Simulation Models of the Ukraine War: Methods and Forecast Accuracy

The Ukraine war has prompted an unprecedented application of quantitative simulation modeling to an active large-scale conventional conflict. Researchers across academic institutions, think tanks, and government agencies have deployed diverse modeling frameworks—from classical Lanchester attrition equations to modern agent-based models (ABMs) and system dynamics simulations—in attempts to understand, predict, and inform policy about the conflict's trajectory. This article surveys the principal modeling approaches, their inputs and outputs, their known limitations, and a retrospective assessment of how well they have matched observed outcomes.

Lanchester Attrition Models

The oldest and most mathematically tractable approach, Lanchester's combat models treat opposing forces as aggregate quantities whose rates of attrition depend on relative strength and firepower. Applied to Ukraine, these models use estimated force strengths, casualty exchange ratios, and equipment replacement rates as inputs, producing projections of relative force balance over time. Lanchester models correctly identified that Russia's initial conventional superiority would erode substantially due to Ukraine's tactical proficiency and Western equipment supply—however, they underestimated the durability of positional warfare and the ability of both sides to sustain losses for longer than classical exchange ratios would suggest.

Agent-Based Models

Agent-based models represent individual soldiers, units, or equipment systems as autonomous agents following behavioral rules. ABMs excel at modeling emergent phenomena—unexpected large-scale behaviors arising from individual-level interactions—such as the collapse of Russian logistics in the initial 2022 offensive or the tactical adaptation of Ukrainian drone warfare. DARPA's Strategic Multi-Layer Assessment program applied ABM frameworks to Ukraine with notable success in reproducing the spatial pattern of front-line evolution in 2022, but struggled to predict the specific timing andlocation of tactical breakthroughs, which depend on local morale, leadership, and terrain factors inadequately captured by current agent rules.

System Dynamics Models

System dynamics models represent the Ukraine conflict as a network of interconnected feedback loops: military production and consumption, economic capacity, population morale, political will, and alliance cohesion. These models are particularly useful for analyzing longer-term sustainability questions—can Russia sustain its defense industrial base output? At what point do Ukrainian manpower pools approach exhaustion?—rather than tactical outcomes. System dynamics modeling by the Stockholm International Peace Research Institute (SIPRI) and the Kiel Institute has generally been more accurate than tactical models in forecasting macroeconomic and production trends, though still subject to significant uncertainty around Russian production statistics and Ukrainian mobilization data quality.

Model Comparison: Forecast vs Actual

Key Inputs and Data Challenges

All simulation models of the Ukraine war face significant data quality challenges. Russian casualty figures are provided by Ukrainian official sources and Western intelligence estimates, with wide confidence intervals. Ukrainian casualty data is classified. Equipment production and loss data relies heavily on OSINT analysis from groups like Oryx, which tracks visually confirmed equipment losses—comprehensive but inevitably incomplete. Morale, a critical variable in model behavior, remains almost entirely unobservable in quantitative form for active combatants. These data limitations translate directly into model uncertainty, which practitioners are increasingly required to characterize explicitly through uncertainty quantification techniques.

FAQ

Which simulation model type has been most accurate for the Ukraine war?

System dynamics models have shown the strongest track record for industrial and economic sustainability forecasts, while agent-based models have performed best on spatial front-line dynamics. No single model type dominates across all dimensions, which is why ensemble approaches combining multiple models are increasingly preferred.

Have any simulations correctly predicted major operational outcomes?

Several models running in late 2021 correctly flagged high probability of major Russian invasion based on force buildup indicators. Some ABM-based models in summer 2022 correctly identified the Kherson axis as the most likely Ukrainian counter-offensive zone based on logistics network analysis.

What is the biggest limitation of current conflict simulation models?

The biggest limitation is modeling human factors—morale, leadership quality, unit cohesion, and political will—which are often determinative in conflict outcomes but resist quantification. Most models handle these as fixed parameters or crude proxy variables rather than dynamic, observable quantities.

Do governments use these simulations for real policy decisions?

Yes. Multiple Western governments have commissioned and used simulation studies, including Lanchester-based force planning models, to inform decisions about weapons system priorities, aid package design, and long-term security planning for Ukraine.

How can simulation models be improved for future conflicts?

Key improvement areas include: better integration of behavioral/psychological variables, real-time data feeds from OSINT sources, uncertainty quantification reporting as standard practice, and ensemble frameworks that combine multiple model types with calibrated weighting schemes.

Sources

1. SIPRI, Quantitative Models of the Ukraine Conflict, Stockholm, 2025.
2. Michael Horowitz & Alex Weisiger, Conflict Duration and Attrition Modeling, Journal of Conflict Resolution, 2024.
3. RAND Corporation, Agent-Based Modeling for Military Analysis: Ukraine Applications, 2025.
4. Kiel Institute, Economic Sustainability Modeling: Russia and Ukraine, 2024.
5. NATO JAPCC, Simulation Methods for Conflict Analysis, Kalkar, 2025.