arxiv: 2604.21067 · v1 · submitted 2026-04-22 · 📊 stat.AP

Recognition: unknown

The geometry of conflict : 3D Spatio-temporal patterns in fatalities prediction

Thomas Schincariol

Pith reviewed 2026-05-09 22:11 UTC · model grok-4.3

classification 📊 stat.AP

keywords conflict predictionspatio-temporal patternsEarth Mover's Distancefatality forecastingpattern matching3D analysisearly warning systems

0 comments

The pith

Recognizing 3D spatio-temporal patterns in conflict fatalities improves prediction accuracy over benchmark models.

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

The paper transforms conflict fatality data into three-dimensional patterns at the Prio-Grid level. It adapts the ShapeFinder model and applies the Earth Mover's Distance algorithm to detect and match similar historical patterns across space and time. These matches are then used to generate predictions for future fatalities. The results show better accuracy than the VIEWS ensemble model, suggesting that pattern-based approaches can enhance understanding of how conflicts spread.

Core claim

By converting spatio-temporal conflict fatality data into three-dimensional patterns and using the Earth Mover's Distance to compare and match these patterns with historical ones via an adapted ShapeFinder model, the study generates fatality predictions that outperform the VIEWS ensemble benchmark.

What carries the argument

The adapted ShapeFinder model with Earth Mover's Distance (EMD) for matching 3D conflict diffusion patterns, which measures the minimal cost to transform one pattern into another to identify similarities for prediction.

If this is right

Early warning systems for violence can incorporate pattern matching to issue more timely alerts.
Analysis of conflict dynamics gains a tool for classifying diffusion shapes in three dimensions.
Predictive models in statistics can benefit from shape-based similarity measures beyond traditional time-series methods.
The framework allows for adaptive capacity in handling complex, dynamic trends in violence data.

Where Pith is reading between the lines

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

Similar pattern recognition techniques could be tested on other spatio-temporal phenomena such as epidemic spread or migration flows.
Future work might explore how changes in conflict drivers affect the stability of these 3D patterns over time.
Combining this method with additional data sources could refine the pattern matches further.

Load-bearing premise

The assumption that past 3D conflict patterns will reliably repeat in the future without significant shifts in conflict drivers or data reporting.

What would settle it

Applying the predictions to a new period of recent conflict data and checking if accuracy gains over the benchmark disappear or reverse.

Figures

Figures reproduced from arXiv: 2604.21067 by Thomas Schincariol.

**Figure 2.** Figure 2: 3D sequences that happened at the border of Kenya and Somalia in 2022 (left section) [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗

**Figure 3.** Figure 3: Three patterns on a 5×5 grid over three months. The x-axis represents longitude, and the y-axis represents latitude. The cell color indicates the number of fatalities per month, ranging from 0 (white) to 9 (dark red), with darker shades representing higher values. Incorporating flexibility into models is essential for matching cases with similar dynamics despite differences in speed, scale, or frequency. … view at source ↗

**Figure 4.** Figure 4: Four steps of the identification of active conflict zone patterns in 2022 in Africa and the [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗

**Figure 5.** Figure 5: Data transformation process illustrated through three steps. (1-left) Active zones are iden [PITH_FULL_IMAGE:figures/full_fig_p016_5.png] view at source ↗

**Figure 6.** Figure 6: Diagram of the overall method of the model. First, we extracted the 3D sequence of interest [PITH_FULL_IMAGE:figures/full_fig_p018_6.png] view at source ↗

**Figure 7.** Figure 7: Rolling Window Scheme for historical search. The red cube represents the input sequence, [PITH_FULL_IMAGE:figures/full_fig_p021_7.png] view at source ↗

**Figure 8.** Figure 8: Example of scenario creation using clustering. The 3D sequence uses longitude on the [PITH_FULL_IMAGE:figures/full_fig_p022_8.png] view at source ↗

**Figure 9.** Figure 9: Outcome of the ShapeFinder model for the Kenya-Somalia example, discussed in Figure [PITH_FULL_IMAGE:figures/full_fig_p026_9.png] view at source ↗

**Figure 10.** Figure 10: The mean and 95% confidence interval for sequences’ similarity metric [PITH_FULL_IMAGE:figures/full_fig_p028_10.png] view at source ↗

**Figure 11.** Figure 11: Log ratio of Mean Squared Error (MSE) between the Views model and the ShapeFinder (SF) [PITH_FULL_IMAGE:figures/full_fig_p029_11.png] view at source ↗

**Figure 12.** Figure 12: Scatter plot of the log ratio of Raw Error (in x) and EMD (in y) at the zone level, as intro [PITH_FULL_IMAGE:figures/full_fig_p030_12.png] view at source ↗

**Figure 13.** Figure 13: Scatter plot of the ratio of increase Rinc (in x) and Mean Average Percentage Error (MAPE) of Views (in y) at the active zone level, with a log-modulus transformation. The color of the point represents the difference in performance between the two models. The darker red, the SF performs better, and the darker blue, the Views model. MAPE values). These are the hardest patterns to forecast, as they involve … view at source ↗

**Figure 14.** Figure 14: Scatter plots showing the relationship between input sequence characteristics and prediction [PITH_FULL_IMAGE:figures/full_fig_p033_14.png] view at source ↗

read the original abstract

Understanding how conflict events spread over time and space is crucial for predicting and mitigating future violence. However, progress in this area has been limited by the lack of methods capable of capturing the intricate, dynamic patterns of conflict diffusion. The complex nature of those trends needs flexibility in the models to untangle them. This study addresses this gap by analyzing spatio-temporal conflict fatality data using an innovative approach that transforms the data into three-dimensional patterns at the Prio-Grid level. In this paper, a shape-based model called ShapeFinder is adapted. By applying the Earth Movers Distance (EMD) algorithm, we detect and classify these patterns, allowing us to compare and match patterns with high adaptive capacity in all dimensions. Using historical similar patterns, we generate predictions of conflict fatalities and compare these with forecasts from the Views ensemble model, a leading benchmark. Our findings demonstrate that recognizing and analyzing conflict diffusion patterns significantly improves predictive accuracy, outperforming the benchmark model. This research contributes to the study of conflict dynamics by introducing a novel pattern recognition framework that enhances the analysis of spatio-temporal data and offers practical applications for early warning systems.

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 adapts EMD-based 3D pattern matching to PRIO-grid fatality data for forecasting but supplies no results, no prediction formula, and no leakage checks, so the outperformance claim cannot be evaluated yet.

read the letter

The core move is taking 3D spatio-temporal fatality patterns on the PRIO grid, running ShapeFinder with Earth Mover's Distance to find historical matches, and using those matches to predict future fatalities. The abstract claims this beats the VIEWS ensemble, which is the kind of practical claim that could matter for early-warning work if it holds up. The adaptation itself is new in this exact combination, and the motivation to treat conflict diffusion as shape matching rather than just counts or simple lags is reasonable for the domain. That part deserves credit as a concrete attempt to bring flexible pattern tools into applied conflict studies. The rest is thin. No tables, no error metrics, no cross-validation details, and no description of how a matched historical window actually becomes a numeric forecast for the next period. The stress-test note is right on this: without that mapping spelled out, you cannot rule out that the lookup pulls in information from the target window or simply averages the same data used to build the pattern library. The paper also skips any head-to-head with other existing spatio-temporal conflict models that already use distance or shape ideas, so the novelty claim stays untested. This is the sort of work that could interest people building operational conflict early-warning systems who are open to non-parametric pattern approaches. A reader already familiar with PRIO-grid data and VIEWS would get the most out of it, but only once the methods and results sections are filled in. It is coherent enough on its own terms to go to a serious referee rather than a desk reject, provided the authors add the missing quantitative comparisons and confirm the forecast step uses strictly prior data only. I would send it for review after those additions.

Referee Report

3 major / 1 minor

Summary. This paper introduces a ShapeFinder model using Earth Mover's Distance (EMD) to identify and match 3D spatio-temporal patterns in PRIO-GRID conflict fatality data. It asserts that predictions derived from similar historical patterns outperform those from the VIEWS ensemble model, offering a new framework for analyzing conflict diffusion.

Significance. Should the outperformance hold under rigorous validation, the work could advance conflict prediction by leveraging geometric pattern matching for spatio-temporal data, with potential applications in early warning systems. The emphasis on adaptive 3D patterns addresses a noted gap in capturing dynamic conflict trends.

major comments (3)

The mapping from EMD-matched patterns to specific fatality predictions is underspecified; it is unclear whether this involves averaging fatalities from matched historical periods, weighted sums, or an additional model, which is essential for reproducing and validating the claimed improvements.
The abstract claims superior predictive accuracy over VIEWS without providing any quantitative results, such as error metrics, confidence intervals, or details on cross-validation and pattern selection criteria, preventing assessment of robustness.
No details confirm that pattern matching for out-of-sample forecasts uses exclusively pre-forecast data, which is necessary to rule out leakage and ensure the method does not inadvertently incorporate target-period information.

minor comments (1)

The phrase 'high adaptive capacity in all dimensions' is used without definition or explanation of its computation.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive and detailed comments. These have highlighted areas where the manuscript can be clarified and strengthened. We address each major comment point by point below, with revisions planned for the next version of the manuscript.

read point-by-point responses

Referee: The mapping from EMD-matched patterns to specific fatality predictions is underspecified; it is unclear whether this involves averaging fatalities from matched historical periods, weighted sums, or an additional model, which is essential for reproducing and validating the claimed improvements.

Authors: We agree that the prediction step requires explicit specification for reproducibility. The ShapeFinder approach computes the fatality forecast for each grid cell and future period as a distance-weighted average of observed fatalities from the k nearest historical 3D patterns, with weights inversely proportional to EMD. We will revise Section 3.2 to include this formula, the chosen value of k, and pseudocode. revision: yes
Referee: The abstract claims superior predictive accuracy over VIEWS without providing any quantitative results, such as error metrics, confidence intervals, or details on cross-validation and pattern selection criteria, preventing assessment of robustness.

Authors: The body of the manuscript reports MAE and RMSE improvements under rolling-origin cross-validation, but we accept that the abstract omits these details. We will revise the abstract to include the key quantitative metrics and a brief statement on the validation procedure and pattern-selection threshold. revision: yes
Referee: No details confirm that pattern matching for out-of-sample forecasts uses exclusively pre-forecast data, which is necessary to rule out leakage and ensure the method does not inadvertently incorporate target-period information.

Authors: All pattern matching for out-of-sample forecasts is performed using only data strictly prior to the forecast origin; the historical database is frozen at t-1 when predicting t. We will add an explicit statement and a supplementary timeline diagram in the methods section to document this temporal separation. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper's chain transforms PRIO-GRID fatality data into 3D spatio-temporal patterns, applies EMD for historical similarity matching, and derives forecasts from those matches before benchmarking against the external VIEWS ensemble. This constitutes a standard similarity-based forecasting procedure whose outputs are not equivalent to its inputs by construction, nor does any step reduce to a self-citation, fitted parameter renamed as prediction, or ansatz smuggled via prior work. The comparison to an independent benchmark supplies external validation, and no equations or procedural descriptions in the manuscript exhibit the specific reductions required for a circularity finding.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no explicit free parameters, axioms, or invented entities can be extracted. The method implicitly assumes that EMD distances on 3D grids capture diffusion dynamics without additional modeling of political or economic covariates.

pith-pipeline@v0.9.0 · 5489 in / 1183 out tokens · 29814 ms · 2026-05-09T22:11:07.473878+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

55 extracted references · 3 canonical work pages

[1]

2025 , note=

ACLED Conflict Alert System (CAST) , author =. 2025 , note=

2025
[2]

arXiv preprint arXiv:2410.14507 , year=

Bin-Conditional Conformal Prediction of Fatalities from Armed Conflict , author=. arXiv preprint arXiv:2410.14507 , year=

work page arXiv
[3]

Journal of Conflict Resolution , volume=

On the accuracy of media-based conflict event data , author=. Journal of Conflict Resolution , volume=. 2015 , publisher=

2015
[4]

International Interactions , pages=

A shape-based approach to conflict forecasting , author=. International Interactions , pages=. 2021 , number = 48, issue=

2021
[5]

2024 , publisher=

Predicting uncertainty in stages: Using a semiparametric hierarchical hurdle model for predicting distributions of conflict fatalities , author=. 2024 , publisher=

2024
[6]

Data & Policy , volume=

Introducing a global dataset on conflict forecasts and news topics , author=. Data & Policy , volume=. 2024 , publisher=

2024
[7]

Information Fusion , volume=

Tabular data: Deep learning is not all you need , author=. Information Fusion , volume=. 2022 , publisher=

2022
[8]

International Journal of Advanced Computer Science and Applications , volume=

Is deep learning on tabular data enough? An assessment , author=. International Journal of Advanced Computer Science and Applications , volume=. 2022 , publisher=

2022
[9]

Journal of Peace Research , pages=

The 2023/24 VIEWS Prediction challenge: Predicting the number of fatalities in armed conflict, with uncertainty , author=. Journal of Peace Research , pages=. 2024 , publisher=

2023
[10]

Proceedings of the National Academy of Sciences , volume=

Point process modelling of the Afghan War Diary , author=. Proceedings of the National Academy of Sciences , volume=. 2012 , publisher=

2012
[11]

American Political Science Review , volume=

Stadl up! the spatiotemporal autoregressive distributed lag model for tscs data analysis , author=. American Political Science Review , volume=. 2023 , publisher=

2023
[12]

Journal of the Royal Statistical Society Series C: Applied Statistics , volume=

A fractional Hawkes process model for earthquake aftershock sequences , author=. Journal of the Royal Statistical Society Series C: Applied Statistics , volume=. 2024 , publisher=

2024
[13]

Journal of the Royal Statistical Society Series C: Applied Statistics , volume=

Self-exciting point processes with spatial covariates: modelling the dynamics of crime , author=. Journal of the Royal Statistical Society Series C: Applied Statistics , volume=. 2018 , publisher=

2018
[14]

Spatio-temporal graph convolutional networks: A deep learning framework for traffic forecasting.arXiv preprint arXiv:1709.04875,

Spatio-temporal graph convolutional networks: A deep learning framework for traffic forecasting , author=. arXiv preprint arXiv:1709.04875 , year=

work page arXiv
[15]

Nature Reviews Methods Primers , volume=

Graph neural networks , author=. Nature Reviews Methods Primers , volume=. 2024 , publisher=

2024
[16]

Statistical Science , volume=

A review of self-exciting spatio-temporal point processes and their applications , author=. Statistical Science , volume=. 2018 , publisher=

2018
[17]

Biometrika , volume=

Spectra of some self-exciting and mutually exciting point processes , author=. Biometrika , volume=. 1971 , publisher=

1971
[18]

Journal of the Royal Statistical Society Series A: Statistics in Society , pages=

Capturing the spatio-temporal diffusion effects of armed conflict: A nonparametric smoothing approach , author=. Journal of the Royal Statistical Society Series A: Statistics in Society , pages=. 2025 , publisher=

2025
[19]

arXiv preprint arXiv:2506.14817 , year=

Next-Generation Conflict Forecasting: Unleashing Predictive Patterns through Spatiotemporal Learning , author=. arXiv preprint arXiv:2506.14817 , year=

work page arXiv
[20]

Journal of Peace Research , pages=

Accounting for variability in conflict dynamics: A pattern-based predictive model , author=. Journal of Peace Research , pages=. 2025 , publisher=

2025
[21]

International Interactions , volume=

Race to the bottom: Spatial aggregation and event data , author=. International Interactions , volume=. 2022 , publisher=

2022
[22]

2024 , publisher=

Taking time seriously: Predicting conflict fatalities using temporal fusion transformers , author=. 2024 , publisher=

2024
[23]

International Interactions , volume=

High resolution conflict forecasting with spatial convolutions and long short-term memory , author=. International Interactions , volume=. 2022 , publisher=

2022
[24]

International Studies Quarterly , volume=

Civil war contagion and neighboring interventions , author=. International Studies Quarterly , volume=. 2010 , publisher=

2010
[25]

Journal of Peace Research , volume=

When have violent civil conflicts spread? Introducing a dataset of substate conflict contagion , author=. Journal of Peace Research , volume=. 2013 , publisher=

2013
[26]

Science , volume=

Predicting armed conflict: Time to adjust our expectations? , author=. Science , volume=. 2017 , publisher=

2017
[27]

Computational Conflict Research , pages=

Analysis of Conflict Diffusion Over Continuous Space , author=. Computational Conflict Research , pages=. 2020 , publisher=

2020
[28]

Political Geography , volume=

Diffusion patterns of violence in civil wars , author=. Political Geography , volume=. 2011 , publisher=

2011
[29]

Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages=

Urbangpt: Spatio-temporal large language models , author=. Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages=
[30]

Conflict Management and Peace Science , volume=

All conflict is local: Modeling sub-national variation in civil conflict risk , author=. Conflict Management and Peace Science , volume=. 2011 , publisher=

2011
[31]

Proceedings of the 1st ACM SIGSPATIAL International Workshop on Geospatial Anomaly Detection , pages=

Large language models for spatial trajectory patterns mining , author=. Proceedings of the 1st ACM SIGSPATIAL International Workshop on Geospatial Anomaly Detection , pages=
[32]

2024 , publisher=

Temporal Patterns in Conflict Prediction: An improved Shape-Based Approach , author=. 2024 , publisher=

2024
[33]

, author=

Earth mover distance over high-dimensional spaces. , author=. SODA , volume=
[34]

Weather, climate, and society , volume=

Water, drought, climate change, and conflict in Syria , author=. Weather, climate, and society , volume=. 2014 , publisher=

2014
[35]

2023 , publisher=

Leveling crowds: Ethnonationalist conflicts and collective violence in South Asia , author=. 2023 , publisher=

2023
[36]

Behavioral Ecology and Sociobiology , volume=

Using time-series similarity measures to compare animal movement trajectories in ecology , author=. Behavioral Ecology and Sociobiology , volume=. 2019 , publisher=

2019
[37]

Journal of Peace Research , volume=

PRIO-GRID: A unified spatial data structure , author=. Journal of Peace Research , volume=. 2012 , publisher=

2012
[38]

International Interactions , volume=

Predicting political violence using a state-space model , author=. International Interactions , volume=. 2022 , publisher=

2022
[39]

International Interactions , volume=

United they stand: Findings from an escalation prediction competition , author=. International Interactions , volume=. 2022 , publisher=

2022
[40]

Journal of peace research , volume=

Networked international politics: Complex interdependence and the diffusion of conflict and peace , author=. Journal of peace research , volume=. 2016 , publisher=

2016
[41]

2024 , institution=

Integrating Spatio-temporal Diffusion into Statistical Forecasting Models of Armed Conflict via Non-parametric Smoothing , author=. 2024 , institution=

2024
[42]

Review of Economics and Statistics , volume=

The promise and pitfalls of conflict prediction: evidence from Colombia and Indonesia , author=. Review of Economics and Statistics , volume=. 2022 , publisher=

2022
[43]

International Interactions , volume=

The role of governmental weapons procurements in forecasting monthly fatalities in intrastate conflicts: A semiparametric hierarchical hurdle model , author=. International Interactions , volume=. 2022 , publisher=

2022
[44]

Journal of peace research , volume=

ViEWS: A political violence early-warning system , author=. Journal of peace research , volume=. 2019 , publisher=

2019
[45]

International Journal of Forecasting , volume=

Conflict forecasting using remote sensing data: An application to the Syrian civil war , author=. International Journal of Forecasting , volume=. 2024 , publisher=

2024
[46]

Conflict Management and Peace Science , volume=

Bad neighbors: Failed states and their consequences , author=. Conflict Management and Peace Science , volume=. 2008 , publisher=

2008
[47]

Journal of Conflict Resolution , volume=

Firewall? Or wall on fire? A unified framework of conflict contagion and the role of ethnic exclusion , author=. Journal of Conflict Resolution , volume=. 2017 , publisher=

2017
[48]

International studies quarterly , volume=

Contagion or confusion? Why conflicts cluster in space , author=. International studies quarterly , volume=. 2008 , publisher=

2008
[49]

2024 , note=

Echoes of Conflict : Quantifying Redundancy in Fatality Time Series , author=. 2024 , note=

2024
[50]

Journal of Peace Research , volume=

Organized violence 1989--2021 and drone warfare , author=. Journal of Peace Research , volume=. 2022 , publisher=

1989
[51]

Journal of Peace Research , volume=

Introducing the UCDP georeferenced event dataset , author=. Journal of Peace Research , volume=. 2013 , publisher=

2013
[52]

Forecasting fatalities in armed conflict:

Hegre, Håvard and Lindqvist-McGowan, Angelica and Dale, James and Croicu, Mihai and Randahl, David and Vesco, Paola , year =. Forecasting fatalities in armed conflict:
[53]

International Studies Review , volume=

Diffusion in the study of civil wars: A cautionary tale , author=. International Studies Review , volume=. 2014 , publisher=

2014
[54]

International Interactions , volume=

When conflict spreads: Arab spring and the limits of diffusion , author=. International Interactions , volume=. 2012 , publisher=

2012
[55]

Journal of Peace Research , volume=

The geo-temporal evolution of violence in civil conflicts: A micro analysis of conflict diffusion on a new event data set , author=. Journal of Peace Research , volume=. 2021 , publisher=

2021