Exploration of wildfire features in consideration of drought impacts and terrestrial influences with a focus in California, USA

Abstract

The rational estimate of fuel product is significant to prevent large wildfires or mitigate their detrimental effects. The better knowledge of interactions among wildfires, droughts and vegetation dynamics can contribute to better estimate of fuels induced by droughts. In the thesis study, statistical modeling, temporal and spatial feature exploration, attribution analysis and impacting relationships were utilized. Due to data availability and accessibility, the study period was from 2001 to 2017 and the study area was California, USA, where large wildfires frequently occur and catches the worldwide attention. In the study, the three wildfire features, i.e. the number of wildfire events, wildfire duration and area burnt, were revealed. The wildfire season in California is found to be from June to October and more wildfires occurred in 2003, 2008 and 2017. The study disclosed that the large wildfires be defined if the burnt area is larger than 5 km2 in terms of percentile distribution of the area burnt. Then, 13 factors, i.e. topographic index, land covers, elevation, aspect, proximity to rivers, slope, precipitation, wind speed, the maximum, minimum and mean temperature, soil moisture and vegetation, were explored to understand the driving factors in triggering wildfires. It is discovered that terrain features and water conditions are the most important variables in association with wildfire occurrence. Moreover, vegetation condition is a primary factor in triggering small wildfires and soil moisture condition is the dominant variable in triggering large wildfires. The temporal and spatial drought conditions in the total, upper, middle and lower soil layers can be used to indicate that severe drought years in California are 2001-2002, 2007 and 2012-2015, especially the latter two droughts. Then, it is found that more wildfires occurred in the ensuing year exactly after the drought year(s). Large wildfires occurred more probably after the longer periods of dry conditions than small wildfires. The dominant negative relationships between soil moisture and natural logarithm of area burnt documented that severe droughts play a crucial role in triggering large wildfires. In the study, a new method was proposed to estimate fuels induced by droughts based on the linear or nonlinear relationships between vegetation and soil moisture, which was optimized by the linear regression and random forests. The proposed method selected the mean value of 10 largest periods of EVI values in each year as the best indicator of vegetation. The fuels was then represented by the negative ratio of the difference of the original vegetation status and the rebuilt vegetation status to the original vegetation status. The rebuilt vegetation is calculated by the vegetation-soil moisture relationships. Then, the valid fuels were identified to be fuels in 2 years according to the largest correlations between fuels in different years and area burnt. The assessments of valid fuels are then availably validated by Thomas fire occurred in 2017 and Mendocino complex fire occurred in 2018. The new proposed method for fuel estimate can be applied to other regions in the word for wildfire risk management.