Adaptive mine planning under geological uncertainty: A POMDP framework for sequential decision-making

Abdellatif Elghali; Hamza Khalifi; Jef Caers; Mostafa Benzaazoua; Yassine Taha

arxiv: 2605.13702 · v1 · pith:RIUKTZYTnew · submitted 2026-05-13 · 💻 cs.AI

Adaptive mine planning under geological uncertainty: A POMDP framework for sequential decision-making

Hamza Khalifi , Jef Caers , Yassine Taha , Mostafa Benzaazoua , Abdellatif Elghali This is my paper

Pith reviewed 2026-05-14 18:14 UTC · model grok-4.3

classification 💻 cs.AI

keywords mine planninggeological uncertaintyPOMDPadaptive policysequential decision makingsimulated annealingensemble smoothernet present value

0 comments

The pith

Mine scheduling as a POMDP produces adaptive policies that shrink the expectation-reality gap from 22.3% to 4.6% and raise realized NPV by up to USD44.6M under prior error.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper argues that conventional fixed extraction plans treat geological uncertainty as a static hedge across scenarios. Instead it models the problem as a POMDP so that each extraction and routing choice explicitly accounts for how future observations will revise beliefs about the orebody. A hybrid solver approximates long-term value with simulated annealing while updating beliefs via ensemble smoother with multiple data assimilation. When the prior matches reality the adaptive policy improves NPV by USD8.4M; when the prior is off by 10% the gain reaches USD44.6M. This shows that sequential belief revision turns uncertainty from a passive constraint into an active source of value.

Core claim

Formulating mine production scheduling as a POMDP allows extraction and routing decisions to be chosen sequentially, each evaluated by its expected long-term value under the current belief state; after each period the belief is updated with new observations via ES-MDA. The resulting policy, approximated by simulated annealing, closes the expectation-reality gap from 22.3% to 4.6% on a copper-gold open-pit complex and yields higher realized NPV than one-shot stochastic optimization, with even larger gains when the initial geological prior is systematically misspecified.

What carries the argument

The hybrid SA-POMDP architecture that approximates action values with simulated annealing and refreshes the belief state with ensemble smoother multiple data assimilation at each decision epoch.

If this is right

Realized net present value rises because decisions adapt to the information actually revealed during mining.
The expectation-reality gap shrinks when future belief updates are folded into the value calculation.
Performance remains superior even when the initial geological model is biased by 10%.
Uncertainty is converted from a fixed hedge into an active driver of extraction sequencing.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same belief-update loop could be applied to other long-horizon extraction problems such as oil-field development or reservoir management.
Replacing the simulated-annealing approximator with a learned value function might further reduce the computational burden for larger deposits.
Operators could test the framework by running parallel static and adaptive plans on a single deposit and comparing actual cash flows.

Load-bearing premise

The hybrid solver produces a policy whose expected value under the true unknown geology can be estimated without large approximation bias.

What would settle it

Run the adaptive policy on a deposit whose true block grades are known in advance and check whether the realized NPV matches the policy's computed expected value within a few percent.

Figures

Figures reproduced from arXiv: 2605.13702 by Abdellatif Elghali, Hamza Khalifi, Jef Caers, Mostafa Benzaazoua, Yassine Taha.

read the original abstract

Strategic mine production scheduling under geological uncertainty is conventionally formulated as a stochastic optimization problem in which a fixed extraction sequence and routing decisions are computed ex ante. This plan-driven paradigm treats uncertainty as passive: decisions are hedged across geological scenarios, but planning does not anticipate how future observations will inform future decisions. We propose a different perspective by formulating mine scheduling as a Partially Observable Markov Decision Process (POMDP), in which extraction and routing decisions are made sequentially with planning explicitly integrating the expectation of future belief updates. To achieve computational tractability, we introduce a hybrid SA-POMDP architecture that combines simulated annealing-based (SA) value approximation with ensemble-based belief updating via ensemble smoother with multiple data assimilation (ES-MDA). At each decision epoch, candidate actions are evaluated through their expected long-term value under the current belief, and the belief is updated as mining observations are assimilated. This yields an adaptive policy rather than a fixed plan. We evaluate the framework on a copper-gold open-pit mining complex with multiple processing destinations. Under a statistically consistent prior, the SA-POMDP reduces the expectation-reality gap from 22.3% to 4.6%, improving realized NPV by USD8.4M relative to one-shot stochastic optimization. Under systematic prior misspecification of 10%, the adaptive framework outperforms static planning by up to USD44.6M (36.9%), demonstrating structural robustness beyond scenario hedging. These results show that sequential belief updating transforms geological uncertainty from a passive constraint into an active component of value creation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper reframes mine scheduling as a POMDP to enable adaptive decisions via belief updates and reports clear NPV gains on a copper-gold case, but the SA approximation's accuracy is unverified.

read the letter

The main point is that this work moves mine production scheduling from a one-shot stochastic plan to a sequential POMDP where decisions adapt as new observations update the geological belief state through ES-MDA. They pair that with simulated annealing to keep the value estimates tractable on a multi-destination open-pit copper-gold example. Under a consistent prior the adaptive policy cuts the expectation-reality gap from 22.3% to 4.6% and lifts realized NPV by about $8.4M versus static optimization; the advantage grows to $44.6M when the prior is off by 10%. That framing of uncertainty as an active learning process rather than just a hedge is the useful shift. The numerical results are presented cleanly and the robustness claim under misspecification is worth noting. The soft spot is the reliance on SA for long-horizon value approximation in a high-dimensional belief space. SA is a heuristic whose error depends on cooling schedule and neighborhood design, yet the abstract gives no convergence diagnostics, ensemble-size sensitivity, or comparison against a higher-fidelity solver. Without those checks it is possible some of the reported outperformance is tied to the solver rather than the POMDP itself. This is for people working on stochastic optimization or sequential decisions in resource extraction. A reader already familiar with POMDPs or ensemble methods will see how the pieces are combined for an industrial scheduling problem. The core idea is coherent and the application concrete enough that it deserves a serious referee, even if the approximation quality will need more evidence in revision. I would send it out for review.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes reformulating mine production scheduling under geological uncertainty as a POMDP to enable adaptive sequential decisions that account for future information from observations. A hybrid SA-POMDP method is introduced, using simulated annealing to approximate value functions and ES-MDA for belief updates via ensemble assimilation. On a copper-gold open-pit case study, the approach reduces the expectation-reality gap from 22.3% to 4.6% and improves realized NPV by USD 8.4M compared to one-shot stochastic optimization, with larger gains (up to USD 44.6M) under 10% prior misspecification.

Significance. If validated, the results would indicate that POMDP-based adaptive planning can convert geological uncertainty from a hedging constraint into an active driver of value creation through sequential belief updating, offering substantial practical improvements in mining operations. The framework's robustness to prior misspecification is particularly noteworthy for real-world applications where geological models are imperfect.

major comments (2)

[Abstract and Evaluation] Abstract and Evaluation: The central claims of gap reduction from 22.3% to 4.6% and NPV improvements of USD8.4M and USD44.6M depend on the accuracy of the SA-based value approximation in the POMDP. However, no independent error bounds, convergence diagnostics for the annealing schedule, or comparisons to exact or higher-fidelity solvers (e.g., on smaller instances) are provided to rule out systematic bias in long-horizon estimates.
[§3] §3 (POMDP formulation and belief update): The definition and computation of the 'expectation-reality gap' is not fully specified, including the number of realizations used, how the true geology is simulated for evaluation, and sensitivity to ES-MDA ensemble size or annealing parameters; this makes it difficult to assess whether the reported 4.6% figure is robust.

minor comments (2)

[Notation] Ensure consistent use of symbols for belief states and value functions across sections to avoid ambiguity in the hybrid architecture description.
[References] Add citations to recent POMDP solvers in mining or resource management contexts for better positioning.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments, which help clarify the validation requirements for our SA-POMDP framework. We address each major comment below and will revise the manuscript to incorporate additional diagnostics and explicit specifications.

read point-by-point responses

Referee: [Abstract and Evaluation] Abstract and Evaluation: The central claims of gap reduction from 22.3% to 4.6% and NPV improvements of USD8.4M and USD44.6M depend on the accuracy of the SA-based value approximation in the POMDP. However, no independent error bounds, convergence diagnostics for the annealing schedule, or comparisons to exact or higher-fidelity solvers (e.g., on smaller instances) are provided to rule out systematic bias in long-horizon estimates.

Authors: We agree that the SA value approximation is heuristic and that stronger validation would improve confidence in the reported metrics. Exact POMDP solvers remain intractable at the scale of the full mine-planning instance (state space >10^6), but we will add (i) convergence diagnostics showing stabilization of the SA value estimates across iterations, (ii) independent error bounds derived from the variance of 20 independent SA runs per state, and (iii) a new comparison on a smaller synthetic instance (reduced blocks and horizons) where exact value iteration is feasible. These additions will appear in a revised §4 and supporting figures. revision: yes
Referee: [§3] §3 (POMDP formulation and belief update): The definition and computation of the 'expectation-reality gap' is not fully specified, including the number of realizations used, how the true geology is simulated for evaluation, and sensitivity to ES-MDA ensemble size or annealing parameters; this makes it difficult to assess whether the reported 4.6% figure is robust.

Authors: We thank the referee for highlighting this omission. The expectation-reality gap is the percentage difference between the a-priori expected NPV (under the initial belief) and the realized NPV obtained by rolling out the adaptive policy on a simulated true orebody; the true orebody is drawn from the same geostatistical model that generated the prior ensemble. Evaluation uses 200 independent realizations, an ES-MDA ensemble of size 100, and the annealing schedule with initial temperature 1000 and cooling rate 0.95. We will expand §3 with these exact parameters and add a sensitivity table showing that the 4.6% gap remains stable for ensemble sizes 50–200 and modest changes in annealing parameters. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical performance from simulation on mining instance

full rationale

The paper's load-bearing claims (22.3% to 4.6% gap reduction, +USD8.4M NPV, +USD44.6M under misspecification) are numerical outcomes of running the SA-POMDP policy on a specific copper-gold open-pit instance and comparing realized values against one-shot stochastic optimization. These are not derived by construction from the POMDP equations, SA value function, or ES-MDA updates; they require external geological realizations and forward simulation. No self-citations, fitted parameters renamed as predictions, or ansatzes smuggled via prior work appear in the derivation. The SA approximation is a computational heuristic whose bias is a separate correctness concern, not a circularity reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities are stated. The POMDP and SA components are standard algorithmic building blocks; the hybrid integration is presented as the novel engineering step.

pith-pipeline@v0.9.0 · 5596 in / 1235 out tokens · 41231 ms · 2026-05-14T18:14:21.092076+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

16 extracted references · 16 canonical work pages

[1]

Simultaneous stochastic optimization of production sequence and dynamic cut -off grades in an open pit mining operation

Paithankar A, Chatterjee S, Goodfellow R, Asad MWA. Simultaneous stochastic optimization of production sequence and dynamic cut -off grades in an open pit mining operation. Resources Policy 2020;66. https://doi.org/10.1016/j.resourpol.2020.101634

work page doi:10.1016/j.resourpol.2020.101634 2020
[2]

Global optimization of open pit mining complexes with uncertainty

Goodfellow RC, Dimitrakopoulos R. Global optimization of open pit mining complexes with uncertainty. Applied Soft Computing Journal 2016;40:292–304. https://doi.org/10.1016/j.asoc.2015.11.038

work page doi:10.1016/j.asoc.2015.11.038 2016
[3]

Simultaneous stochastic optimization of mining complexes - mineral value chains: an overview of concepts, examples and comparisons

Dimitrakopoulos R, Lamghari A. Simultaneous stochastic optimization of mining complexes - mineral value chains: an overview of concepts, examples and comparisons. Int J Min Reclam Environ 2022;36:443–60. https://doi.org/10.1080/17480930.2022.2065730

work page doi:10.1080/17480930.2022.2065730 2022
[4]

Production scheduling with uncertain supply: A new solution to the open pit mining problem

Ramazan S, Dimitrakopoulos R. Production scheduling with uncertain supply: A new solution to the open pit mining problem. Optimization and Engineering 2013;14:361–80. https://doi.org/10.1007/s11081-012- 9186-2

work page doi:10.1007/s11081-012- 2013
[5]

Adaptive open -pit mining planning under geological uncertainty

Armstrong M, Lagos T, Emery X, Homem -de-Mello T, Lagos G, Sauré D. Adaptive open -pit mining planning under geological uncertainty. Resources Policy 2021;72. https://doi.org/10.1016/j.resourpol.2021.102086

work page doi:10.1016/j.resourpol.2021.102086 2021
[6]

A framework for adaptive open-pit mining planning under geological uncertainty

Lagos T, Armstrong M, Homem-de-Mello T, Lagos G, Sauré D. A framework for adaptive open-pit mining planning under geological uncertainty. Optimization and Engineering 2022;23:111 –46. https://doi.org/10.1007/s11081-020-09557-0

work page doi:10.1007/s11081-020-09557-0 2022
[7]

Artificial Intelligence Planning and acting in partially observable stochastic domains

Pack Kaelbling L, Littman ML, Cassandra ’, ’ AR. Artificial Intelligence Planning and acting in partially observable stochastic domains. vol. 101. 1998

work page 1998
[8]

AI -driven optimization under uncertainty for mineral processing operations

Xu W, Eskanlou A, Arief M, Yin DZ, Caers J. AI -driven optimization under uncertainty for mineral processing operations. Sustainable Earth Resources Communications 2025;1:100 –12. https://doi.org/10.46690/serc.2025.02.07

work page doi:10.46690/serc.2025.02.07 2025
[9]

Managing Geological Uncertainty in Critical Mineral Supply Chains: A POMDP Approach with Application to U.S

Arief M, Alonso Y , Oshiro C, Xu W, Corso A, Yin DZ, et al. Managing Geological Uncertainty in Critical Mineral Supply Chains: A POMDP Approach with Application to U.S. Lithium Resources 2025

work page 2025
[10]

G. D. Barros E, Van den Hof PMJ, Jansen JD. Value of information in closed-loop reservoir management. Comput Geosci 2016;20:737–49. https://doi.org/10.1007/s10596-015-9509-4

work page doi:10.1007/s10596-015-9509-4 2016
[11]

A sequential decision -making framework with uncertainty quantification for groundwater management

Wang Y , Zechner M, Mern JM, Kochenderfer MJ, Caers JK. A sequential decision -making framework with uncertainty quantification for groundwater management. Adv Water Resour 2022;166. https://doi.org/10.1016/j.advwatres.2022.104266

work page doi:10.1016/j.advwatres.2022.104266 2022
[12]

Planning treatment of ischemic heart disease with partially observable Markov decision processes

Hauskrecht M, Fraser H. Planning treatment of ischemic heart disease with partially observable Markov decision processes. vol. 18. 2000

work page 2000
[13]

BetaZero: Belief -State Planning for Long -Horizon POMDPs using Learned Approximations 2024

Moss RJ, Corso A, Caers J, Kochenderfer MJ. BetaZero: Belief -State Planning for Long -Horizon POMDPs using Learned Approximations 2024

work page 2024
[14]

Ensemble smoother with multiple data assimilation

Emerick AA, Reynolds AC. Ensemble smoother with multiple data assimilation. Comput Geosci 2013;55:3–15. https://doi.org/10.1016/j.cageo.2012.03.011

work page doi:10.1016/j.cageo.2012.03.011 2013
[15]

A simulated annealing approach to mine production scheduling

Kumral M, Dowd PA. A simulated annealing approach to mine production scheduling. Journal of the Operational Research Society 2005;56:922–30. https://doi.org/10.1057/palgrave.jors.2601902

work page doi:10.1057/palgrave.jors.2601902 2005
[16]

Garbage Detection using Advanced Object Detection Techniques,

Fathollahzadeh K, Asad MWA, Mardaneh E, Cigla M. Review of Solution Methodologies for Open Pit Mine Production Scheduling Problem. Int J Min Reclam Environ 2021;35:564 –99. https://doi.org/10.1080/17480930.2021.1888395

work page doi:10.1080/17480930.2021.1888395 2021

[1] [1]

Simultaneous stochastic optimization of production sequence and dynamic cut -off grades in an open pit mining operation

Paithankar A, Chatterjee S, Goodfellow R, Asad MWA. Simultaneous stochastic optimization of production sequence and dynamic cut -off grades in an open pit mining operation. Resources Policy 2020;66. https://doi.org/10.1016/j.resourpol.2020.101634

work page doi:10.1016/j.resourpol.2020.101634 2020

[2] [2]

Global optimization of open pit mining complexes with uncertainty

Goodfellow RC, Dimitrakopoulos R. Global optimization of open pit mining complexes with uncertainty. Applied Soft Computing Journal 2016;40:292–304. https://doi.org/10.1016/j.asoc.2015.11.038

work page doi:10.1016/j.asoc.2015.11.038 2016

[3] [3]

Simultaneous stochastic optimization of mining complexes - mineral value chains: an overview of concepts, examples and comparisons

Dimitrakopoulos R, Lamghari A. Simultaneous stochastic optimization of mining complexes - mineral value chains: an overview of concepts, examples and comparisons. Int J Min Reclam Environ 2022;36:443–60. https://doi.org/10.1080/17480930.2022.2065730

work page doi:10.1080/17480930.2022.2065730 2022

[4] [4]

Production scheduling with uncertain supply: A new solution to the open pit mining problem

Ramazan S, Dimitrakopoulos R. Production scheduling with uncertain supply: A new solution to the open pit mining problem. Optimization and Engineering 2013;14:361–80. https://doi.org/10.1007/s11081-012- 9186-2

work page doi:10.1007/s11081-012- 2013

[5] [5]

Adaptive open -pit mining planning under geological uncertainty

Armstrong M, Lagos T, Emery X, Homem -de-Mello T, Lagos G, Sauré D. Adaptive open -pit mining planning under geological uncertainty. Resources Policy 2021;72. https://doi.org/10.1016/j.resourpol.2021.102086

work page doi:10.1016/j.resourpol.2021.102086 2021

[6] [6]

A framework for adaptive open-pit mining planning under geological uncertainty

Lagos T, Armstrong M, Homem-de-Mello T, Lagos G, Sauré D. A framework for adaptive open-pit mining planning under geological uncertainty. Optimization and Engineering 2022;23:111 –46. https://doi.org/10.1007/s11081-020-09557-0

work page doi:10.1007/s11081-020-09557-0 2022

[7] [7]

Artificial Intelligence Planning and acting in partially observable stochastic domains

Pack Kaelbling L, Littman ML, Cassandra ’, ’ AR. Artificial Intelligence Planning and acting in partially observable stochastic domains. vol. 101. 1998

work page 1998

[8] [8]

AI -driven optimization under uncertainty for mineral processing operations

Xu W, Eskanlou A, Arief M, Yin DZ, Caers J. AI -driven optimization under uncertainty for mineral processing operations. Sustainable Earth Resources Communications 2025;1:100 –12. https://doi.org/10.46690/serc.2025.02.07

work page doi:10.46690/serc.2025.02.07 2025

[9] [9]

Managing Geological Uncertainty in Critical Mineral Supply Chains: A POMDP Approach with Application to U.S

Arief M, Alonso Y , Oshiro C, Xu W, Corso A, Yin DZ, et al. Managing Geological Uncertainty in Critical Mineral Supply Chains: A POMDP Approach with Application to U.S. Lithium Resources 2025

work page 2025

[10] [10]

G. D. Barros E, Van den Hof PMJ, Jansen JD. Value of information in closed-loop reservoir management. Comput Geosci 2016;20:737–49. https://doi.org/10.1007/s10596-015-9509-4

work page doi:10.1007/s10596-015-9509-4 2016

[11] [11]

A sequential decision -making framework with uncertainty quantification for groundwater management

Wang Y , Zechner M, Mern JM, Kochenderfer MJ, Caers JK. A sequential decision -making framework with uncertainty quantification for groundwater management. Adv Water Resour 2022;166. https://doi.org/10.1016/j.advwatres.2022.104266

work page doi:10.1016/j.advwatres.2022.104266 2022

[12] [12]

Planning treatment of ischemic heart disease with partially observable Markov decision processes

Hauskrecht M, Fraser H. Planning treatment of ischemic heart disease with partially observable Markov decision processes. vol. 18. 2000

work page 2000

[13] [13]

BetaZero: Belief -State Planning for Long -Horizon POMDPs using Learned Approximations 2024

Moss RJ, Corso A, Caers J, Kochenderfer MJ. BetaZero: Belief -State Planning for Long -Horizon POMDPs using Learned Approximations 2024

work page 2024

[14] [14]

Ensemble smoother with multiple data assimilation

Emerick AA, Reynolds AC. Ensemble smoother with multiple data assimilation. Comput Geosci 2013;55:3–15. https://doi.org/10.1016/j.cageo.2012.03.011

work page doi:10.1016/j.cageo.2012.03.011 2013

[15] [15]

A simulated annealing approach to mine production scheduling

Kumral M, Dowd PA. A simulated annealing approach to mine production scheduling. Journal of the Operational Research Society 2005;56:922–30. https://doi.org/10.1057/palgrave.jors.2601902

work page doi:10.1057/palgrave.jors.2601902 2005

[16] [16]

Garbage Detection using Advanced Object Detection Techniques,

Fathollahzadeh K, Asad MWA, Mardaneh E, Cigla M. Review of Solution Methodologies for Open Pit Mine Production Scheduling Problem. Int J Min Reclam Environ 2021;35:564 –99. https://doi.org/10.1080/17480930.2021.1888395

work page doi:10.1080/17480930.2021.1888395 2021