Wildfire incidence, characteristics, and the influence of weather at landscape to global scales

Student thesis: Doctoral ThesisDoctor of Philosophy

Abstract

This thesis investigates patterns of wildfire incidence and relationships between fire incidence/fire characteristics and weather. It does so over landscape (100s to 10,000s km2), regional (millions of km2) and global spatial scales, and over time periods ranging from several days to several decades. Specifically, the role of weather is examined through the Canadian Forest Fire Weather Index (FWI) System and its underlying meteorological variables. Three broadly related research areas are investigated, to determine: 1. whether a percentile-based calibration of the FWI System can be used to better identify periods of high potential fire incidence in the United Kingdom (UK) than an existing implementation of the System; 2. whether the FWI System can be used to support forecasting of the evolution of fire activity captured by the Global Fire Assimilation System (GFAS; Kaiser et al. 2012) at regional to global scales at 1 to 5-day lead times. Such developments could improve the representation of biomass burning emissions in air quality forecasting models; 3. whether a new global night-time active fire dataset can be created using the extensive (> 30 year) Advanced Very High-Resolution Radiometer (AVHRR) Global Area Coverage (GAC) archive, for the purpose of investigating the potential impacts of climate change on regional patterns of fire over recent decades. Results of the first study highlighted that a percentile-based calibration of the FWI System does appear to better capture periods of extreme fire weather and fire incidence across the UK and Great Britain than the existing approach. The new method also removes a spatial bias observed in the existing fire weather index, the Met Office Fire Severity Index (MOFSI). Furthermore, a comparison with historic fire activity for Great Britain indicates that the vast majority of wildfires occur at extreme values of the percentile-calibrated FWI: during 2010-2012, the 75th, 90th and 99th percentiles of at least one FWI component were exceeded during 85, 61 and 18 % of all wildfires respectively. Operational implementation of this percentile-based approach to assessing fire weather conditions may enhance wildfire preparedness and planning in the UK. The second study involved the development of a GLM-based methodology for forecasting daily regional observations of Fire Radiative Power (FRP) using meteorological variables and the moisture codes/fire weather indices of the FWI System. Results of simulated forecasting with reanalysis data showed that this forecasting method improved upon persistence forecasts at the global scale (global mean RMSE reduction = 7-24 %, lead time dependent) and regional scale (regional mean RMSE reduction = 8-17 %, lead time dependent). The GLM-based methodology 3 also showed improvement in forecasting skill over the recently developed forecasting methodology of di Giuseppe et al. (2017) that utilises just the FWI component of the FWI System; forecast RMSE was reduced at both the global scale (global mean RMSE reduction = 7-16 %, lead time dependent) and the regional scale (regional mean RMSE reduction = 3-6 %, lead time dependent). The third study developed a new contextual fire detection algorithm for use with AVHRR GAC data that resulted in the creation of a new global record of fire activity spanning the 1986-2016 period. This dataset was intercompared with several other established – but more temporally or spatially limited – datasets: the MODIS Collection 6 Active Fire Product (Giglio et al., 2016), the Canadian National Fire Database (Stocks et al., 2002) and the recent MODIS-era global fire analysis of GFED burned area (Andela et al. 2017). Interannual variability and multiyear trends observed in the AVHRR dataset were in good agreement with those noted in these benchmark datasets at both global and regional scales. A global analysis of regional scale trends over the 1986-2016 period highlighted several statistically significant (Kendall’s τ, p < 0.1) linear trends and evidence of non-linear trends in fire activity that broadly reflect observations found in the wider literature. Notable regional trends identified over the 1986-2016 period include: statistically significant (Kendall’s τ, p < 0.1) decreasing linear trends in Europe and Northern Hemisphere Africa, likely related to changes in fire management and land use; statistically significant (Kendall’s τ, p < 0.1) increasing linear trends in the western USA, likely related to a combination of factors but including anthropogenic climate change; and a non-linear trend in South America, likely linked to changing rates of deforestation practices in the Amazon Basin. Further refinement of this AVHRR dataset may result in the creation of a new global fire database of use to the wider research community.
Date of Award1 Jun 2019
Original languageEnglish
Awarding Institution
  • King's College London
SupervisorMartin Wooster (Supervisor) & James Millington (Supervisor)

Cite this

'