4.1. Analysis of Active Fires and Their Intensity over Various Vegetation Types
The results of the spatial distribution of Active Fires (AF) and Fire Radiative Power (FRP, MW) for the three study sites (
Figure 2) show the vast distribution of high-density fires across all study sites. These dense fires indicate the areas (i.e., hotspots) where these fires were intense and became disastrous. For example, the fires with the highest FRP were found around 40° and 48° N and 120° W (i.e., Washington and California) in the western USA, 10° S and 54° W in Brazil and 30° S and 150° E (i.e., New South Wales) in eastern Australia. Statistically, Moran’s
I statistic show the presence of statistically significant clusters and outliers, at all the three sites, at 95% confidence level (see
Table 1). The results show that the most intense fire clusters (HH), with an average
FRP of 762.02 MW (i.e., category 3) were recorded in western USA, followed by eastern Australia with an average
FRP of 494.37 MW (i.e., category 2) and Brazil with an average
FRP of 356.77 MW (i.e., category 2). In contrary, the less intense fire clusters (LL) with an average of 7.45 MW (i.e., category 1) were recorded in Brazil, followed by Australia with an average of 18.13 MW (i.e., category 1) and the USA with an average of 59.27 MW (i.e., category 1).
It is also evident from the summary statistics (
Table A1) that forest cover (FC) had the highest fire intensity (FRP, MW), followed by shrublands (SL) across all sites, then cultivated land (CL) in Brazil and Australia and GL in the USA. The higher fire intensity in FC and SL can be attributed to highly combustible tree crown fuel load characterised by leafy-closed canopies and thin branches. Such fires can rapidly propagate, given favourable conditions, to nearby crowns especially in areas with non-fragmented vegetation types such as the Forests. This is consistent with Vadrevu et al. [
66], who found that higher fire intensity for closed broadleaved evergreen and semi-deciduous forests in India than other vegetation cover types based on empirical cumulative distribution functions. Depending on their severity (not studied here), fires may result in higher tree mortality and permanent land-use conversion, especially in Brazil, where food production needs (i.e., agricultural expansion) surpass conservation [
41]. Similarly, the number of AF over FC at the three sites were the highest, followed by SL and CL for the western USA and Brazil and GL in eastern Australia (
Table A1). In contrast, grassland (GL) and other vegetation covers (OVC) showed the lowest fire intensity (FRP, MW) across all sites. This observation is expected since herbaceous vegetation types such as GL have relatively lower above-ground-biomass (AGB) than tree cover types such as FC and SL. Overall, the results indicate that the number of AF and fire intensity characteristics at the three sites varied by vegetation cover type. In fact, the null hypothesis that there is no significant difference in vegetation type-specific fire intensity was rejected at 95% confidence level (
p-value < 2.2 × 10
−16) at all sites.
The class-wise comparisons of vegetation type-specific fire intensity using Pairwise Wilcox test (
Table 2) reveal even more exciting results, showing a significant difference in fire intensities between CL vs. GL, GL vs. FC, FC vs. OVC and FC vs. SL across all sites. In contrast, the fire intensities between CL vs. FC and GL vs. SL were significantly different in Brazil and eastern Australia only (see
Table 2). These differences were mainly driven by different landscape compositions, such as different vegetation types and conditions, temperature and various sizes of the fires at the three sites. For example, the areas with a high number of AF in Brazil and eastern Australia are characterised by an extensive mosaic of CL, GL, SL and FC, while the predominant
AF in western USA was over FC and SL. Therefore, fire intensity over GL was not significantly different from SL and OVC in western USA and OVC only in eastern Australia due to the low abundance of these vegetation types. The observed differences were due to relatively lower herbaceous biomass in GL and CL, which had corresponding lower intensity than FC and SL, which are characterised by thick, woody stems and branches and high density of leaves. When drier, these woody vegetation characteristics result in higher flammability and combustibility; thus, the propagation of wildfires to the broader spatial extent subject to fuel availability and favourable meteorological conditions. Although Brazilian fires were found to be intense, i.e., category 2, in this study, a recent related study [
33] found that it is within long-term normal, while Kelley et al. [
67] showed that meteorological conditions had an insignificant role, i.e., <7%, in their propagation.
Previous studies [
68,
69], based on MCD14ML.005, reported possible omission errors in the detection of active fires as a result of clouds and smoke plumes. It is therefore acknowledged that these errors are inevitable; however, expected to be reduced due to improvements in MCD14ML.006 product [
37] used in this study. Other sources of error, such as the omission of smaller fires, are probable; however, they were not critical in this study since the focus was on major wildfires characterised by large fires that are detectable at MODIS 500 m spatial resolution. Contrarily, errors in the spatial location of
AF and detection efficiency of sensors may be problematic for fire authorities seeking to locate and extinguish actively burning fires and identify the burning vegetation types for habitat-specific fire fighting, to protect endangered fauna and flora. For example, the exact locations of AF are unknown, since each AF point represents the centroids of 1-km MODIS pixels; thus, several fires may be burning within 1-km pixel as the size of the fire relative to pixel size is always small [
50]. Moreover, Hyer and Reid [
70] showed that using MODIS, an accuracy of identifying the correct vegetation type of a single
AF is approximately 88%. This error may be exacerbated by classification errors in land cover maps. Although algorithmic advances for characterising AF at Landsat resolution were achieved [
71], the current temporal configurations (i.e., repeat cycle of ≥ 5 days) of medium to high resolution sensors limits their operational application for fire management.
4.2. Spatio-Temporal Variations in the Burned Area over Various Vegetation Types
The results (
Figure 3) show various peaks in the burned area for specific dates. In western USA, three prominent peaks were observed in August 2018, with the highest BA per day of ~500 km
2. In contrast, the BA, in Brazil, increased consistently in August 2019 and peaked to the highest (i.e., ~4300 km
2) in September 2019. Although extensive BA are common in Brazil [
68,
72], this finding is consistent to Lizundia-Loiola et al. [
33], who found that the total BA from these fires was more than twice the previous year (i.e., 2018); however, it was similar to an average BA over 17 years. Therefore, the 2019 Brazilian fires BA accumulation and distribution were not extraordinary. On the other hand, the highest peaks, of up to ~2700 km
2 in BA per day, were recorded in November and December 2019 in eastern Australia. Comparatively, the peaks in BA per day in the western USA were the lowest among the three sites, followed by eastern Australia, while the highest peaks were recorded in Brazil. The observed higher daily BA can be attributed to its vast area with high continuous fuel load relative to other sites. Moreover, August and September are at the end of the dry season in most of Brazil, providing favourable conditions, i.e., drier and expansive fuel load, for propagation wildfires. Since these fires are mostly anthropogenic, i.e., mainly related to common clearing activities and pasture management [
67], little or no effort is taken to combat their spread until they encroach on protected areas. On the other hand, western USA fires were confined to smaller areas, attributable weaker winds (Ws) (i.e., ranging between 4 to 6 m s
−1, see
Figure 3 and
Figure A1) that allowed prolonged burning of various vegetation types. Similar to Brazil, the fires in the western USA occurred mainly towards the end of the season, when preceding higher temperature have dried-out the vegetation. On the other hand, the fluctuating daily BA in eastern Australia, i.e., several peaks (maxima) and troughs (minima), can be attributed to varying fuel (i.e., vegetation) type and conditions. For example, the observed higher temperature (i.e., 30 °C to 40 °C) increased evapotranspiration, resulting in drier twigs and leaves especially in eucalyptus forest, thus supporting propagation of fires and higher BA [
45]. Conversely, patches of rainforests impede the spread due to higher moisture content [
45]. Across all sites, Ws were weak (i.e., 4 to 8 m s
−1); thus they cannot be considered an important driver to the propagation of these fires (during the period under study).
Consequently, the fire scars (see
Figure 4) were also variable per site. In agreement with the results in
Figure 3, the fire scars were vastly distributed in Brazil and eastern Australia. In contrast, the fire scars were the lowest and widely scattered in western USA. In Brazil, the fire scars correspond to areas characterised as Cerrado (i.e., Brazilian Savanna) and fire-sensitive ecosystems such as the Amazon and Atlantic rainforests [
52]. This is expected since these biomes occupy the majority of Brazilian territory, i.e., more than 70%. This finding is consistent with Moreira de Araújo et al. [
68], who found the largest proportion of burned areas within Brazilian Cerrado and Amazon Rainforest, i.e., 73% and 14%, respectively. While Cerrado vegetation is more adapted to regular fires, these wildfires may result in considerable damage and loss of the Amazon and Atlantic rainforests which are not tolerant to fires [
52].
As can be observed in
Figure 5, SL, characteristic of Brazilian Cerrado vegetation, burned extensively with a
BA of ~28,000 km
2, followed by FC with ~14,000 km
2, which also had the highest number of
AF, then CL with ~8000 km
2 of BA. It is anticipated that the fires, commonly used for grassland management, hunting activities and ‘slash and burn’ agricultural practices in Brazil [
52,
73], spread into adjacent SL and FC, where they became difficult to control and widespread. However, the higher number AF in FC suggests relatively smaller and isolated fires that could not spread to more extensive areas due to constantly wet conditions in the Amazon rainforest. This is consistent with Lizundia-Loiola et al. [
33], who found newly burned pixels in the Amazon tropical forest in the 2001–2019 period, linked to new policies supporting agricultural expansion. Therefore, relatively lower BA over FC in Brazil can be attributed to fuel availability rather than flammability, as the forests have a deeper root system to withstand dry conditions. Nevertheless, the amount of FC loss to wildfires, in Brazil, was marked and similar to that in eastern Australia, i.e., ~16,000 km
2. This is alarming considering that the entire Brazilian territory was analysed, relative to only a portion, i.e., eastern Australia. Similarly, FC was the most extensively burned vegetation type in the western USA with ~1800 km
2. However, it was markedly lower (i.e., six to seven times lower) than the burned FC in Brazil and Australia, respectively. In western USA and eastern Australia, the BA per vegetation type mostly corresponded to the high number of AF and higher fire intensity. It is expected that the antecedent higher Ts and Ws and low RH resulted in a rapid loss in canopy moisture and dry-up of the herbaceous plant components such as leaves, branches and grasses [
74] (see
Figure A2). At the time of the fires, higher
Ts (i.e., 30 °C to 40 °C) coupled with lower Prec. (i.e., <0.4 mm hr
−1), and availability of continuous fuel (
Figure 1 and
Figure 3) influenced the establishment of new fires, propagation and intensity of the fires in FC, SL and GL, as evidenced by the higher number and clustering of AF and fire intensity (
Figure 2,
Table 1), and extensive losses in vegetation cover (
Figure 5). In fact, at the three sites, moderate to extreme dry (or drought) conditions were evident (
Figure A2), causing drier leaves. This is consistent with Nolan et al. [
45], who showed that the widespread fires in eastern Australia were driven by dry fuel conditions, caused by prolonged drought, i.e., before and during the fire. In another study, Turco et al. [
75] show that fire occurrence is significantly related to drought conditions. Similarly, Moreira de Araújo et al. [
68] found that the higher concentration of fires in 2007 and 2010 over Brazilian Cerrado and Amazon Rainforest was related to low rainfall resulting from La Niña climatic pattern. Contrarily, Kelley et al. [
67] showed that the 2019 Brazilian fires had a low meteorological influence, i.e., <7%, and concluded that land management was the most probable driver of the increase in the burned area. Overall, the interaction of various factors, i.e., meteorological, vegetation availability and conditions and human activities influenced the ignition, flammability, combustion, spatial patterns and intensity of wildfires at the three sites.
Although not studied here, the influence of topography on wildfires is considerable. For example, topographic variables such as elevation influence the temperature, while aspect influences the availability of moisture and heat, and slope influences the vegetation types, intensity and rate of spread of fires [
76]. Moreover, Estes et al. [
76] showed that lower slope positions have a higher probability of experiencing lower flame lengths and lower rates of spread in a backing downhill fire. Future studies should integrate more variables including meteorological, vegetation conditions and topographic variables. Moreover, there is a possibility of exclusion of smaller fires MODIS (MCD64A1, 500 m) product [
19,
77,
78]; therefore, the BA and associated burned vegetation types, in this study, may have been underestimated considering vast climatic regions studied. In fact, a recent study, i.e., Boschetti et al. [
79] found the lowest errors in Boreal Forest, Tropical and Temperate Savanna, and the highest errors in the Tropical Forest, Temperate Forest and Mediterranean biomes. The highest errors in these biomes were attributed to small and temporally non-persistent burns. Therefore, research into operational methods for characterising BA from higher resolution data such as from Landsat 8 or Sentinel-2 should be prioritised to enable representative BA estimate to support fire management and recovery efforts. Moreover, higher resolution and detailed land cover products from sensors such as Landsat-8 OLI Sentinel-2 MSI should be considered in future to characterise burned vegetation types. In the future, we will study the longer periods to characterise meteorological, fuel and environmental parameters before, during and after the wildfires as well as multifactor correlation analysis incorporating topographic (i.e., slope, aspect and elevation). Due to vast fire regimes at the sites chosen for this study, conclusions about local drivers of these wildfires cannot be made; thus, further exploration in future studies is needed.