Using decomposed MODIS NDVI time series to improve wildfire risk mapping in Mediterranean forests

Oren Glickman, David Helman, Steve Brenner, Itamar Lensky, Yaron Michael, David Gabay

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Forest fires are a common disturbance in Mediterranean climate. Current wildfire risk maps integrate
weather conditions, topography and fuel, but they do not consider the greening or browning of woody
vegetation. Using a sub-pixel decomposition method for time series of MODIS NDVI, Helman et al. (2015)
showed that the greening or browning of woody vegetation may be used to predict the risk of fire
spreading in Mediterranean forests and woodlands. They used the change of woody vegetation from the
decomposed NDVI time series as a metrics for fire risk prediction in Mt Carmel, Israel.
This study aims to expand the study of Helman et al. (2015) by using a large wildfire dataset in
Mediterranean climate. Date and perimeter of 135 wildfires from 2007 in Greece, Italy and Spain were
acquired from the European Forest Fire Information System (EFFIS) for this study. We defined a buffer of
3000m around each fire perimeter and extracted land-cover information at 300m spatial resolution from
the year before the fire from ESA CCI Land Cover project. Based on Helman et al. (2015)’s methodology,
time series of MODIS NDVI were decomposed for 2001-2006 to woody vegetation component
(NDVIwoody) and woody vegetation trend (NDVItrend). NDVIwoody was used to estimate fuel density,
while NDVItrend was used as an indicator for accumulated dry matter. The decomposition was
performed inside the wildfire perimeter and compared with the buffer zone (control). We analyzed the
effect of NDVIwoody and NDVItrend on wildfire probability using logistic regression with common
wildfire predictors.
Our results show that in areas with a large percent of woody vegetation cover fire risk prediction
improves by 4% (Area Under the Curve) when using NDVIwoody. Areas with a greater amount of
herbaceous vegetation show smaller improvement of only 2%, as expected. In general, NDVIwoody and
NDVItrend showed to improve the wildfire risk prediction by c. 3.5% and 0.5%, respectively. We
conclude that NDVIwoody and NDVItrend may be used as a standalone value for wildfire risk modeling.
Original languageEnglish
Title of host publication7 th INTERNATIONAL CONFERENCE ON FIRE EFFECTS ON SOIL PROPERTIES UNIVERSITY OF HAIFA, ISRAEL 18-21.02.2019
StatePublished - 1 Jan 2019

Fingerprint

Dive into the research topics of 'Using decomposed MODIS NDVI time series to improve wildfire risk mapping in Mediterranean forests'. Together they form a unique fingerprint.

Cite this