Anatomy of the "Las Máquinas" wildfire using remote sensing tools

Abstract

In the austral summer of 2017, Central and Southern Chile was severely affected by a series of wildfires, burning an area of 575,000 ha (5,750 km2), leaving 11 dead, more than 4,000 people affected and destroying over 1,000 structures. One of the fires, named the “Las Máquinas” wildfire, has been the largest single event ever recorded in Chilean history, burning an area of around 193,000 ha (1,930 km2) and killing 4 people. In this work, the anatomy of the “Las Máquinas” wildfire is studied by using the advantages given by Earth Observation tools. By definition, anatomy corresponds to “a study of the structure or internal workings of something”. For this purpose, the pre-, per- and post-fire conditions were described using open-access satellite imagery obtained from multiple remote sensing instruments, giving a new perspective to the description of this event. In addition, the Canadian Fire Weather Index (FWI) was calculated for the entire region of interest on a daily basis, based on remotely sensed data and physical simulations. The findings suggest that within the region of interest (ROI), there was abundant and healthy vegetation in terms of the Normalized Difference Vegetation Index (NDVI). Moreover, some affected municipalities had not experienced wildfires for several years prior to the event, and therefore a large amount of fuel availability was observed. In terms of the flora, Empedrado was the most affected municipality, lowering its average NDVI from 0.67 to 0.36. The species affected were mainly Pinus radiata, Vachellia caven and Nothofagus glauca, whose total burned area was 129,700, 13,700 and 13,500 ha, respectively. On average, the maximum temperatures at the whole ROI achieved values of around 29◦C in December 2016, the month prior to the event, being the highest since 1979. Additionally, the standardized precipitation index (SPI) indicates that there had been a severe and persisting drought since 2009. Burned area was calculated using Landsat-8, Sentinel-2 and combined Aqua/Terra-MODIS images. From MODIS data, the total burned area was 193,804 ha (1,938 km2), while Landsat-8 and Sentinel-2 burned area was studied for multiple threshold values of the delta Normalized Burn Ratio (dNBR). High severity dNBR values were found mainly in Constitución and Empedrado municpalities, particularly close to the village of Santa Olga, which was completely decimated by the wildfire on the night of January 25th. According to MODIS data, only on that night, the fire propagation left a total burned area of 85,000ha (850 km2). Using both the VIIRS-SNPP and MODIS products, the fire radiative power indicated thatthere were some locations where the total amount of energy released by the wildfire exceeded 2,000 MW,mainly in the Constitución municipality and within a 7 km radius of Santa Olga around the time the villagewas destroyed. During the event, the air quality was studied by using the VIIRS-SNPP Deep Blue Aerosol product, and the results exhibited an elevated presence of both high altitude smoke and pyrocumulonimbus clouds. Furthermore, carbon emissions for January 2017 resulted in mean values of 197.49 g·C·m−2 , with maximum values of 1,270 g·C·m−2 , 3,750% higher than in previous years for the same month. The Canadian Fire Weather Index (FWI) was calculated using near-surface air temperature derived from multiple linear regression models, machine learning regression algorithms, and from a physicallybased surface energy balance. Overall, the best performing models were those using the machine learning random forests regression. The FWI was computed using Aqua, Terra, VIIRS-SNPP, Energy balance and through an approximation approach which simplifies its calculation (FWIapp). In general, all FWI results exhibited the same behavior, achieving the highest FWI values on January 25th, 2017. The results show that with this methodology it is possible to estimate a daily landscape-scale FWI at a medium resolution derived from Earth Observation tools without strongly overestimating nor underestimating its value compared to the FWI calculated at weather stations.