On October 8th of this year, a series of wildfires broke out in Northern California. While the causes remain unclear, synoptic atmospheric conditions were such that hot, dry offshore winds provided optimal conditions for rapid fire growth. Autumnal fall winds such as these have historically been associated with devastating fires in the region, as well as California in general. Sadly, these fires are no exception, with the week following October 8th being the deadliest week of wildfires in California's history.
Here I present a preliminary assessment of the extent of the severity of these wildfires using publicly-available multispectral images. I use the differenced Normalized Burn Ratio (dNBR) to give a first-order approximation of the extent of burned ground. My conclusions are no different than those that are available from Cal Fire, but the analysis is simple enough and available to anyone with access to a GIS software package (using the open-source QGIS, this is everyone with a computer). The entire process took me, a remote sensing amateur in every sense of the phrase, one lazy Sunday afternoon to put together. It would be a pretty decent exercise for a dedicated GIS class or environmental change module.
I first became interested in this topic in January, when Chile suffered through a summer of unprecedented wildfires. I found it occasionally difficult to encounter up-to-date information on the state of the fires as they developed, and so discovered the host of open-access near-realtime satellite data available to the general public. The USGS' Landsat 8 and the ESA's Sentinel-2 are the premier primary sources of open-access multispectral data spanning the visible to the shortwave infrared spectra. For this analysis I use Sentinel-2 data, principally because I haven't worked with Sentinel-2 products before, but also because they have a shiny new satellite (Sentinel-2b) as of March 2017. The Sentinel-2 pair will ultimately have 5-day revisit coverage for its operational area, but as they are still working out the kinks I am limited to Level-1C images from 27 September and 12 October for this project.
After downloading the pre-fire and post-fire images from Copernicus, I calculate the Normalized Burn Ratio (NBR) for each scene. The NBR takes advantage of the vast difference in reflectivity between vegetation and burned ground in the near-infrared and shortwave infrared spectra. For Sentinel-2, the equation is:
NBR = ( R8 - R12 ) / ( R8 + R12 )
where Rx = reflectivity of band x.
The difference between pre-burn and post-burn NBR gives the differenced NBR:
dNBR = pre-burn NBR - post-burn NBR
The general correspondence between burned areas according to the dNBR and what I can see in the visible spectrum images is quite good, although for quality assurance one would need some time for field work (not an option for me unfortunately). There are some notable false positives that largely correspond to cultivated land--perhaps not surprising given the time of year, when the dNBR might pick up on fields recently harvested and plowed under. Conversely, burned sections of dry grass often yield a negative dNBR, implying post-burn regrowth, which is almost certainly not the case. This could also be explained by the reduced NIR reflectance of dead matter, and its subsequent effect on pre-burn NBR.
Assuming all dNBR values > 0.1 indicate burned area, we can estimate a maximum burned area using the Sentinel-2 band 12 cell size of 20x20m.
- In Napa County, ~200 km² burned, equivalent to 50,000 acres or ~10% of the total land area.
- In Sonoma County, ~180 km² burned, equivalent to 44,000 acres or ~4% of the total land area.
This occurred in less than four days.
Here's the LA Times on how to support survivors and the displaced.