Date of Award

Spring 5-1-2020

Semester of Degree


Document Type

Open Access Dissertation

Degree Name

Ph.D. in Forest and Natural Resources Management


Forest and Natural Resources Management

Major Professor

Lianjun Zhang

Steering Committee Member

Stephen V. Stehman

Steering Committee Member

Lindi J. Quackenbush


In recent decades, the occurrence of forest fires has risen in the world and led to significant, long-lasting impacts on ecological, social, and economic systems. Along with the traditional tools for fire prediction, statistical modeling has been playing an important role in understanding the nature of forest fires and providing guidelines for decision making of fire prevention and management. In this dissertation, a large data set was collected from 2001 to 2016 in Fujian province, China, including the occurrence of forest fires and many environmental factors. I developed spatial generalized linear models and spatial quantile models under the framework of geographically weighted regression (GWR) to investigate the relationships between the counts and proportion or rate of forest fires and driving topographical, meteorological, human, vegetation, and land coverage factors. The corresponding global models were used as the benchmarks for model comparisons. These spatial models included: (1) geographically weighted Poisson and geographically weighted negative binomial models designed for the counts of forest fires; (2) geographically weighted quantile models for the counts of forest fires at different quantiles or risk levels; and (3) geographically weighted beta model for the proportion or rate of forest fires. The results indicated that the observed forest fires were highly likely to occur in lower elevation, smaller aspect index (meaning stronger sunlight), heavier precipitation, smaller population density, less vegetation, wider grassland, and/or less cropland, while other environmental factors varied greatly with the forest fire occurrence. This study showed the great superiority of these GWR models to the corresponding global models in terms of characterizing the spatial nonstationary relationships, producing better model fitting and prediction, providing a more complete view on the spatial distribution of forest fires, and highlighting the risky local “hot spots” of forest fires as well as environmental factors across the Fujian province, China. Hopefully, the more detailed and localized information would help and assist the forest and fire managers to better understand the behavior of forest fires and influences of the environmental factors across the study area. Thus, the government agencies can make wiser and better decisions on where and what the fire management and prevention should be focused on with reduced economic expenses and improved the efficiency of forest fire management.