Modelling forest fire spread using Cellular Automata

Forest fires are common events that have been a part of the human race since 1.5 millions years BP. Despite forest fires being a natural part of our environment, many consider them problematic due to the perception of a rising occurrence, severity and losses. With higher global temperatures, forest fires are predicted to increase around the globe, negatively affecting the risk of lives, health and infrastructure. In an environment where the forest fire risk increases, mitigating and preventive actions are required and this thesis focus on the mitigating actions of fire fighting operations in Sweden, and the possibility of predicting fire spread using simulation tools. A simulation tool using Cellular automata was developed using a probabilistic mathematical model by Alexandrisis et al. [21], including an additional parameter for the Fire Weather Index to account for fuel moisture levels. The cellular automata is a 2D grid consisting of a number of cells distributed in a matrix form and the tool simulates fire spread between the cells based on a number of rules and cell states. The software development has focused on minimizing user input to enable operational use and automatically gathers data (weather, Fire Weather Index, elevation, land-use etc.) through Geographic Information Systems such as Google Maps and the Swedish Meteorological and Hydrological Institute (SMHI). When using the tool the user specifies a geographical location (latitude and longitude) for the fire starting point and the date/time. The results are exported to a .csv datafile that is imported and visualized using Kepler.gl. As a part of the thesis a software proof-of-concept was developed and tested through three steps. The first step verifies the fundamental functions of the software, testing fire spread without external factors, with wind, and with slope to ensure that all parameters are working as intended. Additionally the Fire Weather Index parameter was tested for each case, which was confirmed to increase or decrease the fire spread rate of the model. Secondly the land-use identification algorithm was tested using three sub-scenario environments; forest only, forest/water, and forest/water/urban. The results confirms that the software can identify the land-use, with reduced accuracy at the interfaces between two land-types. Finally a case study was conducted and the simulation tool was tested and compared to the Västmanland fire in Sweden, 2014. The results from the simulation was that the software did manage to detect the land-use and correctly show a similar fire spread direction as the real fire. There were some discrepancies with regards to the temporal development and the total area burnt, but the foundational parts of software are shown to be satisfying and will provide a good basis for future work