fire-spread

Version 1.5

Background

Fire spread information on a large scale is still a missing key layer for a complete description of fire regimes. We developed a novel multilevel object-based methodology that extracts valuable information about fire dynamics from Moderate Resolution Imaging Spectroradiometer (MODIS) burned area data. Besides the large area capabilities, this approach also derives very detailed information for every single fire regarding timing and location of its ignition, as well as detailed directional multitemporal spread information. The approach is a top–down approach and a multilevel segmentation strategy is used to gradually refine the individual object membership. The multitemporal segmentation alternates between recursive seed point identification and queue-based fire tracking. The algorithm relies on only a few input parameters that control the segmentation with spatial and temporal distance thresholds.

For details on methods, please, refer to our paper:

David Frantz, Marion Stellmes, Achim Röder, Joachim Hill (2016): Fire spread from MODIS burned area data: obtaining fire dynamics information for every single fire. International Journal of Wildland Fire 25, 1228-1237. https://www.publish.csiro.au/wf/WF16003

Open Access Postprint: https://www.uni-trier.de/fileadmin/fb6/prof/FER/Publikationen/frantz_etal_ijwf_postprint.pdf

Install

Suggested procedure

Use fire-spread through Docker:

docker pull davidfrantz/fire-spread

Install it yourself

1) clone repository
2) cd fire-spread
3) make

Linux is required. GDAL is required. The program will be installed in $HOME/bin

If this fails, consider using docker instead.

Test data

A small testing dataset is included in this repository (testdata). You have to extract the data before using it.

Usage

fire-spread \
    input-stack input-dates output-dir basename \
    init-searchdist track-searchdist \
    temp-dist density-dist min-size smooth-dist \
    ncpu queue-size verbose

Argument	Short Description
input-stack	input data (see below)
input-dates	list of input dates (see below)
output-dir	output directory
basename	basename, which is used for naming output
init-searchdist	see paper
track-searchdist	see paper
temp-dist	see paper
density-dist	see paper
min-size	see paper
smooth-dist	see paper
ncpu	number of CPUs
queue-size	buffer size of first-in-first-out queue
verbose	print detailed information (v) or be quiet (q)

Sample call:

docker run \
    -v ~/testdata:/data \
    -u $(id -u):$(id -g) \
    -t --rm \
    davidfrantz/fire-spread \
    fire-spread \
    /data/CCI_1M_DOY_2001-2020_LAEA_example.dat \
    /data/timeseries.txt \
    /data/output \
    example_esacci \
    10 10 5 12 2 3 10 100000 v

Input data

input-stack

• monthly stack of burned area data (DOY)
• mosaic of complete area
• reprojected to some projection which minimizes distortion of both area and angles (e.g. equidistant, see https://doi.org/10.1016/j.cageo.2014.07.005)
• 16bit unsigned data format

Hints

• We developed the code for the MODIS burned area product MCD45A1 collection 5.1 but used it also successfully for MCD64 collection 6 (https://modis-fire.umd.edu/guides.html) and the ESA CCI data versio) 5.1 (https://geogra.uah.es/fire_cci/firecci51.php)
• The different fire seasons (one per year) will be automatically identified, and parallel processing (one CPU per season) will be used for speeding things up.

input-dates

• text file, which contains the dates of the input-stack as YYYYMM
• each date in one line
• last line should be empty
• use Unix EOL

queue size

• the queue size indicates the maximum number of items that can be held in the queue array. If this number is too low, just give a higher value until it works (we recommend iteratively increasing by a factor of 10). Admittedly, this is not the smartest implementation, but much more time-efficient than re-allocating memory dynamically.

A short note on the code

This is dirty research code from the early phase of my C-coding career ;) Not polished, but it gets the job done.