Embedding project-specific, unsystematic (idiosyncratic) risks into discount rates substantially reduces the expected monetary value of geothermal developments, inflates appraisal expenditures, and impairs decision-making accuracy without reducing expected financial losses, according to a modeling study published in Geothermal Energy. Researchers evaluated low-enthalpy geothermal field development planning using decision-theoretic artificial intelligence agents designed to optimize sequential decisions under deep subsurface and market uncertainty. In capital-intensive renewable infrastructure, project planners frequently inflate discount rates above the corporate cost of capital by appending risk premiums to reflect technical or reservoir risks. Evaluating European low-enthalpy geothermal reservoir models under varying discount rate assumptions—a baseline cost of capital of 7 percent alongside risk-adjusted rates of 10 percent and 13 percent representing 3 percent and 6 percent risk premiums—the investigation reveals that this widespread corporate budgeting practice distorts development strategy selection and encourages excessive, redundant data acquisition while failing to provide meaningful downside risk protection.
Geothermal energy currently provides roughly 0.8 percent of global energy demand, despite operating with exceptionally high capacity factors and possessing broad generation potential across multiple continents. Within Europe, geothermal energy production has achieved steady expansion, posting growth in excess of 10 percent over the past decade, with low-enthalpy geothermal systems projected to supply up to 20 percent of district heating expansion between 2035 and 2050. Nevertheless, widespread commercial adoption remains constrained by marginal baseline economics, public concerns surrounding induced seismicity, social acceptance hurdles, and prolonged development timelines. The comprehensive cycle spanning initial surface exploration to operational heat production regularly surpasses seven years. These extended appraisal and construction schedules severely weigh on investment economics, deferring cash flows and slowing the deployment of geothermal energy as a scalable tool for mitigating greenhouse gas emissions. Overcoming these barriers demands streamlined field appraisal strategies that minimize front-end evaluation expenditures while accurately resolving subsurface uncertainties.
To evaluate development planning under realistic geological conditions, the researchers analyzed a case study centered on a low-enthalpy saline aquifer in Europe targeted for large-scale district heating. The subsurface reservoir formation is primarily composed of loosely to moderately cemented, clast-supported conglomerates containing medium to coarse gravel and frequent cobbles of limestone, dolostone, and crystalline lithologies, interspersed with subordinate intercalations of fine sand and silty marly clay. Interpreted sedimentologically as braided river deposits marked by overlapping channels and gravel bars, the reservoir exhibits minimal marine influence confined to distal sections. The reservoir zone extends across an area measuring 8 kilometers by 6 kilometers and encompasses 820,000 grid cells, with the reservoir top dipping from a depth of 1,500 meters in the north to 3,800 meters in the south, maintaining an average gross thickness of approximately 200 meters. Using Latin Hypercube Sampling across uncertain geological, petrophysical, and dynamic thermal parameters—including porosity, permeability, rock heat capacity, and thermal conductivity—the research team generated an ensemble of 250 distinct numerical reservoir models to capture initial subsurface beliefs.
Building upon the 250 subsurface reservoir realizations, the authors simulated production forecasts across 11 discrete field development scenarios designed for commercial heat extraction over a 20-year operational horizon. The investigated scenarios encompassed vertical well trajectories arranged in Line Drive configurations oriented either east–west or north–south, alongside central 5-spot patterns where central injection wells are encircled by peripheral production wells. While single doublet systems typically maintain inter-well spacings exceeding 1 kilometer to prevent premature thermal breakthrough, large-scale commercial implementations may require tighter well spacing to satisfy thermal delivery quotas, resulting in variable cold-water front propagation and reservoir cooling rates across different geological realizations. To evaluate project viability, the team coupled physical reservoir performance with economic parameters, including capital expenditures, operating expenditures, and heat pricing schedules. By sampling 50 distinct economic parameter configurations across each of the 250 geological models using Latin Hypercube Sampling, the simulation generated 12,500 net present value realizations for each development scenario, establishing baseline cumulative distribution functions under a 7 percent discount rate representing the standard cost of capital.
Project appraisal in geothermal development frequently relies on net present value and internal rate of return calculations, yet the study underscores that conventional internal rate of return metrics provide an unreliable basis for ranking capital investments. In the modeled reservoir realizations, an internal rate of return of 20 percent corresponded to net present values ranging widely from 166 million euros to 439 million euros. Because the internal rate of return represents solely the mathematical discount rate at which net present value equals zero, it fails to account for the total monetary scale of cash flows, leading to situations where projects of drastically different financial magnitudes exhibit identical rates of return. Furthermore, established corporate finance literature highlights that the metric relies on problematic assumptions regarding interim cash flow reinvestment rates, generates ranking inconsistencies relative to net present value, struggles with fluctuating cash flow signs, favors short-lived projects with small capital commitments over high-value long-term assets, and miscalculates delayed investments. Consequently, the authors conclude that internal rate of return should not be used to prioritize field development scenarios.
In established financial theory, an asset’s cost of capital reflects systematic market risk, whereas unsystematic or idiosyncratic risks—such as subsurface drilling failures or local technical malfunctions—can theoretically be diversified away across an investor’s broader portfolio. Models such as the Capital Asset Pricing Model quantify systematic market volatility through beta to estimate equity return requirements, which feed directly into a firm’s weighted average cost of capital. Theoretical consensus maintains that asset-specific risks should be addressed through probabilistic cash flow modeling rather than inflated discount rates. Nevertheless, empirical surveys reveal that corporate executives routinely add discretionary risk premiums to hurdle rates, with earlier survey evidence demonstrating that only 48 percent of chief financial officers would approve projects exceeding the weighted average cost of capital by 4 percent if earnings per share declined. Such inflated hurdle rates frequently stem from corporate agency conflicts, cognitive biases including retrievability bias, historical miscalibrations from past project failures, desires to signal capital allocation conservatism to financial markets, or fundamental misunderstandings regarding the true cost of capital.
To resolve the sequential decision challenges of field development and data acquisition without resorting to ad hoc risk premiums, the research team implemented a Partially Observable Markov Decision Process solved by a decision-theoretic artificial intelligence agent. Traditional Value of Information methodologies evaluate data gathering steps in isolation using Bayes’ theorem, adhering to a greedy heuristic that cannot account for the compounding informational value of sequential measurements. The formulated POMDP framework explicitly accounts for unobservable reservoir states across the parameter distributions through a formal mathematical tuple comprising environmental states, actions, transition functions, rewards, observations, observation models, and discount factors. The agent architecture integrates a recursive Bayesian belief updater with an optimal decision policy solved via the Successive Approximations of the Reachable Space under Optimal Policies (SARSOP) algorithm. Within this architecture, the agent navigated 32 distinct action choices, comprising 20 sequential data acquisition actions—including three appraisal wells, dynamic pressure testing, and thermal property analyses—alongside 11 field development options and a neutral walk-away option yielding zero financial reward.
Evaluating project economics in the absence of sequential data acquisition demonstrated that introducing risk premiums into the discount rate directly distorts development decision-making. When evaluated under the baseline 7 percent cost of capital, development scenario 7 emerged as the optimal commercial choice, generating an expected monetary value of 202 million euros alongside a 63 percent probability of economic success, defined as the proportion of simulated outcomes achieving a positive net present value. However, when an idiosyncratic risk premium was added—raising the discount rate to 13 percent—the apparent financial ranking inverted, incorrectly identifying development scenario 10 as the most attractive option. When this chosen scenario was subsequently evaluated under the true 7 percent economic baseline, its actual expected monetary value declined from 202 million euros to 180 million euros, wiping out more than 22 million euros in expected project value due solely to the distorting effect of the inflated discount factor on project selection.
When sequential data acquisition was integrated into the agent-driven workflow, inflating the discount rate triggered severe compounding operational and financial inefficiencies. In the agent simulations, policies optimized under 10 percent and 13 percent discount rates caused expected monetary value to plummet by up to 26 million euros when implemented against actual project economics. Mean data acquisition expenditures escalated sharply from 0.55 million euros under the 7 percent baseline to 0.96 million euros under the 13 percent policy, while the average number of acquired appraisal actions climbed from 4.98 to 7.32. Because higher discount rates penalize future cash flows more aggressively, the artificial intelligence agent attempted to compensate for discounted future revenues by executing additional, redundant testing to compress uncertainty prior to committing capital. This compensatory behavior lengthened the appraisal phase, inflated development budgets, diminished the accuracy of development scenario selection from 45 percent down to 39 percent, and reduced correct go/no-go decisions from 79 percent down to 65 percent.
Crucially, the simulation results confirmed that incorporating idiosyncratic risk premiums into discount rates fails to deliver its primary intended benefit: mitigating downside financial risk. An evaluation of expected loss—defined as the conditional mean of all simulated negative net present value outcomes—demonstrated no meaningful reduction across the 7 percent, 10 percent, and 13 percent discount rate scenarios. Instead of protecting capital, the practice simply curtailed project viability, driving a severe decline in the probability of maturation, particularly when projects faced a required probability of economic success below 70 percent. To eliminate these distortions, the authors advocate separating physical subsurface uncertainty from discount rate estimation. By pairing physical numerical reservoir models with Latin Hypercube sampling and probabilistic metrics such as the probability of economic success, operators can rigorously quantify asset-specific risk without inflating hurdle rates, ensuring accurate development planning, cost-effective appraisal programs, and disciplined capital allocation across sustainable energy portfolios.
Subject of Research: Decision-theoretic modeling of risk integration in low-enthalpy geothermal field planning
Article Title: Incorporating risk in geothermal field development planning: applying decision-theoretic agents
Article References: Clemens, T., Corso, A., Chiotoroiu, M.-M., & Kochenderfer, M. J. (2025). Incorporating risk in geothermal field development planning: applying decision-theoretic agents. Geothermal Energy, 13(1), Article 46. https://doi.org/10.1186/s40517-025-00369-y
Image Credits: AI Generated
DOI: 10.1186/s40517-025-00369-y
Keywords: geothermal energy, low-enthalpy reservoirs, decision-theoretic agents, partially observable Markov decision processes, discount rates, idiosyncratic risk, systematic risk, expected monetary value, data acquisition, SARSOP algorithm, probability of economic success, probability of maturation, capital asset pricing model, weighted average cost of capital, district heating, Latin hypercube sampling, subsurface reservoir modeling
News Source: Violet Maxwell. (November 13, 2026). Inflated Geothermal Discount Rates Reduce Project Value and Increase Exploration Costs. Scienmag.



