Addressing extreme weather events for the renewable power-water-heating sectors in Neom, Saudi Arabia

Saudi Arabia targets at least 50% renewable electricity under Vision 2030, aligned with global decarbonization goals. The integration of variable renewable energy (VRE) poses challenges for reliability due to intermittency and the need to consider high-resolution spatiotemporal variability in planning models. Multi-sector coupling across power, water, and heat can yield more robust and cost-effective designs than optimizing sectors independently, especially in arid regions where desalination tightly links water and power. Yet, high-resolution data cause computational complexity, prompting use of temporal and spatial aggregation methods that risk smoothing out rare but critical extremes. This study addresses the research question of how to plan reliable, cost-effective, fully or near-fully renewable multi-sector systems when extreme weather and demand events are present, using Neom, Saudi Arabia, as a case study.

Prior works use clustering and other time-series aggregation techniques (e.g., k-means, k-medoids, hierarchical, k-shape) to generate representative periods for planning, capturing variance but often missing outliers. Studies show co-optimizing power and water can reduce costs through coordinated capacity and operations. However, clustering smooths extremes, risking under-preparedness for rare events. Several methods incorporate extreme periods: Souayfane et al. found extremes recur seasonally and reduced-variability datasets underestimate storage; Kotzur et al. define extremes via high load and low PV and seed clusters with extremes; Bahl et al. add feasibility constraints iteratively; Teichgraeber et al. identify and embed extreme periods iteratively for residential systems; Li et al. extend to power systems. Most prior studies are single-node and sector-specific, often deterministic, with limited integrated multi-sector focus and exploration of the role of specific technologies (e.g., CSP, geothermal) under extremes.

The model is a two-stage stochastic program over a 10-year horizon with hourly resolution. First-stage variables are annual investment decisions in generation and storage across power, water, and heat; second-stage variables are hourly operations under uncertain future demands. Three demand scenarios (low, medium, high) are used with probabilities 0.3, 0.5, and 0.2, respectively. The objective minimizes total cost (annualized CAPEX, fixed O&M, expected variable O&M) subject to investment and operational constraints for all technologies. Technologies include: power (PV, CSP with thermal energy storage (TES), wind, batteries, CCGT, CHP), water (RO, MED, water tanks), and heat (resistive heaters centralized/decentralized, ASHP, geothermal district heating, CHP, CSP). Spatially, the Neom region is partitioned into nine nodes via k-means on renewable resource attributes; power and storage can be sited at any node, water desalination at coastal nodes, and heat at demand nodes. Temporally, representative days are derived by k-means clustering of 11 years (2008–2018) of hourly weather (DNI, GHI, wind speed, temperature) and normalized demand data; diurnal variability is preserved with 24-hour representative days. To address the limitation that representative days miss extremes, an iterative clustering-optimization workflow is used: (i) cluster to get representative days; (ii) solve design optimization on these days; (iii) simulate fixed-capacity operations over full hourly multi-year data; (iv) assess external supply slack (energy, water, heat); (v) identify extreme days as those with largest cumulative slack; (vi) add extremes to the representative set with computed weights; (vii) re-optimize; iterate until slack below tolerance. Slack variables in energy, water, and heat balances quantify unmet demand requiring external supply and guide extreme-day selection. Notes: transmission constraints for power and water are omitted (assume sufficient build-out), heat is nodal without inter-node transfer; fully renewable cases yield linear programs; inclusion of fossil generation introduces binary variables (MILP). Costs and techno-parameters are from NREL ATB 2022 for power; literature for desalination, storage, and heat; geothermal is modeled via a scalable hydrothermal doublet at 2500 m depth (105 °C, 25 MPa) with utilization factor 0.85. Weather resources are from WRF-based reanalyses validated against in-situ observations. Computations are implemented in GAMS/CPLEX with 1% optimality gap and 24 h limit.

- Fully renewable baseline without extremes (first iteration) invests by 2029: total power 22 GW (~40% CSP, ~30% PV, ~30% wind), 0 GWh batteries, 88 GWh TES; water: 56.2 thousand m³/h RO and 1.26 million m³ tanks; heat: 1.6 GW (32% geothermal, 33% resistive heaters, 5% ASHP, 30% CSP). Simulation over full data reveals substantial KSA grid purchases in winter, highlighting infeasibility under extremes. - Adding extreme days increases total power capacity by 28% to 28 GW; wind +68%, PV +28%; CSP and TES unchanged; RO +8%; water tanks nearly triple. Heat sector capacity falls 29% to 1.2 GW with CSP no longer used for heat and more resistive heaters. Grid electricity purchases drop by 91%, from 2295 GWh (no extremes) to 197 GWh (with extremes) annually; December purchases fall from 706 GWh to 64 GWh. Total system cost rises by 13% (power +12%, water +37%; heat largely unchanged) to achieve higher reliability. - Allowing up to 10% fossil-based annual power generation (CHP, CCGT) yields first-iteration capacity of 29.7 GW (44% wind, 37% PV, 19% fossil) with only 15 GWh batteries; operations simulation purchases just 4.44 GWh from KSA in year 10, a reduction of 99.9% and 97.8% relative to fully renewable cases without and with extremes, respectively. RO is favored over MED despite CHP availability, as replacing RO with MED would require ~3.5 GW additional heat and increase total system cost by ~6%. - Role of CSP: With extremes, a fully renewable system without CSP requires 40.7 GW (61% PV, 39% wind) and 114 GWh batteries, whereas allowing CSP reduces total power capacity by ~31% and substitutes batteries with ~84 GWh TES. CSP cuts total system cost by ~31% versus a no-CSP design but reduces reliability: grid purchases with extremes are 197 GWh (with CSP) vs 116 GWh (without CSP), a 41% decrease without CSP; without extremes, purchases drop from 2295 GWh (with CSP) to 357 GWh (without CSP), an 84% decrease. - Role of geothermal: Geothermal heating displaces resistive heaters, eliminating CSP-for-heat, and reduces total heat capacity by ~45%. Power capacity decreases by ~9% (same mix proportions), desalination and water storage fall by ~3% and ~6%, respectively. Total system cost decreases by ~5.3% while maintaining reliability benefits from lower electric heat demand. - Spatial siting: Considering extremes expands wind siting from node 4 to nodes 6 and 9 along the coast, improving winter reliability. - Summary performance (high-demand scenario): Accounting for extremes substantially reduces external energy needs at increased cost; limited fossil generation substantially improves reliability at lower cost and storage needs; CSP is cost-reducing but less flexible and thus less reliable under extremes; geothermal provides cross-sector cost and capacity benefits.

The research question concerns how to design reliable, cost-effective multi-sector renewable systems under extreme weather and demand variability. The integrated clustering-optimization with extreme-day augmentation directly addresses this by iteratively exposing candidate designs to full historical variability, identifying failure modes via slack variables, and re-optimizing with those extremes included. Results show that ignoring extremes leads to under-built systems that rely on external power during winter low-solar and low-wind periods. Incorporating extremes strategically increases wind and PV capacities and water storage, and adapts the heat mix toward more electrically driven backup, notably improving reliability (91% less external power) albeit at a 13% higher total cost. Allowing a small (≤10%) share of fossil generation greatly improves system adequacy with minimal external purchases and reduced storage requirements, providing a pragmatic transitional pathway. CSP offers significant cost reductions by displacing batteries with TES but at the expense of flexibility, reducing resilience to non-solar-dominated extremes; excluding CSP increases battery use and reliability. Geothermal heating reduces electric load for heat, lowering required power capacity and system cost without compromising reliability. Collectively, these findings underscore the necessity of including extreme events in sector-coupled planning and highlight the technology trade-offs between cost and operational flexibility.

An integrated clustering-optimization framework that iteratively incorporates extreme days yields more reliable multi-sector renewable designs for Neom. Neglecting extremes underestimates capacity needs and increases reliance on external electricity, especially in winter. Including extremes drives higher investments—particularly in wind, PV, and water storage—and reduces external power purchases by 91%, though at a 13% higher total cost. Allowing up to 10% fossil generation dramatically enhances reliability and reduces storage needs, outperforming fully renewable cases even when extremes are considered. CSP substantially lowers total system costs but reduces reliability due to TES’s limited flexibility compared with batteries; systems without CSP purchase less external power under extremes. Geothermal heating is an economically attractive option that reduces power sector capacity and total cost by roughly 5.3% through lower electric heat demand. Future work should extend the framework to include transmission constraints, technology cost uncertainties, and strategies for long-duration storage and demand-side flexibility, alongside broader robustness analyses across multi-year extremes.

- Transmission constraints for power and water are not modeled; it is assumed transmission can be built to accommodate capacities. Heat is modeled nodally without inter-node transfer. - Technology price uncertainty is not considered; future cost trajectories could change investment strategies. - Representative-day methods can under-represent long-duration storage dynamics and may over- or under-estimate needs; adding multiple extreme days in sequence may overstate capacity needs versus dispersed real-world occurrences. - Greenfield assumption may limit applicability where legacy assets and infrastructure exist. - Deterministic technology performance parameters and omission of operational contingencies (e.g., outages) may affect generalizability.