Estimating Grassland Curing for Wildfire Danger Assessment from Satellite based microwave Data.
Chaivaranont, W, J.P. Evans and Y. Liu
MODSIM2015, 21st International Congress on Modelling and Simulation. Modelling and Simulation Society of Australia and New Zealand, Gold Coast, Australia, 29 November – 4 December 2015.
During an extreme summer in Australia, wildfire can become a catastrophic natural hazard.
Multiple meteorological and geographical elements can determined the severity level of fire, ranging from
small scale bushfire to uncontrollable fire. There are various types of fire danger level assessment systems,
where the most common systems used here are McArthur's Forest Fire Danger Index (FFDI) and Grassland
Fire Danger Index (GFDI). The degree of curing is one of the important factors, as a fuel moisture
representation, in determining the fire danger level in GFDI for the grassland environment. It is a
measurement of how dry the grass is in percentage, where a completely dry grass field is indicated by a 100%
cured value. Since it is very tedious and costly to accurately and consistency obtain curing field
measurements, the degree of curing for calculating grassland fire risk is usually assumed constant.
The aim of this study is to accurately estimate the degree of curing using a passive microwave based satellite
product referred to as vegetation optical depth (VOD). VOD illustrates the vegetation dynamics at a continental
to global scale and acts as a proxy for vegetation water content. It is a unit-less measurement of vegetation
water content with a value ranging from 0 to 1.3, where 0 indicates that there is no above ground biomass
present or it is completely dry and 1.3 indicates a dense, fully saturated above ground biomass. VOD used here
is the most recent version with a 10 kilometres spatial footprint and an 8-day temporal interval. Note that the
VOD retrieval is usually accompanied with top layer soil moisture retrieval at a same spatial and temporal
By correlating the observed degree of curing with VOD, a significant relationship can be found. However, it
should be noted that the VOD values for each observation site across Australia are very site-specific with a
distinct base value, such that VOD value at one site may range from 0.7 to 0.9, and another may range from
0.4 to 0.6. Since this study is aiming for a continental scale prediction model, site classifications and additional
remote sensing data were used to supplement VOD. For instance, using the MODIS MCD12C1 land cover
type map, the sites located in the forests were removed, since at a 10 kilometres spatial resolution, if the
majority area within the pixel where the observation were taken is classified as forest, then the data interpreted
from the satellite signal is not going to be a good representation of the grass or cropland site data; it is usually
found that the forest sites have a much higher minimum VOD value, because the total moisture in above ground
biomass is much more higher in forest than grassland.
An additional satellite reflectance data for determining the vegetation greenness called Normalised Difference
Vegetation Index (NDVI) is also used as a supplement data in the multiple linear regression model for
estimating curing, since NDVI has a more normalised range across all sites and a resolution of 500 metres.
Major drawbacks of NDVI are that they are prone to cloud cover interferences and a rapid saturation in dense
vegetation. NDVI used during this study is computed from MODIS MOD09A1 reflectance dataset.
After several site selection criteria and multiple linear regression models were tested, the optimal models for
estimated curing data can be derived based on the curing and VOD with NDVI multiple linear regression. The
computed estimated curing data is then evaluated with the observed curing data again to ensure the robustness
of model performance. Note that the observed data that were used for evaluation phase also contains data that
were omitted during the calibration phase. The results suggested that the observed overall temporal trend in
curing across various sites can be reproduced with estimated curing models. By utilising the estimated curing
models, the curing dataset for Australia can be vastly expanded spatially and temporally.
This page is maintaind by Jason Evans |
Last updated 31st January 2013