updating paper

paper/paper.bib

@@ -9,6 +9,101 @@ @misc{towers_gymnasium_2023
     doi = {10.5281/zenodo.8127026},
 }
 
+@misc{tapley2023reinforcementlearningwildfiremitigation,
+      title={Reinforcement Learning for Wildfire Mitigation in Simulated Disaster Environments}, 
+      author={Alexander Tapley and Marissa Dotter and Michael Doyle and Aidan Fennelly and Dhanuj Gandikota and Savanna Smith and Michael Threet and Tim Welsh},
+      year={2023},
+      eprint={2311.15925},
+      archivePrefix={arXiv},
+      primaryClass={cs.LG},
+      url={https://arxiv.org/abs/2311.15925}, 
+}
+
+@article{KULIGOWSKI2021103129,
+title = {Evacuation decision-making and behavior in wildfires: Past research, current challenges and a future research agenda},
+journal = {Fire Safety Journal},
+volume = {120},
+pages = {103129},
+year = {2021},
+note = {Fire Safety Science: Proceedings of the 13th International Symposium},
+issn = {0379-7112},
+doi = {https://doi.org/10.1016/j.firesaf.2020.103129},
+url = {https://www.sciencedirect.com/science/article/pii/S0379711220302204},
+author = {Erica Kuligowski},
+keywords = {Wildfires, Human behavior, Evacuation, Modeling, Bushfires, WUI fires},
+abstract = {Wildfires are becoming more common around the world, and households are frequently advised to evacuate when these fires threaten nearby communities. Effective evacuation requires an understanding of human behavior in wildfires, which is an area that needs further exploration. The purpose of this article is to present current research performed and data collected on evacuation decision-making and behavior during wildland-urban interface (WUI) fires, identify gaps in the research, and develop a future research plan for further data collection of important WUI fire evacuation topics. Research in this area can support developments of evacuation simulation models, and improvements in education programs, planning, decision-making, and design requirements for community-wide WUI fire evacuation.}
+}
+
+@article{WANG201686,
+title = {An agent-based model of a multimodal near-field tsunami evacuation: Decision-making and life safety},
+journal = {Transportation Research Part C: Emerging Technologies},
+volume = {64},
+pages = {86-100},
+year = {2016},
+issn = {0968-090X},
+doi = {https://doi.org/10.1016/j.trc.2015.11.010},
+url = {https://www.sciencedirect.com/science/article/pii/S0968090X15004106},
+author = {Haizhong Wang and Alireza Mostafizi and Lori A. Cramer and Dan Cox and Hyoungsu Park},
+keywords = {Agent-based modeling, Tsunami evacuation, Multimodal, Decision-making, Life safety},
+abstract = {This paper presents a multimodal evacuation simulation for a near-field tsunami through an agent-based modeling framework in Netlogo. The goals of this paper are to investigate (1) how the varying decisn time impacts the mortality rate, (2) how the choice of different modes of transportation (i.e., walking and automobile), and (3) how existence of vertical evacuation gates impacts the estimation of casualties. Using the city of Seaside, Oregon as a case study site, different individual decision-making time scales are included in the model to assess the mortality rate due to immediate evacuation right after initial earthquake or after a specified milling time. The results show that (1) the decision-making time (τ) and the variations in decision time (σ) are strongly correlated with the mortality rate; (2) the provision of vertical evacuation structures is effective to reduce the mortality rate; (3) the mortality rate is sensitive to the variations in walking speed of the evacuee population; and (4) the higher percentage of automobile use in tsunami evacuation, the higher the mortality rate. Following the results, this paper concludes with a description of the challenges ahead in agent-based tsunami evacuation modeling and simulation, and the modeling of complex interactions between agents (i.e., pedestrian and car interactions) that would arise for a multi-hazard scenario for the Cascadia Subduction Zone.}
+}
+
+@article{BELOGLAZOV2016144,
+title = {Simulation of wildfire evacuation with dynamic factors and model composition},
+journal = {Simulation Modelling Practice and Theory},
+volume = {60},
+pages = {144-159},
+year = {2016},
+issn = {1569-190X},
+doi = {https://doi.org/10.1016/j.simpat.2015.10.002},
+url = {https://www.sciencedirect.com/science/article/pii/S1569190X15001483},
+author = {Anton Beloglazov and Mahathir Almashor and Ermyas Abebe and Jan Richter and Kent Charles Barton Steer},
+keywords = {Wildfire, Evacuation planning, Dynamic factors, Model composition, Behaviour and risk modelling},
+abstract = {Wildfires cause devastation on communities, most significantly loss of life. The safety of at-risk populations depends on accurate risk assessment and emergency planning. Evacuation modelling and simulation systems are essential tools for such planning and decision making. During a wildfire evacuation, the behaviour of people is a key factor; what people do, and when they do it, depends heavily on the spatio-temporal distribution of events in a scenario. In this paper, we introduce an approach that enables the behaviour of people and the timing of events to be explicitly modelled through what we term dynamic factors. Our approach composes several simulation and modelling systems, including a wildfire simulator, behaviour modeller, and microscopic traffic simulator, to compute detailed projections of how scenarios unfold. The level of detail provided by our modelling approach enables the definition of a new risk metric, the exposure count, which directly quantifies the threat to a population. Experiments for a wildfire-prone region in Victoria, Australia, resulted in statistically significant differences in clearance times and exposure counts when comparing our modelling approach to an approach that does not account for dynamic factors. The approach has been implemented in a high performance and scalable system – the architecture of which is discussed – that allows multiple concurrent scenarios to be simulated in timeframes suitable for both planning and response use cases.}
+}
+
+@article{doi:10.1061/JTEPBS.0000221,
+author = {Paolo Intini  and Enrico Ronchi  and Steven Gwynne  and Adam Pel },
+title = {Traffic Modeling for Wildland–Urban Interface Fire Evacuation},
+journal = {Journal of Transportation Engineering, Part A: Systems},
+volume = {145},
+number = {3},
+pages = {04019002},
+year = {2019},
+doi = {10.1061/JTEPBS.0000221},
+
+URL = {https://ascelibrary.org/doi/abs/10.1061/JTEPBS.0000221},
+eprint = {https://ascelibrary.org/doi/pdf/10.1061/JTEPBS.0000221}
+,
+    abstract = { Several traffic modeling tools are currently available for evacuation planning and real-time decision support during emergencies. This paper reviews potential traffic-modeling approaches in the context of wildland–urban interface (WUI) fire-evacuation applications. Existing modeling approaches and features are evaluated pertaining to fire-related, spatial, and demographic factors; intended application (planning or decision support); and temporal issues. This systematic review shows the importance of the following modeling approaches: dynamic modeling structures, considering behavioral variability and route choice; activity-based models for short-notice evacuation planning; and macroscopic traffic simulation for real-time evacuation management. Subsequently, the modeling features of 22 traffic models and applications currently available in practice and the literature are reviewed and matched with the benchmark features identified for WUI fire applications. Based on this review analysis, recommendations are made for developing traffic models specifically applicable to WUI fire evacuation, including possible integrations with wildfire and pedestrian models. }
+}
+
+@Article{Pel,
+  author={Adam Pel and Michiel Bliemer and Serge Hoogendoorn},
+  title={{A review on travel behaviour modelling in dynamic traffic simulation models for evacuations}},
+  journal={Transportation},
+  year=2012,
+  volume={39},
+  number={1},
+  pages={97-123},
+  month={January},
+  keywords={Evacuation; Travel behaviour; Departure time choice; Destination choice; Route choice; Dynamic traff},
+  doi={10.1007/s11116-011-9320-6},
+  abstract={No abstract is available for this item.},
+  url={https://ideas.repec.org/a/kap/transp/v39y2012i1p97-123.html}
+}
+
+@article{McCaffrey_2017, 
+title={Should I Stay or Should I Go Now? Or Should I Wait and See? Influences on Wildfire Evacuation Decisions}, 
+volume={38}, ISSN={1539-6924}, 
+url={http://dx.doi.org/10.1111/risa.12944}, 
+DOI={10.1111/risa.12944}, 
+number={7}, 
+journal={Risk Analysis}, 
+publisher={Wiley}, 
+author={McCaffrey, Sarah and Wilson, Robyn and Konar, Avishek}, 
+year={2017}, month=nov, pages={1390–1404} }
+
 @book{rothermel1972mathematical,
   title={A mathematical model for predicting fire spread in wildland fuels},
   author={Rothermel, Richard C},
@@ -45,6 +140,31 @@ @article{https://doi.org/10.1002/eap.1898
     year = {2019}
 }
 
+@inproceedings{rempel_shiell_2022, 
+title={Using Reinforcement Learning to Provide Decision Support in Multi-Domain Mass Evacuation Operations},  
+url={https://review.sto.nato.int/index.php/journal-issues/2023-fall/sas-ora-conference-2022/68-using-reinforcement-learning-to-provide-decision-support-in-multi-domain-mass-evacuation-operations},
+ booktitle={NATO Operations Research and Analysis Conference}, 
+ author={Rempel, Mark and Shiell, Nicholi}, 
+ year={2022}, 
+ month={October},
+ address = {Copenhagen, Denmark}}
+
+ @article{10.1063/5.0209018,
+    author = {Budakova, Dilyana and Vasilev, Velyo and Dakovski, Lyudmil},
+    title = "{A reinforcement learning algorithm for the optimal evacuation route finding from an electrical substation}",
+    journal = {AIP Conference Proceedings},
+    volume = {3078},
+    number = {1},
+    pages = {040005},
+    year = {2024},
+    month = {04},
+    abstract = "{In this paper, the Intensity of the Characteristic Q-learning (InCh Q-learning) algorithm is proposed. It allows for finding the shortest and at the same time the safest escape route. Data on the intensity and spread of a fire occurring in a virtual electrical substation are used. Matrices are entered with the intensity values of each considered fire characteristic. As the fire spreads in space, zones of intensity are formed. Rules are introduced that take into account intensity zones for finding an escape route. The learning algorithm finds the shortest path for which the intensity of each of the dangerous features is the least. A threshold of the intensity of each character is introduced at which a person can pass and evacuate successfully. The use of priorities for the dangerous features allows one to choose a path with a greater intensity of the safer features and at the same time with a lower intensity of the more dangerous ones. The optimal escape route found is used to build a decision tree to find the location of an injured user as quickly as possible.}",
+    issn = {0094-243X},
+    doi = {10.1063/5.0209018},
+    url = {https://doi.org/10.1063/5.0209018},
+    eprint = {https://pubs.aip.org/aip/acp/article-pdf/doi/10.1063/5.0209018/19898413/040005\_1\_5.0209018.pdf},
+}
+
 @InProceedings{Diao2020,
 author = {Tina Diao and Samriddhi Singla and Ayan Mukhopadhyay and Ahmed Eldawy and Ross Shachter and Mykel J. Kochenderfer},
 booktitle = {AIAA Fall Symposium},

paper/paper.md

@@ -32,11 +32,12 @@ A major effect of climate change today is the increased frequency and intensity
 
 # Statement of Need
 
-There has been significant traction in the use of computational methods to study wildfires. In particular, reinforcement learning -- a subdomain of artificial intelligence where models learn through interaction with their environment -- has seen growing interest from researchers. Applying reinforcement learning requires modeling the spread of wildfires. Traditionally, modeling was primarily done using physics-based methods [@rothermel1972mathematical; @Andrews_1986]. However, newer methods are more data-driven, enabling the use of a higher diversity of features [@https://doi.org/10.1002/eap.1898; @Diao2020].
+There has been significant traction in the use of computational models to study wildfires. Historically, much work has focused on accurately modeling the spread of wildfires. While a lot of older methods were primarily done using physics-based methods  [@rothermel1972mathematical; @Andrews_1986] – with Rothermel being one of the most popular, as well the one we utilize in our package – newer methods rely on machine learning and other data-driven approaches, incorporating a higher diversity of features  [@https://doi.org/10.1002/eap.1898; @Diao2020;@ross2021being].
 
-Researchers have recently been studying wildfire surveillance and monitoring. While various forms of machine learning, such as computer vision [@ganapathi2018using], have been used to solve this task, the most popular method by far has been reinforcement learning [@Julian2019; @altamimi2022large; @9340340]. Research in surveillance and monitoring has been supported by open-source environments for modeling wildfire spread and surveillance [@cellular_automata; @forest_fire].
+Reinforcement learning (RL), a subdomain of artificial intelligence where models learn through interaction with their environment – has also been increasingly used in the context of wildfires. In combination with other traditional statistical methods and computer vision [@ganapathi2018using; @satelliteimages2017], RL has been applied to both the surveillance and monitoring of wildfires [@Julian2019; @altamimi2022large; @9340340]. An area where there has been little work in regards to RL is wildfire evacuation. Understanding the effective approaches for evacuating populated areas during wildfires is a key safety concern during these events [@KULIGOWSKI2021103129, @McCaffrey_2017]. As a result, work has been done to better model traffic during wildfire evacuation scenarios [@Pel, @doi:10.1061/JTEPBS.0000221], and agent-based evacuation simulations have been used for not only wildfires but also other natural disasters like tsunamis [@BELOGLAZOV2016144, @WANG201686]. RL has been previously identified as a potentially helpful tool during evacuation operations [@rempel_shiell_2022] and has been used to model evacuation during electrical substation fires [@10.1063/5.0209018]. The application of RL techniques to the wildfire evacuation task could thus prove beneficial. 
+
+Given the growing interest in studying wildfires through a computational lens, there have been developments in simulators for wildfires. A lot of open-source software focus on providing a visualization of wildfire spread [@cellular_automata; @forest_fire]. The most relevant piece of work to our paper are SimFire and SimHarness, which provide a system for wildland fire spread and a way for appropriate mitigation strategy responses via RL [@tapley2023reinforcementlearningwildfiremitigation]. Nonetheless, the focus is still on wildfire surveillance and mitigation, not on the task of evacuation.
 
-There has also been emerging interest in optimizing the evacuation process using computational methods [@https://doi.org/10.1111/risa.12944]. However, no reinforcement learning environments exist for the task of evacuation. We hope that open-source tools for evacuation will spur development in this area.
 
 # Methods