Date of Award

Spring 1-1-2018

Document Type

Thesis

Degree Name

Master of Arts (MA)

First Advisor

Jennifer Karareka. Balch

Second Advisor

Stefan Leyk

Third Advisor

Carson Farmer

Abstract

Fire events, which delimit the extents of fires as well as fire start and end dates, provide important information for understanding fire regimes, including ignition points, spread rates, seasonality, and final size. Rates and patterns of spread associated with individual events are critical for modeling fire behavior across varied ecoregions and informing fire management. Fire events are not available from satellite-derived burned areas directly, and therefore a consistent process for aggregating burned pixels into fire events is needed. This study presents a flexible flood-fill algorithm that aggregates burned pixels into fire events based on varying spatial and temporal proximity. We tested our approach using data from the Moderate Resolution Imaging Spectroradiometer (MODIS) Burned Area product (MCD64A Collection 5.1) for the western U.S. in the year 2007. We used a range of spatial (number of pixels) and temporal (number of days) distances to cluster Burned Area pixels into events. We then compared Burned Area perimeters to perimeters based on the Landsat-derived Monitoring Trends in Burn Severity (MTBS) product for the same events. Based on this comparison, we determined that for the western U.S., a distance of five pixels and nine days constituted the best spatial and temporal thresholds to define individual fire events with the MODIS Burned Area product. We present a fire event dataset for the continental U.S. based on MODIS burned area for 2001-2016. Our fire event maps include important metrics of the fire regime, such as ignition point, fire size, fire spread rate, and the start and end dates of each individual fire event. The comparison between the MODIS Burned Area derived fire events and the MTBS, as the ground truth data, showed that omission, commission errors, and final accuracy are 12%, 38% and 88% respectively. We expect that the five-pixel, nine-day criteria may vary by ecoregion or land use type, and have designed the code so that these aggregation criteria can be adjusted for different regions. Our aggregation method increases the utility of the MODIS Burned Area product for fire modeling and management by enabling the analysis of fire behavior for individual fire events.

Share

COinS