Date of Award

8-2018

Degree Type

Thesis

Degree Name

Master of Science (MS)

Department

Forestry and Natural Resources

Committee Chair

Douglass F. Jacobs

Committee Member 1

Guofan Shao

Committee Member 2

Kristen C. Nelson

Abstract

Fire prediction systems rely on meteorological descriptors and fuel characteristics to determine fire risk at national scales. However, at a regional scale, anthropogenic dynamics play an important role in determining fire ignition, as well as spatial and temporal distributions. Under an increasing fire activity scenario projected for the next century, Mediterranean ecosystems are particularly fragile regions. Fire variability driven by human stressors is the main threat to the native vegetation and human populations in these regions. The inclusion of anthropogenic indicators on fire prediction systems, especially within Mediterranean ecosystems, is key to developing accurate predictions and effective fire management efforts. As the relationship between human dynamics and fire is complex, it is important to first understand the landscape and socioeconomic perspectives of the human component in these regions and then to identify which specific anthropogenic indicators have the most significant effects on fire in order to include them in the fire predictions systems.

The first case study (CHAPTER 2) focuses on understanding and selecting the landscape transitions, intensity rates, and patch characteristics that have a significant effect on fire variability in Chile. Landsat eight scenes were classified based on spectral signatures to derive four land use categories between two-time intervals. The classification outputs were used to perform a change detection and intensity analysis. The second case study (CHAPTER 3) focuses on selecting the most significant socioeconomic variables that affect fire in Chile and integrates all the significant anthropogenic descriptors into a fire prediction model. To do so, spatial analysis tools were used to understand spatial distribution patterns of fire frequencies. Furthermore, regression models were used to select the most relevant human variables affecting fire frequency change. Finally, based on the data over dispersion and zero frequency characteristics, a zero-inflated model was used to simulate fire frequency predictions. The output predictions were then compared against a climate-based prediction model to evaluate fire prediction accuracy at a regional scale.

In the first case study (CHAPTER 2), regional differences were found in land use transition and characteristics. Twenty-seven percent of the area experienced a change in land use mainly associated with decreases in agriculture and increases in forest/plantation areas. Both transitions significantly decreased the landscape homogeneity. Across space, both landscape transitions and characteristics significantly affected fire frequency changes. The highest increases in fire frequency were related to increases in landscape heterogeneity, increases in forest/plantations (patch mean area) fragmented into multiple (patch number) distant patches (patch density), and decreases in urban and bareland areas.

In the second case study (CHAPTER 3), the spatial distribution of the fire activity was clustered towards the southern regions in years with extreme fire events, categorizing the area as an oscillating hotspot. The socioeconomic variables had a significant effect on fire frequency. Increases in fire frequency were related to increases in poverty percentage and road access. Opposite socioeconomic characteristics were related to decreases in fire frequency. Furthermore, 50% of the fire frequency was explained by the integration of the socioeconomic and landscape descriptors. Furthermore, all the socioeconomic characteristics affecting the fire frequency also had a significant effect on reducing the landscape homogeneity. All the significant descriptors were incorporated into a fire prediction model (LE Social model) and the output was compared to current climate model outputs and observed fire frequency. The LE Social model had a better goodness of fit (1.52) than the climate model (1.73). The LE Social model had a higher accuracy in the predictions in regions located towards the southern areas of the country. On the other hand, the climate models had higher accuracy in regions located towards the north. Finally, the accuracy of both models was reduced when predicting extreme fire frequencies due to the reduce seasonality and spatial distribution of those events that might be explained by a different driver not included in this study.

Results from the study highlight the strong impact of landscape and socioeconomic variables on the fire frequency. At the landscape level, both intensity and transition played an important role in fire frequency change. Furthermore, sites that meet the landscape criteria described in the first case study (CHAPTER 2) have a higher susceptibility to increases in fire frequency. Therefore, those areas should be considered as priority areas for management. Furthermore, despite the previous conceptions about the relevance of climate variables on fire predictions, the second case study (CHAPTER 3) found that the accuracy of the fire predictions using climate descriptors is regional-dependent. The effectiveness of the fire prediction models was highly dependable to the socioeconomic, landscape, and climate differences but temporal dynamics (year differences) as well. Therefore, the incorporation of the internal anthropogenic characteristics on fire predictions accuracy does have an effect in areas with high landscape heterogeneity and poverty levels. These results may provide important insight to help improve current fire prediction systems.

Download

Share

COinS