In Greece, Canada or Spain’s Tenerife Island – This summer we’ve witnessed quite a number of examples of how devastating and dangerous wildfires are. This phenomenon is generally linked to summer months because of temperature rise and dryness, but it’s a year-round disturbance given the ecoregion and climate zone diversity in our planet.
In this entry we want to delve into the fire danger indexes monitoring capacity and compared three of them with total burned area observed data to present here the correlation between them. So in this data viewer (figure 1) you can see some sort of fire danger index validation – what’s the total burned area in a given ecoregion compared to what a given index alerted about?
Let’s have a look at the area the mouse is pointing, a South Apennine mixed montane forest area in Southern Italy:
By clicking on the region we obtain figure #2, where it can be seen how in this segment the selected indexes (the Fire Weather Index -FWI- from the Canadian Forest Service Fire Weather Index Rating System, the Fire Danger Index -FDI- from the Australian McArthur Mark 5 Rating System and the Burning Index -BI- from the U.S. Forest Service National Fire-Danger Rating System) alerted about the risk and the extent of it quite accurately.
The correlation level can be checked in the Region Information widget, as shown on figure #3:
Does this mean the Burning Index is the best, the most reliable and accurate one? The answer is yes – for this particular region. But no, in general. It depends on the ecoregion and the aggregation method used. In this case the mean has been used, whilst our data viewer provides information for other aggregation methods such as percentile 90 and percentile 95, allowing for more extreme events. If we keep using the mean metrics, the colour map will show which index offers a higher correlation in each ecoregion:
Fire seasons
In the menu to the right of the map you may also click on Fire Season. The map will then show the world’s surface in purple, the darker they are, the longer the fire season is (figure #5).
In this particular case, a ten-month fire season, an especially long one, caught our eye. It’s the ChangJiang Plains Evergreen Forests, a Humid Subtropical area around the Yangtze River Delta, in Eastern China, where important cities such as Shanghai, Ningbo, Hangzhou or Nanjing sit.
Again, by clicking there you can get the ecoregion information and see its fire season spans from June to April, one of the world’s longest. But what really made us raise our eyebrows was the drastic burned area reduction over the past few years – as you may see in the diagram below, from 2015, the total burned area is virtually nonexistent. What is this due to? A drastic improvement of the nature protection system? A sudden increase in the timber industry? Maybe due to changes in land uses?
Part of the answer can be found in journal publications such as this article on Nature, in which you can read that “there was a significant increase in plantation in SE China to meet rising demand for forest products”. “Forest fires in SE region accounted for around 84% of total fire occurrences in China, which were largely of anthropogenic origin […] However, there was an overall decreasing trend in forest area disturbed by fires from 1986 to 2020, suggestive of increasing intensity of forest operations […]”.
It’s also partly due to the implementation of forest protection policies and to the progressive urbanisation of the area, as can be seen in the image below showing a random spot of that ecoregion. We observed it via Google Earth Engine TImeline, which allows you to check the evolution of a satellite picture over the years over a given part of the Earth.
Burned area
Furthermore, you can select the “Total burned area” option in the viewer menu. The map will then show several shades of ochre, the darker shades indicating those areas with a bigger extension of burned area.
In this case, our magnifying glass is pointing to a particular strip of the Sahel desert, south to the Sahara, savanna territory populated by lonely acacias and scattered bushes…
...that nevertheless is registering massive fires every year, as can be seen in the following image:
How come such a vast area, around a million hectares a year, is burned in a region with such low vegetation density? How are these fires computed?
A first answer to these questions is related to the huge extension of the relevant area, prone to fire due to the sun’s constant presence and its resulting dryness. A second answer has more to do with the fact that, despite being savanna, there are some vegetated pockets apart from the ones differentiated in the viewer as flooded savanna along the Niger river or around Lake Chad. In some parts, the vegetation has enough connectivity for the fire to expand, and all of it counts as "burned area".
How did we put all this together?
As a first step, we compiled the following data:
- Fire danger indices: The Copernicus Climate Data Store (CDS) catalogue entry named “Fire danger indices historical data from the Copernicus Emergency Management Service” [ECMWF 2019] offers historical –1979 to present– information for a set of indices related to fire danger. These indicators are retrieved in a regular latitude-longitude grid which covers the entire globe (180ºW to 180ºE, 90ºS to 90ºN) with a spatial resolution of 0.25º and a daily temporal resolution. From this CDS catalogue entry, data for 3 variables was downloaded: the already-mentioned FWI, FDI and BI indexes.
- Fire burned area from satellite observations: We have used monthly data of Burned Area (BA) at 0.25º resolution from the “Fire burned area from 2001 to present derived from satellite observations” database (C3S 2019a; C3S 2019b) which is publicly available through the .Copernicus Climate Data Store as part of Copernicus, the European Union’s Earth Observation Programme managed by the European Commission. The BA data used are derived through the analysis of reflectance changes from the medium resolution sensors Terra MODIS and Sentinel-3 OLCI, helped by the use of MODIS thermal information. The algorithms used are adapted to the native data from these sensors to produce an homogeneous gridded dataset of global coverage containing monthly data of BA at the grid scale, extending the database to the present.
- RESOLVE Ecoregions dataset (2017): RESOLVE Biodiversity and Wildlife Solutions provides access to this dataset containing information about 846 in-land ecoregions which depict an overall representation of our living planet (Dinerstein 2017). There is an online data viewer showing the above mentioned ecoregions.
After that, we proceeded to preprocess these data -- the process to clean and homogenise the raw data –described above– can be divided in the following tasks (schematized in figure 6):
Despite having the same spatial resolution (0.25º), the fire danger indices data needed to be re-gridded to the burned area observations grid due to the fact that their meshes presented some differences. Afterwards, the data is temporally aggregated to the same temporal resolution. Therefore, the fire danger indicators are converted from daily to monthly data and the burned area observations remain in a monthly resolution.
As a last step, the ecoregion's metadata are used to perform a spatial aggregation on both the fire danger indices and the burned area observations.
The aggregation methods used remain constant in both temporal and spatial aggregation algorithms. As established in Table #1, the observed burned area was computed by the sum in order to have the total amount of area burned on time and space. However, the fire danger indices had a different approach and were computed using several statistical options such as the mean, the 90th percentile and the 95th percentile.
As a consequence, there will be 10 different variables after the preprocessing task (+10 significance test results) for a total number of 693 ecoregions. This number is lower than the total number defined by the RESOLVE dataset due to the following specific facts:
- Some ecoregions do not have enough extent to perform the temporal and spatial aggregation over them (no grid points inside them).
- Only terrestrial information was provided by both the fire danger indices and the burned area observations.
- There are ecoregions with no presence of fire on their surface.
We hope you found this interesting! As you very well know, should you need any data visualisation solutions, you may reach out to our team via predictia@predictia.es.