Searching for a COVID-19 effect on wildfire activity in Portugal but not finding it: A comment on Sci. Total Environ. 765, 142793.

Societal response to the COVID-19 pandemic implied diminished human mobility and activity. A decrease in the number of fires is expected to be a direct impact of COVID-19, with possible further effects caused by pandemic-related restrictions to wildfire suppression preparedness and response. To assess the impact of such measures on wildfire activity, Rodrigues et al. (2021a) examined the anomaly in burned area at country level (Portugal, Spain, France, Italy, and Greece) in the March–May period of 2020 during which lockdowns and other restrictions were in place in those countries. The important influence of fire weather was controlled for by using a drought index, SPEI, the Standardized Precipitation Evapotranspiration Index, retrieved for the study period plus the preceding 3 months (December–February) to account for antecedent weather anomalies.

The SPEI has been used in a number of studies connecting fire activity to seasonal drought. However, drought indices are not expected to depict weather-related fire danger and fire activity as good as fuel dryness indexes specifically developed for such purpose, especially when the former are calculated for lengthy periods of time (Riley et al., 2013; Russo et al., 2017). Consequently, the SPEI should be seen as a supplement, rather than a proxy for fire danger rating (McEvoy et al., 2019). Additionally, the fuels governing fire-spread rate, thus individual fire size and overall burned area, are mostly the fast-drying finer fuel elements (e.g., Cruz et al., 2015). While a given fuel moisture threshold is required for fire to ignite and spread, fire behaviour will respond increasingly less to increasingly higher values of indexes for the moisture of slow-response fuels. For example, starting from a moisture content of ~250%, nearly all fine forest floor fuels in Mediterranean pine forests will be available to burn within one month in the absence of rain (Fernandes and Loureiro, 2013). Finally, wind speed is the primary engine of fire propagation, and more so in open vegetation such as shrubland, where fire spreads as a function of the atmospheric rather than the drought component of fire weather (Anderson et al., 2015). Additionally, fire activity displays regional to local synchronicity, e.g., Jiménez-Ruano et al. (2019), but the fire-weather analysis in Rodrigues et al. (2021a) was, i.e., burned areas per country were associated with the 6-months SPEI for the entire study area.

Rodrigues et al. (2021a) asserted that the “potential effects of weather on fire incidence” can be accounted for by using the SPEI alone; the statement was subsequently moderated in the reply (Rodrigues et al., 2021b) to Resco de Rios (2021). The authors concluded that spring 2020 burned-area in southern Europe was off by one order of magnitude given the level of observed seasonal drought, thus supporting a COVID-19 effect. As the potential shortcomings of the SPEI are manifest and Rodrigues et al. (2021a) results are assumed as preliminary, further analysis is warranted.

Here, and for the specific case of Portugal, we explore to what extent fire activity relates with drought and atmospheric conditions, assessed from the unambiguous perspective of fire danger rating, and then inspect how the March–May period of 2020 compares with previous years. Then we examine whether the number of fires in 2020 diminished, using 2008–2019 as the baseline and controlling for confounding effects, including time, as the number of fires in Portugal has been steadily decreasing since the early 2000s (Rego et al., 2020). Instatement in 2006–2007 of a national fire management plan including changes in policy recommends 2008 as the starting year for the time series under analysis. Further changes in fire management were apparent after the disaster fires of 2017, mostly focused on tighter ignition control measures, fuel treatments in rural-urban interfaces and expanded fire suppression capability.

The Portuguese wildfire information management system allows associating each fire record (defined by a measurable area burned) with the daily fire weather indexes calculated from the nearest weather station data. Portugal, like many other countries, adopted the Canadian Forest Fire Weather Index (FWI) System for fire danger rating. Within the FWI System, the Initial Spread Index (ISI) and the Buildup Index (BUI) respectively describe the atmospheric and the drought (or fuel availability) components of fire weather. Thus, the ISI accounts for the influences of fine dead fuel moisture content and wind speed, and the BUI is an outcome of rainfall, temperature, and relative humidity. Summing the annual (2003−2020) number of fires and area burned for the months of March to May, as in Rodrigues et al. (2021a), and regressing those metrics on the corresponding mean ISI and BUI values enables assessing the relative influence on fire activity of the two fire weather components (Fig. 1 ). The superiority of the ISI in explaining variation (about half of it) in fire activity in comparison with the BUI is manifest. Interestingly, Fig. 1 points to 2020 as the year with both the least fire activity and the mildest fire weather since 2003, differently from outputs reported in Rodrigues et al. (2021a), thus obfuscating the detection of a COVID-19 effect.

Fig. 1

Number of fires and burned area in mainland Portugal as a function of the atmospheric (Initial Spread Index) and drought (Buildup Index) components of the Canadian Forest Fire Weather Index (FWI) System, for the March–May periods of 2003 to 2020 (red dot). The fitted curves are of the form y = a exp(bx).

We fitted a generalized linear model to the 2008–2019 data assuming a Poisson distribution and a log link function to describe daily fire counts in Portugal (n = 3774). The model explains 56.8% of the variability and includes the following independent variables, sequenced by decreasing order of importance and all significant at p < 0.0001: the ISI; year, to account for the ongoing temporal trend of decreased number of ignitions; month, to account for seasonal variability in human-caused fire patterns, e.g., use of fire for land management; and the BUI. Interestingly, the model accounted for 73.2% of the variability in fire counts during 2020. The model was used to estimate the daily number of fires in 2020 from the 1st of January to the 30th of September 30. Then the predictions were summed for each of four 2020 periods, defined by the severity of governmental measures as per INSA (2020), respectively Pre-COVID-19 (weeks 1–12), Emergency (weeks 13–18), Post-Emergency (19–38), and Contingency (39–40). Deviations in the number of recorded fires relative to the number of predicted fires are given in Table 1 , where the deviation is the difference between the observed and the predicted number, divided by the number predicted and then expressed as a percentage. Deviation from the expected number of fires indicates a near half decrease during the most stringent phase (Emergency, from late March to the beginning of May), suggesting an effect of COVID-19 related measures. However, the previous period (prior to any governmental measures) experienced an even higher decrease. While the hypothesis of a pandemic-related decrease in number of ignitions cannot be excluded, the results are more likely to reflect enforcement of stricter policies and greater compliance by people in the aftermath of the catastrophic fire events of 2017. Nonetheless, this putative effect seems to disappear latter in the fire season, as the numbers of fires observed and predicted for the Post-Emergency phase are nearly coincident.

Table 1

Numbers of fires predicted by the model developed with 2008–2019 data and observed in Portugal from the 1st of January to the 30th of September 2020 by COVID-19 phase. Strength of the governmental measures follows the gradient Emergency > Contingency > Post-Emergency.

Phase	Predicted	Observed	Deviation (%)
Pre-COVID-19	1064	385	−63.8
Emergency	551	236	−57.2
Post-Emergency	7701	7574	−1.6
Contingency	477	409	−14.3

Additional insight can be obtained by formally examining the effect of COVID-19 phase on the number of fires. Adding COVID-19 phase to the previous model increases the r 2 to 0.793 (p < 0.0001). A Dunnett test for the model least-square means shows no difference (p = 0.960) between the Emergency phase and the control (the Pre-COVID-19 phase). However, a significantly lower (p < 0.0001) number of fires occurred during the Pre-COVID-19 phase compared with the Post-Emergency phase. The latter includes the summer months, hence presumably higher incidence of intentionally-caused fires that are less likely to be affected by fire prevention measures. Additionally, one can speculate that the unusually mild fire weather of March–May (Fig. 1) may have moved pastoral burning into the summer months.

Models for the annual burned area in Portugal indicate that the relative influence of weather- and fuel-related variables exceeds the influence of the number of fires (≥1 ha) by a factor of 2.5 (Fernandes et al., 2019). Note that these fires are just 20% of the total fire count and that a decrease in the number of fires does not decrease burned area in the same proportion, as a very small fraction of the total number of fires accounts for most of the burned area. It is reasonable to consider, like Rodrigues et al. (2021b), that humans are more decisive to burned area in winter-spring than in summer. Still, the top-down control should not be overlooked: while the use of fire in the frame of rural land management is widespread in regions such as northern Portugal and NW Spain, fire weather alone will decide whether it will be materialized into fire activity. Regardless of variation in the seasonal role of people in starting and terminating a fire, whether or not an ignition will actually spread and attain significant size is conditional on the spatiotemporal context for the ignition, as determined by fire weather and the landscape. This further complicates attempts at attributing changes in fire activity in relation to human behaviour in response to COVID-19, making the link elusive. To conclude, analysis of fire-weather relationships and fire activity, even if preliminary, should consider the mechanisms at play as thoroughly as possible. Adequate understanding of cause-effect relationships will not be achieved otherwise, as shown here for fire activity in Portugal in March–May 2020.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.