Will Fires Become More Frequent in the Western Cape?

Where there’s smoke, there is fire… literally if you are in any part of the Western Cape during summer. Smoke travels far, making the air hazy and stink of whatever is burning. If you are close enough, your eyes and lungs will burn. Smoke, the evidence of fire, is there but the whereabouts of the fire is not always clear, so how worried should you be? Before you start to panic and call your local fire station or Remote Sensing expert at Umvoto, checkout www.afis.co.za (unless you can see flames, then call emergencies services ASAP!). The Advanced Fire Information System (AFIS) is a South African developed satellite-based fire information tool providing near real-time information to users across the globe. The AFIS screenshot of the recent fires (5^th February 2022) around Steenbras, the N2 and Grabouw area was sent to a friend in Cape Town who wanted to know where all the smoke was from. Satellite information makes seeing fire frequency and impact, quick and easy – no need to wait for a news update or politically-heated social media post to find out what is happening.

Wildfires in the Western Cape can cover large areas of mostly mountainous regions, making both assessment and response to fires difficult. Using remote sensing tools makes assessment quicker and easier, and allows for pre-emptive planning to occur. The value of fire-related data and platforms can be seen by looking at a recent fire event. On the 8^th January 2022 a wildfire started near Kleinmond, and according to a News24 article it burnt for 4 days, destroying at least 5 400 ha of land.

afis app — *Screenshot of the AFIS interface showing the 5^th February 2022 fire near Grabouw.*

Sentinel Hub Playground

To visualise the Kleinmond fire extent (yellow box in the image) the European Space Agency’s (ESA) Sentinel-2 satellite imagery can be processed and mapped using Sentinel Hub Playground. The false colour composite image highlights vegetation in shades of red, with the ocean and Bot River Estuary in darker shades of black and blue. The pre-fire image was captured on 31^st December 2021, 8 days before the fire started. The post-fire image was captured on the 30^th January 2022, about 2 weeks after the fire, which clearly shows the burn scar.

kleinmond pre post fire map — Top: Pre-fire scene from 31^st December 2021, with vegetation in shades of red. Bottom: Post-fire scene from 30^th January 2022 with the extent of burn area (fire scar) highlighted by the yellow polygon.

Fire Weather Index

To better understand why the fire happened when it did, we can look at the Fire Indices dataset. This data from the Copernicus Emergency Management Service (CEMS) for the European Forest Fire Information system (EFFIS) has, among its many parameters, a Fire Weather Index (FWI), which is used for public information about fire danger conditions. This dataset provides a complete reconstruction of meteorological conditions favourable to the ignition, spread and sustenance of fire and is calculated from weather forecasts from historical simulations provided by the European Centre for Medium-Range Weather Forecast (ECMWF) ERA-5 reanalysis. By combining model data and a vast set of quality-controlled observations ERA-5 provides a globally complete and consistent dataset and can be a good proxy for observed atmospheric conditions.

Climate Engine

Using Climate Engine, a web application powered by Google Earth Engine, the FWI can be plotted with the “Make Map” functionality for custom time periods anywhere on the globe. The map below illustrates mean FWI for summer season (01/12/2020 to 28/02/2021) highlighting the high FWI values for the Western Cape in red shades.

map fire weather index — A global snapshot of the December to February mean Fire Weather Index, where reds indicate higher FWI. Note the higher FWI scores for areas around the equator and southern hemisphere. However, FWI scores should not be used without context as can be seen by the high FWI scores for the Saharan and Arabian deserts.

Climate Engine also allows you to plot FWI over time and display a vegetation index, NDVI (Normalised Vegetation Difference Index). This NDVI comes from NASA’s MODIS (Moderate Resolution Imaging spectrometer) on board the Terra satellite. NDVI is a good proxy for vegetation health, and after a fire event the NDVI value drops, as seen for March 2011 below. EFFIS classifies FWI range 21.3 to 38.0 as high, 38.0 to 50 very high and 50-100 as extreme. Even though FWI might be high it does not mean that fire events will occur. This just means conditions are more conducive to fires occurring.

ndvi fwi timeseries — A comparison of Normalised Difference Vegetation Index (NDVI) and Fire Weather Index (FWI) values from Jan 2010 to Jan 2021. By comparing 11 years-worth of data for the Kleinmond region, seasonal trends can be observed. One such trend is the correlation between higher FWI and lower NDVI values are seen.

Fire Regime

By looking at the pattern of behaviour of NDVI (vegetation) and FWI (fire) data we can see that conditions that promote fynbos growth, reduces the FWI, and vice versa. This makes the Western Cape austral winters, where most of the rain is received (i.e. wet and cool conditions), not very good months for fires but good for plant growth. In contrast, the FWI peaks in summer when moisture is low, and temperatures are high. Both variables are inextricably linked to local climate and will therefore be impacted by climate change (which predicts hotter summers and drier winters for the southwestern Cape).

Another strength of using Climate Engine to understand fires, is its ability to summarise big datasets as seen in the time series for NDVI and FWI that date back to 1979. By plotting the annual mean NDVI and FWI values it is possible to see which years have averaged higher FWI values. From this simple plot, derived from lots of data, it is expected that for the Kleinmond region fires take place around the summer season or just before or after summer every 8 -12 years. This is correlated by plotting the previous three major fire events, March 2011 (12 721.04 ha lost), November 1999 (13 291.12 ha lost) and April 1991 (2336.88 ha lost) sourced from Cape Nature, as the vertical red dotted lines.

graph of fire weather index — A +30 year comparison of NDV (dashed green line and green shaded region) and FWI (solid line changing from red to blue) values, with known large fire (dashed red vertical line) shown. From this it can be seen in 1991 and 2011 fire events were preceded by a period of high FWI, and as the vegetation is burnt the NDVI decreases sharply after a fire event. Important to note is that the 1999 was not preceded by high FWI values but the NDVI values had reached pre-1991 fire levels, suggesting there was sufficient biomass to sustain a fire.

Fire Predictions

By keeping track of fynbos age (age is set to 0 by a fire event) and FWI, it is possible to predict which areas are most likely to burn and when. The dominant predictor for wildfire appears to be age of fynbos, also interpreted as amount of biomass available to burn. Should FWI values become higher, or persist for longer periods of time (extending beyond summer season and peaking earlier), we could expect to see more frequent fires. Possibly with more frequent fires, wildfires will become less severe as there will be less biomass build up to burn. Although fire is part of the various fynbos species’ life cycle, more frequent fires would negatively affect fynbos recovery and species diversity.

Burn Severity using Normalised Burn Ratio

Once a fire has happened, mapping and quantifying the area burned can be done using the Normalised Burn Ration (NBR) following the recommended practice from UN-SPIDER (a platform from the United Nations for Outer Space Affairs facilitating space-based technologies for disaster management and emergency response). This can more accurately delineate fire scar and which land types experienced more vegetation loss. This information assists with developing emergency rehabilitation and restoration plans post-fire and can also give insights into possible erosion and landslides.

NASA’s Applied Remote Sensing Training Program (ARSET) in Using Earth Observations for Pre- and Post-Fire Monitoring showcases a Google Earth Engine script that make NBR analysis a quick job to undertake. The only required inputs are dates of pre- and post-fire and your area of interest. You also have a choice between using Landsat-8 or Sentinel-2 satellite imagery.

Maps showing burn severity — The Normalised Burn Ration (NBR) index highlights burnt areas. The index uses near infrared (NIR) and shortwave infrared (SWIR) bands either Landsat-8 or Sentinel-2 pre- and post-fire satellite imagery, post-fire burn severity can be assessed. Healthy vegetation shows a very high NIR reflectance and a low reflectance in SWIR (left hand side middle panel). Whereas fire scars have low NIR reflectance and a high SWIR reflectance (right hand side middle panel). The NBR index uses this difference, after creating a NBR for pre- and post-fire imagery, a comparison between the two results in the Burn Severity of the fire (bottom panel).

The News24 article stated that at least 5 400 ha were burnt in the Kleinmond area, while the NBR analysis shows the fire extent as +/- 6 700 ha. While the NBR may be a slight overestimation it has the advantage of being able to distinguish areas that have been burnt with high severity (382 ha) to low severity (1670 ha) as well as pockets of unburned land (243 ha) within the fire scar and areas that have already shown signs of recovery (1 ha). Pockets that are left unburned and show fast recovery not only aid ecological recovery but may also provide clues about water resources, as these areas may have elevated soil moisture or springs/seeps (possibly indicating zones of groundwater and surface water interaction).

Continued Monitoring

The NDVI on the 7^th January 2022, one day before the Kleinmond fire, was 0.41 (which is below the summer average NDVI of 0.5 in the preceding five days), showing the conditions were very hot and dry. Post-fire NDVI is recorded at 0.28. NDVI time series also reveals that the fynbos in this region was approximately 11 years old, and thus there was a high risk of burning because of biomass build up. This was coupled with a high FWI of 32.3 on the 7^th January and 27.7 on the 8^th January 2022 (sourced from the Copernicus Climate Data Store as real-time data which is not available in Climate Engine yet). This combination of drying vegetation (decreasing NDVI), high biomass (11 years since last fire) and high fire threat made the ideal conditions for fires, and a fire did occur.

Monitoring NDVI and FWI can go a long way in determining which areas are most at risk to fires. Monitoring these indices and linking them to observed events is key in building up a region’s fire risk profile, and allows it to be determined whether preparation for more devastating fires is required in the future (i.e., forming early warning systems as part of disaster risk reduction). All of this is enabled by satellite information that makes mapping fire frequency and impact easy and effective.

For assistance with your geographical information system and remote sensing needs, contact us today.