A Computational Pipeline for Modeling and Predicting Wildﬁre Behavior

Nuno Fachada

Lus

ofona University, COPELABS, Campo Grande, 376, Lisboa, Portugal

Keywords:

Agent-based Modeling, High-performance Computing, Computational Intelligence, Veriﬁcation and

Validation, Wildﬁres.

Abstract:

Wildﬁres constitute a major socioeconomic burden. While a number of scientiﬁc and technological methods

have been used for predicting and mitigating wildﬁres, this is still an open problem. In turn, agent-based

modeling is a modeling approach where each entity of the system being modeled is represented as an inde-

pendent decision-making agent. It is a useful technique for studying systems that can be modeled in terms

of interactions between individual components. Consequently, it is an interesting methodology for modeling

wildﬁre behavior. In this position paper, we propose a complete computational pipeline for modeling and pre-

dicting wildﬁre behavior by leveraging agent-based modeling, among other techniques. This project is to be

developed in collaboration with scientiﬁc and civil stakeholders, and should produce an open decision support

system easily extendable by stakeholders and other interested parties.

1 INTRODUCTION

Wildﬁres represent a major socioeconomic burden in

affected regions, often resulting in widespread devas-

tation. This is particularly true in the case of Por-

tugal, which has the highest density of wildﬁre ig-

nitions among southern European countries (Catry

et al., 2010). While several scientiﬁc and techno-

logical approaches have been used for understanding,

predicting and mitigating the problem (Collins et al.,

2013), recent tragic wildﬁres show this is very much

an open problem (G

omez-Gonz

alez et al., 2018).

Agent-based modeling (ABM) is a modeling ap-

proach where each entity of the system being modeled

is represented as an independent decision-making

agent. It is a useful technique for exploring systems

that can be modeled in terms of interactions between

individual components (Fachada et al., 2015), and is

thus a good ﬁt for modeling the behavior of wildﬁres

(Niazi et al., 2010).

This position paper proposes a scientiﬁc project

to study wildﬁre propagation by developing a com-

plete agent-based modeling and simulation pipeline.

It aims to advance the state of the art by simultane-

ously offering the following novel speciﬁcations: i)

simulations should be able to run on commodity hard-

ware; ii) models should be retrainable with new data

sources, as these become available; iii) models should

https://orcid.org/0000-0002-8487-5837

work robustly with missing data; and, iv) the pipeline

should be a template for others to build upon.

The project is to be developed in collaboration

with several stakeholders, namely scientiﬁc and civil

authorities. Its main output will be a wildﬁre mod-

eling and simulation pipeline with two applications in

mind: i) act as a decision support system for civil pro-

tection stakeholders; and, ii) rather than a ﬁnal prod-

uct, serve as a platform to be extended by interested

parties.

2 BACKGROUND

Agent-based modeling (ABM) is a bottom-up model-

ing approach where each entity of the system being

modeled is represented as an independent decision-

making agent. It is a useful technique for simulating

and exploring systems that can be modeled in terms of

interactions between individual components, such as

cell cultures, military units in a battleﬁeld or epidemi-

ology scenarios (Fachada et al., 2015). ABM shares

characteristics with, and adds capabilities to, discrete-

event simulation, system dynamics and Monte Carlo

methods (Macal, 2016), and in a broad sense, gen-

eralizes techniques such as cellular automata (CA)

and network models (Trucchia et al., 2020). ABM is

thus a good ﬁt for modeling the behavior of wildﬁres

(Yassemi et al., 2008; Widyastuti et al., 2020). Al-

Fachada, N.

A Computational Pipeline for Modeling and Predicting Wildﬁre Behavior.

DOI: 10.5220/0011073900003197

In Proceedings of the 7th International Conference on Complexity, Future Information Systems and Risk (COMPLEXIS 2022), pages 79-84

ISBN: 978-989-758-565-4; ISSN: 2184-5034

though a few modeling approaches have been used to

characterize Portugal’s wildﬁres (Catry et al., 2010;

Collins et al., 2013), the use of ABM is relatively

uncommon given the socioeconomic impacts of the

problem (G

omez-Gonz

alez et al., 2018).

Large scale emergent behavior in ABMs is popu-

lation sensitive, and is thus desirable that the number

of agents in a simulation is able to reﬂect the reality

of the system being modeled (Fachada et al., 2017c).

Additionally, stochastic models in general, and ABMs

in particular, usually require the exploration of large

parameter spaces (Hill et al., 2013). Consequently,

simulating realistic models can be computationally

demanding (Fachada and Rosa, 2017), often requir-

ing parallel and/or distributed implementation ap-

proaches in order to leverage the high core count in

modern processors (Xiao et al., 2019). This has in-

deed been the case for a number of wildﬁre ABMs

(Smith et al., 2016).

While parallelizing ABMs is in itself a complex

task, other difﬁculties exist. ABMs are very sensi-

tive to implementation details, and the impact that

seemingly unimportant aspects such as data struc-

tures, algorithms or order of events can have on results

is tremendous (Wilensky and Rand, 2007; Fachada

et al., 2017b). By deﬁnition, parallelization requires

changes in many of these aspects. Furthermore,

the partitioning of pseudo-random number generator

(PRNG) streams in parallel simulations may bring

problems such as hidden correlations or overlaps in

different substreams, potentially invalidating simula-

tion results (Hill et al., 2013).

In any case, veriﬁcation and validation (V&V) is

crucial for ABMs used in making real-world deci-

sions (David et al., 2017). This is especially the case

for complex dynamical systems in general (Williams,

2018), and wildﬁres in particular (Niazi et al., 2010;

Achtemeier, 2013). Knowledge gaps are often un-

covered by the V&V iterative process (David et al.,

2017; Fachada et al., 2017b; Fachada et al., 2016),

notably when integrating external data (e.g., satellite

data (Cardil et al., 2019; Han et al., 2021)) in the

model—a requirement for making accurate ﬁre prop-

agation forecasts. In such cases, computational in-

telligence techniques can be used to ﬁll such gaps

(Pathak et al., 2018; Fernandes et al., 2019; Fernandes

et al., 2020) and aid in data integration (Ntinas et al.,

2017; Denham et al., 2012).

Here, we propose a scientiﬁc project to study

wildﬁre propagation by developing a computational

simulation pipeline which includes, but is not to lim-

ited to, parallel high-performance ABMs, V&V tools,

machine learning (ML) techniques and data fusion

methods, with the goal of successful V&V of mod-

els against real world wildﬁre propagation scenarios

in Portugal.

3 RESEARCH PLAN AND

METHODS

In this project, we propose to develop a complete

wildﬁre simulation pipeline, with the aim of address-

ing the scientiﬁc question of whether it is possible

to accurately and efﬁciently predict the propagation

of such ﬁres using computational modeling and com-

putational intelligence methods. The project has the

following novel speciﬁcations: i) model simulation

and visualization should be able to run on commod-

ity hardware; ii) models should be easily retrainable

with new data sources, as these become available; iii)

models should work robustly in the absence of one

or more data streams (i.e., the best possible forecast

should be generated given the available data); and, iv)

the pipeline should be easily extendable by third par-

ties, i.e., the project aims to provide a template for

others to build upon.

Collaborations are essential for a project of this

type. Concerning basic research, collaboration will

be sought with IPMA (Portuguese Institute for the

Sea and Atmosphere) and possibly NOAA (National

Oceanic and Atmospheric Administration). For ap-

plied research and ﬁeld tests, cooperation with the

ANPC (Portuguese Civil Protection Authority) is crit-

ical.

The project is divided into the following objec-

tives/tasks:

1. Parallel ABM for conjectural wildﬁre simulations

(18 months). Develop a parallel ABM framework

aimed at wildﬁre simulation with multiple lay-

ers, namely, but not limited to, CA, mobile agents

and network/graph layers. Framework should be

highly performant and capable of incorporating

different types of data (e.g., GIS for geographic

features, GPS for location of wildﬁre response

means, drone video ﬁre tracking (Costa et al.,

2021)), though not able to directly account for

external data sources. Tentative models should

broadly predict ﬁre behavior given initial condi-

tions in some scenarios, allowing users to pose

“what-if” questions, but are not ﬁt for actual de-

cision making. The following questions will have

to be answered:

• What is the best shared-memory architecture to

run simulations on these models? Given highly

heterogeneous model components, are GPUs

a feasible target architecture, or do multicore

COMPLEXIS 2022 - 7th International Conference on Complexity, Future Information Systems and Risk

CPUs offer a more viable prospect in terms

of man–hour/performance balance? Will it be

necessary to use a distributed memory architec-

ture (e.g., multiple computers, supercomputers)

to run these models?

• Given the different types of data to be ac-

counted for, how and where should these be in-

tegrated in the various model layers?

• Does the massive partition of models (and con-

sequently, of PRNGs), for the purpose of con-

current simulations, skew simulation results

one would obtain with serial executions (i.e.,

with no PRNG partitioning)? If so, what strate-

gies can be employed to minimize the issue?

2. Data integration and ofﬂine model training (18

months). Integrate publicly available data in

the framework, namely from satellites such as

SMAP (moisture), Suomi NPP (visible and in-

frared imaging), Meteosat (visible and infrared

imaging), Sentinel-2 (Han et al., 2021) or the

various sources available from LSA SAF, such

as land surface temperature, wildﬁre monitoring,

vegetation and land cover. Use ML techniques

to train models using historical data (from the

same sources), focusing on a limited number of

wildﬁre-prone patches of land in Portugal. Mod-

els should be data-driven, retrodictively valid and

robust in their forecasts when not all of the driving

data is available. Questions to be answered:

• It is possible to automate the process of data

fusion? What types of data transformations are

required?

• What are the most appropriate ML methods for

ofﬂine model training? Is it necessary to de-

velop custom ML approaches?

• How is model robustness affected when sup-

pressing different types of driving data?

3. Automatic model optimization (12 months). An

automated V&V procedure is to be implemented.

Models should be tuned using stochastic opti-

mization methods in the presence of new infor-

mation. Framework improvements based on basic

research stakeholders feedback. Models are not

yet expected to be used for decision making in the

ﬁeld. Questions to be answered:

• It is possible to consolidate the V&V process in

the overall modeling and simulation pipeline?

• What kind of stochastic optimization methods

work best for online model tuning?

4. Field trials (12 months). Simulations tested dur-

ing wildﬁres in the modeled land patches in the

presence of project stakeholders (IPMA, ANPC,

others). Further framework improvements based

on stakeholders feedback. The following ques-

tions will have to be answered:

• Can the models accurately and efﬁciently pre-

dict the propagation of wildﬁres? This is essen-

tially the main scientiﬁc question of the project.

• Can the framework be integrated in the stake-

holders workﬂow (e.g., ANPC) as a decision

support system?

• Can stakeholders such as IPMA build upon the

developed framework?

The proposed plan also accommodates 12 months

to explore innovations in the ﬁeld (e.g., new

data sources), late collaborations and documentation

write-up.

4 EXPECTED OUTCOMES

The main output of this project will be a wildﬁre mod-

eling and simulation pipeline with two applications in

mind: i) predict wildﬁre propagation in the modeled

land patches of Portuguese territory, acting as a deci-

sion support system for civil protection stakeholders;

and, ii) rather than a ﬁnal product, serve as a platform

to be extended by all interested parties.

This is a unique opportunity for the author to con-

solidate his academic experience (discussed in detail

in Section 6) into a project with potentially wide so-

cietal and economical beneﬁts. In the ﬁrst 18 months

(task 1 of the project), we plan to publish three journal

articles, namely:

1. “The inﬂuence of pseudo-random number gener-

ators on parallel agent-based simulations”, Com-

puter Physics Communications.

2. “High-performance implementation of a wildﬁre

agent-based model with OpenCL”, IEEE Transac-

tions on Parallel and Distributed Systems.

3. “Data fusion challenges in a multilayered agent-

based model of wildﬁre propagation”, Interna-

tional Journal of Image and Data Fusion.

While it is difﬁcult to predict exact titles of the pa-

pers to be written after this period, we can summarize

some of the potential contributions for tasks 2 and 3:

1. Automating the process of data fusion in wildﬁre

ABM.

2. Proposing a clear V&V framework for wildﬁre

ABMs in the context of a well deﬁned modeling

and simulation pipeline.

3. Effectively use such models for predicting wild-

ﬁre outcomes under different conditions.

A Computational Pipeline for Modeling and Predicting Wildﬁre Behavior

Scientiﬁc contributions for the ﬁnal stages of the

project will greatly depend on forged collaborations

and stakeholder feedback.

5 SPIN-OFF RESEARCH

PROJECTS

This project has considerable potential beyond wild-

ﬁre forecast in Portugal—i.e., it can potentially be

reused for other regions and countries—and indeed

beyond wildﬁre forecast, period. The most obvi-

ous beneﬁt, as already highlighted in the previous

sections, is the potential for advancing fundamental

knowledge in the ﬁelds touched by this project, from

ML and ABM to parallel computing and multivari-

ate data fusion. A second and less obvious beneﬁt

consists of taking the experience gained with wild-

ﬁre modeling and apply it to other important soci-

etal problems such as modeling refugee migration

patterns, metabolic cellular networks and pandemic

outbreaks—such as the one we are currently experi-

encing. In our opinion, this is a process of increasing

returns.

6 DISCUSSION AND

MOTIVATION

The main goal of this project if to develop a complete

wildﬁre modeling and simulation pipeline for the ef-

fective and efﬁcient prediction of ﬁre propagation.

The framework is expected to be usable as decision

support system, for example by the ANPC, and serve

as a template to be built upon by project stakeholders.

As such, it is directly aligned with goal 15 of the UN’s

2030 Agenda for Sustainable Development (United

Nations, 2015), namely “Protect, restore and promote

sustainable use of terrestrial ecosystems, sustainably

manage forests, combat desertiﬁcation, and halt and

reverse land degradation and halt biodiversity loss”.

Concerning the last 10 years, the author has dedi-

cated considerable part this period to basic research in

the ﬁeld of ABM, focusing on high-performance exe-

cution and reliable replication (namely, parallel reim-

plementation) of these simulation models. In par-

ticular, in reference (Fachada et al., 2017c) the au-

thor proposed a number of ABM parallelization tech-

niques, and statistically compared the output of the

parallelized implementations with the original serial

one.

Due to the amount of work involved in properly

setting up a model comparison experiment, the author

also developed and proposed a model-independent

framework for comparing the output of simulation

models (Fachada et al., 2017b), with concrete imple-

mentations for the R (Fachada et al., 2016) and MAT-

LAB/Octave (Fachada and Rosa, 2018) programming

environments. This framework vastly improves the

cumbersome approaches typically used for replicat-

ing and comparing simulation models, namely when

the goal is that of parallelizing serial realizations.

During these investigations, we have come to the

conclusion that there are several questions that still

need answering with respect to faithful and efﬁcient

model parallelization. Of these, we highlight two:

1. Determining if the architecture of GPUs is feasi-

ble for performing large-scale ABM simulations

with highly heterogeneous components. There

are some conﬂicting views on this topic (Coak-

ley et al., 2012), namely the point to which per-

formance gains can be had by developing GPU-

focused ABM implementations.

2. Determining if the partition of PRNGs for the

purpose of parallelizing ABMs skews results one

would obtain with no PRNG partitioning (i.e.,

with serial model implementations).

In any case, the grounds for this proposal go further

back. In the last 15 years the author gained extensive

knowledge and experience as a researcher in many

of the ﬁelds related with the project, namely in com-

plex systems modeling (Fachada et al., 2009; Isidoro

et al., 2011; Fachada et al., 2015), parallel/GPU pro-

gramming (Fachada et al., 2017a), parallel ABM im-

plementations (Fachada et al., 2017c; Fachada and

Rosa, 2017), model replication and V&V (Fachada

et al., 2017b; Fachada et al., 2016; David et al., 2017),

ML (Fachada et al., 2014), stochastic optimization

methods (Fernandes et al., 2019; Fernandes et al.,

2020; Fernandes et al., 2022) and image processing

(Fachada et al., 2012). This proposal offers a unique

opportunity to capitalize on this experience by devel-

oping a project with potentially vast societal and eco-

nomical beneﬁts.

7 CONCLUSIONS

We expect that this project will promote activity in

a number of scientiﬁc areas, while making contribu-

tions to current scientiﬁc, societal and possibly indus-

trial problems, with positive impacts on society and

wealth. Since this is a rapidly evolving ﬁeld, and we

plan to address prominent challenges in the area, it is

our hope that this proposal yields a number of high

impact publications and captures considerable fund-

COMPLEXIS 2022 - 7th International Conference on Complexity, Future Information Systems and Risk

ing. Given the present-day interest and importance of

the topics discussed here, we expect that the proposed

research attracts top computer science and engineer-

ing students. Importantly, the public and private in-

terest in AI, big data, machine learning and model-

ing and simulation, not to speak of forest and wildﬁre

management, justiﬁes adequate funding for research

and development, while also stimulating innovation

and the creation of value in related areas.

ACKNOWLEDGEMENTS

This work is supported by Fundac¸

ao para a Ci

encia

e a Tecnologia under Grant No.: UIDB/04111/2020

(COPELABS)

REFERENCES

Achtemeier, G. L. (2013). Field validation of a free-

agent cellular automata model of ﬁre spread with ﬁre–

atmosphere coupling. International Journal of Wild-

land Fire, 22(2):148–156.

Cardil, A., Monedero, S., Ram

ırez, J., and Silva, C. A.

(2019). Assessing and reinitializing wildland ﬁre sim-

ulations through satellite active ﬁre data. Journal of

Environmental Management, 231:996–1003.

Catry, F. X., Rego, F. C., Bac¸

ao, F. L., and Moreira, F.

(2010). Modeling and mapping wildﬁre ignition risk

in portugal. International Journal of Wildland Fire,

18(8):921–931.

Coakley, S., Gheorghe, M., Holcombe, M., Chin, S., Worth,

D., and Greenough, C. (2012). Exploitation of high

performance computing in the FLAME agent-based

simulation framework. In Proc. of the 14th Int. Conf.

on High Performance Computing and Communica-

tions, pages 538–545, Liverpool, UK. IEEE.

Collins, R. D., de Neufville, R., Claro, J., Oliveira, T., and

Pacheco, A. P. (2013). Forest ﬁre management to

avoid unintended consequences: A case study of Por-

tugal using system dynamics. Journal of Environmen-

tal Management, 130:1–9.

Costa, R. L. C., Miranda, E., Dias, P., and Moreira, J.

(2021). Experience: Quality assessment and improve-

ment on a forest ﬁre dataset. Journal of Data and In-

formation Quality, 13(1):1–13.

David, N., Fachada, N., and Rosa, A. C. (2017). Verifying

and Validating Simulations, pages 173–204. Springer

International Publishing, Cham.

Denham, M., Wendt, K., Bianchini, G., Cort

es, A., and

Margalef, T. (2012). Dynamic data-driven genetic al-

gorithm for forest ﬁre spread prediction. Journal of

Computational Science, 3(5):398–404.

Fachada, N., Figueiredo, M. A. T., Lopes, V. V., Martins,

R. C., and Rosa, A. C. (2014). Spectrometric differ-

entiation of yeast strains using minimum volume in-

crease and minimum direction change clustering cri-

teria. Pattern Recognition Letters, 45:55–61.

Fachada, N., Lopes, V. V., Martins, R. C., and Rosa, A. C.

(2015). Towards a standard model for research in

agent-based modeling and simulation. PeerJ Com-

puter Science, 1:e36.

Fachada, N., Lopes, V. V., Martins, R. C., and Rosa, A. C.

(2017a). cf4ocl: a C framework for OpenCL. Science

of Computer Programming, 143:9–19.

Fachada, N., Lopes, V. V., Martins, R. C., and Rosa, A. C.

(2017b). Model-independent comparison of simula-

tion output. Simulation Modelling Practice and The-

ory, 72:131–149.

Fachada, N., Lopes, V. V., Martins, R. C., and Rosa, A. C.

(2017c). Parallelization strategies for spatial agent-

based models. International Journal of Parallel Pro-

gramming, 45(3):449–481.

Fachada, N., Lopes, V. V., and Rosa, A. (2009). Simulating

antigenic drift and shift in inﬂuenza A. In Proc. of the

2009 ACM symposium on Applied Computing, SAC

’09, pages 2093–2100, New York, NY, USA. ACM.

Fachada, N., Rodrigues, J., Lopes, V. V., Martins, R. C.,

and Rosa, A. C. (2016). micompr: An R package for

multivariate independent comparison of observations.

The R Journal, 8(2):405–420.

Fachada, N. and Rosa, A. (2018). micompm: A MAT-

LAB/Octave toolbox for multivariate independent

comparison of observations. Journal of Open Source

Software, 3:430.

Fachada, N. and Rosa, A. C. (2017). Assessing the feasibil-

ity of OpenCL CPU implementations for agent-based

simulations. In Proceedings of the 5th International

Workshop on OpenCL, IWOCL 2017, pages 4:1–4:10,

New York, NY, USA. ACM.

Fachada, V., Fachada, N., Turpeinen, T., Rahkila, P., Rosa,

A., and Kainulainen, H. (2012). Topocell – an image

analysis tool to study intracellular topography. The

FASEB Journal, 26(1

supplement):578.2–578.2.

Fernandes, C. M., Fachada, N., Laredo, J. L. J., Merelo, J.-

J., and Rosa, A. C. (2020). Population sizing of cellu-

lar evolutionary algorithms. Swarm and Evolutionary

Computation, 58:100721.

Fernandes, C. M., Fachada, N., Merelo, J.-J., and Rosa,

A. C. (2019). Steady state particle swarm. PeerJ Com-

puter Science, 5:e202.

Fernandes, P. M. A., In

acio, P. M. A., Feliciano, H., and

Fachada, N. (2022). SimpAI: Evolutionary heuristics

for the ColorShapeLinks board game competition. In

Barbedo, I., Barroso, B., Leger

en, B., Roque, L., and

Sousa, J. P., editors, Videogame Sciences and Arts, VJ

2020, volume 1531 of Communications in Computer

and Information Science, chapter 9, pages 113–126.

Springer International Publishing, Cham.

omez-Gonz

alez, S., Ojeda, F., and Fernandes, P. M.

(2018). Portugal and chile: Longing for sustainable

forestry while rising from the ashes. Environmental

Science & Policy, 81:104–107.

Han, A., Qing, S., Bao, Y., Na, L., Bao, Y., Liu, X., Zhang,

J., and Wang, C. (2021). Short-term effects of ﬁre

A Computational Pipeline for Modeling and Predicting Wildﬁre Behavior

severity on vegetation based on sentinel-2 satellite

data. Sustainability, 13(1):432.

Hill, D. R. C., Mazel, C., Passerat-Palmbach, J., and Traore,

M. K. (2013). Distribution of random streams for sim-

ulation practitioners. Concurrency and Computation:

Practice and Experience, 25(10):1427–1442.

Isidoro, C., Fachada, N., Barata, F., and Rosa, A. (2011).

Agent-based model of dengue disease transmission by

Aedes aegypti populations. In Advances in Artiﬁcial

Life. Darwin Meets von Neumann, volume 5777 of

Lecture Notes in Computer Science, pages 345–352.

Springer.

Macal, C. (2016). Everything you need to know about

agent-based modelling and simulation. Journal of

Simulation, 10(2):144–156.

Niazi, M. A., Siddique, Q., Hussain, A., and Kolberg, M.

(2010). Veriﬁcation & validation of an agent-based

forest ﬁre simulation model. In Proceedings of the

2010 Spring Simulation Multiconference, SpringSim

’10, pages 1:1–1:8, San Diego, CA, USA. Society for

Computer Simulation International.

Ntinas, V. G., Moutaﬁs, B. E., Trunﬁo, G. A., and Sirak-

oulis, G. C. (2017). Parallel fuzzy cellular automata

for data-driven simulation of wildﬁre spreading. Jour-

nal of Computational Science, 21:469–485.

Pathak, J., Wikner, A., Fussell, R., Chandra, S., Hunt,

B. R., Girvan, M., and Ott, E. (2018). Hybrid fore-

casting of chaotic processes: using machine learn-

ing in conjunction with a knowledge-based model.

Chaos: An Interdisciplinary Journal of Nonlinear Sci-

ence, 28(4):041101.

Smith, J., Barfed, L., Dasclu, S. M., and Harris, F. C.

(2016). Highly parallel implementation of forest ﬁre

propagation models on the GPU. In 2016 Interna-

tional Conference on High Performance Computing

& Simulation (HPCS), pages 917–924. IEEE.

Trucchia, A., D’Andrea, M., Baghino, F., Fiorucci, P., Fer-

raris, L., Negro, D., Gollini, A., and Severino, M.

(2020). Propagator: an operational cellular-automata

based wildﬁre simulator. Fire, 3(3):26.

United Nations (2015). Transforming our world: the 2030

agenda for sustainable development. Department of

Economic and Social Affairs.

Widyastuti, K., Imron, M. A., Pradopo, S. T., Suryatmojo,

H., Sopha, B. M., Spessa, A., and Berger, U. (2020).

Peatﬁre: an agent-based model to simulate ﬁre igni-

tion and spreading in a tropical peatland ecosystem.

International Journal of Wildland Fire.

Wilensky, U. and Rand, W. (2007). Making models match:

replicating an agent-based model. Journal of Artiﬁcial

Societies and Social Simulation, 10(4):2.

Williams, R. A. (2018). Lessons learned on development

and application of agent-based models of complex dy-

namical systems. Simulation Modelling Practice and

Theory, 83:201–212.

Xiao, J., Andelﬁnger, P., Eckhoff, D., Cai, W., and Knoll,

A. (2019). A survey on agent-based simulation us-

ing hardware accelerators. ACM Computing Surveys,

51(6):131:1–131:35.

Yassemi, S., Dragi

cevi

c, S., and Schmidt, M. (2008). De-

sign and implementation of an integrated GIS-based

cellular automata model to characterize forest ﬁre be-

haviour. Ecological Modelling, 210(1):71–84.

COMPLEXIS 2022 - 7th International Conference on Complexity, Future Information Systems and Risk