 according to Instituto Meteorológico Nacional, 2019, the most recent GHG emissions inventory. The natural question arises: if Costa Rica’s emissions are about 0.024% of global emissions, why does the country commit to ambitious emission reduction targets? Out of the 11.5 Mton of CO2e, almost 8 Mton are emitted in the energy system (or energy sector). According to IPCC, 2014, an energy system comprises ”all components related to the production, conversion, delivery, and use of energy.” In Costa Rica, the 2017 GHG inventory shows that 6 Mton of CO2e come from transport activities, followed by the manufacturing industry with 1.3 Mton. Moreover, according to SEPSE, 2021, the final secondary energy consumption was 56.7% gasoline and diesel in 1CO2e includes the global warming potential of carbon dioxide, methane, and nitrous oxide gases. 2 2018; transport consumed 93.2% of gasoline and diesel. Despite having a highly renewable electricity generation matrix, the final secondary energy consumption in the country was only 23.4% electrical energy. Therefore, it is clear that Costa Rica’s emissions and energy demands orbit around gasoline and diesel consumption, both fossil fuel derivatives imported by the state-owned company RECOPE (from the Spanish, Refinadora Costarricense de Petroleo). In 2018, Costa Rica imported US$1.28 billion in diesel and gasoline, according to RECOPE, n.d.- a. This magnitude comprises the Cost, Insurance, and Freight (CIF) value and is equivalent to 2% of its nominal GDP in the same year (see World Bank, 2019). Its total imports in that same year were about US$1.6 billion, considering jet fuel, liquified petroleum gas (LPG), fuel oil, asphalt, and other products. In 2014, the total oil bill -another term for fossil fuel derivative imports- surpassed US$2 billion. This statistic precedes the shale oil revolution, i.e., an increased oil and gas production in the United States as a result of fracking and horizontal drilling for extraction, as Center, n.d. explains. The higher production resulted in lower international crude oil prices after 2015, reducing Costa Rica’s oil bill as shown in RECOPE, n.d.-a. World Bank, 2021 mentions the trends of higher energy prices worldwide since the last quarter of 2021, which are related to supply chain constraints and will likely increase Costa Rica’s oil bill. Other events can keep oil prices upward or downward, increasing volatility and impairing economic growth, as explained by van Eyden et al., 2019. The NDP, by Government of Costa Rica 2018-2022, 2019, is emphatic in defining decarbonization as a process of transforming the economy such that national emissions per unit of economic output decrease. This work adopts that decarbonization definition and justifies its importance for Costa Rica not because of climate change mitigation (Costa Rica’s contribution to global emissions is 0.024%) but for energy security and affordability. Decarbonization will drive governments, companies, and individuals worldwide to alter their col- lective energy management, transformation, and consumption. Low-emission electricity generation and energy use technologies can make the country avoid energy costs, decouple GHG emissions from economic growth, and reach climate mitigation commitments as a byproduct. Moreover, if developed economies reverse the historical tendency to outsource environmental harm to the Global South, as explained in Jorgenson et al., 2022, countries with decarbonized energy systems can become more attractive for foreign investment. Such a reversal can occur through increased international private sector responsibility or regulations in the European Union and the United States. Finally, a decar- bonized energy system could be less vulnerable to oil price volatility, providing energy cost stability 3 to private firms and households. Considering the technological characteristics of vehicles2, the trends of fuel costs projected by the International Energy Agency in International Energy Agency, 2019, and the cost reduction trajectory of batteries and renewables, Godínez-Zamora et al., 2020 found that decarbonizing the energy and transport sectors in Costa Rica by 2050 produce US$20 billion, discounted at 5%. Then, Groves et al., 2020 evaluated the NDP under deep uncertainties about the future, i.e., variables that cannot be known or agreed upon by stakeholders. They found vulnerabilities of decarbonizing in terms of emission targets and high costs, which would make the NDP unsuccessful: i) low adoption of electric private vehicles and buses, ii) high economic growth with cheap and efficient conventional vehicles, and iii) expensive vehicles and low adoption of electric trucks. This work builds on these previous studies by developing tools that support robust energy system planning for Costa Rica. Planning results in policy, which produces outcomes measured in metrics. Marchau et al., 2019 define a robust policy as one that produces the most favorable outcomes across possible future scenarios. Moreover, there are policy objectives and policy instruments, as distinguished by van den Bergh et al., 2021. Victor-Gallardo, Roccard, et al., 2022 explain that plans such as the NDP propose mitigation measures like transport electrification, which have associated specific objectives (e.g., 95% of the private fleet in 2050) and instruments to achieve them (e.g., through tax subsidies for electric vehicles). The focus of this work is the understanding of robust mitigation measures, objectives, and instruments quantifiable through energy system modeling tools, which have internally consistent engineering and economic features. The decarbonization of the energy system requires a pathway, often called the energy transition (see IRENA, n.d.), comprising technology adoption and policy objectives and instruments that should occur by mid-century. Pye and Bataille, 2016 and Bataille et al., 2016 explain that energy system modeling supports energy transition by estimating future costs, emissions, and energy requirements, gathering and providing insights from and to stakeholders. Lopion et al., 2018 mention that energy models provide insights for policy questions with varying methodologies, spatiotemporal resolutions, and bottom-up or top-down analytical approaches. In Costa Rica, Godínez-Zamora et al., 2020 devel- oped OSeMOSYS-CR, which is an example of an energy system optimization model (ESOM). This chapter presents the state of the art in the following fields: long-term strategies for the energy transition; 2For example, battery electric vehicles waste less energy than internal combustion vehicle counterparts. 4 the applications, characteristics, and limitations of ESOMS; robustness for the energy transition. It also presents a justification for the work, the problem statement, a hypothesis, and the research objectives. Chapter 2 develops this work’s theoretical framework, covering the underpinning concepts of this research: ESOMs, Robust Decision Making (RDM), and transactions in energy systems (energy- related prices and taxes). Chapter 3 presents the methodological developments of the work, Chapter 4 presents and discusses the results, and Chapter 5 presents the conclusions, recommendations, and future work. Many of the developments of this work have been written in five original research articles submitted for peer review. This document compiles some elements from those articles to answer the research objectives, and they are cited accordingly. 1.1. State of the Art IPCC, 2018b reports the need for carbon dioxide (CO2) emissions to fall 45% 2030 relative to 2010 and reach practically zero by 2050 to avoid harsh climate impacts. Therefore, the time for the 2030 emission reduction targets worldwide is relatively short. Costa Rica did not pledge GHG emission reductions by 2030 in its NDC Government of Costa Rica 2018-2022, 2020; instead, the country defined an emissions budget more or less equivalent to keeping emissions constant from 2018 onwards. The NDC and the NDP pledge to reach net-zero GHG emissions by 2050. This decade will give the country time to prepare for more aggressive transformations after 2030 and import low and zero-carbon technologies at competitive costs, which wealthy countries will most likely develop and pay for at early adoption. This state of the art will describe the existing research on how countries advance the energy transition in the 2022-50 horizon. 1.1.1. Strategies for the Long-term Energy Transition Energy systems supply, transform and carry energy in different forms to satisfy the demands of society. Priesmann et al., 2019 state that ESOMs are the most popular tool to analyze energy systems. Sass et al., 2020 highlight the relevance of ESOMs to study decarbonization and decentralization in the design, operation, and control of sector-coupled energy systems. DeCarolis et al., 2017 explain that the solution of an ESOM is the selection of activity and capacity of technologies out of a pool of 5 alternatives, based on differences in the relative cost of competing options, performance, fixed demands, and constraints (i.e., inputs). ESOMs deploy the technologies over a horizon with perfect foresight. Energy system modelers can use diverse implementations of ESOMs to perform a cost-benefit analysis, as Xiang et al., 2020 have done to analyze demand response policies in an integrated system of electricity and gas. Schlachtberger et al., 2018 explain that, typically, ESOMs estimate a least- cost combination of technological options under physical, environmental, and societal boundaries. For example, OSeMOSYS-CR developed by Godínez-Zamora et al., 2020 minimizes cost in the long-term for a countrywide energy system, subject to emission and technological restrictions, and has a yearly temporal resolution. In contrast, Prina et al., 2020 present a multi-objective optimization method with an hourly temporal resolution and multi-node (or regional) approach to consider the spatial resolution. ESOMs support decision-making but do not guarantee a perfect policy design since they can ignore social, economic, or engineering realities. To gain more insights, ESOMs and Computable Gen- eral Equilibrium Models (CGEMs) have been integrated to explain demands through macroeconomic drivers, prices, elasticities, fiscal policies, and levels of income across the population, as mentioned by DeCarolis et al., 2017 and Helgesen, 2013. ITF, 2019 and Bhattacharyya and Timilsina, 2009 explain that the results of these models are not forecasts, but plausible scenarios about the future, i.e., stories describing how the future can unfold through illustrative pathways. Generally, ESOMs have static demands and do not reflect the participation of multiple agents in the energy system. Some efforts have attempted to improve demand modeling. For example, Blanco et al., 2019 linked a behavioral transport model with an ESOM to study hydrogen options for the European Union (EU) by having both models exchange parameters describing the same scenario. Moreover, Zhang et al., 2020 used an agent-based model to determine the demand of an energy system by adding multiple stochastic individual demands of households. Technical cost minimization should not be the only relevant objective of ESOMs. Algunaibet et al., 2019 show there are indirect costs or externalities of deploying technologies in the energy system, often not accounted for in ESOMs; they found that keeping the current mix of energy resources can cost the world up to 1.1 ± 0.2 trillion US$, including direct and indirect costs. Also, Zvingilaite, 2011 presents that planning for the future considering externalities is cheaper than paying for damages. A decarbonized future energy system will have a larger power system based on renewable gener- ation sources, substituting fossil fuel consumption. The trade-off between cost minimization and land requirements has been highlighted by kuang Chen et al., 2022, where costs 10% higher could reduce 6 land requirements for renewable energy by 58% in the Northern European energy system. Accord- ing to Dominković et al., 2022, research about individual technology has been declining, while recent literature has focused on energy system flexibility and integrated energy systems. A specific line of research for integrated energy system analysis is the role of electrification in sup- porting the electrical grid, which can reduce implementation costs through vehicle-to-grid mechanisms (V2G) (see Aghajan-Eshkevari et al., 2022). Still, storage options that diversify the power system tech- nology portfolio are of interest to enable highly renewable systems, like pumped hydropower, biogas, and heat pumps, as exemplified by Nadolny et al., 2022 and Mittelviefhaus et al., 2022. 1.1.2. Applications, Capabilities, and Limitations of ESOMs ESOM approaches vary widely, from deterministic linear programming to stochastic optimiza- tion that considers uncertainties subject to partial equilibrium constraints. Besides the optimization, models vary in temporal and spatial desegregation and scope. Wei et al., 2020 showcase an example of a multi-planning problem (varying temporal scope) for the long and short terms, which in turn have corresponding uncertainties: e.g., technology costs and renewable energy availability. In terms of scope, there are two contrasting examples. Yang et al., 2019 optimize for the operation of a metro system, a specific energy system component. More broadly, Shen et al., 2020 optimize for investment and operation of industrial energy systems, and Ju et al., 2020 optimize for demand response schemes in power markets. Schlachtberger et al., 2018 and Usher and Strachan, 2012 point out that solutions of ESOMs are sensitive to input parameters and have inherent uncertainty. For deterministic energy-economic models, such as OSeMOSYS-CR, Fais et al., 2016 mention that uncertainty can be addressed by analyzing the variation of outputs as a function of changing inputs. Pye et al., 2018 give notice of structural or qualitative uncertainty, which are related to biases in modeling choices. Ruhnau et al., 2022 have used different models to study the same problem using equal data inputs and have shown that models produce different results, evidencing model uncertainty. Ju et al., 2020 propose flexible constraints in the optimization to reflect how decision-makers face risk. Tan et al., 2020 suggest an iterative two-stage optimization for an electricity and heating system considering different uncertain conditions that modify constraints, ensuring continuous operation de- spite worst-case wind power availability. Similarly, but for individual buildings, Wang et al., 2020 use bi-objective trade-off optimization, along with a posteriori analysis based on Monte Carlo simulations 7 to verify performance between environmental and cost outcomes. The above examples use stochastic optimization and complex models for specific systems. Nonethe- less, although accuracy and detail are desirable in ESOMs, Priesmann et al., 2019 warn that the com- plexity must be balanced with acceptable use of computational resources and DeCarolis et al., 2017 recommend increasing the complexity if policy questions make it necessary. Nolting and Praktiknjo, 2022 notes that the complexity of models is increasing, which makes the management of uncertainty harder. In contrast to the stochastic optimization models, tools such as OSeMOSYS-CR trade com- plexity for flexibility and modularity. Howells et al., 2011 mention that tools such as OSeMOSYS-CR have useful prototyping and testing capabilities and Ringkjøb et al., 2018 state that the bottom-up setup of such tools enables a detailed technology characterization, suitable for a national energy sys- tem. Niet et al., 2021 mention the importance of communities of practice around open source models such as OSeMOSYS-CR, which contributes to increased use and model improvements. Below are examples of countrywide ESOM applications. Fais et al., 2016 provide insights for the long-term low-carbon transition of the UK, considering uncertainty, using a national bottom-up ESOM and sensitivity analysis (SA). In Latin America, mostly regional studies have been developed to analyze climate policy: Kober et al., 2014 showed existing commitments would attain CO2e reductions of 40% relative to a baseline scenario in 2050. van der Zwaan et al., 2014 found carbon taxes and biomass resources will play a relevant role in decarbonizing the region. Marcucci et al., 2019 developed an assessment of decarbonization pathways for different regions of the world by treating uncertainties as random variables, e.g., economic growth, resources, and technology costs. According to DeCarolis et al., 2017, the SA approach identifies the input parameters that have the largest influence on results and can strengthen policy insights. Yue et al., 2018 refer to this approach as Monte Carlo Analysis (MCA): to systematically perturb the inputs and then evaluate the outputs with statistical techniques. Wagener and Pianosi, 2019 highlight the relevance of SA because even base year data affects decision-related inputs, particularly in developing countries where data is sparsely available, if at all, as Yeh et al., 2017 reflect. 1.1.3. Robustness for the Energy Transition ESOMs seek insights on what investments or operation schemes comply with an objective (e.g., minimal cost). How do modeling insights influence energy policy and vice-versa? On the one hand, according to Süsser et al., 2021, models help investigate policy options, define targets, and estimate 8 impacts. On the other hand, policymaking also influences data sources, assumptions, the study scope, and how results are used. Fodstad et al., 2022 mentions that there is a lack of studies that models uncertainties of emerging technologies and consumer behavior. Therefore, there is an opportunity to connect ESOM capabilities with policy support processes aiming at finding robust policies, which are the ones that produce the most favorable outcome across multiple scenarios, according to Marchau et al., 2019. Marchau et al., 2019 also say that robust policies have the least regret, i.e., deviate least from the best possible policy provided perfect information was available. To assess energy investment risks, Colla et al., 2020 recommend that decision-makers consider fac- tors like resource availability, installation site, socioeconomic implications, and environmental impact of energy projects. The transition can present vulnerabilities, like the case of Uganda explored by Srid- haran et al., 2019: hydropower dependency and future requirements compete with water allocation for human consumption. Ji et al., 2020 show another example of modeling for the robust planning of elec- tricity systems, considering energy-water interactions. Zhu et al., 2022 show that policy instruments with low GHG abatement costs are more politically feasible, thus, increasing robustness. The cost-benefit analysis of the energy transition will be subject to many debates in the coming decades. With that prospect, N. Kalra et al., 2014 indicate that stakeholders should engage together in a decision process rather than agreeing on assumptions about uncertain matters of the future. R. J. Lempert et al., 2011 introduce Robust Decision Making (RDM) as a methodological tool to help stakeholders improve their decisions under conditions of deep uncertainty. Deep uncertainty is an expression of self-recognized high uncertainty about the future outcome of a system, or in the context of a group of decision-makers, collectively not agreeing on a future outcome of a system. RAND, 2013 explains that RDM relies on computer models and data to explore outcomes of interest with many simulations instead of looking for predictions of the future. Hence, decision-makers change the question of ”what will happen in the future?” to ”what steps can be made today to shape the future?”, as explained by RAND, 2013. RAND, 2013 enlists some RDM applications: water management, management of energy re- sources, flood risk management, and national defense. In the water management field, with RDM methods, N. R. Kalra et al., 2015 found that the Master Plan for the water of Lima (Perú) was overde- signed and suggested specific policies (i.e., demand-side management, pricing, and soft-infrastructure) to save up to 25% of investment costs. Notably, RDM has been used in the United States by Groves and Lempert, 2007 for the 2005 California Water Plan, by Groves et al., 2013 for the Colorado River 9 Basin, and by Finucane et al., 2018 for the Patuxent River Basin. Matrosov et al., 2015 sought robust strategies for London’s water supply, using multi-objective optimization models and visualization tools to study key trade-offs. Callihan, 2013 used RDM to study the implications of climate variability on water management. Singh et al., 2015 studied ecosystem man- agement with varying stakeholder preferences using RDM. Cervigni et al., 2015 address infrastructure decisions for climate adaptation, considering the trade-off of water availability for hydropower and irrigation in Africa. This issue is further explored by Taliotis et al., 2019 for Eastern Africa and the implications of climate on the resilience of the power sector, and how picking a dry climate future strategy contrasts with a wet future one. In the energy sector, Guthrie et al., 2009 and Popper et al., 2009 used RDM to support planning in Israel, providing policy recommendations. Mahnovski, 2007 evaluated decisions of energy companies investing in hydrogen fuel cell technologies. R. Lempert and Trujillo, 2018 affirm that RDM can suitably adress decarbonization planning. Eker and Kwakkel, 2018 explain that RDM evaluates a discrete and pre-specified set of alternative policies, whereas Many-Objective Robust Decision Making (MORDM) generates a large set of alternatives with computational search. Sahlberg et al., 2021 presented the first application of the scenario discovery approach, which sustains RDM (see Section 2.2), in geospatial electrification modeling. Li et al., 2022, Liu et al., 2022, and Ding et al., 2022 took a different approach to find robustness; they embedded robustness within specific energy-related optimization problems to deal with uncertainty. Fiscal policy is an instrument to implement decarbonization. Taylor et al., 2017 propose taxes to favor the reallocation of assets from fossil-based to renewable portfolios. Freire-González and Puig- Ventosa, 2019 suggest options to discourage polluting economic activities, making a case for taxing fossil-based electricity production instead of all options. Moreover, Lin and Jia, 2019 explored the impacts of taxes on energy demand, and OECD, 2019b comprehensively studied the effects of different carbon pricing policies. Finally, even though many countries rely on taxing fuels for government revenue, the literature on the post-decarbonization fiscal policy is not abundant. OECD, 2019a explores how taxes can adapt to declining fossil fuel use in Slovenia and Cesar et al., 2022 define a framework to analyze the fiscal impact of electromobility, i.e., the loss of tax revenue associated with fossil-based transport. Hence, energy system analysis efforts should aim at understanding transactions within it and the costs and benefits for the interacting players, all while seeking robustness in the policy recommendations, for which RDM is suitable. 10 1.2. Justification Groves et al., 2020 and Quirós-Tortós et al., 2021 suggest that a low carbon future is beneficial for countries comparable to Costa Rica. Decarbonization will also produce jobs (see Saget et al., 2020) and advance achieving sustainable development goals (see Haines et al., 2017). J. H. Williams et al., 2021 argue that the sustained implementation of the transition remains a challenge. Each country’s most suitable decarbonization options will depend on their uncertain specific conditions. Therefore, having tools to assess energy system planning that can adjust to national and regional conditions and cater to uncertainties will support the implementation of the energy transition in the oncoming years. Government agencies that formulate energy policy and related policies can benefit from this re- search by continually performing scenario analysis and comparing the simulation outputs with mea- sured indicators. These agencies will better understand the implications of their policy options regard- ing energy and investment requirements, CO2 emissions, overall system cost, and economic benefits per energy system actor. Through this understanding, policies could more easily comply with the public interest in minimizing regret of over or under-investing in energy sectors, having a coherent fiscal policy, and having affordable transport and energy services and products. 1.3. Problem Statement The uncertainty of low or zero-carbon technological availability and cost-effectiveness can hinder effective planning of energy policies that seek energy transformations to accomplish socially-wide bene- fits, e.g., energy security and affordability. In the context of the worldwide energy transition, countries with firm commitments to reach a low-carbon energy system can face barriers caused by: intermittent political support at national and international levels; high investment requirements and cost and revenue changes for different economic actors; public sentiment about the threat of climate change or the available technologies for mitigation. The impossibility of predicting the events of the oncoming decades urges energy policies to have a long-term perspective and adapt to adverse conditions. These policies should guide technology and infrastructure choice, level of investment, and pricing that maximize benefits for actors and minimize regret understood as a deviation from the best possible outcomes. The design of any tool that supports policies with those characteristics is itself a problem that can be stated as follows: 11 How is the energy system modeling tool that supports the planning of technology and infras- tructure choice, investment, and pricing that aims at maximizing benefits and minimizing regret understood as a deviation from the best possible outcomes? 1.4. Hypothesis Computational experiments applied to models that represent the energy system can stress-test policy objectives and instruments from a quantitative perspective. Policy objectives and instruments can be more costly than beneficial under certain conditions, i.e., risk conditions. Consequently, the energy system planning tool must shed light on specific technology adoption and investment targets and price changes through taxes, subsidies, or discount rates that avoid such risks. Hence: Policies conceived with the support of energy system modeling tools can maximize their benefits by estimating robust technology adoption targets, investments, and market price modifications via computational experiments and statistical analysis of possible futures. 1.5. Objectives 1.5.1. Main Objective To develop computational tools to support energy system planning by characterizing robust policy options that maximize benefits and minimize regrets -deviations from the best possible outcomes-. 1.5.2. Specific Objectives 1. To develop a computer experiment software tool to analyze uncertainties and the sensitivity of multiple policy options in the transport and energy sectors for long-term planning horizons. 2. To implement price estimations within an integrated software tool based on energy system mod- els, including taxes, transport service and energy prices, and transactions amongst public trans- port operators, private and public transport users, energy companies, and the government. 3. To develop a methodology that estimates the investments, technology adoption rates, tax rates, service prices, and asset financing rates that cause robust economic outcomes despite uncertainty. 12 Chapter 2 Energy System Modeling and Robust Decision Making This chapter assembles key concepts for the development of the work. Three major topics are joined in this research: ESOMs, RDM, and transfer estimation (taxes, electricity prices, and bus fares). Figure 2.1 shows how the topics are linked. The ESOM is the main energy system model because it enables the techno-economic representation of interacting technologies. Later in this work, a Transaction Estimation Module (TEM) is developed, referring to the modeling of the energy system from the perspective of interacting actors. The ESOM and TEM produce baseline scenarios established on reasonable assumptions, although subject to uncertainty about the future. Since the ESOM and TEM scenarios reflect specific policies and exogenous conditions about the energy system (i.e., demand and technology cost), the RDM-inspired analysis and tools are relevant to evaluate the policies and find which are robust. First, the tools generate an experiment, i.e., a database with possible futures looking toward 2050. Then, the input and output datasets are analyzed to determine what are the robust policies: investments and adoption rates from the perspective of the ESOM and taxes and prices from the perspective of the TEM. This chapter presents Sections 2.1 and 2.2 to conceptually develop ESOMs and RDM, respectively. Section 2.3 lays the groundwork for the development of the TEM. -up a ESOMs: Bottom technological modelling RDM: input and output nalysis TEM: transfer estimation within the energy system Figure 2.1: Connection of topics. 13 2.1. Energy System Optimization Models Considering the complexity cautions by DeCarolis et al., 2017 and taking advantage of OSeMOSYS- CR (i.e., the model started by Godínez-Zamora et al., 2020), the ESOM scheme presented here is based on the Open Source Energy Modeling System (OSeMOSYS), i.e., a set of equations that describe an energy system following a parameterization. Howells et al., 2011 explain that OSeMOSYS models support long-run energy planning and are designed to generate insights based on flexible prototyping and testing capabilities. The open-source feature of OSeMOSYS allows advancement in understanding energy systems through the collaboration of researchers, provided adequate documentation of datasets, code sets, and analyses, according to Pfenninger et al., 2017 and Pfenninger et al., 2018. Groissböck, 2019 warn that for short and mid-term analyses, models that consider operational aspects of energy systems in depth may be more suitable than OSeMOSYS. Models like OSeMOSYS-CR generally have perfect foresight, although there are alternative for- mulations that split solutions over time, according to DeCarolis et al., 2017. The models endogenously calculate the activity and capacity of technologies or processes, the cost of deploying and operating them, building and maintaining required infrastructure, importing energy, or extracting resources while satisfying demands linked to economic activities. Figure 2.2 shows how the energy system is structured in a Reference Energy System (RES). The RES represents the flow of energy from primary or secondary sources to the final energy uses. It includes the processes that transform the energy in different stages. For instance, Figure 2.2 shows how renewable resources (e.g., water, geothermal, wind, solar, or waste) can be used as input to generate electrical power. The electrical power is then distributed to the users via the power distribution infrastructure, which comprises transmission and distribution systems. Electricity is then delivered to households and buildings that use energy in different economic activities. Distributed generation (DG) schemes can also be represented. Electric transport is another demand that electrical power can supply. Traditionally, vehicle technologies are powered by fossil fuel derivatives (also reflected in the RES), which satisfy passenger and freight demands. The characteristics of these vehicles vary according to size, sophistication, and the fuel they use. Such characteristics are energy efficiency, operational life, number of technology units needed to satisfy a yearly demand or commodity output, unit capital costs, unit fixed operation and maintenance costs, and variable costs. 14 Production and Blend Renewable Resources Imports Power Plants Energy Distribution Transport Technologies Agricultural Industrial Residential Mobility Commercial Service DemandsFinal Energy Demand Public services In fr a st ru ct u re Freight Primary and Secondary Sources Final Energy Demand Figure 2.2: Reference Energy System (RES) of OSEMOSYS-CR. The RES has interconnected blocks representing energy transformation processes to supply end- use devices with energy. The system’s energy supply must be in equilibrium with the demand as a fundamental restriction to the optimization of the ESOM with the objective function in Equation 2.1; Equation 2.2 defines the total cost. The model estimates the output variables shown in Figure 2.3. Any block has two general characteristics referenced throughout ESOM analyses: Capacity: represents the quantity of technology in the unit defined by the modelers. For example, the capacity of power plants is defined in MW or GW. Activity: a technology’s level of activity is the production of a commodity, i.e., a demand or another form of energy converted by a technology. Most activities are measured in Petajoules (PJ). In the case of transport demand, units are passenger or ton-kilometers (pkm or tkm). min ∑ (y,t,r) (Total Discounted Cost(y,t,r)), y : year, t : technology, r : region (2.1) ∀(y,r,t)Total Discounted Cost(y,t,r) = Discounted Operating Cost(y,t,r) + Discounted Capital Investment(y,t,r) + Discounted Emissions Penalty(y,t,r) − Discounted Salvage Value(y,t,r) (2.2) 15 Figure 2.3: Outputs of RES elements. Parameterizing the model blocks completes the modeling. Figure 2.4 shows the relationships between technologies and commodities in OSeMOSYS, as well as their respective inputs, listed below: Energy efficiency: transforming or transporting energy from the primary (e.g., renewable resource) or secondary (e.g., barrels of gasoline) sources to the end-use devices causes losses. Each block in the RES has an efficiency that considers the losses due to conversion or transport. Operational lifetime: the number of years the technology are used over the planning horizon. Capital cost: the capital cost of technology investments per unit of capacity. Fixed O&M cost: fixed operation and maintenance (O&M) technology costs per unit of ca- pacity. Variable O&M cost: variable O&M technology costs per unit of activity. Availability in a year: the availability factor changes the time that technologies are functional in a year and the capacity factor defines the fraction of the installed capacity used in one year. Emission factor: emission factor of a technology per unit of activity. 16 Emissions penalty: penalty per unit of emission used to include externalities. Shaw et al., 2014 say that using less emitting technologies can bring about health improvements associated with carbon reductions, which are monetized by Coady et al., 2019. Conversion of output to capacity: connects activity and capacity by defining the quantity (capacity) of technology per unit of commodity produced (activity). Residual capacity: the capacity (quantity of a technology) available from before the modelling period. This value allows calibrating the amount of existing technologies. Maximum or minimum quantity of technology or commodity (constraint): upper and lower limits, as well as fixed values, that can force levels of capacity. This parameter is applicable for commodities too. It allows modeling adoption rates of technologies. Figure 2.4: Inputs of RES elements. 17 2.1.1. Modeling Demand in ESOMs The energy demands and adoption rates of technologies are also input variables. These are not technological inputs; they are related to people’s behavior and economic activities. Defining energy demands is crucial for ESOM analyses and often one of the most debated inputs because of uncertainty. For transport, RAND, 2006 explains that uncertainty is linked to the variability of travel on a physical transportation network. Although there are engineering models that describe aggregate patterns in the use of transport (i.e., the peak time demand), RAND, 2006 mentions persistent uncertainties: i) which transport modes are used for trips, ii) differences in time of use, vehicle occupancy rates and purposes, iii) variability in origins and destinations, iv) cancellation or postponement of trips. Fraser et al., 2018 say the effects of climate change on extreme weather conditions put the trans- portation systems around the globe at risk, showcasing a broader link between transport sector planning and climate change (e.g., adaptation of infrastructure in power, water, and communications systems). Lovrić et al., 2017 assess the infrastructure capacity utilization and its link to the energy system. Emissions from transport are quantified with an emission factor proportional to driven distance or per gallon of fuel consumed. The latter approach is easier to implement and used in OSeMOSYS-CR. The IPCC Tier 1 methodology consists of multiplying emission factors per unit of energy times an activity level (i.e., consumption of fuels). The National GHG Inventory uses data from the National Energy Balance. Bhattacharyya and Timilsina, 2009 present transport demand methods based on fuel consumption econometric models, as well as the following mixed approaches: Identity model: fuel demand equals the product of vehicle utilization and stock. Structural model: derives energy demand of transport from the demand of transport services, which in turn are services used as inputs to minimize costs of production. Market-share model: considers inter-fuel substitution possibilities. Bhattacharyya and Timilsina, 2009 explain that end-use approaches like the above consider the diversity of transport modes, types of vehicles, and efficiencies. According to these authors, the number of trips and modal distribution influence passenger traffic. In contrast, the production of goods, average distance, and characteristics of the traffic structure to move goods determine freight traffic. Yeh et al., 2017 concluded mobility demand is a function of population and Gross Domestic Product (GDP), with integrated modeling options using logit functions or least-cost optimization. They also explain that freight projections are generally dependent on GDP forecasts. However, Keshavarzian et al., 2012 18 signal non-linearities in the relationship between GDP and vehicle ownership. Sakamoto et al., 2016 warned that trends are faulty when systems undergo structural change. One alternative proposed by Keshavarzian et al., 2012 is using non-linear Gompertz or logistic curves to estimate vehicle ownership. Bhattacharyya and Timilsina, 2009 identified that simpler models are virtuous for insufficient data assumptions and have results similar to more complicated ones. Yeh et al., 2017 explain that passenger and freight demands, along with occupancy rates, yearly distance traveled, and vehicle survival rates, can estimate the number of vehicles on the road. Mobility demands are in passenger-kilometer units, i.e., the number of persons times their traveled distance in one year. Similarly, freight demand -in ton-kilometers units- represents the movement of goods instead of persons. Figure 2.5 shows an example of how transport demand is managed in OSeMOSYS-CR. First, the modeled parameterizes an exogenous demand and the mode shift of the scenario. Second, the modeler computes the necessary vehicle capacity with distance and occupancy rates assumptions. Third, the efficiencies of vehicles in kilometers per liter (converted to PJ/km) convert mobility to energy. The modeling process is formalized in Victor-Gallardo, Rodríguez-Zúñiga, Rodríguez-Arce, et al., 2022 and presented in Section 3.1. OSeMOSYS-CR only covers road transport, although other models such as ICCT, 2012 explore aircraft and marine vessels in-depth. All Bus Technologies Gasoline Sedan Electric Sedan Occupancy Rate Occupancy Rate Distance Distance Gasoline Electricity Private mobility demand Public mobility demand Diesel Bus Electric Bus Diesel All Sedan Technologies M o d a l sh i Petajoules Number of vehicles kilometers Passengers/Vehicle Passenger-kilometers Vehicle-kilometers Energy Supply Sub- Model Transport Demand Sub-ModelE ciency (PJ/Gvkm) Figure 2.5: Transport modeling in OSEMOSYS-CR. OSeMOSYS-CR characterizes vehicles and energy use devices but lacks similar technologies for other demand sectors like industry and buildings. For these cases, the demand was introduced in 19 PJ directly. Godínez-Zamora et al., 2020 recognize OSeMOSYS-CR uses an autoregressive integrated moving average (ARIMA) model to define demand projections based on historic energy balance data; the transport sector uses additional constants to reach mobility and freight demands. Victor-Gallardo, Rodríguez-Zúñiga, Rodríguez-Arce, et al., 2022 adjusts these considerations to include the industry sector modeling based on data from Escuela de Ingeniería Eléctrica - Universidad de Costa Rica, 2020 and Escuela de Ingeniería Eléctrica - Universidad de Costa Rica, 2019. Similar to transport, other sectors have demands projected with various features. For exam- ple, Edelenbosch et al., 2017 used a linear estimator for electric load data with a daily and hourly resolution with peak demand, temperature, GDP, population, industrial production, and day dura- tion. Bhattacharyya and Timilsina, 2009 enlist demand alternatives: combinations of econometric and engineering-economy models, system dynamics models, scenario approaches, decomposition models, process models, input-output models, and artificial neural networks. Bhattacharyya and Timilsina, 2009 clarify that econometric methods are employed at a national level, whereas accounting methods (i.e., bottom-up technology-related descriptions) are used at sectoral or end-use levels. The International Energy Agency, 2019 affirmed that worldwide energy demand has a 1% yearly growth projection by 2040, which is significantly lower than the historical 2.3% increase seen in 2018. If such energy demand increase worldwide continues, the strain on the global energy system would be considerable. According to J. Williams and Waisman, 2018, for decarbonization pathways such as the Decarbonization Plan, stakeholders are interested in defining physical transformations and subsequent investments that achieve socioeconomic and emissions objectives. The RES representation is a practical way to define where the investments occur and what technological options are available. The ESOM is relevant to inform investment decision making if translated to a cost-benefit analysis (CBA), which according to N. Kalra et al., 2014 consists of: calculating the value of financial and non-financial costs over a time period; translating future costs and benefits into present value using a discount rate; ranking each investment with a metric for later selection when compared to alternatives, and thus, the investments that bring the greatest benefit are chosen. Energy systems are not, in practice, centrally planned. Thus, ESOMs are criticized for often modeling single decision-makers in the energy system. DeCarolis et al., 2017 present technology- specific discount rates as a solution to reflect agent preferences and other non-financial costs, although 20 empirical evidence is often nonexistent for parameterization. Other solutions to ESOM criticisms, according to DeCarolis et al., 2017, are merging consumer utility and cost objectives and further endogenization (e.g., technological learning) to reduce the need for exogenous assumptions. Yue et al., 2018 explains that these modeling choices face trade-offs between data gaps, the ambition of the studies, level of effort, and thirst for more impactful insights. 2.2. Robust Decision Making Groves et al., 2014 define RDM as a methodology that supports decision-making with the potential to identify policy alternatives that are desirable despite uncertainties, i.e., have good performance across many possible combinations of conditions. It uses statistical and software tools iteratively with the participatory engagement of stakeholders, as shown in Figure 2.6. RAND, 2014 explains that RDM involves deliberation by firstly meeting with stakeholders, planners, and decision-makers to define the scope of a policy study. At that stage, stakeholders help identify the policies’ likely risks or relevant outcomes. Following Groves et al., 2014 and Figure 2.6, the evaluation starts with a large database of simulations, resulting in visualizations of strategies. Iterative consultation with stakeholders leads to refinement of the strategies, leading to fewer choices for further evaluation. Figure 2.6: Iterative steps of a robust decision making analysis. Based on Groves et al., 2014. Groves et al., 2014 explains that, when implementing RDM, there is an evaluation of policy options across multiple futures. Vulnerabilities linked to adopting an option are then identified. The following concepts are relevant to distinguish: 21 Cases: run of a simulation model. Future: a specific set of assumptions about the future. Condition: set of futures that are similar along one or more dimensions of uncertainty. Scenario: set of cases that share a decision-relevant attribute. Strategy: amount, location, and timing of investments and programs (i.e., levers). Hence, there is a distinction when referring to ”policy”: i) synonym of strategy, or ii) it can be short for ”policy lever,” i.e., individual component of a strategy. For decision structuring, stakeholders (e.g., decision-makers, experts, agency professionals) reunite and perform the following activities aiming to narrow down the analysis: 1. identify key goals for the policy; 2. define critical uncertain factors that could influence planning conditions or strategy success; 3. preliminary set of options (strategies) to evaluate; 4. define performance metrics to assess strategies across futures; 5. compile data and models to estimate performance. According to RAND, 2014, one tool that facilitates the XLRM matrix facilitates decision struc- turing. It systematizes the following components: Exogenous uncertainties (X): factors outside the control of decision-makers that may affect the ability of actions to achieve the desired goals. Policy levers (L): actions that decision makers may consider. Relationships (R): describe how the policy levers perform, as measured by the metrics, under the various uncertainties. Simulation models are often used. Metrics (M): of performance to evaluate if a choice of policy levers achieves the desired goals. Mahnovski, 2007 defines robustness as the relative insensitivity of a strategy (or policy) to the unknown probabilities of any state, i.e., the difference between the performance of a strategy in a future 22 state of the world and the best performing strategy of the same future. More robustness metrics in diverse contexts are defined by Doumpos et al., 2016 and McPhail et al., 2018. For the simulation of many futures, data and models evaluate future conditions across a wide range of future possibilities. This step generates a large database of quantitative information through compu- tational experiments to explore the implications of varying assumptions and hypotheses. RAND, 2014 explain that the computational experiment explores uncertainties by sampling variations broadly and uniformly, unlike a Monte-Carlo probabilistic sampling approach requiring a probability distribution function per uncertainty. Thus, by defining broad and uniform samples, strategies are stress-tested without judging whether a future is more or less likely than another, according to RAND, 2014. Experiments can be full factorial, i.e., including all combinations of uncertain factors and their assigned possible ranges. RAND, 2014 explains that this approach is computationally expensive, and thus, a Latin Hypercube Sampling (LHS) scheme is convenient to sample across the factor space without requiring all combinations uniformly. Experimental designs must weigh the computing time needed to simulate the management system for a single future and the number of futures developed to reflect uncertainty, strategies evaluated, and iterations of analysis to perform. Groves and Lempert, 2007 explain the produced scenarios are linked to a story about how drivers affect the trend of variables of interest (e.g., GHG emissions) and how users of that scenario face decisions. Bryant and Lempert, 2010 addressed scenario quality and recommended carefully choosing the number of scenarios that explain key driving forces to avoid false expectations among users of the analysis. In an approach of many futures, the analytics to discard noisy scenarios and extract relevant information is a crucial step to comply with clarity, practicality, and credibility of an RDM project. For the vulnerability analysis, the underlying concept is Scenario Discovery, i.e., an application of statistical and data-mining algorithms to databases generated with simulations, according to Bryant and Lempert, 2010. R. J. Lempert et al., 2008 describes the two general algorithms to choose: Patient Rule Induction Method (PRIM): identifies regions in an uncertain model input space that are highly predictive of model outcomes that are of interest (see Kwakkel and Cun- ningham, 2016 for a clear explanation of the PRIM algorithm). Classification and regression trees (CART): machine-learning methods for constructing prediction models from data. The models are obtained by recursively partitioning the data space and fitting a simple prediction model within each partition. As a result, Loh, 2011 explains that partitioning can be represented graphically as a decision tree. 23 Moreover, decision-makers and stakeholders work together to define a few key scenarios that tell a compelling story about its implications in evaluating the policy. Groves et al., 2014 explain that these actors do not need to agree on the results but rather provide information to refine them. RAND, 2014 explains two measures often used in the vulnerability analysis: Coverage: the ratio of the number of futures represented by the vulnerable conditions that do not meet goals to all futures that do not meet goals. Density: the ratio of futures that are represented by the vulnerable conditions and do not meet goals to all futures that are represented by the vulnerable conditions. Uncertainties are then statistically evaluated across simulated cases with the variables of interest from the experiment. For example, a two-dimensional plot of two key performance measures can statistically evaluate cases, illuminating the importance of visualization tools. Then, clustering cases in the two-dimensional plot leads to finding a scenario: a centroid of a cluster of cases that gives similar interpretations and, thus, an associated narrative. In the trade-off analysis, stakeholders engage in the analysis again through visualizations. In the previous step, a collection of static two-dimensional plots is useful to find scenarios. Interactive plots are required to discuss and analyze qualitative results. Groves et al., 2014 argues that the visualiza- tion must highlight key trade-offs and compare performance measures among strategies. RAND, 2014 explains that trade-offs are examined in terms of how alternative strategies perform in reducing vul- nerabilities. According to Groves et al., 2014, the deliberation involves participating experts, gathering additional information, and providing context about the likelihood of scenarios. The discussions can lead to further inquiring about the results. As a result, more iterations across the RDM process can be necessary to reconcile the robustness criteria. For new futures and strategies, the outcome of the RDM methodology is a robust strategy, i.e., policies that have good performance despite changing conditions over time developed with awareness of possible futures (see Walker et al., 2013). Gong et al., 2017 mention that robust scenario analysis allows stakeholders to understand worst-case futures. Different analysis types exist, such as Dynamic Adaptive Policy Pathways by Haasnoot et al., 2013. Kwakkel et al., 2016 found that RDM and Dynamic Adaptive Policy Pathways are complementary by offering an alternative to handling the vulnerabilities that RDM identifies. They also explain that an Adaptive Policy Pathways approach offers guidance 24 to make choices by steering the adaptation of policies over time Kwakkel et al., 2016. Hermans et al., 2017 designed how observation of technical or political metrics trigger strategy adaptation actions. Below are examples of available RDM toolkits. Kwakkel, 2017 developed the Exploratory Model- ing Workbench to support the generation and execution of experiments and to support the visualization and analysis of the obtained results. More recently, Hadjimichael et al., 2020 develop a Python library for MORDM analysis. Although Dreier and Howells, 2019 presented an option to study OSeMOSYS models under uncertainty, this work develops a computational experiment framework as an exten- sion of OSeMOSYS-CR, based on the Python programming language. This implementation enables uncertainty exploration, policy levers comparison, and coherent TEM modeling. 2.3. Transfers in Energy Systems OSeMOSYS-CR includes exogenously established costs for power plants, energy infrastructure (including transmission and distribution systems), vehicles, and civil infrastructure. Moreover, fossil fuels have an import cost. All these costs are subject to uncertainty. Nonetheless, there are different types of costs in the energy system. The ESOM reflects costs from the country’s or region’s perspective, i.e., how much the imported vehicles and fuels would cost to the nation or region if it were a single agent. Zooming in on the energy system, the government taxes the imports and property of vehicles. It also applies a tax to the sales of gasoline, diesel, and other fuels. Electricity supply prices are subject to the power system investments, the cost of its operation, and the demand. The customer perceives the supply price plus a value-added tax (VAT). Hydrogen, one of the technological choices for transport decarbonization in the long term, will have a consumer price. Finally, the users of public transport pay a fare relative to the demand and investments made by the operators, similar to electricity. Two objectives could have opposite effects: 1) tax policy can punish the use of fossil fuels and related technologies, 2) with uptake in electrification of transport and other energy system components, taxing clean electricity can support the government’s finances. Pursuing either objective depends on the timing and the economic context of the country. If decarbonization is successful and benefits society, a government can tax a percentage of the benefits. Nonetheless, the benefits are not the same for every actor participating in the energy system. Thus, the proposed TEM in this work is an important complement to the ESOM that responds to the weaknesses identified by DeCarolis et al., 2017. The TEM was applied by Rodriguez et al., 2021 for the Costa Rican Ministry of Finance to inform the NDP’s fiscal effects. 25 Chapter 3 Methodology Figure 3.1 shows the framework that accomplishes the work’s objectives and respond to the best en- ergy modeling practices suggested by DeCarolis et al., 2017, here named the Multipurpose OSeMOSYS- based Modeling Framework (MOMF). Figure 3.1a shows the steps and principles of energy modeling that DeCarolis et al., 2017 formalized. Figure 3.1b shows the MOMF contributions, listed below: The block A in Figure 3.1b represents a tool to structure and parameterize a model efficiently, following the logic of Section 3.1. This tool was a component of the contribution in Victor- Gallardo, Rodríguez-Zúñiga, Rodríguez-Arce, et al., 2022, who developed OSeMOSYS-CR-v2. Its data sources and assumptions are in the online documentation1; complementary software programs are available in the open-source license repository2. This block responds to Principle IV (orange in Figure 3.1), offering flexibility to model technology variety, future performance and cost assumptions, and restrictions. The block B in Figure 3.1b builds multiple scenarios using the modeling tool described above. Victor-Gallardo, Rodríguez-Zúñiga, Rodríguez-Arce, et al., 2022 developed scenarios for each possible NDP mitigation measure, with different levels of ambition (uncertainties of outcome), e.g., transport electrification magnitudes. Hence, this process responds to Principle VI (in purple) by evaluating different mitigation measures and implementation possibilities. The tools have Python-based processing of inputs and outputs that increase the model’s trans- parency and communicate insights using data visualization tools, i.e., Principle VII (green in Figure 3.1). The results from block B offer Tier 1 policy insights, i.e., costs, emissions, and technological capacity and activity per scenario. The block C in Figure 3.1b shows the TEM component (in Section 3.2). It expands the sectoral detail in the modeling and produces insights beyond those offered by Godínez-Zamora et al., 2020 1https://osemosys-cr-v2.readthedocs.io/en/latest/ 2https://github.com/EPERLab/osemosys-cr-v2 https://osemosys-cr-v2.readthedocs.io/en/latest/ https://github.com/EPERLab/osemosys-cr-v2 26 because it considers actor disaggregation. Thus, it also contributes to Principle VI by dealing with uncertainty about actor-related questions, e.g., what is the change in government revenue. 1. Formulate research ques ons 2. Spa otemporal boundaries 3. Consider model features 4. Conduct and re ne analysis 5. Quan fy uncertainty 6. Communicate insights I. Let the problem drive the analysis, not the other way around. II. Make the analysis as simple as possible and ascomplex as necessary. III. Develop quality assurance procedures and apply them to input data. IV. Consider the range of sectoral detail across the model. V. Re-evaluate the modelling approach and objec ves throughout the analysis. VI. Consider uncertain es that are both endogenous and exogenous to the model. VII. Make transparency a goal of model- based analysis. DeCarolis et al. steps DeCarolis et al. principles A n a ly si s p ro ce ss (a) (b) Figure 3.1: Overview of best practices and modeling framework for robust planning analysis. (a) Best practices according to DeCarolis et al., 2017. (b) MOMF developed in this work. 27 The block C results offer Tier 2 policy insights, i.e., a financial evaluation of scenarios on ac- tors. Best practices suggest conducting and refining the analysis by re-evaluating the modeling approach according to the objectives of a specific study. The contributions listed above (boxed in a black dashed line) apply for that analysis and re-evaluation process. While the contributions above cover some aspects of uncertainty, block D in Figure 3.1b sys- tematically addresses uncertainty-specific questions, covering Principle VI (green in Figure 3.1). This block identifies desirable policy effects, technology use, and risks (Tier 3 results, Figure 3.1b). To achieve this, block E is necessary: it processes results and creates visualizations to interpret model input variations. Crucially, the modeling tools are deterministic, i.e., there are no specific probability distributions for any inputs in this work. Hence, the MOMF uses the RDM practice of sampling inputs with uniform distributions and defined intervals. Figure 3.1 shows the evolution of the analysis process and the steps, which follow the alphabetical order logic described in Figure 3.1b and the list of contributions. Modelers should analyze uncertainties after the steps 1 to 4 have been advanced. The MOMF can address uncertainty without block D, although limitedly, which helps control the complexity of the analysis. When many parameters are moved simultaneously in experiments, deriving insights from the results can be challenging if previous specific questions have not been clearly defined. This methodological chapter is structured as follows: Scenarios and Modeling (see Section 3.1): describes the scenarios, equations, model structure, and policy ranking process. It covers block A of the MOMF and is based on Victor-Gallardo, Rodríguez-Zúñiga, Rodríguez-Arce, et al., 2022. Transaction Estimation Module (see Section 3.2): develops the equations and logic to model prices and prices between energy system actors. The TEM responds to Objective 2, increasing the complexity of the analysis. The specific policy questions addressed by the TEM are: • The fiscal impact of decarbonization: the TEM served as the basis to estimate the fiscal impact of road transport decarbonization in Rodriguez et al., 2021. • Estimation of bus and electricity prices: these are metrics of interest for the analysis devel- oped in Section 3.4 and in Victor-Gallardo, Quirós-Tortós, et al., 2022. 28 • Evaluation of discount rates profit margins: the TEM was used by Victor-Gallardo and Quirós-Tortos, 2022 to estimate the effect of discount rates and profit margins in electricity prices, which affect the NPV benefits for transport actors caused by decarbonization. Model Experiments (see Section 3.3): describes how model experiments are developed. This work shows three different experiments to address specific objectives: • Wide experiment: it varies most OSeMOSYS-CR inputs to produce 2000 futures, which fundamentally responds to Objective 1. The results are then used to explore the magnitude and timing of investments, technology adoption rates, and service prices that produce robust performance metrics for future energy plans, responding to Objective 3. • Narrow experiment: it varies some OSeMOSYS-CR inputs to produce 800 futures. It ex- plores the electricity sector and the financial costs of new power investments. It responds to Objective 3 by searching for asset financing rates that are more convenient, enhancing the robustness of the policy recommendations. • Tax adjustment evaluation: it varies the contribution to eliminating the fiscal impact per tax type for one specific future. The experiment allows exploring robustness criteria for tax rates, thus, responding to Objective 3. Robustness Analysis (see Section 3.4): describes the approach to find robust decarbonization pathways defined in Victor-Gallardo, Quirós-Tortós, et al., 2022 as the combinations of levers and uncertainties that produce desirable and avoid risk outcomes. This methodological contribution responds to Objective 3 of this work and belongs to blocks D and E of the MOMF. 3.1. Scenarios and Modeling This section describes the scenarios and model versions developed in different studies. Godínez- Zamora et al., 2020 developed the seminal model. Victor-Gallardo, Quirós-Tortós, et al., 2022 modified it to model the costs of additional energy system elements, e.g., distance variation, freight rail, electric fast-charging, and hydrogen refueling stations; the same version was used in Victor-Gallardo and Quirós-Tortos, 2022. Victor-Gallardo, Rodríguez-Zúñiga, Rodríguez-Arce, et al., 2022 expanded the model to include the industrial sector. Finally, a policy objective ranking approach is presented. 29 3.1.1. Scenarios There are two main types of scenarios, both developed for the 2018-2050 period: i) A business-as-usual (BAU) scenario represents how the energy system could evolve if the current energy carrier use proportions remain constant under higher production -as GDP grows-. The BAU is used as a benchmark to compare other scenarios with decarbonization efforts. ii) Alternative scenarios, one of which is the National Decarbonization Plan (NDP). They reflect mitigation measures: policy objectives in different subsectors within the energy system. Table 3.1: Measures and interventions per parameter of the NDP scenario. Measure Parameter Intervention Mode shift and passenger rail transport Public passenger transport demand Increase its participation in motorized transport by 7.5% in 2035 and 20% in 2050 Non-motorized transport demand It reduces motorized transport by 4% in 2035 and 10% in 2050 relative to BAU Electric passenger rail demand Transports 0.1 Gpkm and enables public passenger transport mode shift Freight rail Electric freight rail demand Transport 10% of heavy freight demand in 2050, increasing linearly every year starting in 2024 Heavy freight ZEV penetration Fleet composition 5% by 2030 and 50% by 2050 with electric and hydrogen technology Light ZEV penetration 5% by 2030 and 50% by 2050 with electric technology. By 2030, 20% of the fleet uses LPG, and the restriction is removed afterward Public ZEV penetration 30% by 2035 and 85% by 2050 with electric technology. 3%by 2035 and 10% by 2050 with hydrogen buses and minibusses. Private ZEV penetration 35% by 2035 and 99% by 2050 with electric technologies Biofuels % of the fuel volume Biodiesel: 1% by 2026 and 5% by 2030. Gasoline (ethanol): 8% by 2022. Boilers in industry thermal / electrical / mechanical kW composition 40% of biomass and 60% electric Heat production in industry 90% by 2050 with electric technology and 10% with biomass Heat production for glass 99% by 2050 with electric technology Lift trucks Entirely with electric technology On-site power generation 60% by 2050 in battery storage. The rest with biomass Power generation renewability % of fossil fuel-based production 0% by 2050 Power generation characteristics Considerations for model restrictions Hydropower is not further developed. Only planned geothermal projects are simulated. It is assumed that wind and solar (with and without storage) supply the growth in demand, including transport electrification. Commercial and residential LPG consumption LPG substitution with electricity The LPG phase-out occurs by 2050, gradually starting in 2024. The demand is substituted with electricity on a 1 to 1 basis. Taken from Victor-Gallardo, Rodríguez-Zúñiga, Rodríguez-Arce, et al., 2022. 30 Table 3.1 shows the mitigation measures included in the NDP scenario. Each measure is modeled through a parameter than can be mapped into the previous equations. The quantitative assumption of the mitigation measure is in the intervention column, reflecting policy objectives per measure. Table 3.2 shows individual mitigation measures for the ranking exercise. Measures 4 and 12 are numbered to distinguish between options with or without hydrogen (H2) vehicle technologies. Table 3.2: Description of energy sector mitigation measures for ranking. Measures Description 1 Biofuels Biodiesel: 1% by 2026 and 5% by 2030. Gasoline: 8% by 2022. We do not consider relative cost differences between biofuels and fossil fuels. 2 Mode shift and passenger rail* 50% of motorized passenger transport in 2050 10% of passenger transport in 2050 3 Freight rail 20% of heavy freight transport in 2050 4.a Public ZEV penetration 30% in 2035 and 85% in 2050 4.b (include H2 for half of the participation) 5 Private ZEV penetration 30% in 2035 and 95% in 2050. 6 Passenger elasticity to GDP reductions Decrease the demand elasticity to GDP by 10% in 2030 7 Freight elasticity to GDP reductions 8 Distances Apply 0.9 in 2050 to passenger and freight distances 9 Renewable electricity Keeps 100% renewable electricity production by 2050 10 Commercial and residential LPG removal Removes LGP by 2050 11 Industry decarbonization Substitutes oil for biomass and electricity in 2050 12.a Freight ZEV penetration 30% in 2035 and 85% in 2050 12.b (include H2 for half of the participation) Taken from Victor-Gallardo, Rodríguez-Zúñiga, Rodríguez-Arce, et al., 2022. * For reference, motorized passenger transport has 24% public and 76% private transport participa- tion throughout the analysis period in the BAU. ** For reference, the demand elasticities to GDP of 1.015 for freight and 0.916 for passenger transport in the base year. 3.1.2. Versions Figure 3.2 of the RES used in Victor-Gallardo, Quirós-Tortós, et al., 2022 and Victor-Gallardo and Quirós-Tortos, 2022. Imported fuels supply transport technologies, final energy demands, and diesel and fuel oil power plants. Ethanol and biodiesel are included as a blend with fossil fuels, only reducing the unit emission factor. Power plants produce electricity and electrolyzers powered by utility-scale photovoltaic (PV) solar produce hydrogen. Electricity and hydrogen are then distributed to transport technologies and energy demands. Finally, transport technologies use the energy to 31 convert it into kilometers, mobilizing people or freight. Figure 3.3 shows an extended model version used in Victor-Gallardo, Rodríguez-Zúñiga, Rodríguez-Arce, et al., 2022. It includes biomass costs and industry technologies, which produce heat, force (e.g., for lifting objects with forklifts), onsite power generation, and other electricity demands. Transport demands - Freight demand - Passenger demand Final energy demands (fossil) - Agricultural, commercial, industrial, residen al, public services Secondary energy sources > Imported fossil fuels (gasoline, diesel, LPG, and fuel oil) - Power plants (diesel and fuel oil) - Ethanol and biodiesel blend > Power plants (distributed solar, distributed solar with storage, solar at u✁lity - scale, geothermal, on -shore wind, hydro run of river, hydro dam, bioenergy) - H2 produc on (with u lity scale solar) Transport technologies* Energy distribu on - Power transmission - Power distribu on - Hydrogen distribu on Charging infrastructure - Electric buses - Electric trucksElectricity Electricity Diesel (with biodiesel) Gasoline (with ethanol) LPG Electricity H2 Final energy demands (electricity and H2) - Agricultural, commercial, industrial, residen al, public services Fuel oil Pkm Tkm H2 Figure 3.2: Reference energy system for the energy and transport sectors. Based on Victor-Gallardo, Quirós-Tortós, et al., 2022. Transport demands - Freight demand - Passenger demand Final energy demands (fossil) - Agricultural, commercial, residential Transport technologies Energy distribution - Power transmission - Power distribution - Hydrogen distribution Electricity Electricity Diesel (with and without biodiesel) Gasoline (with and without ethanol) LPG Electricity Hydrogen Final energy demands (electricity and hydrogen) - Agricultural, commercial, residential, public services Passenger-kilometer Ton-kilometer Electricity Hydrogen Industry technologies Biomass LPG Natural gas Fuel oil Diesel (with and without biodiesel) Gasoline (with and without ethanol) Industry demands - Heat from boilers - Heat for cement - Heat for glass - Heat for food and general industry - Lift trucks - Onsite power generation - Other electricity demands Charging infrastructure - Electric buses - Electric trucks Heat, force, and electricity PJ Secondary energy sources > Imported fossil fuels and distribution infrastructure: gasoline, diesel, LPG, fuel oil, and natural gas - Power plants (using fuel oil) - Ethanol and biodiesel blend > Biomass variable costs > Power plants: distributed solar (with and without storage), utility- scale solar (with and without storage), geothermal, on-shore wind, hydro run of river, hydro dam, and bioenergy - Hydrogen production (using utility scale solar) Figure 3.3: Reference energy system for the energy, transport, and industry sectors. Based on Victor-Gallardo, Rodríguez-Zúñiga, Rodríguez-Arce, et al., 2022. 3.1.3. Relationships The equations here describe the modeling done outside of OSeMOSYS3. They support the pre and post-processing of OSeMOSYS inputs and outputs, following the logic described in Figure 2.5. 3See https://osemosys.readthe